Reinforcement Learning Based Energy Management System to Maximize Efficiency for the Hybrid Marine Propulsion System

Hybrid maritime power typically employs energy management technologies such as rule-based or reactive control. The general purpose of this project is to investigate energy management and control approaches for hybrid marine boats using Reinforcement Learning (RL). This project aims to explore the potential of using online adaptive approaches to replace traditional energy management tactics. The initial study focused on a straightforward hybrid maritime power system. A Markov Decision Framework Q-Learning – RL algorithm was used to design a decision-making process. Integrated power systems, combining fuel cells and batteries, present a promising solution for marine propulsion systems, offering high energy density and power while reducing CO 2 emissions, aligning with the International Maritime Organization's goal of achieving net zero emissions by 2050. This study focuses on the implementation of fuel cell and battery systems, a dc-to-dc converter, typically a boost converter, to manage power flow between the fuel cell and battery, enhancing peak power output at propeller where induction machine utilized as a primary mover in a hybrid powertrain system. Using modeling and analysis, this research investigates power distribution scenarios based on varying power demands within onboard marine propulsion systems. Reinforcement Learning is employed to assess the State of Charge (SOC) of the battery and fuel cell density, optimizing power flow under different operational states to maximize total rewards and overall system efficiency. This study contributes to the advancement of marine propulsion systems by integrating Reinforcement learning control mechanisms to optimize power distribution.