The brain that’s turned UoQ’s 2.2MWh Tesla Powerpack battery into a private piggybank
If you want to build something that’s really smart, make it at a university. That’s where the brainy people hang out. When keen minds are set a task to imagine the potential in expensive new technology, a whole lot of value can be unleashed. It happened at the University of Queensland as it prepared to connect $2 million worth of Tesla Powerpack batteries to the National Electricity Market last year, in its quest to be energized only by renewable energy.
The institution was off to a good start, with the 1.1MW/2.2MWh asset having been paid for with glorious compounded savings from the rooftop PV system on its St Lucia campus in Brisbane.
The battery was commissioned in late November and connected to the wholesale market in the beginning of this year. By mid-May the university proudly announced the Tesla kit had helped it save $74,000 in electricity costs over three months.
But how? Buying a battery won’t guarantee you a steady stream of revenue unless you know how to use it, which is where campus smarts come into play. The secret is to design a trading algorithm that looks at the energy price forecasts sent out by the Australian Energy Market Operator and figure out when to charge and discharge – to make money along the way while not ignoring the primary job of pushing down the campus power bill.
It’s not as easy as it sounds. And it doesn’t sound easy in the first place.
“Because UoQ has transitioned to becoming a participant in the NEM we are looking at forecasts for energy prices and scheduling the battery appropriately, based on [predicted prices],” says University of Queensland energy engineering officer Dominic Hains, who designed the algorithm that has turned the $2 million storage asset into a private piggybank.
The algorithm developed for the project looks to the future – based in AEMO’s forecasts – and compares what it sees against previous forecast scenarios, which are constantly updated. “That forecast keeps getting updated on a five-minute resolution,” Hains says, pointing out that AEMO releases two forecast models.
With the Tesla Powerpack in service for a few months now Hains has been able to look back at the battery’s trading activity and carry out virtual performance reviews for his robot prodigy.
“We’re able to run a perfect foresight analysis on how the battery has performed in retrospect,” he says, “[to see], if the battery knew exactly what the prices would have been, how would it have changed its operation.
“There are a lot of parallels between this and predicting the stock market – it’s got the same dynamics, people bidding in – so there is always going to be that huge unknown factor that is very difficult to model. That is something we are going to focus on.” He hints that enhancements to the algorithm will likely focus on outsmarting AEMO’s forecasts.
“What we’ve found is that, where there are forecasts for spikes in the future they typically tend to be the periods that have the biggest uncertainty in price,” he says. “Where prices have been forecast for the peak of the day they typically tend to come off – it’s just that the scale of that peak is very variable.”
The peak will typically occur close to AEMO’s predicted time window but the price will often settle lower than forecast, he says. Curve balls will always arrive out of the blue, of course, such as outages and generators tripping offline. “That will introduce some pretty sudden changes in prices, but not necessarily in the peak periods,” he says, “which is where the strength of the system lies – it’s able to pick up those changes pretty quickly and adapts the way it’s going to use the battery.”
The battery also bids into the “more lucrative” frequency control ancillary services (FCAS) market, Hains says, choosing the contingency FCAS market over the regulation FCAS market.
“Contingency FCAS is more of a back-up mechanism, if regulation markets don’t help the NEM enough. If frequency drops below a certain amount, assets that are bidding into the contingency markets will be called upon,” he says. “We may only be called upon three or four times a quarter.” Revenue is earned in the contingency market regardless of whether the UoQ battery is called on or not “because we have the ability to respond if an event occurs”.
The easiest money, of course, flows when PV systems in the state flood the grid around noon to such an extent that wholesale prices are plunged into negative territory. Hains cites an example where the battery anticipated negative prices over a weekend and decided to discharge the day before at a humdrum $33-53/MWh to make room to fill up on solar the next day and be paid $46/MWh for it. “It was able to get that arbitrage spread the next day by charging up at negative intervals.”
AEMO energy price forecasts are available to the public through NEMweb, but because the university owns the 64MW Warwick Solar Farm project in the Southern Downs region of Queensland it qualifies as a dedicated generator and, upon project completion, will earn a private connection to AEMO.
“We’ll be able to get a bit more insight into exactly what sort of data is being passed around in the NEM and sensitivity data as well, in terms of how generators are bidding and how that potentially could make the price in the coming half-hours a bit more volatile,” Hains tells EcoGeneration.
Hains sees plenty of scope for the UoQ battery – and batteries in general – to be put to use as a tool to hammer down demand charges. “There is potential for savings [at commercial entities],” he says. “If you’re able to shave 300-400kW off your monthly peak demand, that could be a pretty substantial revenue stream.”
If peak load days can be forecast with adequate accuracy, a battery can be used to cap energy use and thus influence a reset of demand charges.
It’s all about forecasting, after all. “It’s a pretty crazy and pretty exciting project – I love it.”