A multi-parameter deep learning model for wind pressure prediction on low-rise buildings

Current wind pressure prediction methods for low-rise buildings primarily address the determination of wind pressure when a single geometric parameter of building changes, with limited research on scenarios where multiple geometric parameters vary simultaneously. This paper establishes a multi-parameter deep learning model for predicting wind pressure statistical values of low-rise buildings. The model is trained and tested for generalization capability using the TPU aerodynamic database. The influence of whether the eave height and roof slope of the tested buildings are included in the training dataset on prediction accuracy, as well as the accuracy of extrapolation predictions, are discussed. Furthermore, the impact of training set size on prediction accuracy is investigated, and the model's predictive capability for low-rise buildings with arbitrary eave heights and slopes but the same planar dimensions, as well as its adaptability to buildings with different planar dimensions, is demonstrated. The study by 15 scenarios shows that the proposed model exhibits strong generalization capability for low-rise buildings with multiple varying geometric parameters. Whether the eave height and roof slope of the tested buildings are included in the training dataset has no significant impact on prediction accuracy, and the accuracy of extrapolated predictions remains relatively stable.