On 28 September 2016, South Australia lost every one of its large synchronous generators within about seven seconds, and the state blacked out. AEMO’s final report into that event, Black System South Australia 28 September 2016, is still the document every system engineer in this country has read at least twice. It’s the reason we’re now having a very technical, slightly overdue conversation about grid-forming inverters, and why AEMO keeps putting the phrase in board papers instead of leaving it in engineering appendices.
Let’s be careful with that number, because it gets mangled in almost every retelling: the state didn’t lose 100% of its generation capacity, it lost the specific combination of wind output and transmission line trips that took out the last synchronous machines holding the system together. What matters for this piece is what happened next, which is that with no spinning steel left on the SA grid, there was nothing left to define what “50 hertz” even meant for long enough to ride through the disturbance. That’s the gap grid-forming inverters are built to close.
The thing synchronous machines used to do for free #
A coal, gas or hydro turbine is a lump of spinning steel connected to the grid. When something goes wrong nearby, a fault, a line trip, a sudden loss of load, that spinning mass physically resists the change in frequency. It has inertia, in the literal Newtonian sense, and it also naturally holds up voltage because it’s a rotating magnetic field, not an electronic circuit deciding what to output. Every synchronous generator on the system is, without any control software at all, quietly agreeing with every other one on what the grid’s frequency and voltage should be at any instant.
Take enough of those machines off the system, replace them with wind and solar and batteries, and you still have plenty of power (megawatts) and plenty of energy (megawatt-hours) available. What you don’t automatically have is anything defining the reference wave that all that equipment is meant to synchronise to. That’s not a capacity problem. It’s a completely different kind of problem, and conflating the two is probably the single most common mistake I see in coverage of this issue.
Grid-following versus grid-forming: the actual difference #
Almost every solar farm and battery built in the NEM over the past decade uses a grid-following inverter. It measures the grid’s existing voltage and frequency, locks onto it with a phase-locked loop, and injects current in step with what it sees. It’s a follower. If the reference disappears, or gets weak and wobbly, a grid-following inverter can lose its lock and trip off, which is precisely what compounds a disturbance rather than helping ride through one.
A grid-forming inverter does something closer to what the old spinning machines did. It behaves as a voltage source rather than a current source: it sets a voltage waveform of its own and lets current flow according to what the rest of the network needs, the same logical role a synchronous generator plays. It can support frequency and voltage through a disturbance rather than depending on someone else’s disturbance-free signal to copy. Done properly, it can also supply a kind of synthetic inertia, an instantaneous energy response in the first cycles of a fault, well before AEMO’s frequency control ancillary services markets even have time to respond.
A rough side-by-side
| Characteristic | Grid-following inverter | Grid-forming inverter |
|---|---|---|
| Behaves electrically as | Current source | Voltage source |
| Needs an existing grid reference? | Yes | No, in principle |
| Response in first ~2 cycles of a fault | Depends on control loop settling | Near-instant, inherent to the control design |
| Typical current NEM deployment | Most existing solar and wind, most early batteries | A handful of newer big batteries and pilot projects |
| Can it “black start” a dead network section? | Generally no | Yes, that’s one of the main selling points |
A worked example, because the milliseconds matter #
Say a 500 MW transmission line trips somewhere in a weak part of the network, one already short on synchronous generation nearby, the kind of scenario AEMO now formally screens for under its system strength framework. In the first 100 to 200 milliseconds, before any market mechanism, before any operator can do anything, the local voltage and frequency start to swing. A grid-following inverter watching that swing through its phase-locked loop can genuinely struggle to track a reference that’s moving fast and getting noisy, and under some conditions it protects itself by tripping offline. A grid-forming inverter, by contrast, is actively setting a voltage of its own during that window and contributing to damping the swing rather than reacting to it a beat late.
Multiply that by a region with several gigawatts of inverter-based generation and not much synchronous plant left, and you can see why AEMO’s Engineering Roadmap to 100% Renewables treats this as core infrastructure planning rather than a nice-to-have feature. It isn’t about whether the grid has enough megawatts on a sunny Tuesday. It’s about whether the grid has a stable electrical reference to hold onto during the bad five seconds.
Where this is already happening in Australia #
South Australia, unsurprisingly, got here first. The ESCRI-SA battery at Dalrymple North on Yorke Peninsula, a 30 MW/8 MWh system part-funded by ARENA and the South Australian government and commissioned in 2018, was one of the earliest grid-forming battery deployments in the country, built explicitly to hold up a section of network prone to islanding. ElectraNet’s synchronous condensers at Robertstown and Davenport do a related but distinct job, adding system strength without adding energy, and they’re a good reminder that grid-forming inverters aren’t the only tool for this, just the newest one that doesn’t need a turbine hall.
Several of the bigger batteries now being built or contracted under state schemes and the Capacity Investment Scheme have grid-forming capability specified into their contracts, and I’d expect that to become close to standard for anything sized above roughly 100 MW within a couple of years. South Australia’s own experience running at very high renewable penetration, which I covered in South Australia’s renewable energy record: what it hides, is essentially a live trial of exactly this problem at grid scale, and the state’s transmission planners have been more explicit about system strength gaps than most.
It’s also worth reading this against the ongoing debate in Are big batteries being built for the wrong job?, because grid-forming capability is one of the clearest cases where a battery’s value has nothing to do with arbitrage or capacity and everything to do with a service the market has historically under-priced.
The catch: standards, cost, and who actually pays for it #
None of this is free, and I’d push back gently on anyone spruiking grid-forming inverters as a simple software upgrade. Retrofitting existing grid-following assets is often impractical; it’s mostly a decision made at the point of building new plant, with cost and design implications for the battery or wind farm developer. AEMO and the AEMC have been working through exactly how these services get specified, tested and remunerated, because “grid-forming” isn’t a single certified feature, it’s a spectrum of control designs and performance, and proving one inverter genuinely rides through a fault the way it claims requires detailed modelling and site testing that adds time and cost to a connection process already notorious for delays.
There’s also a fair question about who pays. If grid-forming capability is a system-wide public good, in the same way transmission is discussed in Transmission versus distribution: who owns the poles and wires, there’s a reasonable argument the cost should sit closer to network charges than to individual developer risk. Right now it sits somewhere uncomfortably in between, decided project by project through connection negotiations, which is slow and inconsistent and, frankly, not how you’d design this system if you were starting from scratch.
Batteries, pumped hydro, and who’s actually best placed to do this #
Batteries have a real advantage here over pumped hydro. A grid-forming inverter’s response happens in milliseconds using power electronics; a pumped hydro unit, even a fast one, still has physical machinery to bring online. That doesn’t make pumped hydro obsolete, projects like the one examined in Kidston pumped hydro: the mine that became a battery still bring long-duration energy storage and genuine synchronous inertia that batteries don’t naturally have, and the trade-offs between the two technologies are laid out properly in Pumped hydro vs big batteries: which firms the grid?. But for the specific job of instantaneous voltage and frequency support in the first cycles after a disturbance, batteries with grid-forming control are, on the current evidence, simply better suited to it than spinning water.
My honest read #
I think Australia is doing the engineering right and the market design slowly. AEMO’s technical work on this, going back to the lessons drawn out of that 2016 SA event, has been genuinely rigorous, and the pilot deployments have mostly performed as advertised. What’s lagging is a clean, predictable framework for who specifies grid-forming capability, who tests it, and who pays for it, so that it stops being negotiated fresh on every connection application. Until that’s sorted, we’ll keep adding gigawatts of new wind, solar and batteries onto a grid where the reference wave itself is getting quietly thinner in places, and hoping the next weak link doesn’t show up on a stormy Tuesday in the wrong postcode.
Somewhat like a long correspondence chess game, actually, where you can have a completely winning position on the board and still lose on the clock. The megawatts are there. It’s the timing that isn’t settled yet.
— Anjali Rao, Grid & Storage Correspondent
Photo by Moritz Kindler on Unsplash