Abstract
Predicting water quality is essential for effective environmental management and pollution control. Dissolved oxygen (DO), one of key water quality parameters, plays a vital role in biological wastewater treatment [1]. This study aims to forecast DO levels in activated sludge tanks of an oil refinery’s wastewater treatment plant (WWTP). Proper oxygen concentration is critical for microbial activity, as inadequate levels can disrupt the biological breakdown of pollutants. The objective is to develop predictive models to identify operational risks early, enhancing treatment efficiency and optimizing resources like chemicals, bacterial cultures, and aeration systems. Additionally, the study aims to provide early warnings to operators, minimizing reliance on laboratory tests and ensuring optimal conditions for bacteria, leading to better operational performance, cost reduction, and improved water quality ultimately promoting sustainable wastewater treatment. Various deep learning models, including Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), and Long-Short Term Memory (LSTM), were applied to a two-year real-time dataset with 24 features and 8-hour intervals. Model performance was assessed using process data for training, validation, and testing. The results reveal that GRU-based models effectively predict DO concentration in sludge tanks, achieving an R² of 0.7, MSE of 0.01, and MAE of 0.07. These soft sensors demonstrate strong potential for system control.