矿井突水水源识别预测研究——以新庄孜矿为例  

Research on prediction of mine water inrush source identification——Xinzhuangzi coalfield as an example

作　　者：温廷新[1] 张波[1] 邵良杉[1] 

机构地区：[1]辽宁工程技术大学系统工程研究所

出　　处：《中国安全科学学报》2014年第2期100-106,共7页China Safety Science Journal

基　　金：国家自然科学基金资助(70971059);辽宁省教育厅基金资助(LT2010048);山东省自然科学基金资助(ZR2010FL012);校企调研基金资助(SCDY2012018)

摘　　要：矿井突水水源识别是煤矿预防突水事故发生的关键工作,为快速、有效判别矿井突水水源,选取Ca2+,Mg2+,K++Na+,HCO-3,SO2-4,Cl-和总硬度7种主要判别指标。利用Logistic回归分析模型,对7种判别指标的重要程度进行回归分析,提取最主要的判别指标作为水源识别的影响因素,建立基于Logistic分析的矿井突水水源识别的随机森林(RF)模型。将煤矿实测的33组数据作为训练数据,进行预测模型训练,另外选用12组数据作为测试数据,利用该模型进行预测,并与Fisher判别方法和神经网络方法进行对比。结果表明:利用Logistic回归分析法能有效地提取影响矿井突水水源识别的因素,去除冗余影响因素,可有效地预测矿井突水水源类型,使矿井突水水源预测模型错误率降低至1/12。Identification of mine water inrush source is essential to prevent mine water inrush accidents. 7 major indexes, including Ca^2＋ , Mg^2＋ , K ^＋ , Na^ ＋ , HCO3^- , SOl^- , C1^- and total hardness, were selected. Using the Logistic regression analysis model, a regression analysis of importance of 7 indexes was done, and major indexes were chosen as influence factors on source identification. An RF model based on Logis- tic analysis was built for mine water inrush identification. 33 groups of data measured in Xinzhuangzi coalfield were taken mine as training data to a train the model. Additional 12 groups as test data to verify it and compare with the Fisher discriminant method and neural network method. The results show that in- fluence factors of mine water inrush identification were extracted effectively by using Logistic regression analysis, it makes the source of mine water inrush prediction model decrease its error rate to 1/12.

关 键 词：矿井突水 水源识别 预测 Logistic回y-3分析 随机森林(RF) 

分 类 号：X924.4[环境科学与工程—安全科学]