PRIMENA MAŠINSKOG UČENjA ZA PREDVIĐANjE POTROŠNjE ELEKTROENERGETSKOG SISTEMA SRBIJE

The paper describes a mathematical model for predicting the load in the Serbian electric power utiliity system. The mathematical model is based on several advanced machine learning techniques used for time series prediction and modeling of complex systems behaviour. Since complex systems have a large number of parameters that affect their operation, input big data of different types is needed to determine the correlation between the input and the system’s load. The data source used is the platform for transparency of the association of European transmission system operators (https://transparency.entsoe.eu/). By combining various statistical, mathematical, and machine learning models, we can obtain highly reliable predictions of consumption within a range of 24 hours. The model can also be used for longer time periods but with proportionally lower reliability. The model's predictions for the next 24 hours will be publicly available on our internet service, along with historical data in order to su inform about the quality of the predictions of the machine learning model (https://www.skalamerie.com/dodona/eeload.php).