| Product(s): | WaterSight |
| Version(s): | 10.00. |
| Area: | Documentation |
Overview
WaterSight automatically calculates patterns and forecasts for each sensor and zone using machine learning together with statistical analysis on top of the collected raw sensor data. By default also a distinction between weekdays and weekends is also done.
Patterns and confidence bands are represented in the graphs by a grey band, while real time measurements are represented by a back line filled with black, yellow and red dots. Whenever real data 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 calculation click here.
Customizing Patterns
Users can confidently rely on the automatic patterns calculated by the solution, however they can also refine/customize the patterns in case that may be needed, for example to improve even more its accuracy and for example minimizing the generation of false positive alerts. The user has the flexibility to:
- Define grouping days
- Add special days
- Increase/decrease pattern offset (move pattern up or down) and increase patterns band width
- Select sensors or zones that might not have any pattern at all
Grouping days
User can customize how patterns are computed in WaterSight by creating different group days. The default is a weekday pattern (that uses past days from Monday to Friday to calculate a pattern), a Saturday pattern (that uses only past Saturdays to calculate a pattern) and a Sunday pattern (that uses only past Sundays to calculate a pattern). However there might exist some cases where there are significant differences in behaviour between each day of the week, and in that case the user can decide to create a different pattern for each week day or grouping some weeks days together.
The definition and configuration of the pattern groups is done in the administration menu >> Settings >> Pattern. After configuring the pattern groups, the admin user needs then to define the corresponding sensors or zones that will use those new pattern groups. This is done through the administration menu, in the sensors and zones configuration page.
For more information about pattern groups please go to the Settings help page, under the Pattern section of the article.
Special days
Some special days such as national holydays or Christmas day have a very specifically and typical behaviour that are very different from a normal week day. It is possible to consider these special days when calculating patterns. By default a national holiday will be treated similar as a Sunday (as people do not work, national holidays tend to be more similar to Sundays than to a normal working week day). When special days are added they will be automatically considered and applied in the pattern calculations of all sensors and zones.
For more information about special days please go to the Settings help page, under the Special Days section of the article.
Pattern Offset and Buffer
Have you ever wondered if you could control or adjust the pattern percentiles? That is possible by adjusting the pattern offset (move pattern up or down) and the pattern buffer (increase pattern bands width). This might be particularly useful when dealing with special periods such as Christmas holydays, Easter holydays, school breaks, or even seasonality transition periods (transitions between summer/winter periods) where the default pattern calculation method (that is based on historical data) may not fit so well. In those cases the user can use its own experience to adjust how the patterns or demands forecasts are expected to increase or decrease (pattern offset) and the uncertainty associated with those user predictions (by increasing the pattern buffer).
The definition and configuration of the pattern bands/percentiles is done in the administration menu >> Settings >> Special Periods. After configuring the special periods (including the definition of a period where those user settings will be applied), the admin user needs then to define the corresponding sensors or zones that will use those new pattern settings. This is done through the administration menu, in the sensors and zones configuration page.
For more information about adjusting pattern bands/percentiles please go to the Settings help page, under the Special Periods section of the article.
No Pattern
Not all sensors and zones will have patterns, as some measurements will be unpredictable and with different behaviour every hour and day. This usually happens for pumps flow, pump status, flow meters measuring entrance of tanks, etc. For measurements that do not follow any trend or pattern, patterns should not be displayed as will not be of value. To do this, the user should go to the administration, to the sensors and zones configuration, and in the pattern field select "None". After that, automatically all graphs will be updated in WaterSight and those sensors or zones will not have anymore any pattern being displayed in the graphs.
See also
Patterns and Demand Forecast Calculation
Pattern Settings (Pattern Groups, Special Days and Special Periods)
OpenFlows WaterSight TechNotes and FAQ's