scientific article; zbMATH DE number 3978968

From MaRDI portal
Publication:3744082





The following linear control system is discussed: \(x_ 0=x\), \(x_{n+1}=A_ nx_ n+B_ nu_ n+C_ nv_ n\), \(n=0,1,...,N-1\), where \(x_ n\), \(u_ n\) and \(v_ n\) are the state variable, control variable, and disturbance of the system, respectively. \(A_ n\), \(B_ n\) and \(C_ n\neq 0\) are matrices; N and \(v_ n\), \(n=1,2,..\). are independent random variables with known distributions. It is assumed that \(N\leq M\) for a constant M, and distributions of \(v_ n\) belong to an exponential family for which the variance is a quadratic function of the mean. The family is indexed by a parameter \(\lambda\in \Delta \subset R\). The M-tuplet \(U=(u_ 0,u_ 1,...,u_ M)\) with \(u_ i\) a Borel vector- function of \((x_ 0,x_ 1,...,x_ i\), \(u_ 0\), \(u_ 1,...,u_{i- 1})\), the data available at time i, is called a control policy. The risk function corresponding to the policy U is of the form: \[ \quad R(\lambda,U)=E_ p\{E_{\lambda}\{\sum^{N}_{i=0}(x_ i',\lambda)S_ i(x_ i',\lambda)'+u_ i'K_ iu_ i\}\} \] where \(E_ p\) and \(E_{\lambda}\) denote the expectation with respect to the distribution of random variables N and \(v_ n\), respectively. \(S_ i\), \(K_ i\) denote symmetric, nonnegative matrices. The aim of the paper is to determine the policy \(\tilde U\in {\mathcal U}\) with \({\mathcal U}=\{U:R\lambda,U)<\infty,\lambda \in \Delta \}\) such that \[ (i)\quad E_{\pi}(R(\lambda,\tilde U))=\inf_{U}E_{\pi}(R(\lambda,U)), \] where \(\pi\) is an a priori distribution of the parameter \(\lambda\), and to determine the policy \(U_{\Gamma}\in U'\) such that \[ (ii)\quad \sup_{\pi \in \Gamma}E_{\pi}(R(\lambda,U_{\Gamma}))=\inf_{U\in U'}\sup_{\pi \in \Gamma}E_{\pi}(R(\lambda,U)) \] where \(\Gamma\) is a known set of a priori distributions of the parameter \(\lambda\) and U' the set of policies for which \(E_{\pi}(R(\lambda,U))\) exists for all \(\pi\in \Gamma\). The policies of type (i) and (ii) are determined under additional assumptions on the distribution of the parameter \(\lambda\).




Cited in
(21)








This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3744082)