2016年1月上施工技术 第45卷第1期CONSTRUCTIONTECHNOLOGY73 DOI:10.7672/sgjs2016010073 基于神经网络的卡尔曼滤波在变形监测中的应用*
韩亚坤',文鸿雁²,陈冠宇”,郭雷,王青涛 3.浙江省测绘大队,浙江杭州310030;4.桂林理工大学广西矿冶与环境科学实验中心,广西桂林541004) [摘要]卡尔曼滤波是基于最小均方误差下递推式的最优估计方法,现已成为最常用的滤波算法之一。
一般来说, 标准的卡尔曼滤波要求动态噪声和观测噪声具有先验已知性,因此这也限制了它在实际生产中的应用,而神经网 络对实际系统辨识具有很好的非线性映射能力,利用神经网络对卡尔曼滤波后的估计值进行补偿,可以在很大程 度上改进卡尔曼滤波的效果,将此方法应用于变形监测中证明了该方法的可行性及有效性。
[关键词]卡尔曼滤波;神经网络;变形;监测;补偿 [中图分类号】TU753[文献标识码】A[文章编号】1002-8498(2016)01-0073-04 ApplicationofKalmanFilterBasedonNeuralNetwork forBuildingDeformationMonitoring HanYakun',Wen Hongyan²,Chen Guanyu',Guo Lei,Wang Qingtao (1.GuangxiKeyLaboraoryfSpatialInformationandGeomatics,Guilin,Guangxi541004,China;2.Colleef Geomatics and Geoinformation,Guilin University ofTechnology,Guilin,Guangxi 541004,China; 3.Zhejiang Brigade of Surveying and Mapping,Hangzhou,Zhejiang 310030,China;4.Guangxi Scientific ExperimentCenterofMining,MetallurgyandEvironmen,GuilinUniersityofTechnology,Guilin,Guangxi541004,China) Abstract;Kalman filteris an optimization estimationmethod ofrecurrence formula based on least mean square error,it has been one of the most frequently-used filter algorithms. Generally, the standard Kalman filter demands dynamic noise and observation noise with the known prior information,because of these reasons,which restricts its application in reality.However,the reality system identification by use of nonlinear mapping ability of neural network to pensate estimate error brought by imprecise system model,which is a optimal filtering algorithm,it solves the problems of Kalman filter on bad stability and low accuracy and improves the effect of Kalman filter.In this paper, the experiment result shows the provided method is effective and available in deformation monitoring application. Key words:Kalman flter; neural network; deformation;monitoring; pensation 1卡尔曼滤波的基本理论[X=Φ-1X-+F--1(1) 卡尔曼滤波是最优估计方法的一种,它通过不断L=BX+ 递推、修正估计过程,且无需对大量观测数据进行存式中:2为第k期的动态噪声;4为第k期的观测 储使其对动态数据处理可以具有良好的效果。
因此噪声;X.为时刻k时的参数;L为第k期的观测值; 在实际应用中通常需要先对...