基于IPSO-BP混合模型的TBM掘进参数预测  

Prediction of TBM tunnelling parameters based on IPSO-BP hybrid model

作　　者：侯少康 刘耀儒[1] 张凯 HOU Shaokang;LIU Yaoru;ZHANG Kai(State Key Laboratory of Hydroscience and Hydraulic Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]清华大学水沙科学与水利水电工程国家重点实验室,北京100084

出　　处：《岩石力学与工程学报》2020年第8期1648-1657,共10页Chinese Journal of Rock Mechanics and Engineering

基　　金：国家重点基础研究发展计划(973)项目(2015CB057904);清华大学水沙科学与水利水电工程国家重点实验室科研项目(2019–KY–03)。

摘　　要：通过TBM上升段数据预测稳定段的掘进参数,可以在每个掘进循环的起始阶段预测出各掘进参数的建议值,辅助进行TBM掘进参数的设置和优化调整。提出一种基于改进粒子群算法优化BP神经网络(Improvedparticle swarm optimization-back propagation,IPSO-BP)的TBM掘进参数预测模型,采用自适应惯性权重对标准PSO算法进行改进,并基于改进PSO算法对BP网络的连接权值和偏置进行优化。基于吉林引松工程TBM3标段802 d的TBM运行数据对训练集和测试集进行划分。选取TBM掘进上升段前30 s的刀盘扭矩、贯入度、刀盘功率、推进速度、总推进力5个掘进参数变化特征(均值和线性拟合斜率),以及岩性、围岩分级和地下水活动情况3个地质参数作为模型的输入,并通过试验法确定模型的3个关键超参数(隐含层节点数、学习率和粒子群种群规模),预测稳定掘进时的推进速度v、总推进力F和刀盘扭矩T。结果表明,所提出的模型对TBM稳定掘进段参数的预测拟合优度均达0.85以上,平均绝对百分误差均小于12.68%,相比于BP模型和PSO-BP模型具有更高的预测精度。It is of great significance to predict the TBM tunnelling parameters in stable phase based on the data of the rising phase,which can predict the recommended values of the tunnelling parameters at the early phase of each tunnelling cycle and assist to set and optimize the TBM tunnelling parameters. A TBM tunnelling parameter prediction model based on improved particle swarm integrated back propagation(IPSO-BP) is proposed,in which the standard PSO algorithm is improved by using adaptive inertia weight and the connection weight and bias of BP network are optimized based on improved PSO algorithm. Based on the 802-day TBM operation data of Songhua River water conveyance project,the training and test sets are divided. The variation characteristics(mean value and linear fitting slope) of cutterhead torque,penetration,cutterhead power,advance rate and total thrust in the first 30 s of TBM rising phase,as well as three geological parameters(i.e.,lithology, surrounding rock level and groundwater level) are selected as the inputs of IPSO-BP model. Three key hyper-parameters including number of the hidden layer nodes,learning rate and population size are determined by experimental method,and the advance rate v,total thrust F and cutterhead torque T in stable phase are predicted. The results show that the R2 of the proposed model is over 0.85 and the mean absolute percentage error is less than 12.68%. Compared with BP and PSO-BP models,the proposed model has higher prediction accuracy.

关 键 词：隧道工程 TBM 改进粒子群算法(IPSO) BP神经网络 掘进参数预测 

分 类 号：U45[建筑科学—桥梁与隧道工程]