MACHINE LEARNING PREDICTION OF VOLUME FRACTION OF GAS-HYDRATES IN NATURAL GAS PIPELINES IN OFFSHORE NIGER DELTA

Abstract

This study employs multiphase simulations with OLGA software to investigate volume fractions of hydrate in an offshore gas system and develops machine-learning models to predict these fractions. Annually, substantial operating expenditures are allocated to hydrate prevention, with significant costs associated with inhibition (Wang et al., 2022). Hydrate formation along natural gas pipelines is recognized as a critical threat to the success of gas field operations. Despite the importance, no machine learning model has been validated for predicting volume fractions of hydrate in the Niger Delta study area, making this development crucial. Key findings indicate significant hydrate jamming risks in Niger Delta offshore flowlines and risers, with a peak volume fraction of 0.54, highlighting the need for proactive management strategies. Hydrate formation begins at 750 m where fluid temperatures fall below formation thresholds, with a sudden increase in volume at 2971 m, peaking at 3022 m before declining. Machine Learning model comparisons show Random Forest's superior accuracy (correlation coefficient of 0.9391, mean absolute error of 0.0271), while Linear Regression provides interpretable insights for future predictions. All models perform well, with Random Forest leading in accuracy. Regression analysis reveals relationships between volume fractions of hydrate and various parameters, guiding management strategies. The Random Forest and Linear Regression models are valuable for estimating hydrate volumes and enhancing management approaches in natural gas pipelines due to their accuracy and interpretability. These findings underscore the importance of proactive hydrate management in offshore gas systems and the potential of Machine Learning models to optimize these strategies.