Lirong Gao, Zeqing Wang, Yuyan Cai, Jiayi Deng, Yanmei Gu, Yiming Zhang, Jia Zhou, Yanfei Zhang, Junbo Zhao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

This letter develops a high-risk regional load attack mechanism in smart grids with incomplete network information. Different from previous research, this attack mechanism enables an attacker to launch a regional load attack with network information about an arbitrary attack region, while minimising the deviation in corrupted data to enhance the stealth of this attack. Such an attack only corrupts a limited number of loads while still overloading multiple lines within a targeted attack region, resulting in significant impacts on smart grids. Case studies conducted on two modified IEEE test systems validate the effectiveness of our proposed attack mechanism and pave the foundation for the future development of practical defensive strategies.

Although machine learning models have been widely used in electricity theft detection, most of them lack interpretability, which hinders user trust and policy enforcement. To this end, this paper aims to investigate the interpretability of machine learning models in electricity theft detection. Specifically, a comprehensive theoretical analysis is conducted to reveal why the interpretability is needed in electricity theft detection. Then, a model-agnostic explainable artificial intelligence (XAI) framework is proposed to uncover the potential start and end times of fraudulent behavior, and to clarify the rationale behind identifying fraudulent users within machine learning models by calculating the importance score of each data point. Simulation results demonstrate that the XAI framework provides class-discriminative data points to interpret fraudulent activities, enabling suspicious users to understand why the machine learning model identified them as suspicious and guiding model improvement. Moreover, compared with benchmarks (e.g., Shapley additive explanations, local interpretable model-agnostic explanations, and gradient-weighted class activation mapping techniques), the harmonic mean of overlap and coverage (HMOC) of the proposed XAI framework is improved by 10.26% to 54.73%, indicating more trustworthy interpretations.

Given the availability of more comprehensive measurement data in modern power systems, reinforcement learning (RL) has gained significant interest in operation and control. Conventional RL relies on trial-and-error interactions with the environment and reward feedback, which often leads to exploring unsafe operating regions and executing unsafe actions, especially when deployed in real-world power systems. To address these challenges, safe RL has been proposed to optimize operational objectives while ensuring safety constraints are met, keeping actions and states within safe regions throughout both training and deployment. Rather than relying solely on manually designed penalty terms for unsafe actions, as is common in conventional RL, safe RL methods reviewed here primarily leverage advanced and proactive mechanisms. These include techniques such as Lagrangian relaxation, safety layers, and theoretical guarantees like Lyapunov functions to rigorously enforce safety boundaries. This article provides a comprehensive review of safe RL methods and their applications across various power system operations and control domains, including security control, real-time operation, operational planning, and emerging areas. It summarizes existing safe RL techniques, evaluates their performance, analyzes suitable deployment scenarios, and examines algorithm benchmarks and application environments. This article also highlights real-world implementation cases and identifies critical challenges such as scalability in large-scale systems and robustness under uncertainty, providing potential solutions and outlining future directions to advance the reliable integration and deployment of safe RL in modern power systems.

Reliance on wind technology as a renewable energy source keeps increasing worldwide. This integration into the grid changes its characteristics and that necessitates accurate system modeling that requires parameter identification. The existing parameter identification methods are often obstructed by the presence of nonlinearities in the power system models, leading to a degradation in the prediction and computation. The fluid community proposed dynamic mode decomposition (DMD) as an effective system identification technique. The DMD prevails in several science applications, including power systems. The algorithm is enhanced to account for dynamical actuation data, which is referred to as dynamic mode decomposition with Control (DMDc). The DMDc provides superior accuracy in system identification. In this article, we propose the DMDc to identify the system parameters of a wind farm system, including an asynchronous machine, a rotor side controller, a dc link, and a grid-side controller. Further, we compare the DMDc with classical parameter identification methods: Prediction error method and the similarity matrix technique. Our results demonstrate the superiority of the proposed method in identifying the wind farm parameters with high accuracy and efficiency.

The intermittent and fluctuating output of wind turbines is increasingly recognized as a major issue affecting the power quality and stability of electrical grids. As wind power integration grows, addressing this challenge is essential. A promising solution involves using a Hybrid Energy Storage System (HESS), combining battery energy storage systems (BESS) and supercapacitors (SC). This paper presents a Q-learning based control strategy for tuning the parameters of a two-stage variable time constant low-pass filter (LPF) in a grid-connected microgrid. The proposed strategy adaptively adjusts the LPF time constant to mitigate wind power fluctuations. It also accounts for practical constraints of energy storage systems and their interfaced converters, such as preventing overcharge/discharge and adhering to maximum power conversion limits. Numerical simulations confirm the effectiveness of the two-stage variable time constant LPF in reducing output wind power fluctuations while considering the practical constraints of HESS.

This paper presents a dual-strategy approach to optimizing charging and discharging schedules for school bus fleets, using the limited charging infrastructure effectively. We aim to ensure that each bus is fully charged for daily operations and aids in grid stability during peak demand. The first strategy utilizes linear programming to schedule overnight charging at available station sockets and strategic discharging during peak periods, efficiently coordinating limited resources. The second strategy employs metaheuristic techniques for continuous optimization, focusing on precise power requirements and offering greater flexibility than the linear model.

Probabilistic analysis tool is important to quantify the impacts of the uncertainties on power system operations. However, the repetitive calculations of power flow are time-consuming. To address this issue, data-driven approaches are proposed but they are not robust to the uncertain injections and varying topology. This article proposes a model-driven graph convolution neural network (MD-GCN) for power flow calculation with high-computational efficiency and good robustness to topology changes. Compared with the basic graph convolution neural network (GCN), the construction of MD-GCN considers the physical connection relationships among different nodes. This is achieved by embedding the linearized power flow model into the layer-wise propagation. Such a structure enhances the interpretability of the network forward propagation. To ensure that enough features are extracted in MD-GCN, a new input feature construction method with multiple neighborhood aggregations and a global pooling layer are developed. This allows us to integrate both global features and neighborhood features, yielding the complete features representation of the system-wide impacts on every single node. Numerical results on the IEEE 30-bus, 57-bus, 118-bus, and 1354-bus systems demonstrate that the proposed method achieves much better performance as compared to other approaches in the presence of uncertain power injections and system topology.

Real-time topology estimation in distribution grid with high penetration of distributed energy resources remains a challenging task due to the insufficient high-precision measurements and frequent topology variations. This article proposes a real-time distribution system topology estimation approach building on the graph theory and Bayesian networks with sparse measurements. The graph theory develops the topology graph bank to effectively leverage the prior knowledge of topology models, including the topology structure and the switching relationship between different topologies. This allows the development of the Bayesian networks for topology tracking using real-time voltage and power injection measurements. A novel discrete method considering the similarity of data correlation information is proposed for the optimal placement ofμPMUs to ensure the performance of topology estimation. Numerical results on the IEEE 33-node and 123-node systems show that the BN-based topology estimation model has better performance against incomplete information, i.e., missing data, than other alternatives.

Due to the rapid growth of distribution systems in urban areas, the increasing complexity of these distribution systems brings challenges to accurate topology identification. The collected voltage data from smart meters have been proven effective in topology identification applications. However, when multiple feeders are connected to the same bus, the accuracy of existing voltage-correlation-based topology identification can be degraded significantly. To address this challenge, a comprehensive inference method is proposed in this article to identify theon/offswitch state in the distribution system considering the same-bus-different-feeder condition. The voltage-power-dependence principle among connected nodes is revealed. Based on this theory, a physical probabilistic network model is proposed to represent the causal relationships between the switch states and the voltage-power dependence in a distribution network. The belief propagation algorithm is introduced to deduce the topology identification model, which can reduce the time consumption of the inference. The performance of the proposed method and its advantage over the existing methods are verified in case studies.

The abnormal events, such as the unprecedented COVID-19 pandemic, can significantly change the load behaviors, leading to huge challenges for traditional short-term forecasting methods. This article proposes a robust deep Gaussian processes (DGP)-based probabilistic load forecasting method using a limited number of data. Since the proposed method only requires a limited number of training samples for load forecasting, it allows us to deal with extreme scenarios that cause short-term load behavior changes. In particular, the load forecasting at the beginning of abnormal event is cast as a regression problem with limited training samples and solved by double stochastic variational inference DGP. The mobility data are also utilized to deal with the uncertainties and pattern changes and enhance the flexibility of the forecasting model. The proposed method can quantify the uncertainties of load forecasting outcomes, which would be essential under uncertain inputs. Extensive comparison results with other state-of-the-art point and probabilistic forecasting methods show that our proposed approach can achieve high forecasting accuracies with only a limited number of data while maintaining the excellent performance of capturing the forecasting uncertainties.

Power system state estimation (PSSE) is the foundation of energy management system applications. Hence, operators impose stringent requirements on PSSE data integrity. False data injection attacks (FDIAs) can cause risks to PSSE data-driven operations and demand mitigation. In this article, we present a two-step FDIA detector design. In step one, we study a novel stealthy attack policy by simultaneously considering the attacker’s cost reduction and damage production. In step two, with the aid of a deep autoencoding Gaussian mixture model (DAGMM), we design an unsupervised detection scheme to detect the stealthy attack. The DAGMM-based detector can meet the requirement of rapidity, unsupervisedness, and data imbalance tolerance. Eventually, we simulate and validate the stealthy attack policy and the corresponding detector using the benchmark IEEE 39-bus and 118-bus systems.

The progression of modern computing technologies assists the development of cyber-physical systems, which are transforming the legacy electrical power systems into smarter ones. The informationalization of the grid poses potential vulnerabilities concerning cyberattacks. With dynamic variations over time, cyberattacks can cause significant impacts on the secondary frequency control with various attack scenarios. In this article, by divulging the characteristics of dynamic attacks, the stability and dynamic responses of secondary frequency control systems are analyzed. The complete attack models considering dynamic load altering attack and dynamic false data injection attack are both derived first. Then the system stability is evaluated with different attack models through mathematical analysis. Eventually, the simulation studies against two benchmark power system models validate the evaluation results.

Data quality issues exist in practical phasor measurement units (PMUs) due to communication errors or signal interferences. As a result, the performances of existing data-driven disturbance classification methods can be significantly affected. In this article, a fast disturbance classification method that is robust to PMU data quality issues is proposed. The impacts of bad PMU measurements on disturbance classification are investigated by analyzing the feature distributions of deep learning methods. A new feature extraction scheme that uses the univariate temporal convolutional denoising autoencoder (UTCN-DAE) is proposed. It allows encoding and decoding univariate disturbance data through a temporal convolutional network to capture the temporal feature representation and is robust to bad data. Based on the features of the frequency and voltage measurements encoded by the UTCN-DAE, a two-stream enhanced network, i.e., the multivariable temporal convolutional denoising network is proposed to achieve optimal feature extraction of multivariate time series by feature fusion. The classification is performed using a multilayered deep neural network and Softmax classifier. Extensive results obtained on the IEEE 39-bus system as well as a large-scale power system in China with field PMU measurements show that the proposed method achieves the highest classification accuracy and computational efficiency as compared to other deep learning algorithms.

The widely used transient stability-constrained optimal power flow (TSC-OPF) method for power system preventive control is very time-consuming and thus not applicable for large-scale systems. This article proposes a new deep learning-enabled surrogate model that can significantly improve computational efficiency while maintaining high accuracy. To achieve that, the deep belief network (DBN) is strategically integrated with the reference-point-based nondominated sorting genetic algorithm (NSGA-III) to develop a new preventive control framework. The DBN allows us to identify the mapping relationship between the transient stability index and system operational features. The identified functional mapping relationship is further used as the surrogate to connect the DBN results with TSC-OPF for preventive control. The integrated NSGA-III and surrogate model enable the multiobjective optimization to consider various constraints and objectives, such as minimization of costs of generation dispatch cost and load shedding while maintaining the system stability. Extensive simulation results on several IEEE test systems show that the proposed method can achieve highly efficient control solutions and outperform other alternatives in terms of computational efficiency and economic benefits.

With the advancement of communication technologies and the development of the smart grid, today's physical power systems present an ever-growing dependency on cyber resources. It increases cyber vulnerabilities, causing safety and system stability concerns. In this article, a data-driven resilient automatic generation control (AGC) scheme is proposed under a false data injection attack (FDIA). The key idea is to identify the relationship between AGC signals and system operational conditions. This is achieved by the proposed regression-based FDIA signal predictions, including sequence-to-point prediction and the long short-term memory network-based prediction. It allows us to reconstruct the AGC control signals without being affected by FDIAs and to attenuate attacks in the closed control loop, thus alleviating the impact of FDIA on system performance. Numerical results carried out on the benchmark systems validate the effectiveness of the proposed method.

The short-term load forecasting is crucial in the power system operation and control. However, due to its nonstationary and complicated random features, an accurate forecast of the load behavior is challenging. An improved short-term load forecasting method is proposed in this article. At first, the load is decomposed into different frequency components varying from the low to high levels realized by the ensemble empirical-mode decomposition algorithm. Then, the smooth and periodic low-frequency components are predicted by the multivariable linear regression method while maintaining the efficient computation capacity, while the high-frequency components with strong randomness are forecasted by the long short-term memory neural network algorithms. Thus, the actual load behavior is obtained by combining these two methods. Finally, the proposed method is validated by experiments, in which the tested data from the west area of China, Uzbekistan, and PJM Interconnection (USA) are used. The prediction of the load behavior is accurate globally along with the local details, as presented in the experiments, which verify the effectiveness of the proposed method.

A probabilistic model for the predicted power sources is typically assumed for the existing uncertain optimal power flow (UOPF). However, obtaining accurate information of that is very challenging in practice due to the limited available data of renewable energy and loads with complex correlations. To address that, this article proposes a belief function based framework for UOPF with a large number of uncertain power sources. Multiple imperfect models with q-least committed joint basic belief density are developed to characterize the knowledge deficient power sources. This yields the integration of the generalized Bayesian theorem with the traditional evidence theory into a unified manner. The former allows estimating the uncertain model of the predicted power sources, whereas the latter is to obtain the probability box of the UOPF variables. Comparison results with the Monte Carlo simulations and the three-point estimation approach show that the proposed method is able to get accurate UOPF results while achieving high computational efficiency for large-scale systems with a large number of knowledge deficient power sources.

In this letter, a two-stage robust unscented unbiased minimum-variance (RU-UMV) estimator is proposed for nonlinear system dynamic state estimation with unknown inputs. In the first stage, by leveraging the statistical linerization and the relationship between unknown input vector and states, we derive a batch-mode regression form. It is shown that the application of weighted least squares for this form yields the same results as the UMV unscented Kalman filter. However, it lacks robustness to outliers. To deal with, robust generalized maximum-likelihood (GM)-estimator together with the projection statistics (PS) is developed, yielding robust state estimates. The latter are further used in the second stage for robust unknown input vector estimation. As a result, both innovation and observation/measurement outliers can be effectively suppressed. Illustrative examples are provided to demonstrate the robustness of the proposed method.

This letter proposes a new polynomial-chaos-based Kalman filter (PCKF) that is able to track the dynamics of nonlinear dynamical systems subject to strong nonlinearities. Specifically, by resorting to the polynomial chaos theory, the uncertainties of the model and the measurements can be effectively propagated through a set of collocation points. However, this polynomial-chaos-based algorithm suffers from the curse of dimensionality. To overcome this weakness, a dimension reduction strategy is proposed based on variance analysis. This allows us to construct more effective collocations points and to significantly improve the computational efficiency of the PCKF without any loss of estimation accuracy. Simulations carried out on various IEEE systems validate the effectiveness of the proposed method.

This letter proposes an unscented Kalman filter (UKF)-based unbiased minimum-variance estimation (UMV) method for the nonlinear system with unknown inputs. By utilizing the statistical linearization, the nonlinear system and measurement functions are transformed into a “linear-like” regression form. The latter preserves the nonlinearity of the system and the measurement models. To this end, the unknown inputs can be estimated by the weighted least-squares. This “linear-like” regression form also allows us to resort to the UMV state estimation framework for the development of new nonlinear filter to handle unknown inputs. Specifically, two approaches have been developed: 1) given the estimated inputs, we derive a filter by minimizing the trace of the state error covariance matrix; 2) without input estimation, we derive the filter by minimizing the trace of the state error covariance matrix subject to a constraint imposed on the gain matrix. We prove that these two approaches provide the same results. Numerical results validate the effectiveness of the proposed method.

This paper analyzes a malicious data attack in which the attacker targets at the generation side of the system and aims to compromise the system security by causing large power imbalance in the real-time operations. Such an attack is called the ramp-induced data (RID) attack in the literature which revealed that the attacker can manipulate the limits of generator ramp constraints in real-time dispatch (RTD) and, thus, impact the power market operations. In this paper, we propose an optimal attack model to analyze the impact of the RID attack on power system operational security and show that the attacker can introduce large power imbalance into real-time operations that can cause security issues or even catastrophic consequences. To address the risk of such an attack, a countermeasure is presented that reassesses the regulation reserve adequacy against a given risk level of the attack. Simulations on the IEEE 118-bus system verify the impact of the proposed RID attack and the effectiveness of the regulation assessment approach.

State estimation is a fundamental function in modern energy management system, but its results may be vulnerable to false data injection attacks (FDIAs). FDIA is able to change the estimation results without being detected by the traditional bad data detection algorithms. In this paper, we propose an accurate and computational attractive approach for FDIA detection. We first rely on the low rank characteristic of the measurement matrix and the sparsity of the attack matrix to reformulate the FDIA detection as a matrix separation problem. Then, four algorithms that solve this problem are presented and compared, including the traditional augmented Lagrange multipliers (ALMs), double-noise-dual-problem (DNDP) ALM, the low rank matrix factorization, and the proposed new “Go Decomposition (GoDec).” Numerical simulation results show that our GoDec algorithm outperforms the other three alternatives and demonstrates a much higher computational efficiency. Furthermore, GoDec is shown to be able to handle measurement noise and applicable for large-scale attacks.

In practical applications like power systems, the distribution of the measurement noise is usually unknown and frequently deviates from the assumed Gaussian model, yielding outliers. Under these conditions, the performances of the existing state estimators that rely on Gaussian assumption can deteriorate significantly. In addition, the sampling rates of measurements from supervisory control and data acquisition (SCADA) system and phasor measurement unit (PMU) are quite different, causing time skewness problem. In this paper, we propose a robust state estimation framework to address the unknown non-Gaussian noise and the measurement time skewness issue. In the framework, robust Mahalanbis distances are proposed to detect system abnormalities and assign appropriate weights to each chosen buffered PMU measurements. Those weights are further utilized by the Schweppe-type Huber generalized maximum-likelihood (SHGM) estimator to filter out non-Gaussian PMU measurement noise and help suppress outliers. In the meantime, the SHGM estimator is used to handle unknown noise of the received SCADA measurements, yielding another set of state estimates. We show that the state estimates provided by the SHGM estimator follow an asymptotical Gaussian distribution. This nice property allows us to obtain the optimal state estimates by resorting to the data fusion theory for the fusion of the estimation results from two independent SHGM estimators. Extensive simulation results carried out on the IEEE 14, 30 and 118-bus test systems demonstrate the effectiveness and robustness of the proposed method.

To enhance the robustness of a power system state estimator to topology errors, bad critical measurements, multiple non-interacting, or interacting bad data (BD), this paper presents a new robust detection method by exploiting the temporal correlation and the statistical consistency of measurements. Particularly, we propose three innovation matrices to capture the measurement correlation and statistical consistency by processing the forecasted states/measurements and the interpolated reliable information from phasor measurement units. The latter is achieved by using a robust generalized maximum-likelihood estimator. We then propose to apply the projection statistics (PS) to the proposed innovation matrices for BD detection. Extensive Monte Carlo simulations and QQ-plots are carried out to obtain an analytical threshold of the statistical test of the PS. Because of the robustness of PS and the enhanced measurement redundancy by the innovations, the proposed method is able to handle various types of BD in both PMU observable and PMU partially observable power systems. Moreover, the proposed method is suitable for parallel implementation, and can be integrated with online applications. Comparison results with existing methods under different BD conditions on IEEE 14-bus, 118-bus, and Polish 2383-bus test systems demonstrate the effectiveness and robustness of the proposed method.

Most existing connectivity-based localization algorithms require high node density which is unavailable in many large-scale sparse mobile networks. By analyzing large datasets of real user traces from Dartmouth and MIT, we observe that user mobility exhibits high spatiotemporal regularity and, more importantly, that user mobility is strongly correlated with the user's social encounters (including so called Familiar Strangers). Motivated by these important observations, we propose a distributed localization scheme called SOMA that is particularly suitable for sparse mobile networks. To exploit the correlation between mobility and social encounters, we formulate the localization process as an optimization problem with the objective of maximizing the probability of visiting a sequence of locations when the user witnesses the given social encounters at different time. Employing the Hidden Markov Model (HMM), we design an efficient algorithm based on dynamic programming for solving the optimization problem. SOMA is fully distributed, in which each user only makes use of the connectivity information with other users. Experimental results based on large-scale real traces demonstrate that SOMA achieves much smaller localization error than many state-of-the-art localization schemes, but requires minimal running time.