Energy Demand Prediction in a Charge Station: A Comparison of Statistical Learning Approaches

Hélène Le Cadre (1) , Cédric Auliac (2)

1 CMA - Centre de Mathématiques Appliquées
2 LIMA (CEA, LIST) - Laboratoire Information, Modèles, Apprentissage (CEA, LIST)

In this article, we compare the performances of 5 learning techniques: artificial neural networks, support vector machines, ARIMA processes and regret based methods. They have been tested over real database which can be associated with the energy demand generated by electric vehicles wishing to reload, in a specific charge station. Using this generic database, our simulations highlight the fact that regret based methods clearly outperform the other learning approaches. This class of methods is all the more interesting as it enables the introduction of game theory to model the interdependences between the agents composing the ecosystem and provides economic guidelines.

Mots clés: Regret algorithms; regret; Supervised Learning; Electric Vehicle Demand; online learning; energy demand; charge station; smart grid; artificial intelligence; ARIMA; machine learning; neural network; support vector machine

Domaines: Apprentissage [cs.LG]; Intelligence artificielle [cs.AI]; Réseau de neurones [cs.NE]; Informatique et théorie des jeux [cs.GT]; Energie électrique

Hélène Le Cadre, Cédric Auliac. Energy Demand Prediction in a Charge Station: A Comparison of Statistical Learning Approaches. European Electric Vehicle Congress 2012, Nov 2012, Bruxelles, Belgium. ⟨hal-00741218v2⟩