基于GA 的神经网络设计及其应用

Neural Network Designing Based on Genetic Algorithm and Its Application

刘克文

Liu Kewen

中原石油勘探局勘探开发科学研究院

Research Institute of Pet roleum Exploration and Development , Zhongyuan Petroleum Exploration Bureau ,Henan 457001 ,P. R. China

传统的神经网络,如BP 网络设计,不仅工作效率降低,网络性能低下,而且会因非线性多极值目标函数而陷于局部最优解。本文采用全局寻优的遗传算法( GA) 来辅助网络设计,实现网络结构、连接权及学习规则的自适应演化。通过利用测井资料与孔隙度参数的学习建模,表明该方法可以克服传统方法的不足,具有一定的推广应用价值。

The conventional neural network design method , such as BP algorithm , it s topological const ruction and parameters is determined by designer’s experience and repeated test . This not only lead to low work efficiency and poor network performance , but also usually lost in local optimal solution because of nonlinear multi-extreme object function. In this paper , we designed the network construction using genetic algorithm as an aid method , and determined the topology , linking weight and learn factor with adaptive evolution. It s application to porosity learning with log data show that this method can improve the network’s performance and is a valuable method.