Prediction of fuel consumption and evaluation of energy efficiency of ships in Bangladesh using machine learning

Abstract

CO2 emissions from ships and their environmental effect are intensifying with the expansion of ship transportation activities worldwide including Bangladesh. Therefore, it is vital to reduce fuel consumption to facilitate energy-efficient shipping for the inland, coastal, and sea-going ships of Bangladesh. This research aims to develop the fuel consumption prediction model of sea-going ships in Bangladesh. Establishing CO2 emission inventories through quantifying the shipping emissions is another aim of this research because of the potential to comprehend the effects of the activities undertaken and the data about the current state of the relevant regions provided by emission inventories. The present study also aims to assess the ships’ operational energy efficiency based on the Energy Efficiency Operational Indicator (EEOI). The current research implements machine learning, one of the data-scienceapproaches, considering its wide range of advantages and avoiding the drawbacks of conventional approaches. Fourteen machine learning algorithms are implemented to develop models having default hyperparameters using pre-processed noon report data of fivesea-going bulk carriers of Bangladesh to predict fuel consumption. Again, from these algorithms, 12 algorithms are used through 10-fold cross-validation to develop prediction models. The Root Mean Square Error, Mean Absolute Error, and Coefficient of Determination are used as the evaluation metrics. The developed models are validated using the data of one of the five vessels. The most effective and feasible model is selected based on the evaluation metrics and validation process. CO2 emissions in cruising, maneuvering, and hoteling phasesdue to fuel consumption of the four concerned vessels are calculated. EEOI values of the three concerned vessels are determined utilizing the determined CO2 emissions.Additionally, EEOI values are recalculated for two different scenarios, such as three distinct percentages of fuel saving and fully laden conditions. The results show that the developed and finally selected model would be a suitable tool for predicting ships’ fuel consumption in Bangladesh. Moreover, CO2 emission inventories of the concerned vessels are developed and analyzed. The energy efficiency of the corresponding vessels is successfully evaluated and investigated how the energy efficiency changes at various operating conditions. However, there are discrepancies between the actual and predicted fuel consumption. Besides, emission inventories development and energy efficiency evaluation could not be accomplished for all five vessels. The reasons are the extreme difficulties of collecting data and the lackof stored voyage data of the vessels of Bangladesh.