Spyros Chatzivasileiadis has just received one of the prestigious ERC Starting Grants for promising young researchers awarded by the European Research Council. The funds will be used to introduce completely new methods for developing and certifying the use of artificial intelligence over the next five years so that electricity sector players can build the missing trust in the new technology.
The supply of electricity is currently undergoing major changes in line with the green transition. Where electricity was previously supplied by large power plants with stable supplies, it will—in future—come from a large and continuously increasing number of renewable energy sources with varying supply volumes depending on weather and wind. Concurrently, our society is becoming increasingly electrified, which significantly increases the demand for electricity.
“Today, the electricity sector uses many tools and resources to create possible scenarios and perform a lot of calculations for these scenarios in order to ensure a stable electricity supply. If artificial intelligence is used for this work instead, we could perform these calculations 1000 times faster, thus saving both time and money,” says Associate Professor Spyros Chatzivasileiadis, DTU Electrical Engineering.
“However—until now—artificial intelligence for the electricity sector has been a ‘black box’, where you only have an idea about the structure of the algorithm but have no detailed insight into how the technology works. This has led to a lack of confidence in artificial intelligence for safety-critical systems, such as power systems. I will develop the tools necessary to change this,” says Spyros Chatzivasileiadis.
Sector efficiency improvements
After completing his studies, Spyros Chatzivasileiadis has worked in a wide range of research fields in the electricity sector. In recent years, he has also become interested in artificial intelligence and has thus spotted the many opportunities this offers for efficiency improvements in the sector if barriers to the use of artificial intelligence can be removed.
“My first initiative will be to provide guarantees for how artificial intelligence will behave when used to optimize the electricity sector. The guarantees are necessary to convince the key stakeholders in the electricity sector. For them, avoiding blackouts is the most significant factor in the security of supply, as a blackout lasting just one-hour costs society several hundreds of million kroner,” says Spyros Chatzivasileiadis.
In the coming years, he will use neural networks for the task. The networks must be trained in predicting the stability of the electricity grid while providing guarantees for their behaviour. Novel neural network training algorithms will then be developed which also incorporate the physical models that have been developed and used by the sector over the past 100 years. Neither of these activities has previously been performed by others.
“The plan is to develop a new training process for neural networks which will enable the extraction of performance guarantees. This will give the players in the industry a guarantee about how methods based on artificial intelligence behave. This will build trust on these methods, which can then be used to increase the speed, optimize operations, and reduce electricity supply costs, thus boosting the green transition,” says Spyros Chatzivasileiadis.
Electricity companies both in Europe and in the USA have already shown interest in the project.
“My hope is to be able to prove that artificial intelligence isn’t a ‘black box’ for systems like the electricity grid, where the security of supply is the top priority. Instead, I want to show that it’s both comprehensible and valuable. This will open the door to the use of artificial intelligence to optimize a wide range of sub-elements of the sector, and it can also contribute to the secure and reliable operation of the electricity grid,” says Spyros Chatzivasileiadis.
Once the algorithms have been developed, they can also be used in other areas in which security of supply is a top priority, and existing physical models can be involved. For example, this could be in aeroplanes or self-driving cars.