EV Fleet Flexibility Estimation and Forecasting for V2X Applications

TL;DR

This paper addresses the challenge of estimating and forecasting V2X flexibility for EV fleets under owner uncertainty, introducing an aggregate polytope representation to capture time-coupled constraints as $A p <= b$. It then leverages multivariate, multi-step deep learning architectures, notably Transformer and attention-based CNN-LSTM, to forecast the aggregate constraint vector $b_{agg}$ by predicting $P_{max}^{agg}$, $C_{max}^{agg}$ and $C_{min}^{agg}$ over horizon $T$ with lead time $k$. Two case studies—long-term flexibility contracts and day-ahead scheduling under dynamic operating envelopes—demonstrate practical applicability and improved decision support for DSOs and charging-station operators. The results show robust, accurate forecasting and scalable integration into optimisation problems, highlighting potential for deploying V2X flexibility in real-world grids despite data gaps and the need for richer datasets.

Abstract

Forecasting the flexibility potential of Vehicle-to-Everything (V2X) systems is important for the future of energy networks, where the integration of renewable energy sources and electric vehicles poses significant challenges. In this paper, we present a novel method for estimating and predicting V2X flexibility potential of an EV fleet, based on an aggregate polytope representation, addressing the need for accurate and reliable forecasting methods in the realm of sustainable transportation. The method is robust against individual uncertainties of EV owners behaviours as it is applied at an aggregate level, and the reformulation of the V2X potential as a set of linear constraints allows the proposed method to be integrated into different optimisation problems and therefore be applied for diverse V2X applications. Case studies showcase the capability of the method in capturing the V2X flexibility potential and demonstrate it effectiveness for different V2X applications.

Figures (7)

• Figure 1: Overview of the EVs statistics for the considered dataset: The majority of the cars come with a SOC$_{arr}$ less than 0.2 and leave with a SOC greater than 0.8. The battery capacities are concentrated around 20 kWh with few of them in the range of 40, 80 and 100 kWh.
• Figure 2: The vectors of the constructed $b_{agg}$ from the considered dataset. $P_{min}^{agg}$ is equivalent to the negative value of $P_{max}^{agg}$ as we considered that the EVs have symmetrical charging/discharging capacities. Roughly, the aggregator can be seen as an EV with 500 kW maximum charging rate and 4000 maximum capacity.
• Figure 3: Multivariate input multi-step multivariate output with n historical data, $k$ lead time and $T$ steps.
• Figure 4: The forecast of $b_{agg}$ vectors for one day.
• Figure 5: Maximum flexibility provision for for two flexibility windows, in the period of feasibility study (left) and period of flexibility activation (right).