Research on Transient Stability Evaluation Method of Power System Based on Improved Convolutional Neural Network

Abstract

Transient stability analysis is a key link in the safe operation of power systems. However, traditional methods (such as time-domain simulation and direct methods) have problems such as low computational efficiency or limited applicability. Although artificial intelligence methods can enhance the evaluation speed, the existing shallow models have insufficient generalization ability in high-dimensional data classification and are mostly limited to binary stability determination, lacking quantitative evaluation. To this end, this paper proposes a transient stability evaluation method based on short-time disturbed trajectories and an improved convolutional neural network (CNN). Firstly, CNN is utilized to establish the mapping relationship between the short-term disturbance trajectory of the electrical quantity at the generator end and the transient stability of the system. Moreover, a sample matrix is constructed by considering the disturbance degree of the generator in the early stage of a fault to enhance feature robustness and reduce misjudgment and missed judgment. Secondly, optimize the network structure based on the inter-layer computing dimension of CNN to improve the accuracy of model evaluation. Furthermore, a composite model is constructed by combining the CNN feature extraction layer with the BP neural network. First, the samples are pre-classified, and then the transient stability margin is predicted to achieve quantitative evaluation. Finally, simulation results based on the IEEE 39-bus system demonstrate that the enhanced CNN model achieves 98.42% assessment accuracy, and maintains margin prediction errors below 3%. By enabling autonomous extraction of high-dimensional trajectory features, the proposed method overcomes the limitations of manual feature selection, offering novel insights for real-time security control in power systems.