Design of BP neural network controller for ball-beam system

The BP NN (back propagation neural network) stable controller for ball-beam system is designed in this paper. The controller is constructed by a three-layer forward neural network, and Fletcher-Reeves conjugate gradient algorithm is used to adjust net weights. The NN training samples for controller is obtained from root locus controller. After off-line trained, controller is put into on-line control for ball-beam system. Through simulation off line and used in real system, the result shows that this NN controller responses fast, and overshoot is smaller and feasible in the control of ball-beam system. It is more effective than the root locus controller.