耦合遗传算法的数据同化系统误差处理方法

Error processing methods of data assimilation systems coupled with genetic algorithms

针对数据同化系统中的误差估计与处理问题,介绍了集合滤波数据同化系统中各种误差来源及特征;侧重于在集合数据同化中为防止滤波发散的乘数放大法、附加放大法和松弛先验法等模型误差处理方案,利用经典的非线性模型——Lorenz模型开展了数值试验。在此基础上,提出了一种耦合遗传寻优算法的数据同化系统,来解决以往的误差调节因子由反复实验法设定的问题;进而结合乘数放大法全局放大和附加放大法局部调节的特点,提出了一种新的混合误差处理方法。结果显示,这些方法可以在适应度函数的约束下自适应地获取最优误差因子,达到最优的同化效果,从而提出了在数据同化系统中为同化实际观测资料可采取的误差处理新思路。

With regards to error estimation and the processing problems in data assimilation,the error sources and the characteristics of ensemble Kalman filter data assimilation systems were briefly reviewed.Concentrating on the model error processing problems,the multiplicative inflation,the additive inflation and the relax-to-prior methods,the commonly used methods for preventing the filtering divergence in ensemble data assimilation,were introduced.The numerical experiments were developed based on the classical nonlinear model-Lorenz model.To solve the hard searching problem for the error adjustment factor usually done by trial and error methods,a new data assimilations system coupled with genetic algorithms was proposed.Moreover,combined with the advantages of multiplicative inflation for global expansion and the characteristic of additive inflation for local adjustment,a new blending error processing method was designed.The results show that all methods can adaptively obtain the best error factors with the constraints of the fitness function,and the assimilation results can be improved consequently.This is a new idea of error processing for assimilating real observations.