Forecasting water flows across distribution networks

Machine-learning models estimate future water flow by station and hour to support planning in distribution networks.

Challenge

Network operators need short-term forecasts of water demand at different stations to plan resources, detect unusual behavior and operate the network with more confidence.

Data

The forecasting service used historical flow data together with weather information, calendar variables and holidays. The system organized predictions by station and forecast horizon.

Solution

Artelnics designed a forecasting workflow that trains predictive models for each station and hour, producing multi-day flow predictions that can be consumed by operational systems.

Results

The system provides structured forecasts for network planning and gives operators a quantitative view of expected demand before it happens.

Illustrations