1388. A computational environment for policy evaluation for the dynamic-stochastic purchasing inventory routing problem in agri-food supply networks
Invited abstract in session HE-15: Multilevel and Stochastic Optimization , cluster Multilevel and Stochastic Optimization Methods.
Thursday, 16:15-17:45Room: FENP208
Authors (first author is the speaker)
1. | Camilo Gomez
|
Industrial Engineering, Universidad de los Andes | |
2. | Juan Betancourt
|
Industrial Engineering, Universidad de los Andes | |
3. | Ariel Rojas
|
Industrial Engineering, Universidad de Los Andes | |
4. | Daniel Cuellar-Usaquen
|
Universidad de Los Andes | |
5. | Sonja Rohmer
|
Department of Logistics and Operations Management, HEC Montreal | |
6. | Marlin Wolf Ulmer
|
Management Science, Otto von Guericke Universität Magdeburg | |
7. | David Álvarez-Martínez
|
Industrial Engineering, Universidad de Los Andes |
Abstract
Reconnecting consumers and producers, local food chains have received growing attention over recent years, offering more transparency and fairer prices to producers. In this context, online platforms play an important role by reducing intermediaries and creating new marketplaces for small suppliers. These newly established marketplaces are often highly dynamic, however, and volatile to changes in demand as well as the availability of products. Making good procurement and inventory decisions (balancing existing trade-offs between costs, service levels and the freshness of products) is, thus, extremely challenging within this uncertain context. This research addresses this problem under consideration of uncertain demand and supply of perishable products, by integrating purchasing, first-mile routing and inventory decisions. To solve this problem, we adopt an approximate dynamic programming framework allowing integration of several modeling and solution approaches into one computational environment. Our contribution is twofold: 1) the environment includes a stochastic instance generator based on Colombian agriculture information systems, providing a baseline for comparisons for the integrated decision problem, and 2) this environment is used to test and compare decision policies based on exact and approximate optimization models, relying on stochastic lookahead approximations. For this purpose, we propose a two-stage decision approach - an all-encompassing model is impractical and separate decision models do not capture the problem’s complexity. We first focus on purchase and inventory decisions, using a stochastic mixed-integer program accounting for the quality loss of perishable products throughout sample paths, capturing variability of unknown information. In the second stage, we then tackle the induced Capacitated Vehicle Routing Problem with exact and heuristic strategies, identifying high-quality feasible routes providing feedback to the first stage. Our instance generator emulates the main agricultural supply network of Bogotá and is used to tune and validate the proposed methodology. The implementation of our virtual environment allows for policy evaluation - providing insights regarding key performance metrics such as profit, costs, product waste and service levels. Finally, we present a sensitivity analysis on how various product quality loss functions affect replenishment and demand compliance decisions.
Keywords
- Supply Chain Management
- Stochastic Optimization
- E-Commerce
Status: accepted
Back to the list of papers