The interference index and its prediction using a neural network analysis of wind-tunnel data

Abstract

This paper reports on recent progress in the authors’ ongoing efforts to quantify the effects of shielding and interference between pairs of buildings located in proximity in a variety of geometric configurations and boundary-layer wind flows. Recent developments in numerical analytical techniques and expert systems have made neural network analysis available as a potentially useful tool in the investigation of this problem. Analysis using neural networks allows the quantification of variables over a continuous range of values, whereas results have previously been limited to the analysis of specific configurations which have been wind-tunnel tested or to the identification of qualitative trends. In this study, we have applied neural network methodology to wind-tunnel data obtained from a variety of sources which describe shielding and interference behavior between two buildings. The results are presented here in terms of a newly defined “Interference Index”. Once a neural network has been properly configured and trained, it can easily generate results for building configurations that have not been tested experimentally, based on the patterns it has derived from the available wind-tunnel data.