Artificial Intelligence for Optimal Chemical Dosing in Water Treatment Plants: A Comparative Study Using Random Forest and XGBoost Algorithms

Abstract

This research relies on artificial intelligence (AI) instead of laboratory testing (jar test) to determine the optimal dosage of chemicals (polymer, alum and micro sand) used to treat water at the Qarmat Ali site in Basra Governorate, which is used for reservoir injection in the Rumaila oil field. The research utilized the XGBoost and Random Forest algorithms. The results demonstrated high consistency with experimental results collected over three years. When comparing these two algorithms, the Random Forest algorithm consistently outperformed XGBoost in accuracy across all variable river water specifications. The results support the use of AI to determine the optimal dosage at any given moment and in all seasons, thereby improving export capacity and produced water specifications.