Comparison of Machine Learning Models in a Data-Driven Method for Design of Laterally Loaded Monopiles Embedded in Layered Clay Sites

1. Home
2. Publications
3. Online Library
4. Comparison of Machine Learning Models in a Data-Driven Method for Design of Laterally Loaded Monopiles Embedded in Layered Clay Sites

Monopiles are widely used as foundations for offshore wind turbine support structures. The PISA project introduced a 1D model, known as the PISA design model, to predict the monotonic response of laterally loaded monopiles. Offshore wind turbine sites typically comprise distinct soil layers with different geotechnical properties. For layered soil configurations, the PISA design model employs soil reaction curves derived from 3D finite element analyses of homogeneous soils, assuming independent behaviour for each layer. However, this approach does not account for interactions between soil layers, which can significantly impact monopile lateral behaviour. In the current work, an extended version of the PISA design model referred to as the data-driven 1D design model has been developed. This model uses machine learning techniques to define the soil reaction curves, allowing for direct calibration across various soil types and layering configurations. This paper presents a calibration procedure for layered clay sites, considering variations in clay strength, stiffness, and layer thickness. Four machine learning models - sparse Gaussian process regression, artificial neural network regression, support vector regression and eXtreme Gradient Boosting - are assessed to determine the most effective approach.

5th International Symposium on Frontiers in Offshore Geotechnics (ISFOG2025)

5 - Data Analytics and Machine Learning