Wastewater Watch
This benchmark uses a municipal wastewater simulator, so it is a demonstration of method and not oil and gas field work. It asks whether existing measurements can give useful warning of an effluent excursion.
Simulated data. No real plant results. Open the interactive reportWhat we found
The black hybrid beat persistence on six-hour rain forecasts, at 1.06 against 1.41 mg N/L. On six-hour storm cases the two were nearly tied.
The grey model struggled under changed conditions and over longer horizons. Simply assuming nothing changes, known as persistence, remained hard to beat.
| Method | Rain | Storm |
|---|---|---|
| Black hybrid | 1.06 | 1.51 |
| Persistence | 1.41 | 1.52 |
| Direct ML | 1.51 | 2.51 |
| Local trend | 1.89 | 1.97 |
| White hybrid | 2.94 | 4.30 |
| Grey hybrid | 3.08 | 4.58 |
| Fixed physics | 3.33 | 4.52 |
How it was tested
Seven methods ran on 28 custom scenarios of the open-loop BSM1 simulator, with six-hour and 24-hour forecasts. The forecasts used no future control data.
What it doesn’t show
Four test scenarios per regime can’t support broad conclusions. The scenarios come from one simulator set-up, and the 4 mg N/L alert level is illustrative, not a regulatory limit.
It claims no real-plant savings, compliance performance or automatic treatment control.