A New Optimality Criterion for Nonhomogeneous Markov Decision Processes
Wallace J. Hopp, James C. Bean, Robert L. Smith
Operations Research · 1987 · Volume 35 · Issue 6
Abstract
We propose a new definition of optimality for nonhomogeneous Markov decision processes called periodic forecast horizon (PFH) optimality. Using measures of discounting and ergodicity, we establish conditions under which PFH-optimal strategies exist and PFH optimality implies the more conventional notions of α-optimality and average optimality. Finally, we use this definition of optimality as the basis for a direct development of forecast horizon results for discounted and undiscounted nonhomogeneous Markov decision processes.
Keywords
118 computations for nonhomogeneous chains, 626 new optimality criterion
Enable JavaScript for citation tools and additional paper interactions.