Multiple Model Algorithms And Its Application In Markov Jump Hybrid System

With the development of science and technology,with the needs of military and civil,the theory of hybrid system estimation has been applied in hybrid systems with Markov jump characteristics,such as maneuvering target tracking,fault diagnosis and recognition,and speech signal recognition.The estimation of hybrid system is a very important branch of multi-source information fusion.It is also important to obtain accurate,reliable and stable estimation results from the hybrid system.Therefore,it has a great theoretical and practical value to improve the estimation algorithms based on Markov jump hybrid system.With the development and improvement of the basic theory of Multiple-Model algorithm,Multiple-Model algorithm has gradually become the mainstream for state estimation of hybrid systems.In order to verify the estimation performance of the multi-model algorithm in the Markov switching hybrid system,a general model of the maneuvering target tracking system with Markov jump characteristics is presented,including the CV model,the CA model,the CT model and the Snake Maneuvering model.Based on the EKF,UKF and PF algorithms in the system estimation,the simulation and comparison are carried out.Secondly,the Multi-Model algorithm is studied,and the Static Multi-Model algorithm and Dynamic Multi-Model algorithm are analyzed,and the most representative Interactive Multi-Model(IMM)algorithm is introduced.The IMM algorithm is widely used in hybrid systems with mode switching because of its high cost-effectiveness.In recent years,researchers have combined the advantage of IMM and Particle Filter algorithm for the problem of nonelinear and non-Gaussian in hybrid systems.Therefore,the Interacting Multiple Model Particle Filter(IMMPF)algorithm has received a great deal of attention.When the system modes change frequently,the peak-error is very large baseded on the algorithm of IMMPF.Using the traditional resampling method is not enough to obtain the satisfactory estimation accuracy.For the problem of particle degeneration in IMMPF algorithm and the disappearance of the particle diversity caused by the resampling process,an improved IMMPF algorithm is proposed,which is called Interacting Multiple Model Particle Filter Optimization Resampling algorithm,Simulation experiments of maneuvering target tracking show that the proposed algorithm can obtain smaller steady-state error and peak-error than IMMPF algorithm.Finally,the Variable-Structure Multiple Model(VSMM)algorithm is studied,and the advantages and disadvantages of existing Multi-Model algorithms such as Model Group Switching algorithm,Likely Model Set algorithm and Expect Mode Augmentation algorithm are analyzed.A new model set adaptive algorithm is proposed by using the Kullback-Leiber(KL)criterion and the expectation model method in combination with the existed VSMM algorithm.Through the simulation of maneuvering target tracking,it can prove that this algorithm can reduce the burden of the system while guaranteeing the performance of estimation,so the algorithm has high cost-effective.