基于支持向量机的煤层气井排采层位水源判识WATER SOURCE IDENTIFICATION OF THE DRAINED-PRODUCED HORIZONS FOR THE COALBED METHANE WELL BASED ON SUPPORT VECTOR MACHINE
李叶朋,申建,陶俊杰
LI Yepeng,SHEN Jian,TAO Junjie
摘要(Abstract)：
基于煤层气井产出水微量元素的测试结果,探讨支持向量机技术在煤层气排采层位水源判识中的应用,确定最优支持向量机参数获取方法。利用所构建的二叉树结构层次支持向量机模型识别了煤层气井产出水源,并对比验证了不同识别方法。结果表明支持向量机能够有效识别煤层气井产出水源,准确率高于80%。比较分析支持向量机、灰色关联分析和神经网络等预测模型发现,在煤层气井排采水源判识上,对小样本学习而言,支持向量机的预测准确率显著高于神经网络法,与灰色关联分析相当,但灰色关联分析对样本依赖性较强,其泛化能力取决于样本深度。研究证明支持向量机是一种可行、有效的煤层气井排采层位水源判识方法。
According to the tested results of the trace elements in the produced water from the coalbed methane wells,the application of support vector machine technique was discussed for the water source identification in the drained-produced horizons of the coalbed methane,and moreover the obtained methods were confirmed for the optimal support vector machine parameters. With the help of the established binary-tree-structure hierarchical support vector machines,the produced water sources from the coalbed methane wells were identified,and furthermore the different identifying methods were compared and verified. The achievements show that the support vector machine can effectively identify the water sources and the accuracies are more than 80%. The contrasts and analyses among the predicting models such as support vector machine,grey correlation analysis,neural network and so on prove that in the identification of the drained-produced water source in the coalbed methane wells,for the learning of the small sample,the predicted accuracy of the former is more obviously higher than that of the later and correspond with the middle,but the middle has strong dependence on the samples and its generalizing capacity depends on sample depth. The studies show that the support vector machine is one kind of feasible and valid identifying methods for the water source from the drained-produced horizons of the coalbed methane wells.
关键词(KeyWords)：
支持向量机;煤层气井;水源判识;灰色关联分析;神经网络
support vector machine(SVM);coalbed methane(CBM) well;water source identification;grey correlation analysis;neural network
基金项目(Foundation): 国家科技重大专项“中-低阶煤层气成藏高产及主控因素研究”(2016ZX05041001-002);; 国家自然科学基金项目“地层流体系统分异与煤系气开采互动”(41672149)
作者(Author):
李叶朋,申建,陶俊杰
LI Yepeng,SHEN Jian,TAO Junjie
DOI: 10.19597/j.issn.1000-3754.201706051
参考文献(References)：
- [1]秦勇,张政,白建平,等.沁水盆地南部煤层气井产出水源解析及合层排采可行性判识[J].煤炭学报,2014,39(9):1892-1898.
- [2]王善博,唐书恒,万毅,等.山西沁水盆地南部太原组煤储层产出水氢氧同位素特征[J].煤炭学报,2013,38(3):448-454.
- [3]张松航,唐书恒,李忠城,等.煤层气井产出水化学特征及变化规律:以沁水盆地柿庄南区块为例[J].中国矿业大学学报,2015,44(2):292-299.
- [4]闫志刚,杜培军,郭达志.矿井涌水水源分析的支持向量机模型[J].煤炭学报,2007,32(8):842-847.
- [5]陈祖云,张桂珍,邬长福,等.支持向量机在矿井突水水源识别中的应用[J].江西理工大学学报,2009,30(5):10-13.
- [6]冯东梅,吴健伟.矿井突水水源的SVM识别方法[J].辽宁工程技术大学学报(自然科学版),2017,36(1):23-27.
- [7]李艳芳,程建远,王成.基于支持向量机的地震属性优选及煤层气预测[J].煤田地质与勘探,2012,40(6):75-78.
- [8]邵良杉,马寒.煤体瓦斯渗透率的PSO-LSSVM预测模型[J].煤田地质与勘探,2015,43(4):23-26.
- [9]Vapnik N V.统计学习理论[M].许建华,张学工,译.北京:电子工业出版社,2015.
- [10]丁世飞,齐丙娟,谭红艳.支持向量机理论与算法研究综述[J].电子科技大学学报,2011,40(1):2-10.
- [11]Vapnik N V.统计学习理论的本质[M].张学工,译.北京:清华大学出版社,2000.
- [12]张学工.关于统计学习理论与支持向量机[J].自动化学报,2000,26(1):32-42.
- [13]杜树新,吴铁军.模式识别中的支持向量机方法[J].浙江大学学报(工学版),2003,37(5):521-527.