1. The trip
Each synchronous generator has a heavy rotor that turns in step with the grid frequency. When supply falls below demand, the grid takes the difference from the kinetic energy in these rotors, so they slow down. The initial rate of fall (RoCoF, rate of change of frequency) depends on the size of the loss and the inertia online.
Move H down towards 1 s. The same loss now gives a faster initial fall.
2. Governors push back
Governors on thermal and hydro units increase power when frequency falls. Droop sets how much extra power they ask for. At 5% droop, a 0.025 Hz fall asks for an extra 1% of system load. A 4 s time constant delays that response. Headroom sets the most power the governors can add.
3. Batteries act fast
A battery can respond faster than a turbine. In this model, the battery measures frequency after a short delay. Its output grows with the fall and reaches full power at a 0.5 Hz deviation. Make the delay shorter or the battery larger, then watch the lowest frequency change. “Battery size” is a power limit, not stored energy.
Evidence and assumptions
4. One real trip, not a replay
On 9 October 2019, the largest generating unit in the National Electricity Market tripped from 748 MW. The unit is at Kogan Creek in Queensland. AEMO reports that frequency fell to 49.61 Hz. The South Australia virtual power plant (VPP) detected the fall and injected power to help recovery.
The fixed comparison below uses the same 748 MW loss on an assumed 25 GW system base. That is 2.992% of load. The other inputs are assumptions too: H = 4 s, droop 5%, governor time constant 4 s, headroom 10% and load damping 1. The model starts at 50 Hz. It has no battery, because the report does not give a total battery power for this model.
- Reported event minimum
- 49.61 Hz
- Model minimum
- Model minus reported minimum
Where the model differs
- One frequency. The model gives one frequency for the whole system. It has no transmission network, regional oscillations, voltage or separate machines. It cannot reproduce local measurements or network constraints.
- Assumed response. Real units have different inertia, governor settings, deadbands and available reserves. The system base and the response constants above are assumptions, not event measurements.
- No event battery response. AEMO documents a VPP response. This comparison leaves it out. A guessed battery size would not show the real share of that response.
- No restoration or protection. The model has no secondary frequency control, no battery energy limit and no load shedding. When frequency crosses a threshold, only the warning text changes. The model can settle below 50 Hz.
What this shows. The model explains how power imbalance, inertia and response affect frequency. In this case its minimum is higher than the real one, so it does not validate the event response. A match on one number would not prove the timing or the cause either. A reconstruction needs measured system conditions, response data and a frequency time series.