A University of New England (UNE) PhD researcher has demonstrated that artificial intelligence can dramatically improve electricity price forecasting, potentially reducing the cost-of-living pressure from power bills while making renewable energy work more efficiently.
Mark Sinclair, who is pursuing his doctorate in Artificial Intelligence at UNE while working as an AI Engineer at Empower Energy in Sydney, has published research showing that transformer models—the same AI architecture that powers ChatGPT—can predict electricity prices far more accurately than current methods, often cutting forecast errors nearly in half.
The research, published recently in the journal Applied Sciences (details below) addresses a problem that directly affects every Australian household and business: the volatility of electricity prices in an increasingly renewable-powered grid.
“The national electricity market still doesn’t provide true forecasts,” Sinclair explains. “They do what we call a linear problem solve based on current bids from generators compared to forecasts of power usage. It’s more of an indication than any form of forecast—like trying to work out what the stock market’s going to do at open based on current bidding.”
This forecasting gap can be costly. When prices are poorly predicted, batteries charge at the wrong times, generators make suboptimal decisions, retailers hedge badly, and grid operators are forced into defensive actions.
“All that inefficiency ultimately gets paid for by consumers,” Sinclair says.
AI better at interpreting volatility
The Australian National Electricity Market is one of the most volatile in the world, partly due to the rapid integration of renewable energy. On a hot day when demand exceeds forecasts, electricity prices can spike from around 20 cents per kilowatt hour to $20 per kilowatt hour.
Current forecasting methods struggle particularly with these volatile periods—exactly when accuracy matters most.
Sinclair’s research tested whether transformer AI models could do better by learning complex patterns across time and combining multiple information sources simultaneously: demand, weather, interconnector flows, and forward-looking market forecasts.
The results provided compelling evidence that he was on the right track.
AI transformer models consistently predicted short-term electricity prices more accurately than official forecasts and other widely used machine-learning models, with relatively strong performance during volatile periods.
“Even a very small improvement in forecast accuracy can have a massive flow-on effect right throughout the whole industry,” Sinclair notes. “It affects everything—from aluminium smelters timing their energy-intensive operations, to retailers managing their margins, to generators knowing when to ramp up production efficiently and burn less fossil fuel.”
Implications for households and business
For households, the implications are direct. Better forecasts enable batteries, hot-water systems, EV charging, and other flexible loads to be scheduled proactively rather than defensively. This reduces waste, smooths out price spikes, and puts downward pressure on electricity prices overall.
The research also supports more efficient renewable energy integration. “Wind and solar are variable by nature, but when market conditions can be anticipated more accurately, the system doesn’t need to rely as heavily on expensive fossil-fuel backups ‘just in case’,” Sinclair says.
UNE’s Deputy Vice-Chancellor – Research, Professor Chris Armstrong, congratulated Sinclair and Dr Shepley on work with potential significance for Australia’s renewables rollout, and for its contribution to understanding of how AI can improve economic and sustainability outcomes.
“It’s a perfect example of the research that UNE is focused on – applying knowledge and technology to solutions that help regional communities become more resilient and self-reliant.”
“…for the betterment of mankind…“
Notably, Sinclair has made all his code and datasets publicly available.
“I’m not doing this for personal gain. I’m doing this for the betterment of mankind,” he says. “Anyone could take it and use it, and I hope they do.”
Sinclair is now transitioning into his PhD proper, focusing on an even harder problem: predicting when price spikes will occur. “None of these models handled price spikes really well. That’s going to keep me busy for the next three years.”
With a Bachelor of Engineering (Electrical) from the University of Sydney, Master’s coursework in Data Science from UNE, and professional experience developing AI systems for battery optimisation, Sinclair brings a unique combination of technical expertise and real-world application to his research.
“The key message is that smarter forecasting is not a theoretical luxury,” Sinclair says. “It’s one of the quiet, structural levers that can meaningfully reduce cost-of-living pressure, improve grid stability, and accelerate the energy transition—without asking households to change how they live.”
Sinclair’s electricity pricing research originated from his UNE Masters thesis, completed in October 2025. His principal Masters supervisor, Lecturer in Computational Science Dr Andrew (Andreas) Shepley, was impressed with the quality of Sinclair’s work and encouraged him to develop it into his first scientific paper.
Dr Shepley is now principal supervisor on Sinclair’s PhD, with UNE Senior Lecturer in Data Science, Dr Farshid Hajati, co-supervising.
Don’t miss any of the important stories from around the region. Subscribe to our email list.