导水裂缝带高度预测的模糊支持向量机模型  

Height Prediction of Water Fractured Zone Based on Fuzzy SVM

作　　者：王正帅[1,2] 邓喀中[1,2] 谭志祥[1,2] 

机构地区：[1]中国矿业大学江苏省资源环境信息工程重点实验室,江苏徐州221116 [2]中国矿业大学环境与测绘学院,江苏徐州221116

出　　处：《地下空间与工程学报》2011年第4期723-727,共5页Chinese Journal of Underground Space and Engineering

基　　金：国家自然科学基金项目(40772191);"十一五"国家科技支撑计划重点项目(2006BAC09B01)

摘　　要：针对传统支持向量机(SVM)模型在导水裂缝带高度预测中存在着易受奇异值干扰而造成的泛化能力降低问题,提出了基于异常样本探测、剔除的模糊支持向量机模型(FS-VM)。采用模糊聚类分析和加权支持向量机(WSVM)相结合的方法,首先根据模糊ISODATA算法求得导水裂缝带高度及其影响因素的最优分类矩阵,剔除分类结果不一致的观测样本,然后以模糊隶属度为样本权重,按照WSVM建模思想建立了导水裂缝带高度预测的FSVM模型。通过实例将FSVM和WSVM、SVM的预测结果作对比分析。结果表明,FSVM避免了异常样本对预测结果的影响,并顾及了建模样本的不同重要程度,其预测精度比WSVM和SVM有较大的提高。Due to the poor generalization of support vector machines（SVM） induced by the outliers in water fractured zone prediction,a novel model named fuzzy support vector machines（FSVM） was proposed on the basis of detecting and deleting the outliers.Fuzzy cluster analysis,together with weighted support vector machines（WSVM）,was adopted in this paper.Firstly,the fuzzy ISODATA was applied to calculate the optimal class matrixes of water fractured zone and its influential factors;Secondly,the samples belonging to different classes were removed from the dataset;Finally,the model of FSVM was established by using WSVM algorithm and selecting fuzzy membership as samples weightiness.The prediction results of FSVM,WSVM and SVM were compared by a case study.The results show that： in addition to considering weightiness of samples,FSVM avoids the negative influence of outliers on prediction results,so its prediction accuracy is obviously better than WSVM and SVM.

关 键 词：导水裂缝带 支持向量机 模糊聚类分析 采空区 

分 类 号：TD853[矿业工程—金属矿开采]