土舱压力是保证盾构隧道施工安全和控制施工风险的关键参数之一。为此，提出一种基于残差神经网络的盾构土舱压力预测方法。首先，通过对南京地铁某盾构区间的掘进参数数据进行收集和分析，构建具有多个残差块的残差神经网络模型。然后，利用所建立的残差神经网络模型对盾构土舱压力进行预测，并评估模型对土舱压力的预测性能。最后，对残差神经网络的关键模型参数（包括残差块数目、网络宽度和学习率）进行参数分析，比较参数变化时土舱压力的预测性能，确定最佳的模型结构。并对模型关键参数进行分析。研究结果表明： 1）所提出的残差神经网络模型可以较准确地预测盾构土舱压力，不同位置的土舱压力预测值与实际值接近； 2）1^#、2^#、3^#、4^#、5^#和6#土舱压力的决定系数（R²）分别为0.95、0.96、0.94、0.90、0.91和0.96，均方根误差（E_RMSE）介于0.017~0.023 MPa； 3）相比于人工神经网络（ANN）、支持向量回归（SVR）和随机森林（RF）模型，残差神经网络模型对土舱压力的预测准确性更高。

The chamber pressure is a critical parameter for ensuring construction safety and managing construction risks during shield tunneling. To address this, a method based on a residual neural network is proposed for predicting shield chamber pressure. Initially, tunneling parameter data from a specific section of the Nanjing metro are collected and analyzed to construct a residual neural network model comprising multiple residual blocks. The developed model is then utilized to predict shield chamber pressure and evaluate its performance. Subsequently, key model parameters, including the number of residual blocks, network width, and learning rate, are analyzed. The model′s prediction performance for chamber pressure under different parameter settings is compared to identify the optimal model structure. The results demonstrate that the proposed residual neural network model predicts chamber pressure with high accuracy, yielding predicted values closely aligned with actual measurements across different locations. Specifically, the determination coefficients for chamber pressures at positions #1, #2, #3, #4, #5, and #6 are 0.95, 0.96, 0.94, 0.90, 0.91, and 0.96, respectively, with root mean square errors ranging from 0.017 to 0.023 MPa. These findings indicate that the residual neural network model outperforms artificial neural network, support vector regression, and random forest models in accurately predicting chamber pressure.