Design Of The Learning Control Algorithm And Its Implementation In Inverted Pendulum

The learning control algorithm is derived from the character of human action, in order to solve the uncertainty of unknown variety which the nonlinearity of the controlled object and system modeling imperfect. The algorithm can reduce the lack of a priori knowledge necessary for set up a control system.Inverted pendulum which is a nonlinear unstable equilibrium control object, as a experimental platform can verify a variety of algorithms and to test various properties of the algorithm.The Q-learning algorithm is researched in this thesis. This algorithm is used to achieve the steady pendulum in inverted pendulum, and it is improved to promote the control effect. The LQR-fuzzy algorithm is used to control the two-stage and three-stage inverted pendulum in this paper. Main topic of the work to complete is as follows:1. The Q-learning algorithm is used to control the one-stage inverted pendulum directly. The learning control algorithm is studied for controlling rules of inverted pendulum stable. Though the analysis of experimental results, and pointing out the problems of the algorithm.2. Q-learning algorithm combined with BP network, and is used to optimize the Q-learning algorithm parameters. Using this algorithm one-stage inverted pendulum stable is achieved.3. A two-wheeled robot model is discussed. The two-wheeled robot model is established and is same as the model of one-stage inverted pendulum. So the optimized Q-learning algorithm can be used to the balance control of two-wheeled robot also.4. The LQR-fuzzy control algorithm is designed for the two-stage and three-stage inverted pendulum. And it is verified though experiment.5. Summary and present shortage in paper. Further work is also mentioned.