基于AIC方法的切换神经网络模型设计  

Design of switched neural networks model based on AIC method

作　　者：连捷[1] 张凯[1] 

机构地区：[1]大连理工大学电子信息与电气工程学部,辽宁大连116024

出　　处：《大连理工大学学报》2011年第6期890-895,共6页Journal of Dalian University of Technology

基　　金：国家自然科学基金资助项目(61004040)

摘　　要：将切换系统设计中的切换思想与神经网络相结合,构建了切换神经网络模型.根据模糊C均值(FCM)聚类方法将样本数据分为多组训练数据,每组数据对应训练一个单一神经网络模型,再利用赤池信息准则(AIC)制定相应的切换规则.根据输入数据特性,选择单一网络或多网络组合的输出作为模型输出,从而达到函数逼近目的.本模型更好地利用了各个子网络在特定区域具有较高逼近精度的特点.仿真结果表明,切换神经网络模型有较高的逼近精度.The switched neural networks model is constructed by combining neural networks with the idea of the switched system.The sample data are classified into groups of training data in terms of fuzzy C-means（FCM） clustering method,and each of them is used to train a sub-neural network model.The switching rule which determines the form of model′s output is established according to Akaike Information Criterion（AIC）.According to the characteristics of input data,the function approximation is achieved by choosing the output of single neural network or the combined multiple neural networks as the model output.The sub-neural network is of high approximate precision in specified regions,which is used efficiently.Finally,the simulation results demonstrate the high approximate accuracy of the switched neural networks.

关 键 词：切换神经网络 AIC FCM聚类 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]