New Model Predicts EV Fast-Charging Demand with Urban Intelligence
As electric vehicles (EVs) continue to gain traction in urban environments, accurately forecasting their charging demands has become a critical challenge for city planners, utility operators, and infrastructure developers. Traditional models have often relied on simplified assumptions, such as dividing cities into functional zones like residential, commercial, or industrial areas, to estimate where and when EVs will need to charge. However, these methods frequently overlook the complex spatial dynamics and human behavioral patterns that shape real-world mobility. A groundbreaking study published in Southern Power System Technology introduces a novel approach that integrates urban spatial structure and driver psychology to deliver a more accurate and realistic prediction of fast-charging load distribution.
The research, led by Keqing Qu, Denghui Zhao, Ling Mao, Jinbin Zhao, and Chuan Yang from the Department of Electrical Engineering at Shanghai University of Electric Power, moves beyond the limitations of conventional zone-based models. Instead of treating cities as a collection of isolated functional areas, the team leverages real-world geographic data to capture the intricate web of urban life. Their method recognizes that modern cities are not rigidly segmented but are instead characterized by overlapping functions, multi-centered development, and dynamic human movement patterns. By incorporating these complexities, the model offers a more nuanced and reliable forecast of where fast-charging infrastructure will be most needed.
At the heart of this new approach is the use of Points of Interest (POI) data. POIs—such as restaurants, shopping centers, offices, parks, and medical facilities—are more than just map markers; they are indicators of human activity and urban function. The researchers collected a vast dataset of over 119,000 POIs from a major Chinese city, categorizing them into 16 types and assigning each a functional weight based on its influence on daily life. For instance, a large business park or a major hospital carries a higher “work” or “medical” weight than a small neighborhood convenience store. This weighting system allows the model to quantify the functional intensity of every part of the city.
To transform this raw POI data into a meaningful representation of the city’s spatial structure, the team employed a technique called kernel density analysis. This method calculates the concentration of POIs within a given radius around each point on a grid, creating a smooth surface that highlights areas of high activity. The result is a detailed map of the city’s “centers”—not just the traditional downtown core, but also secondary hubs and emerging suburban nodes. This analysis revealed a polycentric urban landscape, with clusters of activity scattered throughout the region, reflecting the modern trend of decentralization and the rise of self-contained “satellite” communities.
This spatial structure directly informs the simulation of EV travel behavior. Instead of assuming that trips occur randomly between predefined zones, the researchers used an improved gravity model to predict the flow of vehicles between different locations. In this model, the attractiveness of a destination is determined by its functional weight and its distance from the origin. A high-weight location, such as a major shopping mall, exerts a strong “pull” on travelers, but this pull weakens with distance. The model also incorporates the concept of “impedance,” which accounts for the friction of travel, making longer trips less likely. By combining these factors, the model can simulate realistic travel chains—sequences of trips that a driver might make in a single day, such as home-to-work, work-to-shopping, and shopping-to-home.
The study’s innovation extends beyond spatial modeling to the psychological aspect of driver decision-making. The researchers acknowledge that EV drivers are not perfectly rational actors who always seek the optimal charging station. Instead, they operate with “bounded rationality,” making decisions based on a combination of objective information and subjective preferences. A driver might prioritize a station that is slightly out of the way if it has a lower price, faster chargers, or is located in a familiar area. To capture this, the team developed a decision-making framework based on an “adsorption model,” which simulates how drivers evaluate their options under pressure.
In this framework, a driver’s urgency to charge is represented as a “friction” force. When a driver’s battery level drops, this friction increases, lowering their standards for what constitutes an acceptable charging station. Conversely, a driver with a healthy battery has higher standards and is more selective. The model also incorporates a driver’s “adsorption capacity,” which is their subjective evaluation of a station based on factors like price, available power, and detour distance. The decision to charge is made not when the best option is found, but when the first option that meets the driver’s current “satisfaction” threshold appears. This mirrors real-world behavior, where drivers often choose the first available charger when they are anxious about running out of power.
To validate their model, the research team conducted a large-scale simulation involving 50,000 virtual EVs navigating the road network of the study city over a 24-hour period. The simulation was built on a detailed digital map of 78,748 road segments and 2,426 simplified road links after grid processing. Each vehicle was assigned a unique travel chain, battery capacity, and initial state of charge based on statistical distributions derived from real-world data. As the simulation progressed, the model tracked the location, speed, battery level, and charging status of every vehicle in one-minute intervals.
The results of the simulation were both revealing and significant. When the model accounted for the city’s complex spatial structure and the drivers’ bounded rationality, the predicted fast-charging demand was markedly different from models that used simple functional zoning. The average travel distance for EVs was 33.4 kilometers, a 24.4% reduction compared to 44.2 kilometers in a scenario that ignored POI data. This shorter travel distance led to a 44.3% decrease in long-distance trips over 60 kilometers, which in turn reduced the overall charging demand.
The temporal distribution of charging load also showed a distinct pattern. Two clear peaks emerged: one in the late morning (between 480 and 660 minutes after midnight) and another in the late afternoon (between 1,020 and 1,200 minutes). The first peak corresponds to drivers who begin their day with a partially charged battery and need to recharge during their initial commute. The second, larger peak reflects the end-of-day charging surge as drivers return home and plug in their vehicles. The model predicted a peak load of approximately 3,200 kilowatts during the first peak and 2,340 kilowatts during the second.
Crucially, the spatial distribution of this load was highly concentrated around the city’s multiple activity centers, as identified by the POI analysis. The highest demand was observed at charging stations located near major commercial hubs, transportation corridors, and employment centers. This finding underscores the importance of placing fast-charging infrastructure not just in residential areas, but along the dynamic routes of daily urban life. In contrast, a model that ignored spatial structure predicted a much more centralized load, concentrated almost entirely in the old downtown core, which would have led to a significant underestimation of demand in the growing suburban and secondary centers.
The implications of this research are far-reaching. For city planners, it provides a powerful tool for optimizing the placement of new charging stations, ensuring that investments are made where they will have the greatest impact. For utility companies, it offers a more accurate forecast of peak electricity demand, enabling better grid management and the prevention of overloads. For EV manufacturers and fleet operators, it can inform the design of navigation and charging recommendation systems that better align with real driver behavior.
The study also highlights the evolving nature of urban mobility. The shift from a single-centered to a polycentric city structure, driven by economic and demographic changes, has profound effects on transportation patterns. By capturing this shift, the model is not just a predictor of today’s demand but a forward-looking tool that can adapt to future urban development. It is equally applicable to the planning of charging networks in emerging cities and the retrofitting of infrastructure in mature urban environments.
One of the key strengths of this research is its grounding in real-world data and its adherence to established scientific principles. The use of kernel density analysis, gravity models, and probabilistic distributions for travel behavior ensures that the model is both theoretically sound and empirically validated. The incorporation of bounded rationality adds a layer of behavioral realism that is often missing from purely technical models. This holistic approach, which bridges the gap between geography, transportation engineering, and human psychology, represents a significant advancement in the field of EV load forecasting.
The research team acknowledges that their model is not without limitations. It relies heavily on the quality and availability of POI data, and its accuracy could be further improved by integrating additional data sources, such as population density, economic activity, and public transit schedules. Future work will explore these avenues to create an even more comprehensive picture of urban mobility.
In conclusion, the study by Qu, Zhao, Mao, Zhao, and Yang presents a paradigm shift in how we think about and predict EV charging demand. By moving beyond simplistic functional zones and embracing the complexity of urban space and human behavior, they have developed a model that is not only more accurate but also more insightful. As cities around the world strive to achieve their carbon reduction goals and build sustainable transportation systems, tools like this will be essential for making informed, data-driven decisions. The future of urban mobility is not just electric; it is intelligent, adaptive, and deeply connected to the fabric of the city itself.
Keqing Qu, Denghui Zhao, Ling Mao, Jinbin Zhao, Chuan Yang, Department of Electrical Engineering, Shanghai University of Electric Power, Southern Power System Technology, DOI: 10.13648/j.cnki.issn1674-0629.2024.10.015