基于掘进物理规律与LSTM的双护盾TBM掘进姿态预测方法  

Prediction Method of Double Shield TBM Tunnelling Attitude Based on Tunnelling Physical Law and LSTM

作　　者：梁芝军 LIANG Zhijun(China Railway 1lth Bureau Group City Rail Engineering Co.Ltd.,Wuhan Hubei 430074,China)

机构地区：[1]中铁十一局集团城市轨道工程有限公司,湖北武汉430074

出　　处：《铁道建筑技术》2024年第12期5-9,共5页Railway Construction Technology

基　　金：中铁十一局集团城市轨道工程有限公司科技研究开发计划项目(2022-09)。

摘　　要：TBM掘进姿态控制是保证隧道建设质量的关键。建立隧道掘进姿态与控制参数之间的关系,并据此预测TBM掘进姿态是该领域亟需解决的关键难题之一。提出一种TBM掘进姿态预测方法,以长短期记忆神经网络(LSTM)为桥梁,将初始隧道掘进姿态及每环掘进参数作为输入,逐一预测各环水平、垂向偏移角与偏移距。为克服LSTM过拟合和误差积累的固有缺陷,基于TBM机械运动原理建立TBM掘进姿态物理规律,并将其作为约束引入常规LSTM算法。依托青岛地铁6号线项目,采集共计140组数据,建立基于改进LSTM方法的掘进姿态预测模型,以验证该方法的预测精度与泛化性。The controlling of the TBM tunnelling attitude is the key to ensure the quality of tunnel construction.Establishing the relationship between tunnelling attitude and control parameters,and predicting the TBM excavation posture based on this is one of the key challenges that urgently needs to be addressed in this field.This paper introduces a TBM tunneling attitude prediction method.Using Long-Short Term Memory(LSTM)neural network as a bridge,the initial tunnel tunneling attitude and control parameters of each ring are used as inputs to predict the horizontal and vertical bias angles and distances of each ring.To overcome the inherent shortcomings of overfitting and error accumulation in LSTM,a physical law of TBM tunneling attitude was established based on the TBM movement principle,and it was introduced as a constraint into the conventional LSTM algorithm.Based on the Qingdao Metro Line 6 project,a total of 140 sets of data were collected to establish a mining attitude prediction model based on an improved LSTM method,in order to verify the prediction accuracy and generalization of the method.

关 键 词：TBM 掘进姿态 长短期记忆神经网络 时序预测 

分 类 号：U455.31[建筑科学—桥梁与隧道工程]