基于LSTM算法的复杂红层超大直径盾构隧道掘进姿态预测

Attitude Prediction of Super-large Diameter Shield Tunneling in Complex Red Layers Based on LSTM Algorithm

摘要:

为解决超大直径盾构机在掘进过程中由于地质分布不均匀、人工操作误差导致掘进姿态、轨迹与隧道设计轴线发生偏差的问题，提出基于LSTM神经网络建立复杂红层超大直径盾构隧道掘进姿态预测方法，充分考虑了复杂红层地质以及掘进机等12个参数，采用Pearson方法分析地质参数与掘进参数的相关性，有效提高模型预测精准度。依托广州市海珠湾隧道工程，以盾构掘进西线为例，验证了方法的有效性。结果表明：该模型能够较为精准的预测盾构姿态，对盾构竖直姿态预测决定系数（R2）超过0.8，水平和竖直盾构姿态均方根误差均在10以内。算法可以协助盾构机在施工过程中保持稳定，减少姿态偏差，有助于提高隧道的施工质量，确保开挖的隧道符合设计要求。

Abstract:

To solve the problems of deviation among the tunneling attitude, trajectory and tunnel design axis caused by uneven geological distribution and manual operation errors during the tunneling process of super-large diameter shield machines, an attitude prediction method of super-large diameter shield tunneling in mixed red layers based on LSTM neural network is proposed to fully consider 12 parameters of complex red layer geology and tunneling machines. Pearson method is used to analyze the correlation between the geological parameters and the tunneling parameters, which effectively improves the accuracy of model prediction. Relying on the Guangzhou Haizhu Bay Tunnel Project, and taking the shield tunneling on the west line as an example, the effectiveness of the method is verified. The results show that this model can predict the attitude of shield tunneling relatively accurately. The determination coefficient (R2) for the vertical attitude prediction of the shield exceeds 0.8, and the root mean square errors of both the horizontal and vertical shield attitudes are within 10. The algorithms can assist the shield machines in maintaining stability during the tunneling process, and reduce the attitude deviations, which is helpful to improve the construction quality of the tunnels, and ensure that the excavated tunnels meet design requirements.