基于小波包分解的TCN-RBF神经网络模型在桥梁沉降预测中的应用  

Application of TCN-RBF neural network model based on wavelet packet decomposition in bridge settlement prediction

作　　者：吴昌程 WU Changcheng(Jiangxi Provincial Architectural Design and Research Institute Group Company Limited,Nanchang,Jiangxi 330000,China)

机构地区：[1]江西省建筑设计研究总院集团有限公司,江西南昌330000

出　　处：《北京测绘》2025年第1期105-110,共6页Beijing Surveying and Mapping

基　　金：江西省自然资源厅重点科技项目(ZRKJ20232424)。

摘　　要：静荷载与动荷载在沉降监测数据中表现出不同的特性,直接对非线性、非平稳性沉降监测数据进行预测,无法体现沉降监测数据的不同特性,限制了预测精度。因此,本文引入小波包分解方法,对沉降监测数据进行自适应分解与重构。对于低频重构结果,使用趋势性预测能力较强的时域卷积神经网络(TCN)模型进行训练与预测;对于高频重构结果,使用规律性预测能力较强的径向基函数(RBF)神经网络模型进行训练与预测,重构不同频段预测结果得到最终预测结果。使用苏通大桥实测静力水准数据进行实验,结果表明,本文模型较对比模型预测精度更高,验证了本文模型的有效性。Static load and dynamic load exhibit different characteristics in settlement monitoring data.Directly predicting nonlinear and non-stationary settlement monitoring data cannot reflect the different characteristics of settlement monitoring data,which limits prediction accuracy.Therefore,this paper introduced the wavelet packet decomposition method to adaptively decompose and reconstruct settlement monitoring data.For low-frequency reconstruction results,a temporal convolution network(TCN)model with strong trend prediction ability was used for training and prediction.For high-frequency reconstruction results,a radial basis function(RBF)neural network model with strong regularity prediction ability was used for training and prediction.Different frequency band prediction results were reconstructed to obtain the final prediction result.The experiment was conducted by using the measured static leveling data of the Sutong Bridge,and the results show that the prediction accuracy of the proposed model is higher than that of the comparative model,verifying the effectiveness of the proposed model.

关 键 词：小波包分解 径向基函数(RBF)神经网络 时域卷积神经网络(TCN) 桥梁沉降预测 精度验证 

分 类 号：P258[天文地球—测绘科学与技术]