岩溶水系统的径向基神经网络仿真  

RBF network simulation for a karst water system

作　　者：罗定贵[1] 王学军[1] 郭青[2] 

机构地区：[1]北京大学环境学院,北京100871 [2]东华理工学院土木与环境工程系,抚州344000

出　　处：《水文地质工程地质》2004年第2期14-19,共6页Hydrogeology & Engineering Geology

基　　金：国家自然科学基金资助(40242018)

摘　　要：岩溶水系统的复杂性决定了其输入与输出间具有非常复杂的非线性关系,利用人工神经网络方法进行系统的仿真是一种十分有效的手段。本文以MATLAB为平台,介绍了RBF网络的基本原理与训练方法,具有结构自适应确定、输出不依赖初始权值的优良特性。试用该方法建立了济南市岩溶水系统地下水位及其影响因子间的RBF网络模型,讨论了训练样本集与检测样本集的构建、原始数据的预处理方法、神经网络训练误差设置等重要环节,并与同结构的BP网络进行了对比,其结果BP网络效果依赖初始权值,表现出极不稳定性,且训练速度更慢,RBF网络具有更好的应用价值。Owing to the complexity of a karst water system, there is a complex nonlinear relationship between its input and output. Artificial neural net means is a kind of fully valid measure to carry out a systematic imitation. On the basis of MATLAB, this paper descriibes the principles and training methods of RBF network. This network has such advantageous properties as the independence of output on initial weight value and the adaptation for determining the structure. An attempt is made in this article to set up the RBF neural network model of karst groundwater levels and their influential factors near Jinan, to discuss the construction of training samples assemble and checking samples assemble, the pretreatment of original data and fixing training error of the neural network. When contrasting against the BP network, which has the same construction as the RBF net , it is found that the BP network's effect is dependent on initial weight values, the instability appears, and the training velocity is slower. The RBF network is believed to have a valuable application.

关 键 词：岩溶水系统 非线性关系 人工神经网络 RBF网络 

分 类 号：P641.134[天文地球—地质矿产勘探] TP183[天文地球—地质学]