Adaptive dynamic TSKCMAC neural networks for prediction and identification

In this paper, a dynamic Takagi-Sugeno-Kang type cerebellar model articulation controller (TSKCMAC) neural network is developed for solving the prediction and identification problem. A dynamic TSKCMAC is combines both the merits of TSK fuzzy model and conventional cerebellar model articulation controller (CMAC). The proposed dynamic TSKCMAC has superior capability to the conventional CMAC in efficient learning mechanism, guaranteed system stability and dynamic response. The recurrent unit is embedded in the TSKCMAC by adding feedback connections in the membership functions space so that the TSKCMAC captures the dynamic response, where the feedback units act as memory elements. The dynamic gradient descent method is adopted to adjust TSKCMAC parameters on-line. Moreover, the analytical method based on a Lyapunov function is proposed to determine the learning-rates of dynamic TSKCMAC so that the variable optimal learning-rates are derived to achieve most rapid convergence of identifying error. Finally, the dynamic TSKCMAC is applied in two computer simulations. Simulation results show that accurate identifying response and superior dynamic performance can be obtained because of the powerful on-line learning capability of the proposed dynamic TSKCMAC.