On the Use of Data Driven and Fuzzy Techniques to Calculate the Wind Speed in Urban Canyons

Abstract

The present per presents the results of an extensive study aiming to develop and validate alternative data driven techniques able to estimate the wind speed in urban canyons. The use of deterministic techniques to calculate the wind speed in canyons present a low accuracy because of the high uncertainty of the input data and the incomplete description of the physical phenomena. (C. Georgakis et al., 2004) Extended experimental data collected from seven urban canyons have been used to create a data base of the main parameters that define the phenomenon. Using fuzzy clustering techniques, clusters of input-output data have been developed using as criteria the inertia and gravitational forces. For each cluster using statistical analysis, the more probable wind speed inside the canyon and the corresponding input values have been estimated. Thus, a reduced data space has been created. This reduced data space has been used to develop four data driven prediction models. The models are : a 3D graphical interpolation method, a tree based model as well as a linear regression model. Using the results of the graphical interpolation model, a fuzzy estimation model has been developed as well. All methods have been compared against the experimental data