Hydraulic Model-Based Water Network Anomaly Localization

The patent introduces a hydraulic model as an intermediary between sensor observations and anomaly detection. The model simulates expected system behavior under normal conditions, and deviations between actual sensor data and model predictions trigger anomaly alerts. This intermediary layer filters out false positives by comparing against physically consistent expectations rather than relying on simple threshold-based detection.

The patent replaces simple mechanical threshold-based detection with a physics-based hydraulic model that incorporates fluid dynamics equations. This substitution allows the system to understand the physical relationships in the water distribution network, enabling more accurate anomaly detection and localization by considering how pressure, flow, and demand interact throughout the system.

The patent segments the water distribution network into hydraulic zones or districts based on the hydraulic model structure. By analyzing anomalies at the zone level first and then progressively refining to specific nodes or pipes within affected zones, the system achieves precise localization without requiring sensors at every node. This hierarchical segmentation reduces the overall sensor density needed while maintaining high localization precision.

The patent adds the hydraulic model simulation dimension to the physical sensor measurement dimension. By comparing actual sensor readings with model-predicted values across the entire network, the system can localize anomalies in areas without direct sensor coverage by inferring from measurements at nearby nodes through the hydraulic relationships defined in the model.

The patent performs preliminary calibration of the hydraulic model using historical data before deployment. During this calibration phase, the model learns the normal operational characteristics and variability of the specific water distribution network. This preliminary action establishes a baseline of expected behavior that makes the system robust when detecting anomalies, as it can distinguish between normal variations and actual leaks based on the calibrated model rather than generic a priori thresholds.

The patent implements a feedback mechanism where the hydraulic model continuously compares predicted versus actual sensor measurements and uses this information to refine anomaly detection. The model is updated with new data over time, allowing it to adapt to changing network conditions and improve its detection accuracy. This feedback loop replaces static a priori thresholds with dynamic, data-driven detection criteria that are more robust to varying operational conditions.