Introduction
As the global energy landscape transitions toward intermittent renewable sources like wind and solar, the electrical grid faces increasing stability challenges. To maintain balance, grid operators require "operating reserves"—capacity that can quickly inject or absorb power. Electric vehicles (EVs), acting as mobile batteries, are ideal candidates for this role. However, for carsharing operators, a conflict arises: a vehicle plugged in for grid stabilization is a vehicle unavailable for rental.
This research introduces FleetPower, an algorithm designed to transform a commercial carsharing fleet into a "Virtual Power Plant" (VPP). By optimizing the trade-off between rental income and grid-stabilization payments, FleetPower demonstrates that carsharing fleets can significantly increase profits with negligible impact on user availability.
The FleetPower Methodology
The FleetPower algorithm operates through a dual-level optimization process that balances the "value of immobility" against potential grid service revenues.
Individual Vehicle Optimization
At the individual car level, the algorithm calculates a Profit Performance Indicator (PPI) for every 15-minute interval. It evaluates three potential states:
- Rental Benefit (RB): Estimated income from a potential user, modeled via multiple linear regression based on time and location.
- Negative Regulation (NRB): The benefit of charging the vehicle when the grid has excess power, often at a discount.
- Positive Regulation (PRB): The benefit of discharging the battery back into the grid during shortages.
Fleet-Level Risk Pooling
To manage the uncertainty of rental demand, the algorithm treats the fleet as a collective. Using a "safety stock" approach, it ensures sufficient vehicle availability. If a vehicle committed to the grid is suddenly requested for rental, the VPP dynamically replaces that capacity with another idle vehicle, effectively pooling the risk across the fleet.
The model explicitly incorporates battery depreciation and charging inefficiencies to ensure that the physical costs of vehicle-to-grid (V2G) services do not outweigh the financial gains.
Data and Evaluation
The study utilized high-fidelity, real-world datasets to validate the model:
- Carsharing Data: Provided by Daimler AG, tracking 500 Smart Fortwo Electric vehicles in Stuttgart, Germany. The data includes 15-minute granularity on location, state of charge (SoC), and rental events across 54 urban districts.
- Grid Data: Secondary operating reserve auction data from Transnet BW, a major German transmission system operator.
- Evaluation Period: A seven-week training period in late 2013 followed by a four-week hold-out evaluation period in early 2014.
Key Insights: The Value of Immobility
A core insight of the study is the treatment of carsharing fees as the reference point for the opportunity cost of grid services. The algorithm is intentionally conservative; because the cost of a missed rental (~$20) far exceeds the revenue from a single grid service event (~$0.40), the model uses a 99% confidence interval to prioritize user service levels.
Interestingly, "Negative Regulation" (absorbing excess energy) emerged as the most profitable avenue. This synergy allows the operator to charge vehicles for their primary rental business while simultaneously receiving payments from the grid, all while avoiding the accelerated battery degradation associated with discharging (Positive Regulation).
Results and Future Outlook
The implementation of FleetPower yielded significant financial and operational benefits:
- Profitability: Under 2015 market conditions, gross profits increased by 7.18%.
- Future Scalability: Sensitivity analysis suggests that as battery costs decrease and grid volatility increases, profit gains could reach 12.11% by 2022.
- Operational Impact: The impact on customers was virtually non-existent. Out of over 154,000 decisions, the algorithm only refused a rental in 40 instances—a decrease in availability of less than 0.01%.
During the four-week test, the 500-car fleet successfully traded 61 MWh on the operating reserve market, proving the viability of carsharing fleets as grid-stabilizing assets.
Limitations and Future Work
While promising, the authors highlight that current market designs often require week-ahead commitments, which are poorly suited for decentralized, flexible assets like EVs. Future research will explore more real-time market structures, the impact of seasonality on battery performance, and the generalizability of the model to other geographic regions with different renewable energy profiles.
Summary updated