摘要
径向基函数(radial basis function,RBF)神经网络用于建筑物变形监测的性能受模型隐层节点数的影响,并且在低信噪比条件下预测精度不高,针对该问题,提出了一种利用主成分分析(principal component analysis,PCA)优化RBF神经网络的建筑物变形监测方法,利用PCA对变形监测数据进行预处理,在得到大特征值个数的同时通过剔除小特征值对应的特征向量实现噪声抑制,在此基础上以大特征值个数为隐层节点数构建RBF神经网络模型进行变形预测。采用实际算例对所提方法在低信噪比条件下的预测精度进行验证,结果表明相对于传统BP(back propagation)神经网络方法和小波方法,所提方法可以获得更高的预测精度,并且在低信噪比条件下具有更高的鲁棒性。
The performance of radial basis function(RBF)neural network for building deformation monitoring is affected by the number of hidden nodes in the model,and the prediction accuracy is not high under the condition of low signal-tonoise ratio.Aiming at this problem,a method for monitoring the deformation of buildings using principal component analysis(PCA)optimized RBF neural network is proposed.PCA is used to pre-process the deformation monitoring data.While obtaining the number of large eigenvalues,the noise suppression is achieved by eliminating the eigenvectors corresponding to the small eigenvalues.Based on this,the RBF neural network is constructed with the number of large eigenvalues as the number of hidden layer nodes.The results show that the proposed method can obtain higher prediction accuracy compared with the traditional BP neural network method and wavelet method,and it is more robustness under low singleto-noise ratio conditions.
作者
张杰
蔡楠
张哲
ZHANG Jie;CAI Nan;ZHANG Zhe(Qinghai Provincial Geographic Information Center,Xining 810001,China;Qinghai Provincial Forestry Engineering Consulting Center,Xining 810001,China)
出处
《测绘地理信息》
CSCD
2022年第4期46-50,共5页
Journal of Geomatics
基金
青海省科技厅科技攻关项目(2020-NK-128)
关键词
变形监测
RBF神经网络
主成分分析
变形预测
deformation monitoring
RBF neural network
principal component analysis(PCA)
deformation prediction
作者简介
第一作者:张杰,工程师,主要从事地理信息应用、工程测量的理论与方法研究。E-mail:gogo565@163.com
