基于卷积神经网络的煤矿井下定向钻进轨迹参数预测  

Trajectory parameter prediction of directional drilling in underground coal mine based on convolutional neural network

作　　者：魏宏超 WEI Hongchao(China Coal Research Institute,Beijing 100013,China;China Coal Technology and Engineering Group Xi'an Research Institute(Group)Co.,Ltd.,Xi'an 710077,China)

机构地区：[1]煤炭科学研究总院,北京100013 [2]中煤科工西安研究院(集团)有限公司,陕西西安710077

出　　处：《煤矿安全》2023年第5期84-91,共8页Safety in Coal Mines

基　　金：陕西省自然科学基础研究计划资助项目(2023-JC-YB-341);中煤科工西安研究院(集团)有限公司科技创新基金资助项目(2022XAYKF01-01)。

摘　　要：为了得到定向钻进过程中钻头处的实际轨迹参数、提高定向钻孔轨迹控制效果保证实钻轨迹精准度;基于一维卷积神经网络建立了煤矿井下定向钻孔的轨迹参数的预测模型;选取了502组、每组12个实钻轨迹的倾角、方位角、工具面向角等参数作为训练集,以每组最后1次测量所得倾角、方位角为输出值,其他参数作为输入值对模型进行训练;另取12组轨迹参数作为测试集对模型的预测能力进行检测,并将测试结果与24名技术人员根据工作经验得出的预测值进行对比分析。研究表明:该一维卷积神经网络对轨迹参数的预测可行,在均方根误差、可决系数这2项评价指标显出略微优势的同时,倾角、方位角预测的绝对误差分别为0.69°和0.79°,比技术人员预测值平均分别低了22.5%和18.1%。A predictive model is established to determine parameters of the drilling trajectory based on one-dimensional convolutional neural network(1DCNN),which can obtain the actual path parameters at the drill bit during directional drilling to improve the effect of directional drilling borehole tracing control and ensure the accuracy of the actual drilling path.There are 502 groups training data and each contains 12 parameters such as dip angle,azimuth and angle of measurement while drilling system.The inclination angle and azimuth angle obtained in the last measurement of each group were taken as output values,and other parameters were taken as input values to train the model.Other 12 groups drilling path parameters are used to test the model and the results are compared with the predicted results based on experience of the technical staff.This study result shows that:the one-dimensional convolutional neural network is feasible to predict the trajectory parameters,showing a slight advantage in the two evaluation indexes of root mean square error and determination coefficient,and the absolute errors of dip angle and azimuth predicted by one-dimensional convolutional neural network model are 0.69°and 0.70°which are 22.5%and 18.1%lower onA predictive model is established to determine parameters of the drilling trajectory based on one-dimensional convolutional neural network(1DCNN),which can obtain the actual path parameters at the drill bit during directional drilling to improve the effect of directional drilling borehole tracing control and ensure the accuracy of the actual drilling path.There are 502 groups training data and each contains 12 parameters such as dip angle,azimuth and angle of measurement while drilling system.The inclination angle and azimuth angle obtained in the last measurement of each group were taken as output values,and other parameters were taken as input values to train the model.Other 12 groups drilling path parameters are used to test the model and the results are compared with the predicted

关 键 词：煤矿工程地质 定向钻孔 钻孔轨迹参数 一维卷积神经网络 预测模型 

分 类 号：TD163[矿业工程—矿山地质测量]