Prediction of Scale Precipitation by Modelling its Thermodynamic Properties using Machine Learning Engineering

Abstract

Abstract During oil and gas production, scaling is a flow assurance problem commonly experienced in most regions. For scale control to be effective and less expensive, accurate prediction of scaling immediately deposition commences is important. This paper provides a model for the prediction of Barium Sulphate (BaSO4) and Calcium Carbonate (CaCO3) oilfield scales built using machine learning. Thermodynamic and compositional properties including temperature, pressure, PH, CO2 mole fraction, Total Dissolved Solids (TDS), and ion compositions of water samples from wells where BaSO4 and CaCO3 scales were observed are analysed and used to train the machine learning model. The results of the modelling indicate that the Decision tree model that had an accuracy of 0.91 value using Area Under Curve (AUC) score, performed better in predicting scale precipitation in the wells than the other Decision tree models that had AUC scores of 0.88 and 0.87. The model can guide early prediction and control of scaling during oil and gas production operations.