engleski

A Knowledge-Base Generating Fuzzy-Neural Controller

This paper presents a modification to the Kandadai and Tien's learning algorithm for tuning a fuzzy-neural controller that is able to automatically generate a knowledge base. Tuning is based on reinforcements from a dynamical system, thus giving a pseudosupervised learning scheme using error backpropagation. Originally, a weak reinforcement in the form of a binary failure signal was assumed which proved to be insufficient in terms of steady-state error. Therefore, a continuous reinforcement signal is applied enabling the system to correct the error as well as decreasing the overall control effort in the learning phase.

expert systems; intelligent control; fuzzy logic; neural networks

nije evidentirano