Development of machine learning models for predicting the deposition of sulfide scales in oil production wells

Abstract Scale deposition in oil wells poses numerous operational challenges that can lead to blockage of the completion string; therefore, it is recommended to predict scale accumulation before its occurrence. The current study aims to develop Artificial Neural Network (ANN) models and other Ensemble Machine Learning (ML) models to detect the presence of sulfide scales in oil wells and estimate the percentage of the sulfide scale composition in the accumulated scale. The studied sulfide scales are zinc, lead, and iron sulfides. The investigated ML models are Artificial Neural Network, Decision Tree, Random Forest, Support Vector Machine, K-Nearest Neighbors, Gradient Boosting, Extreme Gradient Boosting, Adaptive Boosting, Light Gradient Boosting Machine, and Ridge Regression. The ML models were constructed using actual field data for lab-analyzed and physically collected scale samples. These samples were collected from 347 production wells in 17 fields over a 22-year period. The original database consists of 1486 scale deposition incidences, and it was randomly split into 80% for training and validation, and 20% for testing. The assigned input data for scale prediction involve the relevant surface and downhole data, including: water ion analysis, water pH, production rates of well fluids, and content of acid gases, downhole pressure, downhole temperature, and injection gas rate for gas-lifted wells. The ANN models were implemented in several field applications and yielded promising results compared to scale tendency calculations from commercial scale prediction software. Moreover, variant models of Gradient Boosting and KNN models showed the highest model accuracy in predicting sulfide scale percentage.