Research On Linear/Non-linear Mixture Particle Filter In Inertial Integrated Navigation System

Filtering technique is one of the key technology in integrated navigation system, in which all kinds of filtering theory is often the first application, while the demands for navigation systems also contributed to the development of filtering theory. Integrated navigation system has a lot of non-linear aspect of the traditional filtering methods which are sometimes difficult to solve such problems. Particle filter can successfully solve the nonlinear filtering problem in theory, but there are some problems such as complexity of the algorithm, computational complexity which restrict the direct application of PF in the integrated navigation system. This paper systematic studies the nonlinear PF and its application in integrated navigation system. Based on characteristics of the integrated navigation system, some innovative hybrid filtering technology has a targeted combination of linear and nonlinear filtering. The innovative filtering algorithm is proposed to solve the problem of the program and implementation for integrated navigation system and achieved better performance than traditional filtering algorithm.Firstly, the paper analyzed systemically the modeling approach about integrated navigation system, in which there often have linear/nonlinear mixture modeling factors. Integrated navigation system itself is a nonlinear system. Whether it is solving the navigation parameters, or establishing the navigation equations are involved in nonlinear problems. While in engineering ofen a small amount of error are used establishing the approximate linear filtering equation. The hybrid model can describe the system filter model more accurately and simply when something are bound by certain conditions, such as accessing the navigation information, establishing the navigation sensor model, or the movement of carrier status. This paper analyzed the filter characteristics of the hybrid filter model and presented the modeling method of the hybrid filter model in the integrated navigation system.To the linear and nonlinear filter model in the integrated navigation system, the paper analyzed the filtering strategies to solve the problem of such filtering by a hybrid particle filter, and proposed a method of mixture particle filter based on Gaussian particle filter. In the filter implementation process, GPF need obtainning Gaussian distribution parameters of the state variable, while still using the algorithm principle and process of the particle filter. This makes the GPF can be combinated well with other Gaussian filter methods such as KF, EKF, or UKF. It will result in a new filtering method which has better overall performance. It is the core idea of these algorithms that the system is divided into the linear part and nonlinear part by the system characteristics of the linear and nonlinear model and the filtering implementation process. Then the traditional filtering methods can be used to the realization for the linear part, such as obtainning GPF's parameters, and the GPF original algorithm flow can be still used for the non-linear part. The improved new algorithm can significantly reduce the computation while ensuring the filter accuracy of the system. So it is more suitable for the integrated navigation system. This improving strategie can also be applied to weakly nonlinear and nonlinear hybrid filter system, thus its scope of application is expanding.For the linear and nonlinear hybrid systems which is composed by part nonlinear state variable appeared in the combined navigation system, the paper gives an algorithm which using RBPF filtering algorithm to implement the process and filter performance characteristics. It is the basic idea of RBPF algorithm that the system is divided into sub-linear space and non-linear space by the way of structural decomposition. The difficulty of the algorithm is the mutual coupling of linear part and nonlinear part after the decomposition, and the complex algorithm implementing. The paper firstly established the SINS/GPS linear and nonlinear mixed filter model. Then it done the structure decomposition to the system model based on the edge probability theory. The specific filtering algorithm process of combined navigation system is presented based on RBPF algorithm. Simulation results show that the RBPF algorithm effectively reduces the dimension of the nonlinear part in the particle filter, and the particles number of the sample has a significant reduction. When the number of particles selected is same, RBPF algorithm will improve the filtering accuracy, and then the filter computation is effectively reduced. At the same time the paper discussed that the data processing to the integrated navigation system can be working by a new algorithm proposed, hybrid Gaussian particle filter method. Different from RBPF filtering methods, the main idea of Gaussian hybrid particle filter is to improve filter performance through optimizing the filter process. Gaussian mixture particle filter divides the GPF into linear and nonlinear stages by the filtering time sequence. The GPF distribution parameters are get by the traditional method in the linear phase. And the rest are implementated still by the particle filter framework. The process of Gaussian mixture particle filter algorithm is simpler. The method has distinct advantages in the case of larger number particles.Finally, this paper advanced a hybrid federated filtering algorithm contraposing the nonlinear part in parts of sub-filters appearing scattered filter. And the method is applied to the multi-information fusion integrated navigation system. When there is some nonlinearity in the system, the traditional federal Kalman filter will make the system linearization and redesign, which will affect the system modeling and filtering performance. It is the main idea of Federal hybrid filtering algorithm that the nonlinear filtering problem in the sub-filters is solved by particle filter while the linear filters are still linear KF. Then the information of all the sub-filters will be combined organically in the main filter. And the filtering effect of better overall performance will be obtained. So the navigation performance is improved. Compare with that the traditional Federal Kalman filter using KF filtering framework, mixed Federal algorithm doesn't have the constraints of the Gaussian linear assumptions in all sub- filters algorithm filters, has a wider range of applications, and establishes the filter model more conveniently.