基于GCN-GRU组合模型的基坑周边管线沉降预测  

Prediction of pipeline settlement around foundation pit based on GCN-GRU combination model

作　　者：秦世伟[1] 陆俊宇[1] QIN Shiwei;LU Junyu(School of Mechanism and Engineering Science,Shanghai University,Shanghai 200444,China)

机构地区：[1]上海大学力学与工程科学学院,上海200444

出　　处：《扬州大学学报（自然科学版）》2023年第4期73-78,共6页Journal of Yangzhou University：Natural Science Edition

基　　金：上海市科学技术委员会科技计划资助项目(21DZ1204202).

摘　　要：为提高基坑变形预测结果的准确性,在传统的单点时间序列预测基础上,引入监测数据的空间特征对预测方法进行改进.基于图卷积神经网络(graph convolutional network,GCN)和门控循环单元(gate recurrent unit,GRU),构建一种能捕获数据时空关联性的变形预测模型GCN-GRU,并将其应用于上海某基坑周边管线沉降的变形预测.结果表明,相比于GRU时间序列预测模型,考虑了空间关联性的GCN-GRU模型在单步预测中的均方根误差(root mean square error,RMSE)和平均绝对百分比误差(mean absolute percentage error,MAPE)分别降低了27.3%和25.0%,多步预测中的RMSE和MAPE降低了37.2%和37.3%,预测结果准确性较高.该方法可为同类基坑工程周边管线沉降变形预测提供参考.In order to obtain more accurate and reliable prediction results,on the basis of traditional single-point time series prediction,the spatial characteristics of monitoring data are introduced to further improve the prediction accuracy.Based on graph convolutional network(GCN)and gated recurrent unit(GRU),a deformation prediction model GCN-GRU,which can capture the temporal and spatial correlation of data,is constructed and applied to the deformation prediction of pipeline settlement around a foundation pit in Shanghai.The results show that compared with the GRU time series prediction model,root mean square error(RMSE)and mean absolute percentage error(MAPE)of GCN-GRU model considering spatial correlation in single-step prediction are decreased by 27.3%and 25.0%,respectively and the average RMSE and MAPE in the multi-step prediction are reduced by 37.2%and 37.3%,respectively.The prediction results of GCN-GRU model considering spatial correlation are more accurate,which can provide reference for the settlement deformation prediction of the surrounding pipelines of similar foundation pit engineering.

关 键 词：基坑工程 变形预测 空间特征 图卷积神经网络 门控循环单元 

分 类 号：TU753.1[建筑科学—建筑技术科学]