基于RBF神经网络的地下水动态模拟与预测  

Simulation and Prediction of Underground Water Dynamics based on RBF Neural Network

作　　者：罗定贵[1] 郭青[2] 王学军[1] 

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,江西抚州344000

出　　处：《地球学报》2003年第5期475-478,共4页Acta Geoscientica Sinica

基　　金：国家自然科学基金专项基金项目 ( 4 0 2 42 0 18)

摘　　要：RBF网络具有结构自适应确定、输出与初始权值无关的优良特性。以matlab为平台将该网络应用于某地的地下水动态模拟与预测 ,较为系统地研究了训练样本集与检测样本集的构建、原始数据的预处理、神经网络的构建、训练、检测及结果评价整个过程 ,取得了良好效果。同时 ,还与BP网进行了对比 ,认为 。This papae introduces the principles of RBF network and the training methods, points out that: RBF network has advantageous properties such as independence of the output on initial weight value and adaptation for determining the construction. Using the “matlab” as the platform,we apply the network for simulation and prediction of underground water dynamics of one place. And reach a good achievement in studying completly a whole process in the construction of training samples assemble and checking samples assemble,pretreatment of original data, establishment, training, inspection and result-evaluation of the neural network. At the same time, drawbacks on BP net such as artificiality for determining the construction, inferiority to RBF net on accuracy and speed of training and random of initial weight value to the outcome are all manifested after comparing RBF net and BP net. In conclusion, RBF network is a neural network model on simulation and prediction of underground water dynamics which is deserved to be popula rized.

关 键 词：人工神经网络 地下水资源 动态模拟 动态预测 BP网络 RBF网络 

分 类 号：P641.74[天文地球—地质矿产勘探] TP183[天文地球—地质学]