基于多重深度学习算法的盾构姿态预测模型  

Shield Attitude Prediction Model Based on Multiple Deep Learning Algorithms

作　　者：林金华 陈哲凡 熊昊 陈磊[3,4] 黄明峰 LIN Jinhua;CHEN Zhefan;XIONG Hao;CHEN Lei;HUANG Mingfeng(China First Highway Xiamen Engineering Co.,Ltd.,Xiamen,Fujian 361000,China;School of Software,Shandong University,Jinan,Shandong 250101,China;Key Laboratory of Coastal Urban Resilient Infrastructures,Ministry of Education,Shenzhen,Guangdong 518060,China;College of Civil and Transportation Engineering,Shenzhen University,Shenzhen,Guangdong 518060,China)

机构地区：[1]中交一公局厦门工程有限公司,福建厦门361000 [2]山东大学软件学院,山东济南250101 [3]滨海城市韧性基础设施教育部重点实验室,广东深圳518060 [4]深圳大学土木与交通工程学院,广东深圳518060

出　　处：《施工技术(中英文)》2025年第5期73-79,95,共8页Construction Technology

基　　金：国家自然基金青年科学基金(52208354)。

摘　　要：在广东某盾构隧道中,通过相关数据比较分析RNN,LSTM,GRU,TCN深度学习算法的预测效果,据此提出适用于隧道盾构施工数据的预处理算法框架,以及基于多变量正态分布的异常值检测方法。通过数据清理、盾构状态数据分割、数据归一化处理后,显著增强深度学习算法拟合能力。建立盾构姿态预测模型,并对模型预测表现进行详细对比分析,最终结合GRU,TCN得到更精确的盾构姿态预测模型。通过MAE和RMSE指标综合比较分析各深度学习算法的预测效果,得出不同姿态下的最优预测模型。将该研究成果应用于多重深度学习模型,进行盾构姿态预测和调整,提高了掘进设备的自动化程度和安全性。In a certain shield tunnel in Guangdong,the prediction effects of RNN,LSTM,GRU and TCN deep learning algorithms were compared and analyzed through relevant data,a pre-processing algorithm framework suitable for shield tunnel construction data was proposed,including an outlier detection method based on multivariate normal distribution.Through data cleaning,segmentation of shield running state data and data normalization processing,the fitting ability of deep learning algorithms was significantly enhanced.A shield attitude prediction model was established,and a detailed comparative analysis of the model's prediction performance was conducted.Finally,a more accurate shield attitude prediction model was obtained by combining GRU and TCN.The prediction effects of various deep learning algorithms were comprehensively compared and analyzed through MAE and RMSE indicators,and the optimal prediction models under different attitudes were obtained.This research result was applied to multiple deep learning models for shield attitude prediction and adjustment,improving the automation level and safety of the tunneling equipment.

关 键 词：盾构姿态 深度学习 时间卷积网络 动态预测 

分 类 号：TU94[建筑科学—建筑技术科学]