Machine Learning and Control for Phosphorous Removal

Abstract

This study addresses the challenge of optimizing control strategies for energy efficiency within legal environmental constraints for the Hias process, Hamar, Norway. The Hias process employs biofilm carriers in both anaerobic and aerobic basins to absorb the nutrients entering the wastewater treatment facility The research is centered around a series of sequential steps beginning with the formulation of research questions that focus on enhancing energy efficiency while adhering to environmental regulations. The methodology encompasses comprehensive data collection, preprocessing, outlier removal, scaling, and the selection of relevant features, followed by the development of dynamic models. Linear regression, deep neural networks, long short-term memory, k-nearest neighbors, gradient boosting regression, random forest, support vector regression, and multilayer perceptron prediction capabilities are examined. Among these, two are selected for further development of control strategies. The core of the research involves selecting, training, and testing machine learning models, designing and testing control strategies integrating with machine learning that can efficiently manage air consumption, thus energy use in an operational setting. Control results are evaluated to ensure they meet the stringent criteria set forth by environmental standards. The paper concludes the discussion, and suggestions for further work, highlighting potential applications of reinforcement learning, and enhancement of analyzer measurement quality to refine the control strategy. This approach not only promises to optimize energy consumption but also ensures compliance with environmental regulations, thereby supporting sustainable operational practices.