The increasing integration of renewable energy resources (RES) introduces uncertainties in modern power systems and makes the dynamic energy dispatch (DED) problem challenging. Uncertainties lead to dynamic grid control, which needs to be addressed for the optimized DED. Moreover, since energy usage and power generation are distributed, multiple parties can be involved in the DED problem. Thus, DED should be optimized in a distributed way for efficiency and privacy. With the development of the Internet of Things (IoT) and machine learning technology, various data can be gathered and analyzed to achieve intelligent energy management, and the dynamics of power grids should be considered for optimality. For this purpose, we investigate how reinforcement learning can be used to solve the DED problem for a dynamic microgrid (MG) environment. The objective is to determine the optimal power generation for each generator using fossil fuels at each time slot, to minimize the cumulative cost of power generation in a given time period. To achieve this goal, we first model the MG with the practical impact of batteries, photovoltaic (PV) panels, and load banks (external grids). Then we formulate the optimization problem of minimizing the total generation from fossil fuels. To solve this problem, we propose a distributed reinforcement learning algorithm to reduce communication costs and improve data privacy. In the proposed scheme, each generator is considered as an agent, which shares a global state and only obtains its own local loss. Then, different agents work jointly to minimize the global cost. Theoretical analysis is provided to prove the convergence of the proposed algorithms, which are also tested with real-world datasets. Results show that the policy learned from the proposed algorithms can balance the production and consumption in the MG for both fully and partially observable MG environments while simultaneously reducing the total generation cost from fossil fuels.

We investigate an optimal distributed energy dispatch strategy for networked Microgrids (MGs) considering uncertainties of distributed energy resources, the impact of energy storage, and privacy. The energy dispatch problem is formulated as a Partially Observed Markov Decision Process (POMDP), and is solved using Deep Deterministic Policy Gradient (DDPG) method. To reduce the communication load and protect privacy, a federated reinforcement learning (FRL) framework is proposed, where each MG trains model parameters with its own local data, and only transmits model weights to the global server. Finally, each MG can obtain a global model that can be generalized well in various cases. The proposed method is communication-efficient, privacy-preserving, and scalable. Numerical simulations are tested with real-world datasets, results demonstrate the effectiveness of the proposed FRL method.

In practical engineering, electricity theft detection is usually performed on highly imbalanced datasets (i.e., the number of fraudulent samples is much smaller than the benign ones), which limits the accuracy of the classifier. To alleviate the data imbalance problem, this article proposes simple data augmentation tricks (SDAT) to boost performance on electricity theft detection tasks. SDAT includes five simple but powerful operations: adding noises to electricity consumption readings, drifting values of electricity consumption readings, quantizing electricity consumption readings to a level set, adding a fixed value to electricity consumption readings, and adding changeable values to electricity consumption readings. In addition, eight potential tricks are also mentioned. Numerical simulations are conducted on a real-world dataset. The simulation results show that SDAT can significantly boost the performance of different classifiers, especially for small datasets. Besides, specific suggestions on how to select parameters of SDAT are provided for its migration use to other datasets.

Deep neural networks have revolutionized many machine learning tasks in power systems, ranging from pattern recognition to signal processing. The data in these tasks are typically represented in Euclidean domains. Nevertheless, there is an increasing number of applications in power systems, where data are collected from non-Euclidean domains and represented as graph-structured data with high-dimensional features and interdependency among nodes. The complexity of graph-structured data has brought significant challenges to the existing deep neural networks defined in Euclidean domains. Recently, many publications generalizing deep neural networks for graph-structured data in power systems have emerged. In this paper, a comprehensive overview of graph neural networks (GNNs) in power systems is proposed. Specifically, several classical paradigms of GNN structures, e.g., graph convolutional networks, are summarized. Key applications in power systems such as fault scenario application, time-series prediction, power flow calculation, and data generation are reviewed in detail. Furthermore, main issues and some research trends about the applications of GNNs in power systems are discussed.

High-quality datasets are of paramount importance for the operation and planning of wind farms. However, the datasets collected by the supervisory control and data acquisition (SCADA) system may contain missing data due to various factors such as sensor failure and communication congestion. In this paper, a data-driven approach is proposed to fill the missing data of wind farms based on a context encoder (CE), which consists of an encoder, a decoder, and a discriminator. Through deep convolutional neural networks, the proposed method is able to automatically explore the complex nonlinear characteristics of the datasets that are difficult to be modeled explicitly. The proposed method can not only fully use the surrounding context information by the reconstructed loss, but also make filling data look real by the adversarial loss. In addition, the correlation among multiple missing attributes is taken into account by adjusting the format of input data. The simulation results show that CE performs better than traditional methods for the attributes of wind farms with hallmark characteristics such as large peaks, large valleys, and fast ramps. Moreover, the CE shows stronger generalization ability than traditional methods such as auto-encoder, K-means, k-nearest neighbor, back propagation neural network, cubic interpolation, and conditional generative adversarial network for different missing data scales.

Scenario generations of cooling, heating, and power loads are of great significance for the economic operation and stability analysis of integrated energy systems. In this paper, a novel deep generative network is proposed to model cooling, heating, and power load curves based on generative moment matching networks (GMMNs) where an auto-encoder transforms high-dimensional load curves into low-dimensional latent variables and the maximum mean discrepancy represents the similarity metrics between the generated samples and the real samples. After training the model, the new scenarios are generated by feeding Gaussian noises to the scenario generator of the GMMN. Unlike the explicit density models, the proposed GMMN does not need to artificially assume the probability distribution of the load curves, which leads to stronger universality. The simulation results show that the GMMN not only fits the probability distribution of multi-class load curves very well, but also accurately captures the shape (e.g., large peaks, fast ramps, and fluctuation), frequency-domain characteristics, and temporal-spatial correlations of cooling, heating, and power loads. Furthermore, the energy consumption of generated samples closely resembles that of real samples.

Scenario generations for renewable energy sources and loads play an important role in the stable operation and risk assessment of integrated energy systems. This paper proposes a deep generative network based method to model time-series curves, e.g., power generation curves and load curves, of renewable energy sources and loads based on implicit maximum likelihood estimations (IMLEs), which can generate realistic scenarios with similar patterns as real ones. After training the model, any number of new scenarios can be obtained by simply inputting Gaussian noises into the data generator of IMLEs. The proposed approach does not require any model assumptions or prior knowledge of the form in the likelihood function being made during the training process, which leads to stronger applicability than explicit density model based methods. The extensive experiments show that the IMLEs accurately capture the complex shapes, frequency-domain characteristics, probability distributions, and correlations of renewable energy sources and loads. Moreover, the proposed approach can be easily generalized to scenario generation tasks of various renewable energy sources and loads by fine-tuning parameters and structures.

The penetrations of wind power and photovoltaic system in distribution network are increasing year by year. The randomness and fluctuation of their output powers bring great challenges to the planning and operation of distribution networks. Aimed at the uncertainty of output power of renewable energy, a stochastic scenarios generation method for photovoltaic and wind power based on generative moment matching network (GMMN) is proposed. In this method, the maximum mean discrepancy is adopted as the loss function of the generator, and the auto-encoder is adopted to reduce the dimension of the generated stochastic scenarios, so as to solve the low dimensional manifold problem of the high-dimensional power curves. According to the characteristics of the power curves, the network structure suitable for the stochastic scenarios generation of renewable energy is designed. The effectiveness and adaptability of the proposed method are verified by real data. The simulation results show that the proposed GMMN can not only simulate the shape characteristics, probability distribution characteristics, fluctuation, and spatial-temporal correlation of the photovoltaic and wind power curves, but also has universality. It can be applied to the random field generation tasks for different generation units only by adjusting the structure and parameters of the network.

Recently, there is increasing interest in detecting electricity thieves for economic benefits for power companies, and many works aim to improve the accuracy of electricity theft detection. Nevertheless, a core obstacle that currently hinders the direct comparison of classifiers for electricity theft detection is the lack of a standard and public dataset, since fraudulent power load profiles are usually difficult to collect for various reasons, including cost, cumber, and confidentiality. Therefore, this paper presents an open-source toolbox, which generates different kinds of fraudulent power load profiles from attack models, and integrates classical classifiers (e.g., support vector machine, multi-layer perceptron, convolutional neural network, long short-term memory, bidirectional long short-term memory) with different performance as baselines for the comparison with new algorithms. Users can easily generate datasets and modify parameters of classical classifiers guided by user friendly interactive interfaces. The codes, toolbox, and user manuals are available online and it is free to use and extend them. 

There are few transformer faults, which makes the methods of transformer fault diagnosis based on machine learning lack of data. For this reason, a method based on improved auto-encoder (IAE) is proposed to augment transformer fault data. Firstly, to solve the problem of limited data and lack of diversity in the traditional automatic encoder, an improved strategy for generating samples for transformer faults is proposed. Secondly, considering that the traditional convolutional neural network will lose a lot of feature information in the pooling operation, the improved convolutional neural network (ICNN) is constructed as the classifier of fault diagnosis. Finally, the effectiveness and adaptability of the proposed method are verified by the actual data. The simulation results show that IAE can take into account the distribution and diversity of data at the same time, and the generated transformer fault data can improve the performance of the classifier better than the traditional augmentation methods such random over-sampling method, synthetic minority over-sampling technique, and auto-encoder. Compared with traditional classifiers, ICNN has higher fault diagnosis accuracy before and after data augmentation.