基于人工神经网络的大直径盾构隧道施工地层变形预测分析

为了预测盾构施工引起的地表沉降规律,以京张高铁清华园大直径泥水盾构隧道工程为背景,结合盾构试验段隧道掘进过程中地层变形的监测数据,建立基于时间序列的NARNN(不含外部输入)和NARXNN(含外部输入)非线性自回归神经网络预测模型,对重要监测断面测点的隧道掘进过程中地表沉降发展趋势进行预测分析,并与传统时间序列ARMA模型预测结果进行对比,发现NARNN模型、NARXN模型、NARMA模型的预测结果与现场监测数据都比较吻合,而NARNN和NARXN非线性自回归神经网络预测模型精度明显高于传统时间序列ARMA模型,而考虑外部输入的NARXNN模型又比不考虑外部输入的NARNN精度高。因此,在考虑施工方法、地质条件和空间效应(埋深)等外部因素条件下建立的NARXNN模型具有良好的预测效果,能够较好地模拟盾构施工引起的地表沉降规律。

Prediction and Analysis of Ground Deformation in Large Diameter Shield Tunnel Construction Based on Artificial Neural Network

In order to predict the law of surface subsidence caused by shield construction, time-series based NARNN(without external input) and NARXNN(including external input) models of non-linear input autoregressive neural network prediction are established based on the large-diameter mud-water shield tunnel project of Qinghua tunnel project of the newly built Beijing-Zhangjiakou High-speed Railway. Based on the monitoring data of the stratum deformation during tunneling of shield tunnel test section, the predictive analysis of surface subsidence development trend in tunneling process of important monitoring section measurement points and the comparison with prediction results of traditional time series ARMA model, it is found that the prediction results of NARNN model, NARXN model and NARMA model are consistent with the on-site monitoring data, and the accuracy of NARNN and NARXN nonlinear autoregressive neural network prediction models is significantly higher than that of the traditional time series ARMA model with external input. While the accuracy of NARXNN model with external input is more accurate than that of NARNN without external inputs. Therefore, the NARXNN model established under the consideration of external factors such as construction method, geological conditions and spatial effects(buried depth) has good prediction result, which can better simulate surface subsidence law caused by shield construction.