An evolving fuzzy framework for dynamic operational regime-aware efficiency monitoring in wastewater treatment plants

This paper introduces the Evolving Multi-layer Stream Mapping (E-MSM) framework, an evolving fuzzy approach for efficiency monitoring of industrial processes under varying operational regimes. The proposed method is demonstrated in this work through its application to Wastewater Treatment Plants (WWTPs), while its generic data-driven architecture may be transferable to other industrial systems. E-MSM extends the Multi-layer Stream Mapping (MSM) methodology by dynamically identifying operational regimes and updating fuzzy rules from process data to estimate reference values. The approach is evaluated on a WWTP under drought, rainfall, and storm conditions using the Benchmark Simulation Model No. 2 (BSM2) and real SCADA data. Results demonstrate that E-MSM provides a more consistent efficiency assessment across the primary clarifier, biological reactor, and secondary clarifier. Specifically, E-MSM achieved an average reduction of approximately 79% in assessment discrepancy compared to the conventional MSM between statistically similar periods, with variations of 0.89% for the primary clarifier, 0.50% and 1.50% for the biological reactor sections, and 0.01% for the secondary clarifier, whereas MSM exhibited discrepancies of 4.96%, 8.09%, 12.76%, and 0.02%, respectively. Validation using real SCADA data confirmed these findings, with E-MSM maintaining a stable 76% efficiency assessment (discrepancy of 0.48%) compared to MSM’s 8.34% discrepancy. Across all case studies, E-MSM performed comparably to or better than established adaptive baselines (RLS and eTS). Furthermore, a comparative analysis on the real-world dataset confirmed that the proposed rule-merging mechanism maintains model compactness without degrading assessment consistency, demonstrating robust performance under real industrial conditions. By enabling adaptive performance monitoring, the proposed framework supports more informed decision-making, facilitates the identification of operational bottlenecks, and supports performance improvement in dynamically evolving systems.