Generalization ability of Universal Learning Network by optimizing network size and time delay

We have already proposed a modeling method for complex nonlinear systems by using the Universal Learning Network(ULN) with switching mechanism. But in the previous paper, the time delays on the branches between the nodes were not adjusted. In this paper, a new modeling method using the ULN:with switching mechanism is studied, where not only parameters but also time delays in the ULN are adjusted to model the nonlinear systems. Simulation results of nonlinear system identification problems show that better performance is obtained by the proposed method than the conventional method that uses only the parameter optimization. And, how the generalization ability of modeling dynamic systems is influenced by:the network size is studied using the switching mechanism in ULN.