Effective anomaly management of wastewater treatment plants (WWTPs) is crucial for environmental conservation and public health security. Traditional monitoring methods often struggle with challenges such as multivariate coupling, nonlinear dynamics, and external interferences inherent in wastewater treatment processes, which has driven growing interest towards artificial intelligence (AI)-based anomaly management solutions. This paper critically reviews recent advancements in AI-based anomaly management strategies for WWTPs, emphasizing three integral aspects: sensor data quality control and self-calibration, early anomaly detection and diagnosis, and fault-tolerant control and resilience enhancement. Systematic comparisons are made among supervised, unsupervised, and transfer learning methods, highlighting the strengths and weaknesses of deep learning, ensemble learning, and intelligent optimization algorithms in addressing practical engineering issues such as sensor noise, multimodal data distributions, imbalanced datasets, and limited cross-facility generalizability. The review further highlights real-world performance metrics beyond conventional accuracy, such as application scalability, anomaly detection timeliness, and technological adaptability. Key findings reveal research gaps hindering for the progress and application of AI-based anomaly management approaches in model interpretability, computational intensity, data quality controls, cross-facility generalization, and cost-effectiveness. More importantly, future research directions cover adaptive learning techniques, explainable AI, integration of AI with digital twin platforms, lightweight infrastructures for real-time edge computing, and environmental and economic analysis of AI deployments in WWTPs.
Yang et al. (Thu,) studied this question.