Smart Charging Network Model Boosts Urban EV Infrastructure Planning
As electric vehicles (EVs) continue their rapid ascent in global markets, cities face mounting pressure to build efficient, future-proof charging networks. The challenge lies not just in deploying more chargers, but in placing them strategically—where they are most needed, when they are most useful, and with the right capacity to serve diverse user behaviors. A groundbreaking study published in Zhejiang Electric Power introduces a novel optimization framework that redefines how urban EV charging stations are planned, factoring in dynamic user demand, real-world traffic patterns, and long-term economic sustainability.
Led by Qiong Wang from State Grid Beijing Electric Power Company, in collaboration with Qing Zou, Le Li, Chaoran Li from State Grid Beijing Daxing Power Company, and Xueying Yan from State Grid Smart Car Networking Technology Co., Ltd., the research presents a comprehensive model that integrates behavioral science, traffic engineering, and robust optimization theory. The outcome is a planning tool capable of minimizing total societal cost—balancing the investment burden on operators with the convenience and time savings for drivers.
The work, titled Optimal Siting and Sizing of Charging Stations Considering Dynamic Charging Demands of Users, was published in the September 2024 issue of Zhejiang Electric Power (Vol. 43, No. 09). It addresses a critical gap in existing EV infrastructure planning: most models rely on static assumptions about charging demand, often ignoring the variability in user behavior, traffic congestion, and the evolving nature of EV ownership patterns. This can lead to underutilized stations in low-traffic areas or overwhelmed hubs during peak hours—inefficiencies that undermine both user satisfaction and operator profitability.
Wang and her team argue that effective planning must begin with a deep understanding of how people actually use their EVs. To capture this complexity, the researchers employed trip chain theory, a method used in transportation planning to model sequences of trips made by individuals throughout the day. For private EV owners, the model simulates daily routines—leaving home, commuting to work, running errands, returning home—based on statistical distributions of departure times, trip durations, and parking durations in different urban zones such as residential, commercial, and employment districts.
For taxi drivers, whose behavior is more erratic and less routine-driven, the team used origin-destination (OD) matrices derived from traffic surveys. These matrices reveal the probability of a taxi moving from one zone to another during different times of the day. For example, between 6 a.m. and 6 p.m., taxis are more likely to travel from residential areas to commercial or work zones, reflecting morning commutes and business travel. In the evening, the flow shifts as drivers head back home or toward nightlife districts. After midnight, the movement becomes more evenly distributed, aligning with late-night shifts and reduced passenger demand.
By combining these behavioral models with Dijkstra’s algorithm—a classic pathfinding method—the researchers simulated the most time-efficient routes for thousands of virtual EVs across a real urban road network. But unlike traditional models that assume constant speeds, this study incorporated a speed-flow relationship that accounts for traffic congestion. The faster a road segment becomes saturated, the slower vehicles travel, which directly affects energy consumption. The model dynamically adjusts energy use based on real-time traffic conditions, ensuring that battery depletion is calculated more accurately.
To handle the inherent randomness in human behavior and traffic patterns, the team used Monte Carlo simulation, running thousands of scenarios to generate a probabilistic map of charging demand across space and time. The results revealed distinct temporal and spatial patterns. Private EV owners predominantly charge overnight in residential areas, with peak demand between 7 p.m. and 3 a.m. This aligns with off-peak electricity rates and the convenience of home charging. In contrast, electric taxis show two major charging peaks: one around midday, when drivers take breaks for meals and rest, and another in the late afternoon, just before the evening rush hour. These stations are typically located near commercial hubs and transit corridors, where fast-charging capability is essential.
This granular understanding of demand forms the foundation of the model’s optimization engine. The goal is to minimize the total annualized cost, which includes both the operator’s expenses—construction, equipment, and maintenance—and the user’s economic losses, such as time spent traveling to a station and waiting in line. The model treats both stakeholders as equally important, reflecting a balanced approach to infrastructure development.
The optimization process unfolds in two stages. First, the model determines the optimal number and location of charging stations. To do this, it uses a weighted Voronoi diagram, a geometric method that partitions space into regions based on proximity and capacity. Unlike a standard Voronoi diagram, which divides space purely by distance, the weighted version accounts for station size and service capability. A larger station with more chargers can “pull” demand from a wider area, effectively expanding its service zone. This ensures that high-capacity stations are placed where demand is dense, while smaller stations serve more localized needs.
Once the locations are set, the second stage focuses on capacity planning—how many high-power (80 kW) and low-power (30 kW) chargers to install at each site. High-power chargers reduce waiting time but are more expensive to install and operate. Low-power chargers are cheaper but may lead to longer queues, especially during peak hours. The model uses queuing theory to estimate average wait times based on arrival rates and service capacity, assuming that vehicles arrive according to a Poisson distribution.
However, the real innovation lies in how the model handles uncertainty. Traditional planning often relies on a single “typical day” scenario, which fails to account for fluctuations in demand due to weather, events, or seasonal variations. To address this, Wang and her colleagues introduced a robust optimization framework. Instead of assuming a fixed demand, they define an “uncertainty set” that includes a range of possible demand levels. The model then seeks a solution that performs well across all scenarios within this set, rather than being optimal for just one.
This is achieved through an adjustable “uncertainty budget,” which controls how much deviation from the expected demand the planner is willing to tolerate. A higher budget means preparing for more extreme scenarios, resulting in overcapacity and higher costs. A lower budget assumes more stable conditions, risking undercapacity during unexpected surges. By tuning this parameter, decision-makers can balance cost and resilience according to their risk tolerance.
To solve this complex, multi-objective optimization problem, the team developed an enhanced version of the particle swarm optimization (PSO) algorithm, called adaptive simulated annealing particle swarm optimization (ASAPSO). PSO is a metaheuristic method inspired by the social behavior of birds or fish, where a population of “particles” explores the solution space by adjusting their positions based on individual and collective best experiences. However, standard PSO can get stuck in local optima—solutions that are good but not globally optimal.
The ASAPSO algorithm improves on this by incorporating elements of simulated annealing, a technique that allows the search process to occasionally accept worse solutions in order to escape local traps. This is governed by a “temperature” parameter that starts high and gradually decreases, mimicking the cooling process in metallurgy. At high temperatures, the algorithm explores widely; as it cools, it converges toward the best solution. The adaptation is further refined by dynamically adjusting the learning rates and inertia weights based on the iteration stage, ensuring both exploration and exploitation are balanced.
The model was tested in a real-world case study: a 58-square-kilometer urban area in a northern Chinese city with a population of 250,000. The road network consisted of 48 nodes and 82 links, serving 3,000 private EVs and 1,100 electric taxis. Using the ASAPSO algorithm and weighted Voronoi partitioning, the model evaluated multiple planning scenarios, varying the number of stations from 5 to 12.
The results showed a clear cost optimum at seven stations. With fewer stations, users faced longer travel distances and higher congestion, increasing both time and energy costs. With more than seven, the marginal benefit of additional stations was outweighed by the rising construction and maintenance expenses. At seven stations, the total annualized cost reached its minimum: 4.0001 million yuan (approximately $550,000 USD). This included 943,600 yuan in annualized construction costs, 1.1072 million yuan in operational costs, and 1.95 million yuan in user-related losses (travel time, energy use, and waiting).
Each of the seven stations was assigned a mix of high- and low-power chargers based on local demand patterns. Stations in commercial zones, where taxis dominate, were equipped with more high-power chargers to support rapid turnover. Stations in residential areas had a higher proportion of low-power chargers, catering to overnight charging needs. The service areas were clearly delineated using the weighted Voronoi method, ensuring that no zone was underserved or overlapped unnecessarily.
To validate the robustness of the model, the researchers compared two scenarios: one assuming fixed, deterministic demand (Scenario 1), and another accounting for uncertainty using the robust optimization framework (Scenario 2). When a 20% increase in demand was simulated—representing future growth or unexpected spikes—the deterministic model struggled. Some stations became severely congested, with wait times exceeding acceptable limits. In contrast, the robust model, having anticipated such fluctuations, maintained manageable queue lengths by allocating extra capacity where needed.
This demonstrates a key advantage of the proposed approach: it builds resilience into the planning process. Rather than reacting to crises after they occur, cities can proactively design networks that adapt to change. As EV adoption accelerates and new vehicle types emerge—such as electric delivery vans or ride-sharing fleets—this flexibility will be crucial.
The implications of this research extend beyond China. Urban planners worldwide face similar challenges in deploying EV infrastructure efficiently. Many cities have invested heavily in charging networks, only to find that usage is uneven or that stations are underutilized due to poor siting. By incorporating dynamic demand modeling and robust optimization, this framework offers a more scientific, data-driven approach to infrastructure investment.
Moreover, the model supports integrated planning across transportation and energy systems. Charging stations are not just parking spots with plugs—they are nodes in a larger energy network. Their placement affects local grid loads, voltage stability, and even renewable energy integration. The study includes constraints on distribution network capacity and voltage levels, ensuring that new stations do not overload existing infrastructure.
From a policy perspective, the model provides a tool for cost-benefit analysis. Governments and utilities can use it to evaluate different investment strategies, assess the impact of subsidies, or simulate the effects of congestion pricing on charging behavior. It also supports equity considerations—by identifying underserved neighborhoods, planners can ensure that EV access is not limited to affluent areas.
The research also highlights the importance of data. Accurate OD matrices, traffic flow measurements, and user behavior surveys are essential inputs. As cities deploy more sensors and connected vehicles, the quality of this data will improve, making models like this even more powerful. In the future, real-time data could be used to dynamically adjust station operations or guide drivers to less congested locations.
While the model is sophisticated, its ultimate goal is practical: to make EV ownership more convenient and affordable. By reducing the time and effort required to charge, it removes a major barrier to adoption. For fleet operators, predictable wait times mean better scheduling and lower operating costs. For individual drivers, knowing that a charger is nearby and likely to be available reduces range anxiety.
In conclusion, the work by Wang Qiong and her colleagues represents a significant step forward in EV infrastructure planning. By moving beyond static, one-size-fits-all models, they offer a dynamic, adaptive framework that reflects the complexity of real-world urban mobility. Their integration of behavioral modeling, traffic dynamics, and robust optimization sets a new standard for smart city planning. As cities strive to meet climate goals and reduce emissions, tools like this will be essential in building sustainable, resilient transportation systems.
The model’s success in a real urban environment suggests it is ready for broader application. With further refinement and integration into municipal planning software, it could become a standard tool for cities worldwide. As the electric revolution accelerates, the ability to plan wisely—not just build more—will determine which cities lead the way.
Qiong Wang, Qing Zou, Le Li, Chaoran Li, Xueying Yan, State Grid Beijing Electric Power Company, State Grid Beijing Daxing Power Company, State Grid Smart Car Networking Technology Co., Ltd., Zhejiang Electric Power, DOI: 10.19585/j.zjdl.202409002