USING ARTIFICIAL NEURAL NETWORKS TO PREDICT INTERIOR VELOCITY COEFFICIENTS

A new method is presented for the evaluation of interior velocity coef ficients using artificial neural networks. The interior velocity coeff icient gives a measure of the relative strengths of inferior air movem ents in the horizontal plane representing an occupied space. Air movem ents in a building depend not only on the external wind velocity, bur also, and indeed principally, on a number of architectural parameters. However, if a meaningful number of such parameters are to be taken in to account, the determination of interior velocity coefficients is ver y difficult. It was therefore decided to look at how artificial intell igence techniques might facilitate the solution of the problems involv ed. After presentation of the background of the study, an introduction to neural networks is given, with their main properties and methods o f implementation. It is shown how these ideas are applied in the prese nt study, and the initial results are presented. The utilization of ne ural networks as a universal predictor is an interesting subject for i nvestigation, given their ability to provide reliable results in situa tions where a large number of parameters have to be taken into account simultaneously. (C) 1997 Elsevier Science Ltd.