浅埋地铁车站施工期地表变形风险预警

Risk state early warning of the surface deformation of shallow buried subway station

为研究浅埋地铁车站主体结构风险预报预警,建立基于实测数据与贝叶斯优化的随机森林模型(random forest model based on Bayesian optimization, BA-RF)机器学习的趋势分析方法,通过研究主体结构上方地表与建筑物沉降、竖井关键点位移、拱顶沉降等实时数据的趋势,预报未来关键位移短时期达到的最大值,并与控制值进行对比,当前序实测数据的时间δ为5、15、25 d时,平均绝对误差(y_(MAE))0.377、0.303和0.270,均方根误差(y_(RMSE))分别为0.853、0.463和0.509,实现了对结构风险的短期预报预警。研究结果表明:预报方法可以在前序实测数据宽度较小的情况下实现极小的偏移比与绝对偏移比,具备良好的预报效果,综合预报方法在实际监测过程中取得良好工程应用效果。

In order to study the risk prediction and early warning of the main structure of the shallowly buried metro station,a trend analysis method of machine learning of random forest model based on measured data and Bayesian optimization was established.By studying the trend of real-time data such as surface and building setlement above the main structure,shaft key point displacement,vault settlement,the maximum value of future key displacement in a short period was predicted and compared with the control value.When the time width of the current sequence measured data was 5,15 and 25,the mean absolute error(y_(MAE))was 0.377,0.377 and 0.270,and root mean square error(y_(TMSE))was 0.853,0.463 and 0.509.The short-term prediction and early warning of structural risks were realized.The research results showed that the forecasting method could achieve a very small offset ratio and absolute offset ratio with a small width of the measured data in the preceding sequence,which had a good forecasting effect,and the integrated forecasting method achieved good engineering application effect in the actual monitoring process.