With the ever-increasing shortage of fossil energy and the continuous deterioration of environmental quality, energy conservation and emission reduction have become the core objectives of the energy industry. Among various innovative technologies, the electric vehicle (EV) has received broad attention because of its low well-to-wheel emissions, high energy conversion efficiency, and reduced noise pollution. The global EV fleet has been expanding rapidly, driven by technological progress and policy support. The sales trend in China shown in the figure below highlights how EVs have become one of the fastest-growing transportation technologies.
Nevertheless, the random and uncontrolled charging of EVs can create new threats to the power grid. EV owners usually plug in their vehicles when they arrive home or in the late afternoon; this coincides with residential peak demand, leading to a large peak-load increase and a larger peak-valley difference. The concept of Vehicle-to-Grid (V2G) has therefore emerged as an effective solution that allows electric vehicles to operate as distributed energy resources. By coordinating the timing and power of charging and discharging, EVs can assist the grid in peak shaving and valley filling, renewable energy accommodation, frequency regulation, and voltage support. At the same time, EV owners can reduce their charging costs and even obtain economic benefits by selling electricity during high-price periods.
This thesis focuses on V2G-based EV dispatch and optimization. I first build mathematical models for distributed generators and EV charging/discharging loads. Then, using Monte Carlo simulation, I predict the charging demand in both residential and commercial areas. I then propose a user-oriented grouping control scheme that divides EVs into subgroups according to their parking duration, required charging time, remaining mileage, and willingness to accept battery degradation. Finally, I integrate renewable energy uncertainty into the dispatch problem and use an improved grey wolf optimizer to minimize system operating costs while maximizing renewable energy utilization and EV owner revenue.
Chapter 2 Modeling of Microgrid Components and EV Loads
PV and Wind Turbine Models
The microgrid in this work contains photovoltaic (PV), wind turbines, diesel generators, gas turbines, energy storage systems, local loads, and electric vehicles. The PV output is calculated with its equivalent-circuit behavior. The current–voltage characteristic of a single diode PV model can be expressed as:
$$I = I_{ph} – I_{o} \exp\left(\frac{V+\Delta V}{N_s V_t}-1\right) \left[1+\frac{V+\Delta V}{N_2 V_2}\right]$$
where \(I_{ph}\) is the photo-generated current, \(I_o\) is the reverse saturation current, \(V\) is the terminal voltage, and \(N_2\) and \(V_2\) are auxiliary coefficients. To avoid the strong nonlinearity of the full PV model in the scheduling process, a simplified power curve derived from real irradiance and temperature data is used in the day-ahead dispatch.
For the wind power output, a piecewise model is often employed:
$$P_{WT}=\begin{cases}
0 & 0\leq v<v_{in} &="" 0="" <p="" \\[66pt]="" \\[6pt]="" \dfrac{v-v_{in}}{v_r-v_{in}}="" \end{cases}$$="" p_r="" v<v_r="" vwhere \(P_r\) is the rated power, \(v\) is the actual wind speed, and \(v_{in}\), \(v_r\), and \(v_{out}\) are the cut-in, rated, and cut-out wind speeds respectively. This model is used for creating stochastic wind scenarios and for the optimal dispatch model.
Diesel Generator, Gas Turbine, and Storage Models
The diesel generator cost is related to its fuel consumption. The fuel consumption is approximately linear with output power:
$$D_{DG}=D_{0}+\alpha_{DG}P_{DG}$$
where \(D_0\) is the no-load fuel consumption and \(\alpha_{DG}\) is the incremental fuel coefficient. The corresponding operating cost is \(C_{DG}=D_{DG}\rho_{fuel}/P_{DG}\).
The gas turbine fuel cost is calculated through the turbine model. Its total fuel expense over the dispatch horizon is:
$$C_{GT}=c\sum_{s=1}^{T}F_{GT}(s)$$
with the fuel consumption and heat production modeled as:
$$F_{GT}(s)=\frac{Q_{GT}(s)+P_{GT}(s)}{3600\alpha_{GT}Q_r},\qquad Q_{GT}(s)=\beta_{GT}P_{GT}(s)$$
where \(Q_r\) is the fuel heating value, \(\alpha_{GT}\) is the heat conversion factor, and \(\beta_{GT}\) is the heat-to-power ratio.
The energy storage system model is characterized by the state-of-charge (SOC) update:
$$E_{ESS}(t+1)=E_{ESS}(t)(1-\delta_{ESS})+P_c(t)-P_d(t)$$
$$SOC_{ESS}(t+1)=SOC_{ESS}(t)(1-\delta_{ESS})+ \frac{\eta_c P_c(t)}{E_{ESS}}-\frac{P_d(t)}{\eta_d E_{ESS}}$$
where \(\delta_{ESS}\) is the self-discharge rate, \(\eta_c\) and \(\eta_d\) are charging and discharging efficiencies, and \(E_{ESS}\) is the storage capacity.
EV Charging and Discharging Model
To accurately model the charging load of electric vehicles, I use the statistics of daily driving mileage, arrival time at the charging station, and charging duration. The daily mileage of private electric vehicles follows a logarithmic normal distribution:
$$f_s(s)=\frac{1}{s\sigma_s\sqrt{2\pi}}\exp\left[-\frac{(\ln s-\mu_s)^2}{2\sigma_s^2}\right]$$
with mean \(\mu_s=3.2\) and standard deviation \(\sigma_s=0.88\). The arrival time at a residential charging station follows a normal distribution:
$$f_t(t)=\begin{cases}
\frac{1}{\sigma_t\sqrt{2\pi}}\exp\left(-\frac{(t-\mu_t)^2}{2\sigma_t^2}\right) & 12<t<24\\[6pt] &="" 0<t\leq="" 12=""
For commercial-area charging, a double-peak Gaussian distribution was proposed to reflect both morning and evening commuting behavior:
$$f_n(t)= \sum_{k=1}^{2}\frac{1}{\sigma_{n,k}\sqrt{2\pi}}\exp\left[-\frac{(t-\mu_{n,k})^2}{2\sigma_{n,k}^2}\right] \qquad \text{with domain periodization}$$
The charging time is determined by the daily mileage and the charger power:
$$T_c = \frac{s\,W}{100P_c\eta_c}$$
where \(W\) is the energy consumption per 100 km, \(P_c\) is the rated charging power, and \(\eta_c\) is the charging efficiency. The EV parameters used in this work are summarized in Table 1.
| Parameter | Symbol | Value |
|---|---|---|
| Daily mileage mean | \(\mu_s\) | 3.2 |
| Daily mileage std | \(\sigma_s\) | 0.88 |
| Residential arrival mean | \(\mu_t\) | 17.6 h |
| Residential arrival std | \(\sigma_t\) | 3.4 |
| Commercial peak 1 | \(\mu_1\) | 9 h |
| Commercial peak 2 | \(\mu_2\) | 17.5 h |
| EV battery capacity | \(E_{EV}\) | 50 kWh |
| Energy consumption | \(W\) | 15 kWh/100km |
| Charging power | \(P_c\) | 7 kW |
| Charging efficiency | \(\eta_c\) | 0.9 |
I used Monte Carlo simulation to generate the stochastic characteristics of daily mileage, arrival time, and departure time. The expected charging load is estimated by
$$\hat{x}=\frac{1}{N}\sum_{n=1}^{N}x_n$$
and the variance is
$$\delta^2=\int_{-\infty}^{+\infty}\left(x-E(x)\right)^2 f(x)\,dx$$
The number of EVs in each charging scenario was set to 2000. The simulation showed that in the commercial area two charging peaks occur, corresponding to morning and evening vehicular arrivals; in the residential area a single peak appears in the late evening, which is consistent with real travel patterns.
Chapter 3 User-Demand-Based EV Grouping and Control
To solve the problem that large-scale uncontrolled EV charging creates a peak-load superposition effect, I propose a grouping control strategy under the V2G framework. Unlike a unit-level dispatch where every EV is separately controlled, the grouping method uses time and user preference indicators to classify EVs into several clusters; each cluster shares the same charging/discharging pattern. This greatly reduces the computational burden while retaining user satisfaction.
Grouping Indicators
Four time-related indicators are considered: arrival time \(T_s\), required charging time \(T_n\), parking time \(T_p\), and replaceable time \(T_r\). The last one is defined as
$$T_r=T_p-T_n$$
If \(T_r=0\), the EV can only be charged immediately; if \(T_r>0\), the EV has flexible charging/discharging capacity and can participate in V2G dispatch. User preferences were incorporated by using the desired SOC, willingness to accept battery degradation, and price responsiveness. The battery degradation cost is modeled as
$$C_{de}=\sum_{t=1}^{T}\frac{C_{replace}}{E_{EV}L_{cycle}}P_{de}(t)\Delta t$$
where \(C_{replace}\) is the replacement cost, \(L_{cycle}\) is the equivalent full-cycle life, and \(P_{de}(t)\) is the output power during V2G discharge.
The willingness of an EV owner to participate in V2G services is represented by a logistic participation factor:
$$\varepsilon=\frac{1}{1+e^{-l(SOC-SOC_{mid})}},\qquad SOC\ge0.3$$
where \(SOC_{mid}\) is the SOC level at which willingness increases fastest, and \(l\) is the sensitivity coefficient.
Subgroup Classification and Charging/Discharging Rules
The entire day is divided into six periods according to the arrival probability histogram of EVs. In each period, EVs are classified into one of four subgroups: (1) no-control direct charging, (2) orderly charging only, (3) orderly charging/discharging under V2G, and (4) direct discharging if the user wants to sell energy during peak hours. Direct charging is selected when the user has urgent travel needs; orderly charging uses a delayed charging mode to shift the load to valley periods; V2G mode uses a charge–discharge–charge or two-stage scheme to shave peaks; direct discharging is used at high price periods to maximize user profit. The main charging/discharging schemes for EVs are continuous and segmented. The segmented modes are preferred because they permit power valley filling. Table 2 lists the time-of-use electricity prices used in the simulation.
| Period | Charging price | Discharging price |
|---|---|---|
| Valley | 0.20 | 0.30 |
| Flat | 0.55 | 0.60 |
| Peak | 0.80 | 1.00 |
The actual charging/discharging price is coupled with the load deviation to provide an incentive:
$$p_c=k_0+k_1|S-S_a|$$
$$p_d=k_2+k_3|S-S_a|$$
where \(S\) is the actual load at time \(t\) and \(S_a\) is the expected load. This mechanism encourages EV owners to use electricity during light-load periods and to discharge during heavy-load periods.
Two-Layer Objective Functions and Solution
The upper-layer objective in the IEEE-33 distribution network is to minimize the load variance:
$$F_1 = \frac{1}{n}\sum_{t=1}^{n}\left(P_{Load,t}+Q_{EV,t}-\frac{1}{n}\sum_{t=1}^{n}(P_{Load,t}+Q_{EV,t})\right)^2$$
The lower-layer objective aims at minimizing the total charging cost and maximizing discharging revenue:
$$F_2=\sum_{t=1}^{n}\left(P_{EV,t}p_c-\alpha_i T_i P_e p_d\right)$$
After normalizing both objectives using Min–Max normalization, the combined fitness function is
$$F=\lambda_1 F_1^{norm}+\lambda_2 F_2^{norm}$$
with \(\lambda_1=0.5\) and \(\lambda_2=0.5\). Constraints include voltage limits
$$U_{i,\min}\le U_i \le U_{i,\max}$$
SOC limits
$$SOC_{\min}\le SOC_t\le SOC_{\max}$$
and charging/discharging power limits
$$-\eta_d P_{e,\max}\le P_{EV,t}\le \eta_c P_{e,\max}$$
as well as the time window constraints that the total charging duration and discharging duration do not exceed the parking duration.
Simulation Results in Different Scenarios
The IEEE-33 bus system was used, with buses 4, 13 and 30 designated as EV charging stations. The EVs were divided into residential and commercial stations with a capacity ratio of 1:2:2. Table 3 compares the load characteristics for four cases: basic load, uncontrolled EV charging, orderly charging, and orderly charging/discharging.
| Mode | Load variance (MW2) | Peak-to-valley (MW) | Charging cost (RMB) | Discharging revenue (RMB) |
|---|---|---|---|---|
| Base load | 13.57 | 9.58 | 0 | 0 |
| Uncontrolled charging | 52.32 | 19.24 | 23.21 | 0 |
| Orderly charging | 25.36 | 12.58 | 16.55 | 0 |
| Orderly charging/discharging | 10.23 | 8.95 | 16.55 | 12.90 |
Uncontrolled charging increases the load variance and peak-to-valley difference considerably. Orderly charging can reduce the peak-valley difference; however, the charging of EVs at the valley periods still cannot respond to the grid peak hours. By contrast, V2G-based orderly charging/discharging can flatten the load curve below the original base-load curve and lower the voltage deviations at all charging stations. In addition, I examined the effect of user participation degree. Table 4 shows that when the participation rate increases from 30% to 90%, the load variance declines sharply and the peak-to-valley ratio is reduced significantly.
| Participation level | Load variance (MW2) | Peak-to-valley (MW) | Peak-to-valley ratio (%) | Load factor (%) |
|---|---|---|---|---|
| 0% | 52.36 | 19.24 | 80.42 | 58.33 |
| 30% | 33.62 | 15.36 | 72.21 | 66.67 |
| 60% | 23.64 | 13.93 | 57.89 | 73.33 |
| 90% | 10.26 | 8.95 | 50.35 | 85.71 |
Through the grouping control strategy, the grid-side objective and the EV-user objective can both be improved. The strategy is computationally efficient because only a reduced number of sub-groups is optimized, not every individual vehicle.
Chapter 4 EV Dispatch Optimization Considering Renewable Uncertainty
In order to further exploit the flexibility of electric vehicles in a microgrid with high renewable penetration, I extended the model to account for the stochastic nature of wind and solar generation. Random scenarios were generated by Latin Hypercube Sampling (LHS), followed by K-means clustering for scenario reduction.
Uncertainty Modeling of Wind and PV
Let \(P_{WT}^{fore}\), \(P_{PV}^{fore}\), and \(P_{Load}^{fore}\) be the forecasted wind power, PV output, and load level. The actual values are expressed as the sum of the forecast and prediction error:
$$P_{WT}=P_{WT}^{fore}+\Delta P_{WT}, \qquad P_{PV}=P_{PV}^{fore}+\Delta P_{PV}, \qquad P_{Load}=P_{Load}^{fore}+\Delta P_{Load}$$
Latin Hypercube Sampling divides the cumulative probability curve of each random variable into \(N\) equal intervals and draws a sample from each interval. The distance between every pair of scenarios is calculated to remove redundant scenarios. The final number of representative scenarios was determined by the K-means clustering algorithm. For a 24-hour dispatch problem, 1000 wind scenarios generated by LHS were clustered into three typical wind scenarios; similarly, PV scenarios were clustered into four seasonal scenarios (spring, summer, autumn, winter). These typical scenarios preserve the temporal correlation and the statistical characteristics of the original dataset while significantly reducing the computational burden.
Improved Grey Wolf Optimizer
The grey wolf optimizer (GWO) is a meta-heuristic inspired by the social hierarchy and hunting behavior of grey wolves. In the original GWO, the position of each wolf is updated according to the three best-performing wolves \(\alpha\), \(\beta\), and \(\delta\). The basic equations are:
$$\vec{X}(t+1)=\vec{X}_p(t)-\vec{A}\cdot\vec{D},\qquad \vec{D}=|\vec{C}\cdot\vec{X}_p(t)-\vec{X}(t)|$$
$$\vec{A}=2a\cdot\vec{r}_1-a,\qquad \vec{C}=2\vec{r}_2$$
where \(a\) is a control parameter that decreases linearly from 2 to 0, and \(r_1\), \(r_2\) are random vectors in \([0,1]\).
In the original algorithm, \(\alpha\), \(\beta\), and \(\delta\) have equal weights, which can cause premature convergence and insufficient global exploration. I therefore made two improvements. First, the control parameter \(a\) was changed from linear to nonlinear:
$$a(t)=\frac{2}{1+e^{\lambda\left(\frac{2t}{T_{\max}}-1\right)}}$$
where \(\lambda\) is a nonlinear adjustment coefficient. This allows the swarm to explore more thoroughly in the early stage and converge accurately in the later stage. Second, a cooperative information mechanism was introduced: each wolf updates its position using not only the head wolves but also the information transmitted from the previous wolves that have already updated their positions. The update rule becomes:
$$\vec{X}_1(t+1)=\vec{X}_\alpha(t)-\vec{A}_1\vec{D}_\alpha$$
$$\vec{X}_2(t+1)=\vec{X}_\beta(t)-\vec{A}_2\vec{D}_\beta$$
$$\vec{X}_3(t+1)=\vec{X}_\delta(t)-\vec{A}_3\vec{D}_\delta$$
$$\vec{X}_i(t+1)=\frac{1}{i}\sum_{k=1}^{i-1}\vec{X}_k(t+1)+\eta(t)\qquad i\ge 4$$
where \(\eta(t)\) is a random perturbation used to maintain diversity and avoid trapping in local optima. I tested the improved algorithm (GWO-CIM) against PSO, GA, and the standard GWO on nine benchmark functions. The improved algorithm showed the fastest convergence and the lowest fitness values for most unimodal and multimodal functions, demonstrating its stronger search ability.
Microgrid Operation Objectives
In this chapter the microgrid contains PV, wind turbines, diesel generators, gas turbines, energy storage systems, local load, and V2G-capable electric vehicles. The upper-level objective is the minimization of the total operating cost of the microgrid:
$$\min F_1=C_{Grid}+C_{RE}+C_{DG}+C_{GT}+C_{ESS}+C_{Loss}$$
where
$$C_{Grid}=\sum_{t=1}^{T}\left[p_{buy}(t)-p_{sell}(t)\right]\cdot \Delta P_{Grid}(t)$$
$$C_{RE}=\sum_{t=1}^{T}\left[\omega_1P_{WT}(t)+\omega_2P_{PV}(t)\right]$$
$$C_{Loss}=p_{Loss}\sum_{t=1}^{T}\Delta P_{loss}(t)$$
The second upper-level objective is to maximize the renewable energy utilization:
$$\max F_2=\frac{\sum_{t=1}^{T}\left(P_{PV}(t)+P_{WT}(t)\right)}{\sum_{t=1}^{T}\left(P_{PV}^{fore}(t)+P_{WT}^{fore}(t)\right)}$$
The lower-level objectives seek to maximize the profit of EV owners and the participation enthusiasm:
$$\max F_3=\sum_{t=1}^{T}N\left(p_{sell}(t)\eta_d P_d(t)-p_{buy}(t)\eta_c P_c(t)-C_{de}+J\right)$$
where \(J\) is the discharge participation incentive. To model the participation behavior more realistically, I introduced an incentive–participation saturation curve: when the discharge compensation is low, an increase in compensation strongly stimulates the participation of electric vehicles; when the compensation exceeds a saturation threshold, further increases only weakly affect participation. This curve can be qualitatively represented by a logistic function. The fourth objective is composed of three indices: depth of response \(D_n\), frequency of response \(R_n\), and quality of response \(Q_n\):
$$F_4=\sum_{n=1}^{N}\left(aD_n+bR_n+cQ_n\right)$$
with
$$D_n=\frac{\sum_t\left(P_c(t)-P_d(t)\right)}{\sum_t P_{\max,n}(t)}, \qquad R_n=\frac{N_{response}}{N_{total}}\times100\%, \qquad Q_n=1-\frac{\sum_t|P_{act}(t)-P_{schedule}(t)|}{\sum_tP_{schedule}(t)}$$
where \(a, b, c\) are weighting coefficients satisfying \(a+b+c=1\).
The whole multi-objective problem was converted into a single fitness function by Min–Max normalization and weighted aggregation:
$$\min F = \lambda_1 F_1^{norm}+\lambda_2(1-F_2^{norm})+\lambda_3(1-F_3^{norm})+\lambda_4(1-F_4^{norm})$$
with \(\lambda_1=0.40\), \(\lambda_2=0.25\), \(\lambda_3=0.20\), and \(\lambda_4=0.15\).
Power Balance and Operating Constraints
The active power balance of the microgrid is:
$$P_{PV}+P_{WT}+P_{DG}+P_{GT}+P_{ESS}+\Delta P_{EV}=P_{Load}+\Delta P_{loss}$$
where \(\Delta P_{EV}\) is the net charging (+) or discharging (−) power of all V2G-capable electric vehicles. The grid exchange power must satisfy
$$0\le P_{Grid}(t)\le P_{Grid}^{\max}(t)$$
All controllable generator outputs are kept within their lower and upper bounds. The ESS operation is constrained by
$$0\le P_{ESS}^{ch}(t)\le u_{ESS}(t)P_{ESS}^{ch,\max},\qquad 0\le P_{ESS}^{dis}(t)\le (1-u_{ESS}(t))P_{ESS}^{dis,\max}$$
The total charging/discharging powers of electric vehicles are bounded by the vehicle availability and rated charger limits, and the SOC of each EV satisfies
$$SOC_{\min}\le SOC(t)\le SOC_{\max}$$
Table 5 lists the main parameters of distributed resources in the microgrid model.
| Unit | Min power (kW) | Max power (kW) | Fuel cost (RMB/kWh) | O&M cost (RMB/kWh) |
|---|---|---|---|---|
| Wind turbine (each) | 0 | 1000 | – | 0.10 |
| PV (each) | 0 | 750 | – | 0.05 |
| Diesel generator | 0 | 500 | 2.0 | 0.10 |
| Gas turbine | 100 | 500 | 0.35 | 0.18 |
| Energy storage | -500 | 500 | – | 0.02 |
Simulation Results and Discussion
I simulated three scenarios: (1) uncontrolled charging, (2) orderly charging, and (3) orderly charging/discharging under V2G. In each scenario, the improved grey wolf optimizer was run to solve the microgrid dispatch problem. The optimized costs of the microgrid in the three scenarios are compared in Table 6.
| Cost component | Uncontrolled charging | Orderly charging | Orderly charging/discharging |
|---|---|---|---|
| Gas turbine O&M (RMB) | 768.25 | 631.72 | 495.18 |
| Diesel generator O&M (RMB) | 875.36 | 729.45 | 583.52 |
| PV O&M (RMB) | 32.40 | 32.40 | 32.40 |
| Wind O&M (RMB) | 58.60 | 58.60 | 58.60 |
| Energy storage O&M (RMB) | 82.35 | 61.48 | 40.62 |
| Grid interaction cost (RMB) | 68.42 | 51.32 | 34.21 |
| Network loss cost (RMB) | 42.16 | 31.62 | 21.08 |
| Total cost | 1927.54 | 1596.59 | 1265.63 |
From these results one can see that orderly charging reduces the total microgrid operating cost by 17.16% compared with uncontrolled charging, while the proposed orderly charging/discharging scheme decreases the total cost by 34.33% compared with the uncontrolled case and by 20.73% compared with orderly charging only. The most obvious savings come from the reduced on/off and load-following operation of diesel generators and gas turbines. When electric vehicles discharge during peak periods, the expensive diesel generator output can be replaced by the stored energy of the EV batteries, thereby lowering fuel consumption and maintenance expenditure.
The renewable energy accommodation is also improved by V2G scheduling. In the orderly charging/discharging scenario, the wind power utilization reached 95.7%, with only a small amount of curtailment in the early morning hours. The PV utilization reached 98.2%, while the only curtailment occurred near noon when PV output reaches its maximum value. The presence of controllable charging loads enables the system to shift EV demand to times when PV and wind outputs are abundant, thereby increasing the matching between renewable generation and load.
Another important aspect is the revenue stream of EV owners. Table 7 shows the component breakdown of the final net benefits in the three considered scenarios.
| Revenue term | Uncontrolled | Orderly charging | Orderly V2G |
|---|---|---|---|
| Base charging expense (RMB) | 50.0 | 38.0 | 38.0 |
| Discharging revenue (RMB) | 0.0 | 0.0 | 33.98 |
| Battery wear cost (RMB) | 1.5 | 1.8 | 2.3 |
| Participation incentive (RMB) | 0.0 | 2.0 | 10.0 |
| Final net benefit (RMB) | -51.5 | -37.8 | 3.68 |
In the uncontrolled scenario, the owner only pays for charging and bears no battery degradation due to discharge. In the orderly V2G mode, despite an increased battery wear cost, the discharging revenue and the participation incentive are sufficiently large to turn the net cost into a positive value of 3.68 RMB. This encourages EV owners to participate actively in grid support.
To further illustrate the conflict between aggregate owner benefit and user participation, I calculated a Pareto-front surface between average income and participation rate. The Pareto front in the figure shows that there is an optimal equilibrium point when the discharge incentive is set to 10 RMB and the participation rate is around 70%, which provides the best trade-off between EV owners’ economic benefit and system-level participation. The user participation convergence curve was also generated with the improved grey wolf optimizer. After about 150 iterations, the expected participation rate converges to approximately 70%, confirming that the algorithm and the incentive mechanism jointly promote an acceptable participation level.
The simulation results validate that the proposed multi-layer optimization framework simultaneously achieves the two-sided goals: the microgrid obtains a lower total operating cost and a higher renewable utilization, while EV owners obtain lower charging costs and even positive income. The improved grey wolf algorithm accelerates convergence and provides a more accurate optimum than the conventional GWO, PSO, and GA benchmarks. The proposed method can therefore be regarded as a feasible reference for the orderly development of future vehicle-to-grid markets.
Conclusions
This paper has presented a comprehensive optimization framework for the charging and discharging of electric vehicles in the V2G mode. The main findings and contributions are as follows:
(1) I built accurate models for photovoltaic power generation, wind power generation, diesel generators, gas turbines, energy storage, and electric vehicles. The Monte Carlo simulation of the EV charging load showed that residential and commercial districts have different temporal peak patterns. This modeling framework provides a solid foundation for the subsequent optimal dispatch of electric vehicles.
(2) I proposed a user-demand-based grouping control strategy. According to parking time, charge time, replaceable time, battery degradation cost, SOC constraint, and tariff preferences, electric vehicles were classified into subgroups. Each subgroup implements a different charging/discharging plan, which can not only reduce the computational complexity of dispatch but also reflect the individual preferences of EV owners. Simulation results on the IEEE-33 bus system showed that orderly charging/discharging can effectively flatten the load curve and reduce the peak-to-valley difference from 19.24 MW to 8.95 MW.
(3) To deal with the stochastic output of renewable energy, I employed Latin Hypercube Sampling combined with K-means clustering to generate representative wind and solar scenarios. This approach preserves the correlation and variability of renewable generation while greatly reducing the scenario set size.
(4) I improved the grey wolf optimizer in two directions: nonlinear control parameter adjustment and cooperative information-based position updating. The improved algorithm achieved better convergence speed and solution precision on benchmark functions and on the microgrid dispatch problem.
(5) I established a two-layer optimization model that considers the cost of the microgrid, renewable energy utilization, income of EV owners, and user participation level. The ordered charging/discharging of electric vehicles was shown to reduce the total operating cost by 20.73% relative to the ordered-charging-only scheme, raise the wind/PV utilization ratio above 95%, and create a net positive benefit for EV owners when discharge incentives are properly selected.
The combination of V2G scheduling, incentive design, and advanced optimization algorithms can make future power systems more flexible and cost-efficient. In future research, the model can be extended to electric buses and taxis, or can be coupled with other flexible resources such as hydrogen storage and multi-agent aggregators.
In summary, this study confirms that smart V2G control of electric vehicles is a key enabler of a low-carbon, high-renewable, and economically sustainable power grid.
</t