A robust line search for learning control

From MaRDI portal





The paper deals with learning control methods, where output measurements are available, but the system model is known only approximately. Examples are 1) The actual on-line measured response is used to evaluate the quality of the input and 2) The output tracking task is to be performed repeatedly. In a more formal (mathematical) problem statement one has to find a solution \(x\in \mathbb{R}^n\) of a certain equation \(f(x)= 0\). For any \(x\), the value \(f(x)\) is available, but only an inaccurate estimate of the Jacobian \(J(x)= {\partial f\over\partial x}(x)\). The classical Newton-Raphson algorithm usually does not converge. Under three hypotheses (Avrachenkov formulated the same three hypotheses in 1998) a new variant of the Newton-Raphson method can be proved to be convergent. The new line search's advantages over the known line searches are demonstrated on a numerical example.











This page was built for publication: A robust line search for learning control

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3151579)