三种RBF网络函数逼近性能对比及应用研究

摘要

非线性函数逼近问题是神经网络数据处理的具体应用之一,在相同误差指标和目标参数的情况下,以具体的非线性函数为例,仿真对比了径向基神经网络(Radical Basis Function,RBF)、模糊RBF和基于遗传算法的模糊RBF网络的逼近性能.结果表明,3种RBF网络结构都能够较好的逼近目标函数,但模糊RBF与GA-RBF网络结构较基本RBF网络结构而言能够更早达到较小的逼近误差范围.在此基础上,仿真验证了模糊GA-RBF网络应用于间接型自校正控制的有效性.%Nonlinear function approximation is one of the specific applications of neural networks to process data.In the case of the same error index and the target parameter,taking the concrete nonlinear function as an example,simulating and comparing Radical Basis Function neural network,fuzzy RBF neural network and fuzzy RBF neural network are based on genetic algorithm.The results show that the three RBF network structures can better approximate the objective function,but the fuzzy RBF and GA-RBF network structure can achieve a smaller approximation error range than the basic RBF network structure.On this basis,simulation results verify the effectiveness of the fuzzy RBF neural network for indirect self-tuning control.