Extreme weather events may simultaneously cause computing-node failures, power-supply degradation, communication-link disruptions, and critical-task migration restrictions in data centers, making conventional scheduling strategies based on stable infrastructure assumptions insufficient for post-disaster service continuity. Meanwhile, post-disaster system states are often scattered across weather reports, operation alarms, power-supply logs, communication-failure messages, emergency rules, and critical-task descriptions, rather than being directly available as unified numerical records. Transforming such heterogeneous and partially unstructured engineering information into computable compute-power constraints is therefore a prerequisite for reliable post-disaster resilience enhancement.To address this problem, this paper proposes an LLM-assisted neighbor-aware dynamic graph reconfiguration framework, termed LA-NADGR. The proposed framework first employs a large language model (LLM) with a schema-constrained extraction mechanism to transform multi-source post-disaster information into structured engineering records, including node states, edge states, task requirements, disaster impacts, and evidence confidence. These records are then mapped into a compute-power collaborative constraint graph that jointly represents computing resources, power margins, communication reachability, migration relations, and task-specific constraints. Based on this graph, the deterministic NADGR module performs neighbor-aware survival perception, task-aware action masking, bottleneck-based cooperation scoring, and cooperative-cluster generation to reconstruct a feasible cooperation topology for subsequent critical-task migration and distributed scheduling. The LLM does not directly generate control actions; instead, its extracted records are verified through deterministic compute, power, communication, and migration constraints, thereby reducing the risk of unsafe LLM outputs in critical infrastructure scenarios.Simulation results show that the proposed framework effectively mitigates pseudo-connectivity caused by communication-only recovery, enables the reconstructed topology to better reflect the true joint feasibility of critical tasks, and reduces the divergence between task deployment and successful execution. The response-time analysis further shows that the deterministic reconstruction core maintains low latency and favorable scalability as the network size increases. These results indicate that post-disaster compute-power resilience depends not only on upper-layer scheduling optimization but also on constructing a trustworthy, sparse, and constraint-grounded cooperation topology before scheduling.

As artificial intelligence models are increasingly applied in high-risk domains, such as power systems, improving the interpretability of the model has become essential. Recently, Kolmogorov-Arnold networks (KAN) have attracted attention due to their enhanced transparency through learnable activation functions and visualizable structures. However, KAN still lack explicit decision rule representations, which hinders their applications in high-risk tasks. To overcome this limitation, this paper proposes a hybrid method that integrates KAN with decision trees to extract key features and analyze the decision rule. Experiments show that decision trees can approximate KAN behavior and quantify the importance of features. Using the pruning mechanism of KAN, we further evaluated the impact of selected features on model performance. The results suggest that the proposed method retains predictive accuracy while enhancing interpretability, offering a promising approach to understand deep learning models in power systems.

Hybrid energy systems, integrating renewable energies, offer a sustainable and low-carbon solution for energy- intensive data centers, addressing the challenges posed by the variability of renewable sources and computational demands. This study proposes a stochastic optimization model of combined energy and computation scheduling of hybrid system and data center, in which a multi-energy storage system of electricity, hydrogen, natural gas, and heat is integrated to increase the flexibility and reliability of system. A scenario generation method for both renewable energy sources and computation loads is developed to characterize their uncertainties, which includes scenario identification, scenario sampling, and scenario generation and clustering steps. An optimization model, considering different objectives of operation cost, penalty cost of renewable power curtailment, and stepped carbon trade cost, is constructed to obtain the best energy and computation task coordinated scheduling strategies. A case study confirms the effectiveness of the proposed strategy for coordinating energy and computation scheduling in data centers. It compares the impact of key variables such as storage configurations and operational objectives. The strategy achieves a 15.26% cost reduction and a 10.79% carbon emission decrease versus traditional methods. Accounting for carbon allowances further cuts emissions by 59.04 %, albeit at a 7.67 % higher cost.

In the context of increasing randomness and complexity in power systems, conventional methods for transient stability assessment, which rely on mechanism modeling and simulation, are insufficient for current needs. Data-driven AI methodologies, characterized by their ability to facilitate intelligent state identification and control strategy generation through offline training and online applications, have garnered significant attention. To address these challenges, we employ Random Forests to build a data-driven model for power system transient assessment and classification. We then establish an interpretable analysis framework using SHAP (Shapley Additive explanations) within the Machine Learning model. This framework quantifies the marginal contributions of input features to model outputs, enhancing interpretability. By analyzing features from both global and local interpretability perspectives, we achieve comprehensive insights: global interpretability ranks feature importance, while local interpretability elucidates the contributions of each feature to the predictions of individual samples. This dual-perspective analysis facilitates timely control strategy formulation by dispatchers. Experimental results demonstrate that SHAP effectively clarifies the contributions of features to classification outcomes, providing a clear and interpretable understanding of model behavior.

The weak grid strength in regions with large-scale renewable energy integration has emerged as a universal challenge, limiting the further expansion of renewable energy development. Currently, the short-circuit ratio (SCR) is widely used to quantify the relative strength between AC systems and renewable energy. To address this issue, this study first analyzes and compares how different reactive power compensation methods enhance the SCR. It then proposes calculation frameworks for both the SCR and critical short-circuit ratio (CSCR) in renewable energy grid-connected systems integrated with reactive power compensation. Furthermore, based on these formulations, a quantitative evaluation methodology for voltage support strength is developed to systematically assess the improvement effects of various compensation approaches on grid strength. Finally, case studies verify that reactive power compensation provided by synchronous condensers effectively strengthens grid strength and facilitates the safe expansion of the renewable energy integration scale.

The operation of new energy base grid-connected systems faces issues closely related to voltage stability, especially under massive operation modes, where traditional voltage stability analysis methods are too time-consuming to meet practical needs. The short-circuit ratio, as an important parameter for measuring the strength of voltage support and stability margin, has limited applicability for systems without synchronous machine power sources. This study extends the concept of short-circuit ratio to fully power-electronic systems. Firstly, based on the voltage support strength index of new energy grid -connected system via VSC-HVDC, it quantitatively analyzes the impact of the shortcircuit ratio of new energy grid-connected system via VSC-HVDC on voltage fluctuations. Secondly, the trajectory sensitivity method is used to quantitatively analyze the impact of different parameters, both operational and flexible direct current control parameters, on the short-circuit ratio, and to identify the most sensitive parameters as key parameters. Finally, through case studies, the effectiveness and practicality of this method in identifying voltage vulnerability points arc confirmed.

With the increasing complexity of power system operations, traditional fault detection methods face challenges in providing interpretability. This paper introduces a novel unsupervised generative model based on Kernel Density Estimation (KDE) to analyze and interpret state estimation in power systems. By modeling fault data with KDE, the model learns the true data distribution and generates samples that mirror the original data distribution. Experimental results demonstrate that the KDE-based model clearly distinguishes stable and unstable states in the latent space, with the generated data preserving the original class proportions (difference $<1 {\%}$). Compared with traditional models, it provides explicit distribution visualization, enabling interpretable analysis rather than relying on opaque decision boundaries.

Abstract In order to achieve fast and accurate transient stability analysis and emergency control, this paper proposes a transient stability emergency control method based on improved deep reinforcement learning. In order to fully explore the temporal and spatial variation trend of transient response, a multi-dimensional feature containing information such as transient situation energy is constructed, and the deep reinforcement learning model is transformed based on the time-space graph neural network. On this basis, an emergency control model is constructed, and the power grid knowledge is integrated into the emergency control decision-making scheme to reduce the exploration of invalid decision-making and improve the performance of the model. The effectiveness of the proposed method is verified in the IEEE-39 system.

With the continuous development and growth of new energy grid integration, the phenomenon of islanding occurs frequently in the new energy grid integration system, which poses a significant threat to the safety and stable operation of the power grid. Islanding detection is an essential means of determining whether an islanding phenomenon has occurred in the new energy grid integration system. This paper first explains the principle and harm of islanding, and introduces the main islanding detection methods currently available, analyzing their advantages and disadvantages. Finally, based on the current status of islanding detection development, this paper proposes the problems that need to be addressed and the future development trends to promote further development in this research field.

With the continuous increase in the proportion of new energy sources in the new power system, the number of distributed power sources is also gradually increasing, leading to a growing possibility of power islanding in the power system. Power islanding refers to the temporary disconnection of a local independent power system from the main grid. The existence of power islands can have an impact on the stability of the power system. Therefore, the detection and optimal partitioning of power islands in the system to minimize their impact during faults, as well as the control of the internal operation of the islands, will become a future research focus. This article will provide a review of the existing technologies for the partitioning, monitoring, and operation control of power islands, and comprehensively analyze and compare the advantages and disadvantages of various techniques, laying the foundation for further research and development of power islands in the new power system.

The interaction between the controllers of doubly-fed induction generators (DFIGs) and series-compensated lines may cause subsynchronous oscillation (SSO). In an analysis of the cause of SSO, it is found that the proportional integral (PI) controller of the current loop of the rotor-side converter (RSC) plays a leading role and amplifies the disturbance. Accordingly, this paper proposes a novel mitigation method that replaces the PI controller of the current loop of the RSC with active disturbance rejection controller (ADRC). The ADRC method can estimate and compensate for change in the rotor current caused by the rotor subsynchronous current in real time, maintaining a constant rotor current, eliminating the effect caused by the disturbance and effectively mitigating the SSO phenomenon. Finally, a simulation model based on a real wind farm is built in PSCAD/EMTDC, where the proposed mitigation method is tested under different operating conditions to ensure the system stability. The results show that the effectiveness of the proposed mitigation method and the ADRC is superior to the PI controller in terms of adaptability to changeable operating conditions.

In response to the problem of unstable interaction in two-stage photovoltaic power generation systems, a mathematical model of the two-stage photovoltaic system was first established, and then a mathematical model of a capacitor converter based on Buck/Boost converter was established to simulate the capacitor. Subsequently, from the perspective of harmonic suppression, the working principles of suppressing oscillations of capacitor converters under parallel adaptation were analyzed, and the effectiveness of capacitor converters was verified through simulation.

This research introduces an algorithm for cooperative management in interconnected multi-regional energy networks, integrating rapid Q-learning with simulated annealing techniques to achieve enhanced automatic power control. The former enhances the convergence speed of the agent through an accelerated learning rate, while the latter addresses local optimization challenges using the principles of simulated annealing. Moreover, this algorithm is designed with a thorough consideration of the characteristics of renewable energy, aiming to enhance the efficiency and sustainability of energy management. Simulation results based on a multi-region load frequency model of a power grid indicate that this approach demonstrates superior performance in achieving optimal cooperative control across multiple regions, offering enhanced control capabilities and faster convergence compared to other reinforcement learning methods, thus providing robust support for large-scale integration of renewable energy sources.

The state identification of the operational status of power systems plays a crucial role in ensuring the safety, efficiency, and reliable operation of power systems, forming the foundation for the management and control of power system operations. In the practical operation of power systems, the duration of fault states is short, while the system remains in a safe and stable state for a significantly longer period, leading to an imbalance in the training samples available. This imbalance can cause traditional machine learning classification algorithms to favor the majority categories of power system operational states, resulting in inaccurate evaluation metrics and reduced model generalization capability. To address this issue, this paper introduces a spatiotemporal dynamic graph neural network model for the identification of power system operational states. By incorporating class cost coefficients, focal loss, and cross-entropy in a joint loss function, the model adaptively focuses on the less represented operational state categories during training, improving classification accuracy on imbalanced and challenging samples. Moreover, the integration of temporal information and spatial topology information within the graph neural network enhances the model’s robustness and generalization ability in state identification. Experimental results on the standard IEEE 68-bus system demonstrate that, compared to traditional machine learning methods, the proposed joint loss with the spatiotemporal dynamic graph neural network model achieves higher accuracy in the identification of power system operational states, significantly improving classification performance on imbalanced samples with a 4.8% increase in F1-score, reaching a peak F1-score of 98.6%.

Renewable energy sources are increasingly integrated into power systems, leading to significant variability in operations. This necessitates robust methods for assessing operational reliability. We propose a novel model–data hybrid approach that incorporates endogenous uncertainty into the reliability evaluation process. First, unlike traditional methods that treat uncertainties as external factors, this approach recognizes that operational decisions can significantly influence how uncertainties are resolved and impact reliability metrics. The proposed method integrates device reliability indices with operational decision variables. This allows us to evaluate the impact of endogenous uncertainty on operational reliability through a reliability-constrained stochastic unit commitment model. Additionally, a model–data hybrid algorithm is introduced for efficient solution of the formulated optimization problem. Case studies demonstrate the effectiveness of the proposed method. Results also show that endogenous uncertainty may cause a 10% error in power system reliability indices.

High-percentage new energy access brings serious challenging for frequency stability in power system, and also presents new demands on the safe and economic operation of the power system. Aiming to improve the inertial situational awareness capability of power systems in the case of high percentage of new energy access. Enhance the frequency adaptation capability of the power system to active disturbances. And to quantify the operation boundary of the system. Based on the power system frequently response model, this paper analyzes the frequently response process of the power grid in different time scales after being subjected to active perturbation. And the mathematical relationship of the system quality response is constructed. Through the study of the whole process of frequency response, this paper proposes a minimum inertia evaluation method for high proportion new energy power system based on improved particle swarm algorithm, and uses particle swarm algorithm as the solution algorithm. Finally, simulation tests based on the IEEE-30 node model validate the effectiveness and the precision of the proposed method.

With the rapid development of distributed power generation technology and microgrid technology, research on the operation and control of new energy storage isolated network systems has received widespread attention. Compared with grid-connected operation, isolated operation can improve the acceptance and application of new energy, increase the flexibility of power grid operation, and solve the problem of difficulty in long-distance transmission in remote areas, which is an important application form and development trend of future new energy. This article reviews the latest progress in the operation and control of new energy storage isolated network systems. The key technologies for the operation and control are analyzed, general methods for the operation and control strategy of are given, and future technology development trends are predicted and discussed.

With the increasing proportion of renewable energy, the power system inertia decreases, and the operation uncertainty rises. It brings concerns about the system frequency and operational reliability. However, the impacts of the power system frequency performance on the reliability parameters of generation units have not been fully investigated. This paper studies the frequency performance and the operational reliability co-evaluation for power systems considering wind turbines. Firstly, a power system frequency regulation model is established considering the regulation capability of wind turbines. Then, the cluster of equivalent wind turbines is incorporated into the frequency regulation architecture of thermal power units, which accelerates the analysis of frequency performance. Then, the frequency performance of the power system with the participation of wind turbines under the operation uncertainty and the unit random faults is quantitatively analyzed. A frequency-dependent generator reliability parameter model is derived. Next, a multi-time scale co-evaluation framework is proposed to realize the co-evaluation of frequency performance and operational reliability. Case studies are carried out on the modified IEEE RTS-79 system and a provincial power system. Results show that compared with the existing research, the proposed method can obtain the frequency performance and reliability results efficiently.