Model identification using fuzzy clustering and applications

During the past decade, the need for development of intelligent control systems has increased tremendously. The more intelligence a system can exhibit, the more sophisticated and complex the system will become. In addition to complexity, other nonlinearity and uncertainty are also introduced to the systems. In order to predict the behavior of these systems prior to their development, approximate models are beneficial. Even though, conventional model identification theories have greatly contributed in obtaining the mathematical model, identifying the model for these systems which have precise mathematical formulations are not a simple task. Currently, Fuzzy Sets and Fuzzy Logic (FL) have shown a promising trend in the area of model identification for complex and highly non-linear systems. These techniques are capable of modeling uncertain situations and can be instrumental in the formulation of interpolative reasoning (1). In this research, a technique is developed for the model identification of an optical servo-tracking system at White Sands Missile Range, New Mexico.; Early techniques of fuzzy modeling, implemented Zadeh's ideas by trying to extract the fuzzy model directly from expert knowledge. Where expert knowledge is not available and only the input and output data of the system is available, clustering techniques appear to be the most appropriate choice in developing a relationship between the input and output of the system.; This research investigates the use of Fuzzy Clustering as a means for model identification of complex systems when only observational data is available. The use of Fuzzy Clustering facilitates automatic generation of rules and its antecedent parameters. The consequent of the model is then formulated in the form of Takagi, Sugeno and Kang (TSK), and its parameters determined by the Least Squares Method (LSM). This algorithm is not only applied to a Kineto Tracking Mount (KTM) at White Sands Missile Range but can also be applied to the identification of other fuzzy system models from observational data. Results are discussed along with suggestions for the fusion of fuzzy logic with other constituents of soft computing, namely, Neural Networks (NNs), and Genetic Algorithms (GAs) for future research.; The contribution of this research can be summarized as follows: (i) A systematic study of fuzzy clustering based on the Mountain Clustering method by Yager and Filev, and its modification called Cluster Estimation method by Chui is presented. (ii) Appropriate software is developed for model identification based on a fuzzy clustering algorithm which can be used for various applications. (iii) The use of fuzzy clustering algorithm to a Kineto Tracking Mount (KTM) at White Sands Missile Range is examined and analyzed. (iv) Results strongly indicate the relationship between the number of clusters (rules) and the cluster centers in a fuzzy model with the number of poles and their locations in a transfer function model. This is a significant observation which has potential for further mathematical development for fuzzy models in relation with the mathematical representation of a transfer function.