针对基于滤波方法的最大似然参数估计步长序列过于单一、算法收敛缓慢并很容易收敛于局部最优解的问题,提出了基于似然权值的在线EM参数估计算法(LWOEM)。通过粒子滤波方法实时估计系统的状态值变化,结合最大似然方法计算静态参数的点估计,然后通过计算更新参数的似然值来动态更新步长序列。与在线EM参数估计算法(OEM)的实验结果比较,表明该算法具有更好的适应性和收敛效果。
The maximum likelihood parameter estimation based on particle filtering,since the updating stepsize sequence is too onefold,the algorithm converges slowly and can easily converge to local optimal solution.Proposed a likelihood weighting online EM parameter estimation algorithm(LWOEM),using particle method to estimate the system state that changes over time.Updated the parameters recursively by calculating the point estimation of the parameters,then calculating the likelihood value of the updated parameters to dynamically updated the stepsize sequence.Compared with the online EM parameter estimation algorithm(OEM).The experiment result shows that the method is of good adaptability and convergence effect.