基于迁移学习的泥水平衡盾构隧道开挖面稳定性预测方法研究

Research on Transfer Learning Based Prediction Method for Face Stability of Slurry Pressure Balance Shield Tunnel

摘要: 为解决传统开挖面稳定性评估方法效率低、难以应对突发情况的难题，以排浆密度、进浆密度为依据，提出了一种基于迁移学习的开挖面稳定性预测方法。相比于理论计算与数值模拟，所提出的基于Attn-LSTM的智能预测模型能够从监测数据中学习不同地层条件下掘进参数与泥浆密度间的关系，且在复杂地层条件下也具备较高的预测精度。利用中俄东线天然气管道工程监测数据的验证结果表明，该方法对排浆密度与进浆密度的预测值与真实值相比仅有0.1858%的平均绝对百分比误差。最后，基于迁移学习方法，在沪通铁路吴淞口长江隧道站前IV标工程监测数据上验证了提出方法的有效性。

Abstract: To address the inefficiency of traditional methods for evaluating face stability and their difficulty in handling unexpected situations, a prediction method for face stability based on transfer learning is proposed, utilizing return slurry density and feed slurry density. Compared with theoretical calculations and numerical simulations, the proposed intelligent prediction model based on Attn-LSTM can learn the relationship between tunneling parameters and slurry density under different geological conditions from monitoring data, and exhibits high prediction accuracy even in complex strata. Validation using monitoring data from the China-Russia Eastern Natural Gas Pipeline Project shows that the average absolute percentage error between the predicted and actual values of discharged slurry density and injected slurry density is only 0.1858%. Finally, the effectiveness of the proposed method is verified on monitoring data from Section IV of the Wusongkou Yangtze River Tunnel station of the Shanghai-Nantong Railway Project using a transfer learning approach.