随着人工智能技术以及信息化技术的快速发展，各种智能化的方法在工程领域被广泛应用，盾构工程的数字化发展需求也随之提升，将人工智能技术以及智能优化方法运用到盾构工程中，有助于总结盾构掘进的规律，提升盾构掘进的效率并且降低工程中的风险。本研究结合大量工程案例的有关盾构掘进参数和地质条件参数，设计了深度学习智能预测算法和盾构掘进参数优化算法以支持盾构施工智能决策，对盾构刀具磨损、推进速度、贯入度三个目标的深度学习预测算法进行设计，以建立的预测算法作为目标函数进行盾构施工参数优化，从而实现盾构施工的智能管控。
针对盾构工程中的掘进参数，本文通过参数分析的方法，用数据筛选得到决定盾构刀具磨损量、推进速度、贯入度的输入参数集合， 运用贝叶斯优化的方法对模型超参数进行寻优，设计基于贝叶斯优化的深度可分离卷积神经网络预测模型进行盾构刀具磨损、推进速度以及贯入度三个优化目标的预测，引入支持向量机回归、极限梯度提升回归算法回归以及未进行超参数优化的卷积神经网络进行对比分析，分析结果表明贝叶斯优化的深度可分离卷积神经网络模型拟合效果最优，可作为多目标优化算法的目标函数，预测结果能够为盾构机的掘进操作提供指导，实时感知刀具服役状态以及盾构机掘进状态。
针对盾构掘进参数的多目标优化控制研究，本文以提升工程效率、降低工程风险为出发点，将刀具磨损量最小、推进速度最大、贯入度最大作为优化控制的三个目标，通过随机森林参数重要性分析对输入参数进行进一步筛选，将筛选后的参数作为模型输入，多目标灰狼优化算法的目标函数选择为建立的深度可分离卷积神经网络预测算法，通过多目标优化流程在三种不同的地质条件下分别得到帕累托最优解集，在每一组帕累托解集中通过分析每一个解的隶属度确定盾构掘进最优参数组合，优化结果表明建立的结合深度学习的多目标灰狼优化算法模型能够对优化目标实现较好的预测与优化。

关键词：盾构掘进；刀具磨损预测；掘进参数预测；深度学习；多目标优化

With the rapid development of artificial intelligence technology and information technology, various intelligent methods are widely used in the engineering field, the digital development needs of shield engineering have also increased. Applying artificial intelligence technology and intelligent optimization methods to shield tunneling research will help summarize the rules of shield tunneling, improve the efficiency of shield tunneling and reduce risks in the project. This research combines relevant excavation parameters and geological condition parameters of a large number of engineering cases to design intelligent prediction and shield construction parameter optimization algorithms to support intelligent decision-making in shield tunneling. The deep learning algorithm for cutter wear, speed, and penetration is designed, and then the shield construction parameters are optimized and controlled to achieve intelligent control of shield tunneling.
In view of the excavation parameters in shield tunneling, this research uses the method of parameter analysis and data screening to obtain a set of input excavation parameters that determine the cutter wear, speed and penetration. Bayesian optimization function is used to select the hyperparameters of the model, design a Bayesian optimization-based depthwise convolutional neural network model prediction model for three optimization objectives of cutter wear, speed and penetration. Support vector regression, XGBoost regression and unoptimized neural networks are introduced for comparative analysis. The analysis results show that the Bayesian optimization-based depthwise convolutional neural network model has the best prediction results for the fitting effect.,can be used as an objective function for multi-objective optimization algorithms, can provide guidance for the tuneling machine, and sense the service status of the shield machine in real time.
In view of the multi-objective optimal control of shield engineering, this research takes improving engineering efficiency and reducing engineering risks as the starting point, and takes the minimum cutter wear, the maximum speed, and the maximum penetration as the three objectives of optimization control. The input parameters are further screened through random forest parameter importance analysis, the filtered parameters are used as model inputs for the model, and the depthwise convolutional neural network regression is used as the objective function of the multi-objective grey wolf optimization algorithm, the Pareto optimal solution sets were obtained in three different geological scenarios through the multiobjective grey wolf optimization process. In each set of the Pareto optimal solutions, the Degree of Membership Function of each solution was analyzed to determine the optimal parameter combination for shield tunneling. The optimization results show that the established multi-objective grey wolf optimization algorithm model combined with deep learning can achieve better prediction and optimization effects on the optimization goals, and provide guidance for parameter selection of shield tunneling.