In recent times, power grid optimization has incorporated ML and optimization algorithms to improve the power grid and reduce losses. Graph Neural Networks (GNN) are efficient in modelling complex dependencies, while Particle Swarm Optimization (PSO) optimizes line losses and operating costs. However, the application of the above approaches together in a single framework has received attention from researchers. The current approaches, like GA and LSTM, face difficulties in scaling and accuracy, which affect the power grid optimization. This study proposes a novel integrated approach for grid load scheduling and line loss reduction using the GNN + PSO approach. The framework is grounded on the IEEE 123-Bus system dataset for training the GNN model for load prediction, and optimized load distribution is determined using PSO to minimize line losses while satisfying grid constraints. The GNN+PSO model resulted in a MAE of 0.005, Mean Squared Error (MSE) of 0.0003, RMSE of 0.008, and R2 OF 0.99. In addition, GNN+PSO approach resulted in a reduction of line losses by 77.1%. This shows that the model outperforms other models like Bi-LSTM+GA, LSTM+GO, and GAT. This approach provides better accuracy and efficiency, which can be used to solve smart grid applications and can be extended to future dynamic grid management.
INTRODUCTION: response to the theoretical bottlenecks of the hybrid energy storage system (HESS) when participating in grid regulation. OBJECTIVES: Such as strong coupling of multi-dimensional state variables, difficulties in coordinating energy storage scheduling with grid stability, and prominent subjectivity in traditional power transfer decisions, a model for AI optimization collaborative control and grid state assessment with HESS as the core control object is proposed. METHODS: Firstly, based on traditional strong transfer features such as fault characteristics, grid topology, equipment status, and environmental disturbances, the operation state variables of HESS (including state of charge, charging and discharging power, health level, etc.) are introduced to construct a multi-dimensional feature matrix integrating energy storage and the grid, and the objective weighting of each dimension is achieved based on the information entropy theory. Secondly, an adaptive weighted kernel principal component analysis (AW-KPCA) algorithm is proposed, which uses the particle swarm optimization (PSO) algorithm to dynamically adjust the kernel function parameters and feature weights, and incorporates the HESS charging and discharging strategy into the optimization space to achieve joint optimization of energy storage scheduling and high-dimensional feature dimension reduction. The core novelty is the synergistic integration of HESS dynamics into grid feature modeling, PSO-based joint optimization of kernel parameters, feature weights, and storage dispatch, and TOPSIS-driven risk quantification for forced transmission decisions. RESULTS: Compared with the traditional kernel principal component analysis algorithm, the proposed method improved data processing efficiency by 41.3% and feature extraction accuracy by 15.7%. Finally, the TOPSIS multi-attribute decision-making theory is integrated to build a comprehensive judgment model on the adaptability of forced transportation, quantify the feasibility and risk level of forced transportation, and achieve optimal decision-making under multiple constraints. CONCLUSION: The simulation test was completed on the IEEE 300-node system and the Guizhou Power Grid Actual Fault Data Set. The accuracy of the proposed model reached 95.7, and its robustness was significantly stronger. The research provides new theoretical and algorithmic support for HESS to participate in the restoration of power grid failures.
Power grid data often contain highly abstract metadata, ambiguous business terms, and complex statistical scopes. These characteristics make general large language models prone to semantic deviation and SQL hallucination in Text-to-SQL tasks. To address this problem, this paper proposes SQL-GRID, a grounded retrieval and interactive disambiguation agent for Text-to-SQL over power grid data. The framework is based on a multi-agent collaboration mechanism. It integrates retrieval, validation, disambiguation, and SQL generation modules to transform natural language intent into accurate SQL statements. To bridge the gap between business semantics and database structures, this paper constructs a Business Knowledge Representation(BKR) based knowledge base, which integrates database structural information, business descriptions, real data samples, and statistical rules. A multi-strategy RAG method is also designed. It combines semantic vector retrieval with keyword matching and provides accurate domain knowledge for SQL generation. To address business ambiguity, this paper further proposes a closed-loop mechanism consisting of pre-generation disambiguation, interactive clarification, and post-generation validation. Through explicit human-machine interaction, the framework guides users to clarify statistical granularity and statistical scope constraints. This process effectively reduces uncertainty during SQL generation. Experimental results show that the proposed method achieves a metadata recall rate of 94% on the power grid dataset. The SQL execution rate increases from 40.2% in the traditional baseline to 84.6%. The semantic matching score reaches 4.85. These results show that the proposed framework significantly improves the accuracy and reliability of self-service queries in complex power grid scenarios.
INTRODUCTION: The increasing integration of renewable energy, distributed storage, and flexible loads has introduced substantial uncertainty and operational complexity into modern smart grids. Existing energy management methods can optimize multi-period dispatch or encode the physical grid topology, but they often do not explicitly represent the execution order and conditional dependencies among heterogeneous control actions. OBJECTIVES: This paper aims to develop a safe and energy-efficient sequential control generation framework for digital-twin smart grids that reduces operating cost and renewable curtailment while maintaining voltage security and action feasibility. METHODS: A topology-constrained latent execution graph learning framework, termed TC-LEG, is proposed. TC-LEG uses a physical grid graph to encode electrical connectivity and operating states, while a distinct latent execution graph represents state-dependent dependencies among storage dispatch, renewable curtailment, demand response, reactive compensation, and topology switching. A hybrid discrete–continuous control sequence generator produces multi-step actions, and a simulation-oriented digital twin synchronizes the current grid state, verifies each candidate action, projects infeasible actions onto a state-dependent feasible set, and feeds the accepted action and updated state back to subsequent generation steps. RESULTS: On the IEEE 33-bus system, TC-LEG achieves a normalized operating cost of 0.811, a voltage violation rate of 0.9%, a renewable curtailment rate of 4.9%, and an action feasibility rate of 98.5%. Its average inference time is 18.6 ms on the IEEE 33-bus system and 28.4 ms on the IEEE 69-bus system, both substantially shorter than the 15-minute control interval. CONCLUSION: TC-LEG provides a topology-aware and interpretable approach to sequential energy management by combining electrical-topology representation, action-dependency learning, hybrid control generation, and digital-twin verification in a closed-loop inference process.
High penetrations of distributed energy resources require energy regulation that combines cloud-level global optimization with edge-level fast response. This paper proposes EC-HDT, a device-edge-cloud hierarchical digital twin in which a lightweight graph-attention-temporal-convolution estimator reconstructs local states under asynchronous, noisy, and missing measurements, while a cloud predictor and model predictive controller perform rolling economic optimization. A five-factor decision weight based on communication latency, information freshness, estimation confidence, operational risk, and edge computational load continuously allocates control authority between edge and cloud, and a quadratic-programming safety layer enforces physical constraints. On the IEEE 33-bus system, EC-HDT achieves a nodal-voltage MAE of 0.0076 p.u., mean/P95 end-to-end latencies of 56.4/89.4 ms, and a 99.2% control success rate; the daily operating cost is 3.51% lower than that of the fixed-fusion scheme. The results indicate that state-aware edge-cloud coordination can improve the latency-economy-safety trade-off in distribution-system regulation.
INTRODUCTION: Transmission-line Internet of Things (IoT) networks provide continuous sensing for power-grid monitoring, but their wireless links are vulnerable to channel fluctuation, burst interference, sparse relay deployment, and limited node energy. OBJECTIVES: This paper aims to improve reliable IoT data transmission over unstable transmission-line channels by jointly considering prediction uncertainty, packet urgency, reliability, latency, and energy constraints. METHODS: A Predictive Cross-layer Anti-interference Optimization (PCAO) method is proposed. PCAO predicts near-future channel and interference states through uncertainty-aware temporal modeling, evaluates packet priority from event severity and information freshness, and jointly optimizes channel allocation, transmission power, redundancy, and relay preference through constrained cross-layer optimization. RESULTS: Experiments on DeepMIMO, RadioML, POWDER, and FlockLab show that PCAO consistently outperforms DQN, DDPG, PPO, and SAC, improving average packet delivery ratio by 2.55 percentage points over SAC while reducing delay by about 17.0%. CONCLUSION: The results indicate that predictive, priority-aware, and robust cross-layer control can enhance IoT transmission reliability under unstable channels in power transmission environments.
To address the challenges in wind-solar microgrids caused by the intermittency and volatility of renewable energy—such as power imbalance, unreasonable energy storage capacity configuration, and the difficulty in achieving coordinated optimization of economy, reliability, and environmental sustainability under limited capacity—a multi-objective energy storage capacity optimization method is proposed based on digital twin technology and a hybrid model combining Temporal Convolutional Network (TCN), Stacked Long Short-Term Memory Network (SLSTM), and Multi-Head Attention (MHA) mechanism (TCN-SLSTM-MHA). First, a digital twin model of the wind-solar microgrid is constructed to enable real-time mapping, monitoring, and simulation analysis between the physical system and its virtual counterpart, overcoming the limitations of traditional models in adapting to dynamic operational scenarios. Second, a TCN-SLSTM model is introduced, enhanced with the MHA mechanism to dynamically assign weights across time steps, thereby improving the accuracy of source (generation) and load forecasting. Finally, a multi-objective optimization function is established that simultaneously considers the minimization of Life Cycle Cost (LCC), maximization of Power Supply Reliability (PSR), and reduction of Carbon Emission Intensity (CEI), enabling the optimal determination of storage capacity. Experimental results demonstrate that, in terms of source-load forecasting, the proposed model improves prediction accuracy by 16.9% and 16.7%, respectively, compared to the conventional TCN-LSTM model. In energy storage capacity optimization, compared to the traditional weighted sum method, the proposed approach reduces LCC by 15.2%, increases PSR by 4.3%, and decreases CEI by 12.5%, validating the effectiveness and superiority of the proposed model.
High renewable penetration increases source-load uncertainty and challenges the stable operation of smart grids. This paper proposes a prediction-driven framework that integrates CNN-Attention-based joint renewable generation and load forecasting with rolling optimization of a battery–supercapacitor hybrid energy storage system. The main innovation lies in coupling source-load forecasting with hybrid energy storage dispatch, so that predicted renewable generation and load trajectories can directly support rolling scheduling decisions. Multi-source temporal inputs, including wind power, photovoltaic power, load demand, meteorological variables, electricity price, and periodic time encodings, are used to characterize source-load coupling. One-dimensional convolution extracts local fluctuation features, while temporal attention assigns adaptive weights to informative historical periods. In the dispatch stage, the battery mainly undertakes low-frequency energy balancing, whereas the supercapacitor suppresses high-frequency power fluctuations. The optimization objective jointly considers grid purchase cost, storage degradation cost, renewable curtailment penalty, load shedding penalty, and grid power fluctuation penalty. Experiments were conducted on one-year operational data from a regional wind–photovoltaic–storage microgrid, with a 15-min sampling interval and 35,040 time steps. Compared with LSTM, GRU, CNN, CNN-LSTM, and Transformer baselines, the proposed model achieved the lowest forecasting errors, with MAE/RMSE/MAPE of 6.692 kW/8.322 kW/3.288% for renewable generation and 3.747 kW/4.490 kW/0.406% for load forecasting. At the dispatch level, the proposed strategy reduced the total operating cost index to 360, decreased grid power fluctuation to 3.4%, improved renewable accommodation to 90.6%, and lowered curtailment to 2.8%.
INTRODUCTION: Secondary systems in power engineering involve complex regulatory documents, operational procedures, dispatching requirements, and maintenance knowledge. Natural-language queries with implicit domain semantics create difficulties for conventional intent recognition in regulation matching and operation-support scenarios. OBJECTIVES: This study aims to enhance regulatory semantic understanding and intent-recognition accuracy, thereby supporting dispatching, maintenance, regulation execution, and operation and maintenance of power engineering secondary systems. METHODS: A knowledge-enhanced intent-recognition method combining Retrieval-Augmented Generation (RAG) and Knowledge Graphs (KGs) is proposed. A seven-label regulatory corpus is constructed, and structured semantic triples together with retrieved regulatory contexts are used as composite inputs. RESULTS: Experimental results show that the proposed method outperforms traditional techniques in terms of Accuracy and Macro-F1 score, improving the recognition of regulation-related, operation-guidance, maintenance-related, and procedure-support intents. CONCLUSION: The proposed approach strengthens the semantic alignment between technical queries and regulatory knowledge, providing an effective foundation for dispatching support, maintenance guidance, regulation execution, and intelligent operation and maintenance in the power industry.
INTRODUCTION: Pumped-storage hydropower plants (PSHPs) are indispensable for maintaining grid balancing amid high-penetration renewable energy integration, yet optimizing their predictive control is severely hindered by the scale of interconnected units and the resulting decision-making complexity. OBJECTIVES: This paper aims to develop a computationally efficient and operationally scalable assignment predictive control framework tailored for modern PSHP fleets. METHODS: To alleviate the computational burden and enhance real-time responsiveness, a model-free assignment predictive control scheme incorporating a cutting-based method is proposed to effectively streamline the solution space while guaranteeing grid stabilization. RESULTS: Furthermore, the framework is extended to multi-PSHP clusters, establishing a cooperative model-free predictive architecture that maximizes collective balancing capacity across diverse geographical regions. Numerical simulations validate the proposed method's efficacy. CONCLUSION: The results indicate that the proposed cooperative framework provides a practical and highly scalable solution for active grid balancing and renewable integration in large-scale PSHP systems.
INTRODUCTION: Offshore wind turbine gearboxes operate under complex conditions and are highly prone to faults. Traditional single-source diagnostic methods are sensitive to noise and load fluctuations, limiting diagnostic reliability. OBJECTIVES: This study aims to improve the diagnostic accuracy and recognition performance of gearbox fault categories. METHODS: Multi-source monitoring data were preprocessed and fused, followed by feature optimization and construction of a deep learning-based diagnosis model for fault detection and classification. RESULTS: The proposed model achieved accuracies of 0.951, 0.947, and 0.938 under different load conditions, with Area Under the Curve AUC values above 0.96, outperforming benchmark models. CONCLUSION: The proposed method improves diagnostic accuracy, robustness, and engineering applicability for offshore wind turbine gearbox fault diagnosis.
INTRODUCTION: Ultra-High Voltage Flexible Direct Current (UHVDC-F) transmission systems have become the preferred choice for long-distance power transmission due to their superior control flexibility and power quality, with their reliability directly determining regional grid stability. OBJECTIVES: To ensure stable operation, it is essential to evaluate system reliability, quantify the enhancement effects of redundant components, and identify critical system vulnerabilities. METHODS: This paper proposes an evaluation framework considering diverse redundancy schemes. For component-level redundancy, a state transition model incorporating spare resource constraints is developed. For sub-module level redundancy, an opportunistic maintenance strategy is introduced. The framework utilizes sequential Monte Carlo simulation to accurately capture the stochastic evolution and dynamic transitions of the system. RESULTS: Results indicate that converter transformer redundancy most significantly boosts overall reliability. Furthermore, opportunistic maintenance for converter valves effectively mitigates degraded operation risks and extends the Mean Time Between Failures. CONCLUSION: This research establishes a high-fidelity digital twin foundation for complex UHVDC-F operational states. Beyond enhancing assessment accuracy, these quantitative matrices can be seamlessly integrated into AI-driven asset management and cost optimization engines, transforming risk evaluation into active O&M economic scheduling to fortify large-scale grid resilience.
INTRODUCTION: Earthquake-induced disturbances require fast, standardized fault reports to support regional power-grid emergency response. OBJECTIVES: To automate report drafting while improving terminology, structure, and traceability compared with manual writing and simple template filling. METHODS: We design a workflow that combines slot filling, metadata-driven image proxy summarization, and controlled prompt-driven paragraph generation, then assembles outputs into a standardized report template and software system. RESULTS: On public earthquake and outage datasets, the proposed method outperforms baseline template filling (BLEU-4 1.00 vs 0.37, ROUGE-L 0.90 vs 0.50, BERTScore 0.93 vs 0.85, field coverage 1.00 vs 0.75) and reduces end-to-end report production from ~180 min (manual) to 2 min 15 s. CONCLUSION: The system enables reliable, high-efficiency fault report generation for earthquake emergency response in regional power grids, with clear potential for deployment.
INTRODUCTION: The output of photovoltaic (PV) arrays has nonlinear characteristics and is affected byambient conditions, so maximum power point tracking (MPPT) is crucial for efficient photovoltaic power generation. OBJECTIVES: This study aims to propose an MPPT control strategy based on improved perturbation and observation (P&O) method to overcome the shortcomings of traditional MPPT methods. METHODS: The method proposed integrates slope control and fuzzy control algorithms for hybrid control to achieve adaptive adjustment of control step size. In addition, the simulation test for start-up and sudden changes in operating conditions of PV were conducted to verify the practicability and progressiveness of the improved method proposed. RESULTS: The results confirm that the improved method proposed can reduce the output fluctuations of PV array and achieve shorter MPPT time and higher MPPT efficiency. CONCLUSION: The improved MPPT method has better MPPT performance, which provide effective reference for further improving the efficiency of PV power generation.
INTRODUCTION: Wind power variability challenges system regulation as renewables expand. Hydrogen storage provides long-term balancing. OBJECTIVES: To optimize the operation and configuration of a wind-storage-hydrogen-gas turbine system. METHODS: An Improved Honey Badger Algorithm (IHBA) with chaotic mapping and nonlinear parameters was combined with Mixed-Integer Programming (MIP) in a bi-level framework. RESULTS: Using real data from Eastern Inner Mongolia, IHBA achieved a fitness of 0.424, outperforming PSO, GA and original HBA by 20.3%, 14.9% and 8.0%. The optimal system included 47.9 MW wind, 94.3 MWh battery, 23.8 MW electrolyzer, 38.9 t H₂ storage, 13.7 MW fuel cell and 7.1 MW H₂-blended turbine, at a minimal cost of 4.57×10⁸ CNY. Annual H₂ output reached 68,900 Nm³, storage level 0.2-0.8, LCOE 0.438 CNY/kWh and payback period 18.4 years. Additionally, comparisons across multiple energy sources, sensitivity and robustness analyses, and ablation experiments indicate that the system configuration is stable and the algorithm optimization performs reliably, providing a solid reference for design. CONCLUSION: The IHBA-MIP framework is efficient and cost-effective for power-hydrogen system design, aiding large-scale renewable integration.
The performance disparities and operational characteristics among diverse devices within low-voltage distributed photovoltaic (PV) systems lead to variations in fault manifestations. The limited sampling frequency inherent in conventional approaches fails to capture critical fault features and temporal information in a timely manner. Moreover, the presence of periodic fluctuations in PV fault data contributes to persistently high false alarm rates. This paper proposes a high-frequency, minute-level data acquisition framework and a lightweight edge diagnosis algorithm for low-voltage distributed PV systems based on a temporal graph convolutional network. The proposed approach employs intelligent PV edge terminals to enable high-frequency data acquisition from distributed PV generation units. A graphical representation of the low-voltage distributed PV plant is constructed, from which dynamic temporal features of PV generation data are extracted using temporal convolutional layers, while topological correlations among PV devices are captured through graph convolutional layers. This enables the spatiotemporal joint modeling of fault characteristics. To accommodate the computational constraints of edge devices, a customized adaptation of the MobileNet-V3 architecture is introduced. By integrating attention mechanisms and implementing layer pruning, the model is tailored for enhanced performance in photovoltaic fault classification, thereby achieving lightweight edge diagnosis. Experimental results demonstrate that the proposed algorithm accurately diagnoses output voltage fluctuation faults in PV inverters and effectively identifies abnormal phase voltage fluctuation faults. It achieves high precision, recall, and F1 scores across various fault types. The model exhibits rapid training convergence with a low loss function value, satisfying the requirements for lightweight edge diagnosis in low-voltage distributed PV systems.
INTRODUCTION: Fire-proof oil systems in thermal power plants are prone to pump wear, leakage, oil degradation, and valve jamming, threatening operational safety. Real-time fault diagnosis is required. OBJECTIVES: To develop a real-time and accurate fault diagnosis method for fire-proof oil systems. METHODS: A multi-sensor monitoring framework integrating pressure, flow, temperature, and oil-quality signals was established. A lightweight deep learning model based on multi-source feature fusion was constructed, with a confidence-driven decision mechanism to reduce the influence of low-quality samples. Experiments were conducted on an open hydraulic-system dataset under multiple operating conditions. RESULTS: The proposed method achieved accuracies of 0.96–0.93 and AUC values of 0.98–0.97. Average diagnostic time was 3.17–3.58 ms, with alarm delays of 49.36–54.82 ms. The model outperformed MBDNN and TDANet in accuracy, efficiency, and robustness under noise. CONCLUSION: The proposed framework enables reliable real-time fault diagnosis of fire-proof oil systems and supports intelligent operation and maintenance in thermal power plants.
The reliability of power information networks is paramount for smart grid security, yet incomplete data transmission paths due to link failures or cyberattacks critically impair topology awareness and operational decision-making. To address the limitations of existing methods in handling the dynamic and structurally complex nature of these networks, this paper proposes a Topology-Enhanced Dynamic Perception Graph Neural Network (TEDP-GNN) for accurate path completion. The core methodology revolves around three integrated innovations. First, we introduce a topological deep learning framework that models high-order node interactions and network connectivity patterns beyond pairwise relationships. This is achieved by employing simplicial complexes to capture multi-node dependencies inherent in power data routing structures. Second, a novel dynamic perception mechanism is designed to continuously monitor link state changes. This mechanism utilizes a temporal gating unit that ingests real-time network alert data, enabling the model to adaptively reweight message-passing pathways in response to topological disturbances such as failures or attacks. Finally, the task is formulated as a topology-constrained optimization problem. Within this framework, we deploy a hybrid attention module that simultaneously computes both node-level and path-level attentions, alongside a multi-dimensional feature aggregation strategy that synthesizes information from topological embeddings, dynamic states, and historical transmission patterns. Preliminary validation on simulated power communication topologies demonstrates that TEDP-GNN significantly outperforms baseline models in predicting missing links, showing marked improvements in precision and recall. The model provides a robust, topology-aware solution for maintaining data path integrity, thereby enhancing the situational awareness and resilience of power information infrastructures.
During the operation of power-system industrial control and monitoring platforms (e.g., SCADA/EMS and substation automation), strong coupling among components and multi-source heterogeneous data often lead to alarm flooding and complicate root cause identification. To address this, this paper proposes an intelligent abnormal-alarm identification and on-site operation behavior analysis method, combining artificial intelligence with an abductive inference framework. Under fixed parameters, the method first aggregates raw alarm streams by events to enhance structure and interpretability. Then, a diagonal-covariance Gaussian Hidden Markov Model (HMM) is trained with normal data, and a path-deviation metric ranks root cause candidates. Multi-source evidence chains—integrating temporal, network, and semantic features—further improve inference interpretability for grid operation and maintenance. Using annotated operation logs, four quantitative metrics (MTTA, MTTR, action rate, consistency) assess the link between model outputs and actual handling behaviors. Experiments on five test sets show the method achieves a 78% alarm compression rate and a 0.43 average silhouette coefficient. Top-1 and Top-3 root cause localization hit rates are 71.8% and 88.5%, with path score fluctuations under 0.05 nats. The average MTTA and MTTR are 186s and 792s, with an 84% action rate and 72% consistency. These results confirm the method’s effectiveness in mitigating alarm flooding, improving root cause localization, and supporting on-site decision-making in power-grid operational scenarios.
Smart grids, increasingly reliant on information and communication technologies (ICT), are vulnerable to complex cyberattacks, thereby mandating the deployment of intelligent and adaptable intrusion detection systems (IDS). However, the efficacy of existing IDS techniques is frequently constrained by their limited capacity to extract distinguishing features from the high-dimensional, heterogeneous data characteristic of grid operations. In order to overcome this, we suggest a novel intrusion detection model that uses a convolutional neural network (CNN) to automatically extract hierarchical features from network traffic. The suggested CNN model outperforms conventional signature-based and SVM-based techniques with an accuracy of 98.8%, precision of 98.6%, and recall of 99.3% using the KDD-CUP99 dataset. Validation on a semi-realistic dataset from the IEEE 14-bus system, which uses IEC 61850 communication protocols, shows that it is 97.3% accurate. This means that it works well when physical and cyber layers are combined. Feature importance analysis shows that cyber-layer features, such as the continuity of GOOSE sequence numbers, are very important for detection. This research introduces a feature-learning-based intrusion detection system (IDS) framework. It works well and shows potential for practical use in improving the cybersecurity of smart grids.
