Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Multi-module Markov decision processes

Thesis/Dissertation ·
OSTI ID:5915536
The author formulates and analyzes an important class of Markov decision processes (MDP). This class, called the multi-module MDP, is distinguished by the characteristic that each scalar element of its vector-state process has dynamics that operate independently of the other elements. Physical systems naturally modeled by this class of MDP are fault-tolerant, multi-component maintenance problems having independently deteriorating components and multiproduct inventory-control problems, where each product demand is independent of the demands of the other products. A key feature that the multi-module MDP invariably possesses is a large state space, which quite often precludes computation of an optimal strategy. A suboptimal design procedure is developed based on the associated upper and lower bounds on the optimal expected cost function. A procedure for improving this suboptimal design is given. The results are extended to the partially observed MDP. Numerical examples demonstrate the significance of the results.
Research Organization:
Virginia Univ., Charlottesville (USA)
OSTI ID:
5915536
Country of Publication:
United States
Language:
English