Accurate electricity load forecasting is essential for the reliable operation of modern energy systems, particularly under increasing variability from renewable integration and environmental factors. This paper proposes a hybrid forecasting framework that combines Long Short-Term Memory (LSTM) networks with a GPT-assisted adaptive optimization strategy for short-term load prediction. An initial LSTM model is constructed to capture temporal dependencies in multivariate time-series data, including temperature and humidity. A feedback-driven refinement process is then introduced, in which GPT is utilized as a heuristic optimization component to iteratively adjust model configurations based on observed training behavior and prediction errors. A simulation-based case study is conducted using a dataset designed to reflect realistic load patterns. The results demonstrate consistent improvements across multiple evaluation metrics and show that the proposed method outperforms baseline and benchmark models. The findings suggest that integrating adaptive optimization mechanisms can enhance forecasting accuracy and robustness, while highlighting the potential of large language models as auxiliary components in time-series prediction tasks.

The inherent second-order power oscillation of the single-phase AC-DC-AC voltage regulator under wide input voltage conditions will cause significant ripple in the DC bus voltage, especially when the load changes suddenly. The traditional synchronous method based on phase-locked loop (PLL) only focuses on phase tracking and does not consider the direct impact of the synchronous timing on the DC bus ripple. This paper reveals that the timing mismatch between the grid voltage and the zero-crossing point of the inverter output voltage is the key factor determining the amplitude of the DC bus ripple, clarifies the intrinsic mechanism of power-time coupling, and establishes the analytical relationship between second-order power mismatch, timing deviation and bus ripple. Based on this, a power-time coupling adaptive synchronization method is proposed, which takes the synchronization timing as a controllable variable, extracts the second-order voltage ripple of the DC bus and introduces a slow adaptive compensation loop. The timing deviation at the zero-crossing point is adjusted without changing the internal structure of the PLL. Through the experimental prototype of the single-phase AC-DC-AC voltage regulator, it is verified that the proposed method can effectively suppress the DC bus voltage ripple under wide input voltage and dynamic load switching. Compared with the traditional PLL synchronization and fixed timing compensation methods, it has better steady-state and transient performance, and maintains a stable output voltage, demonstrating engineering practicability.

The integration of variable wind and photovoltaic generation with proton exchange membrane (PEM) water electrolysis is a key enabler for green hydrogen production in smart multi-energy systems, but it exposes coupled multi-physics dynamics, conflicting operational objectives, and strict safety constraints. This paper proposes a hierarchical data-model hybrid scheduling framework for grid-connected PEM green hydrogen production. At the modeling layer, a physics-informed representation with six governing equations in the main text and twenty-two supporting equations in Appendix A defines explicit "red-line" constraints for electrochemical efficiency, thermal safety, and equipment degradation. At the coordination layer, a high-level module-designed with the aid of a large language model during a 90 day design phase-maps heterogeneous operational context (renewable forecasts, electricity prices, stack health indicators) into adaptive multi-objective preference weights and operating bounds. At the execution layer, Takagi-Sugeno fuzzy controllers provide real-time set-point tracking for multiple PEM units. Numerical studies using European wind, photovoltaic, and day-ahead price data over a 7 day test period demonstrate that the proposed framework achieves similar to 4.9% power tracking root mean square error [compared with 7.8% for proportional-integral-derivative (PID) control], thermal safety margins approximately 174% larger than conventional baselines, and roughly 11.8% faster fault recovery. These results indicate that the proposed physics-informed, data-model hybrid architecture effectively supports the integration of green hydrogen into smart multi-energy systems.

This paper investigates coordinated day-ahead scheduling of an island micro water-energy nexus (MWEN) under multi-source uncertainties, where low-probability but high-impact composite extreme conditions may lead to severe tail risks in system operating costs. To explicitly address this issue, a two-stage stochastic scheduling framework is developed with an objective function formulated based on conditional value at risk (CVaR), enabling systematic control of tail-risk exposure beyond conventional expected-cost minimization. A hierarchical scenario processing pipeline that integrates extreme-scenario retention with k-medoids clustering is further proposed to preserve risk-relevant characteristics while maintaining computational tractability. The MWEN model leverages water storage and water-treatment/desalination flexibility as cross-domain buffering resources to enhance system resilience under adverse conditions. Comparative case studies demonstrate that deterministic scheduling can result in pronounced cost deterioration under uncertainty, with the worst-case daily operating cost reaching $3401.78. In contrast, the proposed strategy significantly suppresses tail risk: at a 95% confidence level, value-at-risk (VaR) is reduced from $2642.22 to $2593.81 and CVaR from $2735.10 to $2684.22. These results highlight the importance of explicitly modeling extreme risks in MWEN scheduling and provide practical decision support for resilient island microgrids with high renewable penetration.

The interaction uncertainty of Distributed Energy Resources (DERs) during grid dispatch poses a challenge to the efficient and economical construction of Virtual Power Plants (VPPs). To address this, this paper proposes a dynamic construction strategy for VPPs that considers the interaction uncertainty of DERs and the response recovery process. First, the interaction effects of DER power variations are quantified through Seasonal Trend Decomposition (STL) and cross-correlation analysis, and an linkage matrix is constructed to depict the uncertainty. Then, a comprehensive evaluation method is proposed, which not only accounts for the direct economic cost of response scheduling but also innovatively introduces the optimization of the post-scheduling response recovery process, using Model Predictive Control (MPC) to minimize the total recovery cost. On this basis, a dynamic construction strategy using a Generative Adversarial Network (GAN)-Copula function is proposed. For resources with significant interaction, a joint probability distribution model is established to accurately predict the power interaction impacts during the scheduling process, thereby dynamically adjusting the final aggregation set and improving its economy. Simulation results based on data from 247 charging stations in Shenzhen and 8,497 communication base stations in Hefei indicate that, compared to existing methods, this approach can reduce scheduling response costs by approximately 12.71%.

This paper presents a low-voltage management device equipped with an upper-layer natural language instruction parsing module. The device is applied to the low-voltage governance of rural power grids and enables intelligent interaction between natural language commands and devices. The front-end adopts a voltage-multiplying PFC (Power Factor Correction) topology, and achieves power factor correction and stable DC output through voltage-current dual-loop control; the back-end uses a half-bridge inverter circuit combined with LC filtering and quasi-proportional resonant control to output standard 220 V AC voltage. A novel BERT-based natural language instruction parsing model (BERT-NLIP) is proposed to parse natural-language commands and convert them into structured parameter settings for the embedded controller; a synchronous phase-locked technology is adopted to coordinate the phases of the PFC and the inverter, thereby suppressing the double-frequency ripple on the DC bus; a multi-frequency quasi-resonant controller is introduced to specifically suppress odd-order harmonics. A supervisory safety mechanism and a goal-oriented automatic scheduling strategy are further introduced to ensure reliable command execution and support practical line-loss-reduction control. The paper presents the detailed main circuit structure and the design process of control loop parameters. Simulation and experimental results verify that the command parsing accuracy is high, the output voltage is stabilized at 220 V, and the total harmonic distortion rates under empty-load and full-load conditions are as low as 2.5% and 2.1% respectively.

Large-scale renewable energy bases increasingly employ automatic voltage control (AVC) to coordinate heterogeneous reactive-power resources. The resulting voltage regulation process inherently involves sampling, communication delay, and nonlinear device characteristics, which may induce nontraditional voltage oscillations and stability degradation that cannot be adequately captured by conventional continuous-time or small-signal analysis. This paper proposes a discrete-time nonlinear voltage stability analysis framework for renewable energy bases with multi-reactive-power-resource coupling under AVC-based coordinated control. The voltage regulation dynamics are formulated as a discrete-time nonlinear closed-loop system by incorporating sampled AVC actions, delayed voltage feedback, and nonlinear voltage–reactive-power coupling. An incremental system representation is constructed, and a strong-contraction-based stability criterion is derived using sector-bounded nonlinearity descriptions and linear matrix inequalities, providing a sufficient condition for global voltage convergence without local linearization. Extensive numerical studies are conducted on a representative renewable energy base with parallel and series coupling topologies. A total of 2916 randomized configurations are evaluated. The proposed criterion achieves consistency rates exceeding 96% for the parallel topology and 99% for the series topology when compared with time-domain simulations, while the probability of dangerous misjudgment remains below 1%. Scenario-based simulations further demonstrate that coupling topology plays a critical role in shaping voltage stability behaviors, and state-space analysis further supports the observed stability behaviors. These results indicate that nonlinear strong contraction offers an effective and practical stability notion for AVC-based voltage regulation in renewable energy bases.

Global climate change has led to frequent extreme weather events, exacerbating supply-demand imbalances and price volatility in power systems. Traditional risk management methods predominantly rely on statistical modeling and single financial instruments, making them inadequate for effectively addressing power shortages and electricity price risks under extreme scenarios. This paper proposes an intelligent financial hedging framework based on Large Language Models (LLMs) that synergistically applies insurance and futures portfolios to electricity markets under extreme weather conditions. First, we construct a CNN-LSTM-Attention prediction model that integrates physical mechanism constraints to achieve medium-to long-term forecasting of power shortages and uncertainty quantification. Second, we design an LLM-driven insurance-futures collaborative decision mechanism, which translates multi-source unstructured information (e.g., insurance clauses and market reports) into quantifiable strategy parameters through semantic understanding, thereby enabling adaptive optimization of insurance coverage and futures positions. Furthermore, we introduce a multi-objective Pareto optimization framework to achieve dynamic balance among risk coverage rate, cost-effectiveness, and return stability. Empirical results using the German electricity market as a case study demonstrate that the proposed framework achieves a 94.8% risk coverage rate under extreme high-temperature scenarios while significantly enhancing the economic efficiency and resilience of the strategy. The research indicates that LLMs can effectively bridge semantic risk identification with quantitative financial allocation, enhancing the intelligence of electricity risk hedging and providing a novel solution for market stability and energy transition under extreme weather.

With large-scale 5G deployment, communication base stations (CBSs) face increasing electricity demand, low backup-storage utilization, and growing demand from power systems for flexible regulation resources. This paper proposes a market-operation coordinated optimization framework for virtual power plant-oriented CBS aggregation. On the market side, a generative adversarial network (GAN) - distributionally robust optimization (DRO) robust bidding strategy is developed to learn the relationship among bid prices, clearing outcomes, and deliverable capabilities, and to construct calibrated perturbation sets. Shortage penalties and worst-case regret are jointly considered to generate robust bids under coupled uncertainty. On the operation side, a probability-driven dispatch strategy embeds historical triggering patterns into day-ahead and intra-day rolling optimization to maintain response readiness. When a dispatch instruction is triggered, response-recovery optimization and CBS selection are performed to minimize additional fulfillment cost. Simulations based on measured data from 7,880 CBSs in Hefei, China, show that the proposed strategy increases cumulative market revenue by 11.39% and reduces response cost by 49.98% compared with the baseline. The results demonstrate that the proposed method improves revenue robustness and reduces fulfillment cost under uncertain market conditions.

The 2025 blackout across the Iberian Peninsula marked a paradigm shift in the nature of power system failures, revealing how modern grids—dense with inverter-based resources and automated controls—can exhibit emergent instabilities that traditional design frameworks may not fully anticipate. This review examines the event through the lens of complex systems engineering, drawing on publicly available information and prior research to illustrate why a well-instrumented, high-renewables grid might still struggle to contain rapidly evolving disturbances. We begin by highlighting structural features—such as bottlenecks and asymmetric interconnections—that have been shown in the literature to influence the propagation of instability in transmission networks. Network-science tools, including algebraic connectivity, spectral radius and betweenness centrality, are discussed as methods used in previous studies to reveal latent fragilities. Attention then shifts to synchronization dynamics, where low-inertia conditions can amplify frequency deviations and interact with heterogeneous inverter response characteristics. Using models of cascading failures and percolation-type transitions, we outline how disruptions can propagate nonlocally, highlighting mechanisms identified in analytical and simulation studies. The analysis extends to fragility scoring systems, modular segmentation strategies and the spatial deployment of virtual inertia and fast-response resources—approaches proposed in prior research as part of a resilience toolkit aimed at containment and stabilization. Finally, we summarize emerging redesign directions emphasizing harmonized protection logic, regional operational cells and complexity-aware digital twins as increasingly important under conditions of uncertainty.

Most studies on grid-forming (GFM) converters have focused on two-level topologies, whereas three-level converters are more prevalent in industrial applications. Compared to two-level structures, three-level converters require additional neutral-point (NP) voltage balancing control. The strong coupling between the nonlinear dynamics of the NP voltage and the GFM control loops presents a critical challenge in the study of three-level GFM converters. Specifically: 1) the inherent NP voltage imbalance in three-level topologies leads to output voltage distortion, degrading the grid-support capability of GFM converters; and 2) the closed-loop voltage regulation in GFM control further exacerbates the voltage distortion caused by NP imbalance, creating an unstable feedback loop. To address this challenge, this article develops a linearized model of a three-level GFM converter that comprehensively integrates the GFM control loops and NP voltage dynamics. Based on this model, theoretical analysis reveals that the zero-sequence control in the GFM control loop is the decisive factor influencing NP voltage balance. Building on this insight, a novel decoupling control strategy is proposed to effectively eliminate the coupling between zero-sequence control and NP voltage dynamics. The proposed method demonstrates robust NP voltage balancing performance under various conditions. Finally, experimental results validate the accuracy of the theoretical model and the effectiveness of the proposed control scheme.

Modern receiving-end power systems are characterized by concentrated load centers, long-distance power transfer, and increasing penetration of inverter-based renewable resources. To support reproducible voltage-security studies for such systems, this paper develops a load-center–oriented equivalent benchmark based on the classical 3-machine 9-bus framework. The proposed benchmark construction explicitly defines the transformation from the original system, including bus and branch data, generator dispatch, load allocation, corridor ratings, renewable locations, SVG/OLTC/ESS/capacitor parameters, and coordinated-control variables. Beyond steady-state voltage profiles, the validation framework incorporates quantitative load-center representativeness indices, renewable-output sensitivity, load-growth screening, selected N-1 contingency results, and a reduced-Jacobian modal indicator. The benchmark reproduces a dominant transfer corridor with 81.50% loading, a minimum bus voltage of 0.9700 p.u., and a high-renewable overvoltage case reaching 1.0807 p.u. Coordinated multiresource control reduces the maximum target-bus voltage magnitude by 0.0526 p.u. relative to the uncontrolled high-renewable case. The model is intended as a transparent steady-state benchmark for voltage-security, renewable-hosting, and coordinated-control studies. Fast electromagnetic and converter-control phenomena are identified as necessary extensions for dynamic studies.

The 2025 Iberian blackout, a highly disruptive large-scale grid failure, underscores the increasing fragility of deeply interconnected and decarbonizing power systems. Traditional security paradigms, such as the N-1 criterion, were designed for isolated and predictable contingencies but are increasingly unable to anticipate or contain cascading failures. Using the Iberian blackout as a reference case, this review synthesizes global case studies to identify common structural drivers of modern power system vulnerability. This review argues that the reliability paradigm must shift from static, component-level robustness toward the management of dynamic failure propagation pathways, reduced system inertia, and cross-sector interdependencies. To operationalize this paradigm shift, this review proposes an integrated, multi-layered resilience framework. From an analytical perspective, this review details the transition toward probabilistic dynamic security assessment for characterizing failure propagation trajectories. At the engineering layer, controllable regional islanding, adaptive protection schemes, and artificial-intelligence-enhanced control execution enable rapid system stabilization within narrow operational time windows. To address computational complexity, this review proposes an integrated interdisciplinary methodological framework that leverages stochastic interaction graphs and digital twins to achieve multi-scale system observability, while employing deep reinforcement learning and coalitional game theory to coordinate fast-timescale physical control with slow-timescale market mechanisms. Finally, this review outlines a governance transition from prescriptive compliance toward continuous, state-aware risk management supported by unified cross-border resilience metrics. By reframing resilience as a dynamically managed systemic attribute, this review presents a structured agenda for safeguarding interconnected power grids against escalating cascading risks.

With the increasing uncertainty in electricity markets and the growing influence of news media, traditional Virtual Power Plants (VPP) lack sufficient information perception capabilities in market behavior modeling to meet dynamic response requirements. To address this problem, this paper proposes a method for modeling and optimizing the energy market behavior of virtual power plants by integrating news sentiment and natural language semantic information. First, we use pre-trained language models to encode news text in multiple dimensions, extracting sentiment tendencies and semantic features, and constructing joint emotion-semantic representation vectors. Second, these vectors are integrated with historical power system operation data and embedded into the state modeling process of virtual power plants, combined with reinforcement learning algorithms to generate optimal strategies. The experiments are based on real electricity price data and news sentiment, comparing with control methods in terms of prediction accuracy, strategy response, and profit capability. Results show our method cuts overall electricity price prediction errors by 28-32 % (RMSE/MAPE) and peak prediction bias by similar to 50 %, while significantly improving VPP profitability and flexibility. This research provides a new perspective for virtual power plant modeling based on natural language information, offering both theoretical significance and practical application value for engineering promotion.

Capacity planning for grid-connected green hydrogen systems is complicated by nonlinear electrolyzer efficiency and time-varying electricity pricing that create operational arbitrage opportunities simplified models cannot evaluate. We develop a bilevel optimization framework coupling upper-level capacity investment with lower-level 5-minute dispatch using piecewise-linear efficiency representation. Seasonally weighted representative days balance computational tractability with temporal fidelity. To ensure fair comparison, all capacity configurations are evaluated by a unified high-resolution operational simulator. Case studies on Inner Mongolia's system with 15% wind and 8% solar curtailment demonstrate 7.0% higher annual profit versus traditional single-level planning, achieved through systematically larger electrolyzer (8.2%) and storage (70.3%) capacities that capture intra-day arbitrage under time-of-use pricing while satisfying renewable energy penetration constraints. Results demonstrate that integrated capacity-operation planning substantially improves economic viability of renewable hydrogen investments.

Wildfires pose a significant threat to urban regions, with cities like Los Angeles facing increasing challenges due to their vulnerability to frequent and severe wildfire events. This study proposes a novel framework for optimizing fire rescue vehicle scheduling and energy system operations during wildfire disasters. By integrating predictive wildfire modeling with microgrid-based energy systems, the framework dynamically allocates energy resources to critical demands such as emergency shelters, hospitals, and rescue operations when grid supply is disrupted. The wildfire model simulates fire growth, wind-driven spread, and infrastructure impact, ensuring that the framework adapts to real-time conditions. A case study focusing on Los Angeles demonstrates the practical application of the proposed methodology, showcasing improved emergency response, minimized infrastructure losses, and enhanced operational efficiency during wildfires. This research highlights the importance of combining energy systems and disaster management strategies to build resilience in wildfire-prone urban areas, offering valuable insights for emergency planners and policymakers.

California's escalating wildfire crisis threatens power infrastructure, leading to widespread blackouts and economic disruptions. Traditional grid-hardening strategies struggle to adapt to dynamic fire propagation, necessitating fire-resilient microgrid optimization. This study proposes an NSGA-III-based multi-objective optimization framework that integrates wildfire spread modelling, energy dispatch strategies, and adaptive islanding to enhance resilience. Capturing key fire-induced uncertainties such as wind-driven fire acceleration and thermal degradation, the model enables proactive decision-making. A case study on California's wildfire-prone energy system, using real fire incident data, demonstrates that strategic islanding and preemptive dispatch reduce blackout durations by over 90%, while renewable energy utilization in post-fire recovery improves by 50%. The proposed method outperforms conventional grid-hardening strategies by 3× in resilience metrics, providing a decision-support tool for wildfire-prone energy systems and laying the foundation for a next-generation resilient power grid.