Feasibility analysis of jacket support structures for offshore wind turbines employing a regression-based artificial neural network model

The use of jacket-structured support systems for offshore wind turbines is growing, particularly in response to the increasing need for deeper water installations and greater distances from shore. However, designing jacket support structures remains computationally demanding due to complex structural analysis and load evaluation requirements. To address these challenges, this study employs regression-based artificial neural network models to assess the structural feasibility of jackets at specific installation sites. A synthetic dataset that incorporates key parameters of wind turbines, site conditions, and jacket configurations, is used for training the neural networks. The effectiveness of predicting the global feasibility of the structure or several partial checks imposed is analysed. Also, different architectures and assembly strategies are analysed. The results indicate that regression-based models achieve great performance in predicting the feasibility of the structures, with high Matthews correlation coefficient scores and strong correlations between predicted utilization factors and actual structural compliance. A comparison against a similar classification-based model suggests that regression-based models offer a more accurate prediction of the border between feasible and non-feasible designs. This characteristic is very useful for including such models in optimization processes, as it provides clear differentiation between viable and non-viable designs. • Regression-based ANNs are used for estimating jacket feasibility for OWTs. • Strategies for aggregating partial checks into a global feasibility are compared. • Estimating global feasibility from partial checks improves performance. • Regression outperforms classification around the feasible/non-feasible boundary.