Title: An Intelligent Energy Management Framework for Hybrid-Electric Propulsion Systems Using Deep Reinforcement Learning

Abstract: Hybrid-electric propulsion systems powered by clean energy derived from renewable sources offer a promising approach to decarbonise the world's transportation systems. Effective energy management systems are critical for such systems to achieve optimised operational performance. However, developing an intelligent energy management system for applications such as ships operating in a highly stochastic environment and requiring concurrent control over multiple power sources presents challenges. This article proposes an intelligent energy management framework for hybrid-electric propulsion systems using deep reinforcement learning. In the proposed framework, a Twin-Delayed Deep Deterministic Policy Gradient agent is trained using an extensive volume of historical load profiles to generate a generic energy management strategy. The strategy, i.e. the core of the energy management system, can concurrently control multiple power sources in continuous state and action spaces. The proposed framework is applied to a coastal ferry model with multiple fuel cell clusters and a battery, achieving near-optimal cost performance when applied to novel future voyages.