scientific article; zbMATH DE number 3970555
A class of non-stationary Markov processes for decision-making (MPDM) is investigated, which has: additional linear constraints connecting the probabilities for going into different states at various moments of time when making certain decisions; the eigenvalue of MPDM, introduced with the help of a non-negative rational flow function v. The decisions optimal choice is reduced to a linear programming problem and it is realized by randomized policies and strategies. It is proved that each MPDM with linear constraints and a flow function v may be represented as an MPDM with a normalized flow function \(v=1\), analog linear constraints and coordinated incomes and coefficients in the left sides of these constraints. The conditions are defined, that allow the use of all the results obtained for MPDM with linear constraints in the case of analog MPDM with a flow function. It is proved that an arbitrary MPDM with linear constraints and a flow- function may be simply represented as a flow with linear capacity constraints on a dynamic graph. The necessary and sufficient conditions are given for the existence of an MPDM with linear constraints - with and without a flow function. It is proved that MPDM with linear constraints and a flow function is a more general mathematical structure than MPDM with the same constraints but without a flow function, the latter class of processes being a partial case of the first.
- scientific article; zbMATH DE number 3974768
- The expected total cost criterion for Markov decision processes under constraints
- Control of a Markov process in a problem with constraints
- Constrained Markovian decision processes: The dynamic programming approach
- Sensitivity of constrained Markov decision processes
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 Q3737255)