Application Research On Robust Filtering Method In Processing Data Assimilation Error

The purpose of the data assimilation is to balance information from the model and the observation,in order to improve the estimating accuracy of state.When the model is non-Gaussian,error statistics property assuming is not true,and there is instant human inputs,data assimilation approaches such as kalman filter and its extensions not any more suitable.In the estimating theory,the purpose of robust filtering is correct system bias,includes the model error and the modeling disturbance leads to estimation unaccuracy.The main contents are as follows:(1)The reason on filtering unaccuracy and low robustness are introduced.In the initial condition,model error and measurement error are worst,data assimilation algorithm of the Kalman filter category with Bayes's rule are not guarantee that the estimation error has a bounded growth rate.(2)The Kalman filter and theH_?filter are discussed:firstly,we introduce the Kalman filter theory;secondly,the traditional Kalman filter method have not robust properties on the change of the model and algorithm parameters,so we introduce theH_?filter which is based on the criterion of minimizing the supremum of a cost function(minimax rule),the original form of theH_?filter contains global constraints on the sequential data assimilation problems,therefore,we discuss a time-localH_?filter;lastly,we test using one-dimensional regression model:the RMSE differences between theH_?filter and the Kalman filter in the change conditions of performance level coefficient and model perturbation.The results show that,the time-localH_?filter has relatively more robust performance than the Kalman filter.(3)Two types of nonlinear filter are considered,the Ensemble Kalman filter and the Ensemble time-localH_?filter,the data assimilation is in nonlinear systems:for suitable high dimensional systems,first of all,we introduce the Ensemble transfor Kalman filter;then,to improve the filtering presicion and robust propertity,we combine ensemble filter algorithm with the time-localH_?filter to form the Ensemble time-localH_?filter,we discuss some specific forms of its,we show that an ensemble Kalman filter with certain covariance inflation is essentially an Ensemble time-localH_?filter.Lastly,we use nonlinear model Lorenz-96 to assess the relative robustness of the Ensemble time-localH_?filter with certain forms.It can be show that the Ensemble time-localH_?filter have more the robustness properity and the filter accuracy than Ensemble transfor Kalman filter.The robust ensemble filtering theory is introduced,which do not need to assume the statistical nature of the model and observation error,and estimation error growth rate is bound,it can be effectively handle error problem in data assimilation.This method can widely apply to the data assimilation of the others nonlinear systems,because of its robustness nature.