论文部分内容阅读
数据同化是在动力学模型的运行过程中不断融合新的观测信息的方法论,Bayes理论是数据同化的基石。从原理、方法和符号系统为Bayes滤波在数据同化中的应用勾勒一个统一的框架。首先对连续数据同化和顺序数据同化的各种方法做了分类,然后给出了非线性系统顺序数据同化的Bayes递推滤波形式,并在此基础上介绍了典型的顺序数据同化方法——粒子滤波和集合Kal-man滤波。粒子滤波实质上是一种基于递推Bayes估计和Monte Carlo模拟的滤波方法,而集合Kalman滤波相当于一种权值相等的粒子滤波。Bayes滤波理论为顺序数据同化提供了更广义的理论框架,从基础的数学理论上揭示了数据同化的基本原理。
Data assimilation is a methodology that continuously incorporates new observational information during the operation of a dynamical model. Bayesian theory is the cornerstone of data assimilation. The application of Bayesian filtering in data assimilation from the principles, methods, and symbology draws a unified framework. Firstly, the methods of assimilation of continuous data and assimilation of sequential data are classified, and then the Bayes recursive filtering form of sequential data assimilation of nonlinear system is given. On the basis of this, a typical sequential data assimilation method Filter and set Kal-man filter. Particle filter is essentially a filtering method based on recursive Bayes estimation and Monte Carlo simulation, and the set Kalman filter is equivalent to a kind of equal-valued particle filter. Bayesian filtering theory provides a more general theoretical framework for sequential data assimilation, and reveals the basic principle of data assimilation from the basic mathematical theory.