大粒径卵石层地铁换乘站地表沉降预测研究

地铁换乘站作为地下轨道交通运营线路中的主要枢纽，其开挖和施工造成的失稳变形将会直接影响到车辆运行、人员安全、地下管线设施和既有线路与建筑物。本文以成都地铁17号线换乘站为例，研究了卵石地层的地铁换乘站在深基坑开挖后沉降变形的发展趋势。通过相关性分析，从15项监测数据中分别选择出适合深基坑挖掘阶段和盾构施工阶段的沉降监测相关影响因素。分别借助径向基函数神经网络（RBF）、小波神经网络（WNN）、非线性回归模型（NARX）和极限学习机（ELM）四种智能算法对不同监测参数与监测点位在开挖阶段及盾构施工阶段的地表高程变形进行预测。研究表明：以上算法中在深基坑开挖阶段，WNN模型的预测结果最为精确，预测值和实测值最为接近；当后期预测的参数类别减少时，NARX模型在预测中表现最好，预测值的范围在单个数据点的误差在-0.4~0.3mm内；且监测数据表明在深基坑开挖的第三、四层阶段施工对沉降变形的影响最大，需要着重监测。由此，证实了智能算法在分析和预测卵石地层的地铁换乘站周边地表沉降变形有着较高的可行性，通过对比分析也得到算法模型在相似工程的研究中具有优势。

Study on Prediction of Ground Subsidence of Subway Transfer Station in Large Particle Size Pebble Layer

As the main hub of the underground rail transit operation line, the instability and deformation caused by its excavation and construction will directly affect the vehicle operation, personnel safety, the normal work of underground pipeline facilities and Existing lines or buildings. The development trend of settlement deformation was studied in pebble stratum after deep foundation pit excavation. Chengdu metro line 17 was taking as an example, trained by radial basis function (RBF), wavelet neural network (WNN), Nonlinear Autoregressive models with Exogenous Inputs (NARX), and extreme learning machine (ELM). The surface elevation deformation of different monitoring parameters and monitoring points were predicted during excavation and shield construction. It shows that the prediction results of the WNN model are the most accurate in the deep foundation pit excavation stage. The predicted value is the closest to the actual measured value; the best performs showed of NARX model at the moment forecast parameter items are reduced in the later period, and the range of the predicted value is within -0.4~0.3mm of a single data point; the third and fourth floors was regard as the greatest impact on settlement and deformation, which needs to be monitored emphatically. It is confirmed that machine learning has high feasibility in analyzing and predicting surface settlement and deformation around subway transfer stations in the pebble stratum. Through comparative analysis, great advantages are shown in the algorithm in similar projects.