Fuzzy rule base derivation using neural network-based fuzzy logic controller by self-learning

This paper presents a new rule base derivation method using neural networks for fuzzy logic control of dynamic systems. The proposed method needs neither dynamic models of the system nor control experts for the control problem. Multi-layered perceptron neural networks are used to form the neural network based quasi-fuzzy logic controller (QFLC). The control performance of the QFLC is achieved on-line using the feedback error learning scheme. The fuzzy control rules are then extracted from the input-output characteristics of the QFLC. They are reduced to form the fuzzy control rule base for an FLC through consecutive procedures such as smoothing, logical reduction, and test running and selection of firing rules. To verify the validity of the proposed rule base derivation method, the authors apply the method to fuzzy logic control of an inverted pendulum