Machine learning-aided prediction of windstorm-induced vibration responses of long-span suspension bridges

Long-span suspension bridges are significantly susceptible to windstorm-induced vibrations, leading to critical challenges of field measurements along with multicollinearity and nonlinearity between wind features and bridge dynamic responses. To address these issues, this article proposes an innovative machine learning-assisted predictive method by integrating a predictor selector developed from regularized neighborhood components analysis and kernel regression modeling through a regularized support vector machine adjusted by Bayesian hyperparameter optimization. The crux of the proposed method lies in advanced machine learning algorithms including metric learning, kernel learning, and hybrid learning integrated in a regularized framework. Utilizing the Hardanger Bridge subjected to different windstorms, the performance of the proposed method is validated and then compared with state-of-the-art regression techniques. Results highlight the effectiveness and practicality of the proposed method with the minimum and maximum R-squared rates of 89% and 98%, respectively. It also surpasses the state-of-the-art regression techniques in predicting bridge dynamics under different windstorms.