Abstract: Aiming at addressing challenges such as the extensive data requirements for feature extraction, prolonged model training times, and reduced computational efficiency in present assessment, a method for evaluating transient voltage stability in wind power integration systems is proposed based on the combination of kernel principal component analysis(KPCA) and chaos particle swarm optimization (CPSO) and back-propagation (BP) neural network. Firstly, the raw feature set is collected according to the input features, followed by nonlinear data processing using KPCA to extract the optimal feature set. Then, the reduced dimension feature set is used as the input of the CPSO-BP neural network for supervised learning. The obtained model is categorized according to the margin of critical fault removal time. The classified samples are used for transient voltage stability evaluation and critical fault removal time margin prediction of wind power grid-connected systems. Finally, the simulation analysis results show that reducing the dimension of the input features, retaining the important input features, and eliminating the redundant features, not only simplifies the model but also improves the accuracy and calculation efficiency of network evaluation.

Key words: wind power grid connection, kernel principal component analysis algorithm, dimension reduction, CPSO-BP neural network, transient voltage stability assessment