NOVEL GAUSSIAN APPROXIMATE FILTER METHOD FOR STOCHASTIC NON-LINEAR SYSTEM

摘要

In the traditional process of nonlinear non-Gaussian filtering system model, there are many disadvantages on filtering results such as inaccuracy initial value selection and low convergence, due to complex system model or the influence of diversity interference. To improve the filter accuracy and solve the above problems, we put forward a new Gaussian approximate filter method, also give the general solution and special solution on the new method. In this paper, we demonstrate that existing Gaussian approximate filter methods are based on our new scheme. The new Gaussian approximate filter method adopts measurement point correction status quadrature points to better utilize the one-step predicted density, non-Gaussian information and high-order moment information of the posterior density, which can directly update quadrature points without repeatedly producing quadrature points. In addition, our new method not only is suitable for deterministic system model but for stochastic system model. At the end of this paper, we apply our method into single variable non-stationary growth model and vertical free-fall model to verify the performance of new method. What is more, we make comparison with the existing Gaussian approximate filter methods, and the results show that our method is more effective and superior.