Multi-Algorithm Fusion Method for State Estimation of Usv Based on Gs-Stsckf and Credibility Theory Coupling

Abstract

To improve the accuracy of ship motion state estimation in complex environments, a multi-algorithm fusion unmanned surface vehicle (USV) state estimation method based on Gaussian sum Strong Tracking Square Root Cubature Kalman Filter (GS-STSCKF) and credibility theory coupling for nonlinear non-Gaussian systems is proposed. First, the existing STSCKF algorithm is improved by using the equivalent dynamic adjustment theory and Cholesky factorization. The fading factor is directly applied to the process noise covariance matrix, and each element of the process noise covariance is finely adjusted using multiple fading factors. Additionally, two methods for calculating the fading factor matrix are proposed. These improvements enhance the model interpretability of the algorithm and significantly improve its estimation performance. Second, the GS-STSCKF algorithm is designed, which uses enhanced Gaussian fitting techniques to improve the non-Gaussian processing capability of the STSCKF algorithm, significantly enhancing filtering accuracy. The improved Gaussian fitting method uses a Gaussian component fusion strategy in the reduction control section, ensuring real-time performance while retaining more Gaussian components to improve filtering accuracy. Next, a trust factor within the GS-STSCKF framework is constructed. Based on the calculation of the trust factor, a condition is derived for designing high-performance filters for nonlinear non-Gaussian systems. This condition, combined with the range of trust factor values and the hunter-prey and dingo optimization fusion algorithm (HPO-DOA) algorithm, achieves real-time adjustment of the observation noise covariance, addressing the shortcomings of the existing STSCKF algorithm in adjusting the observation noise covariance. Additionally, coupling analysis of credibility and strong tracking theory is conducted to study their closed-loop optimization iterative relationship. Then, an improved roulette genetic algorithm is used to optimize the initial weights of the Broad Learning System (BLS), and the HPO-DOA algorithm is used to optimize regression parameters, thereby improving the convergence speed and network accuracy of BLS. The nonlinear mapping and adaptive capabilities of the improved BLS network are used to correct the results of the credible GS-STSCKF algorithm from a compensation perspective, optimizing and correcting the linear approximation error caused by pseudo-matrix construction and further improving the estimation accuracy of the filtering algorithm. Finally, the advanced nature and superiority of the proposed algorithm are verified through simulations and actual data.

Keywords: Unmanned surface vehicles, Nonlinear non-Gaussian, Broad learning system, Gaussian sum strong tracking square root cubature Kalman filter, Credibility filter, Closed-Loop Coupling