Water scarcity is a global challenge, and one of the promising ways to mitigate the water resource crisis is via wastewater reclamation. Reclaimed water can generate non-potable water to substitute the use of drinking water for irrigation or industrial processes. Water quality and aesthetics are the primary concerns in reclaimed water since undertreated water can pose health risks, and the unpleasant colour is likely to induce public misgiving. Ammoniacal nitrogen (NH₃-N) and colour substances exist in the reclaimed water and can severely affect the reclaimed water quality in different ways. Chlorine is commonly used for reclaimed water disinfection and requires precise dosing to satisfy endorsed quality standards. However, NH₃-N consumes chlorine and affects the chlorine dosing. Colou...[ Read more ]

Water scarcity is a global challenge, and one of the promising ways to mitigate the water resource crisis is via wastewater reclamation. Reclaimed water can generate non-potable water to substitute the use of drinking water for irrigation or industrial processes. Water quality and aesthetics are the primary concerns in reclaimed water since undertreated water can pose health risks, and the unpleasant colour is likely to induce public misgiving. Ammoniacal nitrogen (NH₃-N) and colour substances exist in the reclaimed water and can severely affect the reclaimed water quality in different ways. Chlorine is commonly used for reclaimed water disinfection and requires precise dosing to satisfy endorsed quality standards. However, NH₃-N consumes chlorine and affects the chlorine dosing. Colour substances do not consume chlorine, but it requires additional efforts and strategies to remove them from the reclaimed water. Therefore, the on-line monitoring of NH₃-N concentrations and colour levels are usually practised in reclaimed water facilities to assist in the removal of both substances. However, the conventional on-line analyzers are wet-chemistry-based, and the measurement takes time. The limitation creates a potential issue: there may not be sufficient time for the downstream chlorine dosing system to respond to sudden surges in colour and NH₃-N levels. To tackle this challenge, this thesis work developed time-series models based on machine learning to forecast the NH₃-N concentrations and colour levels in the reclaimed water three hours into the future. For the training dataset, the NH₃-N and colour data were collected by an on-line analyzer and a customized auto-sampling spectrophotometer, respectively. Both are installed in a reclaimed water treatment facility in Hong Kong. Baseline models for forecasting NH₃-N concentrations and colour levels were first developed with five machine learning algorithms. Long Short-Term Memory (LSTM) was found to be the most effective algorithm, with the lowest MSE values of 0.0405 and 0.0148 for NH₃-N and colour forecasting models, respectively. In the training processes, novel data pre-processing methods and feature engineering techniques were implemented to enhance forecasting model performance. The data pre-processing methods were proved to enhance the quality of training datasets and improve the performance of NH₃-N and colour forecasting models by reducing the MSE values by 4.2% and 8.1%. The feature engineering results supported that the daily fluctuations in NH₃-N and colour have correlations with the urban water consumption patterns. This finding further enhanced the NH₃-N and colour forecasting model performance by reducing MSE by 8.9% and 28.6% compared to baseline models. The established models can be used to assist the disinfection control strategies based on the model predictions using traditional process control systems. This research offers novel methods and feature engineering techniques for NH₃-N concentrations and colour levels forecasting in reclaimed water for treatment optimization.