基于Transformer的盾构泥水舱液位智能预测与控制  

Intelligent Prediction and Control of Liquid Level in Shield Slurry Chamber Based on Transformer Network

作　　者：卢靖 李刚 胡珉[2,4] 王伊 刘玲玲 LU Jing;LI Gang;HU Min;WANG Yi;LIU Lingling(School of Mechanical and Electrical Engineering and Automation,Shanghai University,Shanghai 200444,China;SHU-SUCG Research Center for Building Industrialization,Shanghai University,Shanghai 200072,China;Shanghai Tunnel Engineering Co.,Ltd.,Shanghai 200032,China;SHU-UTS SILC Business School,Shanghai University,Shanghai 201800,China)

机构地区：[1]上海大学机电工程与自动化学院,上海200444 [2]上海大学-上海城建集团建筑产业化研究中心,上海200072 [3]上海隧道工程有限公司,上海200032 [4]上海大学悉尼工商学院,上海201800

出　　处：《隧道建设（中英文）》2024年第11期2213-2222,共10页Tunnel Construction

摘　　要：泥水液位控制是泥水平衡盾构施工的关键环节,其通过调节泥水循环系统参数进行控制。由于长距离的管道泥水运输,系统存在显著的时滞效应,其控制特性也随地质和管道长度的变化而改变,依靠人工或传统控制方法难以实现准确控制。为解决这一难题并确保盾构推进过程中切口压力的稳定,通过对盾构泥水循环系统控制机制的深入分析,提出一种基于Transformer的盾构泥水舱液位智能预测与控制方法,该方法采用Transformer网络对泥水循环系统的动态特性进行建模,通过迭代多步预测方法对给定控制参数序列下的未来液位情况进行预测。此外,为了优化控制性能并满足系统约束,引入自适应梯度下降方法来解决优化问题及其约束条件,以获得系统的最优控制参数。该方法在苏州河段深层排水隧道的施工数据集上进行仿真验证,试验结果表明:1)在粉砂夹粉质黏土地层的泥水盾构隧道施工中,所提出的控制方法能够有效提高泥水循环系统的控制效果;2)通过与传统控制方法的比较,该智能控制方法显示出更高的控制精度和稳定性,证明其在盾构施工中的应用价值。Slurry level control achieved by adjusting the parameters of the slurry circulation system is crucial for slurry balance shield tunneling.Owing to the long-distance pipeline transportation of slurry,the system exhibits considerable time delays,with the control characteristics changing depending on the geological conditions and pipeline length.Realizing accurate control is challenging with traditional manual control methods.To address this issue and ensure a stable excavation face pressure during shield tunneling,the authors propose the use of an intelligent prediction and control method for estimating the shield slurry chamber level based on a Transformer network.The proposed method leverages the Transformer network′s ability to model the dynamic characteristics of the slurry circulation system using an iterative multistep prediction strategy to forecast future slurry levels for a given sequence of control parameters.Additionally,to optimize control performance and meet system constraints,an adaptive gradient descent method is employed to solve the optimization problem and its constraints,thereby obtaining the optimal control parameters for the system.The proposed method is validated through simulation using a construction dataset of the deep drainage tunnel in the Suzhou river section.Experimental results indicate that the proposed control method substantially improves the control effectiveness of the slurry circulation system during slurry shield tunneling in silty clay layers.Compared to traditional control methods,this intelligent prediction and control method offers higher accuracy and stability,demonstrating its potential for practical applications in shield construction.

关 键 词：泥水平衡盾构 泥水舱液位控制 迭代多步预测 Transformer模型 智能控制 

分 类 号：U45[建筑科学—桥梁与隧道工程]