Assessing seismic reliability & feature importance of steel jacket type offshore platforms using CatBoost algorithm

Predicting the vulnerability and failure probability of steel jacket-type offshore platforms (SJTOPs) during earthquakes is challenging due to uncertainties in structural, soil and loading parameters. This study used machine learning, specifically the CatBoost algorithm, to assess the seismic reliability of SJTOPs by analysing the feature importance of 17 variables related to the superstructure, pile and soil. The study simulated 150 models considering these uncertainties using the Latin Hypercube method and conducted nonlinear time history analyses in OpenSees for artificial earthquakes at two seismic hazard levels. The results indicated that leg diameter is the most critical superstructure parameter, while shear wave velocity and the shear modulus to maximum shear modulus ratio significantly influence the soil's impact on structural response. The study found that SJTOPs have a higher failure probability in the serviceability limit state than in the ultimate limit state, highlighting the importance of accurate seismic assessments for offshore platforms.