Master's thesis under ARISE is awarded 1st place in the REN 2023 Prize
Nuno Mendes, supervised by Pedro Moura, from ISR-UC, and Jérôme Mendes, from CEMMPRE (members of the Associated Laboratory ARISE), won 1st place in the REN 2023 Prize. Nuno Mendes' dissertation, titled “Federated Learning for the Prediction of Net Energy Demand in Communities of Buildings”. The award ceremony took place in Lisbon at the Salão Nobre of the Ritz Hotel on November 13. The Secretary of State for Higher Education, Pedro Teixeira, will present the award.
Nuno Alexandre Gonçalves Mendes, currently PhD student at the University of Coimbra, supervised by Pedro Moura, from the Institute of Systems and Robotics (ISR-UC), and Jérôme Mendes, from CEMMPRE (both members of the Associated Laboratory ARISE), won 1st place in the REN 2023 Prize. Nuno Mendes' dissertation, titled “Federated Learning for the Prediction of Net Energy Demand in Communities of Buildings”. O REN Prize, created in 1995, is awarded every year to the best master's thesis in energy developed at Portuguese universities. The award ceremony took place in Lisbon at the Salão Nobre of the Ritz Hotel on November 13.
Future energy communities allow for the local optimization of resources, namely through energy trading between buildings, but for this, it is crucial to have forecasts of energy consumption and generation in buildings. Conventionally, these forecasts use historical data on net energy consumption in buildings, but they can be improved by also including private building information. Building automation systems allow for the collection of large amounts of data that can be used by forecasting systems, but this data is mostly classified as private. In this context, Federated Learning (FL) has been used in several areas with the main objective of protecting users' private data.
This dissertation proposes a new approach for predicting net energy in energy communities based on a FL system. The structure implemented includes the integration of third-party entities as data providers, and two forecasting systems (one for consumption and one for generation), both managed by the same server, which ensure the forecasting of net electricity consumption, independently, in each of the buildings belonging to the energy community. The results obtained show that the forecasting systems have achieved a high level of accuracy. Above all, they enable a high level of adaptability, for example, to seasonal variations, the entry and exit of buildings from the community or even new communities made up of other buildings.
More information about the “REN Award” is available here.