基于综合算法的补偿模糊神经网络建模方法

针对常规模糊神经网络和补偿模糊神经网络的不足,提出了一种综合聚类算法和梯度下降法的补偿模糊神经网络。该网络的学习分为两步:结构辨识和参数辨识。在结构辨识中,采用关系度聚类方法,自动地划分输入/输出空间,确定模糊规则的数目及每条规则中前提部分和结论部分的初始参数,即构造一个初始模糊模型;在参数辨识中,采用具有五层结构的补偿模糊神经网络,并根据梯度下降法调整所建的初始模糊模型参数,使其具有更高的精度。通过对一非线性系统的建模,仿真结果表明,该网络在建模精度和收敛速度上均优于常规模糊神经网络和补偿模糊神经网络。

A Compensatory Fuzzy Neural Network Modeling Method Based on Hybrid Approaches

In order to overcome the drawbacks of conventional fuzzy neural networks(FNN) and compensatory fuzzy neural networks(CFNN),a compensatory fuzzy neural network based on hybrid approaches is proposed.The identification of the proposed network is composed of two steps: structure identification and parameter identification.In the process of structure identification,the clustering method by relational grades is used to automatically separate the space of input-output data,obtain the numbers of inference rules of fuzzy model and rough estimates of the parameters describing the fuzzy sets in the premise and consequent parts.This is,the initial fuzzy model is constructed.In the process of parameter identification,the gradient descent method is used to tune the parameters of the constructed fuzzy model by a compensatory fuzzy neural network with five layers which can obtain a more precise fuzzy model.Finally,the modeling of a nonlinear system shows that the proposed network is superior to the conventional FNN and CFNN in modeling precise and convergence rate.