The $1.2 million project will use data generated by real-time sky cameras, satellite images and statistical modelling to design a world-first, short-term forecasting model to more accurately predict weather conditions from five minutes up to two hours.
Working alongside colleagues from CSIRO, the University of NSW and Genex Power, the University of South Australia is focusing on the statistical modelling component.
UniSA Professor of Environmental Mathematics John Boland said inaccurate short-term forecasts relating to wind and solar generation have cost Australia’s renewable energy sector about $5 million in the past decade.
He said precise self-forecasting would also help solar farms with battery storage capabilities predict when best to sell or store their electricity.
“Accurately forecasting the output of grid-connected solar systems is critical to increasing the overall penetration of solar and renewables. This is important for the stability and management of the electrical system as a whole,” Professor Boland said.
“Clouds can move and form very quickly, creating complex atmospheric layers which often move in different directions. The existing forecasting systems for wind and solar are designed for longer-term timeframes and have led to multiple issues over the years.
“This highlights the need for reliable short-term forecasts to provide confidence to both renewable generators and the entire industry.”
The 18-month project will implement short-term solar forecasting systems at five operational solar farms in Queensland, NSW and Victoria.
The electricity spot price in Australia is calculated every five minutes with a settlement period of 30 minutes. The settlement time will be reduced from 30 minutes to five minutes from July 2021.
Professor Boland said it was hoped the five systems would be operating at the solar farms by the end of the year so that almost a full year of testing of the forecasting tool could be done before the end of the project.
He said Australia’s five-minute pricing system was as short or shorter than anywhere else in the world, making it the ideal place to develop the forecasting tool.
“Because of the type of market we’ve got here it really invigorates the research area to get things right so it will probably be better than anywhere else in the world because it presents more difficulties and opportunities.
“Some of the other markets in the world are moving towards shorter time scales so what we can develop here will actually be very useful.”
Similar projects, also funded through ARENA’s Advancing Renewables Program, will focus on forecasting for wind farms.
Professor Boland said the Australian Energy Market Operator had previously used a forecasting method developed for a longer time scale that was based on a system developed for wind forecasting in Europe.
“Previously it was good for between a few hours and a day ahead but once you try to scale that down to a forecast in five minutes it’s difficult,” he said.
“But that’s the time scale the market works on in Australia so if the energy market operator can know better what all the possible generators are going to produce in the next five minutes then they can manage the system much more robustly.
“When that happens it not only makes it easier to manage the system but it makes it easier to keep the wholesale prices down as well.”
Sun rises, coals sets
Australia leads the world in household solar power, with around 15% of the nation’s roofs now fitted with PV panels, lowering energy costs and reducing our reliance on fossil fuels.
South Australia leads the nation in the uptake of wind energy and rooftop solar with renewable sources accounting for more than 50% of the electricity generated in the state.
For the project, CSIRO will use skyward-facing cameras to look at the development and movement of clouds, UNSW will use satellite images to look at slightly longer time scale – 15 minutes or half an hour – while Professor Boland’s team at UniSA will use statistical methods and mathematical tools for short-term forecasting.
“I can use statistical tools to model the seasonality, for instance, and then time series forecasts. From knowing the outputs and the solar energy available in the last few time steps we can forecast for the next time step using a simple time series model and when that is added to the seasonality model you get your total forecast.
“But the key thing at the end is to put these three models together and use different compilations of them to suit different times so we have a blended model that performs better than any of the individual parts.”
Republished with permission of The Lead South Australia.