电气技术2026，Vol.27Issue(2)：40-45,52,7.
结合生成对抗网络与支持向量回归提升发电厂励磁系统参数辨识的准确性
Improving the accuracy of excitation system parameter identification in power plants by combining generative adversarial network and support vector regression
覃莹 1王保国 1项文雅 1蒋宗良 1黄荣峰1
作者信息
- 1. 龙滩水电开发有限公司合山发电公司,广西 来宾 546501
- 折叠
摘要
Abstract
The existing methods for parameter identification of the excitation system in power plants usually have problems such as low identification accuracy and poor training stability.These issues lead to insufficient accuracy of parameter identification in the system,which affects the dynamic regulation and stability of the excitation system in power plants.Therefore,this paper combines the generative adversarial network(GAN)and the support vector regression(SVR)model to improve the accuracy of parameter identification in the excitation system of power plants.Firstly,the Wasserstein GAN structure is adopted,and through the adversarial training of the generator and the discriminator,feature data similar to the actual excitation system is generated.Then,the principal component analysis is used to extract and reduce the features of the generated data,mapping the high-dimensional data to a low-dimensional space,providing input for the subsequent SVR model.The SVR model uses the radial basis function kernel function to handle complex nonlinear relationships,thereby accurately predicting the key parameters of the excitation system.Finally,through the dual optimization of combining GAN and SVR,the simulation results show that the mean square error of the GAN+SVR optimization method is generally low,with the highest being 0.000 9 and the lowest being 0.000 1,indicating that the model can effectively capture the dynamic characteristics of the excitation current and maintain high accuracy.关键词
发电厂/励磁系统/生成对抗网络(GAN)/支持向量回归(SVR)/参数辨识Key words
power plant/excitation system/generative adversarial network(GAN)/support vector regression(SVR)/parameter identification引用本文复制引用
覃莹,王保国,项文雅,蒋宗良,黄荣峰..结合生成对抗网络与支持向量回归提升发电厂励磁系统参数辨识的准确性[J].电气技术,2026,27(2):40-45,52,7.
