Multi-product inventory modeling with demand forecasting and Bayesian optimization

The complexity of supply chains requires advanced methods to schedule companies’ inventories. This paper presents a comparison of model forecasts of demand for multiple products, choosing the best among the following: autoregressive integrated moving average (ARIMA), exponential smoothing (ES), a Bayesian regression model (BRM), and a Bayesian dynamic linear model (BDLM). To this end, cases in which the time series is normally distributed are first simulated. Second, sales predictions for three products of a gas service station are estimated using the four models, revealing the BRM to be the best model. Subsequently, the multi-product inventory model is optimized. To define the policies for ordering, inventory, costs, and profits, a Bayesian search integrating elements of a Tabu search is used to improve the solution. This inventory model optimization process is then applied to the case of a gas service station in Colombia. Keywords: Dynamic Linear Models, Inventory Models, Forecasts, Bayesian Statistics.

• Autores: Valencia Cárdenas, Marisol; Díaz Serna, Francisco Javier; Correa Morales, Juan Carlos
• Tipo de recurso: Article of investigation
• Fecha de publicación: 2016
• Institución: Tecnológico de Antioquia
• Repositorio: Repositorio Tdea
• Idioma: eng
• OAI Identifier: oai:dspace.tdea.edu.co:tdea/4015
• Acceso en línea: https://dspace.tdea.edu.co/handle/tdea/4015
• Palabra clave: Cadenas de suministro; Chaîne d'approvisionnement; Supply chains; Prognosis; Pronóstico; Prognóstico; Pronostic; Modelos dinámicos lineales; Dynamic linear models; Estadística bayesiana; Bayesian statistics; Modelos de inventario; Inventory models
• Rights: openAccess
• License: https://creativecommons.org/licenses/by-nc-nd/4.0/