New Model Predicts EV Charging Load with Traffic Flow Insights
A groundbreaking study introduces a novel approach to forecasting electric vehicle (EV) charging demand by integrating traffic flow dynamics and vehicle state-of-charge behavior. This advancement offers a more accurate and efficient method for predicting when and where EVs will require charging, a critical factor for managing the stability of power grids as EV adoption accelerates.
The transition to electric mobility is reshaping not just our roads but also our power infrastructure. As millions of new EVs connect to the grid, their collective charging patterns create a complex and unpredictable load that can strain electricity networks, particularly during peak hours. Traditional methods for modeling this charging load have often fallen short, relying heavily on simulating the behavior of individual vehicles. While useful, these micro-level simulations can be computationally intensive and may fail to capture the emergent, large-scale patterns that arise from the interactions of thousands of drivers making routing and charging decisions simultaneously. The fundamental challenge lies in the inherent randomness of human behavior—when people decide to drive, where they choose to go, and when they decide to plug in are influenced by a myriad of factors, from daily routines to real-time traffic conditions. This randomness translates into a probabilistic distribution of charging demand across both time and space, making precise forecasting a significant hurdle for grid operators and urban planners.
Recognizing these limitations, researchers have long sought a more holistic, macroscopic view. The core insight is that EVs are not just batteries on wheels; they are integral components of a dynamic transportation system. Their movement is governed by the same principles of traffic flow and user equilibrium that dictate the behavior of all vehicles on a network. When one driver chooses a route, it affects the travel time for others, creating a feedback loop that ultimately leads to an equilibrium state where no single driver can improve their travel time by unilaterally changing their path. This concept, known as traffic equilibrium, has been a cornerstone of transportation engineering for decades. However, its application to the specific problem of EV charging load has been limited, with many existing models either ignoring traffic congestion effects or treating them as static, average conditions.
The new research, published in the journal Power System Technology, takes a significant leap forward by directly linking the dynamic state of the transportation network to the evolution of EV battery levels. The authors, Zhu Junliang and Wu Zhigang from the School of Electric Power at South China University of Technology, along with Liu Jianing from the Electric Power Dispatching Control Center of Guangdong Power Grid Co., have developed a sophisticated framework that marries a semi-dynamic traffic equilibrium model with a novel concept called “Combined State of Charge” (CSOC). This fusion allows them to move beyond the limitations of individual vehicle simulation and instead model the collective behavior of an entire EV fleet as a single, evolving probabilistic entity.
The methodology is built on a multi-layered foundation. At its core is the semi-dynamic traffic equilibrium model, which divides the day into multiple time periods. For each period, it calculates how traffic flow distributes itself across the road network based on the principle of user equilibrium, where drivers choose routes to minimize their travel time. Crucially, this model is “semi-dynamic” because it accounts for “traffic residual flow”—vehicles that start a trip in one time period but do not finish it until the next. This captures the real-world phenomenon of congestion, where a delay in one hour can ripple into the next, affecting when and where people arrive at their destinations. By using this model, the researchers can generate a highly realistic picture of how EVs are spatially distributed across a city at any given hour, including the impact of road capacity and length on travel times.
On the other side of the equation is the CSOC model. Instead of tracking the battery level of each car, this approach treats the entire population of EVs as having a combined state of charge that follows a probability distribution, initially assumed to be normal. As EVs travel, their collective SOC “consumes” at a rate proportional to the distance and energy efficiency of their chosen routes. The model then calculates how this probability distribution shifts and evolves over time. When an EV’s SOC drops below a predefined threshold, it generates a charging demand. The CSOC model can then determine the probability that a vehicle arriving at a destination (like a charging station) at a specific time has a low enough battery to require charging. By combining this probability with the number of arriving vehicles and the charging power, the model can compute the expected charging load for that location and time.
The true power of this integrated approach lies in its ability to capture the feedback loop between the transportation and power systems. For instance, if a major road becomes congested, the semi-dynamic traffic model shows that travel times increase. This means EVs take longer to reach their destinations, consuming more energy on the road. This increased energy consumption shifts the CSOC distribution, leading to a higher probability of vehicles needing to charge upon arrival. Furthermore, the congestion might cause some drivers to take longer, less direct routes, which further increases their energy consumption and charging demand. This cascading effect, where traffic conditions directly influence charging patterns, is something previous models struggled to represent accurately. The new model, however, captures this interdependence naturally within its framework.
To validate their approach, the research team conducted extensive simulations on two distinct networks. The first was a well-known 13-node test network, often used in academic research, which allowed them to perform detailed comparisons with established methods. The second was a large-scale, real-world network extracted from the OpenStreetMap data of the Tianhe District in Guangzhou, China, comprising nearly 6,000 road segments and 4,800 nodes. This large-scale test was crucial for demonstrating the model’s practical applicability to a real urban environment.
The results were compelling. When compared to traditional Monte Carlo simulation—a method that relies on running thousands of random vehicle trips—the new model produced charging load profiles that were virtually identical, confirming its accuracy. However, the difference in computational efficiency was staggering. In the 13-node network, the new method achieved its result in just 4.8 seconds, while a Monte Carlo simulation with 100 runs took over 4 hours. In the massive real-world network, the new model took 978 seconds (about 16 minutes), compared to over 11 hours for a 50-run Monte Carlo simulation. This represents a speedup of over 40 times, making the model not only accurate but also feasible for real-time or near-real-time applications, such as day-ahead grid scheduling.
The study also yielded valuable insights into the factors that influence charging demand. One key finding was the temporal lag between travel demand and charging demand. The peak in vehicle travel occurred at 6:00 PM, but the peak in charging load was observed at 7:00 PM. This one-hour delay is a direct consequence of the fact that EVs cannot charge while driving; they must complete their trip before plugging in. This insight is critical for grid operators who need to anticipate when the surge in electricity demand will hit.
Another significant finding concerned the impact of EV penetration rates. The research confirmed that as the percentage of EVs on the road increases, the total daily charging load increases in a directly proportional, linear fashion. This suggests that forecasting future charging demand can be relatively straightforward if the growth rate of EV adoption is known. However, the model also highlighted the profound influence of the underlying road network’s physical structure. When the researchers simulated a reduction in road capacity—simulating a bottleneck or a road closure—they observed a dramatic effect. Not only did the peak travel demand shift later into the evening, but the peak charging load also increased and shifted. This is because congestion forces vehicles to spend more time on the road, burning more energy, and sometimes taking longer detours, both of which lead to higher charging needs. Similarly, when road lengths were artificially increased, the charging peak was further delayed, and the average daily load rose, again due to increased energy consumption from longer travel distances.
These findings underscore a critical point: the power grid and the transportation network are not separate systems but are deeply intertwined. The design and operation of one have a direct and measurable impact on the other. A poorly designed road network that is prone to congestion will not only frustrate drivers but also create a more volatile and higher peak charging load for the power grid. Conversely, a well-optimized transportation system can help smooth out charging demand, making it easier and cheaper to manage.
The implications of this research are far-reaching. For utility companies, this model provides a powerful new tool for load forecasting. By incorporating real-time or predicted traffic data, they can generate much more accurate predictions of charging demand, allowing for better unit commitment, reduced need for expensive peaking power plants, and improved grid stability. For city planners and transportation authorities, the model offers a way to evaluate the “energy footprint” of different urban development and traffic management strategies. A new highway or a change in traffic signal timing can now be assessed not just for its impact on travel time but also for its impact on the city’s overall electricity consumption and carbon emissions.
For the burgeoning EV charging industry, the model can inform strategic decisions about where to place new charging stations. By predicting not just the volume of demand but also its precise timing and spatial distribution, operators can avoid overbuilding in some areas while underserving others. This leads to a more efficient and resilient charging infrastructure.
The authors acknowledge that the model can be further refined. Future work will incorporate the influence of charging prices into the route-planning decisions, as drivers may choose longer routes to reach a cheaper charging station. The model also currently assumes that charging only happens at the end of a trip, but in reality, many drivers engage in “opportunistic charging” during their journey. Incorporating these more complex user behaviors will make the model even more realistic. Additionally, the rise of autonomous and shared mobility could fundamentally change travel patterns, and the model will need to adapt to these new paradigms.
In conclusion, this research represents a paradigm shift in how we think about and model EV charging load. By moving from a micro-simulation of individuals to a macro-analysis of system-wide equilibrium, it provides a more accurate, efficient, and insightful tool. It successfully bridges the gap between transportation engineering and power systems analysis, demonstrating that a holistic, systems-level approach is essential for managing the complex challenges of our electrified future. As the world races toward a zero-emission transportation system, tools like this will be indispensable for ensuring that our power grids can keep up with the demand, paving the way for a smoother, more sustainable transition.
Zhu Junliang, Wu Zhigang, Liu Jianing. Power System Technology. DOI: 10.13335/j.1000-3673.pst.2023.0095