基于粗糙集理论BP网络的地铁深基坑监控预测分析与优化

Analysis and Optimization of the Application of RS-BP Neural Network in Prediction of Deep Subway Foundation Excavation Monitoring

以25个深基坑工程地表沉降实测资料为训练样本,综合考虑多个主要影响因素,应用粗糙集对次要影响因素进行约简,然后建立地表沉降的7-15-1粗糙集BP(RS-BP)神经网络预测模型对5个检验样本进行预测及预测精度分析,并将该模型与传统BP神经网络预测模型进行对比。结果表明:传统BP神经网络预测其平均相对误差达到15.04%;而RS-BP神经网络预测平均相对误差较小,为5.55%,满足精度要求。因此,基于粗糙集BP神经网络预测模型在预测精度上优于传统BP神经网络预测模型。

Based on twenty five samples of actual measurements of ground surface settlement around deep foundation excavation,these factors of surface settlement are comprehensively considered.Firstly,the rough set theory is adopted to reduce secondary attributes for the secondary factors and then obtain the optimal attribute set.Secondly,this paper established the 7-15-1 RS-BP neural network prediction models for surface settlement to predict five validating sample and analyze the precision.Finally,the prediction values of RS-BP model are compared with traditional BP model.The results show that the average relative error of surface settlement prediction with traditional BP model is 15.04%,while RS-BP model is 5.55% and meets the precision requirement.Therefore,using RS-BP model to predict surface settlement is superior to traditional BP model in prediction accuracy.