A novel optimization procedure for training of fuzzy inference systems by combining variable structure systems technique and Levenberg-Marquardt algorithm

This paper presents a novel training algorithm for fuzzy inference systems. The algorithm combines the Levenberg-Marquardt algorithm with variable str ucture systems approach. The combination is performed by expressing the par ameter update rule in continuous time and application of sliding mode contr ol method to the gradient-based training procedure. The proposed combinatio n therefore exhibits a degree of robustness to the unmodeled multivariable internal dynamics of Levenberg-Marquardt technique. With conventional train ing procedures, the excitation of this dynamics during a training cycle can lead to instability, which may be difficult to alleviate due to the multid imensionality of the solution space and the ambiguities concerning the envi ronmental conditions. This paper proves that a fuzzy inference mechanism ca n be trained such that the adjustable parameter values are forced to settle down (parameter stabilization) while minimizing an appropriate cost functi on (cost optimization). In the application example, control of a two degree s of freedom direct drive SCARA robotic manipulator is considered. As the c ontroller, a standard fuzzy system architecture is used and the parameter t uning is performed by the proposed algorithm. (C) 2001 Elsevier Science B.V . All rights reserved.