Design of Reinforcement Learning Algorathm for Single Inverted Pendulum Swing Control

In order to solve the time of the swinging up of the single inverted pendulum is quite long. This paper adopts an improved reinforcement learning algorithm, which is simulated in a double-layer BP neural network. The off-line BP neural network was trained by simulation data in order to acquire reinforcement learning swinging up controller, which was applied in the GLIP2003 inverted pendulum system. Experimental results show that the swinging up controller of reinforcement learning has a certain speed, and the adjustment time of the entire system is less than 8s. The steady state error of the pendulum arc is 0, and the steady state error of small car's position is 0.05m.