基于神经网络的煤矿采动区地面直井产能预测  

Productivity Prediction of Surface Vertical Well in Coal Mine Mining Active Area Based on Neural Network

作　　者：李延河[1,2] 王保玉 吕闰生[3,4] 倪小明[3,4] 张昆[3,4] 郭恒宜 LI Yanhe;WANG Baoyu;LYU Runsheng;NI Xiaoming;ZHANG Kun;GUO Hengyi(Pingdingshan Tian′an Coal Industry Co.,Ltd.,Pingdingshan 467000,China;China Pingmei Shenma Group Co.,Ltd.,Pingdingshan 467000,China;Institute of Efficient Development and Utilization of Coal Gas Resources,Henan Polytechnic University,Jiaozuo 454000,China;College of Resources and Environment,Henan Polytechnic University,Jiaozuo 454000,China)

机构地区：[1]平顶山天安煤业股份有限公司,河南平顶山467000 [2]中国平煤神马集团有限责任公司,河南平顶山467000 [3]河南理工大学煤系气资源高效开发利用研究院,河南焦作454000 [4]河南理工大学资源环境学院,河南焦作454000

出　　处：《煤炭技术》2023年第11期81-86,共6页Coal Technology

摘　　要：基于平顶山首山一矿煤系地质条件、采动区上“三带”发育规律和1口地面采动试验井排采数据,通过理论分析选取埋深、煤厚、渗透率、井口负压、管道压力和瓦斯浓度为地质控制参数,建立了采动直井多参数BP神经网络煤系气产能预测模型。结果表明:随着采煤工作面的推进,日产气量随排采时间呈逐渐递减趋势;不同排采阶段产气量变化较大,整体上产气规律划分为3个阶段分区:Ⅰ产气高值阶段,Ⅱ产气峰值阶段,Ⅲ产气衰减阶段。当BP神经网络隐含层神经元个数为10、30和50时,训练集、验证集和测试集数据拟合度达到90%左右,模型预测结果显示实际日产气量和预测日产气量相对误差较低,分别为3.01%、5.22%和3.35%,实际累计产气量和预测累计产气量数据相关性为0.9991、0.9997和0.9998。Based on the coal measure geological conditions of Shoushan No.1 coal mine in Pingdingshan,the distribution law of"three fracture zones"in the mining active area and the drainage data of a test well,a multi-parameters BP neural network of coalbed methane productivity prediction model and method for goaf wells are established by using the geological control parameters,such as buried depth,coal thickness,permeability,wellhead negative pressure,pipeline pressure and gas concentration.The results show that with the advance of the coal face,the daily gas production decreases gradually with the time of extraction.The gas production varies greatly in different production stages.As a whole,the gas production law can be divided into three stages:Ⅰhigh gas production stage,Ⅱpeak gas production stage,Ⅲgas production decay stage.When the number of neurons in the hidden layer of BP neural network is 10,30 and 50,the data fitting degree of training set,verification set and test set reaches about 90%.The model prediction results show that the relative errors of actual daily gas production and predicted daily gas production are lower,which are 3.01%,5.22%and 3.35%,respectively.The correlations between the actual cumulative gas production data and the predicted cumulative gas production data were 0.9991,0.9997 and 0.9998.

关 键 词：采动区 煤系气 神经网络 产能预测 

分 类 号：TD712[矿业工程—矿井通风与安全]