针对石油化学工业中的某一典型对象的建模过程,介绍一种以可变长度的自然数编码、以AIC(Akaike’sinformationcriterion)为优化目标的遗传算法(GA)设计径向基函数(RBF)神经网络隐含层结构。文中阐述了该方法的原理,实现步骤及网络泛化性能检验,并与正交最小二乘 (OLS)算法相比较,发现前者设计的网络结构相对简单且网络泛化能力有所改善。

Aimed at the modeling of a typical object in the petrochemical industry, the genetic algorithms,is introduced to design the architecture of RBF neural network, in which the technique of variable length encoding with natural numbers is involved and Akaike's information criterion is chosen as the optimal objective. The fundamentals,concrete procedures of GA and generalization performance tests are presented. The RBF neural network derived by GA has relatively simple configuration and improved generalization performance compared with that derive by the orthogonal least square leaming algorithm.