Deep reinforcement learning from demonstrations for mission-aware multi-objective energy storage management of solar-powered UAVs

Solar-powered unmanned aerial vehicles perform specified flight missions under diverse operating conditions. Mission requirements and varying flight conditions call for a mission-aware energy storage management strategy that improves energy efficiency and battery health while maintaining mission performance. To this end, this paper proposes a mission-aware multi-objective energy storage management framework based on an enhanced deep reinforcement learning approach. An integrated model is developed to characterize the interactions among energy dynamics, flight mechanics, and time-varying environmental inputs, thereby enabling the representation of coupled energy flows and mission-related operational constraints throughout flight. The energy storage management problem is formulated as a multi-objective Markov decision process. Charging efficiency, battery health, and altitude tracking are jointly optimized through a weighted reward design. Building upon the twin delayed deep deterministic policy gradient algorithm, an exponentially dilated temporal-difference scheme is incorporated into a reinforcement learning from demonstrations framework to improve learning efficiency and policy stability. Simulation results show that the proposed framework achieves strong performance across the considered mission scenarios and outperforms representative baseline methods. Hardware-in-the-loop experiments further validate the effectiveness and implementation feasibility of the proposed framework for practical energy management of solar-powered unmanned aerial vehicles.