Application Of Intelligent Filtering Algorithms In Data Assimilation

Iterative intelligent filtering algorithms are becoming a new kind of control technology;the development of the technology has opened up a new way to solve the problem of complex nonlinear system. Data assimilation method is aimed at the integration of the physical model and multi-source data, traditional data assimilation techniques have been widely used in the field of linear system, however, when the system is nonlinear, the filtering based data assimilation methods will have filter divergences problems.With regard to the filtering divergence problems of nonlinear system, the main research works of this thesis includes the following aspects:(1) Some kinds of iterative intelligent learning algorithms are introduced into the thesis for nonlinear systems, we discussed the initial state of the algorithm, the learning speed and convergence problems.the research highlights the iteration of this kind of intelligent algorithm and rapid learning speed.(2) In this thesis, with regard to iterative learning control problem in strongly nonlinear system, an IEnKF is further derived by thoroughly analysis and comparison. Within the framework of Lorenz-63 model, we compared the different performances among the following intelligent filtering methods: EnKF, IEnKF and IEKF by changing ensemble numbers,observation error variance, the inflation factors and the model steps, thus we can obtain more accuracy and optimal assimilation algorithm for strong nonlinear filtering system.(3) With regard to an IEnKF Local Convergence problems, by analyzing IEnKF in terms of Gauss-Newton iteration, a modified IEnKF algorithm was obtained with global convergence Strategy, this algorithm was used to ensure each iteration approaching to the true state in data assimilation process.Based on the low-dimensional systems(eg, Lorenz-63 chaotic systems) and the high-dimensional systems(eg, Lorenz-96 chaotic systems), the numerical experiments were developed to test the sensitivity of the modified intelligent method from ensemble numbers, observation error variance, the inflation factors and the model steps.we maked it some Comparative Studies among the IEnKF and EnKF,The results show that Modified IEnKF can follow the true state of a highly nonlinear system very well.In this thesis, potential applications of three kinds of intelligent filtering algorithms are analyzed and discussed for nonlinear data assimilation systems. We compared the different performances among the three methods, an optimal filtering method for the strong nonlinear system is proposed. The results and conclusion will be reference for nonlinear data assimilation system in the future.