THE ESTABLISHMENT OF WIND SPECTRUM ESTIMATION MODELS FOR DOME-LIKE STRUCTURES USING ARTIFICIAL NEURAL NETWORKS

Wind tunnel test results of 35 dome models with rise/span ratio (f/D) from 0 to 0.5 and height/span ratio (h/D) from 0 to 0.5 in boundary layer flow with power law index 0.27 were collected. A wind pressure database for the dome-like roofs was established. The focus of the research reported in this paper was on the differences of wind pressure spectra on the meridian with the change of curvature and height. Random center selection method was used to write Radial Basis Function Neural Network (RBFNN) programs to train, validate and test the ANNs. Several network architectures, data processing and data grouping methods were investigated. The final estimation models found not only accurate but also theoretically consistent. Models were also compared with previous regression formula, and the results were much better. In the future, the ANN models will be implemented using a network platform and a simple web browser user interface. Wind pressure spectra calculated by the server can be easily obtained with simple parameter inputs, which can be used as preliminary estimations before wind tunnel tests.