Clustering Applied to Large Sets of Environmental Conditions for Selecting Typical Scenarios for Ship Maneuvering Real-Time Simulations

Abstract

The methodology described in this paper is used to reduce a large set of combined wind, waves, and currents to a smaller set that still represents well enough the desired site for ship maneuvering simulations. This is achieved by running fast-time simulations for the entire set of environmental conditions and recording the vessel’s drifting time-series while it is controlled by an automatic-pilot based on a line-of-sight algorithm. The cases are then grouped considering how similar the vessel’s drifting time-series are, and one environmental condition is selected to represent each group found by the cluster analysis. The measurement of dissimilarity between the time-series is made by application of Dynamic Time Warping and the Cluster Analysis is made by the combination of Partitioning Around Medoids algorithm and the Silhouette Method. Validation is made by maneuvering simulations made with a Second Deck Officer.