基于混合核函数支持向量机的顶板砂岩富水性研究

摘要

In order to find a better method to predict the level of sandstone water enrichment of coal roof,taking Sangshuping Coal Mine as an example,the BP neural network,K-nearest neighbor classification,decision tree and support vector machine algorithm were used to establish the level of sandstone water enrichment of coal roof.By comparison,the accuracy of prediction model based on SVM was 87.5%,the node error rate was the lowest,better than the other three models.In order to further improve the prediction accuracy,a predictive model of mixed kernel function support vector machine was established,and the prediction accuracy was 100% when λ1= 0.05 and λ2= 0.95.The results showed that:the mixed kernel function with conditional attribute as input and decision attribute as output can predict the grade of sandstone water enrichment in coal seam roof,and the effect is good.%为了寻求一种能够较好地预测煤层顶板砂岩富水性等级的方法,以桑树坪煤矿为例,分别采用BP神经网络、K最近邻分类法、决策树和支持向量机算法对其顶板砂岩富水性进行预测.比较发现,基于支持向量机的预测模型准确率最高为87.5%,节点错误率最低,优于其他3种模型.为了进一步提高模型预测准确率,建立了煤层顶板砂岩富水性的混合核函数支持向量机预测模型,当λ1=0.05与λ2=0.95时预测准确率达到100%.研究结果表明,以条件属性作为输入、决策属性作为输出的混合核函数支持向量机预测模型能较好地预测煤层顶板砂岩富水性等级,效果较好.