项目名称: 进化计算类智能算法在数据同化误差处理中的应用研究
项目编号: No.41461078
项目类型: 地区科学基金项目
立项/批准年度: 2015
项目学科: 天文学、地球科学
项目作者: 摆玉龙
作者单位: 西北师范大学
项目金额: 51万元
中文摘要: 基于控制理论中的稳定性、可控性和可观测性定义,探讨Kalman滤波类顺序数据同化方法产生滤波发散的原因;诠释集合数据同化中协方差放大法的数学机理,力争揭示数据同化系统中的误差表现、误差特征和演进规律,寻找在模型算子维数高、强非线性等应用背景下处理误差问题的方法;发展包括遗传杂交类误差处理法(包括凸杂交、仿射杂交等)、进化策略误差处理法等一系列数据同化系统误差处理方法;以Lorenz-98和SiB2模型作为模型算子,设计基于进化计算方法的实验型数据同化系统误差问题研究平台,进行数据同化系统案例实验;通过大量数值试验和敏感性分析,系统地比较进化计算类误差处理方法和传统误差处理方法(乘数放大法、附加放大法和松弛先验法)之间的优缺点,开展各种误差处理方法在数据同化系统中的应用研究。在以上研究基础上,旨在发展通用算法,为数据同化理论研究提供一个较为全面的误差问题研究平台。
中文关键词: 数据同化;误差传播;同化策略
英文摘要: The reasons of filter divergence for Kalman filter based sequential data assimilation methods will be discussed using the stability, controllability and observability definitions of control theory. We will try to reveal the mathematical mechanism of covariance inflation methods of ensemble data assimilation, error characteristic and error propagation laws for data assimilation. The methods for handling the error problems for high dimensional and strong nonlinearity data assimilation systems are tested. We will develop a series of error propagation methods, such as convex crossover, affine crossover, evolutionary strategy, multiplicative inflation, additive inflation, and relaxation to prior error propagation methods. Choosing Lorenz-98 and SiB2 model as the model operators, we will build an experimental data assimilation error problem research framework to test all proposed methods with the case studies of data assimilation systems. Through a large amount of numerical test and sensitivity analysis, all kinds of error parameterization methods will be compared and the advantages and disadvantages of those methods will be concluded. Based on above methods, a fully research framework for error problems will finally be built for data assimilation systems.
英文关键词: Data Assimilation;Error Propagation;Assimilation Strategy