An adaptive energy management and control framework for long-endurance UAV missions in dynamic environments: A reproducible simulation study

Long-endurance uncrewed aerial vehicle (UAV) missions are limited by onboard energy, and hybrid battery/fuel-cell systems increasingly extend flight time. Prior work largely treats power-source allocation and flight-path planning as separate problems, which can strand one energy source while the other is still available. This paper proposes an Adaptive Energy Management and Control Framework (AEMCF): a nonlinear, wind-compensating guidance law that sets heading and airspeed, coupled to a receding-horizon quadratic program (QP), solved every control step, that allocates power between a battery and a fuel cell using an equivalent-consumption principle so that neither energy source is exhausted while the other still has capacity. We implement the full model in an open, reproducible simulation (UAV point-mass dynamics, a stochastic wind-gust process, a battery state-of-charge model, a load-dependent fuel-cell efficiency curve) and evaluate it against two baselines: a fixed 50/50 power split, and a static equivalent-consumption split computed once before flight. Across 50 Monte Carlo wind realisations, AEMCF achieves a mean endurance of 124.6 minutes, versus 108.0 for the fixed-split baseline and 123.4 for the static offline split; AEMCF and the static split both consume close to 100% of the true onboard energy budget, whereas the fixed split strands roughly 16% unused. This resource-balancing advantage is robust, holding under harsher wind (30 seeds) and realistic sensor bias and noise (20 seeds). The online controller’s further, smaller advantage over the static split is less robust: it does not grow under harsher wind, and does not survive the sensor imperfections tested, indicating AEMCF’s principal contribution is the power-balancing objective itself rather than wind-aware guidance specifically. We report this honestly, including where it complicates the original motivation for the wind-compensating layer, and identify state estimation as a concrete next step for recovering it. All simulation code and results are openly available and run end-to-end without modification.