On identification of nonlinear regressions models: Application to the pole-zero cancellation problem in adaptive control (Q1286746)

The authors consider the problem of estimation of nonlinearly parametrised regression models of the form \(y=f(\theta^*)^T \phi\), \(y \in \mathbb{R}\), where \(\phi\in \mathbb{R}^n\) is a measurable signal, \(f:\mathbb{R}^n\rightarrow \mathbb{R}^n\) is a known differentiable function of the unknown parameter vector \(\theta^*\). Under certain (non convex) assumptions it is shown that a suitable estimation \(\widehat \theta\) of \(\theta\) leads to the convergence of the prediction error \(y-f(\widehat\theta^*)^T \phi\) to zero.