LOR1: reserve getting low. LOR2: reserve low, so losing one large unit could mean load shedding. LOR3: load shedding possible or under way.
Loading…
Grid Desk watches Australia's National Electricity Market (NEM) the way an operations desk would: demand, prices, generation and interconnector flows every 5 minutes, AEMO's reserve notices, and units dropping out. For each region it learns what demand normally looks like for the time of day, day of the week, temperature and rooftop sunshine, and says when today departs from that, and why.
The data-centre tab shows how much electricity data centres use and plan to use, as honestly as public data allows. It keeps three kinds of number apart: published (from AEMO's reports), announced (operators' stated capacity, a maximum) and estimated (our own inference, always with a range). It never claims to measure one facility's live use and never blames a facility for a grid event.
It is an independent project, not affiliated with AEMO or any operator, and nothing here is trading or investment advice.
Grid demand is AEMO’s dispatch total demand for each region: what the grid’s scheduled and semi-scheduled generators and interconnectors supply each 5 minutes. Rooftop solar isn’t in it, because it’s used behind the meter; AEMO estimates it separately (half-hourly), and adding it back gives an approximate underlying demand. Totals are sums of the regional figures as shown.
Expected demand (learned normal) is fitted to each region’s own last 28 days: daily shape for weekdays and weekends, temperature (heating and cooling), rooftop solar once its history is long enough, and a slow trend. The fit down-weights unusual days so they don’t become the normal. The 90% band is the range the model’s errors stayed within on 90% of the intervals it was fitted to. σ (sigma) is the typical size of those errors; a demand alert needs three intervals in a row at least 3σ from expected. Typical is what the model expects for that time on the same kind of day with ordinary weather and sunshine (the median for that hour), so the gap between typical and expected is what today’s weather explains.
Reserve comes from AEMO’s Lack of Reserve notices. Forecast notices are for a stated future period, shown with each one; a condition drops off when AEMO cancels it or its period ends. Price spikes are dispatch prices at or above the alert threshold. Possible trips are units above 200 MW falling by at least 80% within one interval (batteries excluded, since they stop by design).
Freshness: the self-hosted console fetches every 5 minutes and says when data is more than 15 minutes old. The public demo is a snapshot rebuilt every 30–60 minutes by GitHub Actions, so it is always somewhat behind; the bar at the top says how far.
The morning summary covers the previous 24 hours: each region’s peak demand, average and highest price, reserve notices and possible trips.