Machine learning methods for anomaly classification in wastewater treatment plants

• DOI: 10.1016/j.jenvman.2023.118594
• Publication Date: 2023-07-18T19:14:58Z
• Authors: Francesca Bellamoli; Mattia Di Iorio; Marco Vian; Farid Melgani
• Keywords: Gradient boosting; Multilayer perceptron; Perceptron

Modern wastewater treatment plants base their biological processes on advanced control systems which ensure compliance with discharge limits and minimize energy consumption responding to information from on-line probes. The correct readings of probes are particularly crucial for intermittent aeration controllers, rely real-time measurements ammonia oxygen in tanks. These data also an important resource developing artificial intelligence algorithms that can identify process or sensor anomalies, thus guiding the choices plant operators automatic controllers. However, using anomaly detection classification is challenging because noisy nature measurements, difficulty obtaining labeled real-plant data, complex interdependent mechanisms govern processes. This work aims at thoroughly exploring performance machine learning methods detecting classifying main anomalies operating aeration. Using oxygen, power a set Italy, we perform both binary multiclass classification, compare them through rigorous validation procedure includes test unknown dataset, proposing new evaluation protocol. explored support vector machine, multilayer perceptron, random forest, two gradient boosting (LightGBM XGBoost). best was achieved ensemble algorithms, up 96% detected 84% 62% classified correctly first second datasets respectively.