next up previous
Next: Einleitung Up: Abverkaufsprognose mit paralleler Backpropagation Previous: Kurzfassung


In this paper artificial neural networks are adapted to time series prediction. Feedforward multilayer perceptron networks are trained by the back-propagation algorithm in order to forecast the future sales of articles of a certain group of related products in supermarkets. The back-propagation algorithm is parallelized and implemented for multi-processor computers to reduce the duration of training time. First empirical results concerning the optimal network topology and training parameters are discussed.

Thu Jul 27 12:34:10 MET DST 1995