Predicting wave and current loads on jack-up offshore platforms using an integrated machine learning and transfer learning framework

Rapid, precise wave and current load prediction is critical for jack-up offshore platforms’ structural safety. Theory-driven approaches are inefficient for multi-condition computations, while machine learning (ML) offers computational benefits but has limited transferability when structural parameters change. Transfer learning (TL) can address this limitation. This study evaluates an integrated ML-TL framework for rapid load prediction. Random forest regression was employed to develop the source-domain model, whose feature importance then guided TL to reduce target-model training samples. A numerical case study based on linear Airy wave theory and static analysis was conducted for a jack-up platform in the Persian Gulf. Multiple target domains were created by varying leg number, spatial arrangement, and diameter. Results show that the framework enables efficient and accurate prediction in both source and target domains, with target models requiring less than one-third of the source-model training samples. The framework demonstrates its potential for rapid load prediction.