Modélisation énergétique des navires auto-déchargeurs à l’aide des méthodes d’apprentissage automatique

Abstract

Energy consumption during unloading operations represents a central issue in the operation of self-unloading vessels, particularly due to the significant role of belt conveyors in the handling chain. This equipment, often used continuously for extended periods, is responsible for excessive energy consumption that requires improvement. To address this challenge, research in recent decades has focused primarily on the development of energy-saving models based on physical modeling or the use of intelligent algorithms through simulations performed on test benches. In a context where reducing energy consumption is a strategic issue from both an environmental and operational perspective, this thesis focuses on optimizing the power consumption of belt conveyors. More specifically, the study proposes a dual approach based on Industry 4.0 using machine learning models, such as random forests, to predict energy consumption based on operational variables such as unloading rate, engine speed, cargo type, and port characteristics. These models are trained on real-world data from several vessels, capturing the complexity and variability of consumption profiles observed in real-world situations. In addition, numerical optimization algorithms, including the L-BFGS-B method, are used to identify operating configurations that minimize energy consumption while incorporating practical constraints such as unloading time. The results show that this dual approach can achieve significant energy savings, particularly for certain vessels with sufficient technical flexibility. It allows for more realistic operating configurations without compromising logistical efficiency and provides a robust decisionmaking framework for adapting operational instructions based on cargo type and port context. This work thus demonstrates the importance of artificial intelligence applied to maritime logistics to enhance operational sustainability while optimizing equipment performance.

Date	29 Jul 2025
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Lokman Sboui (Supervisor) & Amin Chaabane (Co-supervisor)