PAVLINA
Controller decisions from PAVLINA. Sacrificial and protective decisions are highlighted — they are the deliberate, physics-informed interventions the baseline model can't reason about.
Neural decision map · live replay
Replaying the seeded decision history in timestamp order. When live decisions begin arriving, the replay will switch to firing in real time.
Decision flow
- Step 01Event detectedGrid or node signal breaches a watch threshold.
- Step 02Constraint evaluatedPhysics rules score feasible actions.
- Step 03Target nodes selectedFleet picked by capability and health.
- Step 04Action executedSacrificial or protective dispatch issued.
- Step 05Impact measuredRealised MW compared to expected.
- Step 06Training log updatedOutcome feeds physics-supervised learning.
15 decisions · live queue
Local 11 kV feeder hit 92% loading as domestic demand peaked; deferring EV charging by 45 min avoids transformer overload while preserving state-of-charge targets.
RoCoF exceeded 0.12 Hz/s after unplanned CCGT trip at Grain; fast-frequency response dispatched from Barking pack ahead of BM instruction.
Day-ahead price signal spiked above £340/MWh and physics model forecast a 15-min reserve shortfall; pre-cool buffer absorbs the shed with no comfort breach.
Conservation residual drifted to 0.7% as Scottish wind output oscillated; smoothing at Trafford holds MW imbalance inside the 1% band without SO intervention.
Sub-station SCADA reported oil temperature rising 1.8 °C/min; taper protects the transformer and keeps 40 of 48 chargers online through the peak.
Midday PV output exceeded local demand by 6 MW with export constrained by the Beauly–Denny corridor; storing energy avoids curtailment payment.
Thermal inertia model shows indoor temperatures drift <0.4 °C over 20 min; delaying compressor start shaves the LV feeder peak coincident with evening cooking load.
Aggregate MW mismatch reached 0.9% after Elexon settlement revision; small distributed trims across South West and Manchester restored balance without BM bids.
Pumped storage ramping created a 3-min reactive dip; halting EV load flattens local voltage until Dinorwig stabilises.
Frequency drifted to 49.87 Hz overnight after Torness output reduction; northern battery response reaches Cheviot boundary faster than southern reserve.
Sub-second imbalance following DC interconnector oscillation; small demand-side trim faster and cheaper than STOR call.
Physics model flagged rising conservation residual trend before it breached threshold; pre-emptive multi-site rebalance keeps grid inside nominal band.
Interconnector switched from import to export within 90 s; local EV taper avoids voltage sag on the Stratford 33 kV ring.
North-West import demand jumped 320 MW in 8 s; Trafford asset closest to the affected GSP and delivers response inside the 1 s FFR envelope.
Physics-supervised model predicted cascading frequency deviation; baseline model missed the constraint violation.
Data provenance
- Mock seeded dataActive (live)
Historical grid, carbon, node telemetry, events, PAVLINA and training series seeded into the demo database.
- Simulated autonomous nodesActive (live)
12 representative nodes stand in for a synthetic fleet of 10,000 distributed assets.
- Live upstream feedsUpstream error
Elexon BMRS (metered generation) + NG ESO carbon + Open-Meteo, refreshed every 20 min via cron. Embedded solar/wind (PV_Live, NESO) disabled pending source-aligned merge. carbonintensity_uk: ok · elexon_demand: ok · elexon_generation: ok · neso_embedded_wind: disabled · openmeteo_weather: error · pvlive_solar: disabled.
- Live control layerPlanned
Future PAVLINA outbound actuation to real DER, EV and BESS operators via signed control APIs.