人工鱼群粒子滤波算法

Research on artificial fish school algorithm⁃particle filter

针对粒子滤波算法存在重要性密度函数难以选取的问题,将人工鱼群算法的随机搜索优化的思想引入到标准粒子滤波中,提出一种人工鱼群粒子滤波算法。该算法的基本思想是在重要性权值计算中引入AFSA的觅食行为和聚群行为,使得先验粒子向高似然区域移动,提高估计精度,并对人工鱼群的视野进行改进,增加有效样本数。将提出的算法应用到标准单变量非静态增长模型中进行仿真验证,对比提出的人工鱼群粒子滤波算法与标准粒子滤波和扩展卡尔曼滤波算法。结果表明,人工鱼群粒子滤波的有效样本多,均方根误差最低,说明其估计精度高于粒子滤波和扩展卡尔曼滤波。在增大过程噪声方差情况下,三种算法的均方根误差均随着噪声方差的增大而增大,但是人工鱼群粒子滤波状态估计精度仍优于粒子滤波和扩展卡尔曼滤波算法。

Since it is difficult for the particle filter algorithm to select the importance density function,the idea of random search optimization of artificial fish school algorithm(AFSA)is introduced into standard particle filter,and an artificial fish school algorithm⁃particle filter(AFSA⁃PF)is proposed.The basic idea of this new algorithm is to introduce the foraging behavior and the clustering behavior of the AFSA into the calculation of the importance weight,make the prior particles move to the high likelihood region,so as to improve the estimation accuracy.The visual field of the artificial fish school is ameliorated to increase the number of valid samples.The proposed algorithm is applied to the standard univariate non⁃static growth model for simulation verification.Simulation results and their analysis demonstrate preliminarily that the proposed AFSA⁃PF is more effective than the standard particle filtering and extended Kalman filtering algorithm.The artificial fish school particle filtering has more effective samples and the lowest RMSE,which indicates that its estimation accuracy is higher than that of the particle filtering and the extended Kalman filtering.When the process noise variance is increased,the RMSEs of the three algorithms increase with the increase of the noise variance,but the state estimation accuracy of AFSA⁃PF is still better than that of particle filtering and extended Kalman filtering algorithm.