UK water utility, Affinity Water, harnessed a new machine learning technique from Tianqi Chen to connect weather and water demand in real-time.
Collating multiple analytical tools
UK water company Affinity Water is sharing an open, artificial intelligence (AI) data tool that has helped it to connect the impact of water use in relation to weather patterns.
Data scientists were able to predict per capita consumption (PCC) in real-time while taking into account seasonal demand and the weather.
Measuring the water balance at a district metered area (DMA) and three areas, the AI system used by the organisation brought together multiple analytical tools.
With multiple developers contributing, the Affinity scientists used “xgboost” a gradient boosting machine learning technique created by Tianqi Chen, assistant professor in the machine learning and computer science department at Carnegie Mellon University.
Real-time consumption data
Completed in four weeks, the model was validated against existing data with a PCC data dashboard then developed to allow access to “real-time data on a daily PCC use”.
“We can now model scenarios, to see the expected PCC based on a normal year and the expected PCC based on changes such as the Covid19 lockdown or other major events,” said chief information officer, Andrew Morris.
Using the AI system, Affinity Water said it is able to distinguish between “previous periods of hot weather and high demand and this year’s increased high demand for water, taking into account Covid19”.
The organisation said for the next stage in the PCC modelling journey, it will model leakage-free DMA consumption on a case-by-case basis, eventually generating an accurate unmeasured consumption model.
Limits on historic PCC measurements
In the past, PCC measurements have only been available for analysis every three months.
Using the new tool, Affinity said it would help to plan for significant weather changes, giving the ability to monitor in real-time the impact of the utility’s campaigns to reduce water consumption.
"As a next step, we have developed a customer level model to predict the water usage movement for each customer at any time,” added Morris.
Affinity is releasing the model openly for all water companies to use freely.
"In the future, we hope there will be more open data and more sharing of proven data models which can ultimately help us all deliver a better service and better value to our customers more quickly,” said the chief information officer.
“We already use similar data science methods to identify interruptions to supply and to identify leakage, and these are already proving highly valuable.”