基于智能优化算法和Bi-LSTM神经网络的盾构隧道刀盘扭矩智能预测方法  

Intelligent prediction method for cutterhead torque of shield tunnel based on intelligent optimization algorithm and Bi-LSTM neural network

作　　者：孔凡超 刘懿晗 刘士吉 路德春[2] 杜修力[2] KONG FanChao;LIU YiHan;LIU ShiJi;LU DeChun;DU XiuLi(School of Water Resources and Hydroelectric Engineering,North China Electric Power University,Beijing 102206,China;Institute of Geotechnical and Underground Engineering,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]华北电力大学水利与水电工程学院,北京102206 [2]北京工业大学岩土与地下工程研究所,北京100124

出　　处：《中国科学:技术科学》2025年第1期171-186,共16页Scientia Sinica(Technologica)

基　　金：国家自然科学基金(批准号:52025084,52408420);国家重点研发计划(编号:2022YFC3800901);北京市自然科学基金(批准号:8244058);中央高校基本科研业务费专项(编号:2024MS066)资助项目。

摘　　要：智能预测合理的刀盘扭矩可有效降低盾构隧道开挖对地层的扰动并减小刀盘磨损,对于盾构安全高效施工具有重要意义.本文基于智能优化算法和深度学习神经网络,提出了盾构隧道刀盘扭矩时序智能预测方法.采用Bi-LSTM神经网络作为深度学习预测模型,分别联合粒子群优化算法(PSO)、遗传优化算法(GA)、饥饿游戏搜索算法(HGS)和麻雀搜索算法(SSA)智能确定Bi-LSTM神经网络的3个超参数,即学习率、神经元数量和丢弃率,进而获得了盾构隧道刀盘扭矩智能预测方法.智能模型预测值与实测值的均方根误差作为4种优化算法的适应度函数,数据预处理采用归一化方法降低不同参数间的量级差异影响.智能预测模型的输入考虑了15个参数对刀盘扭矩的影响,包括1个隧道几何参数、6个地质力学参数和8个盾构掘进参数.结合盾构隧道现场实测数据,从精度、效率和稳定性3个角度评价4种优化算法寻优Bi-LSTM神经网络超参数的性能,结果表明相比于其他3种优化算法,HGS方法具有较好的稳定性、较快的效率和较高的计算精度,是确定Bi-LSTM超参数的适用性方法.Intelligent prediction of reasonable cutterhead torque can effectively reduce the stratum disturbance caused by shield tunneling and reduce cutterhead wear,which is of great significance for the safe and efficient construction of shield tunnels.A time-series intelligent prediction method for shield tunnel cutterhead torque is proposed based on intelligent optimization algorithms and deep-learning neural networks.Bi-LSTM neural network is used as the deep learning prediction model.The particle swarm optimization algorithm(PSO),genetic optimization algorithm(GA),sparrow search algorithm(SSA),and hunger game search algorithm(HGS)are jointly used to intelligently determine the three hyperparameters of the Bi-LSTM neural network,respectively,namely that,learning rate,the number of neurons,and dropout rate.Furthermore,an intelligent prediction method for the cutterhead torque of the shield tunnel is obtained.The root mean square error between the predicted values of intelligent models and measured values is used as fitness functions for four optimization algorithms.Normalization method is used for data preprocessing to reduce the impact of magnitude differences between different parameters.The input of the intelligent prediction model considers the influence of 15 parameters on the cutterhead torque,including 1 tunnel geometric parameter,6 geological mechanical parameters,and 8 shield tunneling parameters.The performances of four optimization algorithms for determining hyperparameters of Bi-LSTM neural network are evaluated from three perspectives:accuracy,efficiency,and stability,based on on-site measured data of shield tunneling.The results indicate that compared with the other three optimization algorithms,the HGS method has better stability,faster efficiency,and higher computational accuracy,which is an appropriate method to determine the hyperparameters of Bi-LSTM.

关 键 词：Bi-LSTM神经网络 超参数 启发式智能优化算法 刀盘扭矩 性能分析 

分 类 号：U455.39[建筑科学—桥梁与隧道工程] TP18[交通运输工程—道路与铁道工程]