Shared EV Demand Tied to Urban Design and Location
A new study from Southwest Jiaotong University reveals that the success of shared electric vehicles (SEVs) in urban environments depends heavily on a complex interplay of location, infrastructure, and city planning. The research, conducted by Liao Yang, Luo Xia, and Wang Hongjie, identifies key factors that influence user demand, offering practical guidance for operators aiming to optimize their services and for city planners seeking sustainable transportation solutions.
The findings, published in the Journal of Transportation Engineering and Information, challenge conventional models used to predict travel behavior. By analyzing operational data from over 1,400 shared electric vehicle stations in Chengdu, China, the researchers discovered that traditional statistical models often fail to capture the nuanced, non-linear relationships that govern user choices. Instead, a more sophisticated approach is required—one that accounts for the spatial and temporal patterns inherent in urban mobility.
Chengdu, a city at the forefront of new energy vehicle adoption, provided a rich dataset for the investigation. The team examined two months of station status information, including vehicle availability, capacity, and location, to infer trip origins and volumes. This granular data, combined with detailed geographic information on points of interest (POIs), road networks, and public transit, allowed the researchers to build a comprehensive picture of the factors driving SEV usage.
One of the study’s most significant contributions is its methodological innovation. The authors developed an improved spatial aggregation technique to define the service areas around each station. Moving beyond simple grid-based methods, they used a hexagonal tessellation with a 500-meter radius, reflecting the typical walking distance users are willing to cover to access a vehicle. This approach minimized errors in identifying spatial interactions between stations, a common flaw in previous research. Furthermore, they introduced a refined framework for quantifying the built environment, expanding the traditional “5D” model (density, diversity, design, destination accessibility, and distance to transit) to include not just the density of POIs, but also their proximity and spatial agglomeration.
The analysis confirmed that SEV usage in Chengdu exhibits strong temporal and spatial patterns. Trips are significantly higher on non-working days, and the daily peak occurs during the evening and night hours, aligning with leisure and entertainment activities rather than traditional commuting. This pattern is consistent with Chengdu’s reputation for a vibrant night life. The data also showed a clear spatial clustering of demand, with significant positive spatial autocorrelation across all time periods. This means that high-demand areas tend to be surrounded by other high-demand areas, a phenomenon driven by user behavior—when a station is empty, users naturally search for the nearest available vehicle.
To uncover the underlying drivers of this demand, the research team constructed and compared three different analytical models: a Generalized Linear Model (GLM), a Random Forest (RF) model, and a Generalized Additive Mixed Model (GAMM). The results were telling. While the Random Forest model achieved the highest raw accuracy in fitting the data, it did so at the cost of interpretability, functioning as a “black box” that could not explain the mechanisms behind its predictions. The GLM, a standard tool in transportation research, performed poorly, failing to capture the non-linear relationships and leaving significant spatial patterns unexplained in its residuals.
The GAMM emerged as the superior model. By incorporating smooth, non-parametric functions, it could accurately model the complex, non-linear effects of various factors. Crucially, it also included a random effect for the station’s location, allowing it to account for unobserved spatial heterogeneity. The success of the GAMM was validated by a key test: the spatial autocorrelation of its residuals. A good model should leave no systematic spatial pattern in its errors. The GAMM passed this test, with its residuals showing no significant spatial autocorrelation for most time periods, proving its ability to fully explain the spatial dependencies in the data. This finding underscores the critical importance of using spatially-aware models for urban transportation analysis.
The insights derived from the GAMM are particularly valuable for shared mobility operators. The study found that station attributes are the most influential factor in determining demand. Among these, parking capacity has a profound, non-linear effect. Contrary to the simple assumption that more parking is always better, the research identified a clear threshold. Demand increases with capacity up to a point—around 70 parking spaces per aggregated area—but beyond this threshold, additional capacity actually suppresses demand. This “inverted U” relationship suggests a saturation point where an oversupply of parking leads to inefficiencies, potentially making it harder for users to find a vehicle amidst a large, empty lot, and increasing operational costs for the provider. This finding provides a concrete target for capacity planning, advising operators to avoid over-provisioning in any single area.
Another critical station attribute is the distance between neighboring stations. The study revealed a “sweet spot” for station spacing. When stations are too close together, they cannibalize each other’s demand, leading to internal competition and underutilization. Conversely, when stations are too far apart, users may not find the service convenient enough to use. The optimal distance, identified as approximately 2 kilometers, represents a balance between minimizing competition and maximizing service coverage. This precise guidance can inform strategic decisions about where to place new stations or whether to consolidate underperforming ones.
Pricing is another powerful lever. As expected, lower pricing is strongly correlated with higher demand. The relationship is nearly linear, indicating that users are highly sensitive to cost. This reinforces the need for flexible and competitive pricing strategies, potentially including dynamic pricing based on time of day or demand levels, to attract and retain users.
The research also provides a nuanced understanding of how SEVs interact with the broader public transportation network. The relationship with buses is relatively straightforward: SEV demand decreases as the distance to the nearest bus stop increases beyond 400 meters, suggesting that SEVs are not a primary competitor to buses but rather a complementary service for areas with poor bus access.
The interaction with the metro system is more complex and reveals a strategic opportunity. Within a 2-kilometer radius of a metro station, SEV demand is suppressed. This is logical, as the metro is a highly efficient mode for longer trips in the city core. However, beyond this 2-kilometer buffer, the relationship reverses. SEV demand increases with distance from the metro. This indicates a powerful “last-mile” and “first-mile” potential. SEVs can effectively serve as a feeder service, connecting residential areas on the urban periphery to the high-capacity metro network. This finding is crucial for city planners, as it suggests that investing in SEV infrastructure in these “transit deserts” can significantly improve overall accessibility without the massive capital expenditure required for extending rail lines.
The study confirms that SEVs are primarily used for non-commute trips. While proximity to office buildings and business districts is important, the strongest positive correlations are with entertainment venues, universities, and medical facilities. The density of restaurants, cinemas, and parks is a major draw. This highlights the role of SEVs in enabling leisure and social activities. Proximity to universities is particularly significant, suggesting that students are a key user demographic, likely due to the affordability and flexibility of the service. The presence of medical facilities, especially when they are clustered together, also boosts demand, possibly because users seek a convenient and private mode of transport for medical appointments, particularly during off-peak hours when public transit is less frequent.
The land use mix, measured by land-use entropy, also follows a non-linear pattern. Demand is highest in areas with a moderate mix of land uses but declines in areas with very high entropy—essentially, the densest, most mixed-use urban cores. This counterintuitive result may be explained by several factors. In the most central areas, the availability of numerous transportation options (metro, buses, taxis, walking) may make SEVs less necessary. High land values also make it more expensive to operate large parking lots, potentially limiting station capacity. Furthermore, traffic congestion in these areas could make driving an SEV less attractive.
The study also evaluated the impact of major transportation hubs. SEVs show strong potential for connecting passengers from long-distance bus stations and railway stations to their final destinations within the city. The presence of a shared EV station adjacent to these hubs is a significant positive factor, as it provides a convenient alternative to taxis or public transit for travelers with luggage or specific destinations. Airports, however, present a different case. The research found a negative correlation between airport proximity and SEV demand. This is likely because airports are typically located on the urban fringe, where SEV penetration may be lower, and because they are well-served by dedicated shuttle services, taxis, and ride-hailing, making the SEV a less competitive option.
From a city planning perspective, the implications are clear. The research advocates for a dual approach. At a macro level, planners should prioritize SEV deployment in areas with moderate development intensity, along secondary road networks (which improve local access without the congestion of major arterials), and in the 2-kilometer buffer zones surrounding metro stations. At a micro level, operators should focus on placing stations near universities, entertainment districts, and medical centers, ensuring adequate but not excessive capacity, and maintaining optimal spacing between stations.
The study acknowledges some limitations, such as the lack of socioeconomic data for different neighborhoods and the inability to analyze trip origins and destinations (OD pairs) due to data constraints. However, its robust methodology and actionable findings provide a significant advancement in the field. By moving beyond simplistic models and embracing the complexity of urban space, the research offers a blueprint for creating more efficient, equitable, and sustainable shared mobility systems. It demonstrates that the future of urban transportation lies not in a single mode, but in a well-integrated ecosystem where services like shared EVs fill critical gaps, and their success is determined by a deep understanding of the urban fabric they serve.
Liao Yang, Luo Xia, Wang Hongjie, Southwest Jiaotong University, Journal of Transportation Engineering and Information, DOI: 10.19961/j.cnki.1672-4747.2024.05.011