CCTEG Xi’ an Research Institute Co., Ltd.

基于淮北某矿区的地震属性参数和钻井数据,利用逐步回归分析方法,优选出与煤层厚度有着显著相关性的参数,通过构建主成分分析算法-径向基函数神经网络模型,预测煤层厚度的变化趋势,并将井旁道的煤层厚度预测值保留,作为下一个未知区域预测模型的输入参数,从而获得更准确的煤层厚度预测值。通过不断扩展预测范围,并对其进行持续分析,从而实现整个研究区域煤层厚度的准确预测。分别对比RBF和PCA-RBF2种神经网络模型预测的煤层厚度与真实值的绝对误差、相对误差以及相关系数,其中PCA-RBF神经网络模型的预测值与真实值之间的绝对误差为0~0.08m,相对误差为0%~4%,相关系数为0.9999。结果表明,PCA-RBF神经网络模型预测得到的煤层厚度变化趋势更接近于真实值,预测结果的精度更高,能够为煤矿安全生产、减少成本、提高效益提供强有力的技术支持。

A principal component analysis algorithm ( PCA) - radial basis function ( RBF) neural network model was constructed to predict the change trend of coal seam thickness. Based on the seismic attribute parameters and drilling data of amine area in Huaibei, a PCA-RBF neural network prediction model was constructed using stepwise regression analysis topreferably select parameters with significant correlation with coal seam thickness in order to predict the change trend of coalseam thickness in the mine area, and the predicted value of coal seam thickness in the well bypass was retained as the inputparameter of the next unknown area prediction model, so as to obtain a more accurate coal seam thickness prediction values.By continually extending the prediction range and analyzing, accurate predictions of coal seam thickness are achieved for theentire study area. The absolute error, relative error and correlation coefficient between the predicted coal seam thickness andthe true value of the RBF and PCA-RBF neural network models were compared respectively, where the absolute error between the predicted and true value of the PCA-RBF neural network model was 0 ~ 0. 08 m, the relative error was 0% ~ 4%and the correlation coefficient was 0. 999 9. The predicted coal seam thickness variation trend is closer to the true value andthe accuracy of the prediction results is higher, which further provides strong technical support for safe production, cost reduction and efficiency improvement in coal mines.

中煤科工西安研究院(集团)有限公司科技创新项目(2021XAYJS04)