Aidan O’Sullivan, the Head of Energy and Artificial Intelligence Research at University College London, will be presenting this year’s international keynote at Australian Energy Week 2019. We caught up with Aidan ahead of the event to learn about the role AI will potentially be playing in the energy markets of the future.
For background to our readers, would you be able to provide us with a brief overview of your projects and area of research?
My research focuses on the use of Artificial Intelligence (AI) and machine learning methods in the energy sector to promote efficiency and decarbonisation. This can be the use of AI at the individual customer level where I have projects with Energy Suppliers looking to estimate the uncertainty around customer consumption, or how best to engage with them through intelligent apps to encourage energy savings. I also look at the grid level on the use of reinforcement learning algorithms, algorithms that learn through repeated play in simulation to try and manage the electricity system more optimally. The availability of new sources of data in the sector, such as smart meter data for example, is enabling innovations and projects that just weren’t possible 10 or even 5 years ago so it’s a really exciting area to be working in as a researcher.
What do you think is the leading example of current applications of AI in energy markets around the world that you’re aware of?
I would say the current leading example of AI in the energy system is the recent work by Google Deepmind on improving the accuracy of their wind farm power forecasts in the US which increased utilisation by 10%. There are other applications as well such as the Siemens active network management project for frequency response.
In terms of overall systems management, could AI replace human decision-making in the market operator role?
As we see greater deployment of renewables in the energy mix, and adoption of new technologies like electric vehicles and demand-side response, we will need to adapt the way we manage the system to make best use of these resources. The current system management methods are based around dispatchable generation, however increasing the levels of uncertainty and complexity in both demand and supply demands a different approach to balancing the system. We see evidence of this in the increasing costs of the balancing market in the UK as well as increasing costs of curtailment of available wind power. There is therefore a need for AI methods to assist with, or even, handle the challenge of making complex distributed decision-making under uncertainty that the market operator will be faced with in this new paradigm.
How would you anticipate the transition to AI to occur?
I would expect we would see adoption by asset owners in the first instance such as generation companies looking to maximise the management of a portfolio of assets operating across a region, depending on the success of this there would be a case to utilise the same technology at a system level to improve the performance of the market as a whole.
How many years are we away for AI having a very significant impact in energy systems?
It’s tricky to say, the energy market is quite highly regulated so even with the technology available there may be regulatory barriers that delay adoption for some time. That said, I think there is a refreshing amount of interest and awareness about the potential of these methods to improve outcomes for operators so we could see trial deployments in the next 3-5 years.
What are the different ways that AI or Machine Learning can assist, and be applied, to give generators, grids and retailers a competitive advantage?
A generator company with the ability to make better predictions on the state of the system, whether it will be long or short or high or low demand, than its competitors has a distinct competitive advantage. Additionally there is a trend for systems to move to shorter gate closure times which in turn provides greater opportunity for companies using methods that enable automated rapid decision making on how much generation to commit, and at what cost while, also understanding the constraints of the system.
In the case of generator companies there is also the predictive maintenance application where data on the performance of an asset can be used to schedule predictive maintenance to pre-empt failure and longer periods of down time for repair.
For grid management, maximising the amount of zero-marginal cost renewables in the mix will bring down the total price of the system but given these assets are typically distributed and located far from centres of demand this requires management of generation to reduce curtailment and constraints on the line. Frequency response also becomes a challenge to be managed as renewables increase.
For retailers, sharp spikes in energy prices are a significant problem that has contributed to a number of bankruptcies of new suppliers in the UK. The ability to intelligently hedge out risk in an automated fashion is key to remaining competitive and being able to reduce costs to customers
If you had to give some key pieces of advice to energy companies looking to start an AI program, on how they should go about the process; what would your bullet points be if you were presenting to them?
The main piece of advice would be to make a genuine long term commitment to such a program. There will be requirements in terms of infrastructure, data and personnel that take time to develop. It is also important to ensure that the critical aspects of the problem being considered can be defined sufficiently well for an AI algorithm, this may mean involve gathering or providing access to more data on the system. Opening up data on the system to make it more transparent is also something I’d like regulators to push for as this would enable faster development of AI applications in this sector.