Online parameter identification and dynamic model reconstruction for AUV based on adaptive extended Kalman filter and...

The online parameter identification problem for a six-degree-of-freedom autonomous underwater vehicle (AUV) under ocean current disturbances and measurement noise is the focus of this paper. To improve the identification accuracy in such complex conditions, an integrated closed-loop identification framework is first proposed, which establishes a collaborative mechanism between real-time state estimation and online parameter identification. To realize this framework, an advanced AEKFPEM algorithm is formulated by integrating the adaptive extended Kalman filter (AEKF) module with the prediction error method (PEM) module. Furthermore, to enhance the robustness and anti-disturbance capability of the state estimation process, a specific fault-tolerant mechanism is designed and embedded into the EKF with Sage-Husa noise estimation to resolve the loss of positive definiteness in noise covariance matrices. To reduce the computational burden caused by numerous unknown parameters in the AUV dynamic model, a recursive update expression based on the Sherman-Morrison-Woodbury formula is derived to replace the direct inversion of the Hessian matrix in the conventional PEM. Finally, a dynamic model reconstruction study is designed to verify the practical utility of the algorithm under model mismatch conditions. In this study, a simplified model is constructed online using real-time navigation data from a high-fidelity model. Subsequently, the performance of the simplified model is evaluated via 5-second motion prediction experiments. Comparative results demonstrate that the proposed AEKFPEM achieves robust online parameter identification and superior predictive performance, even under model structural mismatch.