基于机器学习的高层建筑风荷载功率谱预测算法研究

Prediction of Wind Load Power Spectrum on High-Rise Buildings by Machine Learning Based Algorithms

摘要: 为了评估机器学习算法预测高层建筑风效应的可行性，本研究开展了机器学习算法对高层建筑标准模型基底力矩系数风荷载功率密度谱预测的研究。机器学习模型的输入是湍流强度、风向角和折算频率，输出是风荷载基底弯矩系数功率谱。训练了三种机器学习算法：梯度提升回归树、直方图梯度提升回归树和XGBoost，采用Tree-structured Parzen Estimator和交叉验证的方法优化算法的超参数。通过对比三种算法在测试集的预测性能，发现梯度提升回归树算法能够很好地预测顺风向、横风向和扭转向的基底弯矩系数功率谱，且预测值与试验值之间的相关系数不低于0.97。研究表明了机器学习预测高层建筑标准模型风荷载功率谱的可行性，为机器学习应用于高层建筑的抗风设计提供参考。

Abstract: In order to evaluate the feasibility of machine learning algorithms for prediction of wind effects on high-rise buildings, machine learning algorithms have been adopted to predict base moment coefficient wind load power density spectrum on standard tall building model. The input of the machine learning model is turbulence intensity, wind directions and the reduced frequencies, and the out-put is the power spectrum of the along-wind, across-wind, and torque base moment coefficients. The three machine learning algorithms including gradient boosting regression tree, histogram gradient boosting regression tree and XGBoost were trained, the hyperparameters of the algorithms were optimized by the Tree-structured Parzen Estimator and cross-validation. By comparing the prediction performance of the three algorithms in the test set, it is found that the gradient boosting regression tree can predict the power spectrum of the base moment of standard tall building model well and the correlation coefficient between the predicted value and the experimental value is not less than 0.97. The study shows the feasibility of machine learning to predict the power spectrum of standard tall building model, and provides a reference for applying machine learning to wind-resistant design of high-rise buildings.