A new time-evolving neural network architecture and algorithm for nonlinear system identification using adaptive filtering techniques

Concepts from adaptive filtering and some heuristics are used to obtain a fast convergent online neural network especially suited for nonlinear system identification. Rather than training a fixed neural network structure, the algorithm presented allocates nodes when required. This provides for an optimal allocation of hidden nodes in this structure. The results obtained show that the neural network model presented is a viable approach for nonlinear system identification and can be applied to a large class of nonlinear systems. Simulations are provided that show the fast convergence of this neural network structure