基于MRA-BP神经网络的桩基托换桥墩沉降预测模型研究  

Research on the Settlement Prediction Model of Pile Foundation Replacement Bridge Piers Based on MRA-BP Neural Network

作　　者：梁贤伟 郭子亮 袁梓文 穆保岗[3] Liang Xianwei;Guo Zilang;Yuan Ziwen;Mu Baogang(China Railway Shanghai Engineering Bureau Group Huahai Engineering Co.,Ltd.,201100,China;POWERCHINA Shanghai Electric Power Engineering Co.,Ltd.,200025,China;School of Civil Engineering,Southeast University,Nanjing 211102,China)

机构地区：[1]中铁上海工程局集团华海工程有限公司,201100 [2]上海电力设计院有限公司,200025 [3]东南大学土木工程学院,南京211189

出　　处：《特种结构》2025年第1期82-86,97,共6页Special Structures

摘　　要：为更加准确有效地进行桥梁桩基托换工程的墩台沉降数值预测,针对传统BP(Back Propagation)神经网络随机赋值、收敛速度慢等问题,提出了多步滚动算法(Multi-step Rolling Algorithm,MRA)优化的BP神经网络预测模型。以南京市某隧道穿越高架桥桩基础工程沉降监测为研究背景,采用MRA-BP神经网络针对桥墩西侧匝道下JC10-13、JC10-14最大沉降监测点进行沉降预测,并与传统BP预测模型对比,对预测结果进行准确度分析。结果表明:MRA-BP神经网络预测模型在JC10-13和JC10-14监测点拟合优度R^(2)的数值均在0.85左右,相比传统BP预测模型提高了0.33,均方误差MSE控制在0.04左右,预测5#桥墩西侧匝道的沉降最终将稳定在25.5mm。MRA-BP神经网络预测模型适用于桩基托换桥墩沉降预测,能够为桥梁墩台施工建设提供更可靠的预测值,为既有建筑和托换结构进行有效变形控制提供依据。To more accurately and effectively predict the settlement of piers in bridge pile foundation replacement projects,a BP neural network prediction model optimized by a multi-step rolling algorithm(MRA)was proposed to address the problems of random assignment and slow convergence of traditional BP(Back Propagation)neural networks.The setlement monitoring of a tunnel crossing viaduct pile foundation project in Nanjing was taken as the research background.Based on the existing monitoring settlement data,the MRA-BP neural network was used to predict the settlement of the maximum settlement monitoring points JC10-13 and JC10-14 under the ramp on the west side of the pier,and the accuracy of the prediction results was analyzed by comparing with the traditional BP prediction model.The results showed that the goodness of fit R^(2)values of the MRA-BP neural network prediction model at the JC10-13 and JC10-14 monitoring points were both around 0.85,which was 0.33 higher than the traditional BP prediction model,and the mean square error MSE was controlled at around 0.04.It was predicted that the settlement of the ramp on the west side of the 5#bridge pier will eventually stabilize at 25.5mm.The MRA-BP neural network prediction model is suitable for the settlement prediction of pile foundation underpinning bridge piers,which can provide more reliable prediction values for the construction of bridge piers and abutments,and can provide basis to effectively control the deformation of existing buildings and underpinning structures.

关 键 词：桥梁工程 沉降预测 BP神经网络 多步滚动法 桩基托换 

分 类 号：TU470[建筑科学—结构工程]