The winning paper tackles a very topical problem - supporting the control of energy storage by a so-called energy cluster, a local group of energy consumers equipped with renewable energy of the photovoltaic type.

Energy storage alleviates problems with the ongoing balancing of energy consumption and generation, and allows for a hedge against temporary high prices for energy taken from the market in the event of periodic solar shortages. Deciding whether to charge or discharge the storage at a particular time therefore requires taking into account uncertainty from three sources: demand, insolation and market energy prices. An additional difficulty is the non-linear characteristics of the efficiency of charging and discharging of the storage.

To solve the problem, a team from our University used a reinforcement learning algorithm based on the Deep Q-learning methodology.

- Such application of this methodology is a novelty,' emphasises Izabela Żółtowska, PhD. - Thanks to the techniques used in the work, it was possible to overcome commonly known problems with the convergence of the algorithm, which was confirmed by comprehensive tests.

Detailed information about the research results of the team from the Warsaw University of Technology is available in the award-winning paper entitled "Real-time energy purchase optimisation for a storage-integrated photovoltaic system by deep reinforcement learning".

The journal "Control Engineering Practice", in which the paper was published, is published by Elsevier. Each article published there receives, according to the list of the Ministry of Education and Science, 100 points (the third highest possible score).