Prolonged high-temperature operation causes thermal aging of DC cable insulation, which accelerates electrical tree growth during short-circuit faults and poses a critical threat to power system reliability. This paper systematically investigates the initiation and propagation characteristics of grounded DC electrical trees in thermally aged XLPE cable insulation, along with the corresponding changes in aging performance. Based on experimental results, prediction methods for grounded DC electrical tree behaviors are developed using a CNN-SVM model and a GM(1,N) model. The results reveal a non-monotonic variation in tree initiation probability P, propagation rate p1, and expansion ratio DL with increasing aging time, which is closely linked to microstructural changes in aged XLPE, particularly recrystallization. To predict tree behaviors, two data-driven models are established: a CNN-SVM and a GM(1,N) model. Both achieve high predictive accuracy, with relative errors (δtest and σtest) below 5% for all validation samples. These findings establish a quantitative relationship between thermal aging characteristics and grounded DC electrical tree evolution, and provide a novel data-driven method for predicting electrical tree behaviors from aging performance indicators.

Owing to the complex operation environment in the ocean, the external structure of submarine XLPE cable is vulnerable to damage, enabling seawater infiltration and subsequent irreversible degradation of the cable's insulation material. This study primarily seeks to clarify the effect of seawater corrosion on the aging performance of XLPE cable insulation. To this end, accelerated seawater corrosion samples are systematically examined through several aging evaluation methods, including SEM, FTIR, DSC, and AC electrical breakdown tests. Further, microscopic structural changes within the XLPE cable insulation during seawater corrosion are analyzed through molecular simulations. Ultimately, based on the above research results, the mechanism by which seawater corrosion contributes to the deterioration of cable insulation is elucidated and then a preliminary study on material modification is conducted. Test results indicate that there is a non-monotonic evolution of insulation performance with exposure time, characterized by an initial improvement followed by progressive deterioration. Increasing seawater concentration accelerates degradation but exhibits a clear saturation effect beyond a threshold. Furthermore, nanofiller incorporation, particularly Al2O3, significantly enhances insulation performance by reducing mass loss and suppressing corrosive species diffusion, which can provide some help in future research to improve the long-term durability of XLPE cable in the marine condition.

Traditional model-based participation factor calculations in power systems often suffer from parameter inaccuracies. While data-driven Koopman methods offer an alternative, they face the curse of dimensionality due to the difficult selection of observables, drastically restricting their scalability, especially when incorporating algebraic variables. To address these challenges, we propose a unified, deep-learning-enhanced Koopman operator framework. First, we introduce an augmented Koopman operator capable of computing participation factors for both state and algebraic variables. Second, we integrate this operator into an autoencoder to automatically learn optimal observable functions, ensuring scalability for large-scale systems. Illustrative examples on the IEEE 9-Bus and the IEEE 118-Bus modified test systems are carried out to demonstrate the method’s effectiveness and verify its ability to guide stability enhancement.

Estimating electric vehicle (EV) Origin-Destination (OD) demands based on collected traffic sensor data, along with strategically placing traffic sensors, are crucial for effective management in Smart Grid and Intelligent Transportation Systems. This paper proposes a pioneering approach that integrates equilibrium-based EV OD demand estimation and traffic sensor placement into a holistic framework. The proposed framework adopts tri-level structure to address the optimal placement of EV traffic sensors at the upper level, followed by a bilevel model for OD demand estimation through inverse optimization of the traffic assignment problem under equilibrium conditions. To overcome the inherent complexity of the tri-level framework, a deep learning-based solution approach is proposed. Deep reinforcement learning is employed to determine the optimal sensor placement, maximizing OD demand estimation accuracy within a predefined budget. Additionally, a novel encoder-decoder architecture with transformer networks approximates the solution to the bilevel inverse optimization problem. Leveraging multi-head self-attention mechanisms, transformer networks capture intricate relationships within collected EV traffic data. Case studies demonstrate the effectiveness of the proposed framework and the superiority of transformer networks. Results validate the accurate estimation of EV OD demands and highlight the transformative potential of deep learning in addressing traffic sensor placement and OD demand estimation challenges.

Traditional methods for power system analysis and control are facing more and more challenges with the increasing penetration of systems containing VSCs (voltage source converters, e.g., high voltage DC transmission systems, PV or wind farms). Considering the difference between VSCs and synchronous generators, researchers have proposed a control scheme named virtual synchronous generator (VSG) to emulate the inertial characteristics of a real synchronous machine (SM) to tackle the problem of frequency stability. When disturbances occur in the system, VSGs may exhibit oscillatory behaviors in the same way as real SMs. At the same time, it also interacts with SMs due to power exchange as well as frequency and angle synchronization. Under this circumstance, how the existence of VSGs influences subsynchronous resonance (SSR) becomes an important issue. Based on the equivalence theorem, this article focuses on the impact on the SSR of a VSG interconnected with a multi-mass SM. After a VSG is equivalently integrated into the SM, it is evident that there are changes in the equivalent inertia and armature impedances, both of which affect the behaviors of the oscillation as a result.

The rapid proliferation of electric vehicles (EVs) presents substantial challenges to effective charging and discharging regulation within distribution networks, particularly under increasingly diverse and dynamic regulatory scenarios. Conventional reinforcement learning (RL) methods, typically limited by task-specific training, struggle to generalize effectively across varying regulatory objectives and heterogeneous spatial EV distributions. To address these limitations, this study proposes a novel offline meta-reinforcement learning (Meta-RL) framework designed to enhance the adaptability of policy networks through a dedicated meta-network for task-specific identification. Specifically, we propose an innovative meta-network enhanced by large language models (LLMs), which simultaneously captures dynamic physical trajectories derived from historical operational data and encodes expert dispatch knowledge extracted from textual instructions. By integrating these physical and semantic modalities, the meta-network effectively generates task-specific embeddings, enabling the downstream policy network to rapidly adapt and formulate context-aware regulatory strategies. Comparative experiments demonstrate that the proposed offline Meta-RL framework achieves a 47.2% average reduction in operation cost compared to conventional single-task RL methods. Ablation studies further confirm that integrating physical and semantic information significantly improves task recognition accuracy and decision-making quality, reducing adaptation time by 68.8% and average operation costs by 18.4%.

Widespread Internet of Things (IoT) devices not only enable deep cyber-physical integration in power distribution systems but also expose grids to stealthy localized false data injection attacks (FDIAs) tampering with measurement data, undermining situational awareness and threatening grid operation. The high penetration of distributed energy resources (DERs) further exacerbates this issue by introducing significant operational uncertainty and volatility, which can be exploited by attackers to mask malicious activities and complicate data recovery. Existing centralized FDIA-mitigation data reconstruction methods suffer from high overhead, single-point failures, and poor robustness when dealing with DER-induced fluctuations. Driven by this motivation, this article proposes a multiarea data reconstruction (MADR) framework to enhance the cyber resilience of IoT-enabled power distribution systems against FDIA, explicitly considering the challenges posed by DER integration. First, a community discovery algorithm partitions the large-scale distribution system into multiple subareas with rational electrical coupling. Then, a novel stealthy FDIA model considering attacker pReferences and subarea topological/operational characteristics is designed for highly targeted concealed attacks. A compressed sensing-based MADR method with adaptive Huber loss, which dynamically switches error-handling functions, is developed to boost attack robustness and effectively distinguish between genuine DER fluctuations and false data. An iteratively reweighted least square (IRLS) strategy with local information interaction enables parallel subarea data recovery and efficient global reconstruction. Extensive tests on the modified IEEE 123-node system with high DER penetration verify the framework outperforms state-of-the-art methods in feasibility and effectiveness.

Deep Reinforcement Learning (DRL)-based charging strategies become increasingly prevalent for smart grid operations, motivating exact lower bounds on their worst-case performance. However, conventional sampling-based testing provides no formal guarantee on the minimum return attained by a return-maximizing policy over mixed discrete-continuous state spaces with bounded uncertain exogenous variables. This paper presents an abstract-domain formal verification framework in which the admissible initial set is represented as a mixed-integer abstract element, while the DRL policy and environment dynamics are modeled as abstract transformers with bound propagation and structured handling of the discrete time-to-leave state. Under exact encoding and certified global solution within the declared solver tolerance, unrolling these transformers yields a mixed-integer nonlinear program that certifies the exact lower bound. Case studies show that the framework identifies guaranteed EV-charging lower bounds missed by extensive sampling and provides complementary benchmark comparisons with representative formal-verification methods.

Residential loads, with their substantial scale, rapid response speed, and flexible controllability, have become a crucial resource for demand side management. However, privacy concerns arising from data communication and the complexity of response strategies due to variations in customer flexibility present significant challenges to the effectiveness of demand response (DR) programs. To address these issues, this paper proposes a load management framework based on a multi-cluster mean-field (MCMF) game. Firstly, customer flexibility is quantified based on historical power consumption data, and an improved kmeans algorithm is employed to cluster customers within the community. Then, considering each customer's optimization objective to minimize the cost function including the electricity cost and the discomfort level, the problem is formulated as an MCMF game. Customers adjust their power consumption strategies according to the group-specific estimated electricity price, while the load aggregator (LA) collects total power consumption values and updates the price information iteratively until the optimal strategies of all customers converge to an epsilon-Nash equilibrium (epsilon-NE). Case studies involving 2000 customers with heterogeneous flexibility are conducted, and the results demonstrate the effectiveness and advantages of the proposed framework compared with existing methods in peak shaving, electricity cost reduction, and computational efficiency.

High penetration of distributed generation (DG) in active distribution networks drives frequent topology reconfiguration, posing stringent requirements for reliable dynamic topology tracking. However, existing methods are limited by missing snapshot measurements, inherent data distribution shifts between simulated and field data, and label scarcity in real-world datasets. This paper proposes a robust dynamic topology tracking framework for active distribution networks to address these limitations. Missing measurements are recovered via node voltage profile similarity, with probabilistic scenarios constructed to capture DG output uncertainty. A Bayesian nonparametric inference-based domain-adaptive automatic labeling strategy is developed to generate high-confidence pseudo-labels via KL divergence minimization for cross-domain distribution alignment. A physics-constrained graph convolutional neural network (GCN) is then established, which exploits the isomorphism between graph convolution operators and node admittance matrices to learn topology-aware features under physical constraints. Case studies on a modified PG&E 69-bus system and a modified IEEE 123-bus multi-microgrid distribution system verify that the proposed method significantly outperforms existing data-driven models in robustness and accuracy.

Online tracking of inertia in AC microgrids with high renewable integration is a challenging yet critical task. However, traditional methods typically rely on known large disturbances, which are infrequent in actual microgrid operations, thereby limiting the applicability of online inertia estimation. To address this issue, we propose a Bayesian online inertia estimation framework utilizing ambient measurements. This framework enables continuous inertia tracking without relying on specific disturbances, substantially alleviating the limitations of conventional approaches. Moreover, to prevent significant estimation bias caused by the high noise level in ambient rate of change of frequency (RoCoF) measurements, the proposed framework extracts inertia directly from frequency measurements. By evaluating the Fisher Information and deriving the Cramér-Rao Lower Bound (CRLB), we theoretically prove that this frequency-based approach inherently yields a substantially lower estimation error bound compared to the RoCoF-based methods. Furthermore, a modified importance sampling method is introduced to accurately characterize the posterior distribution of inertia. Simulation studies on an AC microgrid with two diesel generators and a wind turbine generator demonstrate that the proposed method can efficiently and accurately track system inertia online across ambient conditions.

The rapid adoption of electric vehicles (EVs) is driving increased charging demand, intensifying the interdependence between power distribution and transportation networks (PTNs). The main challenge in coordinating the EV routing and charging management within the PTN lies in the intractable nonlinear relationship between electricity price signals in the power distribution network (PN) and EV charging and routing behaviors in the transportation network (TN). Therefore, this paper proposes a novel constraint learning approach using XGBoost to solve the nontrivial bilevel optimization problem of vehicle charging within the PTN. XGBoost is first employed to capture the complex relationship between charging prices and optimal charging demand, leveraging its strong approximation capability to construct a tree-based surrogate model that approximates the TN decision model. Then the trained surrogate model is equivalently reformulated as mixed−integer linear programming (MILP) constraints and integrated into the vehicle charging optimization framework. Shapley values are further used to analyze the explainability of the tree-based surrogate model. Based on the constraint learning, a data-driven hybrid optimization model is developed for vehicle charging. Case studies validate the approximation accuracy and explainability of the tree-based surrogate model, as well as the effectiveness of the optimal strategy within the PTN.

The increasing penetration of renewable generation introduces significant uncertainty into short-term power system operations, making cost-effective dispatch under network constraints more challenging. In conventional predict–then–optimize pipelines, uncertainty bounds are learned upstream and treated as fixed inputs to robust optimal power flow (OPF), which prevents operational feedback from improving the uncertainty representation. This paper aims to bridge multiple probabilistic prediction models and robust optimization by proposing a robust semi-end-to-end coordinated decision model for power system prescriptive dispatch. Probabilistic prediction models are used for intraday renewable-generation and load forecasting to estimate upper and lower bounds of one-step-ahead power output, and a robust OPF model is formulated based on the interval uncertainty set induced by the probabilistic forecasts of renewable generation and load. Building on the forecasting and optimization components, we develop a robust semi-end-to-end framework that couples upstream probabilistic prediction with downstream robust OPF via a machine learning layer, a robust decision-making layer, and a decision evaluation layer. In the forward pass, the learned uncertainty bounds are mapped to robust dispatch decisions; in the backward pass, gradients are back-propagated through the optimization layer to train the probabilistic prediction models using task-level feedback. Numerical experiments demonstrate stable training and consistent economic gains over a predict–then–optimize baseline, achieving average total-cost reductions of about 14.3% and 16.0% on the IEEE 5- and 30-bus systems, while the improvement on the larger IEEE 118-bus system is more modest because the uncertainty-related cost components there represent only a small fraction of the total operating cost. Meanwhile, under the adopted single-interval DC-based dispatch setting, the learned intervals are tightened (approximately 25% narrower) while reducing slack activation under stressed network and reserve limits.

Belief propagation (BP)-based algorithms, typified by their efficient probabilistic inference abilities, are widely used for distributed reasoning in complex systems. However, BP has seldom been studied by the existing literature on unbalanced power distribution networks (PDNs), where state estimation (SE) is faced with exacerbated cybersecurity threats posed by malicious adversaries. To this end, the Gaussian BP (GaBP) with superior efficiency and interpretability is employed in this paper, making it the first attempt to propose a GaBP-based detection approach for unbalanced PDNs against various cyber-attacks. In the proposed approach, the nonlinear SE model is emulated by factor graphs and iteratively solved by GaBP, which is further enhanced using message and factor damping techniques to improve algorithm convergence. By analyzing the belief update procedure in a local factor graph, the measurement update term (MUT) corresponding to the local measurement data is designated as the preliminary attack detection variable. To make the attack-induced temporal shape distortion more perceptible, the proposed approach defines the multivariate derivative dynamic time warping (DDTW) distance between the baseline MUT and the latest MUT sequences as the test statistic. Finally, a generalized likelihood ratio test (GLRT) is accordingly designed to detect potential cyber-attacks. Extensive case studies on two modified IEEE test feeders validate the efficacy and superiority of the proposed detection approach.

The non-stationary response of distributed energy storage (DES) users, primarily driven by participation fatigue, poses a significant challenge to the reliability and economic efficiency of aggregator-based grid services. This paper proposes a comprehensive aggregation control and dispatch framework to address this challenge. A fatigue-aware behavioral model is developed, employing online Maximum Likelihood Estimation (MLE) to infer the latent dynamics of user fatigue—a critical non-stationary factor overlooked in prior models. This model is then integrated into a CUCB-MLE-based distributed energy storage aggregation control (DESAC) strategy, where the upper confidence bound is structurally corrected by both the inferred fatigue state and the physical state of charge (SoC) to ensure adaptive and feasible user selection. Finally, the strategy is embedded within a feedback-corrected multi-timescale economic scheduling framework, designed to mitigate execution bias by incorporating realized aggregation power into rolling corrections. Simulation results on a modified IEEE 30-bus system validate the proposed method’s effectiveness. Compared with the CUCB-based benchmark, the proposed approach reduces the relative power adjustment mismatch from 34.66% to 4.63%, decreases the wind power curtailment rate by 68.45%, and lowers the total operating cost by 7.08%, demonstrating its effectiveness in improving aggregation reliability and system economics.

The growing integration of hydrogen energy systems with electrical power grids introduces considerable operational complexities, particularly in relation to the physical limitations of hydrogen transmission such as pipeline congestion. To address these challenges, this study develops an optimal scheduling framework for an Integrated Electricity-Hydrogen System (IEHS) that, for the first time, incorporates hydrogen truck transportation as a flexible and dispatchable resource for the proactive management of pipeline congestion. The inherently complex and nonconvex scheduling problem is reformulated as a tractable Mixed-Integer Second-Order Cone Programming (MISOCP) model. In addition, a relax-and-repair warm start algorithm is proposed to enhance the computational tractability of this large-scale optimization problem. Results indicate that the truck coordination strategy reduces the peak hydrogen flow in congested pipelines to 30%-50% of constrained levels and facilitates a more cost-effective re-dispatch across the entire system. Furthermore, the proposed relax-and-repair warm start algorithm reduces the total computation time by 18% to 35% in various scenarios, without compromising the optimality of the solution. These findings demonstrate that the proposed model and algorithm offer a robust and efficient approach to achieve efficient congestion management and stronger operational security of the IEHS.

Electric vertical takeoff and landing aircraft (eVTOLs) are expected to play an important role in future urban air mobility (UAM) and introduce high-power charging loads to urban power systems. This paper reviews eVTOL charging load modeling and forecasting methods, with particular emphasis on distribution network impacts, grid planning, and coordinated operation. First, the charging load formation mechanism of a single eVTOL is analyzed from the perspectives of flight energy consumption and charging strategies, with emphasis on the key factors affecting charging demand. Second, the operational uncertainty and charging scheduling methods of aggregated eVTOL charging loads at vertiports are discussed, followed by an analysis of urban-scale spatiotemporal charging load modeling and forecasting. Finally, the paper summarizes the impacts of high-power eVTOL charging loads on distribution networks, and discusses infrastructure planning, network control methods, and coordinated flexibility management. This paper summarizes current research progress and, by comparing relevant methods developed for EV charging, identifies future research directions for mission-driven charging load modeling and the coordinated operation of power grids and three-dimensional transportation systems.

Utilizing numerous distributed energy resources (DERs) for frequency regulation is a promising pathway to enhance the security and economy of future power systems. However, a fundamental challenge arises from the emergent collective behavior of these resources: their aggregated response, intended to support macro-level system frequency, can jeopardize micro-level local voltage security within distribution networks. To this end, this paper proposes a unified hierarchical framework based on the principle of synergistic co-design. The framework employs a stage-dependent architecture that systematically integrates distinct control strategies for different regulation stages. For security-critical response, it utilizes a novel stage-aware adaptive virtual synchronous generator (VSG) control method with stability formally proven by Lyapunov analysis, while for economic efficiency, it transitions to a distributed control method for cost-optimal restoration. The entire process is underpinned by a proactive voltage security management strategy that enforces micro-level safety. To enable efficient analysis of these complex dynamics, a dual time-scale information interaction mechanism is designed to accelerate the transmission-distribution (T&D) co-simulation process. Simulations on a hybrid transmission-distribution test system validate the framework's effectiveness, demonstrating that the proposed approach enables scalable, secure, and economical frequency support while effectively managing the emergent voltage challenges from collective DER control.