ABSTRACT
The paper introduces a machine learning method for the evaluation of probabilistic grounding risk accounting for ship motion uncertainties in real operational conditions. The approach makes use of big data from Automatic Identification System (AIS), nowcast and gridded bathymetry (GEBCO) data. A machine learning method that combines Principal Component Analysis (PCA) with Multiple-Output Gaussian Process Regression (MOGPR) methods is used to predict selected features of ship motion dynamics namely ship sway, surge accelerations and drift angles. Predicted ship motion trajectories are quantified to present ship position distribution functions in the time domain using a Gaussian Progress Regression (GPR) flow method while the dynamic safety contours of bathymetry maps are extracted based on the ship’s draft and Under Keel Clearance (UKC). The method is applied for probabilistic grounding risk evaluation of a Ro-Pax ship operating between two ports in the Gulf of Finland. The results demonstrate that the presented methodology can predict probabilistic grounding risk reflecting ship motion uncertainties. Thus, it may assist with the development of novel ship decision support systems for proactive grounding risk mitigation.
