A novel AI-based CNN model to predict the structural performance of monopile used for offshore wind energy systems

Abstract

This study builds an AI-based Convolutional Neural Network (CNN) model to guess 50-year extreme wind and wave conditions and assess structural loads on the monopile foundation of the NREL 15 MW offshore wind turbine. The model was trained and validated by means of 7 years of measured wind and wave data, applying an organized filtering process to check data quality. The CNN projections were evaluated via a multi-step validation approach, integrating extreme value investigation and structural load approximation. The AI-CNN model forecasted a 50-year extreme wind speed (EWS) of 21.61 m/s, 5.3 % higher than the Gumbel algorithm, guaranteeing conventional load calculations. Structural analysis by means of BLADED software demonstrated that critical load sub-components, such as the y-force and x-moment, amplified by up to 10 %, strengthening safety limits under extreme circumstances. Additionally, the AI-CNN model was well validated alongside psychrometric data to expand prediction stoutness further than established extreme value modeling. Additionally, comparative assessment of training dataset sizes (100–800) validated increasing model accuracy and reliability with bigger datasets, highlighting the effectiveness of long-term measured data for CNN training. These conclusions validate the AI-CNN model as a dependable tool for extreme environmental load calculations, advancing enhanced optimization and structural safety for OWT monopile foundations.