ABSTRACT
During a voyage, ship fuel consumption varies with different weather conditions because there are many complicated nonlinear relationships between the fuel consumption and the weather conditions so that it is difficult to directly find these correlations and predict the ship fuel consumption in the specified weather condition. In this paper, we first propose to employ Deep Belief Networks (DBN) algorithm that is able to resolve the nonlinearity problem by exploiting the latent information from the large number of the historical ship voyages and weather data, and dynamically predict the fuel consumption under various ocean meteorological condition. Then we present a system framework of predicting fuel consumption based on the DBN learning framework. Finally, the proposed approach is validated by comparing with some typical machine learning methods, e.g. SVR and BPNN. Experimental results show that the proposed method completely outperforms other methods in accuracy and efficiency, which substantially demonstrates the effectiveness of the proposed DBN-based fuel consumption prediction method.
