Predictive Energy Management Using Random Forests for Fuel Cell-Powered UAVs

Unmanned Aerial Vehicles (UAVs) have gained significant growth and demand across various applications in recent years. Battery-powered UAVs, however, face challenges due to limited endurance. An innovative alternative solution is a fuel cell hybrid power system, but it requires an efficient energy management system (EMS) to coordinate power distribution to meet load among hybrid sources. The drone application's capabilities are restricted by its limited computational resources; thus, the development of lightweight computation EMS methods is required. This paper proposes a computationally efficient random forest (RF)-based EMS that is trained on data extracted offline from another optimization-based EMS in order to optimize the utilization of hybrid sources and to coordinate the power distribution at high computational efficiency. Simulation demonstrated the strategy's efficacy; it reduces the battery state of charge (SOC) deviation, which extends the battery durability, and it also maintains a stable DC bus voltage stability close to its reference value. The strategy successfully reduces hydrogen consumption, showing economical and technical benefits at low computational demand.