为保障盾构施工安全并提升掘进效率，提出基于长短时记忆神经网络(LSTM)算法的大直径泥水平衡盾构掘进姿态预测方法。选取泥水平衡盾构掘进过程中的参数，并采用Pearson相关系数对盾构姿态的关联因素进行分析，获取影响盾构姿态的主要因素，以此构建盾构姿态预测数据集；采用长短时记忆神经网络建立盾构姿态预测模型，并利用自适应估计(Adam)算法对其进行优化以获取最优的盾构姿态预测结果。盾构姿态的预测参数主要包括： 盾头水平偏差(HDSH)、盾头垂直偏差(VDSH)、盾尾水平偏差(HDST)、盾尾垂直偏差(VDST)、俯仰角(R)、滚动角(P)。影响盾构姿态预测结果的主要因素为盾构参数和地层参数，其中，盾构分组油缸压力和地层平均抗压/抗剪强度对盾构姿态的影响最大。经过优化的Adam-LSTM神经网络模型对盾构角度的预测效果最优，均方差在0.1以下；对盾构姿态各项参数预测的平均误差小于5%的占比超过80%。
The authors present a
tunneling attitude prediction method, based on the long short-term memory(LSTM) algorithm, to enhance tunneling safety and
efficiency for large-diameter slurry balance shields.
Shield tunneling parameters are analyzed using the Pearson correlation
coefficient to identify the primary factors that affect the tunneling attitude.
These factors are then used to establish a prediction dataset. An LSTM neural
network model is developed for shield attitude prediction, with the Adam
algorithm applied to optimize the LSTM′s performance.
The key prediction parameters for shield attitude include horizontal deviation
of the shield head, vertical deviation of the shield head, horizontal deviation
of the shield tail, vertical deviation of the shield tail, pitch angle, and
roll angle. The identified primary factors that influence the shield attitude
are the shield operational parameters and geological conditions, with hydraulic
cylinder pressure and the average compressive/shear strength of the strata showing
the most significant effect. The optimized Adam-LSTM
neutral network demonstrates a superior shield-attitude
prediction performance, achieving a mean square error of <0.1 and an average
prediction error of <5%, with more than 80% of the output results.
