Comparative analysis of machine learning prediction models of container ships propulsion power

Regulations on Greenhouse Gas (GHG) ship's emissions and air pollutant are becoming more restrictive. Therefore, a big effort is being put into ship efficiency discussion, specially on predictive models related to route optimization, fuel consumption and air emissions. This paper compares machine learning predictive algorithms, based on the following techniques: least-squares, decision trees and neural networks, to estimate ship propulsion power between two 8400 TEU container ships from the same series. Additionally, the influence of having a predictive algorithm trained with data of its sister ships is invesitgated. The data used in this study were recorded from 2009 to 2014 reaching almost 290,000 entries. The results indicate that random forest regression model and decision trees ensemble models have the best fit for this purpose. It has also confirmed the feasibility of predicting the delivered power of a ship having a machine learning algorithm feed with a sister ship information despite differences in the route and/or operating conditions.

Automated summary

This study compares machine learning predictive algorithms for estimating ship propulsion power and finds that random forest regression model and decision trees ensemble models perform the best. It confirms the feasibility of predicting a ship's power using a machine learning algorithm trained with data from sister ships, despite differences in routes or operating conditions.