Learning priorities in scheduling steel production

Details

Supervisors

Nijssen, Siegfried

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] en science des données, orientation technologies de l'information, à finalité spécialisée

Abstract

Production order planning problems are complex issues that are frequently encountered by companies in today's world. Indeed, in industries such as metallurgy, the order in which materials are produced is important, as each material requires different conditions before it can be processed. Nowadays, thanks to machine learning, algorithms are available to learn and understand from historical data the optimal order in which materials should be produced. However, these learning models may in some cases need a lot of data before they can effectively predict the optimal schedule. Unfortunately, obtaining a lot of data can be very costly for companies. We, therefore, carried out this work with the aim of studying whether it is possible for a learning model to effectively predict the optimal order with access to only a small amount of data during training. To artificially simulate scheduling problem situations, we used the Prize-Collecting Salesman Problem. This concept has often been used in the literature to artificially represent this type of problem.