2822. Dynamic Resource Allocation for a Multi-Class System with Returns
Invited abstract in session TC-54: Methods & stochastic modelling , stream Stochastic modelling.
Tuesday, 12:30-14:00Room: Liberty 1.08
Authors (first author is the speaker)
| 1. | Fikri Karaesmen
|
| Dept. of Industrial Engineering, Koc University | |
| 2. | Arash GandomkarGhalhar
|
| Dept. of Industrial Engineering, Koç University | |
| 3. | Oktay Karabağ
|
| Department of Industrial Engineering, İzmir University of Economics |
Abstract
Motivated by the case of a company that sterilizes used surgical instruments and medical devices to make them available for reuse in hospitals, we investigate a production-inventory model with multiple products and shared resources where the used products return to the facility with limited capacity for sterilization. We formulate a Markov Decision Process to explore the optimal dynamic resource allocation policy that minimizes the shortage and holding costs over the long run. Due to usage-dependent return rates, the optimal policy is complicated. To cope with this complication, we test a number of plausible and simpler policies. However, even the proposed class, there are many candidates to test and choose. To speed up the process of choosing the best among the class of proposed policies, we generate a training data by numerically solving instances of the Markov Decision Process model and labeling the performance of each proposed policy. Using machine learning, we then generate rules for determining the best proposed policy as a function of the problem parameters. The results are promising and suggest that the procedure can be automated.
Keywords
- Stochastic Models
- Inventory
- Queuing Systems
Status: accepted
Back to the list of papers