An event detection case study
Yorkshire Water provides water and waste-water services to more than five million customers in the North of England.
The company produces over 1 billion litres of high-quality drinking water every day and delivers this through over 30,000 km of distribution network. Continuously providing this service with minimal interruption is a challenge that Yorkshire has historically performed well in. However, they recognised that in order to continue to meet challenging targets over the next 5 years more needed to be done.
Until recently, confirmation of bursts usually only occurred after customers notified Yorkshire about them. And it was understood that there were also bursts that could go unnoticed for quite some time. Yorkshire Water wanted a solution that would help to identify visible and invisible bursts earlier so that they could resolve them more quickly – minimising water lost from the network and the effect on customers.
Yorkshire Water embarked on a large-scale “Intervention Enabled Networks” project, and as part of this they carried out a trial with a number of providers to evaluate the accuracy of their anomaly detection algorithms. For the first 6 months 100 locations were monitored 24 hours a day using both flow and pressure data, increasing to 400 locations by the end of 12 months. In the second year this was expanded to 1,400 locations from year 2.
The evaluation was extremely thorough and scientific. It was based on accuracy, cost and the ability to scale. Accuracy was assessed by the percentage of bursts for which there was a Detected Event notification, the percentage of these that was detected before customer contact, and the percentage of notifications that related to a burst.
Artesia’s eVader algorithm was assessed to be the most accurate.
Artesia then partnered with i2O to build the eVader algorithm into a robust, reliable and secure software service as part of a wider network monitoring and advanced analytics solution.
This was then rolled out to the whole network. Currently, data is ingested from more than 3,000 loggers right across the existing Yorkshire Water network every 5 minutes and prepared for analysis.
As new data is received, the eVader algorithm runs and checks if it is normal or not, and categories any abnormalities that are identified. It also updates its model of what ‘normal’ is. Between the hours of 10pm and 6am, Yorkshire Water receives emails of Events Detected from the iNet system as soon as they are identified. The email includes information about the location of the issue as well as the size of the anomaly, the percentage change from normal, and the categorisation.
This data is used by the control room, the leakage team and by analysts to enable quick operational decisions to be made before customers make contact.
- Yorkshire Water saved more than 300 l/s over a single 12-week period with Event Detection ranging from 5 l/s to 27 l/s.
- A large proportion of leaks are now identified by Event Detection; previously a number of different systems were used.
“It is a valuable and useful tool for the control room…eVader is a much better system than [previous system]; it’s smarter, gives more information and automatically updates. All of the other engineers would agree.” – Duty Operations
“From a leakage perspective, it’s great. We usually have at least one a night that we will [deal with]. It works.” – Leakage Operations Manager
“Event classification is a really useful tool…it only takes around 10 to 15 minutes to go through the whole summary email. This often picks up leaks not detected by our customers.” – Data Analyst