Comparison of Using Artificial Neural Network and Gaussian Process in Ship Energy Consumption Evaluation

Abstract

Shipping is a major transportation mode over the world, which also contributes a large proportion of the global CO2 emissions. In order to better control the shipping emissions, it is important to evaluate the ship energy consumptions and further propose appropriate abatement options to reduce shipping emissions. For more accurate evaluation, the relationship between influencing factors and ship fuel consumption has to be identified. To do so, a systems approach is commonly applied which can consider the interactions in the ship energy system. However, due to the complexity of the system, it can be computationally expensive to implement a systems approach. In this paper, two statistical modeling approaches, including the artificial neural network and the Gaussian process, are developed to evaluate the ship energy consumption. The methods based on statistical models are more efficient as they require less evaluation of the original complex system. The performances of two statistical models are further compared in a case study of a chemical tanker. The results show that both models have similar performance in ship fuel consumption prediction. Gaussian process model can account for the uncertainty in the prediction. However, it is sometimes time consuming to implement. The case study indicates that speed reduction has the largest abatement potential. Weather routing and trim optimization can also save the fuel consumption.