基于BP神经网络的地铁盾构掘进参数预测研究  

Prediction of study on subway shield tunneling parameters based on BP neural network

作　　者：刘桂东 LIU Guidong(China Railway Fourth Bureau Group Third Construction Co.,Ltd.,Tianjin 300011,China)

机构地区：[1]中铁四局集团第三建设有限公司,天津300011

出　　处：《地基处理》2025年第2期201-208,212,共9页Journal of Ground Improvement

摘　　要：盾构法施工技术已广泛应用于城市地铁建设,建立盾构掘进参数预测模型在减小施工不确定性、保证盾构机安全高效掘进方面具有重要意义。本文基于杭州某在建隧道工程,以地层实际参数与盾构机预设指标等7项参数为网络输入特征,以盾构机掘进参数等4项参数为输出特征,针对该工程构建了一系列不同超参数组合架构的BP神经网络。通过对比模型在验证集上的表现,选取最适用于该工程的盾构掘进参数预测模型,并在测试集上验证模型的泛化能力。随后,针对5类模型超参数开展了敏感性分析,探究超参数取值对BP神经网络泛化能力的影响。结果表明:该模型对于盾构掘进时的注浆量、出土量、总推力及注浆压力的预测效果均较好;训练批次越大或训练轮次和神经元数量越小,模型总体泛化能力越强,而学习率和隐藏层数量越大时,模型总体泛化能力分别呈现两端大、中间小的U型分布和两端小、中间大的钟型分布。The shield tunneling method has been widely used in urban metro construction.Establishing a prediction model for shield tunneling parameters plays an important role in reducing construction uncertainty and ensuring safe and efficient tunneling of the shield machine.Based on an ongoing tunnel project in Hangzhou,a series of BP neural network models with different hyperparameter combination were constructed for this project,using seven input features(including real parameters of the stratum and preset shield machine indicators)and four output features(shield tunneling parameters).By comparing the performance of these models on the validation set,the most suitable prediction model for shield tunneling parameters was selected,and its generalization ability was further verified.Subsequently,a sensitivity analysis was conducted on five types of model hyperparameters to investigate their impact on the generalization ability of the BP neural network.The results show that the model has a good prediction effect on grouting volume,soil discharge volume,total thrust and grouting pressure during shield tunneling.The larger the training batch or the smaller the number of training rounds and neurons,the stronger the overall generalization ability of the model.The overall generalization ability of the model will show a U-shaped distribution and a bell-shaped distribution with larger learning rate and larger number of hidden layers.

关 键 词：BP神经网络 盾构隧道掘进参数 预测模型 泛化能力 

分 类 号：U45[建筑科学—桥梁与隧道工程]