Ensemble learning for prediction of inorganic scale formation: A case study in Oman

Inorganic scale formation is one of the major flow assurance issues in geothermal energy, oil, and water production due to its damaging effects on formation rock, wells, and transportation facilities. Owing to the intricate nature of scale formation, developing a closed-form mathematical formulation for its prediction is difficult. Thereby, the ability of six machine learning algorithms and a Power Law Ensemble Model (PLEM) to predict inorganic scale formation in carbonate formations is examined in this study. A new dataset of scale formation from realistic wells in Oman, which included temperature, pressure, artificial lift, ionic composition, pH, total dissolved solids, and scale formation tendency of each well, was collected from two reservoirs (Natih and Shuaiba). The machine learning models are Naive Bayes (NA), Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Decision Tree (DT). The results revealed that the RF, KNN and DT provided the best predictions among the individual experts with a F1-scores of 78.6%, 75.9%, and 71.0%, respectively. By integrating the predictions of the individual experts using the PLEM method, the F1-score increased to 90.3% on the test subset. Moreover, a study was conducted to find out some rules of thumb for each input that cause scale formation, but the results showed that there was no explicit condition, indicating that scale formation is a complicated phenomenon requiring advanced modeling approaches. Finally, a new water analysis report was given to the PLEM (as the best model) to predict scale formation in a well that demonstrated a match between the predicted class and the report outcome.