1. Introduction and Research Background
In my research, I recognize that China’s energy structure, which has long relied heavily on fossil fuels, is facing the dual challenges of environmental pollution and energy resource depletion. The transportation sector accounts for approximately 50% of China’s total petroleum consumption. To accelerate the transformation toward a green and low-carbon energy structure, China has been actively promoting the adoption of new energy power generation equipment and electric vehicles to replace traditional fossil-fuel-based power generation systems and internal combustion engine vehicles.
However, the integration of a high proportion of renewable energy generation equipment and high-penetration electric vehicles introduces significant uncertainty to the power grid, posing substantial challenges to its reliable and cost-effective operation. Given this context, developing effective scheduling strategies for electric vehicles to balance grid stability and economic efficiency—while simultaneously enhancing renewable energy utilization—has become an imperative pathway for fostering the deep development of both new energy generation equipment and electric vehicles.
From the outset of my study, I acknowledged that accurately predicting the uncontrolled charging load of electric vehicles is a fundamental prerequisite for effective scheduling. I began by conducting a comprehensive literature review on the current state of research in electric vehicle load forecasting and scheduling. Both domestic and international research efforts were examined, with particular focus on existing forecasting methods and their limitations. I observed that conventional forecasting methods frequently rely on single-source data and overlook the subjective psychological factors that influence electric vehicle users’ decision-making processes. To address these shortcomings, I turned my attention to the principles underpinning travel chain models and Cumulative Prospect Theory, which provided me with theoretical grounding for modeling complex user behavior. Furthermore, I studied the dynamic adaptive sparrow-cuckoo algorithm, demand-side response mechanisms, and vehicle-to-grid technology—frameworks that collectively formed the theoretical foundation of my proposed scheduling strategy and model-solving approach.
2. Theoretical Framework
2.1 Travel Chain Model
In simulating the mobility patterns of electric vehicle users, I applied the travel chain model to characterize their spatial and temporal travel behavior. This model captures the sequence of trip destinations together with critical attribute parameters including the departure times, the duration of each trip, the arrival and departure times for each stop, and the distances traveled. Employing the Markov chain theory, I established a time-varying Markov chain model for electric vehicle trip destination transitions. Since electric vehicle users choose future destinations without the outcome being influenced by historical states, user travel behavior aligns with the fundamental principles of Markov processes. In my model, the destination set is defined as {H (residential area), W (work area), O (other areas)}, and a dedicated Markov chain transition probability matrix is constructed for each of the 24 one-hour time intervals, expressed as follows:
$$P_t = \begin{bmatrix} p_{HH}^{t} & p_{HW}^{t} & p_{HO}^{t} \\ p_{WH}^{t} & p_{WW}^{t} & p_{WO}^{t} \\ p_{OH}^{t} & p_{OW}^{t} & p_{OO}^{t} \end{bmatrix}$$
I fitted the remaining travel chain feature quantities using heterogeneous data obtained from a fusion of the NHTS 2017 dataset and the *China New Energy Vehicle Big Data Annual Report (2023)*. This approach differs from conventional methods that rely solely on NHTS data. The strategy I adopted involved preserving the micro-statistical structure of the NHTS dataset while calibrating it to match localized macro distribution data by applying segment-based calibration parameters. Specifically, leveraging the interval probability distribution statistics from the report, I constructed piecewise probability density functions in which each interval’s cumulative distribution aligns with the statistical results reported from Chinese data sources.
Table 2-1: Fitted parameters for different trip distance distributions
| Trip distance interval (km) | (0,10] | (10,20] | (20,30] | (30,40] | (40,50] |
|—|—|—|—|—|—|
| First-type trip σ | 2.259 | 7.235 | 13.475 | 18.723 | 24.486 |
| Second-type trip σ | 2.483 | 6.473 | 11.730 | 16.720 | 21.713 |
| Third-type trip σ | 2.734 | 6.734 | 11.424 | 16.424 | 22.943 |
2.2 Cumulative Prospect Theory
I recognized that the charging decision of electric vehicle users is a risk-based judgment process influenced by numerous psychological factors, such as range anxiety. To quantify bounded rationality, I adopted Cumulative Prospect Theory (CPT), which systematically explains deviations from rational behavior in risky decision-making. The core components of CPT include the value function, which accounts for loss aversion and diminishing sensitivity. Mathematically, the subjective value function is defined as follows:
$$v(c) = \begin{cases} (c_0 – c)^\alpha, & c_0 \ge c \\ -\lambda (c – c_0)^\beta, & c_0 < c \end{cases}$$
In this formula, \(c_0\) is the reference point, \(\alpha\) is set to 0.88, \(\beta\) to 0.92, and \(\lambda\) to 2.25. The weighting function for cumulative probabilities, which accounts for the overweighting of small probabilities and the underweighting of large probabilities, is expressed as \(w^+(p) = \frac{p^{\gamma}}{(p^{\gamma} + (1-p)^{\gamma})^{1/\gamma}}\) for gains and a similar expression for losses.
The overall subjective value of a decision outcome is then expressed as the sum over weighted decision components as \(V = \sum \pi_i^{+} v(x_i) + \sum \pi_j^{-} v(x_j)\), where the cumulative decision weights \(\pi\) can be represented in the following manner:
$$\pi_i^{+} = w^{+}\left(\sum_{c=i}^{k} p_c\right) – w^{+}\left(\sum_{c=i+1}^{k} p_c\right)$$
$$\pi_j^{-} = w^{-}\left(\sum_{c=m}^{j} p_c\right) – w^{-}\left(\sum_{c=m}^{j-1} p_c\right)$$
I applied CPT to the electric vehicle charging decision model by constructing a cost function comprising the temporary charging cost, the psychological cost (associated with range anxiety), and the cost of alternative transportation. To model uncertainty, I partially captured the uncertainty in energy consumption via the stochastic distribution of the next trip distance. Multiple discrete decision scenarios were then generated based on possible values of that distance, enabling the calculation of cumulative prospect values for charging versus not charging.
Table 2-2: Simplified trip length distribution probabilities
| Trip length interval (km) | (0,10] | (10,20] | (20,30] | (30,40] | (40,50] |
|—|—|—|—|—|—|
| Probability (%) | 40.3576 | 43.0106 | 14.0176 | 2.3529 | 0.2613 |
2.3 Dynamic Adaptive Sparrow-Cuckoo Algorithm
To solve the high-dimensional and nonlinearly constrained scheduling models that arose in my work, I identified substantial limitations in traditional intelligent algorithms, particularly regarding slow convergence speeds and their tendency to become trapped at locally satisfactory solutions. In response, I developed a novel hybrid optimization technique—the dynamic adaptive sparrow-cuckoo algorithm. This algorithm synergizes the global exploration capabilities of the sparrow search algorithm with the Lévy flight mechanisms inherent to the cuckoo search algorithm. Its distinguishing features are fourfold. First, the algorithm leverages high-dimensional chaotic mapping based on a Sine-Tent-Cosine map, thereby enhancing the diversity and traversal properties of the initialized population. This can be mathematically represented as:
$$x_{i+1}^{j} = \begin{cases} \cos\left(\pi\left(a \cdot \sin\left(\pi x_i^{j}\right) + (1-a) x_i^{j}\right) – 0.5\right), & \text{if } x_i^{j} < 0.5 \\ \cos\left(\pi\left(a \cdot \sin\left(\pi x_i^{j}\right) + (1-a)(1-x_i^{j})\right) – 0.5\right), & \text{otherwise} \end{cases}$$
Second, the standard fixed step size in the cuckoo Lévy flight is replaced with a dynamic inertia weight that adapts based on both the iteration number and current fitness values. This dynamic step size allows for extensive global searches during early iterations and more focused local searches during later iterations. Third, the discovery probability is no longer static; instead, it is subject to dynamic adjustments, balancing exploitation and exploration. Finally, the host nest position update strategy integrates the exploratory principles of the sparrow algorithm’s producer formula, considering both global best positions and search randomness based on individual fitness levels.
2.4 Demand-Side Response and V2G Technology
Within the context of interactive power systems, I integrated price-based demand-side response mechanisms to reshape load profiles. This usually involves dynamic tariffs which are set to reflect real-time grid conditions and to incentivize users to shift consumption to periods of lower demand. Meanwhile, V2G technology was a cornerstone for the intraday scheduling. It enables bidirectional energy flow between an electric vehicle’s battery and the grid. This effectively converts a random electric vehicle load into a distributed dynamic energy storage resource. Utilizing V2G requires robust bidirectional charging infrastructure, low-latency communication and control capabilities, and advanced battery management systems to ensure safety and prolong battery lifespan during discharge events.
3. Electric Vehicle Uncontrolled Charging Load Forecasting Model
3.1 Fitting Travel Chain Feature Quantities
A crucial aspect of my forecasting work was the precise determination of probability distributions for various travel chain elements. For the destination transition probabilities, I constructed a time-varying Markov chain where the probabilities change based on the time of day. For instance, in the extract of the morning rush hour between 8:00 and 9:00, the transition probability matrix for an electric vehicle is much higher when moving from the residential area to a non-residential area, reflecting typical commute behavior.
Multiple statistical models were adopted for the fitted characteristic quantities. The duration of a single trip was modeled using a log-normal distribution. For daily total driving distance, I parameterized a gamma distribution, which is represented by the following formulas:
$$f(L_d) = \frac{\beta^{\alpha}}{\Gamma(\alpha)} L_d^{\alpha-1} e^{-\beta L_d}, \quad \Gamma(\alpha) = \int_{0}^{\infty} L_d^{\alpha-1} e^{-L_d} dL_d$$
After fitting a raw statistical distribution using NHTS data, I applied piecewise calibration parameters derived from the report. This step was essential to correct discrepancies between international behavior patterns and the actual reported trends from China’s electric vehicle fleet. The parking durations for different location types were modeled with Weibull distributions.
Table 3-1: Parameters for modeling parking durations
| Destination type | Scale parameter λ | Shape parameter σ |
|—|—|—|
| Residential | 1.067 | 10.037 |
| Workplace | 5.889 | 10.248 |
| Other | 1.572 | 2.026 |
3.2 Charging Decision Model based on Cumulative Prospect Theory
In order to mimic the realistic charging decisions of electric vehicle users, I established a charging decision model based on Cumulative Prospect Theory. The cost function was fundamental to this process. If a user decides to charge, the resulting immediate cost is calculated based on the amount of energy needed to reach the target SOC. If the EV range after arriving at the next destination is less than the critical range threshold, the model also accounts for a psychological cost associated with range anxiety, which rises proportionally to the depletion of remaining range. If the model projects that the electric vehicle will be unable to reach the next destination, the user is assumed to need alternative transportation, thereby adding an additional and significant cost element.
I then defined a reference point from which the gains and losses of different decisions are measured. For most users, this served as their expected cost of charging at home, factoring in future mileage needs. The evaluation of potential outcomes involved generating distinct scenarios tied to next-trip distance variability. After calculating the cumulative prospect values for both the “charge” and “don’t charge” alternatives, the probability of selecting a particular decision was quantified using a Logit-type function:
$$p_{ch} = \frac{1}{1 + e^{V_{unch} – V_{ch}}}$$
Further refining this, I developed a threshold probability that considers the day’s remaining travel fraction. The user will only charge if the decision probability calculated from the prospect values exceeds this threshold; otherwise, they postpone charging, assuming more favorable opportunities might arise later in the trip chain.
3.3 Forecasting Process and Result Analysis
I adopted the Monte Carlo simulation method to simulate the daily travel and charging processes of an electric vehicle cluster. The number of vehicles was set to \(N\), and the simulation was run iteratively for each vehicle. The process began with the extraction of the day’s first travel departure time and total daily mileage. For each individual trip, I extracted the destination type, trip duration, and mileage from the fitted distributions. Upon arriving at each destination, the vehicle user enters the decision stage: deciding whether to charge in the parking area or wait.
Table 3-2: Simulation parameters for electric vehicle forecasting
| Parameter | Value | Parameter | Value |
|—|—|—|—|
| Battery capacity | 60 kWh | Energy consumption | 20 kWh/100km |
| Fast charging power | 50 kW | Slow charging power | 7 kW |
| Simulated vehicles | 1000 | | |
Based on the simulation of 1,000 electric vehicles, I analyzed the time–space distribution of the uncontrolled charging load. The results showed a clear concentration of charging activity in residential areas during the evening, with a peak around 17:30. In contrast, the peak charging load in the workplace area was concentrated in the morning, corresponding to the arrival of commuters. The “other area” load was more evenly distributed throughout the day, but its uncertainty contributed to unpredictable midday fluctuations. These results, with their distinct peak patterns at different locations, validated that my model accurately captured the behavioral differences across geographically distinct urban zones against a backdrop of mixed reality. They further revealed the underlying issue of “peak superimposed on peak”, where residential charging peaks coincide with evening base-load peaks.
4. Two-Stage Optimal Scheduling Model for Electric Vehicles
4.1 Electric Vehicle Scheduling Model and Demand Response Model
My approach centers on a two-stage optimization framework. The day-ahead stage addresses the large-scale scheduling scenario by leveraging predicted data, while the intraday stage dynamically corrects for real-time deviations and unforeseen conditions. At the heart of the electric vehicle dispatch model lies the aggregation of data pertaining to each vehicle’s arrival time, planned departure time, and target SOC upon disconnection. The scheduling center coordinates vehicle charging and discharging during the window from connection to disconnection. The charging/discharging actions of an electric vehicle are governed by the matrix \(X \in \mathbb{R}^{N \times T}\), where \(x_{i,t}\) denotes the power charged or discharged.
The scheduling process must satisfy critical constraints. It must adhere to maximum charging and discharging power limits, ensure that charging and discharging of the battery do not happen simultaneously, and prevent excessive battery degradation. The battery’s state of charge must remain within a permissible operating window, and the target SOC must be achieved by the time the vehicle disconnects:
$$\sum_{t=t_i^{in}}^{t_i^{off}} (S_{i,t}^{ch} \cdot x_{i,t}^{ch} – S_{i,t}^{dis} \cdot x_{i,t}^{dis}) \cdot \Delta t \ge E_i^{off} – E_i^{in}$$
In parallel, I designed a demand-side response model using a price elasticity matrix to simulate how charging load shifts in response to tariff changes. The demand elasticity matrix E describes the relationship between price changes and demand changes across time slots. The diagonal elements represent self-elasticity, whereas the off-diagonal terms represent cross-elasticity. The load after applying a specific price signal is then calculated as follows:
$$\begin{bmatrix} R_1′ \\ R_2′ \\ \vdots \\ R_n’ \end{bmatrix} = \begin{bmatrix} R_1 \\ R_2 \\ \vdots \\ R_n \end{bmatrix} + \mathbf{E} \cdot \begin{bmatrix} R_1 / p_1 \\ R_2 / p_2 \\ \vdots \\ R_n / p_n \end{bmatrix} \odot \begin{bmatrix} \Delta p_1 \\ \Delta p_2 \\ \vdots \\ \Delta p_n \end{bmatrix}$$
4.2 V2G Responsiveness Evaluation Model
A significant part of my study focuses on the practical feasibility of V2G scheduling. Historically, models often assume that all vehicles will comply with discharge commands, overlooking that economic constraints and battery wear deter users from participating. To address this, I developed a comprehensive model to evaluate the willingness of electric vehicles to participate in V2G. This evaluation integrates the economic benefits derived from charging costs and discharging revenues with battery degradation costs. The economic utility is then defined by the following expression:
$$U_{i}^{V2G} = \frac{C_i^{dis} – C_i^{ch} – C_i^{wastage}}{C_i^{dis}}$$
In my model, if a vehicle performs charging at a stop, it is assumed to have lower SOC and is less likely to participate in V2G discharging. For parked electric vehicles that do not charge, the probability of their participation is further influenced by the SOC after discharging. If discharging would deplete the battery below the range required for the remaining trip, the probability of participation drastically decreases. Thus, the probability of a parked electric vehicle participating in V2G is jointly influenced by economic utility and its SOC-related willingness factor:
$$p_{i,park}^{V2G} = \frac{e^{U_i} \cdot p_{i,park}}{e^{U_i} \cdot p_{i,park} + e^{U_i}}$$
For electric vehicles that have already completed their day’s travel and are parked for the night, the participation probability is directly modeled by the Logit formula. The aggregate V2G responsiveness for the entire cluster is calculated as the arithmetic mean of the participation probabilities of all individual electric vehicles. I incorporated this responsiveness into the intraday scheduling as a constraint, ensuring that the discharging commands issued by the optimization algorithm do not erode user engagement over time.
4.3 Day-Ahead Scheduling Model
During the day-ahead phase, my primary goal was to smooth the load curve and enhance renewable energy consumption without relying on real-time bidirectional power flow. I formulated an optimization problem that maximizes the net operating revenue of the microgrid. The revenue is computed as income from selling electricity to customers minus the costs of operating the micro-turbine, maintenance costs, and the cost of purchasing electricity from the upstream grid.
In this phase, I introduced a dynamic time-of-use charging tariff. Rather than being static, this tariff reflects the forecasted uncontrolled charging load for each time slot. In periods during which the forecasted electric vehicle load is high, the tariff rises to incentivize users to shift their charging tasks to times of lower grid stress. The price lever is defined as:
$$p_{ch,t} = \alpha \cdot p_{ch,dy} \cdot \frac{P_{EV,t}}{\sum_{t=1}^{96} P_{EV,t}} + (1-\alpha) \cdot p_{ch,base}$$
The day-ahead scheduling strategy uses typical scenario data for wind, solar, and traditional loads, which I generated through Latin hypercube sampling followed by k-means clustering to manage computation time. The model then optimizes the operation of the micro-turbine, energy storage system, and power exchange with the main grid, under the dynamic electric vehicle charging price. In this scenario, users are allowed to respond to the price signals by shifting their demand but are not instructed to discharge.
4.4 Intraday Real-Time Scheduling Model
The intraday stage moves beyond merely shifting load and activates the full bidirectional charging/discharging capacity of electric vehicles. I identified three core challenges that this phase must overcome: the highly stochastic arrival and departure of electric vehicles that may render long-term schedules ineffective; the need to dynamically balance grid economic performance with user participation in V2G; and the continuous accumulation of forecast errors in renewable generation and load. To solve these issues, the intraday model is built around a dynamic real-time price for both charging and discharging that is refreshed every 15 minutes based on current conditions.
The real-time optimization strategy operates based on periodically updating electric vehicle information. At each scheduling step \(i\), the controller identifies the maximum schedulable horizon from the current time until the departure of the last connected electric vehicle. Within this horizon, it solves a new optimal power flow problem that considers the current state of all connected electric vehicles. The objective function seeks to maximize the microgrid’s net profit. For the first \(t\) interval, the command is sent to the local controllers. The optimization process itself is governed by a crucial constraint—the V2G responsiveness constraint—which is formulated as follows:
$$|R_i – R^{threshold}| \le \Delta R^{threshold}$$
This limits the discrepancy between the responsiveness level predicted by the economic dispatch solution and the predefined threshold, thereby ensuring the strategy is both economically attractive to the operator and does not lead to excessive battery degradation that would undermine engagement. When the forecast errors are detected, the algorithm samples and updates the actual renewable and load data at every step to avoid errors accumulating for too long.
5. Simulation Results and Discussion
5.1 Simulation Setup
I validated my model using a modified microgrid demonstration project architecture comprising wind turbines, two photovoltaic units, a micro-turbine, a battery storage system, AC load, DC load, and multiple electric vehicle clusters. Relevant parameters of the microgrid are presented in the tables below.
Table 5-1: Microgrid operating parameters
| Parameter | Value | Parameter | Value |
|—|—|—|—|
| Price of natural gas / (yuan/kg) | 5.14 | Storage max Power / kW | 300 |
| Micro-turbine efficiency | 0.50 | Max power exchange / kW | 300 |
| BC converter efficiency | 0.95 | Initial SOC | 0.5 |
| Micro-turbine max power / kW | 400 | Storage capacity / kWh | 1000 |
Table 5-2: Time-of-use electricity tariff
| Period type | Time period | Price (yuan/kWh) |
|—|—|—|
| Valley | 00:00-06:00, 23:00-24:00 | 0.4 |
| Shoulder | 06:00-09:00, 15:00-18:00, 21:00-23:00 | 0.8 |
| Peak | 09:00-15:00, 18:00-21:00 | 1.2 |
5.2 Day-Ahead Scheduling Analysis
I compared two scheduling strategies within my study: Strategy 1 applies a traditional flat time-of-use charging tariff without demand response; Strategy 2 uses the dynamic charging price that guides electric vehicles to shift their load. Under Strategy 1, my results showed significant renewable curtailment from 9:00 to 14:45, with total wind and solar curtailment reaching 10.83% due to valley load periods.
From the subsequent analysis of the load curves, it became apparent that after response implementation (Strategy 2), the evening peak of the electric vehicle charging load successfully shifted to the midday valley. The occurrence of abandoned wind and solar power completely disappeared, leading to a renewable energy utilization rate of 100%. The net revenue of the microgrid increased substantially. I attribute the increase to reduced operational costs from generating units and lower procurement costs from the main grid. In the isolated grid scenario, Strategy 2’s operating revenue was also significantly higher than that of Strategy 1, and the total amount of curtailed load dropped from 4988.89 kW for Strategy 1 to merely 309.42 kW, indicating substantially improved grid reliability.
Table 5-3: Day-ahead scheduling results under grid-connected mode
| Strategy | Operational cost (yuan) | Revenue (yuan) | Net profit (yuan) | Curtailment rate |
|—|—|—|—|—|
| 1 | 3482.16 | 15876.05 | 12393.88 | 10.83% |
| 2 | 2396.27 | 15609.39 | 13213.12 | 0.00% |
5.3 Intraday Scheduling Analysis
In the intraday analysis, I examined the performance of the model under four different strategies in an extreme scenario where electric vehicle users participate in V2G but do not respond to demand-side price incentives. This setting forced the scheduling model to manage peaks through V2G alone. Strategy 1 was presented with no V2G discharging, leading to high peak demand at night. In Strategy 2, the electric vehicles were dispatched with no consideration of V2G responsiveness. The results demonstrated that electric vehicles could provide significant discharge capacity during periods of grid stress. However, the resulting battery degradation costs caused user participation rates to drop to only 82.71%. This decline in willingness indicated that the strategy was excessively aggressive.
When I applied Strategy 3 by adding the V2G responsiveness constraint, the control algorithm automatically reduced the depth of discharge to maintain user satisfaction. It still provided needed grid services, but with a more moderate load valley. The net load curve showed less severe negative values. Under this strategy, I found that the V2G response ratio was maintained at 92.94%, which demonstrated an effective balance between grid economics and long-term user participation.
In Strategy 4, the microgrid transitioned into island mode with an increased number of electric vehicles (500 vehicles). Under this isolated grid scenario, the strategy aimed to avoid load curtailment while ensuring system reliability and respecting the established V2G response ratio. By integrating the V2G management functionality, the model succeeded in eliminating load shedding, keeping frequency and voltage stable, and achieving high economic performance.
Table 5-4: Comparative results for different intraday scheduling strategies
| Strategy | V2G response degree | Characteristic |
|—|—|—|
| 1 | N/A | Large grid import cost |
| 2 | 82.71% | Deep discharge, lower participation |
| 3 | 92.94% | Balanced grid economy and EV user |
| 4 | 90.14% | High EV penetration, no load shedding |
Finally, I compared the computational performance of the proposed dynamic adaptive sparrow-cuckoo algorithm against traditional sparrow and cuckoo algorithms. The proposed algorithm required only 1286 seconds to converge to a satisfactory solution for the third strategy. In contrast, the traditional sparrow algorithm required 1875 seconds and the traditional cuckoo algorithm required 1832 seconds. These simulation outcomes strongly suggest that my proposed optimization algorithm is significantly better suited to high-dimensional, nonlinearly constrained problems and is effective in solving practical implementations of electric vehicle scheduling models.
6. Conclusion and Outlook
In conclusion, my research systematically tackled the significant uncertainties that the integration of a high proportion of renewable energy and high-penetration electric vehicles brings to power grid dispatch. I first developed a sophisticated load forecasting model that integrates time-varying Markov chains and multi-source data. By integrating Cumulative Prospect Theory into the decision model, I was able to account for the inherently bounded rationality of electric vehicle users. This approach yielded time–space distributions of charging loads that accurately reflect real-world travel patterns and decision-making behaviors. The key findings of the forecast analysis highlighted the strain placed on evening residential peaks and the morning workplace peaks that align with existing grid demand patterns.
To unlock the potential of electric vehicles as dynamic grid resources, I further developed and implemented a two-stage optimal scheduling model. During the day-ahead stage, I introduced a dynamic, multi-tier charging tariff that efficiently shifted electric vehicle load to periods of high renewable output. The simulation results confirm that this strategy achieves a reduction in net load variance and fully curtails renewable waste. In the intraday stage, I implemented a real-time, closed-loop scheduling model that uses bidirectional V2G smart charging to maximize the microgrid’s net return. Critically, the model embeds a V2G responsiveness forecasting engine, which prevents the optimization algorithm from creating dispatch commands that over-discharge electric vehicle batteries, thereby promoting enduring user participation.
Several future research directions could bring significant benefits. Future work on vehicle-to-grid integration might add revenue-based sequential game predictions for the cost of dynamic grid services and improved smart charging at scale. Expanding the validation datasets by extracting more realistic trip feature quantities from Chinese user behavior, integrating road network topologies into the travel chain model, and developing a more holistic price-based model for electric vehicle load response to electricity tariffs all remain promising areas for future study. Ultimately, the two-stage scheduling framework I have proposed offers a robust and practical approach to supporting the transition to more sustainable, electric-vehicle-integrated power systems.