Neural Network-Based Model Predictive Control with Active Disturbance Rejection for Quadrotor Systems

Rejecting external disturbances and addressing model uncertainties are critical objectives in control system design. In this paper, we propose a neural network-based model predictive control with active disturbance rejection (NNADR-MPC) strategy for quadrotor systems. This approach utilizes neural networks to learn model uncertainties and designs a neural network-based disturbance observer (DO) to achieve active disturbance rejection control. In combination with model predictive control, the system can rapidly adapt to and handle disturbances caused by partially known suspended loads on a quadrotor during trajectory tracking tasks in the presence of model uncertainty. The designed observer has bounded input bounded output (BIBO) stability guarantees. A comparative study is conducted with existing anti-interference methods for quadrotors under suspended load disturbances. The results show that the proposed control strategy can effectively estimate the disturbance and improve the tracking effect of the quadrotor on the reference trajectory.