深埋隧道TBM掘进参数LSTM时序预测模型及应用研究

LSTM time-series prediction model for TBM tunneling parameters of deep-buried tunnels and application research

基于直径为7.8 m的深埋硬岩隧道掘进机(TBM)隧道现场掘进数据与地质数据,提出一种基于长短期记忆(LSTM)神经网络的TBM掘进参数时序预测模型,选择转速、扭矩、推力、净掘进速度、施工速度和开挖比能作为模型的输入变量和输出变量。为了评估地层条件对LSTM模型预测精度的影响,分别建立不同围岩等级下的掘进参数预测模型,对预测结果进行误差分析,并将LSTM模型与传统回归模型的预测结果进行对比。研究结果表明:围岩等级越高模型预测精度越高,施工速度在各级围岩中的预测效果最差,推力和净掘进速度的预测效果最好;LSTM模型的相对误差率、拟合优度、平均绝对百分比误差、均方根误差均明显比传统回归模型的更优。

Based on the field tunneling data and geological data of deep buried hard rock tunnel boring machine(TBM) tunnel with a diameter of 7.8 m, a time-series prediction model of TBM tunneling parameters based on long-short term memory(LSTM) network was proposed. The input and output variables of the model were rotational speed, torque, thrust, penetration rate, advance rate and excavation specific energy. In order to evaluate the influence of formation conditions on the prediction accuracy of LSTM network, the prediction models of tunneling parameters at different surrounding rock grades were established. The error of the prediction results was analyzed and the prediction results of LSTM model were compared with that of the traditional regression model.The results show that the higher the surrounding rock grade, the higher the prediction accuracy of the model. The prediction effect of advance rate in all grades of surrounding rock is the worst, while the prediction effect of thrust and penetration rate is the best. The relative error rate, goodness of fit, mean absolute percentage error and root mean square error of LSTM model are significantly better than those of traditional regression model.