An intelligent energy management framework for hybrid-electric propulsion systems using deep reinforcement learning
Peng Wu, Julius Partridge, Enrico Anderlini, Yuanchang Liu, Richard Bucknall
Abstract
Hybrid-electric propulsion systems powered by renewable energy offer a promising solution for decarbonising transportation, but their performance relies on effective energy management systems. This study proposes an intelligent energy management framework using deep reinforcement learning to address the challenges of stochastic operating environments and concurrent control of multiple power sources. A Twin-Delayed Deep Deterministic Policy Gradient (TD3) agent is trained on extensive historical load profiles to develop a generic strategy for continuous state and action spaces. Applied to a coastal ferry with fuel cell clusters and a battery , the framework achieves near-optimal cost performance in simulations of novel voyages, demonstrating its potential to enhance efficiency and sustainability in maritime transportation . Validation results demonstrate that the 4-cluster strategy achieves an average voyage cost which is only 2.7% higher than the TD3 uniform strategy, while emitting 1.8% fewer Global Warming Potential (GWP) emissions.