SUSTAINABLE ENERGY GRIDS & NETWORKS, sa.29, ss.2-11, 2022 (SCI-Expanded)
In
order for vehicle-to-grid (V2G) services to participate in the power and energy
market, they must provide as much aggregated capacity as the market needs. In
order to provide this capacity, the population of electric vehicle batteries is
used. For this participation it is necessary to estimate the available
capacity, which makes it possible to reliably distribute the existing reserves
in the future. In
this study, Long Short Term Memory (LSTM) and Nonlinear Autoregressive Neural
Network (NAR) were used and developed to predict the next 59 hours aggregated
available capacity (AAC) of a small fleet of 7 electric vehicles for a ten-day travel
adapted from 72 real driving records. Market activities were simulated to
include the delivery of reserves to meet the needs and included in the dataset.
The ability of the developed LSTM deep learning network and NAR machine
learning networks to successfully adapt their predictions to such market events
has been demonstrated. The authors highlight the conclusion that this
capability is critical to the viability and success of future V2G services by
supporting multiple market events.