Physics-constrained interpretable baseline modelling of wastewater lift-pump energy efficiency and its engineering validation.

Influent lift pumps in wastewater treatment plants operate under variable hydraulic conditions, while raw supervisory control and data acquisition (SCADA) records commonly contain transients, asynchronous sensor responses and physically inconsistent measurements, complicating pump-specific energy-performance assessment. This study developed a physics-constrained and interpretable baseline-modelling framework using 1-min SCADA data from a full-scale urban wastewater treatment plant. Steady-state extraction, electrical self-consistency screening, an affinity-informed frequency-power baseline and soft-sensing-based flow-plausibility screening were integrated to construct a reference dataset satisfying predefined physical-consistency criteria. Specific energy productivity (SEP, m3/kWh) was modelled from liquid level and frequency feedback as a compact, leakage-free and physically interpretable representation of pump operating conditions. Feature-ablation analysis demonstrated strong complementary predictive contributions of the two inputs, while comparison with conventional nonlinear response-surface and spline models and machine-learning candidates supported XGBoost as a pragmatic nonlinear baseline. Feature-Space Grid Weighting (FSGW) improved prediction in sparsely sampled operating regions, with the sparse-region benefit persisting across alternative grid resolutions. Chronologically blocked and subsequent out-of-time validation supported short-term temporal predictive stability, while independent pump-specific modelling supported workflow applicability across other same-type units. Model interpretation was used to translate the learned H-f response patterns into practical operating guidance. Under a common liquid-level band, model-guided operation from June to August 2025 was associated with a 4.57 % increase in pump-group SEP and a 4.80 % reduction in specific electricity consumption relative to May. An empirical commissioning-stage healthy-state baseline further supported degradation tracking and descriptive maintenance-related assessment. The framework provides a practical basis for pump-specific energy benchmarking, model-guided operation and maintenance decision support.