Predicting site-specific suction caisson installation pressure with limited data: a multi-fidelity temporal fusion framework

The use of suction caissons as foundations and anchors for offshore wind turbines is anticipated to grow as governments work towards achieving offshore wind power generation targets. However, challenges remain that must be addressed before suction caissons can be widely deployed in offshore wind farms, particularly in accurately predicting installation suction pressure for site-specific conditions. This paper proposes a multi-fidelity temporal fusion framework that integrates low-fidelity data at an early stage from an empirical model with high-fidelity data from field measurements. The proposed framework achieves substantially higher predictive accuracy than the current industry-standard model for site-specific predictions of soil resistance to caisson penetration. Notably, this level of accuracy is attained even when trained on a limited dataset of caisson installation data.