基于TSNE降维算法的基坑水平位移预测研究  

Prediction of horizontal displacement of foundation pit by TSNE dimensionality reduction algorithm

作　　者：孟凡丽[1] 王小刀 王逸晨 曾东旭 MENG Fanli;WANG Xiaodao;WANG Yichen;ZENG Dongxu(College of Civil Engineering,Zhejiang University of Technology,Hangzhou 310023,China;College ofComputer Science and Technology,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江工业大学土木工程学院,浙江杭州310023 [2]浙江工业大学计算机科学与技术学院,浙江杭州310023

出　　处：《浙江工业大学学报》2023年第4期403-411,共9页Journal of Zhejiang University of Technology

摘　　要：通过引入一种可用于基坑土体参数反演及水平位移预测的TSNE降维算法,结合BP神经网络,研究基坑施工过程中深层土体水平位移的变化,解决了传统BP神经网络处理高维数据时存在的泛化能力低、过拟合和局部极小化等问题。该算法对高维数据进行可视化降维及聚类分析,进而反演土体参数并预测水平位移。研究结果表明:TSNE-BP较BP神经网络反演结果更稳定,取值范围更小,泛化性能更强;加入了各工况变形标准误差的预测,结果差异较小且符合原有规律,可作为选取反演参数的指标;进行TSNE-BP参数反演后的模型位移反演误差减小60%~86%,较BP神经网络反演误差减小18%~50%,预测误差较设计误差减小60%~80%;将TSNE降维算法首次运用于土体参数反演及水平位移预测研究有较重要的工程意义。This paper introduces a TSNE dimensionality reduction algorithmfor the inversion of soil parameters and the prediction of horizontal displacements.Combined with a BP neural network,the change of the horizontal displacement in deep soil during the construction of foundation pits is studied and the low generalization ability,overfitting and minimization in the existing BP neural networks are improved.The algorithm is used for the visual dimension reduction and cluster analysis of high dimensional data,and the soil parameters are back-calculated and the horizontal displacement is predicted.The results show that the TSNE-BP is more stable with a smaller value range and stronger generalization performance.With the TSNE-BP parameters back-calculated,the relative errors of model displacements are reduced by 60%~86%,which is 18%~50%lower than those by the BP neural network.The prediction error is reduced by 60%~80%compared with the design error.The TSNE dimensionality reduction algorithm is applied to the inversion of soil parameters and prediction of horizontal displacements for the first time,which is of great engineering significance.

关 键 词：TSNE降维算法 土体参数 水平位移 BP神经网络 可视化 

分 类 号：TU443[建筑科学—岩土工程] TP183[建筑科学—土工工程]