一种基于进化泛函网络的建模与函数逼近方法

Modeling and Function Approximation Approach Based on Evolutionary Functional Networks

提出了一种进化泛函网络的建模与函数逼近方法,该方法把泛函网络建模过程转变为结构和泛函参数的优化搜索过程,利用遗传规划设计泛函网络神经元函数,对网络结构和参数共存且相互影响的复杂解空间进行全局最优搜索,实现泛函网络结构和参数的共同学习,并用混合基函数实现目标函数的逼近,改变了人们通常用同类型基函数来实现目标函数逼近的方式。数值仿真结果表明,提出的网络建模与逼近方法具有较高的逼近精度。

A new genetic programming designing neuron functions, combining genetic programming and evolutionary algorithm, was proposed for hybrid identification of functional network structure and functional parameters by performing global optimal search in the complex solution space where the structures and parameters coexist and interact. The method using hybrid base functions is different from traditional method approximate object function. The computing results show the high precision by the new method.