Predicting few disinfection byproducts in the water distribution systems using machine learning models.

Area of Science:

• Environmental Science
• Water Quality Monitoring
• Predictive Modeling

Background:

• Disinfection byproducts (DBPs) in drinking water are a persistent concern.
• Measuring DBPs in water distribution systems (WDSs) is challenging due to accessibility.
• Machine learning (ML) offers potential for improved DBP predictions in WDSs.

Purpose of the Study:

• To develop and evaluate ML models for predicting key DBPs in WDSs.
• To assess the performance of various ML models using real-world data.

Main Methods:

• Collected 13 years of tri-monthly DBP data (2008-2020) from 113 Ontario water supply systems.
• Trained and tested four ML models: Linear Regressor (LR), Random Forest Regressor (RFR), Support Vector Regressor (SVR), and Artificial Neural Networks (ANN-SV/MV).
• Evaluated model performance using R-squared (R²) values on training and testing datasets.

Main Results:

• High R² values were achieved for training (0.449-0.993) and testing (0.437-0.973) datasets across different DBPs.
• ANN-SV models excelled in predicting Trihalomethanes (THMs), Haloacetic Acids (HAAs), and Dichloroacetonitrile (DCAN).
• SVR models demonstrated superior performance for N-nitrosodimethylamine (NDMA) prediction.

Conclusions:

• Developed ML models reliably predict DBPs in WDSs, offering an alternative to traditional sampling.
• These models can enhance DBP control strategies in WDSs.
• Improved DBP prediction and control can reduce human exposure and associated health risks.