Product(s):	WaterSight
Version(s):	10.00.
Area:	Documentation

 

WaterSight enables near real time anomaly detection, by automatically triggering alerts whenever real sensor or zone measurements are outside the expected patterns or other user defined thresholds. The methodology used follows the below main steps:

1. Generate the real (resampled) time series

• Raw data is collected
• Raw data gets cleaned up and resampled in the following intervals:
  • 15 min resampled time series
  • 1 hour resampled time series
  • Daily average resampled time series
  • Daily minimum resampled time series
  • Daily maximum resampled time series

Those real (resampled) time series are represented in the graphs by a back line filled with black, yellow and red dots.

Below is shown a real resampled 15 minutes time series (black color) with the raw data at green. It is possible to observe some of the cleanup that is automatically done by the software, where some bad data was removed.

Figure 1 - Raw data at green and real 15 minutes time series (processed) at black.

2. Generate the patterns and confidence bands

• Machine learning together with statistical analysis is applied to the processed resampled time series to generate the patterns and confidence bands.

Patterns and confidence bands are represented in the graphs by a grey band.

Whenever real data (re-sampled) goes outside the pattern confidence bands (above percentile 95 or below percentile 5 of the pattern), an outlier is identified and the real value is color coded with a red circle. For more information about patterns or forecasts, please click here.

Figure 2 - Real resampled time series (back line with black, yellow and red dots) and pattern confidence bands (grey band).

3. Alert generation

Alerts can be automatically generated by the software based on user defined triggers or can be manually created directly from the graphs. There are four type of alerts that can be configured:

• Pattern based: are evaluated using the pattern as a reference. Alert gets triggered if values are outside the pattern confidence bands (values are color coded red), for a user defined minimum duration
• Absolute based: are evaluated directly by comparing if real time series values are above or below the user defined threshold.
• No data: No Data alert is triggered whenever there is no data at all for the specified sensor for the minimum duration defined by the user.
• Flat reading: A Flat Reading alert is triggered whenever there are a specified number of consecutive flat readings, greater than the minimum duration defined by the user

Alerts can be triggered to sensor or to zones. 

In case the system is well sectorized in small DMAs (district metered areas), flow measurements exist at all zones inputs and outputs and there are several pressure measurements available inside the DMA, then all those measurements can be used to better geolocate the potential leak inside the zone, by: 1) configuring a flow increase alert for each zone, based on pattern type alarms and 2) configuring pressure decrease alerts for each pressure sensor, based on pattern type alarms. Whenever a flow alert is triggered for a zone, this can also generate a pressure decrease somewhere inside the zone near the burst location.

For more detail about alerts configuration, please click here.

Figure 2 - Example of an alert generated based on pattern deviation. 

4. Event Management

Generated alerts get automatically available in the Active Event List. These events should represent anomalies that are happening or that occurred in the network (such as pipe breaks, zone changes, fire, flushing, meter failures) and should be managed by the user. 

Figure 3 - Active Event List

 

See also

OpenFlows WaterSight TechNotes and FAQ's

WaterSight Learning Resources Guide