Water Network Anomaly Detection Using Hydraulic Model Residuals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting anomalies in water distribution systems, such as leaks and pressure variations, are inefficient due to high false positive rates and imprecise localization, making it difficult to effectively manage water loss and system performance, especially in large-scale systems.
Innovation Solution
A method combining a hydraulic model with statistical processing of time series data and machine learning algorithms to refine control variables and classify anomalies, using sensors to acquire observations and calculate predicted values, thereby improving detection reliability and localization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If sensor-based detection methods are used to automatically detect anomalies, then detection efficiency is improved, but false positive rate increases and localization precision deteriorates
Solution Approach 1:
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 observed sensor data and model predictions indicate anomalies. This intermediary layer enables automatic detection while improving localization precision by identifying which model parameters deviate from expected values.
Solution Approach 2:
The patent replaces direct sensor-based anomaly detection with a model-based approach. Instead of relying solely on sensor thresholds and rules, the system uses a hydraulic model to simulate system behavior and compares it with actual sensor readings. This substitution reduces false positives by accounting for the complex, non-linear behavior of water distribution systems that simple sensor rules cannot capture.
2Measurement precision
If high density of sensors is deployed to locate anomalies precisely, then localization precision is improved, but system cost increases
Solution Approach 1:
The hydraulic model acts as a virtual sensor network that provides localization information without physical sensors. By simulating system behavior and comparing with actual measurements, the model can identify which specific pipes or nodes are experiencing anomalies, achieving precise localization with minimal physical sensors.
Solution Approach 2:
The patent changes the approach from increasing physical sensor density to adjusting model parameters and using statistical analysis of existing sensor data. The system varies model parameters (such as pipe roughness, demand coefficients) and observes how changes affect the match between model predictions and sensor readings, thereby localizing anomalies without adding sensors.
3Reliability
If human inspection is used to detect leaks, then false positive rate is reduced, but detection efficiency and coverage deteriorate
Solution Approach 1:
The hydraulic model serves as an intelligent intermediary that processes sensor data and identifies anomalies with high reliability. The model incorporates expert knowledge about system behavior and uses statistical methods to distinguish true anomalies from normal variations, achieving low false positive rates while operating automatically without human intervention.
Solution Approach 2:
The system implements continuous feedback by comparing model predictions with actual sensor readings and using the discrepancies to detect and localize anomalies. This closed-loop approach enables automatic detection with high reliability, eliminating the need for periodic human inspections while maintaining or improving detection quality.
Data Source
AI summary
A method and a system for detecting anomalies in a water distribution system comprises a network of nodes, and is equipped with sensors of at least water velocity at a subset of the nodes. The water distribution system is modeled by a hydraulic model. The method comprises parametrizing the hydraulic model with initial values of a set of control variables, using the sensors to obtain values of state variables of the network at the nodes, using the hydraulic model to calculate predicted values of state variables, recursively calculating the values of control variables which, applied to the hydraulic model, permit to obtain the predicted values of state variables the closest to the observed values, and classifying nodes of the network based on the values of control variables.