latest research topics in power system engineering

• Next Article

ABBREVIATIONS

I. introduction, ii. fundamentals of electric power systems, a. background, b. power system model, c. power system analysis, d. power distribution system data, iii. preliminaries on ml, a. learning foundations, b. taxonomy of learning paradigms, iv. overview of learning paradigms, a. supervised learning, b. unsupervised learning, c. reinforcement learning, d. semi-supervised learning, v. selected ml applications in distribution systems: categorization and overview, a. building models unattainable by first principles, b. accelerating computation and simplifying modeling and simulation, c. handling incomplete data, vi. future directions, a. domain-informed ml, b. trustworthy ml, c. probabilistic ml, d. synthesis and outlook, vii. conclusion, acknowledgments, author declarations, conflict of interest, author contributions, data availability, machine learning for modern power distribution systems: progress and perspectives.

• Split-Screen
• Article contents
• Figures & tables
• Supplementary Data
• Peer Review
• Open the PDF for in another window
• Reprints and Permissions
• Cite Icon Cite
• Search Site

Marija Marković , Matthew Bossart , Bri-Mathias Hodge; Machine learning for modern power distribution systems: Progress and perspectives. J. Renewable Sustainable Energy 1 May 2023; 15 (3): 032301. https://doi.org/10.1063/5.0147592

Download citation file:

• Ris (Zotero)
• Reference Manager

The application of machine learning (ML) to power and energy systems (PES) is being researched at an astounding rate, resulting in a significant number of recent additions to the literature. As the infrastructure of electric power systems evolves, so does interest in deploying ML techniques to PES. However, despite growing interest, the limited number of reported real-world applications suggests that the gap between research and practice is yet to be fully bridged. To help highlight areas where this gap could be narrowed, this article discusses the challenges and opportunities in developing and adapting ML techniques for modern electric power systems, with a particular focus on power distribution systems. These systems play a crucial role in transforming the electric power sector and accommodating emerging distributed technologies to mitigate the impacts of climate change and accelerate the transition to a sustainable energy future. The objective of this article is not to provide an exhaustive overview of the state-of-the-art in the literature, but rather to make the topic accessible to readers with an engineering or computer science background and an interest in the field of ML for PES, thereby encouraging cross-disciplinary research in this rapidly developing field. To this end, the article discusses the ways in which ML can contribute to addressing the evolving operational challenges facing power distribution systems and identifies relevant application areas that exemplify the potential for ML to make near-term contributions. At the same time, key considerations for the practical implementation of ML in power distribution systems are discussed, along with suggestions for several potential future directions.

Artificial intelligence

Artificial neural network

Distribution optimal power flow

Distributed energy resource

Deep neural network

Distribution system state estimation

Electric vehicle

Fault location, isolation and service restoration

Generative adversarial network

Gaussian process

Graph neural network

k-nearest neighbors

Machine learning

Optimal power flow

Principal components analysis

Phasor measurement unit

Photovoltaic

Power and energy systems

Supervisory control and data acquisition

State estimation

Semi-supervised learning

Support vector machines

Vehicle-to-grid

Weighted least square

Countries worldwide have set imperative targets for lowering greenhouse gas emissions to combat the worst impacts of climate change. 1 As part of these climate measures, efforts to decarbonize the electric power sector rely heavily on deploying renewable energy sources—predominantly wind and solar—at scale. 2 Reaching greenhouse gas emissions reduction targets also necessitates shifting transportation energy demand from fossil fuels to electricity through extensive vehicle electrification and supporting charging infrastructure, leading to a substantial increase in electricity demand. 3 As the demand for electricity grows and the requirements for decarbonization persist, both the importance and the complexity of electric distribution systems are rising.

The integration of small-scale distributed energy resources (DERs), such as solar photovoltaics (PVs) and electric vehicles (EVs), into power distribution systems has increased dramatically in recent years. However, due to their intrinsic variability, uncertainty, and limited controllability, these new assets and technologies bring new challenges to the reliable and secure operations of electric distribution networks. Increasing numbers of DERs are exposing the limitations of existing analytical tools for operating and managing electric distribution networks, primarily because these tools were designed for traditional distribution networks wherein uncertainty arises only from the consumption profiles, and power is supplied predominantly from the transmission system to the distribution system. Meanwhile, the variability and uncertainty from weather-dependent renewable generation, along with limited sensing and visibility into distribution network operations, pose new challenges. These challenges will loom larger in future grids without advanced operation and planning methodologies to improve operational flexibility and guarantee network security, thereby maintaining stable voltages and frequencies. 23 Artificial intelligence (AI) and, particularly, machine learning (ML) have attracted considerable research attention as promising solutions to help address these challenges. The implementation of ML solutions is envisioned to complement or potentially even replace traditional, long-established physics-based modeling approaches used in various aspects of power system analysis. In this context, ML applications cover nearly every area of interest, including generation, transmission, distribution, and consumption, and extend into coupled energy infrastructure systems, such as heating, gas, and transportation. Similarly, ML applications in power systems can cover a broad range of timescales, ranging from sub-second intervals for transient stability to decades for planning. Table I summarizes recent review papers on various power and energy system (PES) domains.

A summary of the most recent review papers on AI/ML for electric power and energy systems in chronological order.

Abbreviations used for different power system domains: G, generation; T, transmission; D, distribution; C, consumption.

A conventional notion of the ML process involves a learning algorithm using input data to find hidden patterns and structures in data, extract new insight from the data, or make predictions automatically. When a priori information (e.g., initial conditions) and knowledge (e.g., physical and mathematical principles) are incorporated into the learning process, what was previously purely data-driven (also commonly referred to as physics-agnostic or model-free ) now becomes a hybrid methodology (also commonly referred to as physics-informed or scientific ML ), more suitable for engineering applications ( Fig. 1 ). A concrete example of this emerging learning methodology is the category of deep learning algorithms referred to as physics-informed neural networks. 24 The latest survey by Huang and Wang 25 provides a thorough review of the application of physics-informed neural networks in the domain of power systems, making it a valuable read for those interested in the topic.

ML alternatives (1) and (2) to physics-based models. Physics-informed models (2), unlike black-box models (1), need less data, which is reflected in the smaller box size pertaining to data sources.

In general, ML encompasses a wide variety of methods that can learn from (i) experimental data (in the case of PES, this is often data derived from computational simulations), (ii) observational data describing physical processes within the system (e.g., power system measurements), or (iii) both (observational and experimental/simulation data), to build predictive or explanatory models. As such, these methods may offer many advantages compared to traditional power system planning, operations, and control methodologies. Some examples include, but are not limited to: (i) reduced computational complexity for solving optimization-based problems prevalent in power systems; (ii) higher accuracy in linear modeling of nonlinear power flow equations and other linear function approximations, leading to improved solution accuracy and/or reduced computation time; (iii) the ability to process and analyze large volumes of measurements collected from heterogeneous data sources to derive valuable insight or predictions from such measured data; (iv) the capability to manage ill-posed problems; (v) more informed decision-making under growing uncertainty from DERs; (vi) intelligent support for fault diagnosis and monitoring in complex distribution networks; and (vii) learning hard-to-model functions, such as consumer behavior for demand response mechanisms that do not provide a closed system of equations. See Fig. 2 for a depiction of some potential application areas of ML in power distribution systems.

New power system operational paradigm shift with prominent ML applications in power distribution systems.

Consistent advances in AI have been fueled by the advent of massive amounts of data and the utilization of high-performance parallel computing using AI accelerator hardware like graphics processing units. 26 More recent advances in ML are characterized by the development of new algorithms, particularly within deep neural networks (DNNs) 27 and graph neural networks (GNNs). 28 As AI/ML advances at an unprecedented rate, it will likely have a significant impact on how power systems are operated and planned. Despite active research devoted to applying state-of-the-art ML techniques to power system problems, there is often a lack of attention paid to evaluating the suitability of the technique for a specific problem, as discussed in Sec. VI . For instance, data-intensive methods like DNNs may not be well-suited for power system problems where large amounts of data are not readily available. Similarly, while GNNs show promising advantages, many obstacles remain to be overcome for their usage in power distribution systems. 29 To address this gap, this article aims to provide insight into important topics concerning (i) ML for electric power systems with a particular focus on electric distribution networks, and (ii) challenges unique to this interdisciplinary research area. Unlike previous works that have either broadly covered these topics or targeted specific application areas or ML methodologies, this paper concentrates on the general ML efforts required for electric distribution systems, specifically the D and C domains from Table I . It does not aim to be a comprehensive survey of all plausible applications nor a thorough literature review, as that would be too ambitious. Instead, this paper discusses applications within the distribution system domain that can benefit from a broad range of ML concepts and techniques by identifying suitable application areas. The goal is to foster and stimulate interdisciplinary research between the two disciplines to guide future research and realize the potential of ML in power distribution systems.

The remainder of this paper is organized as follows: Secs. II and III provide background reading on power systems and ML, respectively, from basic concepts and theoretical fundamentals to state-of-the-art techniques. Section IV provides relevant examples from the distribution system domain in the context of each main ML paradigm. Section V overviews selected distribution system application areas where ML can realize greater potential but still needs to be explored. Finally, Sec. VI discusses open problems and recommendations on research directions, while Sec. VII concludes the work.

This section provides fundamental background reading on electric power systems, specifically focusing on power distribution systems. It is intended for readers with little familiarity with the topic and sets the stage for later discussions by considering various distribution system problems where ML can realize great potential. The section begins with an introduction to key concepts pertinent to power system operations and planning. It then proceeds to describe the basic power system modeling aspects. Additionally, it briefly touches on the advantages offered by ML in power system analysis, reserving more detailed discussions for Secs. IV and V . Finally, the section concludes with an overview of power system data.

An electric power grid is a complex system consisting of a transmission network that transmits electricity over long distances using high-voltage transmission lines and a distribution network that delivers electricity to consumers using low-voltage distribution lines. In traditional power systems, electricity flows unidirectionally from centralized, large-scale synchronous generators to loads. However, the increasing prevalence of DERs in modern power distribution systems has given rise to bi-directional power flows, resulting in a greater incidence of over-voltages, as illustrated in Fig. 2 . These and other issues arising from the clean energy transition pose challenges to the operational reliability of power grids.

An essential requirement for reliable and secure power grid operations is a continuous balancing of electricity supply and demand, differences between which are manifested in changes in the grid frequency (preferably a constant 60 or 50 Hz depending on the country). 30 Any imbalance leads to a change in frequency, which drops from its nominal value in the case of insufficient generation and increases with oversupply. Traditionally, generation and load balancing have been achieved by scheduling (as part of the unit commitment problem) and dispatching (as part of the economic dispatch problem) a controllable generation fleet, such as coal and natural gas power plants, one day ahead and varying their output in real-time to match the varying load (as part of automatic generation control ). Unit commitment and economic dispatch comprise a production cost model formulated as a mixed-integer programming problem that minimizes the bulk power system's total operating costs while adhering to the transmission network's and other physical constraints. In contrast, the capacity expansion model —also an optimization problem but with different objectives—focuses on planning and policy aspects of bulk power systems. The so-called optimal power flow is at the heart of these and other optimization problems and is ubiquitous in power systems (see Sec. II C 2 ).

Security of supply, paramount to reliable and secure power grid operations, is becoming more challenging with a high share of variable renewable energy. 23 These changes call for new services at all system levels, from ancillary services provided by renewable energy sources to optimal market frameworks for obtaining those services. Distribution utilities can only meet these new requirements by significantly improving current operational and planning practices, creating opportunities for active ML research in various distribution system contexts, as shown in Fig. 2 .

The structure of the electric power system (known as the topology )—referring to the physical arrangement and connection of system components—can be conceptualized as a graph composed of nodes ( buses ) to which generators and loads are connected and edges ( lines ) connecting these nodes, as illustrated in Fig. 2 . Depending on the problem at hand, the topology of the power system can be represented by a directed or undirected graph. 31 This representation facilitates the mathematical derivation of the power system model used in all power system problems, from operations and control to planning. In general, the power system model comprises a set of equations describing the relationships between variables of interest within the timeframe under study while accounting for various components and their more-or-less simplified models, such as transformers, circuit breakers, loads, lines, and cables. Notably, the distribution network model differs from the transmission network model in that it has (i) a radial or weakly meshed network topology that frequently changes; (ii) high resistance-to-reactance ratios; and (iii) unbalanced phases due to the presence of many asymmetrical loads (single-phase and two-phase loads, primarily in North America). These network model differences imply that the distribution and transmission systems analysis also differs. For example, a direct implication of (ii) is that line resistances cannot be neglected, a simplification commonly used for traditional transmission system analysis.

This section presents a brief overview of three fundamental problems in power system analysis—the power flow problem, optimal power flow (OPF), and state estimation (SE)—with a focus on challenges related to distribution system analysis. The discussion frames the challenges in the context of ML applications, thereby serving as an introduction to the use of ML in power system analysis. As such, this section is a good primer for readers interested in exploring the intersection of power systems and ML.

1. Power flow problem

The power flow problem, also known as the load flow problem, is instrumental for steady-state power system analysis. It entails determining the state of the system under normal operating conditions, including the voltage magnitudes and angles at all nodes within the network. 32,33 Once the system state is known, any other relevant quantity such as the real and reactive power flow and losses on each line can be analytically calculated. 32,33 This information is crucial for power system operators (hereafter referred to as operators for short) to ensure reliable and secure power system operations as it enables them to understand system behavior under different operating conditions, identify potential issues, and take corrective measures to maintain safe operating conditions. As a result, power flow analysis is a fundamental aspect of key power system operational tasks, such as OPF and SE, which will be further discussed below.

To avoid physical inconsistencies or implausible outcomes resulting from imperfect data or noisy measurements that can negatively impact the performance of ML models, it is necessary to move beyond purely data-driven models like (5) . 34 This is where physics-informed models ( Fig. 1 ) come into play and offer a promising research direction to pursue; see Sec. VI A 3 for a detailed discussion.

2. Optimal power flow

Linear approximations. This set of solution approaches is computationally efficient; however, it yields less accurate, sub-optimal solutions that may violate operational constraints.

Convex relaxations. Although computationally more intensive than the previous solution methods, the main advantage of convex relaxation approaches is convergence to the global optimum and feasibility guarantees.

Black-box or hybrid solvers. This set of solution approaches is ML-based, may be data-prohibitive, and typically does not provide optimality guarantees. Within this group, we can distinguish between end-to-end learning, which is the most popular in the PES literature, and learning-to-optimize approaches for OPF, as discussed in Ref. 36 . The end-to-end learning approach employs supervised learning to map an input, such as nodal net power injections, to an optimal solution, such as voltages or power generation, resulting in high-fidelity optimization proxies for OPF solutions. Conversely, the latter approach leverages ML techniques to accelerate the computational speed of existing optimization algorithms used to solve OPF; one such example is using ML to facilitate the warm starting of OPF solvers. 36 As such, some of the learning-to-optimize approaches for OPF can be sub-categorized into physics-based ML models, which are also known as hybrid solvers.

For a detailed survey on approximations and relaxations of power flow equations in the context of OPF, particularly for the first two solution approaches listed above, the reader is referred to Ref. 37 . More details on the third solution approach listed above can be found in Ref. 36 .

a. Challenges

OPF techniques have been effectively used to optimize bulk power system operations, such as economic dispatch of the generation fleet. However, applying the same OPF methods to distribution systems is challenging due to the differences outlined in Sec. II B . Moreover, the increasing presence of DERs in modern distribution systems exacerbates uncertainty, making distribution OPF (D-OPF) even more complex. As a result, operators need D-OPF solutions that can be executed in near real-time to accurately reflect system operating conditions and make timely decisions under uncertainties. Despite the significant number of published papers on the D-OPF problem, it remains an active area of research. 37  

b. ML applications

The recent literature has seen many instances of ML being applied directly to OPF/D-OPF or its variants, including security-constrained OPF and probabilistic OPF that takes uncertainty factors into consideration. For a brief overview of the use of ML in OPF, refer to Ref. 38 . One exciting and promising research direction involves physics-informed neural networks for solving OPF while imposing constraints on neural network optimization functions. 39 The interested reader is referred to Ref. 40 and references therein for additional details.

3. State estimation

SE is a critical inference task for enabling system-wide situational awareness by determining the most-likely operating state of the power system, typically represented by complex voltage phasors comprising voltage magnitudes and phase angles. To achieve this, SE requires two key inputs: (i) telemetry data available in sufficient quantity to make the network observable, and (ii) an up-to-date network model that includes network parameters and topology. The primary objective of SE is to provide an accurate real-time estimate of the system state (denoted by x ) that is consistent with the available measurements, primarily current and voltage magnitudes in distribution systems. 41 Specifically, the measurement vector z can be represented as z = h ( x ) + e ⁠ , where e denotes the measurement error. The nonlinear function h ( · ) relates system states to these measurements by using the previously defined (inverse) power flow equations.

Remark 3. Measurement redundancy and accurate knowledge of the network model are prerequisites for a classic SE framework, such as weighted least squares (WLSs). 41,42 The WLS state estimator has been used for decades in transmission systems; however, it is unsuitable for distribution systems where the SE problem remains undetermined due to the low observability of the network, as discussed in the Challenges section below. Furthermore, even if the network is observable, the WLS state estimator, or similar formulations, cannot guarantee a solution in all cases and may encounter convergence problems, especially for large-scale systems. Moreover, these state estimators are sensitive to measurement errors and bad data. 41,42

The direct application of the classic SE framework in distribution systems is severely challenged for three reasons: 41,42 (i) low observability due to scarce measurement devices and insufficient communications infrastructure; (ii) poor network models resulting from network model uncertainty, that is, imprecise network parameters and incomplete topology information; and (iii) unbalanced operations. Presently, distribution state estimators rely on so-called pseudo-measurements 43 due to the difficulties in addressing (i) and (ii) without substantial upgrades to existing communications and metering infrastructures. Unfortunately, such infrastructure upgrades are economically impractical due to the vast scale of distribution systems. On the other hand, load-derived pseudo-measurements used to compensate for insufficient telemetry data can propagate significant errors through the distribution state estimator, leading to their unreliable performance. 41,42 Moreover, (iii) renders the decoupled versions of state estimators used in transmission networks inadequate for distribution systems, requiring the development of three-phase state estimators. Hence, there is a growing interest in exploring techniques that enable accurate distribution system state estimation (DSSE) despite limited measurements and uncertainties in system models. Among these techniques, researchers have focused on ML 44,45 and sparsity-based approaches to DSSE, which are discussed below.

ML techniques have shown promising performance in various aspects directly or indirectly related to SE, such as bad data detection, 46 topology identification, including transformer-to-customer mapping and phase connectivity identification, 47 and generation and modeling of higher fidelity pseudo-measurements. 48–50 Furthermore, numerous ML-based state estimators have been proposed that can be broadly classified into two different categories, depending on whether they require difficult-in-practice knowledge of the distribution network model (model-augmented) or not (model-agnostic or model-free). Notably, many ML-based state estimators proposed in the literature assume a level of observability that exceeds what is usually observed in real-world distribution systems or employ sensor placement optimization strategies. 51 Therefore, there are numerous opportunities for further research in this field.

Last, but not least, it is worth highlighting a specific methodological concept utilized within the realm of recommendation (or recommender) systems that is relevant to research in power system SE. For an introduction to recommender systems, see, for example, Ref. 52 . While recommendation systems are commonly linked to platforms such as Netflix, YouTube, and Amazon, where ML is utilized to suggest items of potential interest to individual users based on data from other users, the mathematical concepts behind these systems can also be applied in the power system domain. Specifically, low-rank matrix completion has been used in the area of DSSE, particularly in the context of low-observable distribution networks. For further information, see Refs. 53 and 54 , which provide examples of model-augmented and model-free approaches to voltage estimation, respectively. Apart from matrix completion techniques, the domain of sparsity-based DSSE has seen active research on other pertinent techniques, such as tensor completion. 55–57  

4. General considerations on ML for power distribution system analysis

In power distribution system analysis, high-fidelity modeling is essential for obtaining trustworthy results. Traditional power distribution system modeling relies on complex physics-based models and rigorous mathematical principles. However, these models can lead to intractable optimization and simulation problems given the large scale of these systems. To address this scalability issue, power distribution system modeling is often simplified through assumptions and approximations that come in many forms, like linearization or convex relaxation. Such simplified models may be satisfactory for most operating points but can lead to sub-optimal solutions for others. This becomes more pronounced with increasing variability and uncertainty in system operating conditions due to the increased integration of DERs.

There are many examples where ML can offer an efficient alternative to conventional simplifications, especially where obtaining timely solutions is a primary concern, at the expense of solution accuracy. One such example is the use of ML to develop surrogate models —high-accuracy, computationally cheap approximations of a detailed analytical model—to accelerate analysis. Examples may include convex and nonconvex optimization problems (e.g., D-OPF, DSSE), Monte Carlo simulations, and approximation of nonlinear power flow equations embedded in optimization-based problems.

Computationally intensive Monte Carlo simulations are traditionally used to generate hundreds to thousands of scenarios pertaining to the different system operating conditions (i.e., power generation, consumption, and voltage level) for various relevant studies, such as security assessment under different DERs penetration levels. Identifying the most critical scenario (i.e., operating point) is highly challenging because the system state space can be vast. 58 To that end, ML techniques can be leveraged to optimally search the state space, thus circumventing the need to run distribution power flow analysis for each scenario individually.

By reducing the time burden of various analyses, ML can allow for more comprehensive and/or more frequent optimization, therefore improving the efficiency and reliability of the system. Additionally, because time-intensive model solving often limits the above problems to offline applications, ML methods are promising tools for shifting model solving from offline to online settings, which are more suitable for time-varying operational conditions. It is important to note that by replacing the original detailed physics-based analysis with ML surrogates, certain guarantees on performance may be lost, therefore limiting the application of ML to high-regret applications, especially in the context of system stability.

We close this section with an overview of the various types of data sources used in power distribution systems, accompanied by a brief discussion of important considerations to keep in mind when working with these data. As depicted in Fig. 3 , diverse data sources, such as supervisory control and data acquisition (SCADA), weather stations, meteorological databases, and metering devices, including μ phasor measurement units ( μ -PMUs) and smart meters, can be exploited for various ML applications. However, it is important to note that these data come with unique challenges, such as asynchronous measurements and limited access to real-world measurements.

Heterogeneous data sources in power distribution systems.

The inconsistency in the temporal resolution of distribution measurements originates from the spatiotemporal heterogeneity of sensors installed throughout the electric distribution network, such as smart meters, μ -PMUs, and field equipment monitors. These sensors have varying sampling rates and reporting times, as shown in Fig. 4 , which results in measurements of different temporal resolutions. To effectively utilize these measurements in ML systems, it is necessary to standardize their resolutions. This requires downsampling the higher-resolution data to match the lower-resolution data or vice versa (the so-called upsampling). This can be done in many ways, from trivial linear interpolation methods to more advanced ML methods; we highlight recent work 59 where low-resolution load profiles are upsampled into high-resolution load profiles using generative adversarial networks (GANs).

A range of temporal resolutions of distribution measurements. Adapted from Ref. 60 .

Another important consideration is the limited availability of measured data from real systems due to a lack of sensors in addition to privacy and security concerns. For these reasons, researchers are often limited to using simulated and open-source data to develop ML systems. However, simulations typically assume error-free measurements and no communication delays, which is unrealistic for real-world measurements. To address this issue, researchers may introduce noise that follows certain assumed distributions, which may not accurately reflect the noise patterns found in actual measurements. Additionally, open-source datasets may have undergone pre-processing and may lack certain features that could be relevant to the problem being studied.

This section serves as a brief introduction to the field of ML for audiences outside that community. It delves into the fundamental principles of ML and categorizes the different learning paradigms. By providing examples from a distribution systems perspective, it illustrates the concepts discussed and serves as a starting point for researchers to explore the potential of ML within the power engineering domain.

A learning task T can be described as the process of optimizing a performance measure L driven by a training experience D ⁠ . 26 This task typically involves three main components: the input data (also known as the training set or feature set), the model (or algorithm), and the output or prediction (also known as the label or target). 26 To simplify the discussion, in the following, we will formally define the learning task assuming the supervised learning setting, which is covered in more detail in Sec. IV A . Not only has this type of learning been widely studied in the field of PES, but it also provides a sound basis for explaining the fundamental concepts of ML.

To begin, let us briefly apply the abstract ML terminology above to a concrete example within the distribution systems context. In power systems, a microgrid is defined as a small electricity network that can operate either connected to the larger grid or in an autonomous mode. Consider the example of microgrid formation to facilitate distribution network recovery during low-probability, high-impact events, such as extreme weather events. The learning task in this scenario can be formulated as segmenting the distribution network into autonomous (microgrid) zones. The performance measure to be improved in such a learning task may be the minimum time required for the distribution network's restoration or alternatively the number of critical customers facing power outages. The training experience may involve defining all possible microgrids within the observed network beforehand.

In (8), ℓ is a loss function used as a learning criterion for the optimization problem (8a) , and R is the optional regularization term used to reduce the risk of overfitting.

Remark 4. What does a good model mean in the definition above? In the field of ML for PES, the term good model is not clearly defined and is open to interpretation. Generally, a good model is one that generalizes well to previously unseen data. However, when it comes to PES, other criteria should be considered when defining a good model. For example, features of a model that address barriers to adoption in the real world should be considered. Nevertheless, the lack of a clear consensus on what constitutes a good model in PES research can lead to difficulties in the practical application of ML methods in this field. Therefore, researchers must establish clear standards and definitions for good models in PES in order to make the use of ML techniques successful in the real world. Developing realistic benchmark problems can help the community at large agree upon a consistent definition of a good model for each application of ML to PES.

1. Loss function

The accuracy of an ML algorithm is quantified using the loss function—a mathematical function that measures the difference between the predicted output (also known as the prediction or estimate ) produced by the ML algorithm and the actual output (also known as the ground truth ). The loss function is optimized during the training process using techniques such as gradient descent to enhance the model's predictive efficacy. 61  

2. Data requirements

As discussed above, the learning task is driven by the training experience, or in other words, the data. However, raw data are rarely used in the learning process; they often require prior processing and formatting. These steps generally fall under the umbrella of data engineering and are vital to developing quality ML systems.

Data cleaning , which is used to detect and correct or remove corrupt observations—missing data and outliers;

Feature selection , which involves selecting an “optimal” subset of independent variables (features) from among many less useful ones; this implies ignoring other features as irrelevant. These techniques are instrumental when there are many features and comparatively few data observations;

Feature scaling , which involves normalizing the range of data features. Examples include mean normalization, min–max normalization, and standardization.

The second step (data formatting) is to ensure that the input data are well formatted. The most common is the tabular format (the so-called vector–matrix data representation) where each row and column of the table represents a particular example (also called instance ) and feature (also called covariate ), respectively. Let us consider a single distribution feeder and the corresponding past measurements at each node as an illustrative example of how this system can be represented in tabular form. In such a table, each row corresponds to a time instance when the measurement snapshots were made, and columns correspond to different measured variables at each node (e.g., voltages, injected complex power, consumed complex power).

Remark 5. In the context of ML applications within the distribution system domain, the data utilized are primarily structured in a tabular format. There are, however, certain exceptions to this, such as satellite imagery that can be used to estimate behind-the-meter solar generation or thermal images from distribution lines and transformers inspections that can be used for predictive maintenance.

3. Model performance evaluation

Training set , typically the largest sample of the data, is used to learn a predictive ML model;

Validation set is a smaller data sample used to evaluate the model's performance during training and fine-tune its hyperparameters to prevent overfitting or underfitting to the training data;

Test set is an independent dataset that is not used in the training or hyperparameter tuning process. Instead, it is used to estimate the performance of the predictive ML model on new, unseen data.

Another technique for evaluating and selecting ML models is cross-validation. 63 In this approach, the data are partitioned into subsamples, and the error rate is estimated as the mean of the error rates calculated from all the data subsamples. 63  

The field of ML encompasses various learning paradigms, including supervised, unsupervised, and reinforcement learning. These paradigms have been extensively studied and widely adopted across various application domains. 26 A summary of their key characteristics and related algorithms is presented in Fig. 5 , color-coded to help differentiate them in Sec. IV , where they will be discussed individually in more detail. In Secs. IV A–IV C , we will focus on these main learning paradigms, including semi-supervised learning (SSL) in Sec. IV D . For a comprehensive overview of recent advancements in the field, including transfer, multitask, and multiview learning, the interested reader is referred to Refs. 64 and 65 .

Main characteristics and representative algorithms of the four ML paradigms of interest.

Among the different ML paradigms, a noteworthy distinction is the difference between offline and online learning. In offline learning, a batch of data samples is processed simultaneously, while in online learning, data samples are processed sequentially as they arrive over time. Although supervised and unsupervised learning methods can be implemented using either offline or online learning strategies, these methods are traditionally associated with offline realization. In contrast, reinforcement learning, which relies on sequential interactions between the agent and the environment, primarily operates online. Nonetheless, there are also less represented forms of reinforcement learning, such as batch reinforcement learning that decouples data collection and policy training processes, thereby updating an agent's policy with a batch of pre-collected data. 66 Another such form is pure batch (or offline) reinforcement learning, which aims to learn policies solely from a suitably diverse and sizable dataset, devoid of any online interactions. 67  

Another distinction is between discriminative and generative models. Discriminative models learn the conditional probability distribution p ( y | x ) while generative models learn the joint probability distribution p ( x , y ). 68 While discriminative models are primarily used for classification tasks, they can also be used for other types of supervised learning tasks, such as regression and structured prediction, wherein the output is a structured object, such as a sequence or a tree. 69 On the other hand, generative models are primarily used for unsupervised learning tasks of generating new data samples that resemble the training data. Generative models are particularly promising for applications where the training dataset is small or difficult to obtain (see Sec. IV B ).

This section provides a succinct overview of the primary learning paradigms outlined in Sec. III B and displayed in Fig. 5 . At the end of each subsection are examples in order to demonstrate how the methods can be utilized in the PES domain. Most of the selected examples, including but not limited to predictive maintenance, phase identification, and voltage control, have been extensively studied over the years and are therefore well-suited to illustrate the considered ML paradigms in the context of power distribution systems.

Classification can be further divided into two tasks. The first is binary classification, which aims to determine the class (out of two possible classes) to which a given data instance belongs. The second task is multiclass classification, which assigns a data instance to one of K categories. In the context of power systems, this could include determining if a fault is an open-circuit fault, symmetrical short-circuit fault, or asymmetrical short-circuit fault. Some popular classification algorithms include logistic regression, support vector machines (SVM), k-nearest neighbors (k-NN), and decision trees. Ensemble variants of decision trees, such as random forests, are also commonly used.

Regression is a task that involves making predictions about future outcomes by learning the relationship between a dependent variable and one or more independent variables. In the context of power systems, an example of a dependent variable could be the generated power from a PV system, while independent variables could include factors such as solar irradiance, ambient temperature, and PV system specifications. The common regression algorithms include linear regression, polynomial regression, SVM, neural networks, and random forests. These algorithms are widely used in the PES field to make predictions and projections based on historical data.

It is important to note that many supervised learning techniques, including decision trees, random forests, and artificial neural networks (ANNs), can describe complex, nonlinear relationships between input–output pairs. Among them, ANNs have been widely recognized as universal nonlinear function approximators, capable of approximating arbitrary functions with high accuracy. 70 Recent advances in DNNs have further improved the capabilities of ANNs. DNNs utilize multiple layers of nodes to capture increasingly complex data relationships, making them the most powerful ML technique studied extensively across various research fields, including PES.

Supervised learning is arguably the most extensively studied learning paradigm in PES research. Two selected examples showcase the application of supervised learning in forecasting tasks related to distribution systems (Example 1) and for predictive maintenance (Example 2).

Forecasting is crucial in mitigating the increasing uncertainty in distribution system operations and planning. Supervised learning can be employed to forecast relevant system variables, such as distribution locational marginal prices 71 and distribution system states, 72,73 by learning from historical time series data. These forecasts can provide valuable information for system operators and decision-makers. For instance, accurate forecasts of renewable energy generation can assist operators in managing grids with a high penetration of distributed generation more economically and reliably, alleviating operational uncertainties. Depending on the forecasting horizon, long-term, short-term, and very short-term forecasting (often referred to as “nowcasting” in the ML community) can be distinguished. The first is used in planning studies with timescales ranging from months to years ahead, while the latter two are used in near- and real-time operations, with relevant timescales as depicted in Fig. 6 .

As the distribution infrastructure ages, more advanced maintenance practices are required to ensure its reliable operations. One such practice is equipment predictive (condition-based) maintenance. 74,75 Supervised learning can be utilized to construct predictive models for this purpose. By utilizing health indicator data of network equipment, such as insulation degradation in transformers, and other readily available data, such as transformer specifications and historical failures, a model can be trained to detect potential equipment failures, thereby allowing operators to take timely preventive measures. This approach can greatly increase equipment longevity and help avoid unplanned equipment failures that can lead to power outages and other service disruptions, thereby increasing the overall reliability of the distribution infrastructure.

Illustration of important distribution system timescales for Example 1: Forecasting pertaining to supervised learning application.

Clustering is a task within unsupervised learning that groups unlabeled data into clusters based on their similarities. Standard clustering algorithms are Gaussian mixture models, k-means, hierarchical, and spectral clustering. For a comprehensive overview of clustering algorithms, see Ref. 76 .

Dimensionality reduction is a technique that is frequently employed in the pre-processing stage of data analysis to represent high-dimensional data with significantly fewer dimensions while maintaining the integrity of the data. There are various dimensionality reduction techniques that can be used, with the most popular being principal components analysis (PCA), 77 and autoencoders. 78  

Anomaly detection is a commonly employed technique for identifying observations that deviate from the norm, also known as outliers or anomalies. Among the various algorithms for anomaly detection, two notable examples are one-class SVM and isolation forest. 79 In the context of PES, anomaly detection can identify abnormal patterns in measured data that may indicate a deviation from expected system behavior, such as diagnosing faults using anomaly detection algorithms on voltage data.

New sample generation involves generating new samples or scenarios with distributions that are representative of the actual distribution of the given data. It is important to note, however, that the quality of the generated scenarios is contingent upon the quality of the input data. One prominent algorithm employed for this task is the GANs, which consist of a pair of neural networks—a generator and a discriminator. 80 The generator is trained to produce data samples that closely resemble the training data, while the discriminator is trained to accurately distinguish between the generated samples and the actual training data. 80 This process is conducted in an adversarial manner where both models are trained simultaneously. 80 Unlike traditional scenario generation methods like Monte Carlo simulations, GANs do not require a priori assumptions about the probability distributions of the input data, making them a more attractive option for PES applications. As a result, GANs have gained considerable attention in recent years, and various studies have demonstrated their efficacy in generating near-realistic scenarios for a wide range of applications. These include profiles for electricity demand, solar PV, and wind generation. 81,82

Unsupervised learning has received less interest in PES research due to its lower accuracy compared to its supervised counterpart. Nonetheless, unsupervised learning can be highly useful, particularly for analyzing unlabeled data or in situations where there is not enough labeled data to use supervised learning techniques. In the context of power distribution systems, two common applications of unsupervised learning techniques are phase identification and customer segmentation. Two selected examples (Example 1: Phase identification and Example 2: Customer segmentation) are presented below to illustrate these applications.

The identification of phase connections in secondary distribution networks presents a significant challenge due to the vast size of these networks. Unlike transmission and primary distribution networks, the phase connectivity data in secondary distribution networks are often unknown, incomplete, or inaccurate. To facilitate capacity hosting analysis for accommodating new DERs in secondary distribution networks (e.g., residential and commercial PV installations), it is essential to correctly identify the phase connectivity of each customer. To address this challenge, unsupervised learning techniques can be utilized to identify phase connections using smart meter data, specifically voltage or power consumption measurements. 83–87 By analyzing and extracting valuable insight from these data, customers with similar characteristics can be grouped into representative clusters (seven in total, assuming single-phase, two-phase, and three-phase loads), each corresponding to the same phase or combination of phases (i.e., phase A, phase B, phase C, and combinations thereof as illustrated in Fig. 7 ). However, one potential drawback is the sensitivity of the model to the distribution feeders' levels of unbalanced phases. Further information can be found in Ref. 88 , which presents a review of the current phase identification methods in the literature.

Consumer segmentation is a methodology for characterizing consumers based on their load profiles, which are determined by the characteristics of their interruptible (e.g., computers, refrigerators) and adjustable appliances (e.g., air conditioners, washing machines), as well as their individual preferences and consumption patterns. This process enables distribution utilities to develop targeted customer recruitment strategies, such as personalized pricing and incentives, to effectively engage demand response program participants. One approach for achieving this is through the use of unsupervised learning techniques, specifically clustering methods, to analyze smart meter data and extract consumption patterns. By doing so, utilities can gain valuable insight that can assist in managing demand response under various conditions, including weather, social activities, and holidays. For further information on this topic, the interested reader is referred to Ref. 89 and the references therein.

Illustration of Example 1: Phase identification pertaining to unsupervised learning application. Each distribution line (lateral) is colored according to its phase.

Reinforcement learning is a popular ML paradigm for learning an optimal policy or set of actions through a trial-and-error search and a reward system. 90 The process, diagramed in Fig. 8 , involves an agent interacting with a dynamic environment through a series of actions, each of which leads to a state transition and affects the agent's subsequent actions, all aimed at maximizing a cumulative reward function. In the context of distribution voltage control, exemplified in Fig. 8 , reinforcement learning is used to govern the power output of a PV system, thereby minimizing voltage fluctuations and maintaining acceptable voltage levels across the distribution grid. This topic has been covered in more detail in Sec. V B 1 .

A distribution voltage control framework exemplifying reinforcement learning tasks. At each discrete time step t , the agent, such as a smart PV inverter, receives a representation of the current environment's state s t , which includes the voltage level. Based on this observation, the agent selects an action a t , such as regulating the power injection level. A time step later t  + 1, the environment perceives this action and transitions to the next state s t + 1 while providing a numerical reward r t + 1 to the agent, which affects its next action a t + 1 ⁠ . For example, in the case of a voltage violation where | V | ≥ 1.05 p.u., the agent may initiate power curtailment.

A key distinction between reinforcement learning and supervised learning is the type of feedback provided during training. In supervised learning, training data provide quantitative feedback in the form of actual output for a given input ( ground-truth ). In contrast, reinforcement learning utilizes qualitative feedback, indicating whether the action taken is correct or not. Algorithmic formulations of reinforcement learning include Q-learning, deep variational learning, policy gradient, actor-critic, and state-action–reward–state–action (SARSA), among others.

Reinforcement learning has been extensively studied in recent years for decision-making and control problems in power systems. Within this framework, power system problems are typically modeled as Markov decision processes or variants thereof, such as partially observable Markov decision processes. Broadly, the potential applications of reinforcement learning in power systems can be categorized into two main groups, namely, game-based studies and search-based studies, as indicated in Ref. 58 . Game-based studies cover various tasks, such as demand-side management (e.g., Example 1: V2G), strategic bidding for different electricity markets, and power system control tasks that involve coordination over multiple agents (e.g., Example 2: Voltage control). On the other hand, search-based studies include 58 system fault diagnosis, security assessment, and cascading outage prediction, to name a few. For a detailed analysis of the advantages and limitations of reinforcement learning in decision-making and control applications for power systems, we recommend referring to the latest review paper by Chen et al. 6  

The issue of optimal management of EV fleets in the context of demand-side management is of growing importance. Not only do EVs impose an extra load on the local distribution network, but they also offer a distributed form of potential energy storage through the use of their built-in batteries, referred to as vehicle-to-grid (V2G) technology. To efficiently schedule EV charging and discharging activities, which enables operators to smooth out fluctuating supply from highly variable renewable generation and achieve desired consumption profiles, the literature suggests the application of reinforcement learning techniques. A recent survey on this topic can be found in Ref. 91 .

Maintaining distribution voltage magnitudes within acceptable operating limits, as defined by standards such as ANSI C84.1, 92 is becoming increasingly challenging due to the growing number of customers adopting DERs. This has prompted the development of reinforcement learning-based voltage control solutions, as seen in studies. 93–96 The definitions of environment, state, action, and reward system may vary depending on the specific problem setting. In one such setting, the learning agent assumes the role of the system operator responsible for maintaining voltage magnitudes within the normal operating range ( ⁠ 0.95   p . u . ≤ | V | ≤ 1.05   p . u . ⁠ ). The environment, on the other hand, comprises the aggregate of DERs, with the dynamic aspect referring to changes in system states and associated controls, such as adjustments of renewable generation outputs. The underlying physical model pertaining to distribution network model is regarded as the unknown environment. Within this framework, states correspond to bus voltage magnitudes and actions may entail regulating the distributed generation outputs. The reward can be defined as positive if the voltage magnitude remains within the upper and lower thresholds and negative if a violation, such as a ± 5 % voltage deviation from the nominal value of 1 per unit, occurs.

Self-training 98 uses an initial set of a small amount of labeled data to train a supervised learning model and then applies that model to classify the unlabeled data. The most confident predicted labels are then added to the labeled dataset and used to retrain the model.

Co-training 99 is a variation of self-training where two or more classifiers are trained on different views of data, and their predictions are combined to predict labels for the unlabeled data.

Transductive SVM 100 is a variation of SVM specifically designed for SSL that leverages the unlabeled data to find the decision boundary that best separates the labeled data.

Pseudo-labeling 101 involves training an initial model to predict “pseudo-labels” for previously unlabeled data. The pseudo-labeled and original labeled data are then combined, and the model is retrained on this larger training set. For example, consider the task of classifying a distribution line as either undergoing an outage or operating normally using historical measurements. If only a small portion of the dataset is annotated with the status of the line, the labeled dataset alone is likely insufficient to robustly train a supervised learning model. In this situation, pseudo-labeling can augment the original labeled dataset with pseudo-labeled data, potentially increasing the accuracy of the resulting outage detection model.

MixMatch 102 combines the concepts of data augmentation and “mixup” on both labeled and unlabeled data to generate new examples, and then trains a model on the mixed data using a combination of supervised and unsupervised loss functions, thereby improving the model's performance.

SSL presents a significant area of research potential within the field of PES, particularly when labeled data are scarce. One example is the classification of rare events that disrupt normal power system operations due to extreme weather events or cyber-events. In such scenarios, traditional supervised learning solutions may result in sub-optimal decision boundaries. However, SSL can overcome this limitation by incorporating both labeled and unlabeled data into the learning process to aid in such classification tasks. Specifically, the inclusion of unlabeled data allows the SSL algorithm to infer close-to-optimal decision boundaries, thereby improving classification accuracy. Previous studies have applied SSL techniques for fault diagnosis and event detection, 103,104 detection of cyber-attacks such as false data injection attacks, 105 electricity theft detection, 106 and nonintrusive load monitoring. 107,108 However, the full potential of SSL has yet to be explored in various distribution system applications. As an illustration of one such application, the detection of voltage harmonic distortions is chosen, although it should be noted that SSL can also be used for the detection of other power quality disturbances.

The proliferation of power electronics-based household appliances, or nonlinear loads, has resulted in an upsurge of voltage harmonic distortion within secondary distribution networks. 109 The identification of instances of single even or odd voltage harmonic distortion can be a time-consuming process. A potential solution to this problem is to utilize the semi-supervised approach for the classification of voltage harmonic distortion. The proposed approach involves conducting simulations on power systems with varying levels of nonlinear loads to generate voltage profiles, and subsequently, utilizing these profiles to construct a classifier through SSL. By leveraging the combination of labeled and unlabeled data, the classifier can achieve improved performance in identifying instances of single even or odd voltage harmonic distortion within a subset of the labeled data. It is important to note that a sufficient amount of representative data is needed to ensure the robustness of the classifier.

Remark 6. The distribution system examples presented in this section assume a particular learning paradigm. However, the formulation of certain tasks can be approached from alternative perspectives, thereby enabling the potential utilization of learning paradigms other than those previously demonstrated. For instance, supervised learning techniques may be utilized for phase identification tasks, 110 while unsupervised learning approaches may be employed for predictive maintenance. Additionally, SSL can be utilized for customer segmentation tasks, such as identifying household profiles, 111 as an alternative to unsupervised learning methods. Furthermore, many problems encountered in distribution systems exhibit a high degree of complexity, which can be mitigated by decomposing them into smaller sub-problems, akin to the divide-and-conquer algorithm approach. Subsequently, various learning paradigms can be utilized to address these sub-problems.

This section aims to identify application areas in power distribution systems that could greatly benefit from ML contributions in the future. The identified areas are organized into subsections according to the shared challenges that make them suitable for ML applications. Figure 9 provides an illustrative representation of these categories, aiding readers in comprehending the potential of ML in each application domain. Although this section does not intend to comprehensively survey all relevant ML applications in distribution systems, the insight presented herein offer valuable perspectives on how ML can improve the operational and planning practices of distribution systems, especially in cases where conventional tools are inadequate or inapplicable.

Selected ML applications according to the advantages they offer.

The following examples showcase the use of ML in tackling the task of constructing models pertaining to distribution system operations that are difficult to develop based on first principles alone. ML-based models can help predict system behavior and enhance system resiliency in response to evolving consumer and environmental factors, where traditional first-principles models are not available. Here, we can distinguish between data-driven models that leverage the vast amount of data to capture complex behaviors that are difficult to model and agent-based models, which allow for the representation of complex interactions between system components and agents.

1. Modeling consumer impacts on distribution systems

A highly promising area of ML applications in the power distribution system domain involves modeling consumer behavior for demand-response programs. For example, intelligent building control systems rely on accurate modeling of occupant behavior in buildings. However, the complexity of consumer behavior—driven primarily by human habits and weather—renders it challenging to model explicitly using mathematical equations. Likewise, modeling the impacts of household-owned batteries and EV charging on power distribution systems is also confronted with similar challenges. Mathematical equations alone may not fully capture the complex consumer behavior regarding energy consumption patterns, whereas ML techniques can excel in capturing such unknown complex relationships. More accurate models can, in turn, facilitate the development of coordination strategies for DERs to provide efficient demand-response. At present, research in this area is largely focused on reinforcement learning and deep reinforcement learning. Interested readers are directed to Refs. 112 and 113 for an in-depth review of occupant behavior modeling in buildings and EV charging management using reinforcement learning techniques, respectively.

Given the many successful examples of ML predicting consumer behavior in other application areas (e.g., recommender systems, targeted advertising), there is a powerful set of techniques available that could be extended and applied to understanding, predicting, and shaping consumer impacts on the distribution system. Nevertheless, leveraging ML techniques to model consumer behavior and/or impacts requires careful consideration of equity and fairness. Reference 114 provides a comprehensive overview for readers seeking a general understanding of the increasing concerns surrounding bias, fairness, and equity in ML-driven applications. There are numerous steps that can be taken to mitigate bias and ensure that ML systems are not only accurate but also fair and equitable. 114 This matter is particularly critical for equitable market predictions in future electricity markets with a plethora of diverse and small-scale participants (commonly known as prosumers), or when executing demand-response initiatives to guarantee an equitable and unbiased outcome for all parties involved.

2. Modeling climate impacts on distribution systems

While many of the detailed power system analyses rightfully focus on analyzing the power system in isolation, in some cases, the power system is dramatically impacted by external systems. Of particular interest is modeling and predicting the impact of natural disasters on the distribution system as such events become more common and more destructive due to the effects of climate change. ML is a promising approach to capture such complex interactions and take advantage of the rich and varied weather datasets to build predictive models relevant to the power system. 115 Wildfires are another example of a natural system that is bidirectionally coupled to the distribution system—wildfires can cause power outages and power lines can ignite wildfires—and becoming more prevalent due to climate change. Recent works have begun to address how wildfire risk should factor into power system operations. 116,117

The subsequent application areas of ML in power distribution systems underscore its capacity to speed up computations or simplify modeling and simulation. To avoid repetition, this section will not reiterate on other pertinent topics that were previously discussed, such as the utilization of ML to accelerate D-OPF by optimizing the candidate space or linearizing power flow equations.

1. Voltage control in power electronics-dominated distribution systems

Power distribution systems worldwide are witnessing an increasing adoption of power electronics in various forms, chiefly inverter-interfaced renewable generators (such as PV systems and wind turbines) and power electronics loads (such as power electronics-backed air conditioners). Accordingly, the traditional approach to voltage control in power electronics-dominated distribution networks is becoming increasingly challenged. 118 Instead, the active participation of distributed local power electronics will play a key role in various system-wide control mechanisms (e.g., voltage and frequency control), further increasing the complexity of these problems. Furthermore, conventional voltage regulation devices, such as transformers' on-load tap changers and capacitor banks, may not be adequate to respond to rapidly changing operating conditions due to their slow response times, which can range from hours to days. Moreover, traditional optimization-based methods for voltage control, rooted in the D-OPF framework, may experience slow computational speed and convergence difficulties.

Given these challenges, ML and data-driven techniques—in particular reinforcement learning and deep reinforcement learning—have emerged as a promising avenue for research that can help to (i) enable faster convergence of existing voltage control methods for real-time control or (ii) develop novel control methods adapted to the use of local power electronics controls. The basic concepts of the application of reinforcement learning to the voltage control problem were presented previously in Sec. IV C , Example 2, and are therefore omitted here for brevity. Reinforcement learning has proven effective in this area to deal with the complexity of the voltage control problem without requiring detailed system models (which are often not available) and in a distributed manner that can scale to large systems. Recent work in this area has focused on algorithms that not only learn good policies for nominal conditions, but also provide some guarantee that safety constraints will be met 96 or that the system will remain stable. 119–121 These works can serve as an example for other application areas in providing guarantees that will aid in the adoption of data-driven techniques (see Sec. VI B 4 ).

2. Adaptive relay protection for active distribution systems

Traditionally, protection relay schemes in distribution systems have been programmed for unidirectional power flows. 122,123 The most common is overcurrent protection, which is designed to operate when the short-circuit (fault) current exceeds some predefined threshold (i.e., I f ≥ I min ) ⁠ . However, the increasing displacement of synchronous generators with inverter-based resources may lead to the miscoordination of overcurrent protection schemes. 123 This is due to these resources' notably lower fault current contributions than those in traditionally operated distribution networks. In addition to potential miscoordination, reduced fault currents in power electronics-dominated distribution networks can make fault detection more difficult. Another concern is related to more frequent reverse power flows that can adversely affect normal network operations by causing relay protection to malfunction (i.e., unwanted tripping). This calls for new protection concepts that should gradually replace the functionality of traditional protection systems to ensure stable and safe distribution system operations during both normal and contingency operations. 123  

The idea of adaptive protection has emerged as a potential solution; 122 however, it is still in its infancy, as is the application of ML for protection. 124–126 Nevertheless, ML has the potential to play a crucial role in advancing the field of relay protection in distribution systems. Specifically, ML techniques can aid in developing adaptive relay protection schemes, making it an exciting future research direction. By utilizing look-ahead search techniques, such as Monte Carlo Tree Search, 127 the accuracy and speed of fault detection and isolation can be improved. Moreover, ML-based solutions can replace the conventional lookup tables used for protection relay settings, thereby enabling the establishment of more accurate and adaptive protection settings. However, implementing such solutions requires extensive offline analysis of all possible future outcomes, which can be prohibitively expensive. Despite this, the advantages that ML can bring to relay protection for active distribution systems make it a worthwhile topic for future research and development.

3. Dynamic modeling of distribution systems

Another potential application is in developing data-driven models that capture the aggregate dynamic behavior of an active distribution system for use in transmission level analysis (modeling both domains in detail is computationally prohibitive). Future research should focus on how ML-based dynamic models can be integrated seamlessly into existing dynamic simulations which are to-date centered around physics-based models.

The following application areas demonstrate how ML can be employed to alleviate issues related to missing, incomplete, or nonexistent data. In addition to the topics covered below, several other ML applications are relevant to this category, including synthetic data generation for creating pseudo load profiles used in DSSE (discussed in Sec. II C 3 ) and topology and phase identification (see Example 1 in Sec. IV B ).

1. Building registers of distributed PV for demand-side management

Advanced demand-side management can help mitigate new operational challenges facing system operators by allowing them to shape and modulate flexible loads optimally. However, systemic missing or out-of-date information (e.g., size and location) on installed solar PV panels on residential and commercial rooftops makes demand-side management difficult at scale. Some reasons for this lack of data include older installations made prior to regulations being enacted, lack of owners' awareness of permitting rules, unauthorized installations to avoid permit fees, and discrepancies between reported and installed configurations. 129 At the same time, registration of unregistered PVs can be highly challenging due to their large numbers in poorly visible secondary distribution networks.

Because accurate and up-to-date repositories of distributed PVs and other flexible loads are a prerequisite to effectively employing demand response, ML-based solutions are gaining research attention. In one line of research, ML has been proposed to classify and segment rooftop PVs using aerial and satellite imagery, thereby enabling the creation of more accurate registers (databases) according to the spatial scale of interest (e.g., street or city level). An overview of opportunities and challenges on this topic can be found in Ref. 130 . Other data-driven solutions proposed in the literature rely on smart meter measurements (time-series data, as opposed to images) to detect abnormal energy consumption behaviors, including unauthorized PV installations. An exemplary work in this category is Ref. 131 , where the corresponding solution is developed using a change-point detection algorithm.

2. Situational awareness for distribution system operations and planning

The traditional operations and “fit-and-forget” approach to distribution network planning (capacity hosting analysis) are becoming less than adequate with the widespread adoption of DERs. The necessary changes in operational and planning practices rely on increasing the grid's last-mile visibility, which has historically been difficult to achieve due to a lack of metering devices and insufficient communications infrastructure. 41,42 However, the massive rollout of smart meters has significantly increased the number of metering devices installed at the grid's edge. 132 In addition to smart meters, new untapped sensing potential comes from broadband cable television networks, with sensors deployed across the countries and their own high-speed, low-latency communications network. 133 Unlocking the potential of these sensors can enable greater insight into the operations of last-mile networks, thus allowing operators to react promptly.

Despite the influx of new data, missing measurements due to metering sensor errors or communication disruptions will become an inevitable challenge. However, leveraging the low-rank property of the streaming data matrix can allow for the utilization of low-rank matrix completion and tensor completion techniques, as well as certain ML methods, to recover these missing measurements. 134–136 Although data recovery is an important topic, the lack of visibility into distribution system operations currently presents more pressing challenges. For instance, outage detection in distribution systems remains a nontrivial task for operators who still rely on customer calls to identify power outages, sending crews to locate and eliminate the cause. To overcome this, fault location isolation and service restoration (FLISR) is a promising application in the advanced management of distribution systems, but it requires improved observability. To improve situational awareness at the distribution system level, the PES community has recognized the numerous opportunities that ML can offer, in addition to those discussed in Sec. II C 3 in the context of DSSE. For example, ML can be applied for outage detection, 137,138 fault location, 139 and service restoration, 140,141 among others. Readers interested in further information on ML for FLISR are encouraged to refer to Ref. 10 , which provides a comprehensive literature review on AI applications for distribution system operations.

ML methods and implementations must be characterized by well-posedness, reliability, and robustness in order to be accepted for practical application in distribution systems beyond the research community. Furthermore, their outcomes must be readily verifiable, validated, and reproducible. To attain this, several key challenges must be effectively addressed, including: (i) data availability, quality, nonstationarity, and privacy; (ii) robustness, scalability, and performance guarantees of selected models; (iii) adherence to physical principles and laws; (iv) explainability and visualization of obtained solutions (i.e., interpretability); and (v) a unified evaluation methodology for ML performance, encompassing well-defined metrics and structured baselines for comparison. Figure 10 outlines the identified research directions aimed at addressing these critical challenges through the lens of applied ML. The discussion in this section is centered on challenges and opportunities, emphasizing the critical research directions necessary for making ML a credible and trusted methodology in distribution systems. Nevertheless, we argue that the presented discussion can be generalized to other systems within the PES domain, including bulk power systems.

Main research themes for ML applications in power distribution systems.

The first research theme is centered around three main strategies: (i) leveraging domain knowledge (i.e., expert feedback) on data acquisition and model selection; (ii) incorporating physical principles and governing laws into ML models; and (iii) domain-aware decision-making or policy-making. Each of these strategies comprises multiple challenges that will be discussed in subsequent subsections VI A 1–VI A 4 .

1. Data acquisition

The centrality of data in ML highlights the fundamental importance of data adequacy for successful ML implementations. Data adequacy refers to data availability, quality, and nonstationarity that may significantly constrain ML applications in distribution systems. ML can provide novel and valuable perspectives on these data adequacy challenge, guiding data acquisition and ensuring the adequacy and quality of data for distribution system applications.

a. Data availability

ML relies heavily on a substantial amount of data—often hundreds or thousands of samples—to train its algorithms. The fastest-growing area of ML, deep learning, requires even more extensive data. In practice, the data used for learning ML models in distribution systems are often obtained through computational simulations, as observational data may be scarce. However, generating these data through simulations can be costly. Therefore, it is of interest to design data generation and collection procedures that minimize the amount of data required and its associated costs. Research is needed to identify the most relevant data to produce or gather to improve model results. Using unsupervised and semi-supervised methods with small samples can also prove to be difficult. Gaussian processes (GPs) 142 and their variants (e.g., sparse approximations 143 ) offer a potential solution in these cases. Another approach to overcome the scarcity of data is through data augmentation using GANs, which can synthesize additional training samples from the existing data to supplement the original sparse datasets. Nevertheless, maintaining diversity in the synthesized data and accurately representing out-of-sample or rare events are major ongoing challenges when using generative techniques.

b. Data quality

Data quality is vital to the efficiency of the learning process and encompasses various aspects, such as completeness, consistency in representation, informative nature, and trustworthiness, of the data. Real-world distribution system measurements and operational data are often plagued by issues such as missing values, noise, and nonuniform sampling rates, requiring extensive pre-processing for effective utilization. Additionally, the available data may not accurately reflect the overall distribution of measurements, leading to unrepresentative results. Therefore, ML methods must be able to accommodate the presence of noise, communication delays, and missing data in real-world scenarios. Furthermore, while ML models are often trained and tested on data obtained from high-fidelity simulators and accurate distribution system models, these simulators may contain modeling errors and simplifications themselves and may not always have access to the exact parameters and topology of the distribution network. Therefore, it is essential to incorporate domain knowledge to guide data acquisition and ensure its quality and adequacy.

c. Data drift

Data drift in the field of ML refers to changes in the distribution of data over time relative to the data distribution upon which the model was originally trained. Within the context of PES, data drift can result from changes in the topology of distribution networks, such as reconfigurations of loads and the addition of DERs, as well as transitions between grid-connected and islanded modes in microgrids. The current literature on this issue generally suggests retraining the ML model in response to each topological change, but this may not be a feasible solution for rapidly changing operational settings with a large number of network topology combinations. Further research is thus necessary to fully understand and address the challenges posed by data drift in power distribution systems.

d. Data privacy

In the PES community, there are growing concerns about the accessibility of unencrypted consumer electricity usage data collected by smart meters and other data sources. 144 The aforesaid concerns represent a significant obstacle to the safeguarding of privacy and, thus, call for the deployment of privacy-centric solutions that prioritize and preserve privacy. The adoption of cloud services is a viable option for addressing these challenges. For instance, software running on a distribution utility's server—collecting data from various end-users—may only transmit aggregated or anonymous statistics to the cloud, thereby allowing for data analysis without exposing individual consumer data. While this approach may benefit certain problems, its usefulness in demand-response applications where granular data are required to generate personalized incentives may be limited. In addition, a major hurdle lies in getting utilities to embrace cloud services as well as overcoming the inherent cybersecurity risks associated with this approach.

2. Model selection

In the PES field, applying new ML methodologies, particularly DNNs and GNNs, requires caution. DNNs and GNNs are rapidly evolving and growing research areas within ML, therefore new models are frequently developed and made available. In turn, PES research is often focused on applying these state-of-the-art neural networks, potentially at the expense of more pressing needs. For example, the reliance on large amounts of data—crucial requirement for deep learning—is frequently overlooked in distribution system research. This shift toward more data-intensive ML methods has also diverted attention away from traditional, less data-intensive ML methods that may be better suited for the problems at hand. To address these concerns, it is imperative to carefully consider the suitability of state-of-the-art ML methods for distribution systems and make informed decisions on their use. This requires a collaborative effort between researchers and practitioners from both disciplines, focusing on understanding the context and needs of the distribution system domain.

3. Hybrid and physics-informed solutions

In response to the increasing complexity of the distribution system, researchers have proposed a variety of ML-based alternatives to existing tools. As discussed in examples throughout this paper, these ML algorithms can effectively deal with complexity, reduce the computational burden of analysis, or leverage existing datasets, to name a few of the many potential benefits. At the same time, there exist decades of progress in PES developing traditional physics-based models and techniques that are governed by domain knowledge and theoretical foundations. In many cases, the best solution is not simply choosing one technique over the other, but rather combining the two in novel ways to achieve the benefits of each, while also limiting the drawbacks of each ( Fig. 11 ). This has numerous benefits, including reduced data requirements or enhanced performance; for example, models can be trained with fewer data points, or the models can converge faster to optimal solutions. 145  

Merits of hybrid models relative to purely data-driven or model-based solutions.

Incorporating physics into ML models can be achieved through the enforcement of soft and/or hard constraints during the training process, prediction time, or both. Hard constraints involve transforming an optimization problem into a constrained one, while soft constraints involve modifying the objective function (i.e., the loss function) by incorporating additional physics-based terms. 145 This results in the creation of physics-constrained neural networks and physics-informed neural networks, respectively, which are characterized by their increased capability for generalization and enhanced interpretability. 145 The interested reader is referred to Ref. 34 for a list of software libraries specifically designed for physics-informed ML. Further research is necessary to explore hybrid and physics-informed approaches to achieve better performance and explainability in various distribution system applications.

4. Operator in-the-loop

The core principles of ML assume minimal human intervention, which often contradicts traditional practices in distribution system management. Despite the growing trend toward digitization and automation in power distribution systems, the adoption of ML-based solutions in these systems requires expert input, particularly in the areas of operations, control, and planning. Thus, it is unrealistic to expect that ML alone can revolutionize the operations of power distribution systems. Instead, a more viable approach is to leverage ML-based solutions in conjunction with expert input to improve system operations by offloading certain tasks and allowing operators to focus on optimizing operations. Such co-occurrence of ML and human expertise would reduce the need for human intervention and specialized expertise.

The second research theme is centered around three main strategies: (i) explainable; (ii) understandable; and (iii) repeatable ML results. To gain a full understanding of these concepts in a general sense, it is suggested that the reader refers to Ref. 146 .

1. Interpretability

Traditionally, power engineers' confidence in the suitability of a model for a particular application is closely tied to their understanding of how the model works, its adherence to fundamental physical principles, and the degree to which its outputs facilitate effective decisions. However, this can be challenging with ML models, which are often hard to understand due to their lack of interpretability. To overcome this issue, research into methods that incorporate domain knowledge is important to make these models understandable and trustworthy. This ties into the research theme of physics-informed ML, which has been previously discussed.

Numerous ML frameworks used in distribution systems predominantly employ data-driven methodologies that produce black-box solutions with restricted information, making them difficult to explain. Although these techniques have exhibited good performance, their lack of interpretability renders them unfit for deployment in safety-critical systems. It is therefore essential to incorporate interpretability to establish trust in ML models and to effectively communicate their outcomes to system operators and decision-makers. However, as ML models become more complex—a trend also evident in PES research—they become harder to explain, let alone understand. This leads to a trade-off between complexity and interpretability. One approach is to quantify this trade-off and focus on developing interpretable solutions tailored to specific distribution system applications. This way, operators can understand and trust the results of the models. Despite the extensive body of literature on interpretable ML, 147,148 much progress still needs to be made in creating interpretable ML models tailored specifically to the PES domain.

2. Visualization

Visualization is a frequently overlooked, yet crucial aspect of decision-making for operators in control centers. Apart from facilitating their comprehension of system processes, it also enables them to swiftly identify critical states in the system and take prompt actions accordingly. When it comes to ML systems, it is imperative to take into account the unique needs and preferences of decision-makers and system operators concerning both the visualization of the learning processes and the resulting outcomes. To advance research in this area, the focus should be on developing advanced but intuitive explanatory approaches that incorporate domain knowledge. Such practices would aid operators in understanding and interacting with complex ML models. Therefore, the promising avenue for future research is the development of post-hoc visualization and explanatory techniques, which provide a way to visualize and explain the results of ML models after they have been trained rather than during the training process. Although showcasing the accuracy of a learning model during training is useful, it is even more useful to identify cases where the predictions were incorrect and focus on them. Another prospective area for future development is how to properly visualize the evaluation metrics employed for evaluating the performance of ML models, the topic discussed next.

3. Operator-friendly model evaluation

Researchers tend to prioritize finding the ML model that optimizes a chosen performance metric. However, solely relying on evaluation metrics to evaluate model performance can be misleading. For instance, when dealing with an unbalanced dataset wherein classes are not equally represented, using a performance metric like accuracy can provide a skewed perspective of the model's performance. For example, consider a classification task that aims to detect three-phase faults—the rarest variety—in power distribution systems. If the dataset used for this task contains 100 000 fault currents, with only 100 of them being the faults of interest and the rest being single- and two-phase faults, a trivial algorithm that classifies all faults as single- and two-phase faults (i.e., “non-three-phase faults”) would still have an accuracy of 99.9%. This could lead to the incorrect conclusion that the classifier is highly accurate, despite missing the faults of interest. Therefore, it is crucial to rigorously quantify the performance of learning systems, including the estimation of prediction quality and effective confidence bounds, to increase the reliability and credibility of ML in distribution systems. Additionally, efforts must be made to make performance measures understandable to operators by aligning with their expectations and translating them into more user-friendly metrics.

4. Performance guarantees

For certain high-regret applications in power systems (e.g., determining system stability), the costs associated with failure are so large that even a very reliable model will not be trusted by operators without performance guarantees. Therefore, an important area of future research will be to augment ML approaches with performance guarantees or adopt hybrid approaches which combine ML with traditional analysis that can provide guarantees. For example, in Refs. 120 and 121 , techniques from robust control theory and Lyapunov stability theory, respectively, are combined with reinforcement-learning controllers to achieve stability guarantees. In the supervised learning domain, some efforts have been made to provide worst-case performance guarantees when learning OPF solutions. 149 Despite these promising examples, there remain many proposed applications of ML in power systems that would benefit from work in this area. By incorporating guarantees into ML techniques researchers can help remove a major barrier to adoption. In some cases, when strict guarantees are not possible, probabilistic ML methods can help to accurately convey the uncertainty inherent in the model (see Sec. VI C ).

Finally, it is worth noting that although ML robustness could be a subsection of its own, we cover it briefly here, as the notion of ML model robustness is closely related to this topic. Namely, many data-driven models assume that the inputs to the model are unaltered; however, those designed for power system applications may be susceptible to various perturbations, both adversarial and nonadversarial. 150 Therefore, another important area of research will be analyzing and quantifying the robustness of the ML-based solutions against altered data instances. A relevant work on this particularly important topic is Ref. 150 .

5. Benchmarks

Many of the most successful applications of ML, such as image classification, speech recognition, and natural language processing, have consistently been improved on by outperforming prior works on well-established benchmarks (available at the UCI repository of ML databases 151 ). In contrast, the lack of a unified and well-established set of baselines or benchmarks in the existing PES literature presents a challenge for accurately comparing and evaluating the performance of ML models and reproducing reported results. This results in a fragmented approach to model evaluation, with individual works often presenting specific quality tradeoffs without a direct comparison to prior works. To improve the characterization of methodological performance and enable a fair comparison, it is essential to establish a systematic and standardized framework for defining datasets and benchmarks for each ML application area. This will provide a more robust foundation for evaluating the performance of models and facilitate meaningful comparisons. Moreover, there is a need for research to develop quantitative measures that are both meaningful and interpretable, or to replace existing measures with qualitative characteristics that are better suited for the intended audience.

The last research theme focuses on probabilistic ML and includes two primary strategies: (i) probabilistic ML modeling and (ii) translating predictions with uncertainty estimates into actionable decisions. In contrast to traditional ML methods that typically provide deterministic predictions, probabilistic ML models generate predictions in the form of prediction intervals or probability distributions, encompassing not only a single-point estimate but also a measure of uncertainty. To gain a better understanding of probabilistic ML, refer to Ref. 152 for a general discussion.

Regarding (i), the use of ML in distribution system applications necessitates incorporating probabilistic modeling to account for three intrinsic types of uncertainty. These uncertainties include those arising from the physics involved (see Sec. VI C 1 for examples), the data used to train ML models, and the model itself. The second type of uncertainty typically involves both aleatoric uncertainty resulting from noise in data and epistemic uncertainty resulting from limited data and knowledge, as discussed in Ref. 34 .

1. Probabilistic forecasting and estimation

Modern power distribution systems must be better equipped to handle uncertainty from various sources (such as measurement errors, load fluctuations, weather-dependent renewable generation, and network parameters uncertainty) to guarantee secure operations. One of the key ways to achieve this is through reliable forecasting. Historically, energy forecasting, including demand, wind, and solar, relied on deterministic or single-valued forecasting methods. 153 However, probabilistic energy forecasting, which provides a range of probable values in addition to single-point estimates, is now considered more suitable for power distribution systems and is expected to become a fundamental aspect of their effective planning and operations. Despite the growing popularity of probabilistic ML techniques, particularly in the field of probabilistic forecasting (see, e.g., Ref. 154 ), the majority of forecasting models in practice still rely on deterministic modeling approaches. The transition from deterministic to probabilistic forecasting in power distribution systems has yet to occur, making it crucial to develop best-practice guidelines for their design and standardize evaluation metrics that are suitable for probabilistic ML models at this early stage. In addition to the emphasis on probabilistic forecasting, there is also a significant focus on developing probabilistic estimation methods, such as probabilistic SE using Bayesian networks, 155,156 which offer many advantages over traditional deterministic SE methods.

2. Uncertainty quantification

In the face of growing uncertainty in power distribution systems, relying solely on deterministic forecasting and point estimation is not sufficient for reliable decision-making and operational risk assessment. Additionally, the predictions made by ML models are often subject to a high degree of uncertainty. To enhance the credibility and reliability of ML-based solutions, it is crucial to incorporate uncertainty quantification and provide quantifiable confidence measures for decision-makers to utilize. This will allow them to make more informed decisions based on a better understanding of the uncertainty involved. This important issue has received much attention in recent years, supplementing traditional analytical methods with stochastic formulations to deal with uncertainty. The most common techniques for quantifying uncertainty in PES research are Monte Carlo simulations, 157 Bayesian inference, quantile regression, and polynomial chaos expansion, 158 with methods such as Bayesian deep learning, 159 Monte Carlo dropout, 160 (deep) GPs, 142,161 gaining in popularity. Future research in this area should focus on developing statistical learning methods that allow for the requisite uncertainty quantification in ML predictions that are specifically tailored to distribution system applications. This also entails characterizing and quantifying all sources of uncertainty during model development and determining the factors that contribute most to uncertainty. Providing insight into sources of uncertainty, let alone quantifying their contributions, can significantly increase the chances of making ML systems more accessible to systems operators and decision-makers.

3. Decision-making under uncertainty

In light of the prior discussion, it is evident that deterministic ML is not best suited for decision-making under uncertainty. Instead, probabilistic ML is more appropriate and can enhance the trustworthiness and robustness of ML model predictions in power distribution systems. The challenge, however, is translating the outputs of ML models (i.e., predictive uncertainty estimates) into effective decisions. 162 Currently, the use of ML methods for decision-making under uncertainty is extensively researched within the PES domain. A prime example of this is cost-oriented ML, which endeavors to improve the practical value of probabilistic forecasting in power system decision-making procedures. 163 Similarly, future research should prioritize evaluating the value of probabilistic ML models in the context of decision-making processes within distribution systems rather than just examining the accuracy and precision of their outputs. Another important area for future research is exploring ways to seamlessly integrate probabilistic forecasts and uncertainty estimates into existing decision-making practices. One potential approach is to utilize the upper and lower bounds of uncertainty estimates as inputs into planning frameworks, considering them as the most favorable and unfavorable scenarios. 164 Nevertheless, there is still ample opportunity for improvement in incorporating the full uncertainty estimates into decision-making processes rather than just the distribution tails. Effective visualization that is specifically tailored for distribution system control centers is important in this context, and it relates to the topic of visualization discussed earlier.

How can ML models' interpretability, robustness, and performance guarantees be enhanced so that they can be utilized confidently in complex distribution systems?

How to effectively present the results of ML models to improve their understanding by system operators and decision-makers?

What steps need to be taken to make the performance evaluation of complex ML models more accessible to system operators?

What measures should be taken to validate ML findings and translate them into actionable insight for decision-makers?

What methods should be employed to combine actual distribution system operational data with simulated data to enhance sparse datasets while ensuring data quality standards are met?

How can the uncertainty in ML model predictions be converted into quantifiable confidence measures for system operators and decision-makers?

A comprehensive list of assumptions, parameters, and algorithmic choices used;

A complete description of the raw input data and data processing steps utilized;

A complete description of the model implementation to enable independent replication;

Verification and validation of the model implementation to ensure accuracy and reliability.

In addition to the information above, careful consideration should be given to standardizing model performance evaluation across benchmark datasets and problems and developing appropriate benchmarks in novel application areas if they do not already exist, thereby encouraging model-consistent testing. Additionally, a concerted effort by researchers will be needed to devise meaningful metrics that test the accuracy of newly proposed ML-based solutions, especially if the assessment of model fidelity at rare events (i.e., tails of distributions) is also factored into such evaluations. Last, but not least, by including the code used to generate results in conjunction with papers, researchers can make the methods more accessible (and comparable) to the community to encourage further improvements.

In this paper, we summarize the current state of ML for power distribution systems in a form that is accessible to readers from different engineering backgrounds in order to encourage interdisciplinary research on this topic. While a vast body of literature has focused on reviewing novel ML techniques and their applications in PES, our attention is directed toward power distribution systems. These systems are at the forefront of the ongoing electric power sector's transformation and come with distinctive data-related challenges related to ML applications. In addition to providing background material on power systems—with a specific emphasis on distribution systems—and relevant ML theory, we also explore the current capabilities and limitations of ML for power distribution systems problems. We highlight various ML applications, ranging from distribution system operations and control to planning. We finally discuss potential research directions that can facilitate the adoption of ML beyond the research community and academic circles, and promote its implementation in real-world applications. One of the key takeaways from our discourse is that the current use of new data-intensive methods—widely explored in PES research—is not practical for power distribution systems at this stage due to the limited availability of data in distribution systems planning, and in particular, operations. Consequently, until data availability increases and the data-related issues discussed in this paper are addressed, it is advisable to redirect research efforts toward ML approaches that require less data or enable trustworthy augmentation of existing limited data. In this regard, physics-based ML appears to offer viable and promising research directions to pursue.

This work was authored in part by the National Renewable Energy Laboratory (NREL), operated by the Alliance for Sustainable Energy, LLC, for the U.S. Department of Energy (DOE) under Contract No. DE-AC36–08GO28308. The views expressed in the article do not necessarily represent the views of the DOE or the U.S. Government. The U.S. Government retains and the publisher, by accepting the article for publication, acknowledges that the U.S. Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this work, or allow others to do so, for U.S. Government purposes. This work was also partially funded by the Climate Change AI Innovation Grants program, hosted by Climate Change AI with the support of the Quadrature Climate Foundation, Schmidt Futures, and the Canada Hub of Future Earth.

The authors have no conflicts to disclose.

Marija Markovic: Conceptualization (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (equal). Matthew Bossart: Conceptualization (supporting); Visualization (supporting); Writing – original draft (supporting); Writing – review & editing (equal). Bri-Mathias Hodge: Supervision (lead); Writing – original draft (supporting); Writing – review & editing (equal).

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

Citing articles via

Submit your article.

Sign up for alerts

• Online ISSN 1941-7012
• For Researchers
• For Librarians
• For Advertisers
• Our Publishing Partners  
• Physics Today
• Conference Proceedings
• Special Topics

pubs.aip.org

• Privacy Policy
• Terms of Use

Connect with AIP Publishing

This feature is available to subscribers only.

Sign In or Create an Account

• © 2022

Recent Advances in Power Systems

Select Proceedings of EPREC-2021

• Om Hari Gupta 0 ,
• Vijay Kumar Sood 1 ,
• Om P. Malik 2

Department of Electrical Engineering, National Institute of Technology Jamshedpur, Jamshedpur, India

You can also search for this editor in PubMed   Google Scholar

Faculty of Engineering and Applied Science, Ontario Tech University, Oshawa, Canada

Department of electrical and computer engineering, university of calgary, calgary, canada.

• Presents selected proceedings of Electric Power and Renewable Energy Conference (EPREC) 2020
• Covers recent developments in the emerging areas of power electronics
• Will be useful for professionals interested in advancements in power system

Part of the book series: Lecture Notes in Electrical Engineering (LNEE, volume 812)

27k Accesses

48 Citations

3 Altmetric

• Table of contents

About this book

Editors and affiliations, about the editors, bibliographic information.

• Publish with us

Buying options

• Available as EPUB and PDF
• Read on any device
• Instant download
• Own it forever
• Compact, lightweight edition
• Dispatched in 3 to 5 business days
• Free shipping worldwide - see info
• Durable hardcover edition

Tax calculation will be finalised at checkout

Other ways to access

This is a preview of subscription content, log in via an institution to check for access.

Table of contents (53 chapters)

Front matter, electrifying hilly remote habitation by solar photovoltaic system.

• Chandramohan, Siva Ramakrishna Madeti, Krishna Murari

Multi-objective Weighted-Sum Optimization for Stability of Dual-Area Power System Using Water Cycle Algorithm

• CH. Naga Sai Kalyan, Chintalapudi V. Suresh, U. Ramaniah

Comparison Between Novel Fault Detection Techniques in Solar PV Arrays: A Review

• Khushboo Verma, Suresh K. Gawre, Shailendra Kumar

Change in Negative Sequence Current-Based Islanding Event Detection

• Shubham Kumar, Munna Kumar, Jitendra Kumar

An Assessment Procedure for Distribution Network Reliability Considering Load Growth

• Umesh Agarwal, Naveen Jain, Manoj Kumawat, Sri Niwas Singh

Integration of Solar PV System with Storage Battery System

• Ch. Sekhar, Ajay Raja Sinha, Saibal Manna, A. K. Akella

Analyzing the Impact of Integration of SSSC/TCPS and Wind Turbine in 2-Area Interconnected Power System Networks

• Mukul Dixit, Anshul Dixit

A Review on the Micro-Phasor Measurement Unit in Distribution Networks

• Bugatha Ram Vara Prasad, Brundavanam Sesha Sai, Joddumahanthi Vijaychandra, Rohit Babu

Different Optimizers-Based Gated Recurrent Unit Network to Forecast One Step Ahead Solar Irradiance

• Pardeep Singla, Manoj Duhan, Sumit Saroha

Single Step-Ahead Solar Irradiation Forecasting Based on Empirical Mode Decomposition with Back Propagation Neural Network

• Anuj Gupta, Kapil Gupta, Sumit Saroha

PV Fed Solar Pump Designing for Fish Cultivation

• Vrishank Tiwari, Suman Kumari, Priyanka Priyadarshini Sahoo

Online Transient Stability Assessment Using Regression Models

• P. K. Chandrashekhar, S. G. Srivani

Congestion Management Using FACTS Devices: A Review with Case Study

• Ashish Singh, Aashish Kumar Bohre

Power Consumption Control at Coal Mining Sites

• V. Petrov, A. Sadridinov, A. Pichuev

An Assessment of Indian Smart Grid Pilot for Selection of Best-Fit Communication Technology

• Jignesh Bhatt, Omkar Jani, V. S. K. V. Harish

Single-Sided Bidding Considering Uncertainty of Renewable Source in a Day-Ahead Energy Market

• Abhilipsa Sahoo, Prakash Kumar Hota

Profit Maximization of Utility by Incorporating Demand Response

• G. S. Sivasankari, S. Prasanthini, M. Hamsa Deepika, K. Narayanan, T. Vigneysh, Tomonobu Senjyu

Profit Maximization in a Hybrid Microgrid Incorporating Demand Response

• S. Prasanthini, G. S. Sivasankari, R. Subhasri, K. Narayanan, Velamuri Suresh, Gulshan Sharma

Demand Response Integrated Energy Management Technique For Grid-Connected Microgrid

• R Subhasri, M Hamsa Deepika, S Prasanthini, Suresh Velamuri, T Vigneysh, K Narayanan

This book contains selected proceedings of EPREC-2021 with a focus on power systems. The book includes original research and case studies that present recent developments in power systems, principally renewable energy conversion systems, distributed generations, microgrids, smart grid, HVDC & FACTS, power quality, power system protection, etc. The book will be a valuable reference guide for beginners, researchers, and professionals interested in advancements in power systems.

• Power system
• Power system Protection
• Power System Operation and Control
• Deregulation
• Power System Restructuring
• Distributed generation
• Renewable energy conversion systems
• Power Quality
• HVDC & FACTS

Om Hari Gupta

Vijay Kumar Sood

Om P. Malik

Dr. Om Hari Gupta is currently an Assistant Professor at the Department of Electrical Engineering, National Institute of Technology Jamshedpur, India. He received his B.Tech (Electrical & Electronics Engg.) from UP Technical University, Lucknow, M.Tech from the MN National Institute of Technology Allahabad, Prayagraj, and the Ph.D. degree in electrical engineering from the Indian Institute of Technology Roorkee, Uttarakhand, India. He has visited the Ontario Tech University (formerly University of Ontario Institute of Technology), Oshawa, ON, Canada, for research on microgrid operation. His major areas of research interests include power system compensation and protection, microgrid control and protection, and control of drives. Dr. Gupta is a Senior Member of IEEE and a recipient of the Queen Elizabeth-II scholarship for pursuing research on microgrid operation at UOIT, Canada in 2017. Dr. Gupta has more than 45 publications including Journals, Books, Book Chapters, and ConferencePapers. He is a reviewer for various International Journals including IEEE Transactions on Power Delivery, Electric Power Components and Systems, IET Generation, Transmission and Distribution, International Journal of Electrical Power and Energy Systems, etc.

Dr. Vijay Kumar Sood is an Associate Professor in the Electrical Engineering Department, Ontario Tech University (formerly University of Ontario Institute of Technology), Oshawa, ON, Canada, where he joined in 2007. Dr. Sood was a senior researcher at the Research Institute of Hydro-Québec, Montreal, Canada. He received a Ph.D. degree from the University of Bradford, U.K., in 1977. He authored over 150 articles and 02 books on HVDC and FACTS transmission systems. He has also been the Editor of the IEEE Transactions on Power Delivery, Associate Editor of IEEE Canadian Journal of Electrical and Computer Engineering, and IEEE Canadian Review quarterly magazine. His current research interests include the monitoring, control, andprotection of power systems. Dr. Sood is a life fellow of the Institute of Electrical and Electronics Engineers (IEEE), a fellow of the Engineering Institute of Canada and Emeritus, and a fellow of the Canadian Academy of Engineering.

Dr. Om P. Malik graduated in electrical engineering from Delhi Polytechnic in 1952 and obtained a master’s degree in electrical machine design from Roorkee University in 1962. He received a Ph.D. degree in electrical engineering from the University of London and a D.I.C. from the Imperial College (now University) of Science and Technology, London, in 1965. Dr. Malik worked in electric utilities in India from 1952 to 1961 and spent one year, 1959–60, as a Confederation of British Industries Senior Scholar with manufacturing companies in the U.K. He taught at the University of Windsor, Ontario, Canada, from 1966-68 and joined the University of Calgary, Alberta, Canada, in 1968. He was acting as a dean of the Faculty of Engineering in 1981, associate dean (Academic) from July 1979 to December 1990, and associate dean (Student Affairs) from July 1995 to December 1998. Presently, he is a professor emeritus in the Department of Electrical and Computer Engineering, University of Calgary. Prof. Malik is a life fellow of the Institute of Electrical and Electronics Engineers (IEEE).

Book Title : Recent Advances in Power Systems

Book Subtitle : Select Proceedings of EPREC-2021

Editors : Om Hari Gupta, Vijay Kumar Sood, Om P. Malik

Series Title : Lecture Notes in Electrical Engineering

DOI : https://doi.org/10.1007/978-981-16-6970-5

Publisher : Springer Singapore

eBook Packages : Energy , Energy (R0)

Copyright Information : The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022

Hardcover ISBN : 978-981-16-6969-9 Published: 15 February 2022

Softcover ISBN : 978-981-16-6972-9 Published: 16 February 2023

eBook ISBN : 978-981-16-6970-5 Published: 14 February 2022

Series ISSN : 1876-1100

Series E-ISSN : 1876-1119

Edition Number : 1

Number of Pages : XII, 714

Number of Illustrations : 120 b/w illustrations, 279 illustrations in colour

Topics : Energy Systems , Power Electronics, Electrical Machines and Networks , Energy Policy, Economics and Management

Policies and ethics

• Find a journal
• Track your research

Research Program

PSERC’s comprehensive research program includes 3 stems to address the needs of the modern electric energy infrastructure . They are:

• Power Markets Markets research focuses on market design, analysis, and mechanisms within the context of the electric power system. Current research topics include implications of greenhouse gas policy options for power systems, coupling wind generation with controllable load and storage, demand response options for integrating renewable energy, and interactions of energy and environmental policies in a transmission constrained electric power market.
• Power Systems Systems research seeks ways to increase use, efficiency and reliability of increasingly complex and dynamic power systems. Current research topics are transmission corridors to support renewable energy resources, integration of storage devices with renewable, special protection schemes, visualization using PMU data, synchrophasor application framework, on-line security assessment, and model and data interoperability in a smart grid.
• Transmission and Distribution Technologies This research improves performance of T&D systems by finding new applications for innovative technologies. Current research projects include new substation designs, communication requirements for a smart grid, PHEV’s for energy storage, smart grid implications for distribution engineering, interoperability of PMU’s and PMU-enabled intelligent electronic devices, and micro and nano dielectrics for utility applications.

Leveraged Research

Through collaboration, PSERC conducts research that leverages the industrial support it receives from its members. Here are some examples.

• 2016:  Funded by the U.S. Department of Energy (DOE) “The Future Grid to Enable Sustainable Energy Systems” project focused on how to integrate higher penetrations of renewable generation and other future technologies into the grid while enhancing grid stability, reliability, and efficiency. It also aims to stimulate discussion among the academic, industry, and government communities on what it will take to shape the future grid for the mid-twenty-first century. For more information see the Future Grid Final Report
• 2009-2015: The Consortium for Electric Reliability Technology Solutions ( CERTS ), was formed in 1999 to research, develop and commercialize new methods, tools and technologies to protect and enhance the reliability of the U.S. electric power system. CERTS conducted research supported by the U.S. Department of Energy’s Office of Electric Delivery and Energy Independence, and by the California Energy Commission’s Public Interest Energy Research Program. PSERC faculty worked with researchers at Lawrence Berkeley National Laboratory, Oak Ridge National Laboratory, Pacific Northwest National Laboratory, Sandia National Laboratories and several energy businesses. For more information see the Phase I report (2009-2013) and the Phase II report (2014-2015)

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals

Electrical and electronic engineering articles from across Nature Portfolio

Electrical and electronic engineering is the branch of engineering that makes use of electricity. Electrical engineering concentrates on systems for generating and transmitting large electrical currents and converting them into other forms of energy, such as mechanical motion. Electronic engineering focusses on lower energy currents for processing and communicating information.

Wireless radiofrequency network of distributed microsensors

Distributed sensing of a dynamic environment is typically characterized by the sparsity of events, such as neuronal firing in the brain. Using the brain as inspiration, an event-driven communication strategy is developed that enables the efficient transmission, accurate retrieval and interpretation of sparse events across a network of thousands of wireless microsensors.

Latest Research and Reviews

Correlative, ML-based and non-destructive 3D-analysis of intergranular fatigue cracking in SAC305-Bi solder balls

• Charlotte Cui
• Fereshteh Falah Chamasemani
• Roland Brunner

Electron beam lithography on nonplanar and irregular surfaces

• Huseyin Ekinci

Urban methane emission monitoring across North America using TROPOMI data: an analytical inversion approach

• Mohammadali Hemati
• Masoud Mahdianpari
• Fariba Mohammadimanesh

Application of optimized Kalman filtering in target tracking based on improved Gray Wolf algorithm

• Zheming Pang

The principle of magnetic flux switch

• Amir Heidary
• Mohammad Ghaffarian Niasar
• Marjan Popov

Sand cat swarm optimization algorithm and its application integrating elite decentralization and crossbar strategy

News and Comment

Scaling up stretchable electronics

• Katharina Zeissler

Searching for MXenes

• Matthew Parker

Hybrid perovskites unlocking the development of light-emitting solar cells

Light-emitting perovskite solar cells are emerging optoelectronic devices that integrate light-emitting and electricity-generating functions in one device. This type of device unlocks new possibilities for applications as outdoor light sources, in multifunctional architecture, smart automobiles, self-powered displays and portable power floodlights.

• Alexey Tarasov

Tailoring materials, structures and fabrication processes for stretchable electronics

An article in Nature presents large-area, high-performance intrinsically stretchable electronics thanks to innovation in materials selection, fabrication processes, device engineering and circuit design.

• Silvia Conti

Van der Waals magnet integration for energy-efficient spintronics

An article in Science Advances reports an integrated van der Waals system that enables field-free electric control of the magnetization of Fe 3 GaTe 2 above room temperature.

After confronting one uncanny valley, another awaits

The ‘uncanny valley’ has guided robot engineers on the limits of human likeness, yielding design principles to mitigate the risk of creepy robots. Yet unease with advancements in AI has exposed a new ‘uncanny valley of mind’, with researchers now exploring acceptable boundaries on simulating human intelligence, emotion, empathy and creativity.

• Jan-Philipp Stein
• Karl F. MacDorman

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Power Systems Research Topics Ideas for MS/PhD

List of power systems research topics ideas for ms/phd thesis.

1. Introduction and literature review of power system challenges and issues 2. Power system inertia estimation: Review of methods and the impacts of converter-interfaced generations 3. Artificial intelligence models in power system analysis 4. Power system protection 5. Technical barriers for harnessing the green hydrogen: A power system perspective 6. Evolutionary game analysis on behavior strategies of multiple stakeholders in maritime shore power system 7. Optimal analysis of a hybrid renewable power system for a remote island 8. Evaluation and optimization of off-grid and on-grid photovoltaic power system for typical household electrification 9. Comprehensive performance evaluation and demands’ sensitivity analysis of different optimum sizing strategies for a combined cooling, heating, and power system 10. Comprehensive weight method based on game theory for identify critical transmission lines in power system

11. Integrated standalone hybrid solar PV, fuel cell and diesel generator power system for battery or super capacitor storage systems in Khor fakkan, United Arab Emirates 12. Power system stability in the transition to a low carbon grid: A techno-economic perspective on challenges and opportunities 13. Multi-objective jellyfish search optimizer for efficient power system operation based on multi-dimensional OPF framework 14. Power system hybrid dynamic economic emission dispatch with wind energy based on improved sailfish algorithm 15. Voltage stability analysis of power system 16. Coordinated automatic generation control of interconnected power system with imitation guided exploration multi-agent deep reinforcement learning 17. Load frequency control based on the bees algorithm for the great britain power system 18. Multi-level interactive unit commitment of regional power system 19. Maiden application of fuzzy-2DOFTID controller in unified voltage-frequency control of power system 20. Network‐constrained rail transportation and power system scheduling with mobile battery energy storage under a multi‐objective two‐stage stochastic programming

21. Fuel-cell sharing for a distributed hybrid power system 22. A sustainable energy distribution configuration for microgrids integrated to the national grid using back-to-back converters in a renewable power system 23. An optimization-based strategy for solving optimal power flow problems in a power system integrated with stochastic solar and wind power energy 24. Impact of fossil-free decentralized heating on northern European renewable energy deployment and the power system 25. Investigation and damping of electromechanical oscillations for grid integrated micro grid by a novel coordinated governor-fractional power system stabilizer 26. Improving the Accuracy of the State Estimation Algorithm in the Power System Based on the Location of PMUs and Voltage Angle Relationships 27. The near-optimal feasible space of a renewable power system model 28. Multi-objective optimization of a combined cooling, heating, and power system with subcooled compressed air energy storage considering off-design characteristics 29. Near-Optimal PI Controllers of STATCOM for Efficient Hybrid Renewable Power System 30. ‘Cooperative game’inspired approach for multi-area power system security management taking advantage of grid flexibilities

31. Exergo-economic assessment and sensitivity analysis of a solar-driven combined cooling, heating and power system with organic Rankine cycle and absorption heat … 32. Equilibrium optimizer‐tuned cascade fractional‐order 3DOF‐PID controller in load frequency control of power system having renewable energy resource integrated 33. Analytical model of power system hardening planning for long-term risk reduction 34. An Optimal Control of Integrated Hybrid Power System with FACTS Devices Using Student Psychology‐Based Optimization Algorithm 35. Robust delay-dependent load frequency control of wind power system based on a novel reconstructed model 36. Power system harmonics 37. Increasing Turkey’s power system flexibility for grid integration of 50% renewable energy share 38. Preliminary propulsion and power system design of a tandem-wing long-range eVTOL aircraft 39. Efficient creation of datasets for data-driven power system applications 40. MVO algorithm based LFC design of a six-area hybrid diverse power system integrating IPFC and RFB

41. Optimal performance of a combined heat-power system with a proton exchange membrane fuel cell using a developed marine predators algorithm 42. A modified moth flame optimisation technique tuned adaptive fuzzy logic PID controller for frequency regulation of an autonomous power system 43. A techno-economic viability analysis of the two-axis tracking grid-connected photovoltaic power system for 25 selected coastal mediterranean cities 44. Decentralized data-enabled predictive control for power system oscillation damping 45. Optimal Design of Biomass Combined Heat and Power System Using Fuzzy Multi-Objective Optimisation: Considering System Flexibility, Reliability, and Cost 46. Design Methodology and Parametric Design Study of the On-Board Electrical Power System for Hybrid Electric Aircraft Propulsion 47. A novel sine augmented scaled sine cosine algorithm for frequency control issues of a hybrid distributed two-area power system 48. Sunflower optimization based fractional order fuzzy PID controller for frequency regulation of solar-wind integrated power system with hydrogen aqua equalizer-fuel … 49. Optimal scheduling for power system peak load regulation considering short-time startup and shutdown operations of thermal power unit 50. Voltage and frequency control of wind–diesel power system through adaptive sliding mode control of superconducting magnetic energy storage

51. Impedance-Based Stability Analysis Methods for DC Distribution Power System With Multivoltage Levels 52. Uncertainty Evaluation Algorithm in Power System Dynamic Analysis with Correlated Renewable Energy Sources 53. Which ship-integrated power system enterprises are more competitive from the perspective of patent? 54. Deep Feedback Learning Based Predictive Control for Power System Undervoltage Load Shedding 55. Conventional and advanced exergy analyses of a vehicular proton exchange membrane fuel cell power system 56. Advanced teaching method for learning power system operation based on load flow simulations 57. A novel application of ALO-based fractional order fuzzy PID controller for AGC of power system with diverse sources of generation 58. A review of the smart grid concept for electrical power system 59. Dynamic power–frequency control in a hybrid wind-PV plant interlinked with AC power system 60. Application of SVC and STATCOM for wind integrated power system

61. Mass optimization of the radiation shadow shield for space nuclear power system 62. Power system operation with power electronic inverter–dominated microgrids 63. Comparative study of four MPPT for a wind power system 64. An empirical approach to frequency droop characterization from utility‐scale photovoltaic plants operation in a power system 65. Bio‐inspired hybrid BFOA‐PSO algorithm‐based reactive power controller in a standalone wind‐diesel power system 66. A Small Hybrid Power System of Photovoltaic Cell and Sodium Borohydride Hydrolysis-Based Fuel Cell 67. A hybrid MFO‐GHNN tuned self‐adaptive FOPID controller for ALFC of renewable energy integrated hybrid power system 68. Nonlinear Model Predictive Control of an Autonomous Power System Based on Hydrocarbon Reforming and High Temperature Fuel Cell 69. A new congestion cost allocation method in a deregulated power system using weighting of contractual preferences and geographical location of users 70. Optimized Power System Voltage Measurements Considering Power System Harmonic Effects

71. Power System Characteristics 72. Fault Tolerant Control Method of Power System of Tram Based on PLC. 73. Priority Lists for Power System Investments: Locating Phasor Measurement Units 74. The concept of vulnerability and resilience in electric power systems 75. A review of different optimisation techniques for solving single and multi-objective optimisation problem in power system and mostly unit commitment problem 76. Modeling Inverters with Grid Support Functions for Power System Dynamics Studies 77. Performance of wide-area power system stabilizers during major system upsets: investigation and proposal of solutions 78. Review of Proactive Operational Measures for the Distribution Power System Resilience Enhancement Against Hurricane Events. 79. Chaos Elimination in Power System Using Synergetic Control Theory 80. Role of Hydrogen in a Low-Carbon Electric Power System: A Case Study

a. Study of Assessing the Stability of Rwanda’s Power System from Big Data Based on Power Generation 81. A simplified virtual power system lab for distance learning and ABET accredited education systems 82. Review of risk analysis methods for failure scenario in power system under typhoon disasters 83. Preliminary Study on Forced Oscillation of Power System with Quadratic Nonlinearity 84. A novel power system operation simulation method considering multi-state models of coal-fired power unit: a case study of coal-fired power transformation in a certain … 85. Comprehensive Analysis and Visualization Method of Online and Offline Operation Modes of Power System 86. Impact of the Power System Stabilizer on Transient Stability of the Power System 87. Research on Data Intelligent Retrieval Method Oriented to Unified Business Center of Power System 88. Technologies and economics of electric energy storages in power systems: Review and perspective 89. How to Use LDA Model to Analyze Patent Information? Taking Ships Integrated Power System as an Example

90. Satin Bowerbird optimization algorithm for the Application of Optimal power flow of power system with FACTS devices 91. Applications of Big Data and Internet of Things in Power System 92. Power system planning methods and experiences in the energy transition framework 93. Study of Interconnections in Vietnam Power System with Asynchronous Back to Back HVDC Links 94. A Convolution Neural Network Method for Power System Oscillation Type Identification 95. DE-Assisted LFC of Three-Area Six-Source Interconnected Power System with Wind Model and Fuel Cell Under Restructured Environment 96. Optimization for Short-Term Operation of Hybrid Hydro-PV Power System Based on NSGA-II 97. Deep learning for short-term voltage stability assessment of power systems 98. Transient Stability-Based Security State Classification of Power System Networks Using Kohonen’s Neural Network 99. Renewable Energy-Based Load Frequency Stabilization of Interconnected Power Systems Using Quasi-Oppositional Dragonfly Algorithm

100. Nonlinear Model Predictive Control of an Autonomous Power System Based on Hydrocarbon Reforming and High Temperature Fuel Cell. Energies 2021, 14, 1371 101. Primary frequency control techniques for large-scale PV-integrated power systems: A review 102. Optimal Operation of a Power System with Cross-border Electricity Trading Considering Demand Response Program: A Case Study of Afghanistan 103. Combined solar power and storage as cost-competitive and grid-compatible supply for China’s future carbon-neutral electricity system 104. Design Optimization Analysis Based On Demand Side Management of a Stand-alone Hybrid Power System Using Genetic Algorithm for Remote Rural Electrification 105. Enhance Power Quality of Grid Connected Wind and Solar Power System with ANFIS Control Scheme 106. Observer-based event triggering H∞ LFC for multi-area power systems under DoS attacks 107. Diagnosis performance assessment of the secondary protection for a 68‐bus power system 108. Results of a 200 hours lifetime test of a 7 kW Hybrid–Power fuel cell system on electric forklifts 109. Optimization of power-to-heat flexibility for residential buildings in response to day-ahead electricity price

110. Impact of renewable generation on probabilistic dynamic security assessment of a power transmission system 111. Prospects of power generation from the deep fractured geothermal reservoir using a novel vertical well system in the Yangbajing geothermal field, China 112. A residential community-level virtual power plant to balance variable renewable power generation in Sweden 113. Evaluation of hierarchical controls to manage power, energy and daily operation of remote off-grid power systems 114. … Multivalent Diagnoses Developed by a Diagnostic Program with an Artificial Neural Network for Devices in the Electric Hybrid Power Supply System “House on Water” 115. Technical features of the computing and geo-information system for research of prospective interstate power grid expansion 116. Two stage unit commitment considering multiple correlations of wind power forecast errors 117. Blockchain-based securing of data exchange in a power transmission system considering congestion management and social welfare 118. An Optimization Technique for Voltage Regulation in Electrical Power Systems 119. Availability importance measures of components in smart electric power grid systems

120. Electromagnetic Transients on Power Plant Connection Caused by Lightning Event 121. … incorporating battery energy storage system, minimum variable contribution of demand response, and variable load damping coefficient in isolated power … 122. Comparatives Studies in Molten Salt Reactor FUJI-U3 with Various Power 123. Future of electrical aircraft energy power systems: An architecture review 124. Fluid power troubleshooting 125. Cascading Failures Assessment in Renewable Integrated Power Grids Under Multiple Faults Contingencies 126. COVID-19-induced low power demand and market forces starkly reduce CO 2 emissions 127. Technology revolution in the inspection of power transmission lines-A literature review 128. Reactive power control of photovoltaic power generation systems by a wide‐area control system for improving transient stability in power systems 129. Techno-economic and environmental assessment of the coordinated operation of regional grid-connected energy hubs considering high penetration of wind power 130. A hybrid grey wolf optimisation and pattern search algorithm for automatic generation control of multi-area interconnected power systems 131. ICA based Multi Converter Strategy for Improving Power Quality in Multi-Feeder System with Renewable Source Integration 132. A Self Monitoring and Analyzing System for Solar Power Station using IoT and Data Mining Algorithms 133. Analysis of Emergency Operation Modes of Micro Power Systems with Small Hydroelectric Power Plants 134. Improving the voltage quality and power transfer capability of transmission system using facts controller 135. Particle swarm optimization in image processing of power flow learning distribution 136. A Modified Teaching—Learning-Based Optimization for Dynamic Economic Load Dispatch Considering Both Wind Power and Load Demand Uncertainties With … 137. An ensemble model for wide-area measurement-based transient stability assessment in power systems 138. An adaptative control strategy for interfacing converter of hybrid microgrid based on improved virtual synchronous generator 139. Quantum computing based hybrid deep learning for fault diagnosis in electrical power systems

140. Performance enhancement and multi-objective optimization of a double-flash binary geothermal power plant 141. Optimal reactive power dispatch using an improved slime mould algorithm 142. Applications of Metaheuristics in Power Electronics 143. A Dynamic State Estimator Based Tolerance Control Method Against Cyberattack and Erroneous Measured Data for Power Systems 144. Establishment of Low Voltage Ride-Through Curves and Stability Analysis with High Photovoltaic Penetration in Power Systems 145. A systematic review of the costs and impacts of integrating variable renewables into power grids 146. A wind power accommodation capability assessment method for multi-energy microgrids 147. Analyzing the effects of economic development on the transition to cleaner production of China’s energy system under uncertainty 148. Enhancing fault detection function in wind farm‐integrated power network using Teaching Learning‐Based Optimization technique 149. Economic analysis and optimization of a renewable energy based power supply system with different energy storages for a remote island

150. A hybrid SATS algorithm-based optimal power flow for security enhancement using SSSC 151. Prediction-based analysis on power consumption gap under long-term emergency: A case in China under COVID-19 152. Dynamic virtual power plant: A new concept for grid integration of renewable energy sources 153. Energy Storage Investment and Operation in Efficient Electric Power Systems 154. Comprehensive Design of DC Shipboard Power Systems for Pure Electric Propulsion Ship Based on Battery Energy Storage System 155. Physics-informed neural networks for minimising worst-case violations in dc optimal power flow 156. A review on fractional order (FO) controllers’ optimization for load frequency stabilization in power networks 157. Resilience analysis and cascading failure modeling of power systems under extreme temperatures 158. Cascade‐IλDμN controller design for AGC of thermal and hydro‐thermal power systems integrated with renewable energy sources 159. Hospital-oriented quad-generation (HOQG)—A combined cooling, heating, power and gas (CCHPG) system

160. Energy, exergy, economy and environmental (4E) analysis and optimization of single, dual and triple configurations of the power systems: Rankine Cycle/Kalina Cycle … 161. Deriving pack rules for hydro–photovoltaic hybrid power systems considering diminishing marginal benefit of energy 162. Safe Reinforcement Learning for Emergency Load Shedding of Power Systems 163. Electric power systems 164. Solving optimal power flow problem with stochastic wind–solar–small hydro power using barnacles mating optimizer 165. An efficient control strategy of shunt active power filter for asymmetrical load condition using time domain approach 166. Analysis of various options for balancing power systems’ peak load 167. An open-source extendable model and corrective measure assessment of the 2021 texas power outage 168. Optimal supply chains and power sector benefits of green hydrogen 169. Improving the Efficacy of the Nigerian Electric Power Transmission Network Using Static Synchronous Compensator (STACOM)

170. Optimal Placement of PMUs for Kerala and Tamil Nadu State Level Regional Indian Power Grid 171. Sharing hydropower flexibility in interconnected power systems: A case study for the China Southern power grid 172. Full energy sector transition towards 100% renewable energy supply: Integrating power, heat, transport and industry sectors including desalination 173. Multi-objective optimal power flow problems based on slime mould algorithm 174. System simulation and exergetic analysis of solid oxide fuel cell power generation system with cascade configuration 175. The importance of temporal resolution in modeling deep decarbonization of the electric power sector 176. MULTI-OBJECTIVE OPTIMAL REACTIVE POWER DISPATCH CONSIDERING THE INTEGRATION OF PROBABILISTIC WIND AND SOLAR POWER 177. Knowledge implementation and transfer with an adaptive learning network for real-time power management of the plug-in hybrid vehicle 178. Solar power satellites research in China 179. Frequency stabilization of solar thermal-photovoltaic hybrid renewable power generation using energy storage devices

i. Design and Comparison of Auxiliary Resonance controllers for Mitigating Modal Resonance of Power Systems Integrated with Wind Generation ii. A Computational Intelligence Approach for Power Quality Monitoring iii. Cascading Failure A nalysis of Cyber Physical Power Systems Considering Routing Strategy iv. Demand Side Management by PV Integration to Micro-grid Power Distribution System: A Review and Case Study Analysis v. Use of a hybrid wind—solar—diesel—battery energy system to power buildings in remote areas: a case study vi. Challenges and opportunities of inertia estimation and forecasting in low-inertia power systems vii. Generation Scheduling of Hydro-dominated Provincial Power Grid: Problems and Solutions viii. A review of machine learning applications in IoT-integrated modern power systems ix. Use of hydrogen as a seasonal energy storage system to manage renewable power deployment in Spain by 2030 x. Assessment of utilization of combined heat and power systems to provide grid flexibility alongside variable renewable energy systems

180. The role of the power sector in net-zero energy systems 181. A new approach to determine maintenance periods of the most critical hydroelectric power plant equipment 182. Application of High‐Performance Computing in Synchrophasor Data Management and Analysis for Power Grids 183. Wide-area monitoring and anomaly analysis based on synchrophasor measurement 184. Marine predators algorithm for load frequency control of modern interconnected power systems including renewable energy sources and energy storage units 185. Comparative study on the thermodynamic and economic performance of novel absorption power cycles driven by the waste heat from a supercritical CO2 cycle 186. Adaptive constraint differential evolution for optimal power flow 187. Optimization of location and size of distributed generations for maximizing their capacity and minimizing power loss of distribution system based on cuckoo search … 188. Solar power-to-gas application to an island energy system 189. Fault Detection in Power Transmission System Using Reverse Biorthogonal Wavelet

190. Line failure localization of power networks part i: Non-cut outages 191. Sunsetting coal power in China 192. Decentralised stochastic disturbance observer-based optimal frequency control method for interconnected power systems with high renewable shares 193. Hydrogen-based systems for integration of renewable energy in power systems: Achievements and perspectives 194. Research on frequency modulation control of photovoltaic power generation system based on VSG 195. Research on Grid Expansion Planning and Reliability Balance under the Fusion of Energy Storage and Wind Power 196. Asynchronous Control for Discrete-Time Hidden Markov Jump Power Systems 197. Nonlinear disturbance observer based adaptive super twisting sliding mode load frequency control for nonlinear interconnected power network 198. A Review of Lithium-Ion Battery Models in Techno-economic Analyses of Power Systems 199. Probabilistic estimation model of power curve to enhance power output forecasting of wind generating resources

• Power Systems Topics of presentation

More Research Topics on Electrical Engineering

Related posts:.

• Wireless Systems Research Topics ideas for MS PhD
• information visualization Research Topics Ideas [MS PhD]
• Molecular Computing Research Topics Ideas [MS PhD]
• Software Security Research Topics Ideas [MS PhD]
• Stochastic Networks Research Topics Ideas [MS PhD]

Contents Copyrights reserved by T4Tutorials

Research Topics

Power Electronics Design

Power electronics architectures are trending increasingly towards modular multi-converter structures that facilitate plug-and-play operation while enhancing reliability and efficiency. Generally speaking, this could take the form of parallel-connected systems that promote current sharing or series-connected systems that enable operation at elevated voltages. Our group formulates high-performance solutions for parallel-connected converters in computational applications, point-of-load setups, as well as microgrids. Innovations in series-connected configurations facilitate medium-voltage energy conversion for batteries, photovoltaics, and solid-state transformer applications. 

Low-inertia Power Systems & Grid-forming Inverters

Modern energy resources, such as photovoltaics, batteries, wind, and electric vehicles are interfaced to the grid through power electronics. These interfaces are fundamentally distinct from conventional synchronous generators in that they do not contain moving parts and their dynamics are shaped with digital controls. As generation shifts from large rotating machines to collections of electronic interfaces dispersed across the grid, system dynamics will accelerate under reduced inertia and system structures will become increasingly decentralized. Our group is reimagining the way grids are built and stands at the forefront of grid-forming inverter technologies that enable scalable and resilient power systems. UW is also a co-lead of the UNIFI Consortium.

Electromechanics & Drives

Electromechanical drive systems for vehicles and modern variable speed mechanical systems entail complex multiphysics phenomena that span across the mechanical, electromagnetic, electrical, and control domains. Untangling this interplay of dynamical systems and unlocking high-performance solutions requires breakthroughs in the realms of modeling, design, and experimentation. On the analytical front, we are leveraging the universality of energy to formulate equivalent circuit models that reveal the operation of closed-loop drive systems in a lucid and visually intuitive manner. These approaches facilitate new design methodologies which are validated on custom-designed SiC-based drive circuits and high power density axial flux machines.

Opportunities

Students with interests in controls, power electronics, or power systems are encouraged to email me their resume along with a brief description of their background and interests.

• Frontiers in Energy Research
• Sustainable Energy Systems
• Research Topics

Emerging Technologies for the Construction of Renewable Energy-Dominated Power System

Total Downloads

Total Views and Downloads

About this Research Topic

Over the past decade, significant breakthroughs have been achieved in renewable energy generation, operation, and control technology, greatly enhancing the safe operation and efficient utilization of renewable energy. However, as the penetration ratio of the renewable energy continues to grow, the ...

Keywords : grid forming, fractional frequency transmission system (FFTS), high voltage direct current (HVDC) transmission, power converter & network topology, power system stability and reliability, advanced control strategies, renewable energy

Important Note : All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.

Topic Editors

Topic coordinators, recent articles, submission deadlines, participating journals.

Manuscripts can be submitted to this Research Topic via the following journals:

total views

• Demographics

No records found

total views article views downloads topic views

Top countries

Top referring sites, about frontiers research topics.

With their unique mixes of varied contributions from Original Research to Review Articles, Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author.

Research areas

The power and energy systems group is involved in.

• Active power filter
• Conduction/breakdown mechanisms in insulation materials under high voltage (HV) stress
• Analyzing power quality problems in deregulated environment
• Ancillary services markets
• Bifurcations and chaos in power systems
• Congestion management
• Current-source converter-based Flexible alternating current transmission system (F ACTS) devices
• Development of power supplies
• Distributed energy resources grid interfacing
• Distributed generation issues in electricity markets
• Distribution system automation
• Distribution system planning and load forecasting
• Disturbance monitoring
• Electrostatic applications to industrial processes
• Extraction and mitigation
• Fuel-cell inverters
• HV testing techniques
• Microsensors
• Microturbine-based distributed power generation
• Modeling, control, and implementation of converters
• Web-based power systems programming applications
• Nonlinear systems theory and applications to electricity markets
• Partial discharges
• Passive and hybrid filters design
• Photo-voltaic grid-connected inverters
• Power quality improvement in distribution systems
• Power sector deregulation and electricity markets
• Reactive power control
• Real and reactive power pricing
• Simulation, modeling, and analysis of FACTS devices
• Static frequency changers
• Super-conducting magnetic energy storage

Electric Power Systems Research Latest Publications

Total documents, published by elsevier.

• Latest Documents
• Most Cited Documents
• Contributed Authors
• Related Sources
• Related Keywords

Protection scheme for VSC-MTDC based on low-frequency reactive power

An efficient algorithm for atomic decomposition of power quality disturbance signals using convolutional neural network, on the solution of robust transmission expansion planning using duality theorem under polyhedral uncertainty set, optimum power flow in dc microgrid employing bayesian regularized deep neural network, suppression strategy on neutral point overvoltage in resonant grounding system considering single line-to-ground fault, a review of scenario analysis methods in planning and operation of modern power systems: methodologies, applications, and challenges, power system transient security assessment based on deep learning considering partial observability, damped nyquist plot for the phase and gain optimization of power system stabilizers, a fault reconfiguration strategy based on adjustable space operator discrete state transition algorithm for ship microgrid system, decision-making framework for power system with res including responsive demand, esss, ev aggregator and dynamic line rating as multiple flexibility resources, export citation format, share document.

Department of electric energy

• Master's programmes in English
• For exchange students
• PhD opportunities
• All programmes of study
• Language requirements
• Application process
• Academic calendar
• NTNU research
• Research excellence
• Strategic research areas
• Innovation resources
• Student in Trondheim
• Student in Gjøvik
• Student in Ålesund
• For researchers
• Life and housing
• Faculties and departments
• International researcher support

Språkvelger

Research - department of electric power engineering.

At the Department of Electric Energy (IEL), the mission is to contribute to the fundamental and applied knowledge of electric power engineering, and to develop technology and systems for the planning, operation and maintenance of efficient, sustainable energy systems.  Both research and research-based education at the Department of Electric Energy cover the broad interdisciplinary aspects of power engineering: generation, transmission, conversion and the use of electric energy, including the accompanying techno-economic aspects.

The Department works in close collaboration with industry partners to develop technology for the production of electric energy from renewable energy sources, and contribute to research that leads to solutions for the future power grid, with high relevance for the society, addressing industrial needs and global challenges.

We have five research groups that are responsible for research and education within their areas; the groups also collaborate among each other in research projects. 

Research Groups

Electrical machines and electromagnetics (eme).

The main research areas of the group are related to the development, design, optimisation and testing of electric machinery, especially permanent magnet machines and hydropower generators. In addition, research is also focused on advanced electromagnetic modelling and analyses of different power apparatus and installations.

Electricity Markets and Energy System Planning (EMESP)

The main research areas of the group are related to the integration of renewable energy sources, energy storage and consumption in the electricity market, and how to optimize the integration of the power system with other parts of the energy system, e.g. heating and transport.

High Voltage Technology (HVT)

The main research activities of the group are related to the design, modeling and operation of electric power components. Research into better insulation materials, both for ac and dc, is being conducted.

Power Electronic Systems and Components (PESC)

The research activities of the group are related to the development, design, optimization and control of power electronic converters and systems. Application areas include onshore and offshore power systems, marine, oil & gas as well as transportation sectors. 

Power System Operation and Analysis (PSOA)

The main research areas of the group are related to the planning, operation, control and analysis of power systems, with applications in smart grids, transmission and distribution grids, microgrids and HVDC systems.  

PhD studies

Phd studies.

The PhD programme in Electric Power Engineering is standardised to 180 credits (3 years). The final plan for the PhD programme is designed in consultation with the candidate, the supervisor and the Department depending on the subject area of the thesis and the candidate's needs and preferences.

• PhD programme in Electric Power Engineering
• Info on PhD studies at NTNU

Vacant positions at the Department

person-portlet

Anngjerd pleym head of department, frank mauseth, kjetil obstfelder uhlen, publications.

Cristin (Current Research Information system) is a nationwide database system for research results and documentation.

• All Publications by the Department of Electric E nergy
• All PhD theses from the Department of Electric E nergy
• All Masters theses from the Department of Electric Energy

Summary of PhD projects

Summary of master projects

Physical Laboratory

The Department has access to a broad spectrum of advanced laboratories and scientific equipment.

• Electrical Machines | Pål Keim Olsen
• Electric Circuits
• High Voltage | Frank Mauseth
• High Current | Kaveh Niayesh
• Light and Lighting | Eilif Hugo Hansen
• Power Electronics | Dimosthenis Peftitsis ,  Anyuan Chen
• Renewable Energy
• Relay Protection Laboratory | Hans Kr. Høidalen
• Service Laboratory | Bård Almås
• Smartgrid | Basanta Raj Pokhrel
• Smarthus | Eilif Hugo Hansen
• Workshop | Morten Flå

Digital Laboratory

The Department has many digital resources

• Digital Lab
• Computing Resources available at IEL
• Scientific Software

Strategic Research Areas

• IE Faculty Strategic Research Area: SMARTGRIDS
• NTNU Strategic Research Area: ENERGY

Gemini Centre

About Gemini Centres

• Gemini Centre for Electrical Energy and Energy Systems

New Technologies for Power System Operation and Analysis, 1st Edition

• Power Systems Engineering
• National Wind Technology Center

Research output : Book/Report › Book

NREL Publication Number

• NREL/BK-5D00-72140
• electricity markets
• forecasting
• monitoring systems
• renewable energy integration
• system operation

Access to Document

• 10.1016/C2018-0-00913-5

Fingerprint

• Power System Operation Engineering 100%
• Systems Analysis Engineering 100%
• Energy Integration Engineering 66%
• Renewable Energy Engineering 66%
• Renewables Engineering 33%
• Power Engineering Engineering 33%
• Transmissions Engineering 33%
• System Operation Engineering 33%

T1 - New Technologies for Power System Operation and Analysis, 1st Edition

AU - Jiang, Huaiguang

AU - Zhang, Yingchen

AU - Muljadi, Eduard

N2 - New Technologies for Power System Operation and Analysis considers the very latest developments in renewable energy integration and system operation, including electricity markets and wide-area monitoring systems and forecasting. Helping readers quickly grasp the essential information needed to address renewable energy integration challenges, this new book looks at basic power system mathematical models, advanced renewable integration and system optimizations from transmission and distribution system sides. Sections cover wind, solar, gas and petroleum, making this a useful reference for all engineers interested in power system operation.

AB - New Technologies for Power System Operation and Analysis considers the very latest developments in renewable energy integration and system operation, including electricity markets and wide-area monitoring systems and forecasting. Helping readers quickly grasp the essential information needed to address renewable energy integration challenges, this new book looks at basic power system mathematical models, advanced renewable integration and system optimizations from transmission and distribution system sides. Sections cover wind, solar, gas and petroleum, making this a useful reference for all engineers interested in power system operation.

KW - electricity markets

KW - forecasting

KW - monitoring systems

KW - petroleum

KW - renewable energy integration

KW - system operation

U2 - 10.1016/C2018-0-00913-5

DO - 10.1016/C2018-0-00913-5

BT - New Technologies for Power System Operation and Analysis, 1st Edition

Atom-by-atom: Imaging structural transformations in 2D materials

Silicon-based electronics are approaching their physical limitations and new materials are needed to keep up with current technological demands. Two-dimensional (2D) materials have a rich array of properties, including superconductivity and magnetism, and are promising candidates for use in electronic systems, such as transistors. However, precisely controlling the properties of these materials is extraordinarily difficult.

In an effort to understand how and why 2D interfaces take on the structures they do, researchers at the University of Illinois Urbana-Champaign have developed a method to visualize the thermally-induced rearrangement of 2D materials, atom-by-atom, from twisted to aligned structures using transmission electron microscopy (TEM). They observed a new and unexpected mechanism for this process where a new grain was seeded within one monolayer, whose structure was templated by the adjacent layer. Being able to control the macroscopic twist between layers allows for more control over the properties of the entire system.

This research, led by materials science & engineering professor Pinshane Huang and postdoctoral researcher Yichao Zhang, was recently published in the journal Science Advances .

"How the interfaces of the bilayer align with each other and through what mechanism they transform into a different configuration is very important," Zhang says. "It controls the properties of the entire bilayer system which, in turn, affects both its nanoscale and microscopic behavior."

The structure and properties of 2D multilayers are often highly heterogeneous and vary widely between samples and even within an individual sample. Two devices with just a few degrees of twist between layers could have different behavior. 2D materials are also known to reconfigure under external stimuli such as heating, which occurs during the fabrication process of electronic devices.

"People usually think of the two layers like having two sheets of paper twisted 45° to each other. To get the layers to go from twisted to aligned, you would just rotate the entire piece of paper," Zhang says. "But what we found, actually, is it has a nucleus -- a localized nanoscale aligned domain -- and this domain grows larger and larger in size. Given the correct conditions, this aligned domain could take over the entire size of the bilayer."

While researchers have speculated that this may happen, there hasn't been any direct visualization at the atomic scale proving or disproving the theory. Zhang and the other researchers, however, were able to directly track the movement of individual atoms to see the tiny, aligned domain grow. They also observed that aligned regions could form at relatively low temperatures, ~200°C, in the range of typical processing temperatures for 2D devices.

There aren't cameras small enough and fast enough to capture atomic dynamics. How then was the team able to visualize this atom-by-atom movement? The solution is very unique. They first encapsulated the twisted bilayer in graphene, essentially building a little reaction chamber around it, to look at the bilayer at atomic resolution as it was heated. Encapsulation by graphene helps to hold the atoms of the bilayer in place so that any structural transformation could be observed rather than the lattice getting destroyed by high-energy electrons of the TEM.

The encapsulated bilayer was then put on a chip that could be heated and cooled quickly. To capture the fast atomic dynamics, the sample underwent half second heat pulses between 100-1000°C. After each pulse, the team would look at where the atoms were using TEM and then repeated the process. "You can actually watch the system as it changes, as the atoms settle in from whatever configuration they were put in initially, to the configuration that is energetically favorable, that they want to be in," Huang explains. "That can help us understand both the initial structure as it is fabricated and how it evolves with heat."

Understanding how rearrangement happens can help tune the interfacial alignment at the nanoscale. "It is impossible to underscore how excited people are about that tuneability," Huang says. "The macroscopic twist between the two layers is a really important parameter because as you rotate one on the other, you can actually change the properties of the entire system. For example, if you rotate the 2D material graphene to a specific angle, it becomes superconducting. For some materials, if you rotate them, you change the bandgap which changes the color of light it absorbs and what energy of light it emits. All of those things you change by altering the orientation of atoms between layers."

• Materials Science
• Spintronics
• Engineering and Construction
• Civil Engineering
• Nanotechnology
• Materials science
• Nanoparticle
• Triboelectric effect
• Electrical engineering
• Carbon nanotube
• Phase (matter)

Story Source:

Materials provided by University of Illinois Grainger College of Engineering . Original written by Amber Rose. Note: Content may be edited for style and length.

Journal Reference :

• Yichao Zhang, Ji-Hwan Baek, Chia-Hao Lee, Yeonjoon Jung, Seong Chul Hong, Gillian Nolan, Kenji Watanabe, Takashi Taniguchi, Gwan-Hyoung Lee, Pinshane Y. Huang. Atom-by-atom imaging of moiré transformations in 2D transition metal dichalcogenides . Science Advances , 2024; 10 (13) DOI: 10.1126/sciadv.adk1874

Cite This Page :

Explore More

• Two Species Interbreeding Created New Butterfly
• Warming Antarctic Deep-Sea and Sea Level Rise
• Octopus Inspires New Suction Mechanism for ...
• Cities Sinking: Urban Populations at Risk
• Puzzle Solved About Ancient Galaxy
• How 3D Printers Can Give Robots a Soft Touch
• Combo of Multiple Health Stressors Harming Bees
• Methane Emission On a Cold Brown Dwarf
• Remarkable Memories of Mountain Chickadees
• Predicting Future Marine Extinctions

Trending Topics

Strange & offbeat.

• Systems Ph.D.
• M.Eng. Degree On Campus
• M.Eng. Degree Distance Learning
• Systems M.S. Degree
• Minor in Systems Engineering
• Professional Certificates
• Student Organizations
• Energy Systems M.Eng. Pathway
• Systems M.Eng. Projects

Research Topics

• Research News
• Ezra's Round Table / Systems Seminar Series
• Academic Leadership
• Graduate Field Faculty
• Graduate Students
• Staff Directory
• Ezra Systems Postdoctoral Associates
• Research Associates
• Faculty Openings-Systems
• Get Involved
• Giving Opportunities
• Recruit Students
• Systems Magazine
• Academic Support
• Experience and Employment
• Graduate Services and Activities
• Mental Health Resources
• Recruitment Calendar
• Tuition and Financial Aid
• Program Description
• Program Offerings
• How to Apply
• Ezra Postdoctoral Associate in Energy Systems Engineering
• Cornell Systems Summit

Research in Systems Engineering at Cornell covers an extremely broad range of topics, because of this nature, the research takes on a collaborative approach with faculty from many different disciplines both in traditional engineering areas as well as those outside of engineering.

Because of the nature of systems science and engineering, the research takes on a collaborative approach with faculty and students from many different disciplines both in traditional engineering areas as well as those outside of engineering such as health care, food systems, environmental studies, architecture and regional planning, and many others.

Artificial Intelligence

Computational science and engineering, computer systems.

Data Mining

Earth and Atmospheric Science

Energy systems, health systems, heat and mass transfer.

Information Theory and Communication

Infrastructure Systems

Mechanics biological materials, natural hazards.

Programming Languages - CS

Remote Sensing

Robotics and autonomy, satellite systems, scientific computing, sensor and actuators, signal and image processing, space science and engineering, statistics and machine learning, statistical mechanics and molecular simulation, sustainable energy systems, systems and networking - cs, transportation systems engineering, water systems.

Oliver Gao | Civil and Environmental Engineering

David Goldberg | Operations Research and Information Engineering

Adrian Lewis |  Operations Research and Information Engineering

Linda Nozick |  Civil and Environmental Engineering

Francesca Parise | Electrical and Computer Engineering

Mason Peck | Mechanical and Aerospace Engineering

Patrick Reed |  Civil and Environmental Engineering

Samitha Samaranayake |  Civil and Environmental Engineering

Timothy Sands |  Mechanical and Aerospace Engineering

Huseyin Topaloglu |  Operations Research and Information Engineering

Fengqi You | Chemical and Biomolecular Engineering

Mark Campbell | Mechanical and Aerospace Engineering

Kirstin Petersen |  Electrical and Computer Engineering

Patrick Reed | Civil and Environmental Engineering

Jose Martinez | Electrical and Computer Engineering

Data Science

Madeleine Udell | Operations Research and Information Engineering

Maha Haji | Mechanical and Aerospace Engineering

Semida Silveira | Systems Engineering

Jery Stedinger |  Civil and Environmental Engineering

Jefferson Tester | Chemical and Biomolecular Engineering

Lang Tong | Electrical and Computer Engineering

Fengqi You |  Chemical and Biomolecular Engineering

Shane Henderson | Operations Research and Information Engineering

John Muckstadt |  Operations Research and Information Engineering

Jamol Pender |  Operations Research and Information Engineering

Rana Zadeh |  Human Centered Design

Yiye Zhang |  Weill Cornell Medicine

Information Theory and Communications

Stephen Wicker | Electrical and Computer Engineering

Programming Languages – CS

Andrew Myers | Computer Science

Fred Schneider | Computer Science

Mason Pack | Mechanical and Aerospace Engineering

Mark Campbell |  Mechanical and Aerospace Engineering

Robert Shepherd |  Mechanical and Aerospace Engineering

Richardo Daziano | Civil and Environmental Engineering

Linda Nozick | Civil and Environmental Engineering

Bart Selman | Computer Science

Timur Dogan | Arts Architecture and Planning

Ken Birman | Computer Science

Hakim Weatherspoon | Computer Science

Richard Geddes | College of Human Ecology

An official website of the United States government

Here's how you know

Official websites use .gov A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS. A lock ( Lock Locked padlock ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

Flood mitigation, rural transportation, supply chains: NSF awards 9 new people-focused research grants

Flood mitigation measures, rural transportation systems and pharmaceutical supply chains are among the subjects being examined by nine new research projects receiving backing from the U.S. National Science Foundation. NSF is investing more than $8 million in the projects through its Strengthening American Infrastructure program. The program supports research that utilizes advances in behavioral and social science to improve the value and usefulness of infrastructure in people's lives, from U.S. roads and highways to state and local power grids.

The projects are centered on social and behavioral science being conducted at institutions across the U.S. in collaboration with researchers from a wide range of other fields, including computer science, engineering, geosciences, mathematics and physical sciences.

"These projects can reveal new ways to enhance safety, reduce pain points in our everyday lives and enable greater prosperity and security for future generations," says Marc Sebrechts, director of NSF's Behavioral and Cognitive Sciences Division. "NSF's Strengthening American Infrastructure program is using the illuminating power of social and behavioral science to examine issues and opportunities at the heart of practically every type of infrastructure that people in the U.S. depend on."

"Strong infrastructure stimulates U.S. job creation, improves our quality of life and protects the well-being of our communities for many years into the future," says NSF Assistant Director for Social, Behavioral and Economic Sciences Kellina Craig-Henderson. "The projects we're supporting are taking science out of the proverbial lab and into a broad new array of areas so communities large and small can directly benefit from their discoveries."

The projects' diverse goals include improving air quality in rural areas with high levels of highway traffic, enabling communities and neighborhoods to retain power during large-scale power outages and helping states and cities effectively plan for increasingly frequent floods and storm surges. In addition to conducting fundamental research, NSF-supported projects will support participation and outreach activities with the residents of several local communities, the creation of new tools and resources that will be freely available to others and multiple educational and training opportunities for STEM students.

The three-year projects will be led by 12 institutions in nine states, including three states within the Established Program to Stimulate Competitive Research which supports areas in the U.S. that have historically received less federal support for research and development:

• A collaboration between Texas A&M University and the University of Michigan will examine the community benefits and risks involved in a proposed new coastal storm barrier designed to protect people in the Houston-Galveston area of Texas from increased storm surges and flooding due to hurricanes.
• Northeastern University researchers will explore how to design a new information infrastructure that can ease the supply-chain disruptions that lead to recurring shortages of medicines and other pharmaceutical products in the U.S.
• Clemson University and the University at Buffalo, State University of New York will collaborate on developing a new cyberinfrastructure system that will provide scientists, cybersecurity professionals and others with a more rigorous way to use artificial intelligence to detect harmful online behaviors such as cyberbullying and inciting violence.
• Purdue University researchers will conduct experiments focused on accurately predicting the long-term community-level impacts of retrofitting older buildings to withstand future earthquakes and other natural disasters. Their study includes buildings such as hospitals and residential complexes.
• University of Virginia and Northeastern University will collaborate on an investigation of how physical infrastructure and social factors affect access to safe drinking water during hurricanes and other natural disasters.
• University of Texas at Austin researchers will investigate how water-utility companies can keep critical water systems operating by more effectively communicating emergency messages to their customers before and during catastrophic events such as winter storms.
• George Mason University researchers will analyze how electrical infrastructure is distributed across the U.S. to understand how the potential costs of supporting an increasing number of electric vehicles can be fairly distributed, particularly for those who do not use electric vehicles and for low and middle-income families.
• University of Vermont researchers will study how investments in rural communities' travel infrastructure, such as roads and bridges, influence people's travel-related decisions and how transportation planners and engineers can most effectively use that information.
• Iowa State University researchers will examine the social and economic benefits of microgrids — small, autonomous electrical grids that deliver power to a local area during natural disasters and other emergencies — for low-income families and other groups who face energy insecurity.

In addition to these nine research projects, NSF is also providing smaller planning grants to other institutions to study the feasibility of future infrastructure-focused research. For additional information, please visit NSF's Strengthening American Infrastructure program webpage .

Research areas

• MyU : For Students, Faculty, and Staff

Fall 2024 CSCI Special Topics Courses

Cloud computing.

Meeting Time: 09:45 AM‑11:00 AM TTh  Instructor: Ali Anwar Course Description: Cloud computing serves many large-scale applications ranging from search engines like Google to social networking websites like Facebook to online stores like Amazon. More recently, cloud computing has emerged as an essential technology to enable emerging fields such as Artificial Intelligence (AI), the Internet of Things (IoT), and Machine Learning. The exponential growth of data availability and demands for security and speed has made the cloud computing paradigm necessary for reliable, financially economical, and scalable computation. The dynamicity and flexibility of Cloud computing have opened up many new forms of deploying applications on infrastructure that cloud service providers offer, such as renting of computation resources and serverless computing.    This course will cover the fundamentals of cloud services management and cloud software development, including but not limited to design patterns, application programming interfaces, and underlying middleware technologies. More specifically, we will cover the topics of cloud computing service models, data centers resource management, task scheduling, resource virtualization, SLAs, cloud security, software defined networks and storage, cloud storage, and programming models. We will also discuss data center design and management strategies, which enable the economic and technological benefits of cloud computing. Lastly, we will study cloud storage concepts like data distribution, durability, consistency, and redundancy. Registration Prerequisites: CS upper div, CompE upper div., EE upper div., EE grad, ITI upper div., Univ. honors student, or dept. permission; no cr for grads in CSci. Complete the following Google form to request a permission number from the instructor ( https://forms.gle/6BvbUwEkBK41tPJ17 ).

CSCI 5980/8980 

Machine learning for healthcare: concepts and applications.

Meeting Time: 11:15 AM‑12:30 PM TTh  Instructor: Yogatheesan Varatharajah Course Description: Machine Learning is transforming healthcare. This course will introduce students to a range of healthcare problems that can be tackled using machine learning, different health data modalities, relevant machine learning paradigms, and the unique challenges presented by healthcare applications. Applications we will cover include risk stratification, disease progression modeling, precision medicine, diagnosis, prognosis, subtype discovery, and improving clinical workflows. We will also cover research topics such as explainability, causality, trust, robustness, and fairness.

Registration Prerequisites: CSCI 5521 or equivalent. Complete the following Google form to request a permission number from the instructor ( https://forms.gle/z8X9pVZfCWMpQQ6o6  ).

Visualization with AI

Meeting Time: 04:00 PM‑05:15 PM TTh  Instructor: Qianwen Wang Course Description: This course aims to investigate how visualization techniques and AI technologies work together to enhance understanding, insights, or outcomes.

This is a seminar style course consisting of lectures, paper presentation, and interactive discussion of the selected papers. Students will also work on a group project where they propose a research idea, survey related studies, and present initial results.

This course will cover the application of visualization to better understand AI models and data, and the use of AI to improve visualization processes. Readings for the course cover papers from the top venues of AI, Visualization, and HCI, topics including AI explainability, reliability, and Human-AI collaboration.    This course is designed for PhD students, Masters students, and advanced undergraduates who want to dig into research.

Registration Prerequisites: Complete the following Google form to request a permission number from the instructor ( https://forms.gle/YTF5EZFUbQRJhHBYA  ). Although the class is primarily intended for PhD students, motivated juniors/seniors and MS students who are interested in this topic are welcome to apply, ensuring they detail their qualifications for the course.

Visualizations for Intelligent AR Systems

Meeting Time: 04:00 PM‑05:15 PM MW  Instructor: Zhu-Tian Chen Course Description: This course aims to explore the role of Data Visualization as a pivotal interface for enhancing human-data and human-AI interactions within Augmented Reality (AR) systems, thereby transforming a broad spectrum of activities in both professional and daily contexts. Structured as a seminar, the course consists of two main components: the theoretical and conceptual foundations delivered through lectures, paper readings, and discussions; and the hands-on experience gained through small assignments and group projects. This class is designed to be highly interactive, and AR devices will be provided to facilitate hands-on learning.    Participants will have the opportunity to experience AR systems, develop cutting-edge AR interfaces, explore AI integration, and apply human-centric design principles. The course is designed to advance students' technical skills in AR and AI, as well as their understanding of how these technologies can be leveraged to enrich human experiences across various domains. Students will be encouraged to create innovative projects with the potential for submission to research conferences.

Registration Prerequisites: Complete the following Google form to request a permission number from the instructor ( https://forms.gle/Y81FGaJivoqMQYtq5 ). Students are expected to have a solid foundation in either data visualization, computer graphics, computer vision, or HCI. Having expertise in all would be perfect! However, a robust interest and eagerness to delve into these subjects can be equally valuable, even though it means you need to learn some basic concepts independently.

Sustainable Computing: A Systems View

Meeting Time: 09:45 AM‑11:00 AM  Instructor: Abhishek Chandra Course Description: In recent years, there has been a dramatic increase in the pervasiveness, scale, and distribution of computing infrastructure: ranging from cloud, HPC systems, and data centers to edge computing and pervasive computing in the form of micro-data centers, mobile phones, sensors, and IoT devices embedded in the environment around us. The growing amount of computing, storage, and networking demand leads to increased energy usage, carbon emissions, and natural resource consumption. To reduce their environmental impact, there is a growing need to make computing systems sustainable. In this course, we will examine sustainable computing from a systems perspective. We will examine a number of questions:   • How can we design and build sustainable computing systems?   • How can we manage resources efficiently?   • What system software and algorithms can reduce computational needs?    Topics of interest would include:   • Sustainable system design and architectures   • Sustainability-aware systems software and management   • Sustainability in large-scale distributed computing (clouds, data centers, HPC)   • Sustainability in dispersed computing (edge, mobile computing, sensors/IoT)

Registration Prerequisites: This course is targeted towards students with a strong interest in computer systems (Operating Systems, Distributed Systems, Networking, Databases, etc.). Background in Operating Systems (Equivalent of CSCI 5103) and basic understanding of Computer Networking (Equivalent of CSCI 4211) is required.

• Future undergraduate students
• Future transfer students
• Future graduate students
• Future international students
• Diversity and Inclusion Opportunities
• Learn abroad
• Living Learning Communities
• Mentor programs
• Programs for women
• Student groups
• Visit, Apply & Next Steps
• Information for current students
• Departments and majors overview
• Departments
• Undergraduate majors
• Graduate programs
• Integrated Degree Programs
• Additional degree-granting programs
• Online learning
• Academic Advising overview
• Academic Advising FAQ
• Academic Advising Blog
• Appointments and drop-ins
• Academic support
• Commencement
• Four-year plans
• Honors advising
• Policies, procedures, and forms
• Career Services overview
• Resumes and cover letters
• Jobs and internships
• Interviews and job offers
• CSE Career Fair
• Major and career exploration
• Graduate school
• Collegiate Life overview
• Scholarships
• Diversity & Inclusivity Alliance
• Anderson Student Innovation Labs
• Information for alumni
• Get engaged with CSE
• Upcoming events
• CSE Alumni Society Board
• Alumni volunteer interest form
• Golden Medallion Society Reunion
• 50-Year Reunion
• Alumni honors and awards
• Outstanding Achievement
• Alumni Service
• Distinguished Leadership
• Honorary Doctorate Degrees
• Nobel Laureates
• Alumni resources
• Alumni career resources
• Alumni news outlets
• CSE branded clothing
• International alumni resources
• Inventing Tomorrow magazine
• Update your info
• CSE giving overview
• Why give to CSE?
• College priorities
• Give online now
• External relations
• Giving priorities
• Donor stories
• Impact of giving
• Ways to give to CSE
• Matching gifts
• CSE directories
• Invest in your company and the future
• Recruit our students
• Connect with researchers
• K-12 initiatives
• Diversity initiatives
• Research news
• Give to CSE
• CSE priorities
• Corporate relations
• Information for faculty and staff
• Administrative offices overview
• Office of the Dean
• Academic affairs
• Finance and Operations
• Communications
• Human resources
• Undergraduate programs and student services
• CSE Committees
• CSE policies overview
• Academic policies
• Faculty hiring and tenure policies
• Finance policies and information
• Graduate education policies
• Human resources policies
• Research policies
• Research overview
• Research centers and facilities
• Research proposal submission process
• Research safety
• Award-winning CSE faculty
• National academies
• University awards
• Honorary professorships
• Collegiate awards
• Other CSE honors and awards
• Staff awards
• Performance Management Process
• Work. With Flexibility in CSE
• K-12 outreach overview
• Summer camps
• Outreach events
• Enrichment programs
• Field trips and tours
• CSE K-12 Virtual Classroom Resources
• Educator development
• Sponsor an event