Predicting ship fuel consumption using a combination of metocean and on-board data

Fuel Oil Consumption (FOC) accounts for a significant proportion of a vessel's operating costs. The cost of fuel for a fishing vessel operation may often go up to 50% or more. Accurate forecasting FOC in voyage planning stage is essential for route optimization decision support system with the objective of fuel-saving, which is difficult because the future state of the vessel and its power and machinery systems for fuel modelling are not available during route planning stage. Moreover, the state of the environment conditions and its impact on vessel performance should be considered. In this paper, machine learning approaches were applied to predict FOC from plannable in-situ variables and modelled speed through water. The latter is estimated from speed over ground and environmental variables in this work, whose prediction is also critical for decision support systems to avoid collisions. By applying the proposed methodology, the final selected Random Forest models can achieve high mean accuracies (over 92%) in predicting fuel consumption on unseen future data. •A novel 2-stage model development methodology for ship fuel consumption prediction.•Stage 1 estimates speed through water from speed over ground and Metocean data.•Stage 2 predicts fuel consumption from stage-1 output and on-board measurements.•Multiple machine learning algorithms developed and compared for evaluation.•The selected 2-stage random forest models achieve high accuracy in future forecast.