A mixed fuzzy recursive least-squares estimation for online identification of Takagi-Sugeno models

Without considering the membership feature of each sampling point, both the local fuzzy recursive least-squares (FRLS) and the global FRLS algorithm cannot get an ideal online estimation precision of a Takagi-Sugeno (TS) fuzzy model. This paper proposes a novel mixed FRLS (MFRLS) algorithm for solving the problem. It dynamically makes a multiobjective cost function by weighting the local estimation and global estimation on the membership feature of the sampling point at each updating instant. Then the mixed co-variance matrix of the local and global estimation is deduced by solving the multiobjective optimization problem. Based on the mixed co-variance matrix, a set of MFRLS formula is obtained by further analytical deduction. The simulation experiments on a time-varying nonlinear model have proved the advantages of MFRLS.