|M.Sc Student||Jad Naamnih|
|Subject||Events Detection in Water Distribution Systems|
|Department||Department of Civil and Environmental Engineering||Supervisor||Full Professor Ostfeld Avi|
|Full Thesis text|
This study presents a model for detecting, classifying, and sourcing identification of abnormal events in water distribution systems. The developed method utilizes a Minimum Volume Ellipsoid (MVE), clustered into operation modes, without the need to divide the system into District Metered Areas (DMAs).
The minimum volume ellipsoid (MVE) estimator is the smallest volume ellipsoid that covers m of n observations. The MVE could be found by a resampling algorithm.
An operation mode is determined by the controlled parameters distributed in the system, such as pumps, valves, pressure regulators, either open/on or closed/off.
The model consists three main steps: event detection step, event classification step and event sourcing identification step.
Observations consist the input data of the model which are received from several sensors and meters distributed in the system. Sensors and meters can observe pressure, water flow, water level in tanks, and chlorine concentration in the water. The input consists yearlong measurements with a recording time step of five minutes.
The main objective of the first step is to construct a suitable MVE considering the input data and detect abnormal events. Measurements fall outside the MVE are defined as outliers. Three consecutive hours of outliers are defined as abnormal event.
However, it may be inaccurate to apply all observations at one operation mode in the same ellipsoid. Clustering the input data according to operation modes, then constructing a suitable MVE for each operation mode is a compulsory step to improve the model accuracy and decrease errors and false alarms.
The second step is the classification step, which consists of the categorization of anomalies between physical events, such as: alteration in pump operation, leakage, closing or opening valves or pipes, substantial changes of water level in tanks etc. and water quality events (e.g., contamination).
The third step is a sourcing identification procedure which is developed to restrict the sources of detected events to three possible options. A simulation of all possible abnormal events in the water system is applied to cover anomalies as much as possible. Subsequently, the simulated events are saved as anomalies “dataset bank”. Detected and classified events are compared to all saved events in the anomalies bank, and the three closest events are selected as the possible causing sources.
Since the input data is the only source of the model, a proper placement of meters should be applied to ensure a high representation of the water distribution system components and operation. A satisfying representation of system components and operation can improve real-time anomaly detection and management costs.
The developed model is demonstrated through two water distribution systems example applications. Good results were obtained, especially for the detection and classification steps, with no recorded false alarms. The first water distribution system example application describes the results of the detection and classification step, and in addition it illustrates the importance of clustering the MVE to operation modes in a 3-D figure. The second water distribution system example application is more complex, demonstrating the model outcome results of all model steps.