Fuzzy neural network implementation of self tuning PID control systems

摘要

The fuzzy cognitive map (FCM) is a powerful universal method for representation of knowledge in various domains. The fuzzy inference engine can be implemented in the form of a network of FCMs. FCM implementation of the inference engine provides a suitable mechanism for expert control systems and information engineers to embed acquired human expertise, which is often imprecise, vague, or incomplete. The exploitation of an online learning algorithm empowers the fuzzy inference engine with the ability to modify its incomplete or possibly inconsistent knowledge base resulting in continuous improvement of the embedded knowledge. The fact that learning is an inherent feature of neural networks has inspired several researchers with the idea of using neural networks to implement fuzzy inference engines capable of learning. This paper presents a method for neural network FCM implementation of the fuzzy inference engine using the fuzzy columnar neural network architecture (FCNA). In this method the available human expertise is mapped first into an initial set of weights for the neurons. A new learning algorithm is then used to enhance the embedded knowledge in the neural network as a result of real time experience. The fuzzy inference engine (the neural network FCM) is used in computer simulations to control the speed of an underwater autonomous mobile robot. Results and computer simulation experiments are presented along with an evaluation of the new approach.