Machine learning-based prediction of scale formation in produced water as a tool for environmental monitoring

Predicting scale formation in produced water (PW) under real-world conditions poses a significant economic and environmental challenge for the oil and gas industries. This is primarily due to the continuous variation in salt concentrations, temperature and pressure affecting inorganic scale composition. Machine learning (ML) as a data-driven method is a powerful tool for uncovering hidden patterns in experimental data necessary for decision-making on scale formation predictions by analyzing the complex relationships between mainly the water chemistry and the pH. We used a database comprised of 2313 quality data points from different locations in the Bakken Shale Play, including values such as ionic compositions, pH, and the saturation index of the potential mineral scales in PW at 60°F and 60 psi to train the ML algorithms and identify what scale will likely form in the PW. We used Multiple Linear Regression (MLR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) techniques, achieving R² values of 0.79, 0.82, and 0.95, respectively, for multi-scale formation prediction. We identified the significant features influencing scale formation by applying a Shapley-Additive-exPlanations (SHAP) analysis to our model. This method not only improves understanding of water chemistry but also enhances environmental analyses and monitoring by predicting scale formation as a potential form of contamination, which can help assess and maintain water quality in industrial settings. By shifting from traditional thermodynamic models to a data-driven, user-friendly platform, this approach offers a practical tool for field workers, researchers, and technical managers, even if they have limited expertise in thermodynamics, kinetics, programming, or numerical analysis.