Ship Data Clustering for Improved Fuel Efficiency Optimization Decision Support Systems

In this paper, we propose a hybrid clustering algorithm based on hierarchical clustering in combination with string clustering, similar to that used in clustering genetic sequences, to extract clusters of operational activities from ship data. After extracting clusters (n=9), the distribution and summary statistics of each cluster are compared to Canadian Coast Guard provided operational definitions to obtain context from the clusters. The clustered data is used as the input to a fuel flow prediction model, which will in turn form the basis of an on-board decision support system for marine vessels with the goal of reducing fuel consumption. Finally, we compare the results of fuel flow predictions with models trained on a) different clusters and b) the full dataset, to determine the usefulness of these clusters in improving the performance of a fuel-optimization decision support system for on-board marine vessel use.