Performance enhancement for neural fuzzy systems using asymmetric membership functions

This paper proposes a method to enhance the performance of interval-valued neural fuzzy systems using asymmetric membership functions (called IVNFS-As). Each asymmetric interval-valued membership function is constructed from parts of four Gaussian functions. The proposed IVNFS-As can capture the essence of nonlinearities in dynamic systems. In addition, the Lyapunov theorem is used to demonstrate the convergence of IVNFS-As, and the corresponding learning algorithm is derived using the gradient method. The asymmetric interval-valued membership functions improve the approximation accuracy of simulation results and reduce the computational complexity. The effectiveness of our approach is demonstrated by results obtained for nonlinear system identification, adaptive control, and chaotic-time-series prediction.