Development of a real-time THMs prediction technology based on artificial intelligence and sensors

This study presents the development of a real-time prediction model for trihalomethanes (THMs) using deep learning and sensor-based data from a drinking water treatment plant. As chlorine remains the most widely used disinfectant due to its cost-effectiveness, the formation of carcinogenic THMs has become a critical concern, especially with multi-point chlorine injection strategies. Traditional THMs prediction models have faced limitations due to the complex nature of organic matter and the lack of real-time data availability. In this study, a deep learning model utilizing artificial neural networks (ANNs) was trained on 171 data points comprising real-time flow and water quality variables. Model performance was evaluated using mean absolute error (MAE), mean squared error (MSE), and R2 metrics under different activation functions including rectified linear unit (ReLU), hyperbolic tangent (tanh), and exponential linear unit (ELU). The ReLU-based model achieved the best performance with an R2 of 0.76, indicating reliable prediction without clear over- or underestimation tendencies. Variants of the model with reduced input variables and adjusted output ranges were also tested for model improvement. These modifications showed reduced error variability and enhanced model stability, albeit with slightly reduced R2 values (approximately 0.72). This approach demonstrates the potential of real-time artificial intelligence (AI)-driven THMs forecasting to support optimized chlorine dosing and regulatory compliance in drinking water treatment facilities. Unlike previous studies focused solely on accuracy, this research emphasizes practical applicability using sensor-acquirable parameters, providing a scalable and real-time decision-support tool for THMs management.