As renewable generation continues to reshape the operating profile of power systems, grid operators are confronting a tighter and more volatile frequency-regulation landscape. Wind and solar resources reduce the share of synchronized rotating mass, while their variable output increases the active-power imbalance that automatic generation control must absorb. In this environment, the electric vehicle is moving from a passive charging load to a potentially high-value flexibility resource. A new body of research focused on regional grid frequency regulation shows how large-scale electric vehicle aggregation, market-based pricing, probabilistic dispatchable-power assessment and real-time simulation can be combined into a coordinated control framework that respects user preferences while improving grid performance.
The research agenda is timely. By the end of 2024, renewable energy installed capacity in China reached about 1,889 GW, representing roughly 56 percent of total installed capacity and exceeding thermal generation for a second consecutive year. Solar capacity reached about 890 GW, up 45.2 percent year on year, while wind capacity reached about 520 GW, up 18.0 percent. Renewable generation exceeded 3.46 trillion kWh, about 35 percent of total electricity generation. At the same time, the electric vehicle stock reached 31.4 million units, or 8.90 percent of the total vehicle fleet, with pure electric vehicles accounting for 70.34 percent, or about 22.09 million units. These two trends create both the need for new frequency-regulation resources and the physical basis for a large-scale electric vehicle response.
1. Frequency Regulation Becomes a Strategic Constraint for High-Renewable Grids
The frequency-regulation challenge is not only a matter of adding more reserve. It is a matter of adding the right kind of reserve at the right time and at the right cost. Traditional thermal units remain essential, but their ramp rates, response delays and minimum stable output levels limit their ability to follow fast renewable fluctuations. Energy storage can respond quickly, but its duration and investment cost remain constraints. The electric vehicle offers a different profile: it can absorb power by increasing charging, reduce demand by decreasing charging, and in vehicle-to-grid mode can even discharge back to the grid. When aggregated by an electric vehicle aggregator, these small individual batteries become a dispatchable resource with response speeds that are well suited to secondary frequency regulation.
The market and control architecture matters as much as the physical capability. The research examines a provincial frequency-regulation market organized through day-ahead quotation, day-ahead pre-clearance, intraday clearance and real-time dispatch. In the day-ahead stage, the dispatch center determines the frequency-regulation capacity requirement and publishes market information. Frequency-regulation providers submit capacity and mileage price offers before the cutoff time. After the day-ahead unit commitment is settled, a pre-clearance sequence is formed. In the intraday stage, the dispatch center organizes formal clearance according to the pre-clearance order, from low to high, until the total awarded capacity meets the requirement. Compensation includes mileage payment and capacity payment, and the performance score links payment to regulation speed, precision and response time. Under the market rules cited in the study, the performance weights are 0.4 for regulation rate, 0.3 for regulation precision and 0.3 for response time.
This market design is important for the electric vehicle because it allows an aggregator to bid on behalf of many small batteries. The electric vehicle aggregator becomes the interface between the wholesale frequency-regulation market and individual electric vehicle users. It converts a system-level control signal into charging-power adjustments at the charger level. It also converts market revenue into user incentives. Without such an interface, individual electric vehicle owners would be too small to meet market entry requirements and too dispersed to be controlled in real time.
2. Evolutionary Game Pricing Defines Electric Vehicle Compensation Zones
Compensation is a central driver of electric vehicle participation. If the mileage price is too low, electric vehicle users may prefer to satisfy their charging needs rather than provide regulation. If the price is too high, the grid may bear unnecessary cost. The research therefore models the interaction between the grid and electric vehicle users as an evolutionary game. The grid can choose to participate in the interaction or not, and the electric vehicle user can choose to participate or not. The payoff matrix captures the grid’s benefit from reduced thermal reserve, the electric vehicle user’s regulation revenue, the loss from unmet interaction and the opportunity cost of being called but not dispatched. The replicator dynamics describe how the share of participating electric vehicle users and the grid’s willingness to engage evolve over time.
The results identify a feasible mileage-price region for electric vehicle frequency regulation. The overall feasible domain is reported as 1.45 to 3.61 CNY per MW for mileage price and 8.87 to 12.40 MW for dispatchable power. Within this region, the interaction willingness of both sides can evolve toward a stable equilibrium. The grid side benefits more as the mileage price rises because a higher price attracts more electric vehicle capacity and reduces overall regulation cost. The electric vehicle side, however, can earn the required return with less dispatched power as the price rises, so its dispatched-power contribution may decline at higher prices. This trade-off is the economic core of the electric vehicle frequency-regulation market.
The research further separates electric vehicle users by risk preference: conservative, neutral and aggressive. Conservative users are less attracted by mileage payments and tend to preserve charging demand. Neutral users balance regulation revenue and charging needs. Aggressive users are more sensitive to price and are willing to provide larger power adjustments. The price intervals differ accordingly.
| Electric vehicle user type | Dispatchable power range (MW) | Mileage price range (CNY/MW) |
|---|---|---|
| Conservative | 8.57 to 8.87 | 2.04 to 2.22 |
| Neutral | 8.57 to 12.45 | 1.45 to 3.61 |
| Aggressive | 8.57 to 21.95 | 1.00 to 5.00 |
The table shows that a wider dispatchable-power range is associated with a wider mileage-price range. Aggressive electric vehicle users can accept a broader price interval because they are more willing to adjust charging power and because their larger regulation capability brings more value to the grid. Conservative electric vehicle users require a narrower and generally higher price band to be induced to participate. This finding gives the dispatch center a practical basis for differentiated compensation rather than a single uniform price for all electric vehicle users.
3. Copula-Based Assessment Captures Electric Vehicle Dispatchable Power
A central technical difficulty is that the electric vehicle is not a deterministic generator. Its dispatchable power depends on state of charge, plug-in time, expected departure time, charging power limit, battery capacity, user behavior and market price. These variables are not independent. A simple correlation coefficient cannot capture the joint dependence among state of charge, mileage price and dispatchable power. The research therefore introduces a multivariate Gaussian Copula model to describe the joint distribution of electric vehicle schedulable power, state of charge and frequency-regulation mileage price. The Copula approach separates the marginal distributions of each variable from the dependence structure among them, which allows the model to represent nonlinear and multidimensional relationships more accurately.
The study first defines the physical dispatchable-power range of an individual electric vehicle. When the system frequency rises, the electric vehicle can increase charging power to absorb surplus energy. When the system frequency falls, the electric vehicle can reduce charging power to release demand. The upward and downward ranges are constrained by the maximum charging power, the baseline charging power, the state of charge and the battery limits. The model also respects the time coupling of charging: the energy required to reach the target state of charge within the remaining plug-in period limits how much upward regulation can be provided. This prevents the optimization from promising regulation capacity that would violate the user’s charging needs.
The research then applies differentiated control strategies according to user risk preference. For conservative electric vehicle users, the dispatchable power is modeled as a convex function of state of charge: participation willingness rises only when state of charge is relatively high. For neutral electric vehicle users, the dispatchable power varies linearly with state of charge. For aggressive electric vehicle users, the dispatchable power follows a steeper convex relationship, allowing larger adjustments across a wider state-of-charge range. These strategies convert the abstract notion of risk preference into explicit power-adjustment curves that can be used by the aggregator.
| Electric vehicle type | Marginal distribution of dispatchable power |
|---|---|
| Conservative | Gamma(4, 0.35) |
| Neutral | N(2.75, 0.7) |
| Aggressive | Beta(10, 2.5) |
The Copula model produces a joint density of dispatchable power, state of charge and mileage price. The study compares Clayton, Gaussian and Gumbel Copula functions and examines tail dependence. For conservative electric vehicle users, the lower-tail dependence between dispatchable power and mileage price is stronger than the upper-tail dependence, indicating that price has limited attraction and that these users prefer lower regulation power to protect charging demand. For neutral electric vehicle users, the upper-tail and lower-tail dependence are more balanced, reflecting a trade-off between regulation revenue and charging needs. For aggressive electric vehicle users, the upper-tail dependence is stronger, indicating that higher mileage prices can induce larger dispatchable power.
The study also compares Gaussian Copula and T-Copula against the empirical rank correlation. The Gaussian Copula shows smaller sample error and better represents the dependence among state of charge, mileage price and electric vehicle dispatchable power. This is why the multivariate Gaussian Copula is selected for the electric vehicle schedulable-power assessment. The model is not merely a statistical description; it becomes the basis for allocating regulation commands among individual electric vehicles according to their risk-adjusted capability.
| Control approach | Upward dispatchable power (MW) | Downward dispatchable power (MW) |
|---|---|---|
| Cluster unified control | 13.79 | 13.08 |
| Proposed risk-preference model | 14.71 | 14.22 |
For a fleet of 5,000 electric vehicles, the proposed risk-preference-aware control approach raises upward dispatchable power by 11.1 percent and downward dispatchable power by 3.72 percent compared with cluster unified control. The improvement is not obtained by ignoring user needs. It is obtained by recognizing that different electric vehicle users have different willingness to adjust charging power and by allowing the aggregator to allocate regulation tasks accordingly. This is a significant result for large-scale electric vehicle participation because it shows that user heterogeneity can be an asset rather than a source of uncertainty.
4. Two-Layer Optimization Coordinates Thermal Units and Electric Vehicle Aggregators
The market and assessment models are embedded in a two-layer optimization framework for intraday regional grid frequency regulation. The upper layer is executed by the dispatch center and minimizes the total frequency-regulation cost while treating regional frequency deviation as a probabilistic constraint. The lower layer is executed by the electric vehicle aggregator and allocates the aggregator-level command among individual electric vehicles using the Copula-based dispatchable-power model. This structure reflects the actual hierarchy of grid operation: the dispatch center does not control each electric vehicle directly, and the electric vehicle aggregator does not set the regional frequency-regulation requirement.
The upper-layer objective includes mileage cost and capacity cost. Mileage cost reflects the compensation for actual power adjustment, while capacity cost reflects the opportunity cost of holding reserve capacity. The constraints include thermal unit output limits, thermal ramp limits, electric vehicle aggregator power limits, frequency-control limits and transmission-related capacity limits. To avoid excessive concentration of regulation duty, the model limits a single thermal unit to no more than 10 percent of the total requirement, a single electric vehicle aggregator to no more than 5 percent, and the total electric vehicle aggregator capacity to no more than 30 percent. These limits are consistent with the market rules referenced in the study and help prevent the frequency-regulation process from causing large shifts in power flow.
The probabilistic constraint is formulated as a chance constraint. The regional frequency deviation after regulation must remain within the allowed range with a probability no less than a specified confidence level. The frequency-regulation demand is treated as a random variable with a normal distribution. The chance constraint is transformed into a mixed-integer linear form through scenario discretization, binary variables and the big-M method. The confidence level is not chosen arbitrarily. The study calculates satisfaction levels for regulation cost and frequency deviation across different confidence levels and selects 90 percent as a balanced choice that supports both economic efficiency and reliability.
| Regulation scenario | Capacity cost (CNY) | Mileage cost (CNY) | Total cost (CNY) |
|---|---|---|---|
| Thermal units only | 2,100.00 | 3,167.85 | 5,267.85 |
| Thermal units plus electric vehicles | 1,768.71 | 2,255.93 | 4,024.64 |
The cost comparison shows a 23.6 percent reduction in total frequency-regulation cost when electric vehicles are introduced. The reason is straightforward: electric vehicle regulation has a lower marginal cost than thermal regulation in the modeled market, so the optimization dispatches electric vehicle capacity first and relies on thermal units when electric vehicle capacity is insufficient. The result does not imply that electric vehicles replace thermal units. It implies that electric vehicles can substitute for part of the thermal reserve and reduce the cost of maintaining frequency quality.
The lower-layer allocation uses the electric vehicle dispatchable-power model to distribute the aggregator command. Electric vehicles with larger risk-adjusted dispatchable power receive larger regulation tasks. In the reported case, the average allocated dispatchable power is 4.39 kW for aggressive electric vehicle users, 2.74 kW for neutral electric vehicle users and 1.41 kW for conservative electric vehicle users. This distribution reflects both capability and preference. Aggressive users have wider regulation margins and stronger price sensitivity, so they contribute more. Conservative users have narrower margins and stronger charging priority, so they contribute less. The allocation is therefore differentiated, but it remains anchored in the physical and behavioral constraints of each electric vehicle.
| Optimization method | Reported cost value (CNY) | Optimization time (s) |
|---|---|---|
| Chance-constrained programming model | 2,255.93 | 0.37 |
| Multi-objective genetic algorithm | 3,659.42 | 64.31 |
The algorithm comparison also favors the chance-constrained formulation for intraday application. The chance-constrained model reports a lower cost value and a much shorter optimization time than the multi-objective genetic algorithm. The shorter time is important because intraday secondary frequency regulation operates on a 15-minute cycle and requires repeated re-optimization. A method that takes more than one minute may be acceptable for planning studies, but it is less suitable for fast dispatch. The chance-constrained model provides a practical balance between probabilistic reliability and computational speed.
5. Real-Time Simulation Platform Links Optimization, LabVIEW and RTDS
To test the framework beyond offline calculation, the research designs a comprehensive simulation platform for large-scale electric vehicle participation in regional grid frequency regulation. The platform combines a MATLAB-based optimization module, a LabVIEW-based dispatch control interface, an Access database and an RTDS real-time simulation model. The optimization module solves the two-layer dispatch problem. The LabVIEW interface mimics the dispatch center, displays system states and sends power-control signals. The RTDS model represents the regional grid, thermal units, charging stations and electric vehicle batteries. Data communication is implemented through UDP and a GTNET interface so that control signals can be exchanged between the digital control system and the real-time simulator.
The platform is designed around three functional modules. The dispatch control module monitors system operation, identifies frequency deviation and issues regulation commands. The optimization module evaluates thermal and electric vehicle regulation capabilities and determines the optimal dispatch strategy. The real-time simulation module reproduces the regional grid response and allows the effect of electric vehicle charging-power adjustments on frequency to be observed. The platform also supports data storage, query and visualization, which makes it possible to examine the dynamic process of electric vehicle participation rather than only the final optimization result.
The single-area test uses an IEEE 39-bus system with electric vehicle charging stations connected at selected buses through equivalent aggregation nodes. The simulation step is 50 microseconds and the time window is 30 seconds. Each charging station is rated at 10 kV and 0.5 MW, with a standard frequency of 50 Hz, a time constant of 0.015, a proportional coefficient of 50 and an integral coefficient of 0.01. The electric vehicle model uses a 60 kWh battery, a maximum charging power of 11 kW, a maximum discharging power of minus 11 kW, an initial state of charge following N(0.5, 0.1), maximum and minimum state-of-charge limits of 0.9 and 0.1, a manual control mode and a charging efficiency of 0.85. These parameters allow the simulation to represent both the energy constraints and the power constraints of the electric vehicle fleet.
| Parameter | Value |
|---|---|
| Charging station rated voltage | 10 kV |
| Charging station rated power | 0.5 MW |
| Standard frequency | 50 Hz |
| Charging station time constant | 0.015 |
| Proportional coefficient | 50 |
| Integral coefficient | 0.01 |
| Electric vehicle battery capacity | 60 kWh |
| Maximum charging power | 11 kW |
| Maximum discharging power | -11 kW |
| Initial state of charge | N(0.5, 0.1) |
| Maximum state of charge | 0.9 |
| Minimum state of charge | 0.1 |
| Charging efficiency | 0.85 |
6. Single-Area Tests Show Faster Frequency Recovery
The single-area simulation introduces a load disturbance and observes the frequency response after primary and secondary regulation. Without electric vehicle aggregator participation, the thermal unit response is delayed by about 6 seconds because of ramp and control constraints. The frequency stabilizes at about 50.11 Hz after primary regulation, and it takes about 11 seconds to recover to 50.05 Hz after secondary regulation. With electric vehicle aggregator participation, the electric vehicle fleet begins to respond shortly after receiving the command, and the recovery time to 50.05 Hz is reduced to about 5 seconds. As the electric vehicle power penetration increases, the recovery time becomes shorter.
The performance improvement is quantified through the market performance coefficients. At an electric vehicle power penetration of 30 percent, the regulation rate coefficient rises by 70.65 percent, the regulation precision coefficient rises by 15 percent and the response time coefficient rises by 6.37 percent. The composite frequency-regulation performance index increases from 0.79 for thermal-only regulation to 1.09 for thermal-plus-electric-vehicle regulation, an improvement of 36.34 percent. This result is consistent with the physical expectation that a fast, distributed electric vehicle fleet can reduce the duration of frequency deviation and lower the minimum frequency nadir.
| Regulation scenario | Regulation rate coefficient | Regulation precision coefficient | Response time coefficient | Composite performance index |
|---|---|---|---|---|
| Thermal units only | 0.92 | 0.60 | 0.80 | 0.79 |
| Thermal units plus electric vehicles | 1.57 | 0.69 | 0.85 | 1.09 |
The study also compares conservative, neutral and aggressive electric vehicle fleets under the same initial conditions. Because the risk-preference-aware model gives aggressive electric vehicle users a larger dispatchable-power range, the aggressive fleet recovers to the allowed frequency range in about 7 seconds, and its minimum frequency deviation is about 0.039 Hz, the smallest among the three groups. This does not mean that conservative users are unimportant. It means that the aggregator can rely on aggressive users for faster and larger adjustments while preserving conservative users for smaller, less disruptive adjustments. Such differentiated participation is essential for maintaining user acceptance over repeated dispatch events.
The simulation further examines individual electric vehicle behavior from three perspectives: real-time charging-power adjustment, state-of-charge evolution and estimated remaining charging time. Aggressive electric vehicle users show larger charging-power variations and stronger participation, while conservative electric vehicle users show smaller variations and a stronger priority on charging. The state-of-charge trajectories and remaining charging times reflect this difference. The aggregator is therefore not only minimizing system cost; it is also respecting the user-side consequences of frequency-regulation participation.
7. Interconnected Systems Benefit from Electric Vehicle Participation
The research extends the analysis to a two-area four-machine interconnected system. Area A has a load of 950 MW, Area B has a load of 1,767 MW and the tie-line power is 400 MW. Electric vehicle charging stations are aggregated and connected to a selected bus in Area A. The study examines three renewable penetration scenarios: no renewable generation, 10 percent renewable penetration and 20 percent renewable penetration. The frequency fluctuation becomes larger as renewable penetration rises, and the need for flexible regulation capacity becomes more pronounced. Under the 20 percent renewable penetration scenario, the electric vehicle fleet is used to test coordinated frequency regulation.
In the disturbance case, a 200 MW load reduction at a bus in Area A causes the frequency to rise to 50.28 Hz. Thermal governor action brings the frequency toward 50.14 Hz, and the tie-line power reaches about 499.84 MW. In the thermal-only case, the load reference of a thermal unit in Area A is adjusted from 0.039 p.u. to 0.031 p.u., the tie-line power deviation is 33.66 MW and the frequency recovers to 50.04 Hz. In the thermal-plus-electric-vehicle case, the load reference is adjusted from 0.039 p.u. to 0.033 p.u. and the electric vehicle charging power is increased by 40 MW. The tie-line power deviation is 28.72 MW and the frequency also recovers to 50.04 Hz.
| Two-area case | Thermal unit reference change | Electric vehicle action | Tie-line power deviation (MW) | Recovered frequency (Hz) |
|---|---|---|---|---|
| Thermal only | 0.039 to 0.031 p.u. | None | 33.66 | 50.04 |
| Thermal plus electric vehicles | 0.039 to 0.033 p.u. | Charging power increased by 40 MW | 28.72 | 50.04 |
The comparison shows that electric vehicle participation reduces tie-line power fluctuation by 14.68 percent while achieving the same frequency recovery target. This is an important result for interconnected systems because large tie-line swings can propagate disturbances and force additional thermal adjustment. By absorbing part of the surplus or deficit locally through electric vehicle charging control, the system reduces the burden on the tie-line and on remote thermal units. The electric vehicle therefore provides not only frequency support but also a form of local flexibility that improves the quality of interconnected operation.
8. Market, Control and Simulation Evidence Point in the Same Direction
The evidence from the market model, the probabilistic assessment, the two-layer optimization and the real-time simulation is mutually reinforcing. The evolutionary game shows that a differentiated mileage-price range can align grid and electric vehicle user incentives. The Copula model shows that electric vehicle dispatchable power is jointly shaped by state of charge, price and risk preference, and that a risk-preference-aware model can unlock more regulation capability than a uniform cluster control. The two-layer optimization shows that electric vehicle participation can lower total regulation cost while maintaining a probabilistic frequency-deviation constraint. The real-time simulation shows that the resulting commands can be executed through a practical communication and control chain and that the frequency response improves in both single-area and interconnected systems.
The practical implication is that large-scale electric vehicle participation should not be treated as a single homogeneous block. An aggregator that ignores user risk preference may either overstate available capacity or underuse willing capacity. An aggregator that applies a uniform price may fail to attract conservative users and may overpay aggressive users. A dispatch center that treats frequency-regulation demand as deterministic may either over-procure reserve or expose the system to frequency risk. The integrated framework addresses these issues by combining a market price signal, a probabilistic user-response model, a chance-constrained dispatch model and a real-time control platform.
For grid operators, the value of the electric vehicle resource is not limited to energy arbitrage or peak shaving. The electric vehicle can provide fast frequency response, reduce thermal unit cycling and lower the cost of reserve. For electric vehicle users, the value is not limited to a small payment. The value includes a measurable return for allowing the aggregator to adjust charging power within user-defined limits. For aggregators, the value lies in the ability to bid a reliable capacity product that reflects the heterogeneous preferences of the electric vehicle fleet. When these values are aligned, the electric vehicle becomes a scalable grid asset rather than a source of unmanaged load growth.
9. Outlook: From Private Cars to Broader Electric Vehicle Behavior
The study also identifies directions for further work. The current analysis focuses mainly on private electric vehicle participation. In practice, electric vehicle fleets include buses, taxis, logistics vehicles and other commercial vehicles. Their usage patterns and charging constraints differ substantially. A bus may have a rigid depot charging schedule, a taxi may have high daily mileage and limited dwell time, and a logistics vehicle may have route-dependent state of charge. These differences affect the opportunity cost of frequency-regulation participation and the acceptable compensation price. Extending the model to multiple electric vehicle types would make the aggregator’s bid more realistic and the market design more inclusive.
Another direction is the treatment of baseline charging power. The current model uses a fixed baseline power within a 15-minute dispatch interval. In reality, the baseline is dynamic because electric vehicle arrivals, departures and charging demand change over time. Accurate electric vehicle power forecasting would allow the aggregator to update the baseline and the dispatchable-power range more frequently. This would improve the reliability of the capacity offer and reduce the risk of failing to deliver regulation when the electric vehicle fleet is called. The study does not build a forecasting model, but it identifies this gap as a priority for future research.
Battery degradation and charging-mode differences also deserve more attention. Frequent charging-power adjustments can affect battery life, and the cost of degradation may influence user willingness. DC fast charging and AC slow charging have different response capabilities and different impacts on battery stress. A more complete model would include degradation cost, charger type and user travel requirements in the dispatchable-power assessment. Such a model would help the aggregator avoid promising regulation capacity that is physically available but economically or behaviorally unacceptable.
10. Key Takeaways for Electric Vehicle Frequency Regulation
- Large-scale electric vehicle aggregation can provide fast, bidirectional frequency-regulation support for regional grids with high renewable penetration.
- An evolutionary game model identifies differentiated mileage-price ranges for conservative, neutral and aggressive electric vehicle users, allowing compensation to reflect user risk preference.
- A multivariate Gaussian Copula model captures the joint dependence among electric vehicle dispatchable power, state of charge and mileage price more effectively than simple correlation methods.
- Risk-preference-aware control increases upward dispatchable power by 11.1 percent and downward dispatchable power by 3.72 percent for a 5,000 electric vehicle fleet compared with cluster unified control.
- A two-layer chance-constrained dispatch model reduces total frequency-regulation cost by 23.6 percent when electric vehicles are included, while maintaining a 90 percent confidence level for frequency-deviation control.
- Real-time simulation with LabVIEW and RTDS shows a 36.34 percent improvement in composite frequency-regulation performance and a 14.68 percent reduction in tie-line power fluctuation in the interconnected case.
- Differentiated allocation gives aggressive electric vehicle users larger average dispatchable power, neutral users intermediate power and conservative users smaller power, preserving user preferences while improving grid support.
- Future work should extend the framework to buses, taxis, logistics vehicles, dynamic baseline forecasting, battery degradation and multiple charging modes to make electric vehicle frequency regulation more realistic and scalable.
The overall message is clear: the electric vehicle is no longer only a transportation asset. With the right market signal, the right probabilistic assessment and the right control architecture, a large-scale electric vehicle fleet can become a flexible frequency-regulation resource that improves grid stability, lowers system cost and respects the preferences of electric vehicle users. The research provides an integrated path from market pricing to real-time execution, and it offers a practical reference for regional grid operators, electric vehicle aggregators and market designers seeking to scale electric vehicle participation in frequency-regulation ancillary services.