DEMAND PREDICTION IN INDUSTRY 4.0 THROUGH A TRANSFORMER-BASED ARCHITECTURE (PREDICCIÓN DE LA DEMANDA EN LA INDUSTRIA 4.0 EMPLEANDO UNA ARQUITECTURA BASADA EN TRANSFORMERS)

Abstract

Given the changing circumstances faced by the industry today product demand is constantly influenced by several factors such as global economic conditions affected by wars, pandemics and recessions. For companies to cope with these challenges effectively, it is crucial to have a dependable demand prediction system that can be swiftly and efficiently communicated to their supply chain. However, this can be challenging for large consortiums with distributed supply chains spanning different countries. In this study a neural network architecture based on Transformers is proposed for demand prediction. This system could be integrated into a cloud service accessible to various locations within a company's supply chain, thus reducing information delays. By evaluating our approach with real product demand data and comparing it with other architectures the experiments prove that our model outperforms other methods.