Control of High-altitude Power Grid Inspection UAV Using Hierarchical PID Based on Reinforcement Learning

Abstract

Unmanned aerial vehicles (UAVs) are now being utilized in a wide range of work scenarios to replace humans in performing dangerous tasks and jobs. In the context of high-altitude power grid inspections, UAV inspections have significant advantages. However, the complex environment often poses challenges for controlling UAVs. Building upon traditional PID control, this paper uses MATLAB software to construct the dynamic model and controller of a quadcopter UAV. A cascaded loop is established by selecting the control objects for the inner and outer loops. The stabilization of the UAV is achieved by setting the PID parameters for the inner and outer loops. Additionally, the deep reinforcement learning algorithm is employed to fine-tune the PID parameters based on the Zigler-Nicholes method. The adjustment is performed by considering the reward function values for the UAV under different disturbances. Finally, response curves and dynamic response parameter indicators are obtained. Compared to traditional cascaded PID control, the quadcopter UAV system tuned using this method exhibits significant improvements in both dynamic response parameters and steady-state responses.