Skip to main content

Project - Data science

Prediction of drinking water consumption

The challenge

For the water distribution network operator, it is difficult to accurately estimate the water consumption for the days or even the hours to come. This can sometimes cause problems with supply limitations if the required volume of water is not available at a given time. In addition, leaks are not easily detectable on large-scale networks with permanent consumption.

The solution

Using historical water usage data, we were able to apply machine learning to model the water usage for each subnet of the overall distribution system. The resulting model made it possible to predict water consumption for the following 24 hours, with more than 85% of errors below the threshold of 2.5 m³/h (on networks with an average consumption of around 50 m³/h).

The potential gain

The use of such a prediction tool gives rise to multiple applications :

  • The water supply can be controlled in a predictive way based on expected demand ;
  • Unusual consumption is detected, which can trigger alarms ;
  • Leaks in the network can be detected.

Project Details

Client Activity

  • Environment

Solution Brand

  • Endress Hauser
  • Wonderware
  • Historian
  • Python

Nature of the work carried out

  • Load prediction


  • Data Science Offline and Optimisation

Notre ThinManager-Integrator Certified

1.22 Mb – file pdf
Download this file

Other references More to be discovered

Let's stay in touch

Subscribe to our newsletter