基于概率机器学习的大跨度桥梁台风效应预测方法对比研究OA

Accurate and efficient prediction of wind-induced responses in long-span bridges is critical for en-suring their wind-resistant safety and operational performance.To address the challenges of unstable predic-tion accuracy,weak uncertainty characterization,and poor real-time applicability in existing methods,the Su-tong Bridge was employed as the research case,and eight probabilistic machine learning models were adopted to predict typhoon effects based on the decade-long datasets of typhoons and their effects collected from the bridge site.The predictive performance of these models was comprehensively compared in terms of prediction accuracy,uncertainty quantification,and computation efficiency.The results indicate that deep ensemble-based models significantly reduce prediction errors and achieve the highest prediction accuracy.Compared with other probabilistic machine learning models,such models also exhibit superior uncertainty quantification performance.Despite the variations in training time across different models,all require only minimal testing time,demonstrating the feasibility of rapid prediction of typhoon effects using probabilistic machine learning approaches.