基于深度学习的地铁盾构姿态失准机理

与智能预测研究

许恒诚

摘 要

伴随我国城市规模的扩张和人口密度的提升，大规模地铁建设对盾构隧道的成型质量提出了更高的要求。盾构姿态失准，即盾构掘进轨迹偏离设计轴线，是影响隧道质量的关键因素之一。盾构掘进姿态控制和管片拼装排布是导致盾构姿态失准的两个主要原因。在盾构掘进时，盾构司机必须依靠自身经验不断调整盾构机的姿态沿隧道设计轴线掘进；而在管片拼装时，则须根据盾构姿态、管片姿态和设计轴线，及时优化管片排布方案实现盾构纠偏。然而，由于多变的地质空间环境和复杂的盾构机操作流程，仅依靠人工经验、事后控制的盾构姿态调整策略，存在控制滞后、准确度低和稳定性不足的问题。本文以盾构失准为研究对象，主要包括以下工作：

首先，本文归纳了盾构姿态的影响因素，建立盾构运动力学模型，据此分析了盾构失准机理和现有盾构姿态控制策略的不足，从盾构掘进姿态预测和管片排布优化两方面着手提出了基于事前控制的盾构姿态调整策略。然后，针对盾构姿态预测，结合深度学习理论，建立了整合小波变化、卷积神经网络和长短时记忆网络的盾构姿态预测混合模型，实现了通过输入盾构机历史运行参数预测其未来姿态变化趋势。其次，根据管片排布原则，设计管片排布优化算法。该算法考虑了盾构机姿态、管片姿态和设计轴线三者间的拟合偏差关系，可自动计算当前环的最优拼装点位。最后，以某地铁隧道为例，收集盾构运行历史数据和管片拼装记录，采用Python语言编程进行验证。

结果表明，本文提出的智能预测模型和管片优化算法可以实现盾构未来姿态的预测和管片拼装点位的推荐，用以辅助盾构司机进行姿态控制，改善盾构姿态失准问题，从而提高隧道衬砌质量。

关键词：盾构机；姿态失准；深度学习；智能预测；混合模型

Abstract

With the expansion of China's Metropolis and the increase of population, large-scale metro construction puts higher requirements on the forming quality of shield tunnels. The shield tunneling misalignment（STM）, which means the movement trajectory of shield deviates from the design axis, is one of the key factors affecting the tunnel quality. The control of attitude and position of shield and segment erection are the two main reasons for the STM. In shield tunneling, the shield driver must constantly adjust the attitude and position of the shield machine along the design axis by relying on his own experience. When the segments are assembled, the posture of ring must be optimized according to the shield attitude and the design axis to correct the shield deviation. However, due to the variable geological space environment and complex shield machine operation, the shield attitude and position adjustment strategy based on manual experience and post-control method has problems of lack of control lag, accuracy and stability. In this thesis, the mechanism of STM is taken as the research object, which mainly includes the following work:

Firstly, this thesis summarizes the influencing factors of shield attitude and establishes the shield motion mechanics model. Based on this model, it analyzes the mechanism of STM and the existing control strategy of attitude and position of shield machine. From the two aspects of shield tunneling attitude prediction and the segment erection optimization, the adjustment strategy of shield attitude and position based on pre-control is proposed. Then, for the prediction of attitude and position of shield machine, a deep learning- based hybrid model integrated wavelet transform, convolutional neural network and long-short-term memory network is established for predicting the attitude and position of shield machine in the future through fedding historical shield operating data. Secondly, according to the principle of segment erection, the optimization algorithm of the segment erection is designed. The algorithm considers the fitting relationship between the the attitude of shield machine, the segment and the design axis, and can automatically calculate the optimal posture of the current ring. Finally, take a metro tunnel as a case study, collect the shield operation data and the segment erction record, and use Python to write the codes for verification.

The results show that the prediction model and the optimization algorithm proposed in this thesis can predict the attitude and position of shield and recommend the posture of the segments erection, assist the driver to control the attitude and position, and improve the quality of the tunnel lining.

Keywords: Shield Machine  Tunneling Misalignment  Deep Learning 

                    Intelligent Prediction Hybrid Model