Online parameter identification and real-time manoeuvring prediction for a water-jet USV based on weighted multi-innovation prediction error method integrated with dynamic window strategy

Abstract

Research on online parameter identification and real-time manoeuvring prediction for a class of water-jet unmanned surface vehicle (USV) is carried out in this paper. Utilizing actual sailing data from a water-jet USV, the weighted multi-innovation prediction error method integrated with dynamic window strategy is proposed to identify the manoeuvring parameters of the USV model online. Subsequently, real-time prediction of the water-jet USV's motion is achieved based on the established time-varying model. The thrust generation of water-jet propulsion system and the effect of rotational current on the USV's motion are analyzed simultaneously, and then a three-degree-of-freedom mathematical model is established for the water-jet USV equipped with two water-jet propulsion systems. Due to the weakening of the correction ability of the prediction error method in the later stage, an adaptive step factor with phase adjustment is designed to improve the response accuracy to the error innovation and maintain the algorithm's correction ability. Since the prediction error method updates the identification value using only a single innovation each time, incorporating multi-innovation theory enhances the utilization of historical data, allowing the algorithm to more accurately reflect the current state or trend. In order to fully consider the differences between data points, an adaptive weighting strategy is developed to assign weights according to the contribution of the data in the innovation window to USV modeling, so as to enhance the tracking performance of the time-varying parameters. Aiming at the outliers in the collected data, a dynamic innovation window strategy is designed, and then the data in this window is filtered by Quartile algorithm and the outliers are detected by local outlier factor, so that the window could contain more effective sailing state information. A large amount of actual test data analysis demonstrates that, the algorithm proposed in this paper could achieve more accurate online identification of water-jet USV model parameters and more precise real-time prediction of USV motion, which would provide strong support for safe navigation and efficient control of USV.