地震预报专家系统中学习样本的构建  

Constructing the Learning Samples in the Expert System for Earthquake Prediction

作　　者：李远[1] 刘悦[1] 王媛[1] 吴耿锋[1] 

机构地区：[1]上海大学计算机工程与科学学院,上海200072

出　　处：《计算机工程与应用》2005年第4期198-200,共3页Computer Engineering and Applications

基　　金：国家自然科学基金项目(编号:60203011);上海市高等学校科学技术发展基金项目(编号:02AK13)

摘　　要：如何选择和处理学习样本是地震预报专家系统中非常重要的问题。该文在分析以前方法的特点和不足的基础上,提出了异常驱动样本构造法,并用基于RBF神经网络的属性约简方法来处理学习样本。使用异常驱动样本构造法可以方便、科学地根据异常属性出现的频率选择学习样本的属性和根据每条样本的空缺属性率选择学习样本。基于RBF神经网络的属性约简方法利用RBF(RadialBasisFunction)神经网络的特点来量化各维属性对结果的影响程度,从而约简对结果影响程度小的属性。通过实验,表明了用该文方法进行地震预报样本的选择和处理,能明显地提高地震预报的精度。How to select and process learning samples is a very important problem in the expert system for earthquake prediction.This paper presents the method of Exception-Driving to select the samples and uses Attribute Reduction Method Based on RBF Neural Network to process the samples.Using the Exception-Driving method,people can easily and scientifically select the attributes of samples by frequency at which the exception attribute arises,and select the samples by vacant attribute frequency.Attribute Reduction Method Based on RBF Neural Network can evaluate different degrees of influence casted by every attribute on results,and then reduce the attribute whose influence is small.The experiments show that using the methods presented in this paper to select and process samples can enhance the performance of the expert system.

关 键 词：专家系统 RBF神经网络 数据预处理 非线性关系 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]