L'UNIVERSITE DES SCIENCES ET TECHNOLOGIES DE LILLE Ecole Doctorale des Sciences de la Matière, du Rayonnement et de l'Environnement

Competition and globalisation imply a very accurate production and sourcing management of the TextileApparel-Distribution network actors. Face to the production and distribution constraints of the products (lead time relatively long and short lifetime of the items, ...), such an organization requires a sales forecasting system adapted to market uncertainties and distributor needs. The uncertain character of the sales is characterized by the influence of many not easily controllable and identified explanatory variables. The distributor must take into account mean-term forecasts (horizon : one season) in order to anticipate its production and its first provisioning. The readjustment of the short-term forecast (horizon: one to three weeks) is also necessary in order to correct the planning of the replenishments, throughout the season. Many forecasting models exist. However, they are generally unsuited to the textile context. Indeed, their capacities of training and modelling are often limited for the reduced and disturbed historical sales data. The user understanding and intervention are also often complicated when using traditional models. Thus, we propose a forecasting system, composed of several models which perform forecasts on various horizons and at different sales aggregation levels. This system is based on techniques resulting from soft computing such fuzzy logic, neural networks or the evolutionary procedures, authorizing the processing of uncertain data. Performances of our models are then evaluated and analysed on real data from a important textile distributor. Lastly, within the framework of a project called AIDE, this forecasting system is integrated in a decision-making aid tool for each actor of the textile network. DISCIPLINE : Productique – Automatique et Informatique Industrielle MOTS-CLES : Prévision des ventes, Prédiction des séries temporelles, Soft computing, Logistique de la filière Textile-Habillement-Distribution, Logique floue, Réseau de neurones artificiel, Algorithmes évolutionnistes KEY-WORDS : Sales forecasting, Time series prediction, Soft computing, Textile-Apparel-Distribution network logistics, Fuzzy logic, Artificial neural network, Evolutionary algorithm LABORATOIRE GEMTEX-ENSAIT 9, rue de l'ermitage – BP 30329 http//www.ensait.fr F-59056 ROUBAIX Cedex 01 INSTITUT FRANÇAIS DU TEXTILE ET DE L'HABILLEMENT 2, rue de la recherche http//www.ifth.org 59650 VILLENEUVE D'ASCQ Cedex