Abstract
Offshore wind turbines (OWTs) operate in harsh and unpredictable environments, where combined effects of wind, waves, and earthquakes produce highly nonlinear structural responses. Although finite element (FE) simulations provide accurate predictions, they are computationally intensive and impractical for large parametric studies or real-time decision-making. This study introduces a data-driven automated machine learning (AutoML) framework to develop predictive models of monopile-supported OWTs under multi-hazard conditions. The proposed methodology establishes mappings between critical input variables such as turbine rating, water depth, wind speed, wave characteristics, and seismic acceleration and key engineering demand parameters, including maximum hub displacement, hub acceleration, and pile-head rotation. To address the substantial variability and skewness inherent in offshore datasets, multiple machine learning algorithms are systematically evaluated, including ensemble methods, Gaussian process regression, multilayer perceptrons, kernel regression, decision trees and support vector machines. The resulting surrogate models achieve rapid and reliable response predictions, offering a practical alternative to computationally demanding FE simulations. Their demonstrated accuracy and robustness support integration into real-time risk assessment and decision-support workflows for high-value offshore assets. Validation on unseen offshore scenarios within the sampled design space further confirms the models’ potential to enhance the efficiency and scalability of offshore structural analysis.
•AutoML framework predicts OWT multi-hazard responses with high accuracy.•Surrogate models replace costly nonlinear FE simulations for rapid assessment.•Multiple ML algorithms evaluated for skewed offshore datasets.•Models generalize well across unseen operating conditions within the sampled space.•Enables real-time risk and decision support for offshore wind infrastructure.