神经网络模糊规则抽取及其应用

从功能的观点出发,提出了一种基于粒子群优化算法的神经网络模糊规则抽取方法。该方法利用所要抽取模糊规则的表达形式,设计了规则的粒子三段表示方式,在粒子群算法优化过程中,采用两种更新方法,即离散化方法和连续化方法。该方法不依赖于具体的网络结构和训练算法,可以方便地应用于各种回归型神经网络。仿真实验表明,该方法可以抽取出保真度较高的符号规则。在实际应用中,采用模糊规则抽取算法,从丙烯腈反应器软测量模型中所得到的规则,提供了一种参数调节的指导性策略。

Fuzzy Rules Extracting and Application of Neural Network

In view of function,an approach based on particle swarm optimization algorithm for fuzzy rule extraction from trained neural network is proposed.A particle rule three parts expressing method is designed to express the fuzzy rule.During the process of particle swarm optimization,two updated methods of discrete and continuous methods are adopted.Without depending on detailed networked structure and training algorithm,the method can be conveniently applied in various diversified neural estimator.Experimental results show that high fidelity symbolic rules are extracted via the proposed method.In practical application,the rules extracted from soft sensing model of acrylonitrile reactor provide a guidance to adjust operational parameters.