Machine learning approach for electric vehicle availability forecast to provide vehicle-to-home services

• Main Authors: Donovan Aguilar-Dominguez, Jude Ejeh, Alan D.F. Dunbar, Solomon F. Brown
• Format: Article
• Language: English
• Published: Elsevier 2021-05-01
• Series: Energy Reports
• Subjects: Electric vehicle; Optimisation; Machine learning; Vehicle-to-grid
• Online Access: http://www.sciencedirect.com/science/article/pii/S2352484721001517

In this study, we propose a machine learning (ML) model to predict the availability of an electric vehicle (EV) providing vehicle to home (V2H) services. Electric vehicles are able to store and give back energy directly to consumers and/or the grid using V2H and/or vehicle to grid (V2G) technologies. However, there is a limited understanding of what impact vehicle availability has on the its capacity to engage in such services. Using five different vehicle usage profiles, classified by the number of trips made per week, the machine learning model proposed is used to predict the availability of an EV. An optimisation model is then used on each profile to obtain the minimum electricity bill for each profile class assuming V2H service provision. PV generation providing power to the house was also considered. The ML model had an accuracy of over 85% and R2value of 0.78 in predicting the location and distance travelled for the EV respectively. Final results showed that the less an EV is used for travelling, the greater its availability to participate in V2H services. Also, all categories of EV user benefited from reduced power bills when deploying V2H. An electricity cost reduction of at least 46% on average was obtained when V2H is implemented with an agile electricity price structure regardless of the level of vehicle usage.