Dosing forecasts for a water treatment plant
A forecast built from a water treatment plant's process data over time, to support its chemical dosing decisions.
Engineering teams
Project snapshot
- Industry
- Engineering teams
- Input modalities
- Time-series process data

The situation
The plant team worked out chemical dosing by reading the process data themselves. That was slow, and it didn't say ahead of time what the plant would need.
What I recommended and set up
I built the forecast to support the team's dosing decisions, not to make them for it.
What changed after delivery
- The team sees what the plant will need before it needs it.
- Dosing decisions rest on a forecast, not only on reading the data by hand.
The result
- An average forecast error of 5.9% on chemical dosing.
How it works
- The model learns from the plant's process history and forecasts the dosing it will need.
- Feature analysis shows which inputs drive each forecast, so the plant team can see why it says what it says.
- The team uses the forecast in its day-to-day dosing decisions.
- The 5.9% is the mean absolute percentage error: on average, the forecast was 5.9% off the actual reading.