基于正则化IRM-NN的棒–板长间隙正极性操作冲击放电电压智能预测  

Intelligent Prediction of Positive Switching Impulse Discharge Voltage of Rod-plane Long Gap Based on Regularized IRM-NN

作　　者：杨冰雪 卢占会[1] 姚修远 时卫东 丁玉剑 YANG Bingxue;LU Zhanhui;YAO Xiuyuan;SHI Weidong;DING Yujian(North China Electric Power University,Changping District,Beijing 102206,China 2.China Electric Power Research Institute,Haidian District,Beijing 100192,China)

机构地区：[1]华北电力大学,北京市昌平区102206 [2]中国电力科学研究院有限公司,北京市海淀区100192

出　　处：《中国电机工程学报》2023年第14期5683-5692,共10页Proceedings of the CSEE

基　　金：国家自然科学基金项目(51861135310,51977198)。

摘　　要：长空气间隙的放电特性是特高压输变电工程外绝缘设计的重要依据,而典型棒–板空气间隙的放电特性一直是研究者们关注的基础问题。目前,棒–板间隙放电电压的计算方法有很多,但大多数方法都无法很好地适应大范围的温湿度变化。为实现极端温湿度条件下放电电压的精确计算,该文提出一种基于正则化不变风险最小化神经网络(invariant risk minimization-neural network,IRM-NN)的棒–板长间隙放电电压预测方法。系统分析了棒–板间隙放电的影响因素,提取关键特征量作为输入训练模型。模型在测试集上的平均绝对百分比误差仅为1.6%,验证了该模型可以有效外推至试验条件外的应用场景。然后,对比该模型与3种常用机器学习模型的预测效果。结果表明,该模型在训练样本试验范围之外的样本上的计算精度明显高于其他模型。所提棒–板间隙50%放电电压计算方法可适应大范围温湿度及一定电极尺寸变化,可为长空气间隙放电特性研究提供参考。The discharge characteristic of long air gap is an important basis for the insulation design of UHV power transmission and transformation projects.The discharge characteristics of typical rod-plane air gap have always been a basic problem concerned by researchers.At present,there are many methods to calculate the discharge voltage of rod-plane gap,but most of them can not well adapt to the large range of temperature and humidity changes.A method for predicting discharge voltage of rod-plane long gap based on regularized invariant risk minimization-neural network(IRM-NN)is proposed for the accurate calculation under extreme temperature and humidity conditions.In this paper,the influencing factors of rod-plane gap discharge are systematically analyzed,and the key characteristic parameters are regarded as input training models.The mean absolute percentage error of the model on the test set is only 1.6%,which verifies that the model can be effectively extrapolated to application scenarios beyond test conditions.Then the prediction effect of the proposed model is compared with that of three commonly used machine learning models.The results show that the computational accuracy of proposed model is significantly higher than that of other models on samples outside the scope of training sample test.The calculation method of 50%discharge voltage of rod-plane gap proposed in this paper can adapt to a wide range of temperature and humidity and a certain electrode size change,which can provide a reference for the study of discharge characteristics of long air gap.

关 键 词：长空气间隙 放电电压 温湿度变化 不变风险最小化 神经网络 

分 类 号：TM855[电气工程—高电压与绝缘技术]