Construction of Asymmetric Type-2 Fuzzy Membership Functions and Application in Time Series Prediction

This paper proposes a method to construct asymmetric fuzzy membership functions (MFs) for improving performance of type-2 fuzzy logic systems. The effect of asymmetric type-2 fuzzy MFs for fuzzy logic systems is discussed by illustration examples. Each asymmetric MF is constructed by four Gaussian functions to introduce the properties of uncertain mean and uncertain variance. Based on the gradient method, the corresponding learning algorithm is derived. This modification improves the approximation accuracy and reduces the computational complexity. Simulation results of nonlinear systems identification and chaotic time-series prediction are shown to demonstrate the effectiveness.