基于思维进化算法的人工神经网络结构优化  

Optimization of the Artificial Neural Network Architecture Based on Mind Evolutionary Computation Algorithm

作　　者：何小娟[1] 曾建潮[1] 徐玉斌[1] 

机构地区：[1]太原重型机械学院系统仿真与计算机应用研究所,太原030024

出　　处：《太原重型机械学院学报》2004年第2期153-157,共5页Journal of Taiyuan Heavy Machinery Institute

基　　金：国家自然科学基金项目(资助号:60174002);山西省青年科学基金项目(资助号:20031031)。

摘　　要：针对神经网络结构难以优化的问题,本文采用思维进化计算(MEC)算法和BP算法相结合的方法来动态优化神经网络结构。随机产生网络结构,对每一结构,利用BP算法评价神经网络结构优劣,找到局部最优结构,再通过MEC算法中的趋同、异化操作,找出全局最优结构。仿真结果说明了算法的有效性。It is us ually difficult to optimize the artificial neural network architecture. A method that combines MEC( Mind Evolutionary Computation) and BP algorithm is presented in this paper to dynamically optimize the neural networks. A network architecture population is generated stochastically at first. For each architecture, the BP algorithm is used to evaluate its performance , by which a local optimum is found. Then, through the similartaxis and dissimilation operators of the Mind Evolution Computation method, the global optimum is acquired. The results of experiments show that this method is efficient.

关 键 词：人工神经网络 结构优化 思维进化算法 BP算法 趋同 异化 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]