Nonlinear Phenomena in Complex Systems, 2005, Vol.8, No.2, pp.186-192

NONLINEAR PHENOMENA IN COMPLEX SYSTEMS
An Interdisciplinary Journal

2005, Vol.8, No.2, pp.186-192

Decision Supporting System Based on Clusters of Nonlinear Models.
M.Bucolo, F.Caizzone, L.Fichera, L.Fortuna, and G.Tomarchio

Sales Forecast is one of the common Business Management issues that any firm has to deal with. Manufacturing Firms particularly have to approach sales forecast to face different problems like production planning, material purchasing, inventory optimization in order to improve time and service to the customers but also costs. In this paper this problem has been approached for the group of Discrete and Standard Products of STMicroelectronics, one of the world wide biggest Semiconductor Firm. The paper deals with the sales forecast in the short time horizon of three months ahead; particularly it will describe the design of the modeling strategy that has been developed after a formal description of the business model to point out the specification issues for the forecast in terms of constraints, available information, and target. Linear and Nonlinear Forecast clusters have been finally developed to face the problems related to the short amount of available data and to guarantee models robustness by minimizing the recurrence error. The development of a suitable identification strategy has allowed to obtained satisfactory results with both the linear and nonlinear structures and, therefore, to satisfy the target of the forecast accuracy. Nonlinear clusters have globally showed a better performance than linear ones.
Key words: support system, forecast, identification, neural-networks models

Full text: Acrobat PDF (100KB) PostScript (224KB) PostScript.gz (86KB)