Graph comparison efficient conditional generative adversarial networks for parameter identification of synchronous generators

作     者：Yin, Linfei Wang, Zixuan 

作者机构：Guangxi Key Lab Power Syst Optimizat & Energy Tech Nanning 530004 Guangxi Peoples R China 

出 版 物：《EXPERT SYSTEMS WITH APPLICATIONS》 (Expert Sys Appl)

年 卷 期：2025年第269卷

核心收录：

学科分类：1201[管理学-管理科学与工程(可授管理学、工学学位)] 0808[工学-电气工程] 08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Natural Science Foundation of China Natural Science Foundation of Guangxi Province (China) [AA22068071] 

主　　题：Synchronous generators Parameter identification Conditional generative adversarial networks Graph comparison learning 

摘      要：Synchronous generator parameters have nonlinear characteristics, and traditional parameter identification methods cannot effectively capture the dynamic nonlinear behavior of synchronous generators in actual operation. Therefore, the parameters recognized by traditional methods have significant deviations from the actual operating parameters. In this work, an algorithm for graph comparison efficient conditional generative adversarial networks (GCECGANs) is proposed. Efficient neural network is an network structure specialized in processing image and graph data, fully utilizing the local information and global structure of image and graph data, with high computational efficiency and low resource consumption. The graph comparison learning network improves feature extraction and representation learning by optimizing the relative distance between data elements of the graph structure, enhancing the understanding of data similarities and differences in the model. Based on conditional generative adversarial networks, the efficient neural network focuses on efficient encoding of graph data, while the graph comparison learning network focuses on optimizing the network parameters by comparing different graph samples to enhance the generalization capability. By designing loss functions and optimization strategies, GCECGANs effectively improve the ability to handle graph-structured samples and data adversarial. Finally, the effectiveness of GCECGANs in generator parameter identification is verified in three algorithms. The accuracy and practicability of GCECGANs are effectively improved compared with traditional parameter identification algorithms.