Ellen Phiddian
Cosmos science journalist
In a highly competitive market, all energy generators rely on highly accurate predictions of how much electricity they’ll be able to make. Australian researchers have figured out a way to improve these predictions for wind and solar farms, using artificial intelligence.
The National Energy Market – “the grid” – requires automatic forecasts every five minutes from electricity generators. This ensures that electricity generation meets demand. It can be very costly if those five-minute forecasts prove to be incorrect.
“The market operates on these five-minute windows,” says Dr Christoph Bergmeir, a senior lecturer in data science and AI at Monash University, and lead researcher on a recently completed ARENA project.
“Every five minutes, the generator has to bid for how much they’ll generate.”
The amount generated can be tricky to predict for wind and solar farms – solar especially. Currently it’s mostly done based on simple (“naïve”) algorithms and weather models.
“If we have a day without clouds, then of course it’s quite predictable,” says Bergmeir.
“But if a cloud shadows the solar panels, in that moment the production has dropped by, let’s say, 40%. And that happens from pretty much one second to the other. For wind it’s maybe a bit less fast, but even there, in one minute it can change quite a lot.”
If a generator over-predicts, the electricity has to be speedily, and expensively, produced elsewhere, and the cost is handed back to the generator. Because of their variability, wind and solar farms face a lot of costs for incorrect calculations.
Monash researchers sought to address this with machine learning. Bergmeir’s team used data produced by wind and solar farms to train an AI.
Bergmeir says they wanted to make something that could easily be integrated into currently operating generators: meaning it used data the farms have already recorded, and no extra hardware would be required. It’s a requirement of AEMO (the Australian Energy Market Operator) that this data be supplied, so every wind farm in the country already collects it.
“Wind turbines already measure wind speed, wind direction, and power generation. And that’s all already sent through to AEMO. We didn’t want to deploy expensive additional sensors on the wind farms and the solar farms.”
With wind farm generation, the tool was a success. “We could see that the system, without any additional sensors, does really well. And it’s not clear to me which additional sensors would make it better,” says Bergmeir.
Predictions weren’t quite as perfect for solar power. “Our system does work, but probably with some additional sensors, measuring cloud cover and so on, it could potentially be better.”
The study used a wind farm (Waterloo Wind Farm in SA) and a solar farm (Ross River Solar Farm in Queensland) to develop the algorithm, but the prediction tool is now on the market. Engineering company Worley, which collaborated with Monash on the project, has made the tool commercially available to other wind farm generators.
Since working on the energy prediction tool, Bergmeir has turned to focus on cost prediction. He’s now running a competition for energy prediction and optimisation, using data generated by a “microgrid” installed at Monash University – a combination of rooftop solar, batteries and other power sources that aren’t linked to the main grid, so that researchers can develop and test more forecasting and optimisation technology.
Read more:
- Solar and wind cheapest energy source in Australia
- Sharing batteries provides better storage
- How to jet power wind turbines