Importing measured electricity data

Where to get the file

Most Australian smart-meter setups expose 30-minute interval data through one of these channels:

• Your distributor’s portal — log into your network distributor (Powercor / Jemena / AusNet / Energex / Endeavour / Ausgrid / Essential Energy) and look for “Download my data” or “Interval data”. The export is usually NEM12 or a simplified per-day CSV.
• Your retailer’s portal or app — AGL, Origin, EnergyAustralia, Powershop, Amber and most smaller retailers all support an interval-data export, often labelled “NEM12” or “usage data”.
• A monitoring service — Powerpal, Energy Locals, Wattwatchers, etc. expose CSV downloads of the channels they track.

Whichever source you use, exporting once a quarter is enough for backtesting; the more recent your data, the better the comparison against the forecast.

Supported formats

NEM12 (preferred)

The AEMO industry-standard CSV used across the National Electricity Market. solarprod parses the 200- and 300- records to extract your NMI, interval length (5 / 15 / 30 / 60 min), and per-channel values. Suffix codes that start with B (B1 / B2) are treated as exports; E (E1 / E2) and others as imports.

Some distributor portals (United Energy, Powercor, CitiPower) export NEM12 without the surrounding 100 header and 900 footer rows — the file just opens with a 200,... NMI row. solarprod accepts that shape directly; no manual editing is needed.

Simple CSV

A header row with at least the column timestampplus one of import_kwh, export_kwh, or solar_kwh. Timestamps may be ISO-8601 (2026-05-02T18:30:00) or 2026-05-02 18:30. solarprod infers the interval length from the first two timestamps.

timestamp,import_kwh,export_kwh
2026-05-01 00:00,0.05,0.00
2026-05-01 00:30,0.04,0.00
...

What gets used vs discarded

When the daily-usage predictor trains on your upload, it automatically scrubs two kinds of rows that would otherwise mislead the model:

• Holiday clusters — runs of two or more consecutive days where total usage is well below your usual rolling-month median. A typical “away on holiday” pattern.
• Post-install net-meter data — if the predictor detects a sustained drop in import that looks like a solar or battery install, it discards the data from the install onward. Once panels are on, the smart meter only records what’s left over after self-consumption, so it’s no longer a record of your household’s full electricity demand. The pre-install months stay in the training set because that’s when the meter saw gross usage.

Each prediction response includes a notesblock listing how many days were used, how many were excluded, and the regime-change cutoff (if any). Going forward, an inverter API integration will let solarprod read solar production directly and reconstruct gross usage from net-meter readings, so post-install data stays usable.

Caps and privacy

Files are capped at 5 MB and 20 datasets per account (about 5 years of quarterly NEM12 exports). Datasets are stored against your account, only ever served back to you, and removed when you delete the dataset or close your account. Nothing is shared with third parties.