Error analysis for nonlinear system identification using dynamic neural networks

We analyze the error of nonlinear identification via dynamic neural network, with the same state space dimension as the system. We assume the system space state completely measurable and the neural network parameters tuned by a known learning algorithm. This error is formulated, and by means of a Lyapunov-like analysis we determine its stability conditions, as our main original contribution, we establish a theorem that gives a bound for it. The applicability of this result is illustrated by one example.