Faculty
Department
Year of Publication
upload
Publication Type
Abstract
Flow assurance remains a critical challenge in the oil and gas industry, where complex interactions among temperature, pressure, corrosion, and flow dynamics can lead to operational inefficiencies, production losses, or complete pipeline blockage. Traditional rule-based and thermodynamic models often fall short in capturing the nonlinear, multi-parameter dependencies underlying these challenges. In this study, a data-driven framework was developed for the prediction and early detection of flow assurance challenges in oil and gas pipelines using ensemble machine learning models.
A dataset comprising twenty-four operational and material parameters including temperature, pressure, pipe size, flow rate, corrosion impact, and energy consumption was analyzed to classify pipeline states into normal, moderate, and critical risk categories. Three algorithms Random Forest, Support Vector Machine (SVM), and Gradient Boosting—were implemented, optimized through grid-based hyper parameter tuning, and evaluated using cross validation and standard performance metrics such as accuracy, precision, recall, F1-score, and confusion matrices.
The results indicated that all three models successfully identified key operational relationships influencing flow assurance risks. The Random Forest model achieved a high training accuracy of 98.71% but showed overfitting with a test accuracy of 40.33%. The SVM model achieved a test accuracy of 45.00% with a recall of 70.6% for critical conditions. The Gradient Boosting model outperformed both, achieving a cross-validation score of 47.71%, test accuracy of 49.33%, and recall of 97.2% for critical flow states, with minimal overfitting (accuracy gap of 4.10%).
The study concludes that ensemble machine learning methods, particularly Gradient Boosting, offer a reliable and interpretable approach for predicting and classifying flow assurance challenges. By enabling early detection and proactive intervention, the proposed framework can support predictive maintenance, reduce pipeline downtime, and enhance operational safety and efficiency in oil and gas transportation systems.
A dataset comprising twenty-four operational and material parameters including temperature, pressure, pipe size, flow rate, corrosion impact, and energy consumption was analyzed to classify pipeline states into normal, moderate, and critical risk categories. Three algorithms Random Forest, Support Vector Machine (SVM), and Gradient Boosting—were implemented, optimized through grid-based hyper parameter tuning, and evaluated using cross validation and standard performance metrics such as accuracy, precision, recall, F1-score, and confusion matrices.
The results indicated that all three models successfully identified key operational relationships influencing flow assurance risks. The Random Forest model achieved a high training accuracy of 98.71% but showed overfitting with a test accuracy of 40.33%. The SVM model achieved a test accuracy of 45.00% with a recall of 70.6% for critical conditions. The Gradient Boosting model outperformed both, achieving a cross-validation score of 47.71%, test accuracy of 49.33%, and recall of 97.2% for critical flow states, with minimal overfitting (accuracy gap of 4.10%).
The study concludes that ensemble machine learning methods, particularly Gradient Boosting, offer a reliable and interpretable approach for predicting and classifying flow assurance challenges. By enabling early detection and proactive intervention, the proposed framework can support predictive maintenance, reduce pipeline downtime, and enhance operational safety and efficiency in oil and gas transportation systems.
Supervisor(s)
co-supervisor