耦合多目标遗传算法的数据同化方法参数优化研究

Research on Parameter Optimization of Data Assimilation Method Coupled with Multi-Objective Genetic Algorithms

针对影响数据同化系统性能的数据同化方法中多参数优化问题,设计了一种耦合多目标优化非支配排序遗传算法(Non-dominated Sorting Genetic Algorithms,NSGA-2)的优化框架。以Lorenz-96模型为研究对象,局地化集合转换卡尔曼滤波(Local Ensemble Transform Kalman Filter,LETKF)为实验算法,对影响其同化性能的分析膨胀因子和协方差膨胀因子进行联合优化实验。同时针对计算时间过长问题,对该框架进行并行改进。研究结果表明:该框架针对变量范围较大的参数空间,具有良好的多变量寻优效果,鲁棒性较强,并行设计节省运行时间,耦合多目标遗传算法的数据同化方法整体寻优框架设计实现简单,进一步将验证其在陆面数据同化中的应用前景。

To solve the problem of multi-parameter optimization in data assimilation method which affects the performance of data assimilation system,an optimization framework was proposedcoupled with objective optimization Non-dominated Sorting Genetic Algorithms (NSGA-2).To choose Lorenz-96 model as the research object,the Local Ensemble Transform Kalman Filter (LETKF) was used as the experimental algorithm.The analysis inflation factor and the covariance inflation factor that affect data assimilation performance were studied together in new framework.At the same time,the problem of over-long computing time was solved by paralleling the framework.The results showed that this framework had a good multivariable optimization for parameter space with large variable range and strong robustness.Parallel design could save the running time.Data assimilation coupled with multi-objective genetic algorithm and overall optimization framework design are easy to use.Further verification of its land data assimilation use were explored in the future.