The rise of rooftop solar and battery storage combined with emerging technologies like electric vehicles and ever-evolving smart devices makes it hard to predict future demands on the electricity grid. Could deep learning be the answer?
Deep learning is a field of machine learning based on artificial neural networks – complex computing systems inspired by the human brain. It has been used in a wide range of applications including brain cancer research and short-term electricity price predictions.
Over the next several years, a team of researchers from RMIT University in Melbourne will work in partnership with electricity distribution companies CitiPower and Powercor to develop a deep learning system to accurately predict electricity demand in Victoria’s power network.
Lead researcher Dr Mahdi Jalili told create that deep learning is evolving into a science that can provide highly accurate predictions. He added that accurate electricity demand forecasting is important to get everyday planning right, and avoid over- or under-investment in the power distribution network.
“If you have over-investment, it’s going to be transferred directly to [electricity bills] without improving the efficiency of the network,” he explained, adding that under-investment could lead to issues such as blackouts and the need for more spending to resolve them.
The RMIT researchers recently received a $321,000 grant from the Australian Research Council (ARC) for a three-year project to further their research, which is scheduled to start in September.
As well as Jalili, the team includes Professor Xinghuo Yu from RMIT, Peter McTaggart from Powercor and Dr Peter Sokolowski, Research Fellow at RMIT and Chair of the Engineers Australia Electrical College.
Jalili said that Victoria’s smart meter network has been collecting energy usage data every 30 minutes for the past decade, and this will be used in the deep learning project. The team will also be using available data on rooftop solar, and will incorporate data on electric vehicles and energy storage such as batteries as it becomes more readily available over the next couple of years.
The researchers’ aim is to use data to optimise existing grid infrastructure, and help industry and government plan where to focus future investments for the best results. They will also look at ways their technology can be used to fill gaps in existing data.
At the end of the project, the team hopes to have deep learning technology which can be used for short, medium and long term forecasting for individual, substation, local area and global electricity demand.
Sokolowski added he was looking forward to delivering data-driven decision making and using these insights to train tomorrow’s expert workforce.
Engineering vs market model
Other efforts are also underway to shift energy network planning to a more data-based model.
In May, energy analysis companies Global-Roam and Greenview Strategic Consulting joined forces to release a 530-page Generator Report Card based on 20 years of National Electricity Market (NEM) data to the end of December last year.
According to Paul McArdle, managing director at Global-Roam, the aim was to keep the analysis as objective as possible – not pro-coal or pro-renewables, but pro-data.
While the information was publicly available from the Australian Energy Market Operator (AEMO), Global-Roam explained that it focused on moment-to-moment transactions. The challenge was to put the data in a form that showed long-term trends.
Among the trends examined was whether the wind and solar generation in the AEMO renewable energy zones balanced each other (when the wind wasn’t blowing, the sun was shining and vice versa).
“We hope that the results presented in the Generator Report Card can raise the level of awareness of what’s actually happening (and possible) in order that we can move on to explore approaches that will actually help the energy transition succeed – rather than just re-hashing old debates,” McArdle said in a blog post.
Sokolowski said that while it was good to examine the market data to see what it reveals, it was important to look at technical and market performance to get a complete picture and produce accurate forecasts.
“We need to look at what we actually need to know, what we need to collect, and how we need to analyse it to allow data-based planning and management,” Sokolowski said.