Adaptive Data-driven Predictor of Ship Maneuvering Motion Under Varying Ocean Environments

Abstract

Modern marine vessels operate increasingly autonomously, enabled by the strong interaction between data acquisition and analysis. The data-driven technology has been widely applied and significantly benefits maritime clusters by providing real-time predictions, optimizations, monitoring, controlling, improved decision-making, etc. While offshore engineering applications are usually operating in highly dynamic environments, which is an unavoidable obstacle when developing motion predictors. To this end, we propose an adaptive data-driven predictor aiming to supply decision support for vessels under varying ocean status. The predictor based on the Gaussian Process can decide whether and when to update itself from the assessment of external situations. By optimizing the ancient model with new observations, the adaptive model better fits the current situation, and efforts of re-training from scratch could be saved. Co-simulation, as an enabling tool, is utilized to simulate the dynamic ocean environments and ship maneuvers. Experimental results have demonstrated the effectiveness of the adaptive predictor, especially when unseen weather is encountered.