基于TCN-SMHA-LSTM-TPA混合神经网络的复杂地层盾构土仓压力预测新方法

Novel Method for Predicting Chamber Pressure in Complex Geological Conditions Based on TCN-SMHA-LSTM-TPA Hybrid Neural Network

摘要: 文章针对传统方法在复杂地层盾构掘进过程中关于土仓压力超前预测精度不高的问题，提出了一种融合时间卷积神经网络(TCN)、多头稀疏自注意力(SMHA)、长短时记忆网络(LSTM)和时间感知注意力(TPA)的混合神经网络模型，用于精准预测复杂地层盾构施工中的土仓压力。TCN模块通过多层卷积捕捉时间序列的局部特征，而多头稀疏自注意力机制进一步增强了模型对长距离依赖关系的建模能力，LSTM层则专注于捕获时间序列的长期依赖信息，TPA模块自适应地调整注意力权重，从而提高预测精度。该模型在盾构施工中的应用表明，其在复杂地质条件下对盾构土仓压力的超前预测具有较高的准确性和适应性，有助于盾构施工的实时监控与决策。

Abstract: To address the issue of low prediction accuracy in traditional methods for forecasting chamber pressure during shield tunneling in complex geological conditions, this article proposes a hybrid neural network model based on Temporal Convolutional Network, Sparse Multi-Head Attention, Long Short-Term Memory, and Temporal Pattern Attention. The model is designed for precise prediction of chamber pressure in shield construction. The TCN module captures local features of time series through multi-layer convolution, while the SMHA mechanism further strengthens the model’s ability to model long-range dependencies. The LSTM layer focuses on extracting long-term dependencies in time series, and the TPA module adaptively adjusts attention weights to improve prediction accuracy. The application of this model in shield tunneling demonstrates its high accuracy and adaptability in forecasting chamber pressure under complex geological conditions, contributing to real-time monitoring and decision-making in tunneling operations.