Grid forming converters can provide frequency and voltage support to the power systems and are enabling units in modern power systems with high levels of power electronics and renewable energy penetration. However, grid forming converters may experience synchronization instability following large disturbances. During synchronization instability, the system state variable is captured by a periodic orbit. This article presents a novel approach to eliminating the synchronization instability of grid forming converters by removing the periodic orbit responsible for synchronization instability. First, an oscillation index is proposed to estimate the distance to the homoclinic bifurcation, and the periodical orbit sensitivity is derived to assess the sensitivity of the oscillation index to control parameters. Then, the oscillation index and its sensitivity are used to estimate the parameter adjustment required to trigger the homoclinic bifurcation and eliminate synchronization instability. An application framework of the proposed method is also provided. Finally, the proposed method and the application framework are validated through simulations and experimental verifications.

Converters can provide reactive power support for the power system and offer advantages such as fast response and flexible control. This paper aims at enhancing power system security by utilizing the reactive power capacity of converters. A voltage security margin index is introduced and an optimization model is developed to improve the voltage security margin index by optimizing the converter control parameters. The effectiveness of the proposed model in enhancing voltage security is validated through case studies on a 39-node system.

Recent climate change has led to more frequent and widespread extreme weather events, causing damage to power system components and resulting in significant outages and losses. Component fragility curves and sequential Monte Carlo simulations have been widely used for power system outage risk assessment. Fragility curves serve as the bridge linking weather events to component failures, while sequential Monte Carlo methods capture the temporal dynamics of weather and balance probabilities with consequences. However, a direct and improper combination of these two tools can introduce significant bias, especially under high time resolution. This phenomenon is referred to as the “failure probability curse.” This paper analyzes the fundamental cause of this issue and proposes a conditional-probability correction method for the sequential Monte Carlo simulation. The method can eliminate the inherent bias from high-time-resolution sampling in power outage risk analysis and reconcile sequential Monte Carlo with different physical failure mechanisms. Furthermore, the conditional-probability perspective provided in this paper offers support and explanation for related existing methods at the level of mathematical principles, enhancing the insights of the “failure probability curse” from both an explanation and solution perspective. Numerical experiments based on an illustrative system and the Tennessee power network confirm the validity of the proposed approach.

The Automatic Generation Control (AGC) system, reliant on real-time measurements over communication networks, is susceptible to stealthy false data injection attacks (FDIAs), risking equipment damage and economic losses. We propose a robust FDIA detection method using maximum likelihood estimation (MLE) of a drifted multivariate Ornstein-Uhlenbeck (OU) process. Independent of load observability, in various cyberattack scenarios, the proposed FDIA detection method delivers accurate and rapid detection of sophisticated FDIAs, outperforming traditional unknown input observer (UIO) methods, which miss detections, and Long Short-Term Memory Autoencoder (LSTM-AE) approaches, which suffer from prolonged detection times.

This paper proposes a novel prediction-free two-stage coordinated dispatch framework for the real-time dispatch of grid-connected microgrid with generalized energy storages (GES). The proposed framework explicitly addresses grid awareness, non-anticipativity constraints, and the time-coupling characteristics of GES, providing microgrid operators with a near-optimal, reliable, and adaptable dispatch tool. In the offline stage, we generate the hindsight state-of-charge (SoC) trajectories of GES by solving the multi-period economic dispatch with historical scenarios. Subsequently, leveraging this historical information (SoC trajectories, net loads, and electricity prices), we synthesize and dynamically update online References for both SoC and opportunity cost through kernel regression. We propose an adaptive Lagrange multiplier-based online convex optimization algorithm, which innovatively incorporates reference tracking for global vision and expert-tracking for step-size updates. We provide theoretical proof to show that the proposed OCO algorithm achieves a sublinear bound of both dynamic regret and time-varying hard constraint violation. Numerical studies using ground-truth data from the Australian Energy Market Operator demonstrate that the proposed method outperforms state-of-the-art methods, reducing operational costs by 5.0-6.2% and voltage violations by 0.8-9.1%. These improvements mainly result from mitigating myopia by reference tracking and the adaptive capability provided by dynamically updated references and adaptive Lagrange multipliers. Sensitivity analysis demonstrates the robustness, computational efficiency, and scalability of the proposed method.

The interaction between extreme weather events and interdependent critical infrastructure systems involves complex spatiotemporal dynamics. Multi-type emergency decisions within energy-transportation infrastructures significantly influence system performance throughout the extreme weather process. A comprehensive assessment of these factors faces challenges in model complexity, heterogeneous differences between energy and transportation systems, and cross-sector privacy. This paper proposes a risk assessment framework that integrates the heterogeneous energy and transportation systems in the form of a unified network flow model, which enables full accommodation of multiple types of energy-transportation emergency decisions while capturing the compound spatiotemporal impacts of extreme weather on both systems simultaneously. Based on this framework, a targeted method for identifying system vulnerabilities is further developed. This method employs neural network surrogates to achieve privacy protection and accelerated identification while maintaining consideration of system interdependencies. Numerical experiments demonstrate that the proposed framework and method can reveal the risk levels faced by urban infrastructure systems, identify vulnerabilities that should be prioritized for reinforcement, and strike a balance between accuracy and speed.

Reasonable carbon emission allocation is the crucial cornerstone of the low-carbon development of power systems. However, electricity transactions change the economic characteristics of electricity carbon emission responsibility and pose great challenges to the present allocation models. This paper fills the research gap by proposing a brand-new model to allocate responsibility among generation companies, load aggregators, and grid companies simultaneously, with the consideration of Peer-to-Peer trading and physical power flow distribution. First, the trading power shift distribution factor is introduced to analyze the directions of line flow pieces corresponding to each transaction. Then, the allocation model is established based on optimal flow tracing and the trading power shift distribution factor, where the power flow section is divided into several portions corresponding to electricity trades, and the power, together with the responsibility within each portion, can be traced to generators through optimal flow tracing. The responsibility allocated to each entity comprises the parts related to trading power and active losses. The model is further reformulated as a Nash bargaining and transformed with rotated second-order cones and the binary expansion method. Case studies demonstrate that the proposed model can manifest the impacts of different forms of trades on the responsibility allocation intuitively. The contrast between the proposed model and the carbon emission flow method indicates the outstanding advantages in physical rationality and in reducing emissions based on transaction characteristics. Eventually, the scalability is verified on a provincial power system.

Artificial Intelligence (AI) has been extensively integrated into smart grids to enhance anomaly detection, with time-series AI detectors such as Long Short-Term Memory Autoencoders (LSTM-AE) and Convolutional Autoencoders (Conv- AE) emerging as pivotal technologies for handling sequential data and intricate temporal dependencies. However, this integration exposes critical infrastructure to novel security vulnerabilities, particularly during the model training phase. This paper investigates an underexplored threat: black-box training-data poisoning attacks on time-series AI detectors in power systems. Unlike gradient-based test-time evasion attacks such as the Fast Gradient Sign Method and Carlini & Wagner, our attack is formulated as a black-box training-data corruption framework that poisons archived load records before detector training. We introduce a physics-informed attack framework that leverages Stochastic Differential Equations (SDEs) to generate physically plausible poisoned load records. The methodology encompasses empirical SDE parameter extraction, template-based synthetic load generation with daily drift compensation, and its integration into an Automatic Generation Control (AGC) system to systematically contaminate the training dataset. Simulation results reveal that time-series AI detectors trained on this tainted data exhibit profound detection failures, misclassifying severe anomalies—such as system frequency drops below 49.5 Hz—as normal operations, thereby underscoring a critical vulnerability in AI-enabled power systems and the urgent need for defenses against such physics-constrained adversarial threats.

Although deep reinforcement learning has demonstrated significant potential in load frequency control, existing methods often exhibit insufficient robustness when facing extreme or worst-case disturbances that lie outside the standard training data. To address this issue, this article proposes an adversarial training framework based on Hamiltonian optimal control. Unlike traditional data augmentation strategies that rely on random noise, this method mathematically derives adversarial load profiles—including sustained frequency drops and oscillatory disturbances designed to induce system resonance—by designing specific Lagrangians for the objective functionals and solving the resulting Hamiltonian canonical equations. We integrate these targeted adversarial samples into the training loops of deep deterministic policy gradient (DDPG), twin delayed DDPG, and proximal policy optimization algorithms to enhance the agents’ adaptability to extreme operating conditions. Simulation results on two-area and extended three-area power system models demonstrate that the proposed method accurately identifies the system’s critical resonant frequencies. Compared to baseline controllers trained solely on random noise, the agents trained with Hamiltonian adversarial loads exhibit significantly superior frequency stability and damping characteristics when subjected to structured, time-varying extreme disturbances, effectively preventing catastrophic system instability.

ABSTRACT The rapid growth of LLM‐driven artificial intelligence (AI) and data center deployment is imposing a new paradigm of massive and highly dynamic power demand on the grid. Volatility in both computing and renewable energy resources introduces severe challenges to power system operation. Although typical computing workloads are technically flexible, their responsive capacity remains largely untapped in real‐world scenarios. This paper argues that the main barriers to scalable computing–electricity coordination are not solely technical, but arise from weak electricity cost pass‐through, limited cost visibility, and misaligned compute‐side incentives. In leasing and cloud‐service modes, electricity costs of data centres are often bundled into broader service charges, reducing their influence on workload scheduling. Meanwhile, compute‐side actors prioritise GPU utilisation, return on investment, latency, throughput, and service quality over flexible load profiles supporting power‐system operation. Therefore, green operational constraints and new pricing mechanisms that expose carbon or electricity signals to compute‐side decision‐makers are critical to making computing–electricity coordination scalable in practice. Flexible computing loads can support power systems only when technical flexibility is matched with actionable economic, contractual, and operational incentives.

The coordinated dispatch of electric and district heating system (DHS) shows a typical master-slave mode. Therein, the heterogeneous decomposition (HGD) method achieves preferable convergence by exchanging power and price information of coupled combined heat and power (CHP) units between the power system and heating system, but also introduces privacy disclosure risks. This paper proposes a privacy-preserving HGD method for combined heat and power dispatch (CHPD) using the Paillier cryptosystem. First, a privacy-preserving primal-dual interior point method (PDIPM), namely Paillier-PDIPM, is proposed to solve general-form quadratic programming (QP) with encrypted parameters. Then, the key management service (KMS) is introduced to enhance PDIPM for two-agent collaborative optimization. Finally, the distributed Paillier-HGD method is proposed based on KMS and Paillier-PDIPM, where exchanged data is encrypted, and sub-models are solved through privacy-preserving computation. Case studies in test systems of different scales show that the Paillier-HGD method can achieve privacy preservation for CHPD without compromising the accuracy of results and support collaborative optimization among heterogeneous energy systems with high efficiency.

Convention studies on the large disturbance power system frequency stability focus on the center of inertia frequency. However, the frequency drop at certain nodes may be much severer than the frequency drop of the center of inertia frequency after a large disturbance in low-inertia systems. This paper investigates the nodal frequency response characteristics of low-inertia power system to optimize the allocation of grid-forming converters. An optimization model is established based on the dynamic system model to improve the nodal frequency response characteristics under multiple contingencies. The proposed model is validated in a 39-bus system.

In the integrated electricity and gas system (IEGS), common failures in the gas system may lead to significant pressure decrease of gas-fired power generators, which causes fluctuations or cascading failures in the power system. In this paper, an applicable early warning and proactive control framework based on the dynamic equivalence of gas transmission networks is proposed to mitigate the impact of such incidents. Firstly, different time scales of the dynamics involved in the gas-electric cascading failures are discussed to get a suitable analysis model of the proposed framework. Secondly, a dynamic equivalent model of the gas network oriented towards the coupling nodes of the IEGS is constructed based on the linearized form of gas transmission equations. Finally, a proactive control method for the power system considering electromechanical transient processes based on the iterated equivalence parameters of the gas network is introduced to minimize the loss of the cascading failure. Case studies demonstrate that the framework proposed in this paper can effectively alleviate the impact of gas failures on the power system and possesses strong scalability.

Due to the intrinsic complexity of time series forecasting within power systems, artificial intelligence has emerged as a promising pathway for predictive analytics. Although time series data from power systems inherently exhibit consistent characteristics across spatial, temporal, and covariate dimensions, current forecasting methods remain scenario-specific and fail to develop unified models. This limitation stems from incomplete utilization of multi-dimensional features, resulting in suboptimal data exploitation. To this end, this paper initially constructs a unified tensor representation for time series data of power systems, encapsulating various forecasting tasks within an integrated forecasting architecture. Following this, a unified model with multi-dimensional attention structure for time series forecasting (UniMATS) is proposed, which is capable of integrated feature extraction across all data dimensions simultaneously. Evaluated on three public datasets of wind power, photovoltaic, and load, UniMATS achieves RMSE reductions of 9.8%, 6.9%, and 14.6% respectively compared to the second-best models in day-ahead forecasting tasks. The results of the case study underscore the superior performance and the efficient model structure of UniMATS over existing benchmarks, affirming its general capabilities for time series forecasting in power systems. The general predictive capability and the modular design of UniMATS enable the development of a foundational large model for time series forecasting in power systems based on UniMATS blocks.

With the increasing demand for diverse test data in power system research, generating power grid models that satisfy power flow constraints have become a critical challenge. In this paper, we propose a progressive fine-tuning approach for large language models (LLMs) for the power grid model generation task. The initial phase applies supervised fine-tuning to enhance structural consistency and task adaptability, while the subsequent enhancement phase employs reinforcement learning with an expert knowledge–guided evaluation mechanism to ensure compliance with AC power flow constraints. To support fine-tuning, we construct a systematic dialogue-based dataset that encompasses label model generation and high-quality question–answer pairs construction, providing a framework that can be extended to other power system applications. Numerical experiments on generating power grid models of varying scales demonstrate that the proposed method significantly improves format accuracy, power flow residuals, and convergence rates, while also exhibiting strong generalization capability to previously unseen grid sizes. Its practical utility is further validated through a downstream Optimal Power Flow (OPF) learning task, where LLM-generated data serve as a benchmark to help identify the model with superior generalization capabilities.

As an increasing number of distributed energy resources (DERs) are managed by virtual power plants (VPPs), intelligent adversaries could generate abnormal loads through compromised VPPs, inducing line overloads. Such attacks are referred to as VPP-based manipulation of demand (VMAD) attacks. Unlike existing studies that focus mainly on post-attack consequence mitigation, this paper proposes a coordinated cyber-physical defense framework for mitigating overloads induced by VMAD attacks. First, the features of VMAD attacks in the cyber domain and their resulting overloads in power systems are analyzed. Then, a cybersecurity neighborhood-watch mechanism for VPPs is proposed, which helps dynamically quantify abnormal loads based on the security status of VPPs. Finally, a coordinated dispatch strategy is designed to enhance system resilience to such VMAD attacks, ensuring power flows on transmission lines remain within secure limits. Simulations on several IEEE test systems and a real-world system validate that the proposed coordinated framework can mitigate the overload risk of high-risk lines with lower overall costs, indicating its potential to support the defense against VMAD attacks.

The risks to cyber-physical power systems stem not only from external disturbances but also from internal anomalies. Traditional control methods are increasingly challenged under the challenges of new controlled entities and cyber risks. This paper proposes a learning-based resilience control method for LFC systems. By incorporating adversarial training methods, a generic control strategy is developed for the system under FDIA conditions. Additionally, based on switching system modeling, the stability of the switching system is integrated into the training process to ensure the stability of the controller. The method is validated through simulations of an LFC system with two types of regulation resources. The results demonstrate that the proposed method ensures frequency stability in the switching system.

Typhoon disasters pose serious threats to the secure operation of wind-integrated power systems by damaging transmission lines and perturbing wind power output. To address the typhoon-induced large-scale unplanned wind farm cut-outs, it is necessary to develop an orderly wind curtailment plan in day-ahead dispatch. However, balancing robustness and wind power utilization in such plans remains challenging. Furthermore, it is difficult for a single uncertain optimization framework to simultaneously account for the distinct uncertainties in both transmission contingency and wind power output, leading to strategies with limited practicality. To tackle these challenges, this paper proposes a day-ahead dispatch model incorporating an orderly wind curtailment strategy, using adaptive robust stochastic optimization (ARSO) to handle the dual uncertainties of line failure and wind generation. The model consists of pre-dispatch and re-dispatch stages. The curtailment plan is determined in the form of curtailment thresholds in pre-dispatch stage and implemented in re-dispatch stage. In addition, this paper designs a wind power sample generation method and develops a solution algorithm for the ARSO model based on the column-and-constraint generation (C&CG) approach. Case studies on modified IEEE 39-bus and 118-bus systems verify the necessity of orderly wind curtailment and the effectiveness of the proposed method.

Traditional data-driven models have limited capabilities to describe topological relations, leading to difficulties in short-term voltage stability (STVS) assessment with strong locality. For the real-time dynamic security analysis (DSA), a novel STVS assessment method based on the heterogeneous edge-integrated graph attention network is proposed. Considering various credible contingencies, the STVS quantitative indicators of buses are obtained, avoiding the time-consuming problem of time-domain simulation in the conventional DSA. First, the mechanism similarity between the STVS and message passing-based graph neural network is analyzed. A virtual homomorphism technique and multi-layer perceptron are introduced to handle the original heterogeneous input features. Then, to focus on the nonlinear impact of transmission lines on dynamic voltage interactions, an edge feature integration method is designed for feature aggregation. The physical processes of STVS in the system under line contingencies can be effectively reflected. Finally, case studies verify the superiority of the proposed method in terms of both accuracy and its generalization ability to new topologies. To understand the mechanism of the model, a post hoc interpretability analysis is conducted based on the attention weight and quasi-steady state sensitivity at the node and feature levels, respectively.

The optimal dispatch of integrated electricity and heating systems (IEHS) becomes a computationally challenging non-convex problem when considering bilinear coupling between mass flow rates and temperatures in district heating networks. This paper proposes a learning-accelerated optimization framework that synergizes a spatio-temporal graph neural network with sequential linear programming (SLP) to achieve fast and high-quality solutions. The approach employs Graph Attention Networks (GAT) and bidirectional Gated Recurrent Units (biGRU) to learn tighter relaxation bounds by capturing both spatial network topology and temporal dependencies. Quantile regression techniques are then used to predict feasible intervals for pipe flow rates. The solution of this problem initializes an SLP algorithm that efficiently recovers a near-optimal and feasible solution for the original nonlinear problem. Case studies on the P39H89 and P118H223 systems demonstrate that the proposed method achieves a near-optimal solution with negligible violations while significantly reducing the computation time. The results validate the framework's ability to enable computationally efficient and scalable dispatch for large-scale integrated energy systems.