Demand Forecast

DemandLSTM P10 / P50 / P90 electricity demand forecast, at the model's own native cadence and horizon for the selected region.

as_of: —

model: —

Region

Peak demand

—

Trough

—

Mean

—

Total energy

—

sum of median × interval

Uncertainty

—

A real walk-forward backtest shows every NEM region currently underperforms its own seasonal-naive baseline with the currently-served model — TAS1 most notably (22.9% vs 6.14% baseline). NSW1/QLD1/VIC1/SA1 are closer (1.2x-1.5x the baseline); select a region above for its own real numbers.

NEM — next —

Loading forecast…

Model info

No model metadata available.

Endpoint

Served live by forecast-api (FastAPI + PyTorch LSTM, MLflow Registry). horizon/interval always reflect the model's real native cadence — requesting a different one isn't supported yet, so there's no horizon selector here.

GET /v1/forecast?region=NEM
→ { region, model, horizon, interval, points: [{ ts, p10, p50, p90 }] }

Now (sparkline)

No data available.

Actual vs Predicted — NEM

A real walk-forward re-forecast of the currently-served model: what it actually would have predicted (P10–P50–P90) at each of several real points over the last 7 days, against real actual demand for those same real timestamps — not the live forecast repositioned to look retrospective.

Running real walk-forward backtest…