Identification of nonlinear time-varying systems using time-varying dynamic neural networks

In this paper, time-varying neural networks are proposed for modeling and identification of continuous-time time-varying nonlinear systems. The neural network undertaken is of time varying weights, and the iterative learning methodology is applied for network training. It is shown that when the used neural network is perfect in approximation, namely, the approximation error being zero, the identification error converges to zero over the entire time interval as the iteration increases. To deal with the non-zero approximation error, the dead-zone modified iterative learning algorithms are used for updating the time-varying weights, and the identification error is ensured to converge to the bound, which is proportional to the approximation error.