A self-learning fuzzy controller based on reinforcement and its application

This paper proposes a self-learning fuzzy logic control system through reinforcements for solving the considered dynamic systems whose input-output training data are unavailable. The learning system consists of an artificial neural network (ANN) and a predicted neural network (PNN). The task is to balance a pendulum hinged to a movable cart by applying forces to the base of the cart. The ANN can have multiple outputs to perform the different tasks. In this case, all the output nodes of the ANN receive the same reinforcement signal from the PNN. With the PNN, the predicted reinforcement signal, p(k), can provide the ANN beforehand as well as more detailed than external reinforcement signal does through the learning mechanisms carried out by TMS320P14 chip.