迭代集合卡尔曼滤波方法的性能比较研究  

Comparative Studies on Iterative Ensemble Kalman Filter Methods

作　　者：徐宝兄 摆玉龙[1] 邵宇[1] 黄智慧[1] 

机构地区：[1]西北师范大学物理与电子工程学院甘肃省原子分子物理与功能材料重点实验室,甘肃兰州730070

出　　处：《遥感技术与应用》2015年第6期1182-1188,共7页Remote Sensing Technology and Application

基　　金：国家自然科学基金项目"进化计算类智能算法在数据同化误差处理中的应用研究"(41461078);甘肃省高校基本科研业务费资助项目

摘　　要：针对数据同化过程中模型的非线性问题,通过分析对比得出了一种适合强非线性系统的迭代集合Kalman滤波(IEnKF)。在Lorenz-63模型的框架内,比较分析集合Kalman滤波(EnKF)、迭代集合Kalman滤波(IEnKF)和迭代扩展卡Kalman滤波(IEKF)在集合数、观测误差方差、放大因子和模型步长不同时同化性能差异,由此探讨这3种方法的优劣。研究结果表明:随着集合数的增加,3种算法的同化性能都得到了一定的改善;放大因子的增大,使其同化性能变差且EnKF呈现出多重波峰波谷的现象;3种方法的均方误差(RMSE)随观测误差方差和模型步长的增大而增大,其同化精度都变差;而IEnKF同化性能最优,更具有鲁棒性。With regard to model non-linear problems in data assimilation process,an Iterative Ensemble Kalman Filter（IEnKF）is derived by thoroughly analysis and comparison.Within the framework of Lorenz-63 model,this paper compared the different performances among the following three methods,Ensemble Kalman Filter（EnKF）Iterative Ensemble Kalman Filter（IEnKF）and Iterative extended Kalman Filter（IEKF）,by changing ensemble numbers,observation error variance,the inflation factors and the model steps.The final comparative studies show that the assimilation accuracy of all three algorithms can be improved when ensemble numbers increase.When we change the inflation factors,the assimilation results are becoming worse and the EnKF presents obvious multihill and multivalley phenomena.The RMSE of all three algorithms increase when observation error variance and the model steps increase,and the results of algorithms get worse as well.The results show that the IEnKF is the most optimal algorithms with a much better robust performance.

关 键 词：数据同化 Lorenz-63模型 集合KALMAN滤波 迭代集合Kalman滤波 迭代扩展Kalman滤波 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]