Abstract As power systems transition towards decarbonised energy mixes with increasing wind penetration, prolonged low‐wind‐power (LWP) events have emerged as a critical source of adequacy risk. However, a comprehensive analysis of LWP events remains constrained by the scarcity of long‐term historical wind‐power generation records in regions with high renewable penetration. To address this limitation, this study proposes a TabNet‐based wind power modelling framework for 90 wind farms in Guangxi, China, leveraging historical meteorological records and measured operational data. ERA5 reanalysis data are used to reconstruct hourly wind‐power generation over a 43‐year period (1980–2022). An adaptive mean‐below‐threshold method featuring a tolerance mechanism is proposed to identify LWP events over the 43‐year period. Based on this, the intra‐annual and inter‐annual characteristics of LWP events are analysed, and their relationship with the El Niño–Southern Oscillation and the Eastern Pacific (EP) index is investigated. The results indicate that both the annual duration and frequency of LWP events exhibit upward trends over the study period, with the upward trends becoming more pronounced after 2000. Furthermore, the EP index demonstrates a lagged association with LWP duration and frequency in Guangxi, with decaying La Niña years exhibiting the clearest positive anomalies in LWP duration and frequency among the investigated ENSO categories. These findings offer climate‐informed insights for wind power adequacy assessment and renewable integration planning.
Abstract Photovoltaic‐storage‐charging microgrids (PSCMs) are being increasingly deployed in extreme environments, including desert, polar, coastal and high‐altitude regions, to provide a reliable and sustainable power supply. However, these harsh environments significantly accelerate the degradation of photovoltaic arrays, battery storage systems and electric vehicle charging infrastructures. Effectively managing PSCMs under these conditions requires energy management systems with control strategies that explicitly account for environment‐dependent degradation dynamics. This paper provides a comprehensive review of degradation‐aware energy management for PSCMs operating in extreme environments. First, extreme operating environments are classified using quantitative thresholds, and their specific degradation effects on PSCM components are systematically analysed. Next, degradation modelling approaches comprising analytical models, data‐driven approximations and physics‐informed enhancements are reviewed and compared. Current energy management strategies, ranging from model‐based methods to data‐driven reinforcement learning and hybrid architectures, are then examined. Finally, future research directions are outlined, focusing on generalizable multi‐stress degradation models, principled degradation‐aware control frameworks and comprehensive datasets and digital twin testbeds.
Abstract Weather‐dependent renewable energy makes modern power systems more sensitive to extreme weather. Therefore, reliable identification of power system‐oriented extreme weather scenarios is important for reliability assessment, resilience analysis, and risk‐oriented operation. However, these scenarios are heterogeneous and suffer from severe class imbalance, which makes reproducible labelling and robust model training difficult. This paper proposes a hybrid classical‐quantum learning framework for power system‐oriented extreme weather scenario identification. First, a structured definition and labelling framework is developed to extract extreme weather scenarios from large‐scale weather‐power datasets by jointly considering meteorological abnormality, power system response, and external event evidence. Second, a quantum generative module is introduced as a training‐only augmentation tool to synthesize minority‐class samples, while validation and testing remain strictly based on real observations. Third, a hybrid classical‐quantum classifier is constructed by embedding variational quantum circuits as trainable feature‐mapping modules within a classical learning pipeline, aiming to examine their feasibility under small‐sample and imbalanced‐data conditions. In addition to conventional generation and classification metrics, risk‐oriented indicators are used to evaluate whether the identified scenarios are meaningful for power system operation. The proposed framework is validated using real‐world weather‐power data from China. The results show that the quantum generative adversarial network preserves minority‐class risk characteristics, while convolutional neural network‐quantum recurrent neural network (CNN‐QRNN) attains the numerically highest minority macro‐F1 of 0.2998 among the compared models, with a macro‐F1 of 0.3972.
Abstract Modern power systems require increasingly flexible and responsive power conversion, driving the need for power converters with enhanced controllability and adaptability, where artificial intelligence (AI) plays a key role in enabling intelligent decision making, adaptive control and system level optimization. However, current AI applications in power systems remain largely restricted to cloud‐based post processing or device‐level demonstrations, lacking end‐to‐end hardware platforms that support the full pipeline from programming to deployment within a closed loop formed by physical converter hardware and AI modules. To address this gap, this paper proposes an edge AI‐powered power converter (EAI‐PC) hardware platform that utilizes locally deployed edge intelligence to establish bidirectional data exchange between converter low‐level control and EAI upper‐layer command dispatch. The platform features an end‐to‐end AI‐to‐converter closed‐loop deployment framework, enabling AI‐generated commands to be directly executed on physical converter hardware. By interconnecting converters (AC/DC and DC/DC) with an EAI module (using NVIDIA Jetson Orin Nano) via an industrial CAN protocol, this platform supports hardware‐level algorithm verification in the power system domain. The proposed EAI‐PC platform is experimentally evaluated across system configurations from single to multiple interconnected converters, assessing converter performance, command delivery and three AI deployment cases (two offline and one online): a multilayer perceptron (MLP) for power‐sharing prediction, a spatio‐temporal graph neural network (STGNN) for photovoltaic (PV) and load forecasting, and a bidirectional long short‐term memory (Bi‐LSTM) for online battery dispatch in load management and peak shaving.
Abstract Renewable power purchase agreements (PPAs) expose generators and offtakers to capture‐price risk because realised revenue depends on prices during renewable production hours. Downside protection written on the production‐weighted settlement price is economically relevant, but difficult to mark and hedge because the index is non‐tradable and depends jointly on power prices and renewable output. This article develops a futures‐anchored statistical mark‐to‐model and hedging framework for long‐dated Asian floors embedded in renewable PPAs when renewable generation remains stochastic. The framework is not interpreted as a unique arbitrage‐free price in a complete market. Instead, it isolates the hedgeable component through liquid German baseload futures and updates the remaining price‐generation exposure using deterministic seasonal profiles and a vector autoregression for deseasonalised German day‐ahead price and wind‐generation residuals. Moment matching and a delta‐method approximation yield a tractable Bachelier‐style floor mark and operational futures hedge ratios. In pre‐delivery backtests for the 2024 and 2025 delivery years, the stochastic hedge reduces daily residual variance by 94.7% and 89.0%, respectively compared with 69.1% and 47.9% for a deterministic‐generation hedge. The results show that stochastic renewable production is a first‐order input for practical risk management of renewable PPA optionality.
Accurate state of health (SOH) estimation is crucial for the safe operation of lithium-ion batteries. To address the limited capability of traditional models in characterizing nonlinear degradation, this study proposes a novel data-driven SOH estimation framework. First, an indirect health indicator (IHI) extraction strategy constructs aging-sensitive features directly from operational data, avoiding complex electrochemical modeling. Second, a hybrid multiple kernel extreme learning machine (HMKELM) integrates radial basis, polynomial, and sigmoid kernels with adaptive mixing weights to capture both local and global characteristics of degradation trajectories. Furthermore, to alleviate the hyperparameter-tuning bottleneck of HMKELM, a gold rush-inspired grey wolf optimizer (GRGWO), incorporating a leader-update strategy and a dynamic convergence factor, is developed to escape local optima and jointly optimize kernel parameters and mixing weights. Validation on the NASA ARC-FY08Q4 dataset shows that the proposed GRGWO-HMKELM framework enables accurate long-horizon SOH estimation using only early-cycle data for training and consistently outperforms representative baseline models, indicating improved generalization for practical in-service SOH monitoring.
Abstract An effective method for detecting cyberattacks is essential to the security of smart grids (SGs). In SGs, data from both cyber and physical domains can support attack detection. However, existing works insufficiently consider the heterogeneity, high dimensionality, and cross‐domain correlations of multi‐source data, affecting model generalization, stability, and detection accuracy. To address these limitations, this paper proposes a hybrid stacked learning method with cooperative feature selection (HSL‐CFS). First, a cooperative feature selection approach is proposed for feature extraction from multi‐source heterogeneous data. An embedded method with adaptive threshold adjustment and a wrapper method with dynamic bidirectional search extract key features from complementary perspectives, and their feature sets are merged via set union to output representative features. Second, a hybrid stacked model combining Extreme Random Trees (ET) and an improved Convolutional Neural Network (CNN) is proposed. The CNN is enhanced with Euclidean‐norm regularization (L2 regularization) and Squeeze‐and‐Excitation (SE) block attention mechanism to mitigate overfitting and recalibrate channel responses. Furthermore, probability alignment calibrates base classifier outputs before stacking, enabling better capture of complementary patterns in SGs cyber‐physical data. Experimental results show that HSL‐CFS selects fewer features and achieves enhanced stability, generalization and accuracy, outperforming existing methods on the selected dataset.
Fast oscillations observed in wind farms and other renewable energy systems interfaced with power grids must be addressed promptly to prevent devastating shutdown and damage. Existing approaches mainly target high-magnitude oscillations using data-intensive convolutional neural networks (CNNs), leading to delayed detection and overlooking mild oscillations that can still cause long-term degradation. This paper proposes a dual detection framework that rapidly identifies high-magnitude oscillations using time-domain analysis while detecting mild oscillations through FFT-based frequency-domain feature extraction. The framework leverages hyperdimensional computing (HDC) and machine learning models, including one-class support vector machine (SVM), 1D-CNN, 2D-CNN, and a one-class autoencoder. To enhance performance evaluation, we introduce true positive confidence (TPC) and false positive confidence (FPC) metrics. Unlike prior work limited to server-based evaluation, we deploy HDC on edge platforms, including a Jetson Nano graphics processing unit (GPU), tensor processing unit (TPU), and field programmable gate array (FPGA). Experimental results on real-world wind farm data and Simulink-generated datasets show that HDC outperforms all baseline models, achieving at least 99% confidence in both TPC and FPC for high-magnitude and mild oscillations. Notably, HDC operates up to 2100 times faster than 1D-CNN on GPU and 6800 times faster on FPGA, while being four times smaller in model size, demonstrating that FPGA-based HDC is the minimum suitable hardware for reliable real-time deployment in wind farms.
Abstract Against the backdrop of renewable energy achieving full market integration, renewable energy forecasting deviations have become a critical bottleneck constraining the clearing efficiency of electricity spot markets, hindering the rational allocation of costs, and threatening the secure operation of the power system. A framework comprising a full‐stage governance system covering planning, medium‐ to long‐term trading, spot markets, and settlement is adopted in this paper. The experiences of global markets in addressing forecast deviations of renewable energy through multi‐stage, tiered governance are systematically reviewed in the paper. During the planning stage, a probabilistic physical defence boundary is established to address hidden physical gaps arising from random fluctuations that deterministic planning cannot account for. In the medium‐ to long‐term stage, risks are mitigated through a combination of physical resource aggregation and financial hedging tools, including bilateral contracts, forward contracts, reliability options, and parametric insurance. In the spot market, approaches such as self‐forecast verification, rolling forecast corrections, and a multi‐level substitution mechanism for abnormal data are adopted. During the settlement stage, incentive‐compatible mechanisms are used to improve the accuracy of renewable energy forecasting. Finally, suggestions on full‐stage risk management for large‐scale renewable energy participating in the spot market are proposed in this paper.
Carbon emissions and energy consumption of urban buildings represent a notable fraction of global energy utilization. In pursuit of the "carbon peaking and carbon neutrality" goals, urban park building integrated energy system (IES) energy saving technology is regarded as a key technology to achieve sustainable urban development. Therefore, a bi-level optimal capacity allocation method for the park energy supply/storage equipment has been developed in this paper. Flexible load demand response and energy supply reliability are fully considered. First, heterogeneous data analysis of smart park is conducted and models of energy conversion devices and storage devices based on the energy supply/storage characteristics of park IES are established, along with operational constraint models for various equipment. Second, at the upper level, the optimization aims to minimize the park IES cost. The operational cost is minimized in the lower level. The source/storage capacity configurations are optimized by genetic algorithm. In this paper, operational optimization is incorporated into planning, and flexible load control schemes are applied to peak load shifting, contributing to alleviated energy demand pressure during peak periods. Finally, the supervisory role of the proposed bi-level optimization method is proved by case studies demonstrated by a real park IES and the effectiveness and feasibility of the data processing methods in improving the quality and accuracy of load data were also verified. The developed methodology enhances economic benefits, fully meeting the multi-energy demand of the system.
To address the challenge of low adaptability in distribution network planning caused by significant regional differences in electricity consumption, this paper proposes a distribution network planning method based on a self-adjusting parameter double deep Q-network (SAP-DDQN). First, considering the disparities in electricity consumption across different locations, distribution areas are classified according to load density. For each type of distribution area, appropriate calibration criteria are selected, and indicator models are established covering reliability, economics and flexibility criteria. Subsequently, key indicators under each criterion are extracted using the analytic hierarchy process and kernel principal component analysis. A deep reinforcement learning model for distribution network planning is then developed to achieve rapid optimization of the network configuration. Finally, the effectiveness of the proposed method is validated using a 24-node distribution network and an actual 84-node urban distribution network in China. Test results demonstrate that the proposed method can accurately select the optimal network configuration for each distribution area and provide a specific planning scheme.
With the increasing penetration of power electronic devices in power systems, virtual synchronous generator (VSG) technology has garnered widespread attention for its ability to provide inertia and damping support to the grid. However, while simulating the external characteristics of synchronous generators, this technology also introduces inherent rotor oscillation issues. Particularly in multi-machine parallel operation, insufficient system damping can easily lead to low-frequency oscillations, threatening system stability and equipment safety. To address this issue, this paper first establishes a small-signal model for multi-VSG parallel grid connection. Subsequently, a Phillips-Heffron model tailored for multi-VSG parallel structures is constructed, revealing the fundamental cause of system oscillations under disturbances. Building upon this foundation and drawing inspiration from traditional power system stabilizer design principles, a virtual power system stabilizer control strategy tailored for multi-VSG parallel grid-connected systems is proposed. Finally, the proposed control strategy is validated through MATLAB/Simulink simulations. The simulation results demonstrate that, compared to conventional control methods, the proposed VPSS control strategy effectively suppresses low-frequency oscillations in the system, significantly enhancing overall stability.
Research in power system is persistently hampered by the scarcity of high-fidelity, public grid data due to security and privacy constraints. Existing unimodal synthesis methods fail to harmonize physical laws, visual representations and semantic descriptions, obstructing the application of multimodal large language models (LLMs) in the energy sector. To address this, we propose a multimodal synthesis framework that generates aligned datasets comprising physical parameters, single-line diagrams and natural language descriptions. The framework combines rule-based topology generation with a two-stage chain-of-thought (CoT) strategy, enabling LLM agents to initialize electrical parameters based on statistical priors. To ensure physical feasibility, an iterative power flow feedback loop is introduced to guarantee convergence. Furthermore, retrieval-augmented generation is employed to enhance component-level visual details. Experimental results indicate that the synthesized grids achieve high structural similarity and physical fidelity compared to real-world benchmarks. We have open-sourced this physically validated multimodal grid dataset to provide critical foundational support for developing physics-informed "energy LLMs."
The widespread integration of photovoltaic (PV) power, energy storage systems, and other demand-side resources highlights the importance of optimal dispatching for the PV-storage-load virtual power plant (VPP). However, the fluctuation of the PV power generation and the uncertainty of the electricity prices exacerbate the economic operation risks of the VPP. To address these challenges, an optimal dispatching strategy for the PV-storage-load VPP is proposed, with due consideration given to the dual uncertainties of electricity prices and PV power output. Firstly, the conditional value-at-risk theory is employed to quantify the uncertainty risk of VPP revenue caused by electricity price fluctuations. Secondly, in view of the asymmetric fluctuation intervals of PV power output, a quantification method for PV uncertainty and dispatch robustness is developed using the confidence gap decision theory. Furthermore, by combining the regulation reserve model of multi-type flexible resources, a robust optimization model for the PV-storage-load VPP is constructed with the objective of maximizing comprehensive operational revenue, which includes the provision of upward and downward reserve services. Finally, case studies based on a PV-storage-load VPP in a Chinese province are conducted to validate the effectiveness and superiority of the proposed model. The simulation results indicate that the proposed robust optimization strategy effectively reflects the relationship between the uncertainty of PV power output and the risk preference of decision-maker, mitigates the fluctuation risks of electricity prices to ensure the stability of the power system, and enhances the economic efficiency and flexibility of the PV-storage-load VPP operation.
Nepal's electricity sector operates under a monopsony, with government entities dominating generation, transmission, and distribution. This market structure suppresses competition, leading to inherent challenges in establishing fair and efficient electricity tariff systems. This study aims to address these inefficiencies by diagnosing the existing tariff structure and proposing restructured time-of-use (TOU) models. A key constraint of the proposal is ensuring revenue neutrality for the utility. The research employs a dual-method approach. First, it models locational marginal pricing (LMP) across six load buses in power world simulator, integrating a 144 MW generator and categorizing demand into residential, industrial, commercial, non-commercial, and other loads. Second, it analyzes daily load profiles from Ghusel village to design optimal TOU tariff models for residential consumers. The LMP modeling revealed significant locational price variations, highlighting congestion and inefficiencies in the current system. Furthermore, introducing distributed generation reduced the LMP for residential consumers by 12.15% compared to base case scenarios. The analysis of load profiles led to the selection of a four-period TOU structure (morning peak, evening peak, normal, and off-peak) as the most effective. Specific tariffs-a demand charge of 60 Nepalese Rupee (NPR)/kW and energy rates of NPR 9.49, 12.55, 6.78, and 4.07, respectively-were found to improve peak load shifting, revenue collection, and consumer fairness. The study demonstrates the strong potential of LMP and structured TOU tariffs to enhance grid reliability, sustainability, and fairness in Nepal. However, successful implementation requires overcoming practical obstacles such as cross-subsidies, regulatory inflexibility, and inadequate metering. Therefore, this study recommends a gradual policy strategy involving pilot initiatives, regulatory reforms, and consumer awareness campaigns.
The high penetration of renewable energy sources introduces uncertainty, posing significant challenges to the secure operation of multiple microgrids interconnected through lower voltage flexible interconnection devices. To address power and voltage fluctuations caused by these uncertainties, this paper proposes a robust optimisation method for multiple microgrids based on a novel online energy storage system (ESS) response strategy. First, a synthetic energy management model is constructed to facilitate the coordinated scheduling of various distributed energy resources within the system. Next, inspired by the mechanisms of automatic generation control, a novel online ESS response strategy is designed to enable ESSs to participate in both day-ahead economic scheduling and real-time uncertainty mitigation. Building on this, a robust optimisation model for multiple microgrids is established and reformulated into a solvable form based on duality theory. Finally, case studies conducted on a physical multiple microgrid system validate the effectiveness of the proposed method. The results demonstrate that the method effectively mitigates uncertainties and maintains voltage security, ensuring efficient and reliable operation of microgrids.
With the deepening of research on smart grids, the distribution system has gradually evolved into a cyber-physical distribution system (CPDS). The refined reliability modelling and assessment methods for CPDS are of great significance. This paper proposes an integrated reliability assessment framework combining distribution information system (DIS) performance analysis with sequential scenario correction. A DIS model is established. By mapping the performance of DIS fault location and isolation to the operational scenarios of the power distribution system (PDS), the impact of DIS performance on the interruption time of PDS can be quantified. A physical operational scenario simulation and correction method is proposed, which uses the quantified interruption time due to DIS impacts for the correction of PDS operational scenarios. Based on the topology, the distribution system is divided into regions, and an optimal decision model for recovery resources is designed and used to calculate reliability indicators. The improved CPDS reliability assessment results of the IEEE RTS BUS6 F4 bus system are presented and analysed, and the key factors affecting the reliability of CPDS are identified and discussed.
The increasing uncertainty caused by volatile renewable generation and random electricity demand has always been a critical challenge in power system operations. Robust optimization (RO) is a powerful tool for effectively addressing this uncertainty. As the interplay between uncertain factors and decision-making becomes more prevalent, RO with decision-dependent uncertainty (DDU) has attracted increasing attention. DDU significantly changes how the uncertainty set in RO is modelled and how the problems are solved. This study provides a comprehensive overview of the recent developments in RO with DDU for power system problems. We begin by introducing various models of DDU, classified according to their underlying causes. Next, we summarize the state-of-the-art solution algorithms for RO with DDU, such as variants of the column-and-constraint generation (C&CG) algorithm, variants of Benders decomposition, and multiparametric programming. Furthermore, we explore the application of RO with DDU in power systems. Based on our findings, we propose several research directions that may be valuable for future studies.
To improve the adaptability of voltage regulation in active distribution networks (ADNs) with high photovoltaic (PV) penetration, this paper proposes a distributed Volt/Var control (VVC) strategy enabled by multi‐agent deep reinforcement learning and implemented through heterogeneous PV inverters. First, a distributed VVC framework is established by partitioning the ADN into multiple sub‐networks, each modelled as an agent, with the goal of minimizing voltage deviation. This control framework considers the heterogeneous operation modes of PV systems and utilizes both reactive power support and active power curtailment to maintain voltage within acceptable limits. Then, the VVC problem is formulated as a Markov game and solved using a multi‐agent soft actor–critic algorithm. Simulation studies conducted on the IEEE 33‐bus and 118‐bus test systems validated the effectiveness of the proposed method, demonstrating its superior performance in reducing voltage fluctuations compared to benchmark approaches.
Low-frequency oscillations (LFOs) remain a major obstacle to maintaining stable dynamic performance in power systems with high levels of renewable energy integration. In particular, inter-area LFOs have emerged as a critical concern in large, interconnected grids. In recent years, wide-area damping control (WADC) has been widely studied in suppressing inter-area LFOs in power systems with renewable energy. This paper provides a comprehensive review of the key technologies underpinning WADC within the framework of wide-area measurement systems. First, it outlines the overall structure of WADC in renewable-integrated power systems. Next, it examines the central technical issues associated with WADC in detail. The paper then summarizes and compares various offline and online adaptive design methodologies for wide-area damping controllers. Finally, it discusses the major challenges facing WADC and highlights future development opportunities. Overall, this review aims to deliver a thorough and meaningful overview of current research on WADC for power systems with high renewable energy penetration.
