论文部分内容阅读
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。
In order to reduce the generalization ability caused by traditional singular value (SVM) model in predicting the height of water-conducting fractured zone, the traditional support vector machine (SVM) model is proposed. The fuzzy support vector machine model based on abnormal sample detection and elimination (FS- VM). By using fuzzy clustering analysis and weighted support vector machine (WSVM), the optimal classification matrix for the height of water-carrying fractures and its influencing factors is obtained based on the fuzzy ISODATA algorithm, and the observed samples with inconsistent classification results are removed. Then, The fuzzy membership is the sample weight. According to the WSVM modeling idea, the FSVM model of water-bearing fractured zone height prediction is established. An example is used to compare FSVM, WSVM and SVM forecast results. The results show that FSVM avoids the impact of abnormal samples on the prediction results, and takes into account the different importance of the model samples, and its prediction accuracy is greatly improved than WSVM and SVM.