Applied machine learning model comparison: Predicting offshore platform integrity with gradient boosting algorithms and neural networks
Alec S. Dyer, Dakota Zaengle, Jake R. Nelson, Rodrigo Duran, Madison Wenzlick, Patrick Wingo, Jennifer Bauer, Kelly Rose, Lucy Romeo
Abstract
Offshore oil and gas platforms operating past their design life can pose significant risk to operators and the surrounding environment, as the integrity of these structures decreases over time due to a variety of stressors. This has important implications for industry and government, which are seeking to safely extend the life of platforms for continued use or reuse for alternative offshore energy applications. As a result, there is a need to quantify the remaining useful life (RUL) of operating platforms by analyzing the effects that stressors may have on structural integrity. This study provides a platform risk assessment by employing two machine learning models to forecast the removal age of existing platforms in the U.S. federal waters of the Gulf of Mexico (GoM): a gradient boosted regression tree (GBRT) and an artificial neural network (ANN). These data-driven models were applied to a large, extensive dataset representing the natural and engineered offshore system. Both models were found to provide promising predictions, with 95–97% accuracy and predictions within 1.42–2.04 years on average of the observed removal age during validation. These results can be applied to inform life extension opportunities for fixed and mobile offshore platforms, as well as localized maintenance strategies aiming to prevent operational and environmental risk while maintaining energy production.