Energy Management for Hydrogen-Lithium Powered UAV Based on Deep Reinforcement Learning

Hydrogen-lithium powered unmanned aerial vehicles have emerged as a promising direction in aviation due to their high energy density and zero emissions. However, current energy management systems for hydrogen-lithium hybrid UAVs still face numerous challenges, including how to precisely and efficiently distribute power during different flight phases to improve overall energy efficiency and extend flight duration. This study proposes a deep reinforcement learning based energy management method for addressing these challenges. At first, a detailed motion model that includes flight mechanics, air density variations, mass changes, and propulsion load characteristics is established to analyze power requirements during each flight phase. Based on this model, a dynamic power distribution framework using Deep Q-Network (DQN) integrated with Equivalent Consumption Minimization Strategy (ECMS) is developed to achieve adaptive power allocation between the hydrogen fuel cell and lithium battery. MATLAB simulation results show that compared to traditional rule-based control strategies, the proposed algorithm effectively handles random disturbances, achieves real-time power adjustment, and maintains stable output from both the hydrogen fuel cell and lithium battery. The findings indicate that this method reduces total energy consumption by $12.7 \%$ and extends lithium battery lifespan by $18.3 \%$. This research improves to enhance the system’s flight duration and adaptability.