Name
Evidence-bound AI for wastewater digital twins: KPI integration and action to cost models for energy-aware operations
Authors
Christian Miguel Pacheco Rodríguez, Juan Pedro Cortés Pérez and Christian Andrés Yanza Pérez, University of Extremadura, Spain
Description

Operational Digital Twin and AI initiatives in wastewater treatment are often constrained by implicit data assumptions that prevent auditable optimisation decisions. We describe an evidence-governed workflow in which contract-aligned signals are declared and QA/QC gates are enforced before KPI computation or optimisation is attempted. Over a representative 180-day operational window, a master daily dataset is assembled from SCADA exports (flow, actuator runtimes and setpoints) and online water-quality sensors. Data acceptance applies engineering-range screening and MAD-based spike filtering using a five-hour window; downstream analytics proceed only when availability meets a configurable seventy-five per cent gate, with flow used as a hydraulic witness variable for cross-stream consistency. High frequency hood off-gas telemetry is filtered and aggregated to one-minute resolution prior to daily alignment, stabilising energy-relevant patterns while preserving traceability. The resulting information substrate enables interpretable KPI integration and supports action-to-cost models that map actuator setpoints to energy use, expressed in kilowatt-hours, and operational expenditure, expressed as monetary cost, supporting cost–energy optimisation under real-world coverage constraints.

Track
Posters