BP和RBF神经网络在气隙击穿电压预测中的应用和对比研究  

Application and contrast analysis of BP and RBF neural network in prediction of breakdown voltage of air gap

作　　者：罗新[1] 牛海清[1] 林浩然[1] 游勇[1] 

机构地区：[1]华南理工大学电力学院,广东广州510641

出　　处：《电工电能新技术》2013年第3期110-115,共6页Advanced Technology of Electrical Engineering and Energy

摘　　要：气隙的击穿电压是决定外绝缘水平的重要因素之一,现有关于击穿电压的理论都是单参数的经验公式,对某一特定大气条件下的击穿电压则很难估计。本文讨论了BP及RBF神经网络在气隙击穿电压预测中的应用,详细说明了在人工气候室中进行击穿试验的过程和BP、RBF神经网络的构建方法。使用人工气候室中获得的样本数据对网络进行训练,用训练好的网络对击穿电压进行预测,结果表明BP及RBF神经网络均能较好地对气隙击穿电压进行预测。并对BP及RBF神经网络进行了比较,RBF神经网络在收敛速度、网络构建、非线性逼近以及泛化能力方面都要优于BP神经网络,更适合于气隙击穿电压的预测。Breakdown voltage of air gap is an important factor to determine the level of external insulation. Almost all the theories about the prediction of breakdown voltage now existing are based on empirical equation of one pa- rameter. This paper discusses the application of BP neural network and RBF neural network in the prediction of breakdown voltage of air gap, illustrates the experiment process of breakdown voltage and describes both BP and RBF network constructions. Firstly both neural networks are trained by the sample data obtained in artificial climate can. Secondly networks which have been trained are used to predict the breakdown voltage. The results indicate that the application of BP neural network and RBF neural network is feasible. Through the comparison between BP and RBF neural networks we conclude that RBF neural network is superior to BP neural network in accuracy, train- ing efficiency and ability of generalization. Thus RBF neural network is more suitable to predict the breakdown volt- age of air gap.

关 键 词：击穿电压 预测 BP神经网络 RBF神经网络 

分 类 号：TM83[电气工程—高电压与绝缘技术] TM85