The traditional dissolved gas analysis （DGA） in oil-immersed transformer fault diagnosis is unable to effectively utilize fault information, and the imbalance of fault sample types results in poor diagnostic model performance. To address this, a transformer fault diagnosis technique based on data augmentation and fault feature optimization, named SCNGO-SVM-AdaBoost, is proposed. Firstly, to handle the imbalanced sample dataset, the safe-level synthetic minority over-sampling technique （safe-level SMOTE） is used for data augmentation of the original transformer fault sample set. Then, kernel principal component analysis （K-PCA） is employed to optimize and extract fault features from the ratio-based oil chromatographic data. Secondly, the northern Goshawk optimization （NGO） algorithm is enhanced by integrating positive and negative cosine and refraction reverse learning strategies. The stability of this algorithm is verified using test functions and optimization capability is improved by the diagnostic algorithm. Through a comparative analysis of actual fault diagnosis using the original samples and other methods, the results show that this approach can effectively improve the performance of transformer fault diagnosis.

False data injection attacks pose serious security threats to cyber-physical power system. Due to the class imbalance property between the attacked samples and the normal samples, machine learning detection methods tend to predict the majority of classes, which affects their detection accuracy of the attacks. Therefore, a false data injection attack detection based on Focal Loss^IM Transformer is proposed. Transformer utilizes itself attention mechanism to capture long-term dependencies in data, thereby identifies imbalanced false data injection attack data. Focal Loss^IM enhances the detection method's ability to identify imbalanced data by introducing modulation factors to better match the distribution and characteristics of false data injection attack samples, thereby improving the detection accuracy of the detection method for attacks.The effectiveness of the proposed method is verified through simulations on IEEE 14⁃node system, IEEE 30⁃node system, and IEEE 57⁃node system. Compared with traditional loss functions and other detection methods, the proposed method exhibits better generalization ability and recognition ability for minority classes, with high recognition accuracy and low false alarm rate.

With the growing integration of renewable energy, load frequency control （LFC） in power systems faces security risks from false data injection attack （FDIA）. Existing detection methods struggle to differentiate control input attacks from measurement attacks, compromising system stability and security. This paper develops a state-space model for LFC incorporating renewable energy and energy storage systems and analyzes the impact of FDIA on system dynamics. A state-space decomposition method is employed to decouple attack signals into control input and measurement attacks, improving detection accuracy. A sliding mode observer-based attack estimation method is proposed for real-time detection. Additionally, an attack-resilient control （ARC） strategy is designed using H_{\mathrm{\infty }} control theory to enhance system robustness. Simulations show that the proposed method reduces the attack estimation mean squared error by nearly 30% and significantly improves frequency response stability compared to traditional methods. These results demonstrate the method′s effectiveness in detecting FDIA and enhancing power system security.

As the international situation becomes increasingly complex and the widespread adoption of intelligent and information technologies in power systems continues, power systems have emerged as prime targets for cyber-attacks, highlighting the critical importance of cybersecurity. To provide a powerful tool for studying the evolution of security risks in power network and support cyber risk prediction and command work, a wargaming-based method is proposed for modeling the evolution of power network security. A deduction framework based on the concepts of composability and minimal operation units is constructed through strategic-level wargaming simulations to achieve large-scale power network security scenario modeling and improve simulation efficiency, while effectively identifying potential risks in the network. Experimental validation demonstrates that the proposed method exhibits superior performance in large-scale power information network security wargaming, providing robust support for research on the evolution of power network security risks, optimization of cybersecurity strategies, and command operations in network security.

In order to improve the operational efficiency and voltage quality of distribution networks, an optimization strategy is proposed based on a state-response framework, combining deep reinforcement learning with the distributed generalized fast dual ascent （SAC-GFD） algorithm. Firstly, the soft actor-critic （SAC） algorithm is employed to model the distribution network operation optimization problem as a Markov decision process （MDP）. The agent interacts and explores in an environment with renewable energy fluctuations and load uncertainties, obtaining a control strategy that is robust to uncertain environments. The distribution network operation optimization problem is thus transformed into a Markov decision process, enabling the training of an agent capable of quickly outputting the optimal active and reactive power for distribution network equipment. Secondly, the current distribution network's power flow distribution, node voltage states, and active and reactive voltage sensitivity matrices are calculated. Then, based on the current state of the distribution network, users apply a distributed method to calculate their optimal load adjustments to ensure the safe operation of the network. Finally, the simulation results on the IEEE 33⁃node system demonstrate that compared to traditional deep reinforcement learning algorithms, the proposed method more effectively reduces network losses and node voltage deviations while achieving faster training speeds and better optimization results.

Facing the flexible and adjustable resources inside the intelligent distribution area of distribution network, a Stackelberg game pricing strategy and electric vehicle ( EV) charging management strategy for distribution network operator based on scenario generation method is proposed to guide the user-side flexible resources to realize efficient interaction with the distribution network through market-oriented means. Firstly, based on conditional density networks (CDNs), Gibbs sampling method is used to predict EV load, which takes into account the joint probability distribution of EV users′ charging characteristics, and EV aggregation model is constructed to reduce the model dimension. The distribution locational marginal pricing (DLMP) based on the second-order conical optimal power flow model of distribution network is obtained, and on this basis, the Stackelberg game pricing strategy and user-side incentive mechanism of distribution network operator are constructed. The bi-level Stackelberg game model is transformed into a mixed-integer linear programming problem using the Karush-Kuhn-Tucker (KKT) conditions and optimal duality theory for solution, and finally the global optimal pricing strategy is obtained, and the power allocation module is used to realize the reasonable distribution of charging power for EV users. Simulation results based on improved IEEE 33-node power distribution system verify the effectiveness of the proposed strategy.

The integrated wind-solar-hydrogen storage coupling system is one of the important means to improve the consumption of new energy and promote the development of hydrogen economy, but the high randomness of wind-solar-hydrogen load and the slow solution speed of the system model bring difficulties to the optimal scheduling of the system. Firstly, in order to analyze the uncertainty factors in the wind and solar load more accurately, the variational mode decomposition （VMD） technology is adopted, and the dynamic time warping （DTW） algorithm is combined with the K-medoids clustering method to obtain the uncertainty of the wind and solar load and the system operation scenario. Then, a wind-solar-hydrogen storage coupling system model and a multi-time-scale optimal scheduling model based on robust MPC-PSO are constructed. Then, the combination of robust MPC and PSO algorithm is used to realize the fast decoupling solution of rolling optimization of system operation. Finally, a simulation example of wind-solar-hydrogen-storage coupling system is established for analysis. The results show that the proposed method improves the economy of hydrogen production and the wind and solar absorption rate, and improves the solution efficiency compared with a single robust MPC.

Accurate short-term power load forecasting is essential for the formulation of daily power generation plans and the real-time dispatching of new power systems. In order to obtain accurate and reliable load forecasting results, this paper proposes an Attention-GAT-LSTM intelligent algorithm targeting the temporal and uncertain characteristics of real power load data, and applies it in practical new power systems. In raw data processing, this algorithm innovatively integrates a self-attention mechanism, introduces a data processing unit to assign weights, and employs a skip-connection mechanism to prevent overfitting. The processed data is transmitted to a graph attention network (GAT) for spatial node feature extraction, then passed to a long short-term memory (LSTM) network for temporal feature extraction. Through forward propagation, back propagation, and gradient descent methods, the weights and biases of the LSTM layer are iteratively updated, effectively minimizing information loss during iterations and highlighting key time-point information. Comparative analyses with various models demonstrate the method’s high prediction accuracy in short-term (hour-level) power load forecasting, providing data support for the operational dispatching, planning and construction of new power systems.

The sea has extremely abundant wind resources, and studying offshore wind power hydrogen production technology and load side demand response is of great significance for wind power consumption and smoothing the volatility of offshore wind power. Therefore, a source-load multi-time scale optimization scheduling strategy for a hydrogen containing integrated energy system （IES） including hydrogen production from multiple demand responses and offshore wind power is proposed. Firstly, the operating mechanism of the offshore wind power hydrogen production （OWHP） system is explored, and an OWHP model is constructed, which includes wind power hydrogen production, seawater desalination, hydrogen compression, hydrogen transmission pipelines and gas hydrogen storage. A hydrogen energy multi energy utilization model is also constructed, which includes gas hydrogen blending, hydrogen methanation and hydrogen fuel cells. Secondly, the regulation characteristics of load side response resources at different time scales are analyzed, and a multi demand response model is proposed. Finally, in order to reduce the impact of prediction errors in offshore wind power on IES operation, a multi-time scale of three stages optimization model consisting of day ahead, day ahead, and real-time is proposed to smooth out system power fluctuations. The simulation results of the case study show that the proposed model can effectively absorb offshore wind power resources, improve IES economy and low-carbon performance, and alleviate the impact of source and load uncertainties on system operation.

As the penetration rate of wind turbines increases, the power system inertia level decreases year by year. At the same time, the frequency response has partition characteristics, and it is more flexible and accurate to evaluate the power system inertia in terms of region. Therefore, a dynamic partition inertia estimation method is presented based on one-dimensional squeeze and excitation residual neural network （1D-SE-ResNet）. Firstly, the trend and value approximation of frequency distance are computed, and the k-means clustering method is used to dynamically partition the system and the number of partitions is determined by the S-C metric. Then, the 1D-ResNet is improved by adding the squeeze and excitation module, which enables it to provide weights for each channel to enhance the network performance. The regional cluster centre node frequency and frequency change rate data under different inertia levels and load perturbations of the system are collected as one-dimensional feature inputs, and the regional effective inertia is the output, training the network to achieve regional inertia estimation. Finally, simulations are carried out on the IEEE 39-node and IEEE 118- node systems containing wind power. The results show that the trained 1D-SE-ResNet can achieve accurate evaluation of regional inertia based on dynamic partitioning.

In recent years, load disaggregation methods based on deep learning have been widely applied. However, current researches mainly focuse on inputs from traditional Euclidean space sequences, which struggle to accurately capture temporal correlations during the operation of electrical devices, thereby reducing the resolution accuracy of electrical equipment analysis. Furthermore, the switching actions of household appliances may have long-distance impacts in time series data, yet existing models often overlook the long-distance dependency issues in load data. To address these challenges, a non-intrusive load disaggregation model is proposed based on graph convolutional network (GCN) and bidirectional long short-term memory (BiLSTM). This method transforms the total load sequence into graph-structured data containing nodes and edges using graph theory, effectively considering inter-node correlation features and utilizing GCN for feature extraction. Additionally, BiLSTM neural networks are introduced to handle the limitations of long-term time series data. Case analysis demonstrates that the proposed model significantly outperforms traditional methods in terms of disaggregation accuracy and effectiveness.

In response to the problem of feature aliasing in the recognition of composite power quality disturbances (PQD), this paper constructs a frequency decoupling layer based on discrete cosine transform (DCT) as the input layer of the deep network, and a deep frequency attention network (DFAN) suitable for composite PQD recognition is proposed. Firstly, based on DCT theory, a frequency domain decoupling layer is constructed to decompose the PQD signal, reducing feature aliasing between different disturbances. Secondly, PQD components with different frequencies are used as input for multiple channels, and high-dimensional features containing different frequency information are extracted using convolutional layers. Then, attention mechanism is used to adaptively allocate weights for each frequency channel, achieving frequency component selection and noise removal. Finally, the calculation process of the proposed frequency domain decoupling layer is optimized. Some parameters are calculated and saved in the network structure during network construction, which avoids repeated calculations during the forward propagation process of network training and reduces the training time of the network. The simulation experiments show that compared to other common PQD recognition methods, the proposed method can effectively handle the mixed problem of multiple PQDs, and has higher accuracy in identifying composite PQDs.

The large number of distributed photovoltaic (PV) access on the user side increases the difficulty of net load forecasting. In order to improve the forecasting accuracy, a short-term net load probabilistic forecasting model based on weather classification and neural network is proposed. First of all, in order to better describe the meteorological conditions, the weighted sum of cloud cover and ambient temperature that have the greatest influence on PV output is carried out to construct a comprehensive meteorological impact factor, and a weather classification method based on the volatility of meteorological impact factor is proposed. At the same time, the maximum information coefficient is used to help the model select the input characteristics in different time frames in view of the time difference of the influence of meteorological factors on the net load. Finally, using the sample selection results as input, probability prediction is made by quantile regression and gated loop unit (QR-GRU) neural network, and probability density curve is generated by kernel density estimation. The simulation results show that the model has good prediction performance.

With the further promotion of the "Double Carbon" goal, China's wind power industry has developed rapidly in recent years. How to accurately and effectively predict wind power is crucial to realize the safe grid connection of wind turbines and maintain the stable operation of the system. Aiming at the problems such as input feature redundancy, insufficient generalization ability, and insufficient capture of inherent wind power output characteristics in existing wind power forecasting methods, a short-term wind power prediction model based on feature selection and similarity based fusion using long short term memory network-attention mechanism (LSTM-AM) is proposed. Firstly, least absolute shrinkage and selection operator (Lasso) regression is used to optimize input features to reduce redundancy. Then, the long short term memory network-attention mechanism（LSTM-AM） fusion network model is established using long short term memory and attetion mechanism. Finally, similar historical samples are extracted by Euclidean distance calculation and weighted with the model output as the final predicted value. The experimental results show that compared with traditional methods, the prediction accuracy of the proposed method is significantly improved, and it shows higher accuracy in wind power prediction, which can provide support for power system planning and operation and the in-depth application of renewable energy.

The new energy source is connected to the grid through the inverter, and the voltage amplitude, phase, frequency and other physical quantities of the grid-connected inverter system have strong interactive coupling in electromagnetic, electromechanical and other time scales. It is a hot issue to accurately grasp the dominant behavior characteristics of synchronous stability of grid-connected inverters. Therefore, a multiple density-based spatial clustering of applications with noise (DBSCAN) algorithm is proposed to analyze the physical characterization of the synchronous stability of grid-connected multi-inverter systems. Firstly, the grid-connected models of grid-following inverters and grid-forming inverters are established, and the synchronization characteristics of their synchronous control units are analyzed. Secondly, the proposed algorithm is used to cluster and classify simulation data of the phase difference between the grid-connected inverter and the common connection point, the inverter power angle, the system frequency and so on. Finally, the simulation analysis shows that the power angle of the grid-connected inverter system is gradually restored to the original operating state and the new steady state. In addition, the system synchronization stability is consistent with the frequency stability. The successful phase locking of the grid-following inverters synchronous control unit can characterize the synchronization stability of the system, while the grid-forming inverters synchronous control unit cannot.