Machine Learning-Based Prediction of Wax-Related Production Losses for Flow Assurance Optimization

Wax deposition poses a significant challenge to flow assurance in oil production systems, potentially leading to production losses and increased maintenance requirements. Accurately forecasting the impacts of wax deposition on hydrocarbon production can enhance the mitigation strategies and optimize field operations. This study developed supervised machine learning (ML) models to predict oil, water, and gas production rates, using key wellhead, surface, and wax-related parameters. Both tree-based and neural network algorithms were used. A dataset containing 306 samples from multiple wells was used, with 80% utilized for training and 20% reserved for testing the models. Ensemble tree-based models delivered the most accurate performance, with Extra Trees achieving an R² above 0.98 for all targets. The performance of the deep learning model, however, was limited by the small dataset size. Singular Value Decomposition analysis was performed to understand how different parameters influence well production dynamics. The results indicated that the wax accumulation rate and wax cutting frequency are essential contributors to well dynamics, alongside tubing head pressure and choke size. Wax-related features were further explored through sensitivity analysis, which showed that increased wax deposition and intervention frequency resulted in reductions of up to 25% in oil rates and 31% in gas rates. Based on the predictions, oil production was more negatively affected by the frequency of wax cutting, whereas the wax accumulation rate had a more severe impact on gas production. Therefore, careful design of flow assurance measures, balancing reactive strategies, such as mechanical wax cutting, with proactive measures, such as chemical inhibition, is critical. The proposed ML workflow provides a reliable predictive framework for hydrocarbon production to support flow assurance decisions, reducing wax-related downtime and enhancing field productivity.