Optimizing the membrane ultrafiltration process using machine learning: A decision making tool based on self-organizing maps

doi:10.1016/j.jwpe.2024.106787

2025-01-01

PRE

DOI：10.1016/j.jwpe.2024.106787

原文链接

原文求助

摘要

En 中文

Ultrafiltration has become essential in most drinking water production facilities. However, managing the variability of water quality, especially during strong fouling events, remains challenging. Therefore, this study aims to develop a decision-making tool capable of predicting membrane fouling based on water quality. To achieve this, a monthly monitoring of six water resources (2 surface and 4 ground waters) over 11 months was conducted. Filtration tests were performed on all resources to assess their fouling potential, and 15 quality parameters were measured. Subsequently, the collected data was utilized to create a self-organizing map (SOM) to classify water resources according to their quality parameters. The resulting map is capable of clustering water resources into three categories with each category exhibiting a distinct fouling potential: (1) Water with high NOM and turbidity levels and significant fouling potential, (2) Water with moderate NOM and turbidity levels and moderate fouling potential and (3) Water with low NOM and turbidity levels with no fouling potential. Finally, the study was concluded by presenting two applications for the SOM. The first application demonstrated the ability to determine an optimal mixing ratio for two resources with different qualities, which reduces overexploitation of the cleaner water while maintaining effective filtration. The second application highlighted the optimization of coagulant dosage during the coagulation process, leading to a lower chemical usage without compromising water quality or causing membrane fouling. These successful applications emphasize the potential of SOMs in water resource management and improving overall process efficiency in drinking water production.

Keyword:

Ultrafiltration

Water resource management

Membrane fouling prediction

Artificial intelligence

Machine learning