On Comparing Performance of Conventional Fuzzy System with Recurrent Fuzzy System

Abstract

Fuzzy controllers are very commonly used as universal approximators. It is seen that while dealing with plants or systems with known order, fuzzy systems prove to be very effective approximators. The concept of fuzzy systems used in the field of identification and control has been extended by introducing a new architecture known as “recurrent fuzzy systems (RFS)”. This new architecture enhances the approximation capacity of the conventional fuzzy structure and empowers it to deal with dynamic process of unknown order and structure. Similar to its neural counterpart, fuzzy systems can also be recurrent. This paper presents the identification of two nonlinear systems using the concept of recurrent fuzzy system assuming that their structure is unknown using a single hidden variable. Although the number of hidden variable is kept same for the identification of both the examples but two different structures of RFS (using different numbers of delayed output feedbacks) have been used in this paper. These systems are then also identified using traditional fuzzy systems where the order of the actual plant and the structure of its transfer function are taken into account. A comparative analysis is done between both the identification schemes. A suitable learning algorithm based on backpropagation has also been given to update the parameters associated with the model. The paper highlights that using RFS for approximation of higher order systems reduces the number of rules and thus the complexity to a great extent. Also it shows that even if the exact structure of the plant is unknown, RFS proves to be of great use.