Artificial intelligence model helps produce clean water

About 2.2 billion people, more than a quarter of the world’s population, lack access to safe, managed drinking water, and about half of the world’s population experiences severe water scarcity at some point during the year. To overcome these shortages, huge socioeconomic costs are being spent on sewer irrigation and alternative water sources such as rainwater reuse and seawater desalination.

Furthermore, these centralized water distribution systems have the disadvantage of not being able to respond immediately to changes in water demand. Therefore, there is a growing interest in decentralized water production technologies, which are electrochemical-based technologies that are easy to adopt, such as capacitive deionization and battery electrode deionization (also known as faradaic deionization).

However, the existing water quality measurement sensors used in electrochemical-based technologies do not measure and track individual ions in water, and have the limitation of roughly inferring water quality conditions from electrical conductivity.

Dr. Son Moon’s research team at the Korea Institute of Science and Technology (KIST) Water Resource Cycle Research Center, in collaboration with Professor Baek Sang-Soo’s team at Yeongnam University, has developed a technology that uses data-driven artificial intelligence to accurately predict the concentration of ions in water during electrochemical water treatment processes.

Their paper is published in the journal Water Research.

The researchers first built a random forest model, a tree-based machine learning technique utilized for regression problems, and then applied it to predict ion concentrations in electrochemical water treatment technologies.

The developed random forest-based artificial intelligence model was able to accurately predict the electrical conductivity of the treated water and the concentration of each ion (Na⁺, K⁺, Ca²⁺, and Cl^–) (R²=~0.9).

They also found that updates were required about every 20–80 seconds to improve the accuracy of the predictions, which means that in order to apply this technique to national water quality networks to track specific ions, it is necessary to measure water quality at least every minute to train the initial model.

The random forest model used in this study has the advantage of being economically superior to complex deep learning models, requiring more than 100 times less computing resources to train.

“The significance of this research is not only in developing a new AI model, but also in its application to the national water quality management system,” said Dr. Moon. “With this technology, the concentration of individual ions can be monitored more precisely, contributing to the improvement of social water welfare.”