Original article in French: Agefi
This winter season has been mild in Europe. The only exception was a polar vortex that hit Scandinavia in November, freezing the thermometer down to -43°. Hardly surprising for Vasilis Pappas, chief meteorologist with MET Group. Together with his colleagues in the Analytics Team, he provides forecasts to traders at the Zug-based multinational energy trading company.
“We develop models for different timeframes, from daily to the coming seasons,” he explains. For this winter, he was anticipating above-average temperatures. “That’s more or less what happened, with a cold snap in mid-January.”
When presented with around fifty scenarios, MET employs machine learning algorithms to determine the most likely – in terms of wind, temperature and precipitation. They are then communicated to traders, who adjust their electricity and gas transactions or optimise the production of their power plants. “Getting it wrong can be costly,” stresses Vasilis Pappas.
Now in use for two years, the method enables MET “to reach satisfactory decisions seven to eight times out of ten”, a proportion “higher than before”, according to the meteorologist, who does not divulge any further details.
Romande Energie is also relying on artificial intelligence (AI) in electricity trading, “not to make a profit, but to cover our customers’ electricity demand”, explains data scientist Arthur Cherubini. The team of 14 engineers he works with draws up consumption scenarios related to the weather, with a view to the potential production of solar, hydro and biomass power stations. Based on this information, the energy company’s traders buy any electricity shortfalls at the best price.
While these technologies may minimise the discrepancy between actual and forecast electricity demand, Arthur Cherubini points out that “human analysts still make the final decision”. The algorithm may lack context for certain peaks or drops in consumption, “school holidays for instance”.
“Trading is certainly the energy sector where AI is the most advanced,” says Nicolas Charton. An energy strategy consultant, he is director of the Lausanne branch of the E-Cube consultancy, advising the Federal Office of Energy (Ofen), the cantons and companies such as Alpiq. “Trading based on production and consumption forecasts already relies on a large amount of data. It offers a quick return.” According to the consultant, these investments have already seen some forerunners, including the Argovian supplier Axpo, achieve profitability.
AI also provides valuable advice on energy infrastructure management. Predictive maintenance of assets such as wind turbines, monitoring wear and tear, is becoming increasingly widespread.
These technologies are also used upstream. As in the case of MET Group, where machine learning is used to determine the best location for a solar or wind power plant. Vasilis Pappas and his team examine “huge volumes of data to determine the most appropriate location in terms of solar irradiation and wind”.
Another example: E-Cube has developed an AI tool estimating the need for electric charging stations in each area, based on topology and vehicle traffic.
Meanwhile, Romande Energie is working on a “smart grid”. This low-voltage grid of tomorrow is needed at a time when decentralised production sources such as solar panels are multiplying, and electric cars and heat pumps are driving demand. “We are working on the design of an autonomous and flexible system which would manage the final portion of our infrastructure,” explains Arthur Cherubini. This project is currently being tested on a neighbourhood scale, in partnership with the Ecole Polytechnique Fédérale de Lausanne (EPFL).
For the record, from 2028 onwards, suppliers are required to install at least 80% smart electricity meters, following the revision of the Electricity Act. According to Nicolas Charton of E-Cube, the automation of consumption readings would allow energy companies to identify anomalies or errors more easily. “Imagine a customer with a faulty battery”, he says.
On the business side, none of the companies contacted has yet set up a virtual assistant open to the public. However, some have already made them available to their call centre employees, enabling them to respond more quickly to enquiries.
Could we envisage an energy company deploying a conversational robot for its customers? For advice, yes, but with no contractual commitment,” says Nicolas Charton. He does not anticipate any short-term adoption, however, “why not in a few years’ time, when bots are going to be able to use personal metering data more effectively”.
Although E-Cube has been incorporating machine learning into its advice to businesses for two years now, these technologies only account for “5% of the projects actually implemented”, according to Nicolas Charton. However fashionable AI may be, its applications are “still modest”, often hampered by budgetary constraints. The experts consulted believe that democratisation is only likely to come with the mass of data, when all infrastructures can communicate with each other.
Even so, AI is not going to drastically reduce bills. In 2023, electricity prices increased by an average of 27%. According to specialists, this surge is due to external factors that are always difficult to anticipate, such as the conflict in Ukraine. “Geopolitics is more complicated than the weather,” sums up Nicolas Charton.