Characteristic analysis and filtering algorithm design for UNGM model

Abstract

The univariate non-stationary growth model (UNGM) is widely used in the verification of nonlinear filters, and the unscented Kalman filter (UKF) is often used as the reference filter for comparative analysis when using this model to evaluate the filter performance. However, due to the strong nonlinearity of UNGM and the change of model properties with different parameter settings, the estimation misalignment problem due to different reasons will occur when UKF is used for filtering. To solve these problems, this paper analyzes the complex characteristics of UNGM in filtering process, and proposes an UKF with sliding sampling module(SSUKF). The algorithm is optimized on the basis of UKF, and can effectively deal with the complex characteristics of UNGM by sampling and analyzing the filtering information in the filtering process and correcting the distribution of Sigma points in real time. SSUKF is applied to UNGM under different parameters and compared with UKF and bootstrap particle filter(BPF). The simulation results show that SSUKF can effectively solve the misalignment problem when UKF is applied to UNGM, and the calculation speed is better than BPF. Compared with UKF, SSUKF is suitable as a benchmark filter for evaluating the performance of nonlinear filters using UNGM.