Application Of The RBF Neural Network In The Wind-induced Interference Effects On Tall Buildings

The problems of wind-induced interference effects on tall buildings have beenattracting great concerns in wind resistant researches. However, most previousinvestigations mainly focus on the interference effects between two buildings. Due tothe high complexity of the wind tunnel experiment and the massive amount of testingdata, the RBF neural Networks are applied to investigate the comprehensiveinterference characteristics of among three and two tall buildings.The main contributions and results of this thesis can be summarized as follows:Firstly, on basis of the comparison study of different algorithm of the RBF neuralnetworks. When the number of input vector is the same as the number of neural cells,an additional damping item is introduced to improve the predictability of the network.The generalized RBF networks, the number of input vector is less then the numberof neural cells, are also used in the study and show a good predictability by thecarefully selection of the multi- net training parameters through a series of effectivetests.Secondly, The RBF neural network approach was applied to model and predictthe complex interference effects between two tall buildings. And the correlation andregression methods are further used for the quantitative analysis of the distribution ofinterference factors on basis of the RBF modeling. The relevant regression equationsare obtained and proposed to simplify the complexity of the multi-parameter windinduced interference effects between two tall buildings with different breadth ratiosand height ratios.Thirdly, For interference effects among three tall buildings, on basis of the RBFmodeling and prediction, the simplified interference factors contours are proposed torepresent the distribution of the interference factor among three tall buildings. Theacceleration based interference effects are also analyzed, and the distributions ofacceleration based interference factors are compared with the distributions of the blending moment based interference factors.Finally, Some suggestions are provided for the assessment of wind inducedinterference effects on design loads for tall buildings.