Laser tech could slash grid penalties for wind farms, because their forecasts are less wrong at the right times

A wind forecasting trial in Victoria’s breezy heartland has found that the use of on-site, real-time data collection can mean bigger savings for wind farm operators — but adding AI is still considered a waste of money.

The Australian Renewable Energy Agency (ARENA) funded a trial that tested two forecasting methods – one that uses data from turbines and another with two lasers to measure physical wind speeds. The idea was to see which could beat the market operator’s estimates and reduce grid instability and frequency-control (FCAS) costs.

But the study, led by Aeolius Wind Systems at the 140-turbine Macarthur wind farm in south-west Victoria last year, found that the most accurate forecast doesn’t necessarily mean it will be the most cost-effective.

A SCADA model that used turbine data was deemed to be the most accurate over five minute intervals, beating the Australian Energy Market Operator’s (AEMO) operational forecast by 21 per cent. 

But it was also worse at avoiding forecasting errors at times when they were most expensive to make, an issue that added an extra $45,000 a year overall in costs.

The LiDAR laser model, on the other hand, was less statistically accurate but better at avoiding those key forecasting errors. It delivered an estimated overall saving of $306,000 a year.

The study found that a combination of both models made redundant an AI-based software that had been developed to make the initial data even more accurate.

“Increasing neural-network architecture complexity beyond a certain point – the strongest linear model, the best multilayer perceptrons, and the best convolutional neural networks all converged on statistically similar minimum error,” the final report on the Aeolius Wind Systems-led study said. 

“Meaning [that] added complexity bought no additional accuracy while substantially increasing the resources needed to build, run and maintain the model.”

Accurate forecasting to cut costs

With wind and solar now making up about half of available generation capacity in the east coast grid, it’s increasingly important that short-term forecasting is accurate to reduce the risk being charged for under- or over-delivering electricity. 

Solar farms, for example, paid out 32 per cent of all frequency performance payments (FPP) penalties between June last year and April this year.

That’s despite producing only 8.7 per cent of total generation on the east coast.

FCAS costs are falling thanks to batteries which are helping to balance the grid, but regulatory costs are still an issue for variable generation. 

Wind and solar power output changes minute-to-minute as wind speed and direction shift, and AEMO has to balance these fluctuations to keep the grid stable

“It means that when a wind or solar farm’s actual output diverges from what it expected to deliver, AEMO has to call on frequency control services to make up the difference. That cost is recovered from the market and, ultimately, from consumers.

A regulation raise cost is charged when a generator can’t meet an AEMO request for more power to come into the market in the event of a sudden dip

A regulation lower cost is the opposite — when suddenly there is too much electricity being put into the network.

Wind and solar generators often face high regulation costs because inaccurate forecasts cause them to deviate from their scheduled output.

Historically, the Macarthur wind farm has paid about $1.5 million a year in regulatory costs, but each model could cut that substantially. 

While the study didn’t test what would happen if the results from both were combined, it suggested that using both turbine and laser data models would deliver even more gains.

To support independent media, and help combat the spread of deliberate misinformation and disinformation about the energy transition, you can click here to make a one off donation or become a regular supporter of Renew Economy.