TOWARDS INCREASING THE LEARNING SPEED OF GRADIENT DESCENT METHOD IN FUZZY SYSTEM

It is investigated in this paper that how learning algorithm of fuzzy system can be arranged by gradient descent method and how the learning speed can be increased in this method. First, the optimal range of le arning speed coefficient not to be trapped in local minima and not to provide too slow learning speed is investigated. With the optimal rang e of learning speed coefficient, the optimal value of learning speed c oefficient is suggested. With this value, the learning algorithm shoul d not give learning oscillations and not provide too slow learning spe ed in any system to be approximated. Modified momentum is developed an d applied to the learning scheme of gradient descent method in order t o increase the learning speed. Simulation results assure that this mod ified momentum provides fast learning speed and also can converge to t he optimal point within stable learning process without selecting the momentum coefficient arbitrarily.