基于EMD和RBF神经网络的大坝形变预测  

Dam deformation prediction based on EMD and RBF neural network

作　　者：刘思敏 徐景田[1] 鞠博晓 LIU Simin;XU Jingtian;JU Boxiao(Faculty of Information Engineering,China University of Geosciences,Wuhan 430074,China;School of Geodesy and Geomatics,Wuhan University,Wuhan 430079,China)

机构地区：[1]中国地质大学(武汉)信息工程学院,湖北武汉430074 [2]武汉大学测绘学院,湖北武汉430079

出　　处：《测绘通报》2019年第8期88-91,95,共5页Bulletin of Surveying and Mapping

基　　金：国家自然科学基金(41874009)

摘　　要：利用长期观测数据结合预测模型对大坝的形变趋势进行估计评价是大坝结构安全监测的必要内容。本文综合利用EMD和RBF神经网络,研究大坝变形时间序列中非线性周期信号变化的内在规律,使用西龙池L022号站4000期数据作为训练样本,对后续80期数据进行预测,并通过对预测结果与实测变形差值的统计分析评价本文方法的预测水平。结果表明,N、E、U 3个方向的RMSE分别为0.878 6、0.360 4和2.235 mm。与BP进行对比,RBF预测效果更好,受数据精度影响较小,MAE、RMSE较BP分别最高可提高63%、57%,且本文方法计算效率高,泛化能力强。The use of long-term observation data combined with prediction models to estimate the deformation trend of dams is an essential part of dam structure safety monitoring. In this paper, the EMD and RBF neural networks are comprehensively used to study the intrinsic law of nonlinear periodic signal changes in the dam deformation time series. The 4000 data of Xilongchi L022 station is used as the training sample, and the subsequent 80 data is predicted and passed. The statistical analysis of the difference between the predicted result and the measured deformation is used to evaluate the prediction power of the method. The results show that the RMSE in the three directions of N, E, and U are 0.878 6, 0.360 4 and 2.235 mm, respectively. Compared with BP, RBF prediction is better, and it is less affected by data accuracy. MAE and RMSE can be increased by 63% and 57%, respectively, compared with BP. The method of this paper has high computational efficiency and strong generalization ability.

关 键 词：GNSS自动化监测系统 经验模态分解(EMD) RBF神经网络 大坝形变预测 

分 类 号：P258[天文地球—测绘科学与技术]