Deformation Behavior and Prediction Model of Newly Built Stations Closely Underpassing Existing Structures
摘要
为降低密贴下穿工程中既有结构附加变形过大可能导致的结构承载力不足、轨道交通线路不平整等工程风险,特此开展新建车站施工对上部既有结构的扰动变形机理及预测模型研究.首先,以某地铁新建车站零距离下穿既有车站为工程背景,介绍零距离暗挖施工方案;其次,结合现场监测数据分析既有车站结构竖向及水平位移、道床沉降、轨道高差、轨距等扰动变形响应特征;最后,结合长短期记忆(LSTM)神经网络、自适应惯性权重粒子群优化(AWPSO)算法及先验信息约束(PIC),构建结构变形预测模型并进行验证分析.研究结果表明,AWPSO-LSTM-PIC模型能够有效利用先验信息约束和超参数寻优模块,在小样本多任务预测问题中具有较为优异预测性能,平均拟合优度R2达到0.938,具备一定的可解释性,可为密贴工程条件下既有车站变形控制提供实时预警与决策辅助支持.
To mitigate safety risks induced by excessive additional deformation of existing structures in closely aligned underpass projects—such as inadequate structural bearing capacity and uneven rail transit lines—this study investigates the disturbance deformation mechanism and develops a prediction model for the overlying existing structures during new station construction.Using the case of a newly built metro station at zero distance beneath an existing station as the engineering context,the zero-clearance excavation method and an automated monitoring scheme are first introduced.Subsequently,based on field monitoring data,the disturbance-induced deformation response of the existing station structure is analyzed,covering vertical and horizontal displacement,track bed settlement,rail level difference,and track gauge variation.Finally,a structural deformation prediction model integrating a Long Short-Term Memory(LSTM)neural network,an Adaptive Weighted Particle Swarm Optimization(AWPSO)algorithm,and Prior Information Constraints(PIC)is constructed and validated.The results demonstrate that the proposed AWPSO-LSTM-PIC model effectively incorporates prior information constraints and a hyperparameter optimization module,exhibiting outstanding predictive performance in small-sample,multi-task prediction scenarios.The average coefficient of determination(R²)reaches 0.938,reflecting considerable interpretability and generalization capability.