With the rapid adoption of electric vehicles (EVs) worldwide, particularly in regions like China where EV penetration is accelerating, the integration of large-scale EV aggregations (EVAs) into power systems has become a critical research focus. The unique idle energy storage characteristics of EVs, when harnessed through vehicle-to-grid (V2G) technology, offer significant potential for enhancing grid stability and efficiency. However, accurately quantifying the dispatchable potential of EVAs across different geographical areas and time periods remains a complex challenge. Existing control systems often fail to fully account for the holistic nature of power systems, leading to suboptimal utilization of EV flexibility. In this paper, we address these issues by proposing a novel method for constructing a multi-stage electric vehicle dispatchable region (MEVDR) that incorporates user decision-dependent behaviors. This approach divides the dispatchable region into two distinct components: the dispatchable energy region (DER) and the dispatchable power region (DPR), providing a comprehensive reflection of the energy and power scheduling characteristics of EVAs over specific time intervals.
The proliferation of electric vehicles, especially in China, has introduced both opportunities and challenges for power systems. While EVs can serve as distributed energy resources, their uncontrolled charging and discharging can lead to grid instability, voltage fluctuations, and increased peak loads. To mitigate these effects, it is essential to develop accurate models that capture the aggregate behavior of EVAs and integrate them into coordinated control frameworks. Traditional methods for assessing EV dispatchable potential often overlook the dual nature of energy and power scheduling capabilities, leading to incomplete evaluations. Moreover, regional variations in EV usage patterns, such as those between office, commercial, and residential areas, further complicate the modeling process. Our work aims to bridge these gaps by introducing a systematic MEVDR construction method that leverages probabilistic modeling and multi-level coordination.
In this study, we utilize the Gaussian mixture model (GMM) to cluster EV data from different regions and time periods, fitting probability density functions to characterize the behavior of various EV clusters. This allows us to construct and analyze the MEVDR for EVAs in diverse scenarios, highlighting differences in dispatchable potential and their implications for power system operations. Building on this foundation, we develop a vehicles-garage-grid multi-level coordinated control system (VGGMCCS) that incorporates the MEVDR model and addresses the characteristics of multiple system layers, including users, garage operators, and the grid. By integrating ReLU functions and model decomposition iterative methods, we optimize control strategies to balance the interests of all stakeholders while improving solution efficiency. Through case studies based on an improved IEEE 33-node system and garage model, we demonstrate the effectiveness of our proposed approach in reducing user costs, enhancing garage revenues, and improving grid operational efficiency.
The remainder of this paper is organized as follows. First, we detail the construction of the MEVDR, including the DER and DPR components, and explain how user behavior characteristics are incorporated. Next, we present the multi-regional analysis using GMM clustering, showcasing the variations in MEVDR across different areas and time periods. We then describe the VGGMCCS framework, outlining its hierarchical structure and optimization objectives. Finally, we provide a comprehensive case study comparing our method with alternative strategies, followed by conclusions and future research directions.
Construction of the Multi-stage Electric Vehicle Dispatchable Region
The MEVDR is designed to evaluate the scheduling potential of individual EVs and EVAs over specific time intervals, considering key variables such as grid connection time, disconnection time, initial state of charge (SOC), target SOC, rated battery capacity, and the user-acceptable SOC lower limit. This model operates under several assumptions: scheduling boundaries are set at midnight, with cross-day charging handled as separate units; the day is divided into 96 time slots (each 15 minutes) for evaluation; and each EV connects to the grid only once per day, with continuous charging and discharging processes. The MEVDR is subdivided into the dispatchable energy region (DER) and dispatchable power region (DPR), each capturing distinct aspects of EV flexibility.
For the DER, we establish constraints based on user demand, including the as soon as possible (ASAP) and latest response (LR) constraints. The ASAP constraints define the minimum time and maximum energy required for an EV to charge to the user-acceptable upper SOC limit, while the LR constraints specify the latest charging time and maximum output energy to meet user needs. For an electric vehicle $i$, the ASAP time constraint $T_i^{ ext{ASAP}}$ is given by:
$$T_i^{ ext{ASAP}} = \frac{(S_i^{ ext{UB}} – S_i^{ ext{arrive}}) E_i}{\sum_{k=1}^{n} P_i^C(k) \eta_C}$$
where $S_i^{ ext{UB}}$ is the upper SOC limit acceptable to the user, $S_i^{ ext{arrive}}$ is the SOC at grid connection, $E_i$ is the rated battery capacity, $P_i^C(k)$ is the charging power in time slot $k$, and $\eta_C$ is the charging efficiency. The ASAP energy constraint $E_i^{ ext{ASAP}}$ is then:
$$E_i^{ ext{ASAP}} = \int_{t_i^a}^{t_i^a + T_i^{ ext{ASAP}}} P_i^C(t) \eta_C \, dt$$
with $S_i^{ ext{LB}} \leq S_i^{ ext{UB}} – \frac{E_i^{ ext{ASAP}}}{E_i}$, where $S_i^{ ext{LB}}$ is the lower SOC limit. The user-defined SOC limits are determined as:
$$S_i^{ ext{UB}} = \max\left(S_i^{ ext{arrive}} + (S_i^{ ext{tar}} – S_i^{ ext{arrive}}) \frac{R_i^{ ext{avg}}}{R_i^{ ext{MAX}}}, S_i^{ ext{tar}}\right)$$
and
$$S_i^{ ext{LB}} = \min\left(S_i^{ ext{arrive}}, \frac{P_i^C (t_i^l – t_i^a) + (S_i^{ ext{tar}} – S_i^{ ext{arrive}}) E_i}{2}\right)$$
where $S_i^{ ext{tar}}$ is the target SOC, $R_i^{ ext{avg}}$ is the average daily driving distance, and $R_i^{ ext{MAX}}$ is the full-charge range. The LR constraints are similarly derived, with the LR time constraint $T_i^{ ext{LR}}$ expressed as:
$$T_i^{ ext{LR}} = \frac{S_i^{ ext{tar}} E_i + E_i^{ ext{LR}}}{P_i^C \eta_C}$$
and the LR energy constraint $E_i^{ ext{LR}}$ as:
$$E_i^{ ext{LR}} = \min\left(\int_{t_i^a}^{T_i^{ ext{LR}}} (P_i^{C, ext{MAX}} + P_i^{D, ext{MAX}}) \, dt, \frac{P_i^{C, ext{MAX}} \eta_C}{P_i^{D, ext{MAX}} + P_i^{C, ext{MAX}} \eta_C}\right)$$
These equations collectively define the boundaries of the DER, enabling a clear visualization of the energy scheduling range for individual EVs and EVAs.
For the DPR, we introduce the concept of dispatchable flexibility (DF) constraints, inspired by resilience assessment methods. The DF upper constraint $\phi_i^{ ext{UB}}$ and lower constraint $\phi_i^{ ext{LB}}$ for an electric vehicle $i$ are:
$$\phi_i^{ ext{UB}} = \int_{t_0}^{t_s} P_i^{ ext{UB}}(t) \, dt, \quad P_i^{ ext{UB}}(t) \leq \max(P_i^C(t))$$
and
$$\phi_i^{ ext{LB}} = \int_{t_0}^{t_s} P_i^{ ext{LB}}(t) \, dt, \quad P_i^{ ext{LB}}(t) \leq \max(P_i^D(t))$$
where $P_i^{ ext{UB}}(t)$ and $P_i^{ ext{LB}}(t)$ represent the schedulable input and output power, respectively. The charging power profile for an EV is modeled based on SOC thresholds, with different power levels during various charging phases. For example, the charging power $P_i^C(t)$ can be expressed as:
$$P_i^C(t) = \begin{cases}
P_i^{C, ext{MAX}} \frac{S_{i,t} + 22.5}{10} & \text{if } S_{i,t} < S_i^{ ext{point}} \\
P_i^{C, ext{MAX}} & \text{if } S_i^{ ext{point}} \leq S_{i,t} < S_i^{ ext{tar}} \\
P_i^{C, ext{MAX}} \frac{46 – S_{i,t}}{10} & \text{if } S_{i,t} \geq S_i^{ ext{tar}}
\end{cases}$$
where $S_{i,t}$ is the SOC at time $t$, and $S_i^{ ext{point}} = \max(S_i^{ ext{arrive}}, S_i^{ ext{LB}})$. This model captures the nonlinear charging behavior of EVs, which is crucial for accurate DPR construction.
For EVAs, the DER and DPR are derived using Minkowski addition of individual EV constraints. The EVA DER is defined by aggregating ASAP and LR constraints across the cluster. The EVA ASAP energy constraint $E_{ ext{EVA}}^{ ext{ASAP}}$ is:
$$E_{ ext{EVA}}^{ ext{ASAP}} = \max_{t \in T} \left( \int_{t_i^a}^{t} \Delta E_{ ext{EVA}}(t) \, dt \right)$$
where $\Delta E_{ ext{EVA}}(t)$ is the energy difference between consecutive time slots. The EVA ASAP time constraint $T_{ ext{EVA}}^{ ext{ASAP}}$ is then:
$$T_{ ext{EVA}}^{ ext{ASAP}} = T_{ ext{span}} + \int_{t_i^a}^{T_{ ext{end}}} u(T) \, dT – \frac{E_{ ext{EVA}}^{ ext{ASAP}}}{\int_{t_i^a}^{t} P_{ ext{EVA}}^C(t) \, dt}$$
where $u(T)$ is the unit step function, $T_{ ext{span}}$ is the update step (15 minutes), and $T_{ ext{end}}$ is the end of the schedulable period. Similarly, the EVA LR energy constraint $E_{ ext{EVA}}^{ ext{LR}}$ is:
$$E_{ ext{EVA}}^{ ext{LR}} = \max_{t \in T} \left( \sum_{i \in \phi_{ ext{EV}}} c_{i,t} E_i^{ ext{LR}} \right)$$
and the EVA LR time constraint $T_{ ext{EVA}}^{ ext{LR}}$ is:
$$T_{ ext{EVA}}^{ ext{LR}} = t_i^l – \int_{t_i^a}^{T_{ ext{end}}} u(T – t_i^l) \, dT – \frac{E_{ ext{EVA}}^{ ext{LR}}}{\int_{t_i^a}^{t} P_{ ext{EVA}}^C(t) \, dt}$$
The EVA DPR is constructed by summing the DF constraints of individual EVs:
$$\phi_{ ext{EVA}}^{ ext{UB}} = \sum_{i \in \phi_{ ext{EV}}} \int_{t_0}^{t_s} P_i^{ ext{UB}}(t) \, dt, \quad P_{ ext{EVA}}^{ ext{UB}}(t) \leq P_{ ext{EVA}}^{C, ext{MAX}}(t)$$
and
$$\phi_{ ext{EVA}}^{ ext{LB}} = \sum_{i \in \phi_{ ext{EV}}} \int_{t_0}^{t_s} P_i^{ ext{LB}}(t) \, dt, \quad P_{ ext{EVA}}^{ ext{LB}}(t) \leq P_{ ext{EVA}}^{D, ext{MAX}}(t)$$
where $P_{ ext{EVA}}^{C, ext{MAX}}(t)$ and $P_{ ext{EVA}}^{D, ext{MAX}}(t)$ are the maximum aggregate charging and discharging powers of the EVA. These formulations allow for a comprehensive representation of the power scheduling potential of EVAs over time.
Multi-regional Analysis of MEVDR Using Gaussian Mixture Model
To account for regional and temporal variations in EV behavior, we employ the Gaussian mixture model (GMM) to cluster EV data from different areas and time periods. The GMM approximates the probability distribution of EV parameters as a weighted sum of Gaussian components, enabling a probabilistic characterization of EVA behavior. The probability density function $p(x)$ for a variable $x$ is given by:
$$p(x) = \sum_{k=1}^{K} \omega_k \mathcal{N}(x; \mu_k, \sigma_k^2)$$
where $K$ is the number of components, $\omega_k$ is the weight of the $k$-th component, and $\mathcal{N}(x; \mu_k, \sigma_k^2)$ is the Gaussian density function with mean $\mu_k$ and variance $\sigma_k^2$. The parameters are estimated using the expectation-maximization (EM) algorithm, which iteratively updates the weights, means, and variances to fit the observed data.
We focus on three distinct regions: office areas, commercial areas, and residential areas, each further divided into weekday and weekend states. This “3 region, 2 state” approach allows for a detailed analysis of MEVDR variations. For each region and state, we fit GMMs to EV data such as grid connection times, disconnection times, initial SOC, and target SOC. The resulting probability density functions are used to generate representative EV clusters, for which we construct and analyze the MEVDR.
In office areas, EV behavior on weekdays is characterized by concentrated grid connection during morning hours and disconnection in the evening. The MEVDR exhibits high and stable DER and DPR during working hours, indicating strong scheduling potential. For example, the DER during peak hours may span up to 2500 kWh, with DPR limits of ±200 kW. On weekends, however, the MEVDR becomes more volatile due to reduced EV numbers and random usage patterns, with DER and DPR values dropping significantly. This highlights the importance of temporal factors in assessing EVA flexibility.
Commercial areas show even greater volatility, with MEVDR fluctuations driven by periods of high vehicle influx (e.g., during shopping hours). The DER may peak at 2000 kWh during busy periods but drop to near zero during off-hours. The DPR similarly exhibits sharp transitions, reflecting the intermittent nature of EV availability in commercial zones. This variability poses challenges for grid integration but also offers opportunities for targeted scheduling during peak demand periods.
Residential areas display more stable MEVDR patterns, particularly on weekends and during non-working hours on weekdays. The DER often remains above 1500 kWh for extended periods, with DPR limits of ±150 kW. However, during weekday morning and evening rush hours, the MEVDR may decline as EVs are used for commuting. This stability makes residential EVAs well-suited for providing consistent grid services, such as load leveling or frequency regulation.
The table below summarizes the GMM-derived parameters for EV behavior in different regions and states, illustrating the mean and variance of key variables:
| Region | State | Mean Connection Time (h) | Variance Connection Time (h) | Mean Disconnection Time (h) | Variance Disconnection Time (h) | Mean Initial SOC | Variance Initial SOC |
|---|---|---|---|---|---|---|---|
| Office | Weekday | 8.0 | 0.5 | 20.0 | 0.5 | 0.3 | 0.1 |
| Office | Weekend | 9.0 | 0.4 | 17.0 | 0.4 | 0.35 | 0.15 |
| Commercial | Weekday | 12.0 | 0.4 | 14.0 | 0.4 | 0.25 | 0.12 |
| Commercial | Weekend | 10.0 | 0.5 | 20.0 | 0.5 | 0.3 | 0.1 |
| Residential | Weekday | 0.0 | 0.4 | 8.0 | 0.4 | 0.4 | 0.08 |
| Residential | Weekend | 0.0 | 0.5 | 11.0 | 0.5 | 0.35 | 0.1 |
These regional analyses demonstrate that MEVDR can effectively capture the scheduling characteristics of EVAs across different contexts, providing valuable insights for grid operators. By understanding these variations, operators can tailor scheduling strategies to maximize the utilization of EV flexibility while maintaining grid stability.
Vehicles-Garage-Grid Multi-level Coordinated Control System Based on MEVDR
The VGGMCCS is a hierarchical framework that integrates the MEVDR model into a multi-level control structure, encompassing the grid, garage, and user layers. The system consists of an upper grid model and a lower garage model, which interact through macro-control instructions and feedback. The upper grid model, based on an IEEE 33-node network, optimizes grid-level objectives such as minimizing losses and voltage fluctuations, while the lower garage model decomposes these instructions into detailed EV charging and discharging schedules, considering user costs and garage revenues.
The upper grid model is subject to constraints including voltage limits, power flow equations, and power balance. For example, the voltage constraint at bus $j$ is:
$$V_{j, ext{min}}^2 \leq V_j^2(t) \leq V_{j, ext{max}}^2$$
and the power flow equation for branch $ij$ is:
$$V_j^2(t) = V_i^2(t) – 2(R_{ij} P_{ij}(t) + X_{ij} Q_{ij}(t)) + (R_{ij}^2 + X_{ij}^2) I_{ij}^2(t)$$
where $V_j(t)$ is the voltage magnitude at bus $j$, $R_{ij}$ and $X_{ij}$ are the resistance and reactance of branch $ij$, $P_{ij}(t)$ and $Q_{ij}(t)$ are the active and reactive power flows, and $I_{ij}(t)$ is the current magnitude. The power injection at node $j$ is:
$$P_{ ext{inj}}(j,t) = \sum_{k \in ext{idx}_v} P_{ij}(k,t)$$
where $ ext{idx}_v$ is the set of branches sending power to node $j$. The objective function of the upper grid model is to minimize grid losses and voltage deviations:
$$ ext{Obj1} = \min \left( \alpha \sum_{j \in ext{branches}} \sum_{t \in T} R(j) I^2(j,t) + \beta \sum_{j \in ext{buses}} \sum_{t \in T} (V^2(j,t) – \bar{V}^2)^2 \right)$$
where $\alpha$ and $\beta$ are weighting factors, $R(j)$ is the resistance of branch $j$, $I^2(j,t)$ is the square of the current in branch $j$ at time $t$, $V^2(j,t)$ is the square of the voltage at bus $j$, and $\bar{V}$ is the desired voltage level.
The lower garage model operates within a node of the upper grid and must adhere to constraints such as transformer capacity limits and MEVDR boundaries. The transformer load constraint is:
$$P_{ ext{load}}^{ ext{trans}}(t) \leq P_m(t)$$
where $P_{ ext{load}}^{ ext{trans}}(t)$ is the transformer load at time $t$, and $P_m(t)$ is the power limit at node $m$. The MEVDR constraints for the EVA include:
$$P_i^D(t) \leq P_{ ext{DPR}}^{ ext{LB}}(i,t), \quad P_i^C(t) \leq P_{ ext{DPR}}^{ ext{UB}}(i,t)$$
for individual EVs, and
$$\sum_{i \in ext{EV}} P_i^D(t) \leq P_{ ext{DPR}}^{ ext{LB}}(t), \quad \sum_{i \in ext{EV}} P_i^C(t) \leq P_{ ext{DPR}}^{ ext{UB}}(t)$$
for the aggregate EVA. The energy constraints are:
$$E_{ ext{DER}}^{ ext{LB}}(i,t) \leq \int (P_i^C(t) – P_i^D(t)) \, dt \leq E_{ ext{DER}}^{ ext{UB}}(i,t)$$
and
$$E_{ ext{DER}}^{ ext{LB}}(t) \leq \sum_{i \in ext{EV}} \int (P_i^C(t) – P_i^D(t)) \, dt \leq E_{ ext{DER}}^{ ext{UB}}(t)$$
The objective of the lower garage model is to minimize user costs and maximize garage revenues, with user cost minimization taking priority. The user cost objective function is:
$$ ext{Obj2} = \min \sum_{i \in ext{EV}} \sum_{t=0}^{96} \left[ c_i P_i^C(t) ext{ReLU}(f_{i,t}) p^{ ext{G2V}}(t) – P_i^D(t) ext{ReLU}(-f_{i,t}) p^{ ext{V2G}}(t) \right]$$
where $c_i$ is a cost coefficient, $f_{i,t}$ is the charging/discharging state (1 for charging, -1 for discharging), $p^{ ext{G2V}}(t)$ is the price for grid-to-vehicle service, and $p^{ ext{V2G}}(t)$ is the price for vehicle-to-grid service. The ReLU function is defined as:
$$ ext{ReLU}(x) = \begin{cases} 0 & \text{if } x < 0 \\ x & \text{if } x \geq 0 \end{cases}$$
The garage revenue objective function is:
$$ ext{Obj3} = \max \sum_{i \in ext{EV}} \sum_{t=0}^{96} \left[ P_i^D(t) ext{ReLU}(-f_{i,t}) (p^{ ext{V2G}}(t) – p(t)) – c_i P_i^C(t) ext{ReLU}(f_{i,t}) (p(t) – p^{ ext{G2V}}(t)) \right]$$
where $p(t)$ is the electricity price at time $t$. These objective functions are formulated as mixed integer programming (MIP) problems, which can be solved efficiently using commercial solvers. The VGGMCCS employs an iterative decomposition approach, where the upper and lower models are solved sequentially, with feedback loops to ensure coordination between layers.
Case Study and Comparative Analysis
To validate the proposed VGGMCCS, we conduct a case study based on an improved IEEE 33-node system, with a garage located at node 28. The garage has a transformer capacity of 500 kVA and can serve up to 50 electric vehicles, each with a maximum charging power of 36 kW and discharging power of 28 kW. The charging and discharging efficiencies are set to 0.92. The EVA behavior follows the office area weekday distribution derived from GMM clustering, with EVs of various models, such as AITO M5 EV, NIO ET5, AVATR 11, TESLA Model 3, and BYD Han EV, each with different battery capacities and power ratings.
Electricity prices are based on time-of-use tariffs from Jiangsu Province, China: 1.7697 CNY/kWh during peak hours (8:00-12:00 and 17:00-21:00), 1.3418 CNY/kWh during shoulder hours (12:00-17:00 and 21:00-24:00), and 1.0139 CNY/kWh during off-peak hours (0:00-8:00). The service fees for G2V and V2G are set as 15% of the electricity price and a compensation formula, respectively:
$$p^{ ext{G2V}} = 1.15 p(t)$$
and
$$p^{ ext{V2G}} = 0.65 + \frac{p(t) – 1.3418}{2}$$
We compare our VGGMCCS with two alternative strategies from the literature: Contrast Strategy 1, which focuses on simple EVA scheduling without MEVDR, and Contrast Strategy 2, which employs a two-stage optimization without multi-level coordination. The simulation results are summarized in the table below:
| Metric | VGGMCCS | Contrast Strategy 1 | Contrast Strategy 2 |
|---|---|---|---|
| Grid Loss Percentage | 18.90% | 28.00% | 22.76% |
| Garage Revenue (CNY) | 64.25 | 69.49 | 54.62 |
| Total User Cost (CNY) | 5,978.38 | 6,490.05 | 5,904.50 |
| Average Charging Cost per EV (CNY) | 119.57 | 129.80 | 118.09 |
| Transformer High-Load Time (hours) | 0 | 8.5 | 0 |
| Max EVA Power Fluctuation during Peak (kW) | 15.00 | 629.58 | 87.83 |
| Solution Time (seconds) | 138.41 | 90.06 | 172.09 |
The results indicate that VGGMCCS significantly reduces grid losses by 9.1% and 3.86% compared to Contrast Strategies 1 and 2, respectively. During peak hours, the improvement is even more pronounced, with loss reductions of 12.17% and 8.69%. This is achieved through better coordination of EV charging and discharging, which smooths the load profile and reduces stress on the grid. The transformer high-load time is eliminated under VGGMCCS, whereas Contrast Strategy 1 results in 8.5 hours of high load, risking transformer overload. The power fluctuations of the EVA are also minimized, with VGGMCCS reducing the maximum fluctuation by 96.36% and 82.59% compared to the contrast strategies.
From an economic perspective, VGGMCCS balances the interests of users and garage operators. User costs are reduced by 7.88% compared to Contrast Strategy 1, while garage revenues are increased by 17.63% compared to Contrast Strategy 2. This demonstrates the ability of VGGMCCS to achieve a win-win situation for all stakeholders. The solution time of VGGMCCS is reasonable, at 138.41 seconds, which is faster than Contrast Strategy 2 and only slightly slower than Contrast Strategy 1. However, the lower garage model can be solved in just 3.16 seconds, enabling rapid response to real-time changes in EV connectivity.
Detailed analysis of individual EV schedules under VGGMCCS shows more aggressive utilization of V2G capabilities, with EVs participating in grid services for longer durations during peak hours. For instance, EV #17 increases its grid participation time by 10% and 175% compared to Contrast Strategies 1 and 2, respectively. Similarly, EV #39 shows increases of 25% and 66.67%. This aggressive scheduling is balanced by the MEVDR constraints, which ensure that user SOC requirements are met, with EVs maintaining SOC above 60% for longer periods, enhancing reliability for unexpected trips.
In conclusion, the case study confirms that VGGMCCS, based on the MEVDR model, effectively improves grid operation, reduces costs, and enhances flexibility utilization. The system’s ability to coordinate multiple levels and incorporate regional behavioral differences makes it a robust solution for integrating large-scale electric vehicles into the power system, particularly in the context of China’s growing EV market.
Conclusion
In this paper, we have presented a comprehensive method for constructing and analyzing the multi-stage electric vehicle dispatchable region (MEVDR) for electric vehicle aggregations (EVAs). By dividing the dispatchable region into energy and power components (DER and DPR), we provide a holistic view of EV scheduling potential, accounting for user behavior and regional variations. The use of Gaussian mixture model (GMM) clustering allows for accurate characterization of EVA behavior across different areas and time periods, enabling tailored scheduling strategies. The vehicles-garage-grid multi-level coordinated control system (VGGMCCS) leverages the MEVDR to optimize grid operations, user costs, and garage revenues, demonstrating significant improvements over alternative strategies in case studies.
The key contributions of this work include: (1) a novel MEVDR construction method that enhances the visibility and understanding of EVA flexibility; (2) a multi-regional analysis framework that highlights the impact of spatial and temporal factors on dispatchable potential; and (3) a practical VGGMCCS that achieves multi-stakeholder benefits through efficient coordination. Future research will focus on extending the MEVDR to dynamic environments with real-time data, incorporating uncertainty in EV behavior, and exploring applications in larger-scale power systems with high renewable energy penetration. As electric vehicles continue to proliferate, particularly in China, such approaches will be essential for realizing the full potential of V2G technologies and building resilient, efficient power systems.