Adaptive PID Control of UAV Variable-Pitch Propellers Based on Lightweight Neural Networks

This paper proposes a novel lightweight neural network-PID composite control method to address the challenges of stability in millimeter-wave (mmWave) and terahertz (THz) wireless communication for UAVs, which are severely affected by nonlinear dynamics and external disturbances in variable-pitch propeller control. On one hand, traditional PID controllers are widely used in variable-pitch systems, but their adaptability to varying flight conditions and external disturbances is limited, leading to suboptimal performance in dynamic environments. On the other hand, existing machine learning approaches, while offering better adaptability, often suffer from high computational complexity, which makes them unsuitable for real-time applications, particularly for UAVs with limited onboard resources. The novelty of the proposed framework lies in the integration of a lightweight backpropagation (BP) neural network with a PID controller and sliding mode control (SMC). This architecture leverages the self-learning ability of the neural network to dynamically adjust the PID parameters in real-time, thus improving adaptability and robustness without incurring high computational costs. To further enhance efficiency, we introduce a separation strategy for offline training and online inference, significantly reducing the computational load during operation. Additionally, an event-triggered mechanism is employed to optimize the frequency of neural network updates, minimizing unnecessary calculations. By using this proposed system, we achieve better dynamic performance, reduced overshoot, and faster settling times compared to conventional PID controllers. In particular, our method demonstrates superior performance in mitigating speed fluctuations under sudden gust disturbances and adapting to diverse flight conditions. In contrast to traditional methods, which rely on fixed control parameters, the adaptive nature of our approach ensures continuous optimization during flight, guaranteeing improved thrust response precision and communication stability. Experimental validation through simulations and UAV flight tests confirms that the proposed method outperforms traditional PID control in terms of adaptability, robustness, and efficiency.