New Model Predicts EV Charging Demand Using Traffic Flow Theory
As electric vehicles (EVs) continue to gain popularity worldwide, one of the most pressing challenges for urban planners and energy grid operators is accurately predicting when and where EV charging demand will occur. Traditional models, which rely on simulating individual vehicle behavior through statistical sampling, have long struggled to capture the complex, large-scale interactions that shape real-world traffic and charging patterns. Now, a team of researchers from South China University of Technology and Guangdong Power Grid has introduced a groundbreaking approach that leverages traffic equilibrium theory to model EV charging load with unprecedented accuracy and efficiency.
The new method, detailed in a recent paper published in Power System Technology, moves away from micro-level simulations and instead adopts a macroscopic, system-wide perspective. By integrating semi-dynamic traffic equilibrium with a novel concept called Combined State of Charge (CSOC), the model captures how the collective behavior of EV drivers—shaped by traffic congestion, route choices, and charging thresholds—directly influences the spatial and temporal distribution of electricity demand.
The significance of this work lies in its ability to address a critical gap in current EV infrastructure planning. As EV adoption accelerates, utilities face increasing pressure to manage the strain on power distribution networks. Unpredictable charging patterns can lead to localized grid overloads, especially during peak hours. Existing models, often based on Monte Carlo simulations, require extensive computational resources and repeated random sampling, making them impractical for real-time planning or large-scale scenario analysis. More importantly, they fail to account for the feedback loop between vehicle routing and traffic congestion—a key factor in urban environments.
“The traditional approach treats each EV as an isolated agent making independent decisions,” explains Dr. Zhu Junliang, lead author of the study and a researcher at the School of Electric Power, South China University of Technology. “But in reality, the path a driver chooses depends on what other drivers are doing. If too many vehicles take the same route, it becomes congested, which in turn influences the decisions of others. This interaction creates a system-level equilibrium that individual-based models simply can’t capture.”
To overcome this limitation, Zhu and his colleagues turned to principles from transportation science. Traffic equilibrium theory, a well-established framework in urban planning, describes how vehicle flows distribute themselves across a network such that no single driver can reduce their travel time by unilaterally changing routes. This state of equilibrium reflects the collective optimization of travel behavior under real-world constraints such as road capacity and congestion.
The team adapted this theory into a “semi-dynamic” model, meaning it divides the day into multiple time intervals—typically 15 to 90 minutes—each treated as a static equilibrium problem. However, unlike purely static models, it accounts for the “residual flow” of vehicles that do not reach their destination within a single time period, carrying over into the next. This dynamic propagation allows the model to simulate how delays accumulate during rush hours and how they ripple through the system over time.
A key innovation in the model is the introduction of random utility theory to reflect real-world driver behavior. In practice, not all drivers have perfect information about traffic conditions. Some may rely on outdated navigation data, while others may prefer familiar routes over objectively faster ones. The model incorporates this uncertainty by assigning a probability to each route choice, based on its perceived travel time. This probabilistic approach ensures that the simulated traffic flow is not only efficient but also realistic, reflecting the diversity of human decision-making.
Once the traffic flow is established, the model shifts focus to the electrical side of the equation: the charging behavior of EVs. Instead of tracking individual battery levels, the researchers use the concept of CSOC—Combined State of Charge—to represent the statistical distribution of battery levels across the entire EV population. At the start of the simulation, the initial SOC of EVs is assumed to follow a normal distribution, reflecting the natural variation in how much charge different vehicles have when they begin their journeys.
As vehicles travel, their SOC decreases in proportion to the distance and energy consumption of their chosen route. The model tracks how this distribution evolves over time, shifting leftward as energy is consumed. When an EV’s SOC drops below a predefined threshold—set at 20% in the study, to ensure the battery is never fully depleted—the driver is assumed to seek a charging station. The model then calculates the probability that a vehicle arriving at a charging node will need to charge, based on its current SOC distribution.
Charging itself is modeled with attention to real-world battery dynamics. The researchers note that fast-charging power typically declines once the battery reaches 80% state of charge, due to electrochemical limitations. Therefore, charging is assumed to occur at full power only up to this threshold. The duration of a charging session, and thus the instantaneous load on the grid, depends on the vehicle’s starting SOC when it plugs in. By integrating the probability distribution of arrival SOC with the charging power profile, the model computes the expected charging load at each time step.
One of the most compelling aspects of the research is its validation on both small-scale and real-world networks. The team first tested the model on the well-known 13-node Nguyen-Dupuis network, a standard benchmark in transportation research. Here, they compared their results against multiple Monte Carlo simulations, which involve running thousands of individual vehicle trajectories to estimate average behavior. The findings were striking: as the number of Monte Carlo trials increased, the simulated charging load converged toward the result produced by the new model. With 100 simulation runs, the Monte Carlo result closely matched the theoretical model, confirming its accuracy.
However, the performance difference was dramatic. While 100 Monte Carlo runs took over four hours to complete, the new model produced its result in just 4.8 seconds—a speedup of more than 3,000%. This efficiency makes the model suitable not only for academic research but also for practical applications such as real-time grid management and long-term infrastructure planning.
To demonstrate scalability, the researchers applied the model to a real-world urban network derived from OpenStreetMap data, covering a significant portion of Tianhe District in Guangzhou, China. This network included nearly 5,000 nodes and over 5,800 road segments, representing a level of complexity far beyond typical test cases. Even at this scale, the model remained highly efficient, completing the simulation in under 17 minutes—still vastly faster than any sampling-based method could achieve.
The results from both networks revealed several important insights. First, EV charging demand closely follows travel demand, but with a consistent time lag. In the simulations, peak travel occurred around 6:00 PM, while peak charging demand peaked at 7:00 PM. This delay reflects the fact that drivers must complete their trips before they can charge, a simple but often overlooked aspect of EV behavior.
Second, the model shows that increasing EV penetration leads to a proportional rise in average charging load, without altering the overall shape of the daily demand curve. This linearity suggests that as more EVs are added to the fleet, the timing of charging peaks remains predictable, provided that driving patterns stay constant. This finding is encouraging for grid operators, as it implies that scaling up EV adoption does not necessarily introduce new forms of unpredictability.
However, the structure of the road network itself plays a crucial role. When the researchers reduced road capacity to simulate congestion, they observed a significant shift in both travel and charging patterns. During peak hours, travel demand became delayed, pushing the charging peak later into the evening. Moreover, average charging load increased, even though the number of trips remained the same. This effect arises because congestion forces some drivers to take longer alternative routes, consuming more energy and thus requiring more charging.
Similarly, when road lengths were artificially increased, the charging peak was further delayed, and total energy consumption rose. This demonstrates that urban design—such as the layout of roads and the placement of charging stations—has a direct impact on electricity demand. Cities with sprawling, inefficient road networks may face higher and more volatile charging loads than those with compact, well-connected layouts.
These findings have profound implications for the integration of transportation and energy systems. As EVs become a major component of both sectors, it is no longer sufficient to plan roads and power grids in isolation. The research underscores the need for coordinated, cross-domain planning that considers how changes in one system affect the other. For example, building a new highway may not only alter traffic flows but also reshape the spatial and temporal profile of electricity demand, potentially requiring upgrades to local substations or the deployment of smart charging strategies.
The model also opens new possibilities for policy design. By accurately forecasting charging demand under different scenarios, planners can evaluate the impact of interventions such as congestion pricing, dynamic tolls, or incentives for off-peak charging. The ability to simulate how such policies influence both traffic flow and grid load allows for more holistic decision-making.
Looking ahead, the research team plans to extend the model in several directions. One key area is the incorporation of charging prices into the route optimization process. In reality, drivers may choose to charge at a particular station not just based on proximity, but also on cost. By integrating real-time electricity pricing, the model could simulate how economic signals influence both travel and charging behavior.
Another important extension is the inclusion of en route charging. The current model assumes that charging occurs only at destinations, but many EV drivers now plan long trips with intermediate charging stops. Modeling this behavior would require a more complex representation of trip chains and driver decision-making, but it would significantly improve the model’s realism.
Finally, the rise of autonomous and shared mobility presents new challenges and opportunities. Self-driving EVs could be programmed to optimize both travel time and charging cost, potentially leading to very different patterns of energy use. Shared fleets might operate continuously, requiring frequent, distributed charging throughout the day rather than concentrated evening loads. Adapting the model to these emerging paradigms will be essential for future urban planning.
In conclusion, the work by Zhu Junliang, Wu Zhigang, and Liu Jianing represents a significant advance in the field of EV load modeling. By grounding their approach in fundamental principles of traffic flow and probability theory, they have developed a tool that is not only faster and more scalable than existing methods but also more insightful. It provides a clear, mechanistic explanation for how the collective behavior of EV drivers shapes the electrical load on the grid—a perspective that is essential for building resilient, sustainable urban energy systems.
As cities around the world accelerate their transition to electric mobility, models like this will play a crucial role in ensuring that the power grid can keep pace. The days of treating EVs as isolated consumers of electricity are over. The future belongs to integrated, system-level thinking—and this research is a major step in that direction.
Zhu Junliang, Wu Zhigang, Liu Jianing, South China University of Technology and Guangdong Power Grid, Power System Technology, DOI: 10.13335/j.1000-3673.pst.2023.0095