论文部分内容阅读
针对非线性、非高斯系统状态的在线估计问题,提出一种改进的粒子滤波算法,该算法综合考虑“优选建议分布函数”和“重采样”两种并行改进滤波性能的方法.首先通过Unscented卡尔曼滤波器产生系统的状态估计,并在协方差预测阶段引入衰减记忆因子,消弱滤波器对历史信息的依赖,增强当前量测信息对滤波器的修正作用,从而产生一个优选的建议分布函数,有效抑制了粒子退化现象;接着在重采样阶段引入MCMC(Markov Chain Monte Carlo)方法来构造马尔科夫链产生服从目标分布的粒子,使样本更加多样化,有效避免了粒子枯竭问题.最后,通过系统仿真及说话人跟踪实验,证明了该算法的有效性.
In order to solve the problem of online and non-Gaussian system state estimation, an improved particle filter algorithm is proposed. The proposed algorithm takes into consideration both “optimal recommended distribution function” and “resampled” two parallel improved filtering performance. Firstly, the state estimation of the system is generated by Unscented Kalman filter, and the attenuation memory factor is introduced in the covariance prediction stage to weaken the dependence of the filter on the historical information and enhance the correction effect of the current measurement information on the filter, thus resulting in a preferred The proposed distribution function effectively suppresses the particle degeneration phenomenon. Then, by introducing the Markov Chain Monte Carlo (MCMC) method into the Markov Chain Monte Carlo (MCMC) at the resampling stage, the Markov chain Monte Carlo method is used to construct the Markov chain Monte Carlo to produce the particles that obey the target distribution and make the samples more diversified. Problem.Finally, through the system simulation and speaker tracking experiments, the effectiveness of the algorithm is proved.