During part-load operation, francis turbines often experience adverse phenomena such as degraded internal flow patterns and increased vibration. To gain deeper insight into the flow characteristics within different components of the turbine under these conditions, this paper conducts a three-dimensional transient numerical simulation of the full flow passage for a specific francis turbine unit. A systematic comparative analysis of the flow characteristics under three typical operating conditions is performed. The results indicate that significant flow instability occurs during part-load operation. Specifically, a large positive attack angle at the runner inlet leads to flow separation, the shear between the jet flow from guide vanes at small openings and the main stream generates large-scale turbulent vortex structures, and the residual swirl at the runner outlet results in the formation of an eccentric vortex rope in the draft tube, thereby inducing intense pressure pulsations with substantially increased amplitude. The results of this study contribute to a better understanding of the complex unsteady flow phenomena inside francis turbines operating under part-load conditions, providing a theoretical basis for the safe operation and structural optimization of francis turbines.

To address the issues of noise interference, poor generalization of single models, and difficulty in adapting to complex peak shaving conditions in traditional wall temperature measurement of coal-fired units, a hybrid model for wall temperature prediction based on whale optimization algorithm-variational mode decomposition-convolutional neural network-bidirectional gated recurrent unit(WOA-VMD-CNN-BiGRU) is proposed. This method follows a technical route of data decomposition-feature extraction-time series prediction. firstly, the Pearson correlation coefficient method is used to screen input features strongly related to wall temperature. Then, the whale optimization algorithm(WOA) is employed to optimize the hyperparameters of variational mode decomposition(VMD) to effectively decompose the original wall temperature signal. finally, a CNN-BiGRU hybrid model is constructed, integrating the local feature extraction capability of convolutional neural network(CNN) and the time series modeling ability of bidirectional gated recurrent unit(BiGRU) for prediction. Using actual operation data of power plant units as samples, the performance of the proposed model is compared with that of the BiGRU model in single-step and multi-step prediction scenarios. The results show that in both single-step and multi-step predictions, the prediction error of the proposed model is significantly reduced, and its ability to explain data is significantly improved. This indicates that the proposed method can effectively enhance the accuracy and stability of wall temperature prediction, providing reliable technical support for wall temperature monitoring and over-temperature early warning of coal-fired units.

Multimodal data throughout the entire life cycle of substation project lead to information silo. The 3D modeling software of departments such as equipment suppliers, design institutes and construction units have not been unified. The problems of data format conversion and integration are obvious. Based on knowledge graph(KG) and retrieval-augmented generation (RAG), this study proposes an Enhanced grid information model(EGiM), which integrates multiple 3D data formats of substation project through the GraphRAG method. KG builds models of complex power grid relationships to obtain various connections among substation, transformers and distribution network. RAG improves the accuracy of relationship description by extracting relevant data from the knowledge base and combining it with the 3D model. This study designs semantic mapping rules adapted to power grid multimodal data, realizing standardized semantic alignment of heterogeneous formats, and constructs a dynamic weight retrieval mechanism based on vector and graph to accurately get implicit equipment associations. This method increases the data integration success rate to 98.2% and the cross-format call success rate to 95.6%, improves data processing efficiency, and provides core technical support for the efficiency improvement of substation engineering design collaboration and fault diagnosis.

Reliable arc suppression of ground faults is a critical guarantee for the safe and reliable operation of distribution networks. Generally, in the analysis of ground faults, it is assumed that the distribution network is three-phase balanced. Existing methods perform poorly in suppressing arcs when considering line impedance and asymmetrical loads in three-phase unbalanced distribution networks during single-phase ground faults. To address this issue, this paper proposes a hybrid method that combines current arc suppression and voltage arc suppression techniques. A distribution network model that accounts for asymmetrical line parameters and load imbalances is established and analyzed. By injecting current twice, the threshold value of ground conductance and the reference values for improving both arc suppression methods are calculated. When the ground conductance is less than the threshold, the voltage arc suppression method is used, and when the ground resistance is greater than or equal to the threshold, the current arc suppression method is employed. The proposed method can mitigate the impact of asymmetrical line parameters, ground impedance magnitude, and load imbalance, reducing the fault current to below the limit. finally, MATLAB/Simulink simulations verify the reliability and feasibility of the proposed arc suppression control strategy.

Aiming at the problem that parameter mismatch impairs the current loop performance in the sensorless control system of surface-mounted permanent magnet synchronous motors(SPMSM), this paper proposes a deadbeat current predictive control method based on online parameter identification.first, in accordance with the characteristics of permanent magnet motor control systems, deadbeat current predictive control is employed to enhance the dynamic response performance of the current loop. Second, an online identification approach for motor parameters including inductance, flux linkage, and resistance are proposed, which integrates the multi-innovation theory with the least squares method. finally, the identified results are fed back to the deadbeat current predictive controller in real time, realizing the online update of control parameters and the compensation for parameter mismatch. The simulation results indicate that this method can effectively improve the steady-state performance of the current loop controller under the condition of motor parameter perturbation.

To address the issues of low accuracy and poor real-time performance in distribution network fault diagnosis, this paper proposes a fault diagnosis method based on the lightweight dual branch feature fusion-AlexNet(DBff-AlexNet). firstly,the collected three-phase voltage and current data from each node of the distribution network are converted to the two-dimensional time-frequency images based on the continuous wavelet transform. These images are then input into DBff-AlexNet for fault diagnosis. Secondly,in the proposed DBff module, the fire skip module adopts squeeze and expand layers with small convolutional kernels to reduce the number of parameters while extracting local features. To preserve information lost in squeeze layers, skip connections are employed for concatenating input features with features from both squeeze and expand layers. Moreover, depthwiseseparable dilated convolutions with varying dilation rates areproposedto expand the receptive field without increasing the parameter count, thereby enhancing global feature extraction capability. finally, the Grad-CAM heatmap method is used to visually illustrate the model's attention regions in time-frequency diagrams, improving the model's interpretability. The simulation results show that this method has higher fault diagnosis accuracy and effectiveness, ensuring the accuracy of fault diagnosis in distribution networks while reducing model complexity.

Machine learning algorithms are widely applied in extracting key featuresoftransient stability in power systems, but their poor interpretability has become a key constraint. Therefore, this paper proposes a key feature extraction and interpretability analysis method based on the improved modified light gradient boosting machine model optimized by Optuna. This method introduces the Optuna automatic hyperparameter tuning algorithm to automatically and optimally adjust the hyperparameters of the model, thereby enhancing the performance and speed of model optimization. A hybrid attribution analysis framework based on SHAP and LiME is proposed. The global feature importance ranking is obtained through SHAP values, and LiME is used to conduct local analysis on individual samples for verification. The test results on the IEEE 10-machine 39-node system show that the proposed method has higher evaluation accuracy and faster evaluation speed, and can provide reasonable and effective explanations for the extracted key features.

Large-scale high-impedance cascaded transmission corridors have been built in some regions of China, where traditional out-of-step islanding devices often underperform.This paper proposes ahigh-impedance cascaded corridorsislanding strategy based on slow coherency theory. firstly, based on slow coherency theory, the electrical connections of nodes are classified. Characteristics of high resistance cascade channel splitting are analyzed accordingly, and indicate that this type of channel has the problem of confusion in islanding division after the splitting device is activated due to edge nodes. Secondly, for the chain impedance structure of high impedance cascade channels, a slow coherency and islanding strategy suitable for high impedance cascade channels is proposed by combining natural breaking point method.finally, the IEEE 39 node standard calculation system is used to simulate and verify the proposed strategy with a provincial power grid in western China.Simulation results show that the proposed strategy can effectively search for the optimal solution section, and has good adaptability and effectiveness for power grids with high resistance cascade channels.

Aiming at the difficulties in high-impedance fault(Hif) identification, such as weak fault characteristics, difficulty in constructing classification boundaries due to sample scarcity, and high misoperation rates under unknown disturbances, a new Hif identification method fusing generative adversarial networks(GAN) and deep metric learning is proposed. This method adopts time-domain data augmentation techniques to randomly perturb and inject noise into the original signal, in order to expand the fault samples and enhance the model's adaptability to noisy environments.first, a GAN adversarial training mechanism is introduced to generate simulated unknown disturbance samples, addressing the incompleteness of negative samples in training data. Second, a deep encoder based on gated recurrent Unit(GRU) is constructed to accurately capture key sequential physical characteristics of Hif, such as the zero-rest phenomenon and asymmetry. Third, an improved triplet loss function combined with an online hard sample mining strategy is adopted for end-to-end training. By incorporating GAN-generated simulated disturbances as strong negative samples, the model is forced to push unknown disturbances away from normal and fault clusters in the feature space. finally, prototypical networks are utilized to adapt to small sample scenarios, establishing standard feature archives for each category. A two-stage threshold discrimination strategy is then employed to achieve precise identification of Hif and effective rejection of unknown anomalies. Simulation verification and comparative analysis demonstrate that the proposed method not only performs well in Hif recognition accuracy,but also achieves zero error response to unknown system disturbances with the enhanced boundary defense capability of GAN, significantly improving the robustness and security of the model in open set environments.

To solve the problem of identifying series arc faults with feature aliasing in low-voltage power scenarios, this paper proposes a time-variance-accuracy coefficient random forest(TVA-Rf) model for low-voltage series arc fault detection. First, analysis of sample feature sensitivity reveals that the core issue stems from reduced sensitivity under specific conditions, leading to inaccuracies in traditional arc detection methods. Secondly, the random forest model is used as the core recognition model due to its anti-interference properties. finally, a TVA coeffit is constructed to achieve multi-objective optimization of random forest hyperparameters, and the random forest model is trained using low-voltage load samples. Through experimental testing of low voltage series voltage, the fault identification method achieves an accuracy rate of 99.97%, and surpasses traditional methods in both accuracy and calculation speed. This proves that the method overcomes feature overlap interference and achieves accurate identification of low voltage series arc faults.

in order to solve the problem of bus voltage oscillation caused by the fluctuation of light intensity and load abrupt change in photovoltaic storage DC microgrid, an iLADRC strategy is proposed. In this paper, a series correction link is added to the traditional improved linear active disturbance rejection control, and the observation accuracy of the extended state observer is improved by eliminating the steady-state error of the dilated state observer. Secondly, the feedforward active disturbance rejection control is introduced into the voltage outer loop control, and a positive feedback channel is drawn from the voltage outer loop to compensate the reference current output of the voltage outer loop, so as to realize the error-free tracking of the signal, improve the control performance and response speed of the hybrid energy storage control system, and reduce the overshoot and adjustment time of the DC bus voltage. finally, the simulation model of DC microgrid is built by MATLAB, and the results show that the overshoot of the improved linear active disturbance rejection control is smaller and the stabilization time is shorter, which proves the effectiveness of the control strategy proposed in this paper.

The state of health(SOH) of electrochemical energy storage batteries is an important indicator for evaluating the degree of battery performance degradation and safety. In order to improve the engineering adaptability and accuracy of the model,this study takes 20 Ah lithium iron phosphate batteries as the research object and proposes an evaluation method that integrates multiple feature parameters: partial charge and discharge data and characteristic frequency points of the electrochemical impedance spectrum are used as input features, a BP neural network model optimized based on the whale algorithm is constructed, and the two types of feature parameters are further fused and input into the model for SOH evaluation. The experimental results show that the evaluation accuracy of the model with fused features is significantly improved: its mean absolute percentage error(MAPE) reaches 1.09%. Compared with the single feature input model, MAPE decreased by 39.8%(compared to impedance method) and 43.5%(compared to voltage method). This study provides an effective method for performance evaluation and system management of lithium iron phosphate batteries for energy storage.

With the rapid increase in the penetration of renewable energy generation, the structure of the power system gradually shifts from being dominated by synchronous generators to a hybrid system composed of both grid-connected and standalone renewable energy sources. Due to the complex dynamic interactions between different generation units, which exhibit strong nonlinear coupling characteristics, the new challenges arise for the synchronization stability of the system. To address these issues, a mathematical model for the transient stability analysis of generation units, considering different synchronization control methods, is established to reveal the underlying instability mechanisms. The phase-plane method is then used to analyze the impact of the power command of the grid-connected renewable energy, virtual synchronous control parameters, and the phase-locked loop control parameters of the standalone renewable energy on the system's synchronization stability. Based on these analyses, design principles for controller parameters suitable for the stable operation of hybrid systems are proposed. finally, simulation results under fault scenarios are used to validate the effectiveness of the proposed model and the correctness of the parameter design method, providing a theoretical basis for the stable operation and controller parameter tuning of high-penetration renewable energy power systems.

In order to achieve grid connection of the PV system according to the scheduling plan and configure mixed energy storage to regulate the online electrical energy, a hierarchical coordinated control strategy for the system is proposed based on the division of the battery and supercapacitor SOC areas. firstly, system level regulation makes timely adjustments to the grid connected dispatch power, providing reference values for matching PV and grid connected power for hybrid energy storage systems. Hybrid energy storage regulation allocates smooth power reference values for hybrid energy storage systems to achieve reasonable charge-discharge of batteries and supercapacitors. Then, a mathematical model of inverter is analyzed and control method is designed. finally, the effectiveness and feasibility of the proposed hierarchical coordinated control strategy are verified through system simulation, and the PV system is able to operate stably and grid friendly under different SOC.

Aiming at the challenge of balancing economic and environmental objectives in the multi-objective optimal scheduling of virtual power plants(VPPs), this paper proposes an optimization method based on the multi-objective artificial hummingbird algorithm(MOAHA). A scheduling model is established with the objectives of minimizing both operational cost and carbon emissions. The multi-objective artificial hummingbird algorithm is introduced to solve this model, leveraging its three foraging behaviors combined with an external archive and a dynamic crowding distance strategy to enhance the convergence and distribution uniformity of the obtained solutions. furthermore, the entropy weight-TOPSiS comprehensive evaluation method is employed to select the optimal compromise solution from the Pareto front. Experimental results demonstrate that the proposed method achieves a widely distributed Pareto solution set. The comprehensive optimal scheme yields an operational cost of 1.544 908 million yuan and carbon emissions of 19.464 0 t. Compared to the solutions obtained by the multi-objective grey wolf optimizer and the multi-objective genetic algorithm, the proposed method reduces operational cost by 2.57% and 33.02% respectively, and reduces carbon emissions by 65.2% and 72.1% respectively. This validates the effectiveness and superiority of the proposed method in simultaneously addressing economic and environmental goals. Additionally, while a purely economic-oriented scheme saves 59.38% in cost compared to the comprehensive optimal scheme, it results in a 465.19% increase in carbon emissions.

in power distribution network construction,drilling and pole-erecting machine as critical equipment, their intelligent operation accuracy and safety are significantly affected by unstructured and complex road environments. To address the obstacle detection problem of drilling and pole-erecting machine during unstructured road operations, this article proposes a detection method based on multi-sensor fusion. By integrating camera and LiDAR data with a decision-level fusion strategy, the method effectively enhances the accuracy and robustness of obstacle detection. Experimental results demonstrate that the proposed approach significantly reduces both the false detection rate and missed detection rate, providing strong support for the accurate and safe operation of drilling and pole-erecting machine.

To address the challenges of small-sample characteristics in transmission line fault data and the identification of single-phase-to-ground fault causes, this paper proposes a fault cause identification method based on multi-dimensional robust fisher feature selection(MDRffS). first, multi-dimensional features are extracted from fault recording data in the time domain, frequency domain, and time-frequency domain to construct a candidate feature set. By jointly considering feature discriminative ability, mutual information relevance, and robustness indicators, an MDRffS-based feature selection mechanism is developed to achieve effective feature screening and dimensionality reduction, yielding a set of high-quality feature vectors. On this basis, a multi-support vector machine(multi-SVM) identification model incorporating a priority-based decision strategy is constructed to accurately identify five typical fault causes, including lightning strikes, bird interference, wildfire, ice damage and wind-induced faults. finally, case studies based on actual power grid fault recording data demonstrate that the proposed method achieves an average identification accuracy of 92.314%, verifying its feasibility and effectiveness under small-sample conditions.

To balance the accuracy and complexity of transmission line insulator defect detection algorithms, a high-precision lightweight detection algorithm based on YOLOv8n is proposed. first, the C2f_MSCA module is constructed by embedding multi-scale coordinate attention(MSCA) into the csp bottleneck with 2 convolutions(C2f) module, replacing it in the backbone network to enhance feature extraction capabilities. Then, light symptotic feature pyramid network(LightAfPN) is introduced and optimized with depthwise separable convolution(DSConv) and CA for lightweight operation and improved attention to critical features. The neck network integrates the small target feature layer P2 while removing the redundant high-level feature P5 from the backbone, further reducing model complexity and enhancing small target extraction. The space to depth convolution(SPDConv) is introduced to compensate for the accuracy loss resulting from the removal of the P5 layer. finally, the differentiated component-wise modulated scylla-ioU(DCM-SioU) is designed to replace YOLOv8n's complete intersection over union(CioU), dynamically adjusting the penalty strategy during different training phases to optimize confidence loss. Experimental results show that the recognition model has achieved an average accuracy of 96.04% for the intersection to union ratio threshold of 0.5 for the insulator body and three different insulator defects. Compared with the benchmark model YOLOv8n, it improves by 3.59 percentage points and reduces the number of parameters by 29.5%, achieving a balance between model accuracy and complexity. This can provide reference for deploying object detection models at the edge.

To address the issues of complex procedures, such as time-consuming data processing, low automation levels, and significant environmental interference,a fully autonomous UAV inspection system for overhead linesbased on laser SLAM technology is proposed. The system fully integrates LiDAR, visible-light cameras, an iMU, and an edge computing module. Utilizing SLAM technology, it achieves real-time mapping of transmission corridors via laser point clouds, real-time pylon extraction, real-time pylon type matching based on an established pylon template library, and real-time waypoint matching for corresponding pylon types using a pre-built waypoint database, thereby completing detailed inspection route planning. The system can accomplish high-precision, high-efficiency channel and detailed inspection tasks in a single flight, significantly reducing inspection procedures and effectively enhancing the automation level of UAV inspection operations.

In the traceability of power data links, large language models(LLMs) rely on fixed corpora for pre-training, making it difficult to adapt to rapidly changing complex relational networks. Their multi-step reasoning is often accompanied by illusions and omissions, especially in dynamic data and long chain dependency scenarios, making it difficult to maintain stable and accurate results. To address these issues, this paper proposes a dynamic iterative framework for the traceability of power data links based on LLMs. it incorporates dynamic adjustment of reasoning depth and introduces entity pruning during entity expansion, thereby reducing redundancy and invalid paths while ensuring reasoning efficiency and enhancing link completeness. The experimental results indicate that the proposed framework significantly improves the accuracy and reliability of link traceability, offering a new solution for knowledge-enhanced reasoning in complex power data environments.