Modeling of nonlinear continuous time dynamic system with fuzzy neural networks

A Runge-Kutta fuzzy neural network (RKFNN) is presented, which is appropriate for modeling a general nonlinear continuous time dynamic system. It is proved that RKFNN can approximate any nonlinear dynamics with any given precision. RKFNN that learns the changing rates of system states is constructed with the traditional Runge-Kutta integral equations, which solves the low precision problems of DMFNN learning systems states. For RKFNN learning, an online recursive algorithm is presented. It is shown experimentally that RKFNN is a better structure for modeling a nonlinear continuous time system.