非高斯随机系统的残差分布和熵的分析与分析.pdfVIP

Abstract Abstract The state-space model is a statistical model, which is a kind of a wide range of applications and strong practicality, to a certain assumptions, the state-space model deduces the various Kalman filter and smoothing, which can be applied to all aspects of model inference, therefore, there are great advantages of the state-space model which deals with inference question, such as the parameter estimates, test ，predictor and update. This dissertation studies inference problem of Gaussian and non-Gaussian dynamic random system, which are based on the linear state-space model. The researches are organized as following: First, in linear time- varying control systems, when the random noise (input noise and observation noise) is Gaussian, the classical Kalman filter is the linear recursive filtering method and uses system status estimation and the current valuation to recur new state valuation. This dissertation raises Kalman filter algorithm and achieve the algorithm through simulation. When the random noise is non-Gaussian, this dissertation tests the result of Kalman filter through simulation. Second, the traditional maximum likelihood parameter estimation method exist deficiencies in the solution of practical problems, so this dissertation introduces the EM algorithm. This dissertation discusses the principle of algorithm and implementation steps, and applies EM algorithm to the parameters estimation problem of the state space model, then deduces the state space model parameter estimation process, based on Kaman smoothing algorithm and the EM algorithm, the simulation results also showed that EM algorithm