基于网格搜索-支持向量机的采场顶板稳定性预测  

Stope roof stability prediction based on both SVM and grid-search method

作　　者：郭超[1] 宋卫华[1] 魏威[2] 

机构地区：[1]辽宁工程技术大学矿业学院,辽宁阜新123000 [2]中南大学资源与安全工程学院,湖南长沙410083

出　　处：《中国安全科学学报》2014年第8期31-36,共6页China Safety Science Journal

基　　金：国家自然科学基金资助(51304110);辽宁省高等学校优秀人才支持计划资助(LJQ2013039)

摘　　要：为准确、快速地预测采场顶板稳定性(SRS),建立基于支持向量机(SVM)理论的SRS评价法。考虑煤-岩石介质与环境条件和工程因素,研究岩石单轴抗压强度、岩石质量指标(RQD)、煤体抗压强度、顶板水文状况和工作面月推进速度对SRS的影响。建立SRS识别的SVM模型。为提高预测模型的泛化能力和预测精度,利用网格搜索法(GSM)及10折交叉确认寻优方法对SVM模型的参数进行优化。用该模型对5组待判工程实例进行判别。研究结果表明,模型训练样本10折交叉确认准确率达91.3%,对测试样本识别正确率为80%,识别结果与实际较吻合。In order to predict SRS accurately and rapidly, a new method based on SVM was worked out for evaluating the SRS. Coal-rock medium properties, environmental and engineering conditions were considered. Effects of five factors on SRS were studied. The factors are the uniaxial compressive strength of rock, rock-quality designation （RQD）, compressive strength of coal mass, hydrological condition of roof and advancing speed of working face. A SVM analysis model was built to predict the dynamic engineering classification of SRS. To enhance the generalization performance and prediction accuracy of the model, GSM and 10-fold cross validation optimization method were applied to optimize the parameters of it. The model was applied to 5 groups of engineering example. The results show that 10-fold cross-validation can achieve accuracy of as high as 91.3% for training samples, and 80% for testing samples, and that SVM classification model prediction of SFI.q ,-,~,,~,, ;~l, ___v~_

关 键 词：采场顶板稳定性(SRS) 支持向量机(SVM) 网格搜索法(GSM) 10折交叉确认 预测 

分 类 号：X936[环境科学与工程—安全科学]