Last week’s outbreak of price volatility in South Australia was the first significant spot price action in the NEM since summer. Like Tolstoy’s unhappy families, every NEM volatility event is different: here I’ll try to unpick the multiple threads woven into this one.

I’ll use versions of this supply mix chart to provide an overview and to highlight specific factors:

Price volatility bookended the night of Sunday 21th June into Monday 22nd June. There was also a precursor event 24 hours earlier that in hindsight was a warning of the volatility to come.

Still air

Through Saturday wind production fell away, reaching negligible levels for 24 hours from mid-Sunday. Gas-fired generation lifted to fill the gap, especially outside daylight hours. Isolating these two sources makes this more obvious:

Outside help – or not?

A superficial analysis would now blame the high prices on low wind. Or expensive gas. But let’s remain curious. We didn’t see price volatility in Victoria, so what were the interconnectors into South Australia doing? Bringing those into the view is illuminating:

The “Transfer” series is the net flow into or out of South Australia across its multiple interconnections to Victoria:

• the DC Murraylink cable,
• and two AC connections via Heywood to the south and the relatively new Buronga – Bundey line (PEC-1) to the north.

As I explain elsewhere, the two AC links are combined into a single “V-SA interconnector” for market dispatch and reporting purposes, but appreciating the distinct physical components is important in understanding what follows.

During the critical periods we don’t see transfers from Victoria increase to fill the gap left by wind. In fact during morning and evening peaks from Saturday afternoon through to mid-Monday there are net outflows from South Australia, coinciding with high prices – clearly something is going awry with transmission, leading to forced flows into Victoria. We’ll return to this in more detail later.

What about batteries?

Let’s add battery discharge – and recharging – to the picture. I’ve removed gas and wind to more clearly highlight battery activity and stored energy levels:

Discharge (generation) occurs in morning and evening periods – and coincides with those periods of low or reverse interconnector transfers. Recharge concentrates in the middle of each day and to a lesser extent the overnight periods – but is noticeably absent in the Sunday-Monday overnight period: the battery fleet entered Monday with close to nothing in the tank. This left batteries with very little ability to respond to that morning’s volatility spike (as noted here earlier), with an extended run of prices at the market cap ($20,300/MWh).

There are important differences between the evening peak periods on Saturday (June 20^th) when prices remained below $300/MWh, and Sunday evening’s volatility. Despite forced transfers out of the region on both days, on Saturday this reversed just after 8pm with net transfers into South Australia of 300-400 MW, relieving pressure on batteries while the fleet still held moderate levels of charge.

But on Sunday batteries began with about 1,000 MWh less energy in storage, and were substantially depleted by the time there was any significant relief from the forced outflows to Victoria. At which point spot prices jumped towards the market cap and stayed volatile all evening – no doubt a factor in batteries not beginning to recharge.

The overall picture

Let’s stand back and summarise findings so far:

• Very low wind and consequently a high call on the gas-fired generation fleet were key background conditions for the volatility outbreak.
• Transfers from Victoria did not rise consistently to help fill the gap left by falling wind – during key periods something significant was limiting imports far below the notional 700-800 MW combined capacity of the AC and DC links, and at times forcing energy out of South Australia.
• Prices “popped” once the battery fleet’s charge ran low on Sunday evening, then again on Monday morning after the fleet had failed to undertake any meaningful recharging overnight.

The fundamentally counter-intuitive factor here is interconnector behaviour, which will occupy most of the rest of this article. Given that behaviour, it’s also puzzling that the battery fleet failed to prepare for a possible repeat event on Monday morning – when the three prior peak periods (Saturday evening, Sunday morning and evening) had clearly been impacted by transmission limitations.

Empty tanks – autobidders beware?

Speculating on reasons for the battery fleet’s failure to recharge ahead of Monday morning, one factor may have been the high Sunday-Monday overnight prices, which hardly fell below $300/MWh and approached $1,000/MWh in a few dispatch intervals. It’s possible that most instances of the autobidding software used by battery operators were set not to recharge at such elevated prices unless there was an imminent threat of much higher prices ahead.

Now, a human operator looking at what had happened on Saturday and Sunday could well have realised that similar restrictions on interconnector flows were quite likely to recur on Monday morning. They may have concluded there was clearly a threat of extreme prices, and set their battery bidding software to at least partially recharge despite the high overnight prices.

But left to their own devices, most autobidders may not have had the capacity to look beyond AEMO’s predispatch price projections (shown here are three price series – for expected regional demand levels, and increments of +100 MW and +200 MW):

While these projections show some risk of prices above $1,000/MWh at higher demand (or lower supply) levels, they may not have been extreme enough to override autobidder reluctance to recharge at prices above $300/MWh.

If this admittedly speculative analysis is correct, then the event may have been an expensive reminder for some operators that it should be mandatory to have a knowledgeable human overseeing, and where necessary adjusting, an autobidder’s default behaviour.

Down the transmission rabbit hole

Anyone who started reading for a relatively high-level analysis might be advised to jump off here. Unpicking the interconnector thread is quite a bit more complex.

What were the transmission limitations about? Just after midday on Saturday, AEMO published the following Market Notice to participants:

Worded in typical AEMO Market Notice style, this terse communication provides very little information on cause or likely effects. It’s helpful to start an explanation with a network diagram:

X-what?

The Balranald – Buronga X3 line sits within New South Wales, forming part of a relatively low capacity 220kV corridor built in the 1980s to reinforce supply to the far south-west of the state – which had originally been supplied entirely from Victoria via Red Cliffs.

Constraint mechanics

With South Australia able to draw power from this area via Murraylink and PEC-1, X3 flows are affected by transfer levels to South Australia. A reduction in X3 capacity (AEMO’s “short notice rating change”) means that these transfers may have to be reduced – or even reversed – to avoid overloading the line. The constraint referred to in AEMO’s notice provides more detail:

Each constraint equation in the NEM is a mathematical instruction to the dispatch algorithm for managing a limit on the power system or its components.

• A constraint’s “Right Hand Side” (RHS) value reflects the size of the limitation – in this case headroom on the X3 line.
• The “Left Hand Side” (LHS) lists controllable items (target generator outputs, battery charge or discharge and interconnector transfers) which impact the limit. These LHS items are levers the dispatch process can use to manage the limitation, with the directional impact and strength of each lever given by the sign and size of its “Factor” in the second column.

Generators like Limondale and Sunraysia (factors of +1.0) sit east of Balranald and increase westward X3 flows when generating; generators around Broken Hill (factors of -0.5318) tend to reduce westward flows on that line. Constraining output from the former, or increasing it from the latter, help to relieve the constraint.

I’ve highlighted terms for Murraylink (“V-S-MNSP1”) and the V-SA combined interconnector which includes PEC-1. Exports to South Australia on either route tend to increase westward X3 flows.

Why “V-SA”?

Observant readers will notice that there is no “PEC-1” term on the constraint LHS.

• PEC-1 is bundled with Heywood as part of the V-SA AC interconnection, and not treated as a separate New South Wales to South Australia interconnector, in part because a significant proportion of its flows come from Victoria via Red Cliffs and Buronga.
• It won’t be until October 2026 that we see PEC appear explicitly, in the form of a NSW1-SA1 interconnector (via PEC Stage 2).

Unlike DC links such as Murraylink whose flows are controllable, there’s no way for the dispatch process to independently target flows on parallel AC links like PEC-1 and Heywood. Each will carry some share of overall transfers between South Australia and Victoria, determined by system conditions and the physical laws of AC networks.

This helps explain why there’s a V-SA term, not a PEC-1 term, in the X3 constraint, as well as its relatively small factor of 0.1689. Part of the AC transfer between Victoria and South Australia will flow via PEC-1, and the network topology means that higher flow on PEC-1 also draws in some additional westward flow on the X3 line itself. The small factor simply means that the relationship between V-SA total flows and their impacts on X3 loading are a lot less than 1 to 1.

For reasons there’s not room to explain here, even if they have smallish factors interconnector transfers are often the first levers the dispatch algorithm uses to manage constraints like this one: once the X3 line reaches its flow limit, to prevent further increases the process will start to reduce or even reverse transfers on Murraylink and V-SA.

Counterprice flows

The tabulation in the constraint screenshot earlier showed Murraylink is flowing into Victoria (target -156 MW), while V-SA’s target flow is only 183 MW into South Australia, well below its 600-odd MW capacity. This was despite the price in South Australia being $159/MWh in this interval, and Victoria’s just $92/MWh.

Jumping ahead to Sunday evening’s volatility event, the same constraint was active and forcing reverse net flows against a much larger price difference:

Here, managing the X3 constraint by limiting and reversing flows on the interconnectors to South Australia was proving to be a very expensive solution for the dispatch algorithm, as it required dispatching extra generation in South Australia bidding at prices up to $15,100/MWh.

Leverage lost

The problem was that the algorithm had very few levers to use except those interconnectors:

I’ve highlighted in the table the capacities available at generators and batteries able to reduce X3 westward flows – either charging at batteries with positive constraint factors (east of Balranald), or generation from units with negative factors (west of Buronga). None of the former had any charging capacity bid available. Most of the latter were solar farms with no generation available since the sun had set. There was very limited wind, and only one battery (KESSB1) was offering to discharge.

With these sources dispatched to maximum, it was left to the two interconnectors to cover the remaining gap in the constraint and hold X3 flows to the line’s reduced capacity, resulting in net exports from South Australia despite its 5-digit spot price.

Monday morning blues

A similar pattern played out during the Monday morning event, but with the South Australian battery fleet unable to contribute much supply, made things even more difficult for the dispatch process. Here we see the X3 constraint violating with its LHS value exceeding the RHS limit.

Physically, this means that the dispatch algorithm was unable to wind back or reverse transfers to South Australia sufficiently to ensure that X3 flows remained within the reduced line rating.

This is because – with the battery fleet largely hors de combat – there simply wasn’t enough available generation in South Australia to meet local demand in this interval without net imports of 120 MW: 233 MW out of South Australia on Murraylink, offset by 353 MW flowing into South Australia on the AC interconnectors.

(Why these offsetting flows rather than (say) just scheduling 120 MW on V-SA and dispatching Murraylink to 0 MW? That’s due to their different factors in the X3 constraint. Murraylink’s 0.3880 versus V-SA’s 0.1689 means that Murraylink flowing 233 MW into Victoria helps relieve the X3 constraint by more than twice as much as an additional 233 MW on V-SA worsens things.)

Action needed – trip the breakers

Lines running beyond ratings represents an “insecure” dispatch position, requiring prompt action to be taken by AEMO. In this case, the decision was made to take the X3 line out of service altogether – after all, a disconnected line can’t be overloaded!

There doesn’t seem to have been an AEMO market notice to this effect, but it’s deducible from the constraint sets invoked by AEMO over this extended event.

“Constraint Sets” are groupings of constraints related to specific physical aspects of the power system – for example variations in multiple limits which might be required when a particular piece of transmission equipment is out of service. The Gannt-style chart above shows the timing with which various constraint sets related to the X3 situation were active.

Starting from the middle of Saturday, the set containing the original X3 derating constraint appeared – around the time of that first AEMO notice. Jumping forward to Monday morning – just after the constraint violation event discussed above – we see that original set being removed and an “N-BABU” set commence. This constraint set covers an outage of the X3 line (Balranald to Buronga – hence “BABU”). From this, the obvious conclusion is that AEMO in consultation with Transgrid decided that taking X3 out of service was the best response to an ongoing overloading risk.

We also see that something similar happened during the precursor event on Saturday evening, where for a short period the N-BABU set replaced the original constraint. It’s not clear why things then reverted, ie why the X3 line went back into service for the next 36 hours.

A new solution – and finally an explanation

On Tuesday afternoon, the N-BABU set was replaced by a new “N-BU_7118” constraint set. AEMO did issue a Market Notice at this point:

Still pretty uninformative, but looking at one of the constraint equations invoked by this new constraint set yields more insight:

Buried in the “Plain English” (!) translation of this constraint, there are finally some details of the underlying physical cause for all these market machinations: a “Buronga 7118 isolator hot joint issue”.

An isolator is a piece of switching equipment at the end of a transmission line, and a “hot joint” is a tell-tale sign of a connection to that piece of equipment starting to lose conductivity and heat up when current through it increases – probably picked up on a thermal-imaging camera. Hence the reduced X3 line rating.

A key difference

Finally, compare the LHS terms and factors on this new constraint with the original:

They are very similar – not a surprise, because they are both designed to manage a physical restriction of capacity on the X3 line. The single key difference is the absence of the V-SA interconnector from the new version of the constraint.

But that one difference has a very substantial market impact: AC transfers into South Australia will no longer be restricted by the “hot joint” situation on the X3 line isolator (which is still ongoing at the time of writing). While moderate volatility persisted over the next couple of days, there has so far been no repeat of the extreme prices on Sunday and Monday.

=================================================================================================

About our Guest Author