Efficient 3D geological modeling for offshore wind farm with sparse CPTU and borehole data

Geological modeling of complex marine strata is critical for offshore wind farm development but faces two challenges: (1) inconsistent soil classification between two commonly used site investigation data, i.e., piezocone penetration tests (CPTU) and boreholes, posing a challenge for integrating multi-source information; (2) sparse distribution of investigation data horizontally, which makes 3D spatial interpolation difficult. To tackle these two issues, this study proposes a novel framework for geological modeling of offshore wind farms. The framework encompasses two crucial steps. First, a tailored CPTU-based soil classification model is trained to resolve interpretation discrepancies. Second, an advanced spatial interpolation model is developed for the 3D modeling of sparse investigation data. Both steps make uses of simple but efficient nonparametric machine learning algorithms, e.g., weighted kNN, for predictions, i.e., soil classification and 3D spatial interpolation, respectively. Sequential model-based optimization technique is employed to determine the optimal model hyperparameter. The proposed methodology is demonstrated and validated through benchmark examples and applied to two case studies of offshore wind farms with complex strata in East China Sea and Yellow Sea. It significantly improves soil classification accuracy, especially for silty soils dominated site, and enhances the capability of complex stratigraphy modeling despite sparse data.