Improved Particle Filter Using Clustering Similarity of the State Trajectory with Application to Nonlinear Estimation: Theory, Modeling, and Applications

Table 2

Accuracy improvement percentage of mean and variance of RMSE for five nonlinear algorithms running 100 times under both the Gaussian and non-Gaussian noise models.

Filter	RMSE	Gaussian noise			Non-Gaussian noise
		SIR	APF	GPF	SIR	APF	GPF
ECSPF	Mean	68.4%	68.0%	69.2%	55.5%	56.1%	56.9%
	Variance	85.4%	81.5%	88.7%	51.4%	30.3%	54.0%
CCSPF	Mean	65.4%	65.0%	66.3%	51.7%	52.3%	53.2%
	Variance	73.7%	66.7%	79.6%	40.1%	14.2%	43.4%

Note: SIR = sequential importance resampling; APF = auxiliary particle filter; GPF = Gaussian particle filter; ECSPF = Euclidean distance clustering similarity particle filter; CCSPF = Chebyshev distance clustering similarity particle filter; RMSE = root mean squared error.