上海市地铁沿线地面沉降预测研究

Study on land subsidence prediction along the subway lines in Shanghai city

针对地铁沿线地面沉降防治的问题,该文建立了基于月度累计沉降量时间序列的滚动预测模式,提出了“线性回归+BP神经网络”的组合预测模型。同时根据2013—2020年地铁沿线地面沉降的合成孔径雷达干涉测量(InSAR)监测成果,选取3个表征不同地面沉降演变特征的永久散射(PS)点作为研究对象,分析评价了多元线性回归、BP神经网络和“线性回归+BP神经网络”等3种模型的预测效果。结果表明,“线性回归+BP神经网络”的预测精度均最高,体现了组合预测模型的有效性和优越性,对提升地面沉降预测精度具有一定的参考价值。

In view of the problem of prevention and control of land subsidence along subway lines,a rolling prediction method based on the time series formed with monthly cumulative settlement was established in this paper,and a combination prediction method named as"linear regression+BP neural network"was proposed.According to the interferometric synthetic aperture radar(InSAR)monitoring results of land subsidence along subway lines from 2013 to 2020,three persistent scatter(PS)points representing different characteristics of land subsidence evolution were used to evaluate the predictive effects of three models,named respectively as multiple linear regression,BP neural network and"linear regression+BP neural network".The results showed that the prediction accuracy of"linear regression+BP neural network"was the highest in the three cases,which reflected the availability and superiority of the combined prediction method,and had reference value for improving the accuracy of land subsidence prediction.