Energy-Aware Hybrid Neural Network and Finite-Time Sliding Mode Control for Efficient UAV Trajectory Tracking
Аннотация
This paper presents a novel energy-aware hybrid control strategy for unmanned aerial vehicles (UAVs), combining the strengths of Finite-Time Sliding-Mode Control (FTSMC) with a Radial Basis Function (RBF) Neural Network. The main objective is to achieve robust and precise trajectory tracking while simultaneously minimizing the energy consumed by the UAV’s propulsion system. The proposed Energy-Aware Hybrid NN-FTSMC (EA-NN-FTSMC) effectively addresses the challenges of uncertainties and disturbances by using the adaptive capabilities of the neural network to approximate unknown dynamics. This adaptive compensation allows for the utilize of smaller robust control gains, leading to a significant reduction in control effort and consequently, lower energy consumption and chattering. The proposed controller is based on a nonsingular fast-terminal sliding surface that guarantees finite-time convergence of the tracking error, together with an explicit Radial Basis Function weight-update law derived from a Lyapunov analysis. It is evaluated against five controllers of increasing complexity: a PID regulator, a classical SMC, a finite-time SMC (FTSMC), an energy-aware FTSMC (EA-FTSMC), and a hybrid NN-FTSMC under a common smooth reference and identical actuator-saturation limits. Relative to the FTSMC baseline, the proposed EA-NN-FTSMC reduces the altitude tracking root mean square error by more than $$80\%$$ (from 0.144 m to 0.026 m) and the roll error by about $$90\%$$, while the energy-aware term roughly halves the control input chattering (a $$52\%$$ reduction in total variation). All control efforts remain within physically realistic actuator limits ($$|u_z|\le 60$$ N for a 2 kg vehicle), and the total propulsion energy is comparable across controllers, with the energy-aware term concentrating its savings in the transient and disturbance rejection phases where control activity dominates. These results highlight the practical benefits of integrating adaptive neural networks into energy-aware control designs for UAVs, extending their operational endurance and improving overall system efficiency.