Analysis of Markov Chain Monte Carlo Methods in Multi-Indenture Inventory Optimization
Abstract
U.S. Navy aircraft are required to meet minimum operational availability targets, while minimizing spare parts procurement costs. The current optimization model written by Salmeron and Buss, uses marginal analysis, as described by Sherbrooke, to determine optimal sparing policies for this highly complex multi-indenture model. The literature lacks alternative optimization methodologies for such a problem, so we propose an alternative approach utilizing simulated annealing (SA), a Markov Chain Monte Carlo algorithm. We present three SA approaches tested in three case studies of varying size and complexity. Our initial findings show that in very simple problems, SA is easily capable of outperforming marginal analysis; however, problems with more complexity have large optimality gaps. This is likely because the SA Markov chain is unable to effectively explore the multi-indenture structure of the problem. We implement a method to account for this structure that intelligently builds initial feasible solutions using an epsilon-greedy approach to marginal analysis. This approach produces better results than NAVARM in more than half of the trials on problems of moderate complexity. We also implement a novel method for calculating operational availability that may allow full scale problems to be optimized more efficiently.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2022
- Accession Number
- AD1200370
Entities
People
- Adam M. Alleman
Organizations
- Naval Postgraduate School