基于径向基网络的金属氧化物避雷器阀片伏安特性拟合

为了解决现有金属氧化物避雷器阀片伏安特性拟合中存在的过冲、收敛慢、误差大的问题，用神经网络方法，通过对金属氧化物避雷器（MOA）的原始伏安数据进行分析，确定了神经元个数，并设计了输入、输出录属函数对数据进行预处理。用一个径向基神经网络系统对MOA阀片的伏安特性曲线进行了拟合，拟合结果与分段线性、多指数、线性与非线性拟合法相比，较大幅度地提高了拟合精度和收敛速度，完全适用于MOA阀片的伏安特性拟合。

Approximation to the V-I Characteristic of Resistors of Metal Oxide Arrester Based on Radial Basis Function Neural Network

The problems encountered by conventional techniques in approximation to the V-I characteristic of resistors of metal oxide arrester (MOA) are described. A radial basis function (RBF) neural network is designed to approximate it and the data generation, data preprocessing required to set up the training data for the neutral network are studied in detail. The test results show that, compared with the traditional methods such as piecewise linear, multi exponential, composite exponential, linear and non linear approximations, the RBF network greatly enhances the fitting accuracy and rate of convergence, and the method is suitable for the approximation very well.