Highlights A data-centric and application-oriented testing regime for fuel consumption prediction (FCP) models. Noon report data comprehensive preprocessing and preparation with data fusion and feature engineering. Two multi-ship FCP models were trained and optimized using XGBoost (XGB) and multi-layer perceptron neural network (ANN). Model behaviors, strengths, and weaknesses by comparing models using the proposed testing regime.

    Abstract Fuel consumption prediction (FCP) of vessels is the core of many decarbonization efforts by the maritime industry. Understanding the capability of machine learning (ML) FCP models is essential in various decision-making processes. However, the current model testing practice does not reflect their uncertainty and resilience in actual applications. To address this gap, this study proposes a testing regime that could provide insights into models’ behaviors, dependency on different features, and potential vulnerabilities to data uncertainties in the deployment phase. Two multi-ship FCP models were developed for testing, using extreme gradient boosting (XGB) and multi-layer perceptron artificial neural network (ANN) algorithms on noon reports of a container fleet operated globally in 2.5 years. Unlike previous studies, which explicitly indicated the superior ML algorithms, results from this study depicted a complicated situation with no decisive dominance of one algorithm over another, suggesting the potential of model combination and cooperation for optimal application performance. Aiding the FCP model development efforts, this study also includes findings regarding (1) the optimal configurations for ANN models, and (2) the reliance of FCP ML models and algorithms on different fuel consumption influencing factors. To our knowledge, this study is among the first to advocate a more comprehensive understanding of AI-based FCP models’ characteristics in realistic scenarios instead of simple selections based on accuracy indicators.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    An application-oriented testing regime and multi-ship predictive modeling for vessel fuel consumption prediction


    Beteiligte:
    Nguyen, Son (Autor:in) / Fu, Xiuju (Autor:in) / Ogawa, Daichi (Autor:in) / Zheng, Qin (Autor:in)


    Erscheinungsdatum :

    2023-08-16




    Medientyp :

    Aufsatz (Zeitschrift)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch




    Machine learning approach to ship fuel consumption: A case of container vessel

    Uyanık, Tayfun / Karatuğ, Çağlar / Arslanoğlu, Yasin | Elsevier | 2020


    Fuel consumption prediction methodology for early stages of naval ship design

    Gheriani, Eran (Eran Y.) | DSpace@MIT | 2012

    Freier Zugriff

    Electric power consumption predictive modeling of an electric propulsion ship considering the marine environment

    Chae-og Lim / Byeong-cheol Park / Jae-chul Lee et al. | DOAJ | 2019

    Freier Zugriff


    Optimizing ship speed to minimize fuel consumption

    Kim, Jae-Gon | Online Contents | 2014