Skip to content

Worked example: Warkworth live operations

The Warkworth calibration example shows a model reproducing a measured spill. This one takes the next step: running that model as a live operational twin. It walks the whole Live Ops loop end to end on a real, inflow-driven network, so you see exactly what a utility operator would.

The model here is warkworth-live.inp - the same Warkworth network, driven entirely by [INFLOWS] hydrographs at the pump stations, with no subcatchments. That matters: the weather forecast cannot move it, so the live driver is the measured SCADA inflow. It is the canonical case for the SCADA connector.

  1. Open and save the model. Download warkworth-live.inp and open it (drag the file onto the map, or use Open). Sign in and Save it to the cloud as Warkworth live - the server loop runs cloud projects.

  2. Create a monitor. Open CSO monitoring & alerts, and create a monitor for the Warkworth live project (the defaults are fine).

  3. Add the SCADA connector. On the monitor’s row click SCADA, then Warkworth pilot preset. It fills in: driver Recorded replay, inflow tag LSPS_Inflow → model node LSPS_WW, sample warkworth-scada.csv. Save connector. (See SCADA connector for the full setup.)

  4. Turn on decisions. Set Decisions: on for the monitor, so the hourly loop logs recommendations automatically.

  5. Run it now. Open Live Ops (real-time). The driver badge reads Inflow-driven (needs SCADA) - confirming the loop will use the measured inflow, not the weather. Click Run live decision (server).

The server pulls the recent measured inflow from the replayed feed, bias-corrects the model’s inflow to that level, simulates candidate control actions, and logs a recommendation:

Driver: SCADA inflow

Chosen automatically because the model has no subcatchments. The trigger names the live inflow, e.g. “Live inflow … (…× modelled) → …”.

Recommendation: lower the pump start levels

Start the wet-well pumps sooner to open headroom before the peak - the action that buys down the most simulated flooding.

Click Sync server decisions to pull the logged decision into the ledger. Each decision carries its trigger, the flooding before → after, the exact setpoint changes, a rationale and provenance, and an Accept / Reject / Execute status. Export audit ledger (PDF) for the regulator-ready record; it is saved with the project.

The preset replays a recorded file so you can evaluate the loop on real data with nothing to wire up. In production you swap the recorded driver for a live REST historian on the same monitor - a PI Web API, InfluxDB or MQTT-bridge endpoint - and nothing else changes: the same decision engine, the same ledger, the same hourly loop. See SCADA connector for the URL-template and auth fields.

If your model does have subcatchments, you do not need a SCADA connector - the loop drives it with the weather forecast instead, recommending when predicted rain would flood and staying quiet in dry weather. Buzet, Oldham and Lebanon are that case; open one, create a monitor with Decisions: on (no connector), and Run live decision (server) to see the Weather-forecast driven badge. If a run logs nothing, that is usually the honest answer - no rain in the forecast, or no controllable rule to tune - and the Explain button hands the outcome to the Copilot.

  • The recorded feed is a replay, not a live plant link. It exercises the whole loop faithfully, but a production pilot points the connector at a real historian.
  • The recommendation is advisory. Setpoints are applied in the model (verify by re-running) or sent to a control system only through an explicitly configured, gated webhook.