Classifying the weights of particle filters in nonlinear systems

Abstract Among other methods, state estimation using filters is currently used to increase the accuracy of systems. These filters are categorized into the following three branches: linear, nonlinear with Gaussian noise, and nonlinear with non-Gaussian. In this paper, the performance of particle filters is investigated via nonlinear, non-Gaussian methods. The main aim of this paper was to improve the performance of particle filters by eliminating particle impoverishment. A resampling step was proposed to overcome this limitation. However, resampling usually leads to a dearth of samples. Therefore, to virtually increase the number of samples, the available samples were broken into smaller parts based on their respective weights via a clustering approach. The experimental results indicate that the proposed procedure improves the accuracy of state estimations without increasing computational burden.