Reformulation of parameter identification with unknown-but-bounded errors (Q1103338)

The usual parameter identification with unknown-but-bounded errors problem consists in finding the parameter uncertainty intervals \(PUI_ j=[\theta\) \(m_ j,\theta\) \(M_ j]\) knowing the measurement vectors \(y_ 1,y_ 2,..\). \(y_ M\) recording the output y at the sampling times \(t_ 1,t_ 2,...,t_ M\), the matrix \(\Phi\) and the bounds \(e\) \(M_ k\geq | e_ k|\) of the ``equation errors'' \(e=(e_ 1,e_ 2,...,e_ M)\) in the system represented by: \(y=\Phi \theta +e\). In the case the matrix \(\Phi\) contains sampled records of the output and the model is described by the input-output representation: \(Ax_ k=Bu_ k\), \(y_ k=x_ k+\eta_ k\) where \(x_ k\) is the hypothetical noise-free output, \(y_ k\) is the measurement vector corrupted with measurement noise \(\eta_ k\) for which the bounds \(\eta\) \(M_ k\geq | \eta_ k|\) are known and \(A(z^{-1})=1+a_ 1z^{-1}+...+a_ nz^{-n},\) \(B(z^{-1})=b_ 0+b_ 1z^{-1}+...+b_ mz^{-m},\) the equation error is shown to be given by: \(e_ k=\eta_ k+a_ 1\eta_{k- 1}+...+a_ n\eta_{k-n}\) and the problem consists in finding the parameter uncertainty intervals depending on the bounds \(\eta_ k\) M determined by simple experimental considerations. Simulations on two examples are given and commented.