With the rapid growth in the number of electric vehicles globally, particularly in China where the adoption of electric vehicles has accelerated, the demand for charging and battery swapping infrastructure has become increasingly critical. As of recent years, China’s electric vehicle fleet has expanded significantly, necessitating efficient planning of charging piles and battery swapping stations to meet user needs while ensuring grid stability. In this study, we address the challenges associated with single-function charging facilities by proposing a hybrid planning method that integrates both charging piles and battery swapping stations. Our approach leverages massive ride-hailing data to extract travel patterns and predict the spatio-temporal distribution of charging and battery swapping demands. We then develop models for charging pile allocation and a two-layer planning framework for battery swapping stations that incorporates ordered charging strategies to reduce costs and enhance grid performance. Through simulations based on real-world traffic networks and distribution systems, we demonstrate the effectiveness of our method in providing convenient, fast, and economical power supply for electric vehicle users, while supporting the stable operation of the distribution network.
The proliferation of electric vehicles, especially in China, has highlighted the limitations of existing charging infrastructure. Traditional charging stations often struggle to accommodate the diverse needs of different electric vehicle types, leading to issues such as long waiting times and inefficient resource utilization. To overcome these challenges, we focus on a data-driven approach that analyzes electric vehicle travel behavior to inform the planning process. By processing ride-hailing order data, we identify key travel features, such as trip origins and destinations, which allow us to model the spatio-temporal distribution of charging and battery swapping demands. This foundation enables us to design a hybrid system where charging piles are deployed in functional areas based on demand density, while battery swapping stations are optimized using a two-layer model that considers operator costs, user convenience, and grid impacts. Our methodology not only improves service quality for electric vehicle users but also promotes the integration of electric vehicles into the power system through ordered charging strategies that mitigate peak load stresses.
In the context of China’s electric vehicle market, the need for robust infrastructure planning is paramount. The Chinese government has implemented policies to support the expansion of charging and battery swapping facilities, reflecting the urgency of this issue. Our work contributes to this effort by developing a comprehensive framework that combines data mining, mathematical modeling, and optimization algorithms. Specifically, we use an improved Grey Wolf Optimization algorithm to solve the complex multi-objective problems inherent in facility planning, ensuring that solutions are both cost-effective and technically feasible. The results from our case study in a urban traffic network show that the hybrid approach reduces overall costs and enhances user satisfaction compared to traditional methods. Furthermore, by incorporating ordered charging, we demonstrate how battery swapping stations can participate in grid services, such as load shifting, which is crucial for the sustainable growth of electric vehicle adoption in China and beyond.
To begin, we outline the data processing and demand prediction methodology. We utilize ride-hailing data from a major Chinese city, which includes information on vehicle trajectories, timestamps, and locations. After data cleaning and transformation, we focus on a specific urban area to analyze travel patterns. The road network is simplified into nodes and edges, and points of interest are classified into functional zones such as residential, commercial, and industrial areas. This allows us to determine the distribution of electric vehicle activities and their associated energy demands. The key step involves modeling the energy consumption of electric vehicles based on factors like environmental temperature and traffic conditions, using a unit energy consumption model. For instance, the energy consumed per kilometer can be expressed as a function of speed and temperature, which we represent mathematically. The state of charge of an electric vehicle battery is updated after each trip, and charging or battery swapping decisions are made when the remaining battery level falls below a threshold, typically set at 20% of capacity. This is formulated as:
$$G_d Q – \omega l_{d+1} < 0.2Q$$
where \( G_d \) is the battery state upon arrival, \( Q \) is the battery capacity, \( \omega \) is the energy consumption per kilometer, and \( l_{d+1} \) is the distance of the next trip. If this condition is met, the user may choose to charge or swap the battery based on availability and convenience. For charging, the decision depends on whether the charging power and time suffice to complete the next trip:
$$G_d Q + P_1 t_i > 0.2Q + \omega l_h$$
where \( P_1 \) is the charging power, \( t_i \) is the dwell time, and \( l_h \) is the remaining travel distance. For battery swapping, users opt for this mode if charging is insufficient or if swapping stations have available batteries. This decision process is simulated using Monte Carlo methods to generate realistic demand profiles.
The demand prediction results reveal distinct patterns across different functional zones. For example, residential areas exhibit peak charging demands in the evening, while commercial areas see higher demands during daytime hours. The spatio-temporal distribution is critical for planning the number and location of charging piles and battery swapping stations. We aggregate these demands into a probability transition matrix that guides the simulation of electric vehicle trips, enabling us to forecast hourly demand variations. This approach ensures that our planning models are grounded in actual travel behavior, which is essential for accurate infrastructure deployment.
Next, we describe the planning models for charging piles and battery swapping stations. For charging piles, we consider the costs from both the operator and user perspectives. The objective function minimizes the total annual cost, which includes investment costs, user travel time costs, waiting time costs, and charging costs. The investment cost for charging piles is calculated as:
$$C_1 = \sum_{i=1}^{H} C_{\text{chg}} N_p \frac{v_0 (1 + v_0)^q}{(1 + v_0)^q – 1}$$
where \( C_{\text{chg}} \) is the cost per charging pile, \( N_p \) is the number of piles, \( v_0 \) is the discount rate, and \( q \) is the operational lifespan. User travel time cost is given by:
$$C_2 = 365 u \sum_{i=1}^{H} \sum (L_{ij} / v)$$
where \( u \) is the time value cost, \( L_{ij} \) is the distance to the charging pile, and \( v \) is the average travel speed. Waiting time cost is modeled using queueing theory:
$$C_3 = 365 C_w \sum_{N_i} W_i \sum_{e} N_e$$
where \( C_w \) is the waiting cost per hour, \( W_i \) is the average waiting time at pile \( i \), and \( N_e \) is the number of electric vehicles served. Finally, the charging cost is:
$$C_4 = 365 \sum_{i=1}^{H} \sum_{t=1}^{24} g (1 – \text{SOC}_i) Q$$
where \( g \) is the electricity price, and \( \text{SOC}_i \) is the state of charge upon arrival. Constraints include limits on the number of charging piles per node and maximum waiting times.
For battery swapping stations, we employ a two-layer planning model. The upper layer focuses on minimizing costs related to station construction, user waiting times, travel times, and swapping costs. The lower layer addresses grid-related costs, such as load fluctuations and network losses. The upper-layer objective function is:
$$\min F_1 = C_b + C_{\text{wait}} + C_{\text{EV}} + C_P$$
where \( C_b \) is the construction and maintenance cost, \( C_{\text{wait}} \) is the waiting time cost, \( C_{\text{EV}} \) is the travel time cost, and \( C_P \) is the swapping cost. The construction cost is detailed as:
$$C_b = \sum_{i=1}^{H} \left( f \frac{v_0 (1 + v_0)^q}{(1 + v_0)^q – 1} + F \right)$$
with \( f = C_{\text{chr}} N_i^d + C_{\text{bat}} M_i^B + C_{\text{eh}} S_i^D \), where \( C_{\text{chr}} \), \( C_{\text{bat}} \), and \( C_{\text{eh}} \) are the costs per charger, battery, and swapping device, respectively, and \( N_i^d \), \( M_i^B \), and \( S_i^D \) are the quantities of these components. Waiting time cost is:
$$C_{\text{wait}} = 365 Z_w \sum_{i=1}^{H} \sum_{t=1}^{24} \bar{W} \cdot N_H$$
travel time cost is similar to that for charging piles, and swapping cost is:
$$C_P = 365 \sum_{i=1}^{H} \sum_{t=1}^{24} g (1 – S^d_{\text{OC},i}) Q$$
Constraints in the upper layer include limits on the number of swapping devices, chargers, and batteries, as well as maximum travel distances and waiting times. The lower-layer objective function minimizes grid costs:
$$\min F_2 = C_{\text{fluc}} + C_{\text{loss}}$$
where \( C_{\text{fluc}} \) is the load fluctuation cost:
$$C_{\text{fluc}} = 365 \sum_{t=1}^{24} \left( P_h(t) – \frac{\sum_{t=1}^{24} P_h(t)}{24} \right)^2$$
and \( C_{\text{loss}} \) is the network loss cost:
$$C_{\text{loss}} = 365 g \sum_{i=1}^{24} \sum I_{ij}^2 R_{ij}$$
Constraints here include voltage and current limits in the distribution network.
To solve these models, we use an improved Grey Wolf Optimization algorithm. Traditional optimization methods like particle swarm optimization may converge slowly or get stuck in local optima, so we enhance the GWO algorithm by incorporating elements from PSO, such as individual memory and differential evolution. The position update equation is modified as:
$$x_1 = x_\alpha – A_1 D_\alpha$$
$$x_2 = x_\beta – A_2 D_\beta$$
$$x_3 = x_\delta – A_3 D_\delta$$
$$x(t+1) = \frac{x_\alpha(t) + x_\beta(t) + x_\delta(t)}{3} + b_1 r_3 [P_{i,\text{best}}^d(t) – x_i^d(t)] + b_2 r_4 [x_j^d(t) – x_i^d(t)]$$
where \( x_\alpha, x_\beta, x_\delta \) are the positions of the alpha, beta, and delta wolves, \( A_1, A_2, A_3 \) are coefficient vectors, \( D_\alpha, D_\beta, D_\delta \) are distance vectors, \( b_1 \) and \( b_2 \) are coefficients, \( r_3 \) and \( r_4 \) are random numbers, and \( P_{i,\text{best}} \) is the best position of individual \( i \). Additionally, we implement an elite retention strategy to preserve the top solutions and maintain population diversity.
In our simulation, we apply this framework to a case study based on a urban traffic network and the IEEE 33-node distribution system. We assume a fleet of 6,000 private electric vehicles and 4,000 taxis, and use real data to parameterize the models. The results for charging pile planning show that the optimal number of piles varies by functional zone, with residential areas requiring more piles due to higher demand. The total annual cost is minimized at a specific number of piles, as summarized in the following table:
| Number of Charging Piles | Investment Cost (10^4 CNY) | Charging Cost (10^4 CNY) | Travel Time Cost (10^4 CNY) | Waiting Time Cost (10^4 CNY) | Total Annual Cost (10^4 CNY) |
|---|---|---|---|---|---|
| 175 | 525.9 | 983.5 | 194.1 | 448.3 | 2,151.8 |
| 208 | 624.7 | 983.5 | 93.8 | 387.5 | 2,089.5 |
| 244 | 732.6 | 983.5 | 38.4 | 304.9 | 2,059.4 |
| 296 | 882.3 | 983.5 | 38.4 | 238.4 | 2,142.6 |
| 335 | 1,056.2 | 983.5 | 38.4 | 182.1 | 2,260.2 |
For battery swapping stations, we compare unordered and ordered charging scenarios. In unordered charging, batteries are charged immediately after swapping, leading to higher peak loads. In ordered charging, time-of-use electricity prices encourage off-peak swapping, reducing costs and grid stress. The optimal number of stations is eight in both cases, but ordered charging yields lower costs, as shown below:
| Number of Stations | Construction Cost (10^4 CNY) | Waiting Cost (10^4 CNY) | Swapping Cost (10^4 CNY) | Grid Loss Cost (10^4 CNY) | Load Fluctuation Cost (10^4 CNY) | Total Annual Cost (10^4 CNY) |
|---|---|---|---|---|---|---|
| 6 (Unordered) | 1,423.9 | 248.3 | 483.5 | 104.1 | 440.5 | 3,139.4 |
| 8 (Unordered) | 1,540.1 | 204.9 | 483.5 | 120.5 | 440.5 | 2,999.8 |
| 6 (Ordered) | 1,195.3 | 248.3 | 251.4 | 98.6 | 220.7 | 2,453.4 |
| 8 (Ordered) | 1,285.7 | 204.9 | 251.4 | 113.1 | 220.7 | 2,286.1 |
The improved GWO algorithm demonstrates faster convergence and better solution quality compared to standard methods like PSO or genetic algorithms. In our tests, it achieved lower objective function values in fewer iterations, validating its efficiency for this application.
In conclusion, our hybrid planning method effectively addresses the spatio-temporal demands of electric vehicles in China. By integrating data mining, optimization models, and ordered charging strategies, we achieve a balance between user convenience and economic efficiency. The results highlight the importance of considering both charging and battery swapping options to cater to diverse electric vehicle users. For future work, we plan to explore the integration of renewable energy sources and vehicle-to-grid technologies to further enhance the sustainability of electric vehicle infrastructure. This approach not only supports the growth of China’s electric vehicle market but also provides a scalable framework for other regions facing similar challenges.