Intelligent particle filter based on bat algorithm

Abstract

Particle filer is apt to have particle impoverishment with unstable filtering precision, and a large number of granules are required to estimate the nonlinear system accurately, which reduces the comprehensive performance of the algorithm. To solve this problem, a new particle filter based on bat algorithm is presented in this paper, where particles are used to represent individual bat so as to imitate the search process of bats for preys. In traditional resampling process, particles are directly discarded, the improved algorithm adopts another approach and solves the problem of particle impoverishment. It combines the advantages of particle swarm optimization algorithm and harmonic algorithm perfectly. New particle filter has capacity of global and local search and is superior in computation accuracy and efficiency. By adjusting frequency, loudness, and impulse emissivity of particle swarm, the optimal particle at that time is followed by particle swarm to search in the solution space. The global search and local search can be switched dynamically to improve the overall quality of the particles swarm as well as the distribution rationality. In addition, the improved particle filter uses Lvy flight strategy to avoid being attracted by harmful local optimal solution, it expands the space of research and further promotes the optimization effect of particle distribution. Using the useful information about particle swarm, improved particle filter can make particles get rid of local optimum and reduce the waste of iterations in insignificant status change. Based on the number of valid particle samples, it can improve quality of particle samples by expanding their diversity. In information interaction mechanism of improved particle filter, the method in this paper sets scoreboard of particle target function to compare the value of particle target function at each iteration sub-moment with the value of target function on scoreboard to gain global optimum of all particles at current filtering moment. Taking information interaction between global optimum and particle swarm, the guiding function of global optimum is realized. The process of particle optimization is ended prematurely through setting a maximum iteration or termination threshold. There is a tendency for the whole particle swarm closing to high likehood area without global convergence so that the advantages of improved particle filter in accuracy and speed will not be damaged. In addition, convergence analysis and computational complexity analysis are given in this paper. Experiment indicates that this method can improve the particle diversity and prediction accuracy of particle filter, and meanwhile reduce the particle quantity obviously which is required by the state value prediction for nonlinear system.