基于动态响应特征学习的综合负荷模型构成在线辨识OA

Real-time and accurate identification of load composition is of great significance for power system simula-tion and analysis.Current identification processes based on conventional optimization methods struggle to handle multidimensional temporal characteristics and are computationally intensive,leading to insufficient identification accuracy and an inability to meet the demands of online applications.To address this challenge,an online load com-position identification method suitable for active integrated load models is proposed,integrating the feature weight-ing capability of attention mechanisms with the feature extraction capability of convolutional neural network(CNN).Firstly,an active integrated load model incorporating photovoltaics and energy storage is proposed from a mechanis-tic perspective.Subsequently,a feature extraction network integrating multi-scale convolution and attention mecha-nisms is constructed to capture heterogeneous load features in parallel and highlight critical information.Finally,key load nodes are screened based on the ratio of global parameter sensitivity among load nodes as an evaluation met-ric,and target nodes are identified accordingly.Case study results demonstrate that,compared to existing methods,the proposed approach achieves higher identification accuracy and robustness,meeting the requirements for online security analysis in most power system operational scenarios.