Particle Filtering Method For Multi-population Cooperative Resampling

As a kind of filtering method suitable for nonlinear and non-Gaussian system,particle filter plays an irreplaceable role in various fields.However,due to the introduction of resampling technology,the problem of particle impoverishment in particle filter seriously affects the state estimation accuracy of particle filter.Therefore,how to improve resampling technology and make particles evenly distributed in high probability area has become an unavoidable research difficulty and hot spot in the improved particle filter algorithm.The main work of this paper is as follows:(1)This paper presents a particle filter method based on adaptive M-H resampling.Firstly,an M-H resampling strategy with two proposed distributions is used to resample low weight particles,that is,randomly extract high weight particles for Gaussian mutation or cross operation of high and low weight particles,so as to promote the movement of low weight particles to the high probability region.Secondly,according to the proportion of low weight particles in all particles,the adaptive functions of these two proposal distribution selection probabilities are constructed to further improve the overall quality of particles,and the effectiveness of the method is verified by simulation analysis.(2)Aiming at the problem of particle impoverishment and sensitivity to initial value,this paper further studies and proposes a particle filter method of multi-population cooperative resampling.Firstly,particles are divided into several populations,and importance sampling is completed independently.Then,a circular transfer model is adopted to distribute the sets of low weight particles in various populations are suggested to be distributed as the M-H sampling of high weight particles randomly selected from the previous population for Gaussian variation or the cross operation of high weight particles and low weight particles,so as to increase the diversity of particles.The simulation results show that the proposed method can improve the state estimation when the initial value deviates from the true value,and make the particles more evenly distributed around the state posterior probability density.(3)Combined with the actual measurement data of liquid level of silicon melt obtained in the liquid level calibration experiment of silicon crystal growth,the above two particle filter methods are used to estimate the liquid level of silicon melt.The experimental results show the superiority of the two methods and their application effect in engineering practice.