在高速铁路建设中，矿山法隧道衬砌类型的确定是一个重要环节，其决策显著受制于地理与地质条件，且当前高度依赖设计人员的主观经验。为解决这一难题，提出隧道衬砌类型智能设计整体框架，建立隧道衬砌类型智能设计方法。系统分析隧道洞身衬砌设计影响因素，收集248座隧道共4 100个洞身衬砌节段的设计资料，搭建隧道洞身勘察设计数据库。在此基础上，分别搭建随机森林、支持向量机、梯度提升树3种机器学习模型，采用贝叶斯与交叉验证的手段优化机器学习模型超参数，提出基于CNN、CNN+LSTM、CNN+EPSAnet的3种深度神经网络衬砌类型预测模型。主要结论如下： 1）基于CNN+EPSAnet的隧道衬砌类型智能设计模型预测效果整体最优，预测准确率为86.8%，该模型充分考虑了隧道洞身勘察设计数据特征，同时也可以有效提取多尺度空间信息； 2）研发出隧道洞身智能决策系统，可实现隧道洞身衬砌类型智能决策模型的参数实时调整与可视化训练； 3）通过BIM技术将智能决策得到的衬砌类型设计参数进行三维展示，实现隧道衬砌类型的智能化设计和数字化呈现的深度融合。
Determining the lining type is crucial in the design of high-speed railway tunnels constructed using a mining method. This is strongly influenced by geography and geology as well as depends on the designers subjective experience. In this study, an overall framework of tunnel lining-type intelligent design is proposed, and a tunnel lining-type intelligent design system is established. A systematic analysis of factors influencing tunnel lining-type design is performed, collecting design data from 4 100 lining segments across 248 tunnels, and a tunnel lining survey and design database is developed. Three classical machine learning models random forest, support vector machine, and gradient lifting tree are employed, and model hyperparameters are optimized using Bayes and crossvalidation methods. Additionally, three deep neural network models—convolutional neural network (CNN), CNN+long short-term memory, and CNN+efficient pyramid squeeze attention block (CNN+EPSAnet)—are also employed. Prediction results of the test set indicate that: (1) The prediction effect of the CNN+EPSAnet-based tunnel lining-type intelligent design model outperforms others, with a prediction accuracy of 86.8%. CNN+EPSAnet not only fully considers the characteristics of the tunnel lining survey and design data but also effectively extracts multiscale spatial information. (2) A tunnel lining intelligent decision system is developed, and the real-time adjustment of model parameters and visual training of the tunnel lining-type intelligent decision model are realized. (3) The lining-type design parameters obtained via intelligent decision are displayed in three dimensions through building information modeling technology, and the deep integration of intelligent design and digital presentation of tunnel lining types is realized.
