A Theory for Semi-Markov Decision Processes with Unbounded Costs and Its Application to the Optimal Control of Queueing Systems
Abstract
Semi-Markov decision processes with countable state and action spaces are investigated. The optimality criteria considered are the average cost criterion, the undiscounted cost criterion, and the discounted cost criterion. The common assumption of bounded costs has been replaced by some considerably weaker conditions. In particular, our assumptions are weaker than those made by Harrison, Hordijk, Lippman and Reed when they considered the same problem. The existence of optimal, stationary optimal and stationary E-optimal policies is investigated. Policy improvement is considered. Necessary and sufficient conditions for the optimality of a policy are given. Then the optimal control of queueing systems is considered by formulating this general problem as a semi- Markov decision process. Finally, four different ways of proving the optimality of an unimprovable policy are developed in the context of queueing systems.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1976
- Accession Number
- ADA030649
Entities
People
- Peter Orkenyi
Organizations
- Stanford University