Predicting sulfate mineral scale solubility with machine learning

Citation:

Cao Z, Hu Y*, Zhang P*. Predicting sulfate mineral scale solubility with machine learning. Journal of Cleaner Production [Internet]. 2024.

摘要:

Mineral scale refers to the hard inorganic solids nucleated on substrates or deposited from the aqueous phase. The formation and deposition of barium sulfate and strontium sulfate in various industries, such as water treatment and oilfield operations, can significantly impact facility operations, posing serious threats. Machine learning (ML) approaches have been adopted recently in scale threat predictions to address the limitations of conventional scaling prediction models. However, there are few reports on collecting sulfate mineral scaling data, employing ML methods for data analysis, and evaluating the modeling results to gain deeper insights of sulfate mineral scaling process and to improve the accuracy of sulfate scaling threat prediction. Despite comprehensive experimental studies, the literature does not provide adequate guidance for identifying the influence on the solubility of barium sulfate and strontium sulfate under different aqueous environments and actual operating conditions. To this end, this study collected 1600 experimental datasets of barium/strontium sulfate from the literature to construct and evaluate the reliability and versatility of a ML-based model for sulfate solubility calculations. Single neural networks, hybrid neural networks, and optimization algorithms were employed to build solubility prediction models for barium sulfate and strontium sulfate across a wide range of temperatures, pressures, and different ions. The model's applicability in predicting sulfate scaling threats in various actual operating environments demonstrated its broad usability, consistent with its actual performance. This study marks the first stride towards constructing a reliable model for identifying the scaling trends of barium sulfate and strontium sulfate across various operating conditions, underscoring the importance of developing robust and accurate prediction models to address challenges in various industrial systems.

Link