Research On Parallel Computing Of Sequential Data Assimilation Methods

With the increase of the complexity of the earth model,the integration of various new types of high-precision information,and the integration of multiple models with multiple angles,the data assimilation technology will take longer to compute.The problem can been effectively solved by parallel data assimilation frame.At the same time,the performance of the data assimilation system is easily affected by multiple parameters in the system.Coupling intelligent algorithm to optimize the selection of data assimilation system parameters is worth exploring,while the parallel performance of the coupling scheme needs to be developed..In view of the above issues,the following work is carried out:(1)The intrinsic parallelism of the ensemble kalman filter is introduced.The principle of the two parallel strategies in the parallel data assimilation frame,the mode decomposition and the domain decomposition,and the Local Ensemble Transfer Kalman Filter are briefly discussed.Based on the Lorenz-96 model,the main interfaces and processes are designed.Finally,based on the single and double processes,in reducing the need for single-machine storage,stability of the data assimilation performance and decreasing of the time,the assimilation system has achieved good results in the three areas.(2)Data assimilation parameter optimization system coupled with DE and NSGA-2 are proposed.The optimization of variable combination in the variable space consisting of the localized radius and the covariance inflation factor is studied,and it is confirmed that the scheme has good convergence and validity from the characteristics of individual population distribution,the convergence of the optimal individual,and the variance of the population.Finally,the correctness of the optimal individual selection is verified from the performance comparison of individuals randomly selected in space and optimal individuals.Similarly,for the coupled NSGA-2 system,The optimization of variable combination in the variable space consisting of the analysis inflation factor and the covariance inflation factor is studied.It is confirmed that the scheme has good convergence and correctness from the characteristics of individual population distribution and the convergence of the optimal individual.(3)For the time-consuming problem of the data assimilation parameter optimization system coupled with intelligent evolutionary algorithms,this essay combines the parallel computing library provided by Matlab to design a parallel program for the data assimilation parameter optimization system.The effect of the number of individuals on the parallel speedup and efficiency is studied.The results show with the increase of individuals,the parallel performance will be better,but it is still constrained by factors such as the switching loss of the computing unit,the efficiency of language operation,and the time occupation ratio of the no-parallel part.In this essay,the parallel performance of PDAF is studied with Lorenz-96 model.the data assimilation parameter optimization system coupled with intelligent evolution algorithm is proposed,and the parallel design scheme is given.The research results provide an effective reference for the further study of parallel computing of sequential data assimilation methods in the future.