Helping remote communities using AI
Using artificial intelligence (AI) to help provide affordable, reliable, sustainable power to remote communities has won a Talent Green Award from the German Federal Ministry of Education and Research for Soheil Mohseni, a postdoctoral research fellow at the School of Engineering and Computer Science’s Sustainable Energy Systems Group.
Selected by a jury of German experts, awards have been made to 25 young researchers from around the world, granting them unique access to the country’s top researchers. For Mohseni, this means being able to use the supercomputing facilities at the Technical University of Berlin and collaborating with leading global experts.
“Most of the existing solutions to provide electricity to remote areas rely on diesel generators,” says Mohseni, “but we’re looking at ways to create sustainable systems using whatever renewable power sources are available in the area—wind, solar, hydropower, tidal, geothermal or bioenergy.”
His research focuses on developing a method to optimise the size of the equipment in a microgrid. “While different combinations of technologies may be feasible, only one combination is optimal,” he says.
“The novel aspect of my research is using AI-based optimisation algorithms inspired by natural phenomena in methods developed for sizing, capacity planning and designing microgrids.”
Existing methods of planning microgrids use exact mathematical optimisation algorithms, which require a problem to be simplified. “Every time you run it you will get the same answer but the parameters of the problem you are looking to solve have been approximated.
“Using AI-based methods avoids these approximations and allows you to access the whole solution space to find a globally optimised solution without having to make simplifications.”
Mohseni says the main disadvantage of the AI algorithms is that a slightly different answer is possible each time the algorithms are applied. “This means it might be a locally optimal solution but doesn’t guarantee that the globally optimal solution can be guaranteed in only one run, so we have to run it several times to make sure. It takes more computing power but should guarantee the user that it is closer to the optimal solution even in just one run than the business-as-usual methods.”
Access to the supercomputing facilities in Berlin will enable Mohseni to run a full model of his simulations.
“Currently, due to computational limitations, all the software packages tailored to optimising microgrids use some kind of simplification to reach a solution,” he says. “To simulate performance over one full year, the time intervals are currently usually one hour. That’s incorporating solar radiation, wind speed, load patterns and wholesale market electricity prices at each hour of the year.
“That one-hour timeslot can be reduced to one second with a supercomputing facility, which means we can create a better efficiency and performance comparison of my algorithm with the industry-leading ones.”
Mohseni hopes this new generation of optimisers can lead to determining lower cost systems for economically under-developed countries and might drive the deployment of sustainable energy system.
As well as finding the optimum size for a microgrid, Mohseni’s work includes using game theory to help consumers and producers co-operate to find solutions for settling and clearing ‘micro markets’.
“In the near future, consumers can become ‘prosumers’—producing and consuming power, selling the excess to the grid or to neighbours,” he says. “We could have a micro market for trading energy and also decide to review our power consumption at certain hours. For example, if my neighbour offers me an incentive of say 5c/kWh for turning off my washing machine because this will be cheaper than they can buy from the grid, then I may do that.
“Game theory is used for studying such interactions, which we can characterise and see how people can cooperate with each other to find a solution that maximises benefits for everyone.”
There are two main classes of game theory—cooperative (coalitional) and non-cooperative (strategic). “In non-cooperative game theory, it is assumed that people are self-interested and want to maximise utility, just thinking about the financial yields,” says Mohseni. “The models can show how different behaviours affect the overall solution for the settlement in such micro markets, which in turn affects the total cost of the system.
“What I’ve found is that if you cooperate with each other, you are all better off. The total size of the system you have can be reduced and excess power that would otherwise be wasted is minimised. As the degree of self-interest increases, so does the curtailed power, the need for capital-intensive storage capacity, and the total cost of the system.”
If the COVID-related restrictions for entering the country for non-residents are lifted, Mohseni will be traveling to Germany by the end of 2022 for his research visit.