Robust Energy-Saving Control Strategies for Agricultural UAV Electromechanical Systems Based on Modular Hybrid Optimization Algorithms

This study extensively explores the energy consumption increase problem of agricultural unmanned aerial vehicles (UAVs) under low-altitude, intense disturbance, and high load fluctuation situations. Based on a modular hybrid optimization method, the study develops an energy-efficient control technique for electromechanical propulsion systems.First, a dynamics and energy consumption model of the propulsion system is established, consisting of a brushless DC motor, an electromagnetic-aerodynamic coupled load, and an inverter control module. Simultaneously, a full-link energy performance index incorporating copper losses, switching losses, and additional aerodynamic losses is constructed. Then, a fractionalorder sliding mode controller (FOSMC) is designed, introducing the Caputo fractional-order operator to characterize the historical energy accumulation effect. A disturbance sensing observer (DSO) is also constructed. The DSO utilizes a nonlinear gain and adaptive filtering structure to achieve real-time estimation of external disturbances. To achieve synergistic optimization of multi-source energy consumption suppression and control performance, a hierarchical multi-objective hybrid optimizer (HM-MOA) is proposed. HM-MOA consists of a threelayer parameter search structure composed of a genetic algorithm, differential evolution, and a particle swarm optimization algorithm. This model can optimize FOSMC parameters, DSO sensitivity, and PWM modulation strategy. The experiment applied a BLDC 1000 W propulsion system as the benchmark platform. Results show that compared to the unoptimized traditional sliding mode controller, the proposed method reduces overall energy consumption by 21.4%, copper loss by 17.8%, motor switching loss by 23.6%, and aerodynamic additional loss by 15.2%. In terms of tracking performance, the response time is shortened by 31.7%, overshoot is reduced by 28.4%, and steady-state error is reduced by 35.1%. Under strong disturbance scenarios, DSO keeps the external disturbance estimation error within$\pm 3.5 \%$, a 46% improvement compared to the model without an integrated disturbance observer.