Optimizing Electric Car Charging Station Layout Under Congestion

1. Introduction and Background

Global energy structure transformation has accelerated, while environmental protection demands grow increasingly urgent. Traditional fuel vehicles are now recognized as a critical barrier to sustainable development due to their high energy consumption and severe pollution. According to the International Energy Agency’s World Energy Outlook 2023 report, the global transportation sector contributes approximately 24% of total carbon dioxide emissions, with traditional fuel vehicles accounting for over 70% of this share. As the number of vehicles worldwide continues to grow rapidly, the eventual depletion of petroleum resources is universally acknowledged, compelling humanity to seek new, sustainable alternative energy sources to reduce dependence on fossil fuels.

Electric power development has created new pathways toward clean energy for human society, driving the transformation of the traditional automotive industry. Electric cars have become central to the new round of automotive technology revolution, demonstrating advantages in industrial competitiveness and low-to-zero emissions. Since the 2015 Paris Climate Conference, countries have implemented various policies to reduce carbon emissions and advance green low-carbon economies. Major economies including China, the European Union, and the United States have established specific targets for transitioning away from traditional fuel vehicles. The European Union formally passed legislation in 2023 requiring member states to halt sales of fuel vehicles by 2035. The United States’ Inflation Reduction Act offers tax credits up to $7,500 for electric car purchasers. China’s New Energy Vehicle Industry Development Plan (2021-2035) targets new energy vehicle sales comprising over 20% of annual vehicle sales by 2025.

The electric car market has shown explosive growth under these policies. By the end of 2023, global annual electric car sales exceeded 14 million units, marking a 35% increase from the previous year. China’s market accounted for over 60% of global sales, maintaining significant industry leadership. Projections indicate that the global electric car fleet will reach 700 million to 900 million units by 2040, representing approximately two-thirds of the total projected vehicle fleet. Major markets including China, the EU, and the United States have published plans to phase out internal combustion engine vehicles over coming decades, providing strong institutional support for long-term industry development.

Despite the clear environmental benefits of electric cars, “range anxiety” remains a major bottleneck hindering broader market penetration. Range anxiety describes consumer concern about the disparity between expected and actual driving range. Modern electric cars typically achieve actual ranges between 400 and 600 kilometers under full charge, but actual range varies considerably with driving behavior, road conditions, temperature, and other factors. To alleviate this anxiety, countries are actively expanding charging infrastructure networks and accelerating補能 system construction.

However, charging infrastructure planning and construction still lag behind market development. According to the China Electric Vehicle Charging Infrastructure Promotion Alliance, by the end of 2023, over 1.8 million public charging piles existed nationwide, maintaining a vehicle-to-pile ratio of 2.5:1—still significantly short of the internationally recognized ideal of 1:1. In Tianjin’s Nankai District, the charging station density is only 0.8 stations per square kilometer, far below actual demand.

This thesis addresses a specific gap in existing research. Most prior studies have concentrated on single-objective optimization using simulation data to determine demand points and candidate station locations, which often diverges from reality. Furthermore, these studies frequently ignore the complexity of urban traffic networks, particularly the impact of congestion on travel time. To address these challenges, I propose a dual-objective charging station layout optimization model that incorporates traffic congestion factors. I employ K-Means clustering to determine candidate station locations and utilize Amap API real-time traffic data to simulate route choice. This approach bridges the gap between theoretical models and actual traffic conditions, enhancing practical applicability and user satisfaction.

2. Literature Review and Theoretical Foundations

2.1 Charging Station Siting and Capacity Planning

Domestic and international research has established comprehensive frameworks for charging station siting and capacity planning. Sha Di (2017) proposed a hierarchical location planning approach with separate optimization strategies for downtown slow-charging facilities and peripheral fast-charging stations, integrating traffic flow and user demand distribution variables. Xu Zuqin (2018) analyzed factors influencing charging loads and developed a gray GA-LM-BP neural network prediction method for electric car ownership, subsequently applying Monte Carlo simulation to construct charging load prediction models. Huang Mengchao (2020) investigated user charging demand uncertainty using robust optimization combined with queuing theory. Bayram et al. (2013) developed fast-charging station optimization methods based on power distribution models. Namdeo et al. (2014) proposed a multi-stage optimization model incorporating real data and spatial distribution information of early EV adopters. Jung et al. presented a stochastic dynamic itinerary interception refueling model for electric taxi charging that accounts for queuing delays.

2.2 Charging Demand Prediction and Load Management

Jia Peng et al. (2021) proposed a charging flow prediction scheme integrating gray prediction theory, road network information, and actual charging records to reduce user waiting time and optimize service capacity. Zhang Binqiao et al. (2021) introduced a charging load scheduling strategy based on full-cost electricity pricing that mitigates the impact of concentrated charging on both congestion and grid stability. Zhu Yongsheng et al. (2023) applied M/M/c queuing theory to optimize charging station capacity allocation, thereby reducing economic losses and construction costs. Xiao et al. (2020) developed an optimization model considering bounded queuing lengths and site selection constraints to reduce waiting times.

2.3 Optimization Algorithms and Model Construction

Domestic research demonstrates sophisticated model building. Luo Jiaxin et al. (2022) constructed a bi-level programming model optimizing social economic costs and carbon emissions using improved coyote optimization. Que Huakun et al. (2022) designed a fuzzy bi-objective programming model for siting that maximizes user satisfaction and minimizes total costs, applying multi-population genetic algorithms. Wang Heping and Gao Rui (2023) proposed a multi-objective mathematical optimization model incorporating construction investment reduction, utilization improvement, and demand satisfaction with improved particle swarm optimization. Energy charging infrastructure research spans high-dimensional attributes. Chen et al. (2021) introduced a balanced mayfly algorithm for optimizing electric car charging configuration to minimize installation costs. Cheng et al. (2022) developed an improved whale optimization algorithm to solve non-convex nonlinear positioning and capacity problems. Zhou et al. (2022) presented a cost-model plus genetic algorithm approach including social total cost assessment with economic and environmental components.

2.4 Traffic Network and User Behavior Coupling

Several researchers have studied traffic and charging interactions. Shi Yunyang et al. (2020) investigated dynamic route planning methods based on traffic conditions and battery level, incorporating real-time traffic prediction with remaining charge. Shang Longlong et al. (2023) proposed a charging station layout optimization approach considering congestion using a hybrid integer second-order cone programming model. Huang Bo et al. (2024) addressed accident impacts on electric car driving and charging behavior through Dijkstra-algorithm-based rolling optimization. Li Chunyan et al. (2025) developed a dynamic traffic model incorporating speed-volume relationships for real-time congestion-adaptive routing.

International studies have similarly advanced traffic-coupled optimization. Ashwani et al. (2022) developed three single-objective models minimizing energy consumption, waiting time, and charging cost while considering battery capability, traffic state, and energy pricing. Bian et al. (2022) integrated POI classification with functional area analysis for rapid charging station planning, explicitly quantifying congestion effects on time costs. Wang et al. (2022) established a model maximizing charging station operating revenue while minimizing additional user charging costs under congestion index constraints. Wu et al. (2024) formulated location-routing optimization for uncertain traffic networks using adaptive co-evolutionary clustering.

2.5 Research Gaps and Contributions

From my literature review, I have identified two principal deficiencies in existing research. First, most studies fail to fully integrate congestion impacts into global charging station optimization. Many rely on static demand models assuming Euclidean distance or shortest-path travel, while others use coarse congestion indices. Some examine traffic impacts on routing but do not globally optimize charging layout. Second, numerous studies depend on simulation data and lack real case validation. Random selection or grid discretization methods often ignore spatial heterogeneity in charging demand, causing allocation inefficiency: some stations remain underutilized while others face excessive queuing.

This thesis contributes to the literature in three innovations:
(1) Incorporating congestion into a practical optimization framework by obtaining actual user travel times via the Amap API, thereby improving model realism and accuracy.
(2) Employing K-Means spatial clustering integrating commercial, residential, and office multidimensional characteristics to identify high-demand candidate zones—significantly more precise than grid or random methods.
(3) Constructing a dual-objective optimization model balancing construction investment and user travel costs, providing governments and enterprises with more actionable decision reference.

3. Theoretical Framework

3.1 Traffic Congestion

Traffic congestion is conceptualized as the phenomenon where vehicle travel demand exceeds network capacity during specific periods, causing impeded movement and speed reduction. China’s Ministry of Public Security identifies congestion when vehicles are stationary or slow-moving with queues exceeding 1 kilometer, or when vehicles at signalized intersections cannot pass after three consecutive green lights. Japan’s legal definition classifies ordinary road congestion as queuing exceeding 1 kilometer or lasting over 10 minutes. The United States defines “severe congestion” as traffic falling below 22 km/h.

Congestion manifests through:
(a) Temporal characteristics: peak commuting hours (approximately 7:00-9:00 and 17:00-20:00) create concentrated congestion;
(b) Spatial characteristics: bottleneck areas at transitions between high-capacity and low-capacity roads, intersections with poor signal configuration, and areas with significant left-turn flows;
(c) Recurrent versus non-recurrent categories: the former has predictable periodicity, while the latter derives from accidents, weather, and special events.

Congestion impacts charging station siting in crucial ways. In the temporal dimension, peak congestion reduces travel speed and increases energy consumption per kilometer, creating concentrated charging demand that may exceed station capacity. Spatially, congestion-prone areas—commercial zones, primary corridors, and transport hubs—demonstrate higher demand densities, making them priority zones for charging placement.

For quantitative analysis in my model, I introduce a congestion coefficient linked to road service levels. When roads are clear, the coefficient equals 1. Under light congestion it becomes 2; moderate congestion yields 3; and heavy congestion produces 4. Table 1 presents these relationships.

Table 1. Congestion Coefficient versus Speed Correlation
Service Level V/C Ratio Traffic Condition Driving Speed Congestion Coefficient
Level I 0-0.4 Clear 0.8V₀ < V < V₀ 1
Level II 0.4-0.6 Light congestion 0.6V₀ < V < 0.8V₀ 2
Level III 0.6-0.75 Moderate congestion 0.4V₀ < V < 0.6V₀ 3
Level IV 0.75-0.9 Heavy congestion 0 < V < 0.4V₀ 4

3.2 Electric Cars and Charging Stations

Electric cars are vehicles driven primarily by onboard power sources. Categories include Battery Electric Vehicles (BEVs), Plug-in Hybrid Electric Vehicles (PHEVs), Hybrid Electric Vehicles (HEVs), Extended-Range Electric Vehicles (EREVs), Fuel Cell Electric Vehicles (FCEVs), and micro-electric vehicles. BEVs and PHEVs currently dominate the market, serving short-distance environmental travel and users with mixed requirements, while FCEVs and EREVs provide extended-range solutions.

Charging stations are classified by power and purpose: fast-charging stations provide high-power rapid service for long-distance travel and are situated along key corridors and service areas; slow-charging stations serve residential and working zones for extended charging sessions; and home charging units satisfy daily routine needs.

Charging station planning principles include:
(1) aligning with actual and forecast electric car populations;
(2) equitable distribution across urban zones;
(3) economic feasibility and cost control;
(4) forward-looking design with expandability.

Planning factors include user demand distributions, infrastructure costs (construction, distribution systems, operation), and grid conditions that influence feasibility of fast-charging implementations.

3.3 Facility Location Theory

Traditional models—the P-median problem, P-center problem, and covering models—provide the theoretical basis for charging station optimization.

The P-median problem, first formulated by Hakimi in 1964, selects P facilities minimizing weighted total cost from demand points to facilities:

$$ \min \sum_{i \in I}\sum_{j \in J} a_i d_{ij} X_{ij} $$

subject to:

$$\sum_{j \in J} Y_j = P$$
$$\sum_{j \in J} X_{ij} = 1, \forall i \in I$$
$$X_{ij} \le Y_j, \forall i \in I, j \in J$$

where \(X_{ij}\) equals 1 if demand i is served by facility j, and \(Y_j\) equals 1 if facility j is opened.

The P-center problem minimizes the maximum distance between any demand point and its closest facility:

$$\min D$$
$$\sum_{j \in J} X_{ij} = 1, \forall i \in I$$
$$\sum_{j \in J} d_{ij} X_{ij} \le D, \forall i \in I$$

The covering model determines the minimum number of facilities needed so that all demand points have access to at least one facility within a preset coverage standard.

My charging station layout model fundamentally synthesizes aspects of both P-median and covering approaches, optimizing station location and quantity decisions simultaneously.

4. Model Construction

4.1 Problem Description and Assumptions

Given candidate station locations identified through K-Means clustering, my model answers the following questions: which candidate stations should be constructed so that all charging demands are served, how many fast/slow charging piles should each station have, and how should demands be assigned given congestion-affected travel times?

Model assumptions are stated explicitly:
A1: Construction costs are borne by a single entity (noting actual systems can adjust through government subsidy coefficients \(\gamma\)).
A2: Candidate locations determined via K-Means clustering of commercial centers, residential zones, office areas, and transportation hubs remain fixed.
A3: Fixed construction cost \(C_j\) is known and geographically correlated.
A4: Each charging pile type \(k\) has different costs and service capabilities; total piles per station cannot exceed station capacity limits.
A5: User waiting time cannot exceed maximum permitted waiting period \(G_{jk}\).
B1: Each user selects exactly one charging station for service.
B2: Users select stations that minimize travel time/distances.
B3: Virtual user points are generated at high-density locations; user positions are fixed.
B4: Charging demand \(L_{ik}\) is known and fixed, derived from daily driving habits, battery capacity, and consumption.
C1: Baseline travel time \(T_{ij}\) obtained from Amap API reflects road-clear conditions.
C2: Under congestion conditions, actual travel time is multiplied by the congestion coefficient \(r\) (values from Table 1).

4.2 Sets, Parameters, and Decision Variables

Sets:
\(I\): set of user charging demand points, indexed by \(i\).
\(J\): set of charging station candidate points, indexed by \(j\).
\(K\): set of charging pile types, \(k=1\) for fast charging, \(k=2\) for slow charging.

Parameters:
\(C_j\): fixed construction cost of station \(j\) (10,000 yuan).
\(P_k\): cost to purchase one type-\(k\) charging pile (10,000 yuan).
\(o\): maintenance cost coefficient as proportion of initial investment.
\(L_{ik}\): charging time required at type-\(k\) pile for user \(i\) (hours).
\(T_{ij}\): travel time from user demand point \(i\) to station \(j\) (hours).
\(W_j\): upper limit on piles per station (quantity).
\(Q_j\): lower limit on piles per station (quantity).
\(G_{jk}\): maximum user waiting time (hours).
\(S\): minimum required number of charging stations to construct.
\(r\): traffic congestion coefficient.
\(E\): total available budget for station construction (10,000 yuan).
\(M\): an arbitrarily large number.

Decision variables:
\(x_j = \begin{cases} 1 & \text{if station } j \text{ is constructed} \\ 0 & \text{otherwise} \end{cases}\)
\(b_{ij}\): number of electric cars traveling from demand point \(i\) to station \(j\).
\(n_{jk}\): number of type-\(k\) charging piles installed at station \(j\).
\(y_{ijk} = \begin{cases} 1 & \text{if user } i \text{ charges at station } j \text{ with type } k \\ 0 & \text{otherwise} \end{cases}\)

4.3 Objective Functions

My first objective minimizes total station construction and operational cost:

$$\min Z_1 = \sum_{j \in J} C_j x_j + \sum_{j \in J}\sum_{k \in K} P_k n_{jk} + o\Big(\sum_{j \in J} C_j x_j + \sum_{j \in J}\sum_{k \in K} P_k n_{jk}\Big)$$

which I decompose into three components:

$$\min Z_1 = \sum_{j \in J} C_j x_j + \sum_{j \in J}\sum_{k \in K} P_k n_{jk} + o\Big(\sum_{j \in J} C_j x_j + \sum_{j \in J}\sum_{k \in K} P_k n_{jk}\Big)$$

where the first term accounts for fixed station construction; the second covers pile purchase costs; and the third represents maintenance operational costs, estimated as a fixed proportion of initial investment in line with industry conventions [42, 40].

My second objective minimizes total user travel time subject to congestion:

$$\min Z_2 = \sum_{i \in I}\sum_{j \in J} r \cdot T_{ij} \cdot b_{ij}$$

The inclusion of congestion coefficient \(r\) modifies user travel time under congested conditions.

4.4 Constraints

Constraints for this model include:

$$\sum_{i \in I} b_{ij} = \sum_{i \in I}\sum_{k \in K} y_{ijk}, \quad \forall j \in J$$
$$b_{ij} \in \{0,1,2,\dots\}, \quad \forall i \in I, j \in J$$

Constraint (1) states that the number of electric cars from point \(i\) to station \(j\) equals those allocated through binary variables \(y_{ijk}\). Each user \(i\) must be served by exactly one station:

$$\sum_{j \in J}\sum_{k \in K} y_{ijk} = 1, \quad \forall i \in I$$

Only stations that are constructed can serve users:

$$\sum_{k \in K} y_{ijk} \le x_j, \quad \forall i \in I, j \in J$$

Charging pile quantities at each station minimum and maximum specifications:

$$\sum_{k \in K} n_{jk} \ge Q_j x_j, \quad \forall j \in J$$
$$\sum_{k \in K} n_{jk} \le W_j x_j, \quad \forall j \in J$$

At least \(S\) stations are constructed in the planning region:

$$\sum_{j \in J} x_j \ge S$$

Piles may only be installed in constructed stations:

$$\sum_{k \in K} n_{jk} \le M x_j, \quad \forall j \in J$$

Charging demand duration limit:

$$\sum_{i \in I}\sum_{k \in K} y_{ijk} \cdot L_{ik} \le \sum_{k \in K} G_{jk} \cdot n_{jk}, \quad \forall j \in J$$

Budget constraints:

$$\sum_{j \in J} C_j x_j + \sum_{j \in J}\sum_{k \in K} P_k n_{jk} + o\Big(\sum_{j \in J} C_j x_j + \sum_{j \in J}\sum_{k \in K} P_k n_{jk}\Big) \le E$$

Binary and integer variable limits:

$$x_j \in \{0,1\}, \quad y_{ijk} \in \{0,1\}, \quad n_{jk} \in \mathbb{Z}_{\ge 0}$$

The model described above is a mixed-integer linear programming problem (MILP). For solution purposes, I apply the weighted LP-Metric method:

$$\min Z = \alpha \frac{Z_1 – Z^*_1}{Z^*_1} + \beta \frac{Z_2 – Z^*_2}{Z^*_2}$$

where \(Z^*_1\) and \(Z^*_2\) represent individual objective function optima, respectively, and parameters \(\alpha\), \(\beta\) denote weight coefficients expressing the relative significance of construction cost versus user time minimization, where alpha is set to 1 and beta to 1 in baseline experiments. By integrating these objectives, the multi-objective optimization problem is transformed into a single-objective problem solvable with the LINGO optimization software.

5. Case Study Application

5.1 Study Area and Data Source

I selected selected areas of Nankai District, Tianjin, as the case site for this thesis. Nankai District represents Tianjin’s central urban zones with high economic development and population density. By 2023, electric car ownership in Nankai reached approximately 20,000 vehicles, accounting for roughly 8% of total motor vehicles. Public charging infrastructure currently totals about 300 piles, but fast-charging piles are relatively few. Charging stations cluster in commercialized areas while residential and other districts face shortages. Real-world issues of congestion-related allocation and accessibility are, therefore, particularly tangible for validating this model.

5.2 Data Collection and Preparation

I collected point-of-interest (POI) data via the Amap Open Platform API. POI categories include: transport hubs (railway stations, airports, bus stations), commercial centers (shopping malls, commercial streets, office buildings), residential zones (residential quarters, apartment groups), public service facilities (hospitals, schools, libraries, museums), and tourist attractions (scenic areas, parks). These POIs serve as proxy user demand locations.

Using K-Means clustering, I grouped POI demand points into defined numbers of clusters for each scenario. Table 2 reports the three scenario configurations.

Table 2. Scenario Configuration
Scale POI Points Cluster Centers (Potential Stations) User Demand Points
Small 74 5 50
Medium 808 20 300
Large 2144 50 800

Figures 2-4 show the K-Means cluster center locations for the small, medium, and large scenarios. Cluster centers are geographically distributed across high-demand zones identified within each POI dataset.

5.3 Parameter Settings

I assume electric cars have charging demand when residual battery ranges from 20% to 80%. Fast charging fills 80% of battery capacity in 1 hour; slow charging requires 6 hours. Station construction costs range between 70 and 90 ten-thousand-yuan, based on site-specific conditions. Table 3 summarizes parameter values:

Table 3. Parameter Settings
Parameter Value Unit Parameter Value Unit
\(C_j\) (70,90) 10,000 yuan \(W_j\) 15 piles
\(P_1\) (fast) 2 10,000 yuan \(Q_j\) 5 piles
\(P_2\) (slow) 1 10,000 yuan \(G_{jk}\) 2 hours
\(o\) 20 % \(S\) varies stations

6. Results and Analysis

6.1 Multi-Scenario Results

I now present results across the three scenarios:

Table 4. Optimization Results for Small, Medium, and Large Scenarios
Scale \(Z_1\) (10,000 yuan) \(Z_2\) (hours) Number of Stations Fast Piles Slow Piles
Small 310.8 1.9578 3 15 1
Medium 1747.2 17.1006 16 90 4
Large 4093.2 26.9058 36 255 28

The small-scale layout (78% of candidate points) exhibited 3 charging stations at candidate expansion sites with 15 fast piles and one slow pile, totaling construction and operations costs of CNY 3.108 million. Fast piles dominate at 93.75%; this configuration prioritizes high-demand zones with minimal capital expenditure while achieving reasonable user outcomes. The medium-scale layout expands to 16 stations among 20 candidate points, with 90 fast and 4 slow piles, representing a total cost of 17.472 million yuan. The large-scale layout places stations at 36 of 50 candidate points with 255 fast piles and 28 slow piles, incurring 40.932 million yuan construction costs, while improving spatial coverage across the network. Across all three scales, fast-charging piles dominate, indicating that minimizing waiting times is the paramount factor when seeking a good charging experience. Slow piles appear primarily at certain stations where they can aid economic balancing.

Specific placement patterns demonstrate these findings. In the small-scale scenario, candidate points 2, 3, and 5 were selected for station construction. Candidate point 2 received 6 fast piles; candidate point 3 received 5 fast charging piles; and candidate point 5 received 4 fast and 1 slow pile. In the medium scenario, candidate locations included distribution of fast charging stations at 16 of the 20 potential points. A larger number of slow piles (4 total) appeared at candidate locations where facility utilization will be lower. Table 5 shows how medium-scale place.

Table 5. Medium-Scale Station Configuration
Candidate Station Fast Piles Slow Piles
J2 4 1
J3 7 0
J4 5 0
J6 5 0
J7 5 0
J9 2 3
J10 10 0
J11 5 0
J13 6 0
J14 6 0
J15 5 0
J16 7 0
J17 5 0
J18 7 0
J19 5 0
J20 6 0

6.2 Sensitivity Analysis

(1) Weight coefficient \((\alpha,\beta)\) analysis

To explore solution flexibility, I systematically varied weights across objective components. Results are summarized in Table 6:

Table 6. Weight Coefficients and Their Effect on Layout
\(\alpha/\beta\) \(Z_1\) (10,000 yuan) \(Z_2\) (hours) Station Count Fast Piles Slow Piles
0.1/0.9 1758.0 17.0614 16 91 6
0.2/0.8 1657.2 17.2114 15 90 6
0.3/0.7 1563.6 17.5003 14 90 4
0.4/0.6 1563.6 17.5033 14 90 4
0.5/0.5 1560.0 17.5325 14 89 3
0.6/0.4 1543.2 17.7458 14 89 2
0.7/0.3 1528.8 18.0714 14 89 0
0.8/0.2 1528.8 18.0867 14 89 0
0.9/0.1 1515.6 19.1686 14 89 0

As user-time weight increases, station counts and service piles correspondingly increase. Optimizing user experience via reduced travel time results in total construction costs rising; conversely, budget-focused strategies may entirely eliminate slow-charging piles. Policy can adapt weight selection dynamically: when public acceptance is at low levels, explicit optimization of user travel time supports market development; as electric car populations mature, adopting cost-weighted models enables prudent infrastructure investment.

(2) Maximum waiting time \(G_{jk}\) analysis

I varied the permitted maximum waiting time (G) across multiple sensitivity tests. Table 7 provides results under medium-scale scenario:

Table 7. Waiting Time Variations
\(G_{jk}\) (hours) \(Z_1\) (10,000 yuan) \(Z_2\) (hours) Fast Piles Slow Piles
1.0 1754.4 17.9053 178 0
1.2 1698.0 17.6364 148 0
1.4 1648.8 17.5658 127 1
1.6 1611.6 17.5333 111 2
1.8 1585.2 17.5186 100 2
2.0 1560.0 17.5325 89 3

Longer allowed waits allow fewer fast piles and more slow piles, reducing cost but marginally increasing travel. Waiting time limits can be adjusted based on local service expectations: in upscale urban zones, strict \(G_{jk}\) values maintain customer satisfaction, while in peripheral districts where demand is less urgent, relaxed limits allow capital economy.

(3) Minimum station number \(S\) analysis

Finally, minimum required station quantity was varied at medium scale to analyze the effect on coverage and cost metrics, as in Table 8:

Table 8. Minimum Stations Sensitivity Results
\(S\) (stations) \(Z_1\) (10,000 yuan) \(Z_2\) (hours) Total Piles Fast Piles Slow Piles
12 1345.2 18.5703 89 89 0
13 1450.8 18.0369 90 89 1
14 1560.0 17.5325 92 89 3
15 1656.0 17.2222 94 90 3
16 1747.2 17.1006 94 90 4

As minimum station counts rise, total pile count increases and construction cost rises; nevertheless, user travel time declines with even more station dispersion, demonstrating that users derive proximity benefits. These findings imply that planning authorities should optimize nationwide infrastructure by deliberately balancing minimum station number against available budgets for infrastructure coverage.

7. Managerial Insights and Policy Implications

The simulation outcomes generate useful practical guidance for city governments, charging network operators, and electric car users:

1) Stage-based deployment: For regions at the adoption stage, orient optimal layouts toward high-density fast-charging stations around commercial hubs and interchanges, avoiding excessive initial investment levels. For mature markets, plan an integrated network with mixed-level coverage in residential and working areas, establishing strong base coverage.
2) Congestion-sensitive placement: Since congestion imposes meaningful time penalties within dense networks, strategically deploy station networks with awareness of both existing flows of congestion as well as ridership zones likely to drive EV charging needs.
3) Public-private collaboration: Since price-sensitive fast-charging expansion demands nontrivial capital, municipalities should proactively remove regulatory barriers and incentivize private investment through subsidies and public-private partnerships (PPPs); anchor installations in dense zones first.
4) Dynamic service standards: Use parameters such as maximum waiting times as planning tools in balance; if density is anticipated to rise in any one segment, consider stricter waiting-time metrics coupled to reduced user wait dissatisfaction as that sub-network matures.
5) Adaptive reconfiguration: Introduce routing data linkage and information systems so that electric car owners can be guided to less congested charging locations, while aggregators shift live charging load distribution across substations based on real-time demand.
6) Technology upgrades: As electric cars see continued integration with vehicle-to-grid operation, charging station systems should include capacity margins for future smart grid operation, load management, and newer high-output systems, so initial siting remains stable amid rises in electric car power demands.
7.1 Limitations

This research nonetheless carries several limitations. First, K-Means clustering, while robust and valid, may underperform when large-scale city multidimensional and very high-dimensional landscapes are presented; other clustering or aggregate demand models may detect subtler spatial structures. Second, generation of virtual points was based on POI data collection rather than recording actual driving and charging data. As a consequence, real users may produce different charging patterns from assumed behavior. Third, charging times generated randomly allowed each run to produce synthetic charging sessions but could not populate input charging sessions with genuine usage data, especially across user categories and varying driving modes.

8. Conclusion and Future Directions

Conclusion

In this thesis I developed a dual-objective optimization model for electric car charging station layout that explicitly incorporates traffic congestion. The optimization minimizes total charging station construction and operational costs and total user travel time. To reflect actual traffic constraints, user travel time, captured via real-time Amap traffic data, is adjusted with road congestion coefficients. I applied K-Means clustering to derive candidate station locations and solved the model with the LINGO software to produce the optimal charging layout under multiple scale scenarios. Detailed validation in Tianjin’s Nankai District showed:

(1) Scale-dependent network structure varies: small-scale installation tends to concentrate fast piles in high-demand zones to fulfill demand within low budgets, while medium-scale layout balances wider coverage with cost, and large-scale integrates significantly larger station counts and broader area with strategic slow-charging piles to moderate escalating expenditure. Overall, fast piles occupy a dominant share at 75%-90% of total piles across scale levels.

(2) Sensitivity tests reveal model adaptability: changing objective weighting toward user travel increases fast pile deployments and total costs; increasing waiting tolerance enables lower-cost slow pile deployments but degrades user experience.

(3) Minimum station requirements can deliver more rapid response and reduce user costs but force greater outlay.

These outcomes provide key implications for charging site selection-policy in urban settings, underlining trade-offs between carbon-efficiency policy support and infrastructure planning constraints.

Future work

Future research directions include adopting advanced clustering approaches, such as DBSCAN or Gaussian mixture models, which are capable of identifying more flexible boundaries and density-based structures. Integrating actual global positioning system (GPS) vehicle trajectory datasets would calibrate user location data from real travel and parking behavior. Complementing model data with actual charge use into charging sessions—via collaborations with charging networks—would quantify specific usage distributions. Modeling combinations of short- and long-wait users with generative adversarial networks could improve validity. Additional directions include incorporating dynamic congestion evolution over time-of-day, renewable generation availability, and power grid constraints in station plans to design integrated electrified transport-energy networks.

Scroll to Top