Abstract-A fuzzy clustering approach is developed to select pole locations for Orthonormal Basis Functions (OBFs), used for identification of Linear Parameter Varying (LPV) systems. The identification approach is based on interpolation of locally identified Linear Time Invariant (LTI) models, using globally fixed OBFs. Selection of the optimal OBF structure, that guarantees the least worst-case local modelling error in an asymptotic sense, is accomplished through the fusion of the Kolmogorov n-width (KnW) theory and Fuzzy c-Means (FcM) clustering of observed sample system poles.
I. INTRODUCTIONIn general, many physical systems and control problems suffer from parameter variations due to non-stationarity, nonlinear behavior, or dependence on independent variables, such as space coordinates. These systems vary in size and complexity from highly advanced aircrafts [1] to induction motors [2], but they share the common need for accurate and efficient control of the relevant process variables, which has to satisfy the rapidly increasing industrial performance demands. However, accurate modelling of such systems is in general a complex and tedious task, involving the use of non-linear partial differential equations, leading to models with many parameters and high computational complexity.For processes with mild non-linearities, the theory of LPV systems offers an attractive framework for modelling and handling non-linear or time-varying dynamics. These systems are generally described in a state-space representation (SSR), where the state-space matrices are usually affine functions of a time-varying parameter vector ζ : Z → Γ. Here Γ denotes the parameter space. Furthermore, control design in the LPV framework can be carried out by using LTI control theory via gain scheduling [3]. Therefore, the LPV approach can offer a useful venue to meet recent industrial demands. However, existing methods for identification of such systems often produce models with high complexity or -for instance with subspace techniques -with substantial computational load. Because most control design methods require low-order models and fast iterations in the identification process, it is a challenge to develop efficient methods for LPV system identification that yield models with limited complexity and computation time. An additional point of concern is that the McMillan degree of the system may change due to variations of ζ, especially when the approach is based on interpolation of local models. One way to overcome these