With the increasing severity of global climate issues and a deepening understanding of environmental pollution caused by traditional fuel vehicle emissions, electric vehicles (EVs) are gaining worldwide recognition as clean and efficient transportation solutions. In China, the rapid adoption of electric vehicles presents unique challenges and opportunities for integrating these mobile loads into the power grid while ensuring user convenience. However, existing scheduling approaches often fail to comprehensively address the dual needs of users and the grid, leading to inefficient and disordered charging practices. This paper addresses these industry pain points by developing a multi-objective decision-making model for fast-charging stations, incorporating road network and grid information for segment weighting, and predicting the spatiotemporal distribution of electric vehicle charging loads. The aim is to provide theoretical support and practical guidance for resolving electric vehicle charging issues, thereby promoting the widespread adoption and sustainable development of electric vehicles in China and beyond.
Current research on electric vehicle charging scheduling has several limitations. Some studies focus on charging station location planning and the impact of large-scale electric vehicle charging on the grid but neglect the analysis of user charging demands. Others prioritize user credit-based scheduling strategies to allocate grid dispatch plans to electric vehicles, overlooking the dynamic characteristics of user needs. Approaches that consider user perspectives, such as dynamic pricing to reduce disordered charging, often only address economic demands and are thus one-sided. Additionally, while some research predicts the transition probabilities of electric vehicles under various conditions to evaluate charging and load situations, they do not incorporate optimal path planning. Although path planning based on factors like travel distance, load, and speed has been explored, these methods lack discussion on whether electric vehicles require charging during travel. These gaps highlight the need for a holistic approach that integrates user behavior and grid stability.
In this paper, we propose a novel methodology that combines traffic and power grid models to optimize electric vehicle charging scheduling. The core of our approach involves predicting the shortest charging path for electric vehicles based on factors such as traffic network topology, charging pile deployment, and battery state of charge (SOC). For instance, when an electric vehicle’s battery level drops to 20% or below, it triggers a charging signal, and the system calculates the optimal path to a charging station. The distance calculation for charging travel considers the vehicle’s position on the road network. Let $M(x)$ represent the total travel distance to a charging station, where $x$ is the distance from a reference point on the road segment. For a road segment between points B and C, with charging stations at A and D, the distance function can be expressed as:
$$ M(x) = \min(l_{ab} + x, l_{ad} – x + l_{cd}) $$
where $l_{ab}$ is the distance between A and B, $l_{ad}$ is the total segment length, and $l_{cd}$ is the distance between C and D. This formulation accounts for the electric vehicle’s location and the proximity of charging stations, ensuring efficient routing. To enhance real-time path optimization, we introduce a segment weighting model that incorporates average real-time speeds from both traffic and grid data. The weighted speed $w$ for a segment is given by:
$$ w = \frac{w_1 + w_2}{2} $$
where $w_1$ is the actual vehicle speed and $w_2$ is the weighted speed based on historical data. This model uses the Floyd algorithm to dynamically adjust paths by inserting intermediate points, reducing overall travel distance if $l > l_1$, where $l$ is the direct path length and $l_1$ is the sum of segments via the insertion point. This method ensures that electric vehicles in China EV networks can adapt to changing traffic conditions, minimizing delays and energy consumption.
The optimization model includes a target function that predicts the time for an electric vehicle to reach a charging station. If an electric vehicle sends a charging request at time $T_0$, and the travel time to the optimal station is $t_1$, the arrival time $T_{\text{reach}}$ is:
$$ T_{\text{reach}} = T_0 + t_1 $$
Additionally, we employ a first-come-first-served (FCFS) queuing model to estimate waiting times at charging stations. Let $a_\pi(T, T+t)$ denote the expected number of electric vehicles arriving at station $\pi$ in the time interval $(T, T+t)$. The waiting time $W$ for the $j$-th time period at the $i$-th charging station is calculated as:
$$ W = \sum_{j} \tau \cdot X_q(i, j) $$
where $\tau$ is the simulation duration per period and $X_q(i, j)$ is the traffic flow at station $i$ at time $j$. To convert time costs into economic terms, we use a conversion factor $\epsilon = 1.8$, leading to the user time cost $C_{\text{time}}$:
$$ C_{\text{time}} = \epsilon \cdot W $$
This cost must satisfy the constraint $C_{\text{time}} \leq C_{\text{limit}}$, where $C_{\text{limit}}$ is the maximum tolerable time cost for users. For grid stability, the load capacity constraint ensures that the grid’s maximum load exceeds the electric vehicle demand by a safety margin of 20%. The constraint is formulated as:
$$ Q_{\text{total}} = \sum_{i} (Q_{\text{sta}}^i + Q_{\text{sun}}^i + Q_{\text{stor}}^i) \leq 0.8 \cdot Q_{\text{grid max}} $$
where $Q_{\text{sta}}^i$, $Q_{\text{sun}}^i$, and $Q_{\text{stor}}^i$ represent the charging pile capacity, solar energy input, and storage capacity at node $i$, respectively, and $Q_{\text{grid max}}$ is the grid’s maximum capacity. This prevents overloading and supports the integration of electric vehicles into the power system.
To solve this model, we utilize an adaptive particle swarm optimization (PSO) algorithm. The particle position vector $N = (n_1, n_2, \dots, n_T)$ represents potential solutions, with each particle $n_i = (n_{i1}, n_{i2})$ updated iteratively based on individual and global best positions. For the $b$-th particle in the $a$-th iteration, the position $N_a^{b+1}$ and velocity $V_a^{b+1}$ are updated as:
$$ V_a^{b+1} = \omega V_a^b + C_1 s_1 (N_{\text{lbest}}^a – N_a^b) + C_2 s_2 (N_{\text{obest}}^a – N_a^b) $$
$$ N_a^{b+1} = N_a^b + V_a^{b+1} $$
Here, $\omega$ is the inertia weight, $C_1$ and $C_2$ are acceleration coefficients for individual and global best positions, and $s_1$, $s_2$ are random numbers in [0,1]. The inertia weight $\omega$ decreases with iterations to enhance local search capability, calculated as:
$$ \omega = \omega_2 + (\omega_1 – \omega_2) \cdot \frac{T_{\text{max}} – t}{T_{\text{max}}} $$
where $\omega_1$ and $\omega_2$ are initial and terminal inertia weights, $t$ is the current iteration, and $T_{\text{max}}$ is the maximum iterations. Additionally, parameters $s_1$ and $s_2$ are adjusted using a hyperbolic tangent function for fine-tuning:
$$ s_j = \tanh(e \cdot j) \quad \text{for} \quad j = 1, 2 $$
where $e$ is a scaling factor. This adaptive PSO efficiently handles the nonlinearities in electric vehicle charging scheduling, ensuring robust optimization for China EV applications.
We conducted simulation experiments based on the road network and grid characteristics of a representative Chinese city, referred to as XA City, to validate our optimization strategy. The simulation environment included a traffic network model tailored for electric vehicle travel, with 16 charging stations each equipped with 80 to 120 charging spots. The time-of-use electricity pricing for this setup is summarized in Table 1, which influences charging decisions and load distribution.
| Time (h) | 0-7 | 7-10 | 10-15 | 15-18 | 18-21 | 21-23 | 23-24 |
|---|---|---|---|---|---|---|---|
| Price (USD/kWh) | 0.17 | 0.21 | 0.25 | 0.21 | 0.25 | 0.21 | 0.17 |
We compared three charging strategies: Scheme A (shortest distance to charging station), Scheme B (lowest electricity price), and Scheme C (our proposed road-electricity coupling model). The distribution of electric vehicles across charging stations at 10:00 AM was analyzed. In Scheme A, Station 9 experienced nearly 300 electric vehicles, causing severe imbalance, while other stations had fewer vehicles. In Schemes B and C, the number at Station 9 decreased by approximately 53.21% and 53.58%, respectively, demonstrating more uniform distribution. Furthermore, the charging load variance among stations at 12:00 PM was calculated, as shown in Table 2. Scheme C achieved the lowest variance, indicating better load balancing and efficiency.
| Evaluation Metric | Scheme A | Scheme B | Scheme C |
|---|---|---|---|
| Variance (MW²) | 1.771 | 1.042 | 0.781 |
The results confirm that our optimization strategy effectively enhances charging efficiency, reduces waiting times, and lowers costs for electric vehicle users. By predicting the spatiotemporal distribution of charging loads and implementing segment weighting based on real-time data, our approach enables “peak shaving” and avoids over-saturation at charging stations. This is particularly crucial for the growing China EV market, where uncontrolled charging could strain grid resources. The adaptive PSO algorithm proved effective in handling complex constraints, such as user time costs and grid capacity, ensuring practical applicability.
In conclusion, the research on electric vehicle charging scheduling optimization is vital for improving energy utilization, ensuring grid stability, and promoting the adoption of electric vehicles. Our proposed method, which integrates road and electricity coupling models, offers a comprehensive solution that addresses both user needs and grid demands. Through accurate prediction of charging load distributions and dynamic path optimization, we achieve a balanced and efficient charging network. This study underscores the importance of collaborative approaches in advancing electric vehicle infrastructure, contributing to sustainable energy use and the broader goals of environmental protection. Future work could explore real-time data integration and machine learning techniques to further refine scheduling strategies for electric vehicles in diverse urban settings.