Deep-learning methods for forecasting electricity demand.
The project
Software for the non-linear forecast of the electrical load for meter clusters.
ACTOR proposes a solution based on Deep Neural Networks which, based on the consumption history, ancillary variables relating to scheduled events, weather conditions, and social factors, performs real-time forecasts of electricity demand for clusters of meters, with horizon both intra-day and day-by-day storms.
Electric power demand forecast
Electric power cannot be stored efficiently in large quantities. It is, therefore, necessary to produce in real-time the amount of energy required by all consumers (families and companies) and manage its transmission so that supply and demand are always in balance, thus guaranteeing continuity and safety of the provision of the service.
The management of these energy flows on the grid is called dispatching. In this context, a system for forecasting electricity consumption in real-time with high precision becomes strategic, aimed at reducing the imbalance that entirely affects end consumers.
Some studies show that electricity consumption in a household depends on geographical and climatic factors, and social factors related to the composition of the family unit, as well as on the scheduling of particular days such as holidays, strikes, and sporting events, …
ACTOR proposes a solution based on Deep Neural Networks which, based on the consumption history, ancillary variables relating to scheduled events, weather conditions, and social factors, performs real-time forecasts of electricity demand for clusters of meters, with horizon both intra-day and day-by-day.
The predictor developed has a percentage error of less than 4% for a generic cluster of a few dozen meters and less than 2% for a regional cluster (around 200-300 meters).
References
The solutions provided by ACTOR for electrical power forecast have been implemented in E4SIGHT software of NECTAWARE