Unmanned aerial vehicle strong-robustness attitude control method based on deep reinforcement learning

Abstract

The invention discloses an unmanned aerial vehicle strong-robustness attitude control method based on deep reinforcement learning, and proposes that a Bayesian probability model is used to better simulate interference and uncertainty in a real flight environment, and a fitted aircraft dynamics model is used as the input of a reinforcement learning framework based on a DDPG algorithm. And through random extraction of the aircraft digital model and interaction of various flight data acquired by the aircraft real flight data, the Shenzhou network parameters are updated. The output is an aircraft rudder mechanism which comprises a rudder, an elevator and an aileron. According to the invention, the Bayesian neural network can improve the accuracy of the aircraft model, so that the aircraft model is closer to a real flight environment; the control system based on the neural network can improve the control effect of the aircraft in various interference environments by utilizing generalization ability; moreover, the controller after offline training can be rapidly transplanted to various airborne platforms, and is very high in practical value.

Description

technical field [0001] The invention belongs to the technical field of attitude control of unmanned aerial vehicles, and relates to a strong and robust attitude control method for unmanned aerial vehicles based on deep reinforcement learning. Background technique [0002] In recent years, the control technology of fixed-wing UAVs has become mature. Traditional UAV attitude control systems, such as PID / sliding mode control and its optimization variables, have shown excellent performance in many situations that are only in a steady state. For example: CN113485437A uses neural network to adjust PID parameters to adapt to different flight environments, but when the drone is in a dynamically changing environment, the controller will vibrate or even diverge; CN111857171B uses the state equation to construct a neural network to solve the optimal solution, but in some In a nonlinear complex environment, the control effect is not good for objects with strong inertia; CN113359440A us...

AI Technical Summary

Problems solved by technology

For example: CN113485437A uses neural network to adjust PID parameters to adapt to different flight environments, but when the drone is in a dynamically changing environment, the controller will vibrate or even diverge; CN111857171B uses the state equation to construct a neural network to solve the optimal solution, but in some In a nonlinear complex environment, the control effect is not good for objects with strong inertia; CN113359440A uses implicit dynamics to convert the control problem of unmanned aerial vehicles into the control input parameters for solving time-varying second-order systems, but the method is theoretically complex and the amount of calculation is large , and when the environment has strong time-varying characteristics, the control effect may produce serious hysteresis oscillations

Therefore, most traditional control algorithms design controllers based on digital six-degree-of-freedom models, but due to environmental errors between the digital model and the real environment, the portability and control effect of traditional algorithms are greatly reduced

Images