USV Motion State Estimation Method Based on Credible Fusion STSCKF and GANs-BLS Closed-Loop Coupling Iterative Technology

Traditional methods for unmanned surface vehicle motion state estimation often struggle to balance the theoretical guidance of model-driven approaches with the adaptability of data-driven methods. To address this challenge, this paper proposes a novel closed-loop coupled iterative estimation framework integrating a credibility-fused strong tracking square-root cubature Kalman filter and a generative adversarial network-broad learning system. First, to overcome the weak adaptability and poor interpretability of conventional fading factor adjustment, the proposed method introduces multiple fading factors that directly participate in the real-time dynamic feedback correction of process noise parameters. Second, for credibility assessment in nonlinear systems, two distinct credibility algorithms (based on linearized analysis and nonlinear analysis) are established, and two credibility fusion strategies are proposed based on fused objective function superposition and mutual evaluation of credibility criteria. Third, to compensate for linear approximation errors, a closed-loop coupled iterative mechanism is designed, using regularization parameters as a bridge between data-driven and model-driven approaches. Both simulation and field test results demonstrate that the proposed algorithm improves state estimation accuracy by at least 19.9% and reduces noise covariance error by at least 20.6% compared to existing methods, while maintaining real-time processing capabilities.