Machine learning models for predicting ship main engine Fuel Oil Consumption: a comparative study

Research output: Contribution to journalArticle

Abstract

As Fuel Oil Consumption (FOC) constitutes over 25% of a vessel’s overall operating cost, its accurate forecasting, and the reliable prediction of the relevant ship operating expenditures can majorly impact the ship operation sustainability and profitability. This study presents a comparison of data-driven, multiple regression algorithms for predicting ship main engine FOC considering two different shipboard data acquisition strategies, noon-reports and Automated Data Logging & Monitoring (ADLM) systems. For this, various multiple regression algorithms including Support Vector Machines (SVMs), Random Forest Regressors (RFRs), Extra Trees Regressors (ETRs), Artificial Neural Networks (ANNs), and ensemble methods are employed. The effectiveness of the tested algorithms is investigated based on a number of key performance indicators, such as the mean and median average error and the coefficient of determination (R2). ETR and RFR models were found to perform best in both cases, whilst the existence of an ADLM system increased accuracy by 7% and reduced the required period for data collection by up to 90%. The derived models can accurately predict the FOC of vessels sailing under different load conditions, weather conditions, speed, sailing distance, and drafts.
LanguageEnglish
Article number106282
Number of pages14
JournalOcean Engineering
Volume188
Early online date26 Aug 2019
DOIs
Publication statusPublished - 15 Sep 2019

Fingerprint

Fuel oils
Learning systems
Ships
Engines
Sailing vessels
Operating costs
Support vector machines
Sustainable development
Data acquisition
Profitability
Neural networks
Monitoring

Keywords

  • FOC prediction
  • ship energy effciency
  • multiple regression
  • support vector machines
  • neural networks
  • ensemble methods
  • machine learning

Cite this

@article{ce0a69cd9220463eb3e2ae8cd3586301,
title = "Machine learning models for predicting ship main engine Fuel Oil Consumption: a comparative study",
abstract = "As Fuel Oil Consumption (FOC) constitutes over 25{\%} of a vessel’s overall operating cost, its accurate forecasting, and the reliable prediction of the relevant ship operating expenditures can majorly impact the ship operation sustainability and profitability. This study presents a comparison of data-driven, multiple regression algorithms for predicting ship main engine FOC considering two different shipboard data acquisition strategies, noon-reports and Automated Data Logging & Monitoring (ADLM) systems. For this, various multiple regression algorithms including Support Vector Machines (SVMs), Random Forest Regressors (RFRs), Extra Trees Regressors (ETRs), Artificial Neural Networks (ANNs), and ensemble methods are employed. The effectiveness of the tested algorithms is investigated based on a number of key performance indicators, such as the mean and median average error and the coefficient of determination (R2). ETR and RFR models were found to perform best in both cases, whilst the existence of an ADLM system increased accuracy by 7{\%} and reduced the required period for data collection by up to 90{\%}. The derived models can accurately predict the FOC of vessels sailing under different load conditions, weather conditions, speed, sailing distance, and drafts.",
keywords = "FOC prediction, ship energy effciency, multiple regression, support vector machines, neural networks, ensemble methods, machine learning",
author = "Christos Gkerekos and Iraklis Lazakis and Gerasimos Theotokatos",
year = "2019",
month = "9",
day = "15",
doi = "10.1016/j.oceaneng.2019.106282",
language = "English",
volume = "188",
journal = "Ocean Engineering",
issn = "0029-8018",
publisher = "Elsevier",

}

TY - JOUR

T1 - Machine learning models for predicting ship main engine Fuel Oil Consumption

T2 - Ocean Engineering

AU - Gkerekos, Christos

AU - Lazakis, Iraklis

AU - Theotokatos, Gerasimos

PY - 2019/9/15

Y1 - 2019/9/15

N2 - As Fuel Oil Consumption (FOC) constitutes over 25% of a vessel’s overall operating cost, its accurate forecasting, and the reliable prediction of the relevant ship operating expenditures can majorly impact the ship operation sustainability and profitability. This study presents a comparison of data-driven, multiple regression algorithms for predicting ship main engine FOC considering two different shipboard data acquisition strategies, noon-reports and Automated Data Logging & Monitoring (ADLM) systems. For this, various multiple regression algorithms including Support Vector Machines (SVMs), Random Forest Regressors (RFRs), Extra Trees Regressors (ETRs), Artificial Neural Networks (ANNs), and ensemble methods are employed. The effectiveness of the tested algorithms is investigated based on a number of key performance indicators, such as the mean and median average error and the coefficient of determination (R2). ETR and RFR models were found to perform best in both cases, whilst the existence of an ADLM system increased accuracy by 7% and reduced the required period for data collection by up to 90%. The derived models can accurately predict the FOC of vessels sailing under different load conditions, weather conditions, speed, sailing distance, and drafts.

AB - As Fuel Oil Consumption (FOC) constitutes over 25% of a vessel’s overall operating cost, its accurate forecasting, and the reliable prediction of the relevant ship operating expenditures can majorly impact the ship operation sustainability and profitability. This study presents a comparison of data-driven, multiple regression algorithms for predicting ship main engine FOC considering two different shipboard data acquisition strategies, noon-reports and Automated Data Logging & Monitoring (ADLM) systems. For this, various multiple regression algorithms including Support Vector Machines (SVMs), Random Forest Regressors (RFRs), Extra Trees Regressors (ETRs), Artificial Neural Networks (ANNs), and ensemble methods are employed. The effectiveness of the tested algorithms is investigated based on a number of key performance indicators, such as the mean and median average error and the coefficient of determination (R2). ETR and RFR models were found to perform best in both cases, whilst the existence of an ADLM system increased accuracy by 7% and reduced the required period for data collection by up to 90%. The derived models can accurately predict the FOC of vessels sailing under different load conditions, weather conditions, speed, sailing distance, and drafts.

KW - FOC prediction

KW - ship energy effciency

KW - multiple regression

KW - support vector machines

KW - neural networks

KW - ensemble methods

KW - machine learning

U2 - 10.1016/j.oceaneng.2019.106282

DO - 10.1016/j.oceaneng.2019.106282

M3 - Article

VL - 188

JO - Ocean Engineering

JF - Ocean Engineering

SN - 0029-8018

M1 - 106282

ER -