Demand-Side Management Using Deep Learning for Smart Charging of Electric Vehicles

• Published in: IEEE transactions on smart grid 2019-05, Vol.10 (3), p.2683-2691
• Main Authors: Lopez, Karol Lina, Gagne, Christian, Gardner, Marc-Andre
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Automobiles; Batteries; Computational modeling; Decision making; Decisions; Deep learning; deep neural network; Demand side management; Driving; Dynamic programming; Electric power demand; Electric vehicle charging; Electric vehicles; Electrical loads; Electricity consumption; Electricity pricing; Energy conservation; Energy costs; Energy management; Machine learning; Neural networks; Petroleum; Pricing; Real time; Real-time systems; Smart charging
• ISSN: 1949-3053; 1949-3061
• DOI: 10.1109/TSG.2018.2808247

The use of electric vehicles (EVs) load management is relevant to support electricity demand softening, making the grid more economic, efficient, and reliable. However, the absence of flexible strategies reflecting the self-interests of EV users may reduce their participation in this kind of initiative. In this paper, we are proposing an intelligent charging strategy using machine learning (ML) tools, to determine when to charge the EV during connection sessions. This is achieved by making real-time charging decisions based on various auxiliary data, including driving, environment, pricing, and demand time series, in order to minimize the overall vehicle energy cost. The first step of the approach is to calculate the optimal solution of historical connection sessions using dynamic programming. Then, from these optimal decisions and other historical data, we train ML models to learn how to make the right decisions in real time, without knowledge of future energy prices and car usage. We demonstrated that a properly trained deep neural network is able to reduce charging costs significantly, often close to the optimal charging costs computed in a retrospective fashion.