Deep Learning for Smart Grid and Energy Context

Abstract

Deep learning is becoming ever more significant for potential applications in the field of smart grid and the energy transition paradigm. Energy production forecasting from renewables and real time monitoring of the smart grid, are just two examples of how advanced neural network architectures can find wide and useful use in these fields. Due to the presence of volatile production and of new consumption profiles arising from electric vehicles recharge and from the interaction with other energy vectors, more efficient forecasting tools are needed to ensure balancing services or effective self consumption in energy districts. The present paper implements deep learning techniques which can be used for these purposes. In particular, training of neural networks of LSTM type for the classification of electrical signals have been compared and CNN-LSTM and ConvLSTM hybrid techniques for power production prevision by PV plants have been proposed.