磷虾群优化的改进粒子滤波算法

Improved particle filter algorithm optimized by krill herd

摘要:

标准的粒子滤波存在着权值退化问题,重采样可以解决权值退化问题,但也会带来样本贫化现象.为解决样本贫化问题,提出了一种利用磷虾群优化的改进粒子滤波算法.该算法结合粒子滤波的求解过程,以磷虾个体的诱导、觅食和随机扩散运动引导粒子向高似然区域移动.首先,将粒子滤波中粒子的状态值作为磷虾群的个体位置,从而将粒子的状态估计转化为磷虾群的寻优；其次,针对粒子滤波的特点,分析了磷虾算法中可以改进的参数,对磷虾算法中个体诱导、觅食运动的权值设计了新的动态更新策略,保证算法前期全局快速寻优后期局部精确寻优,同时为保持粒子的多样性,对磷虾个体进行遗传算法中的交叉操作,并设计了新的交叉概率更新公式；最后,在标准磷虾算法的基础上分析了改进算法的收敛性,并选用一种单静态非增长模型进行仿真试验. 仿真结果表明, 所提出的算法与标准粒子滤波以及粒子群、蝙蝠算法优化的粒子滤波相比具有更高的状态估计精度和更小的均方根误差,粒子的分布更合理.

Abstract:

The common problem with standard particle filters is weight degradation. Resampling can solve this problem, but it causes another noted hindrance—sample impoverishment. To overcome sample impoverishment, an improved particle filter algorithm based on krill herd optimization is proposed. The algorithm combines the solving process of particle filters and guides the particles to move to the high likelihood area by the induced motion, foraging motion, and physical diffusion of krill individuals. Firstly, the state values of particles were regarded as the individual positions of the krill herd. Thus the state estimation of particles was transformed into the optimization of krill herd. Secondly, according to the characteristics of the particle filter, parameters that can be improved in the krill herd algorithm were analyzed. A new dynamic updating strategy was designed for the weights of the individual induced motion and foraging motion to ensure fast global optimization in the early stage and accurate local optimization in the later stage. At the same time, to maintain the diversity of particles, the crossover operation in the genetic algorithm was carried out on krill individuals, and a new crossover probability updating formula was designed for krill individuals. Then, the convergence of the improved algorithm was analyzed based on the standard krill herd algorithm. A single static non-growth model was selected for simulation. The simulation results show that the proposed algorithm has a higher accuracy of state estimation, smaller root mean square error, and more reasonable particle distribution compared with standard particle filters, particle swarm optimization algorithm optimized particle filters (PSO-PF), and bat algorithm optimized particle filters (BA-PF).