Implementation of a Back-Propagation Neural Network for Demand Forecasting in a Supply Chain - A Practical Case Study

Abstract

Demand forecasting is a key way to the efficient management of SCM (supply chain management) in a logistics information system. A poor forecasting approach for the product demands in marketing must cause to decrease competitive capability, lose customers and increase costs. A real case of the product demand forecasting was studied by an artificial neural network (ANN) approach demonstrated in this paper. The studied case is a medium-scale electrical connectors production corporation in Taiwan, which manufactures a variety of the connectors to supply marketing needs of diverse assembly products including mobile telephone, TFT, PDA, CD-ROM, CD-RW, DVD-ROM, DVD-player, notebook computer, digital camera, etc.. The types of the connectors produced by the studied firm are over 50. Owing to the insufficient experimental data provided by the studied corporation, a simulation tool called AweSim was used to simulate the orders of the various types of connectors, according to the historical received orders, and a set of the simulated data was used to train the proposed back-propagation network (BPN) so as to offer a proper demand forecasting tool to the studied firm. Four BPN structures were trained and tested and the best one was determined by ANOVA analysis. The BPN demand forecasting has being used by the studied corporation