We got the power: Predicting available capacity for vehicle-to-grid services using a deep recurrent neural network

Shipman, Rob; Roberts, Rebecca; Waldron, Julie; Naylor, Sophie; Pinchin, James; Rodrigues, Lucelia; Gillott, Mark

Abstract

© 2021 Elsevier Ltd Vehicle-to-grid (V2G) services utilise a population of electric vehicle batteries to provide the aggregated capacity required to participate in power and energy markets. Such participation relies on the prediction of available capacity to support the reliable delivery of agreed reserves at a future time. In this work real historical trip data from a fleet of vehicles belonging to the University of Nottingham was used and a simulation developed to show how battery state-of-charge and available capacity would vary if these trips were taken in electric vehicles that were charged at simulated charging station locations. A time series forecasting neural network was developed to predict aggregated available capacity for the next 24-h period given input data from the previous 24 h and its increased predictive capability over a regression model trained using automated machine learning was demonstrated. The simulations were then extended to include delivery of reserves to satisfy the needs of simulated market events and the ability of the model to successfully adapt its predictions to such events was demonstrated. The authors conclude that this ability is of critical importance to the viability and success of future V2G services by supporting trading and vehicle utilisation decisions for multiple market events.

Citation

Shipman, R., Roberts, R., Waldron, J., Naylor, S., Pinchin, J., Rodrigues, L., & Gillott, M. (2021). We got the power: Predicting available capacity for vehicle-to-grid services using a deep recurrent neural network. Energy, 221, Article 119813. https://doi.org/10.1016/j.energy.2021.119813

• Journal Article Type: Article
• Acceptance Date: Jan 6, 2021
• Online Publication Date: Jan 9, 2021
• Publication Date: Apr 15, 2021
• Deposit Date: Jan 20, 2021
• Publicly Available Date: Jan 10, 2022
• Journal: Energy
• Print ISSN: 0360-5442
• Electronic ISSN: 1873-6785
• Publisher: Elsevier
• Peer Reviewed: Peer Reviewed
• Volume: 221
• Article Number: 119813
• DOI: https://doi.org/10.1016/j.energy.2021.119813
• Keywords: Vehicle-to-grid, V2G, Deep learning, CNN-LSTM network, Machine learning, Neural networks
• Public URL: https://nottingham-repository.worktribe.com/output/5234817
• Publisher URL: https://www.sciencedirect.com/science/article/abs/pii/S0360544221000621