A New Framework for Energy-optimized Biological Treatment in Wastewater Treatment Plants using Machine Learning Techniques

Wastewater treatment plants (WWTPs) are critical infrastructures that play a vital role in protecting the public health and environment. However, they consume significant amounts of energy whose reduction becomes a decisive challenge. Efficient operation and management of WWTPs require accurate prediction and real-time monitoring of the plant's performance. Previous research tackled specific challenges, such as influent load prediction, effluent quality estimation, and modelling relationships between process parameters, without addressing strategies for energy consumption optimization or reduction. Additionally, reliance on single-plant studies limits the generalizability of their findings. Besides, their models were more complex to implement on wider scales. In this paper, the Stacking Wastewater Optimizer (SWO) is proposed as a novel framework that integrates Ensemble Stacking Learning with different optimization strategies specifically tailored for energy consumption to improve WWTP performance. Unlike previous approaches, SWO not only provides accurate prediction but also deduces actionable insights and optimization recommendations to achieve substantial energy savings. A real dataset generated from the largest WWTP in Africa and Middle East was conducted as a case study. The experimental results show that SWO significantly outperforms existing models, achieving predictive accuracy of influent organic loads with an average improvement of 56.8% and 28% compared to baseline models while removing and transforming outliers respectively. The assessment demonstrates adaptability across various WWTPs operational conditions, with a maximum optimization rate of 10.66%. Additionally, integrated energy optimization techniques, particularly Gradient Descent, achieved substantial reductions in energy consumption ranging from 29.52% to 55.84%, highlighting SWO's potential for enhancing WWTP efficiency on scale.