Le 29/11/2023 par gicquel :
Bonjour à tous,
Veuillez trouver ci-dessous une offre pour un stage au sein du laboratoire LISN de l'Université Paris Saclay. Le sujet porte sur la planification de production industrielle avec incertitude sur la demande des clients.
Risk-averse models for industrial production planning under demand uncertainty
Production planning is among the key processes encountered in a manufacturing company. Basically, production planning consists in deciding what, when, and how much to produce on a production resource (such as a chemical reactor) over a finite horizon, typically spanning several days or weeks. The resulting production plan should satisfy as best as possible the customers’ demand for the finished products while minimizing the total production costs. Dynamic lot-sizing arises in production whenever set-up operations (e.g., tool changing or cleaning) are required to prepare the resource before the processing of a new type of product.
Thus, one of the key input data in lot-sizing is the time-varying customers’ demand to be met. However, information about this future demand is most often obtained through imperfect forecasting procedures. Neglecting forecasting errors while planning production may lead to stock-outs and unsatisfied customers’ demands and/or to inventory levels higher than planned. Hence, dynamic lot-sizing should be considered as a combinatorial optimization problem involving uncertainty. In this internship, we consider that an accurate probabilistic description of the random demand is available under the form of probability distributions and will thus study stochastic programming approaches for lot-sizing under uncertainty.
However, most previously published approaches on stochastic lot-sizing rely on risk-neutral models seeking either to minimize the expected cost of the production or to comply with service level constraints setting an upper bound on the expected backorders. Such approaches rely on the assumption that the production planner is risk-neutral, i.e., is ready to accept a very high production cost (or a very poor service level) for certain unfavorable realizations of the stochastic demand as long as this is offset by a smaller production cost (or a better service level) in more favorable realizations. However, this assumption does not always hold in practice and the decision-maker may be risk-averse. In this case, she would be concerned, not only by the average performance of the production plan but also by its performance under unfavorable realizations of the demand. For instance, she may wish to control the production cost increase and/or service level deterioration when the actual demand is much higher than the expected one. It appears that such risk-averse models have been scarcely studied in the lot-sizing literature.
The aim of this internship will thus be to study how the risk aversion of the production planner may be incorporated in the modeling of production planning and dynamic lot-sizing problems under stochastic demand. More specifically, we will focus on a variant of the lot-sizing problem (the single-item capacitated lot-sizing problem) and investigate how risk measures may be introduced in the problem formulation. Our objective will be to gain some insights about the potential practical interest of these approaches.
Production planning, lot-sizing, stochastic programming, mixed-integer linear programming, risk aversion
Environment: The internship will take place at the LISN (Laboratoire Interdisciplinaire des Sciences du Numérique) of the Université Paris Saclay (91190 Gif-sur-Yvette).
It will be jointly supervised by Céline Gicquel (Associate Professor at the University Paris Saclay), Franco Quezada (Assistant Professor at the University of Santiago of Chile), Bernardo Pagnoncelli (Associate Professor at SKEMA Business School, Lille) and Safia Kedad-Sidhoum (Professor at the Conservatoire National des Arts et Métiers, Paris).
Time and duration: The start of the internship is flexible but should be in Spring 2024. The expected duration is 4 to 6 months.
Profile Candidates: Candidates must be M2 level student (2nd year of MSc or last year of “cycle ingénieur”). They must have a solid background in applied mathematics and computer science, good programming skills, and a particular liking for operational research.
Contact: Candidates must send their CV, a letter of motivation and, if available, their master marks to Céline Giquel (celine.gicquel_at_ lisn.upsaclay.fr).