A multimode data reconciliation method for wastewater treatment processes

Accurate data acquisition in wastewater treatment plants (WWTPs) is crucial for effective pollution mitigation. However, measurement errors and varying operating conditions pose significant challenges to data reliability. This paper proposes a multimode data reconciliation method to address these issues. First, slow feature analysis (SFA) extracts static and dynamic features, capturing subtle process variations that conventional methods overlook. Next, a two-level clustering approach is applied, where self-organizing maps (SOM) perform coarse clustering to reduce noise, and K-means refines cluster boundaries for more precise operating condition classification. Finally, a Correntropy-based robust reconciliation estimator mitigates measurement noise and outliers, enhancing accuracy across diverse operating states. Validated on Benchmark Simulation Model No. 1 (BSM1), the proposed method outperforms state-of-the-art techniques, achieving RMSE and MAPE reductions of at least 79.62 % and 88.44 %, respectively. This framework significantly improves data quality and provides a scalable solution for multimode industrial processes.