Neural network prediction of across-wind aerodynamic spectrum of rectangular plane super high‑rise buildings

In the wind-resistant design of super high-rise buildings，it is often found that when the aspect ratios are relatively large，the across-wind load recommended by the Code will be too conservative and overestimate the wind load and wind-induced response. The high-frequency base force balance technology is used to carry out wind tunnel tests on 10 kinds of rectangular plane super high-rise models with different aspect ratios in the wind fields B and C. A method of combining genetic algorithm（GA）and back propagation（BP）neural network is adopted. Firstly，GA is used to optimize the initial weight and threshold of BP neural network. Then，optimal parameters are assigned to the BP neural network to train and solve the problem. The k-fold cross-validation method is used for simulation verification. Finally，an across-wind aerodynamic prediction model of the structure with satisfactory accuracy is obtained，which shows that the GA-BP model has the advantages of fast convergence and strong generalization ability. The model is used to predict and compare with the experimental results. The results show that the aerodynamic model based on GA-BP can predict the across-wind aerodynamic spectrum of the structure that is not involved in the modeling. The across-wind load and wind-induced response of the structure calculated by the model and the original wind tunnel data are in good agreement，but they are significantly less than the results of the current load code method at a large depth width ratio，which shows that the results obtained by the code method are conservative. © 2023 Nanjing University of Aeronautics an Astronautics. All rights reserved.