Examining the performance of retrofit measures in real ship operation using data-driven models

The quantification of the effect of ship design measures on its energy efficiency during real operation and using data-driven models is presented. Data collected from automated logging and meteorological service provider, received before and after the ship retrofit are utilized. Initially, a data preparation analysis is carried out and the input variables for the prediction of the main engine's fuel oil consumption are selected. Afterwards, artificial neural networks (ANNs) design techniques are investigated utilizing modern machine learning methods. The generalization competence of a model is assessed based on its efficiency to cope with data that were not used in its training. Furthermore, the accuracy achieved by models incorporating different input parameters is examined. Then, the assessment of fuel consumption savings due to the design measures is performed using the ANN models which have been trained on separate data before and after the adoption of measures.