With the rapid growth of the electric car market, particularly in China where government policies and technological advancements have fueled the adoption of new energy vehicles, understanding the safety implications of these vehicles in complex traffic environments has become paramount. The proliferation of China EV models, such as those produced by local manufacturers, has transformed urban and highway landscapes, necessitating detailed studies on driving behavior under specific conditions. Interchange merging areas on highways are critical zones where vehicles from ramps merge into mainline traffic, often leading to conflicts and accidents due to the need for speed coordination and lane-changing maneuvers. Statistics indicate that a significant portion of highway accidents occur in these areas, highlighting the urgency for targeted safety evaluations. For electric cars, which exhibit distinct acceleration, deceleration, and handling characteristics compared to traditional internal combustion engine vehicles, drivers may adopt more aggressive behaviors, potentially increasing the risk of collisions. This study focuses on assessing the safety of driving behavior for pure electric vehicles in interchange merging areas using a driving simulation approach, incorporating vehicle interaction patterns and advanced modeling techniques to provide actionable insights for improving road safety.
The unique attributes of electric cars, such as instant torque delivery and regenerative braking systems, can influence driver behavior, especially in scenarios requiring rapid decision-making. In China, the EV industry has seen exponential growth, with models like the BYD Han and NIO ES6 becoming common on roads, yet drivers may lack experience in handling these vehicles in high-stakes environments like highway interchanges. Previous research has explored general driving behavior in merging areas, but few studies have specifically addressed the safety aspects of electric cars, considering the interactive dynamics between vehicles. This gap is critical, as the increasing share of China EV in the global market underscores the need for tailored safety frameworks. By employing a six-degree-of-freedom driving simulator, this study replicates real-world interchange scenarios to collect high-frequency data on vehicle operations and driver inputs, enabling a comprehensive analysis of safety performance.
To quantify driving behavior safety, this study utilizes the Slack-Based Measure Data Envelopment Analysis (SBM-DEA) model, a non-parametric method that evaluates efficiency based on multiple inputs and outputs without requiring predefined weights. This approach is particularly suited for assessing the safety of electric car operations, as it accommodates the multi-dimensional nature of driving metrics, such as longitudinal acceleration and time to collision. The model’s ability to handle non-radial slacks makes it ideal for identifying inefficiencies in driving behavior that could lead to accidents. In the context of China’s evolving transportation infrastructure, where electric cars are increasingly integrated into highway systems, this methodology offers a robust tool for policymakers and automakers to enhance safety standards. The subsequent sections detail the experimental design, data processing, and results, emphasizing the role of driver experience and scenario-specific factors in determining safety outcomes for electric vehicles.
Methodology: Driving Simulation Experiment and SBM-DEA Model
The driving simulation experiment was conducted using a high-fidelity, six-degree-of-freedom simulator that replicated the dynamics of a popular electric car model, similar to the Roewe Ei5, which is a common China EV. This simulator included a full-scale驾驶舱, motion base, and panoramic LED displays to create an immersive environment, with audio systems simulating road and vehicle noises. The electric car’s kinetic parameters, such as acceleration curves and braking response, were calibrated to mirror real-world performance, ensuring that the simulation accurately reflected the behavior of a typical electric car. A total of 40 drivers participated in the study, comprising 25 novice drivers (with less than 2 years of experience and under 1,000 km of driving mileage) and 15 experienced drivers (over 4 years of experience and more than 10,000 km). This division allowed for a comparative analysis of how driving proficiency affects safety in electric cars, particularly in the context of China’s growing EV market where novice drivers may be transitioning from conventional vehicles.
The experimental scenarios were based on a real interchange in Shanghai, where two highways converge, and focused on the eastern merging area. Five distinct interaction patterns were designed to cover common merging behaviors: passive cut-in, lateral interaction, longitudinal interaction, active front cut-in (right side), and active rear cut-in (right side). Each scenario involved controlled interactions with other vehicles to simulate typical traffic densities, ensuring that drivers faced realistic challenges while operating the electric car. For instance, in the passive cut-in scenario, a vehicle from the ramp merged into the mainline, requiring the participant to adjust speed and position accordingly. Data were collected at a frequency of 100 Hz, capturing variables such as vehicle position, speed, acceleration, yaw rate, steering wheel angle, and time to collision (TTC). The TTC metric, defined as the time until a collision would occur if vehicles continued on their paths, was used to identify conflict segments, with severe conflicts defined as TTC values below 2.8 seconds and general conflicts between 2.8 and 4.7 seconds. In total, 115 valid interaction segments were extracted from the data for analysis, representing various scenarios encountered by drivers of electric cars.
The SBM-DEA model was applied to evaluate the safety efficiency of driving behavior, with an input-oriented approach under variable returns to scale. This orientation was chosen because the goal is to minimize risky inputs (e.g., harsh accelerations) while achieving desired safety outputs, aligning with the objective of enhancing electric car safety. The input indicators included longitudinal acceleration, lateral acceleration, yaw rate, and steering wheel angle rate, which reflect vehicle dynamics and driver control actions. The output indicators consisted of driving speed (as a measure of mobility), TTC (representing safety margins), and severe conflict exposure time (the duration where TTC < 2.8 s, indicating high risk). The mathematical formulation of the SBM-DEA model for n decision-making units (DMUs), each with m inputs and s outputs, is given by:
$$ \min \rho = 1 – \frac{1}{m} \sum_{i=1}^{m} \frac{s_i^-}{x_{ik}} $$
subject to:
$$ X \lambda + s^- = x_k $$
$$ Y \lambda \geq y_k $$
$$ e \lambda = 1 $$
$$ \lambda, s^- \geq 0 $$
where ρ is the efficiency score (with ρ = 1 indicating full efficiency), s_i^- represents the slack variables for inputs, X and Y are the matrices of input and output vectors, λ is the intensity vector, and e is a unit vector. This model allows for a relative comparison of driving behavior safety among participants, accounting for the specific characteristics of electric cars, such as their responsive acceleration, which might lead to different input patterns compared to traditional vehicles. The stability of the DEA model was verified by assessing the ranking consistency across subsets of data, ensuring that the results were robust and reliable for drawing conclusions about China EV safety.
| Scenario | Interaction Type | Description | Number of Valid Segments |
|---|---|---|---|
| 1 | Passive Cut-in | A vehicle merges from the ramp into the mainline, requiring the electric car driver to react. | 37 |
| 2 | Lateral Interaction | A vehicle changes lanes from an adjacent lane, posing a side conflict. | 40 |
| 3 | Longitudinal Interaction | A leading vehicle brakes suddenly, testing following distance. | 7 |
| 4 | Active Front Cut-in (Right) | The electric car merges from the ramp, interacting with a vehicle ahead. | 14 |
| 5 | Active Rear Cut-in (Right) | The electric car merges from the ramp, interacting with a vehicle behind. | 17 |
The data processing involved extracting the interaction segments based on TTC thresholds, with each segment representing a unique driving episode for the electric car. For example, in Scenario 3, only 7 segments were valid due to the specific nature of longitudinal interactions, where drivers often maintained safe distances, reducing the number of conflicts. The input and output indicators were normalized to ensure comparability, and the DEA model was implemented using software tools to compute efficiency scores for each DMU (i.e., each driver-scenario combination). The weights for aggregating scores across scenarios were derived from the proportion of segments in each scenario relative to the total, reflecting the relative risk exposure. This methodology not only addresses the safety of electric cars in interchange merging areas but also contributes to the broader understanding of how China EV technologies can be optimized for better driver adaptation and road safety.
Results and Analysis: Model Stability and Safety Performance
The stability of the SBM-DEA model was assessed by examining the consistency of efficiency rankings when the dataset was perturbed, such as by removing random subsets of DMUs. Results showed that over 85% of the DMUs exhibited ranking changes with a standard deviation of less than 4, indicating high model stability. This robustness is crucial for applications in real-world safety evaluations, especially for electric cars where driving behavior data can be volatile due to the vehicles’ unique dynamics. The stable rankings validate the use of the DEA model for comparing novice and experienced drivers, providing confidence in the subsequent safety analysis for China EV operations.
The efficiency scores, representing driving behavior safety, were calculated for each of the 115 interaction segments, and then aggregated by driver type and scenario. The overall safety score for each driver was computed as a weighted average based on the frequency of scenarios, using the formula:
$$ Q_k = \sum_{i=1}^{5} w_i \rho_{ik} \times 100 $$
where Q_k is the safety score for driver k, w_i is the weight for scenario i (proportional to the number of segments), and ρ_ik is the efficiency score of driver k in scenario i. A score of 100 indicates optimal safety, meaning the driver minimized risky inputs while achieving good safety outputs. The results, summarized in Table 2, reveal that experienced drivers consistently achieved higher safety scores (mean of 76.66) compared to novice drivers (mean of 61.30), with lower standard deviations indicating more consistent performance. This trend underscores the importance of driving experience in safely operating electric cars, particularly in complex merging areas where quick decisions are required. The higher scores for experienced drivers suggest that they are better at managing the responsive nature of electric cars, such as modulating acceleration to avoid conflicts, which is a key consideration for the expanding China EV market.
| Scenario | Driver Type | Mean Safety Score | Standard Deviation |
|---|---|---|---|
| 1 | Novice | 62.84 | 29.83 |
| 1 | Experienced | 65.15 | 29.81 |
| 2 | Novice | 55.41 | 30.01 |
| 2 | Experienced | 82.48 | 26.66 |
| 3 | Novice | 75.19 | 22.04 |
| 3 | Experienced | 91.17 | 12.79 |
| 4 | Novice | 69.57 | 39.12 |
| 4 | Experienced | 59.33 | 30.67 |
| 5 | Novice | 75.69 | 35.84 |
| 5 | Experienced | 98.63 | 3.35 |
| Overall | Novice | 61.30 | 19.49 |
| Overall | Experienced | 76.66 | 18.38 |
Breaking down the results by scenario, notable differences emerged. In Scenario 2 (lateral interaction), experienced drivers scored significantly higher (82.48) than novices (55.41), indicating that experienced drivers are more adept at handling side conflicts, possibly due to better situational awareness and control of the electric car. Similarly, in Scenario 3 (longitudinal interaction), both groups performed well, but experienced drivers had a higher mean score (91.17) and lower variability, suggesting that maintaining safe following distances is a skill enhanced by experience, which is critical for electric cars with regenerative braking that may alter stopping patterns. In contrast, Scenario 4 (active front cut-in) showed novice drivers outperforming experienced ones (69.57 vs. 59.33), though with high variability. This anomaly might be attributed to the task being an expected merging maneuver for novices, who may prepare more carefully, whereas experienced drivers could become complacent. For Scenario 5 (active rear cut-in), experienced drivers nearly achieved perfection (98.63), with minimal deviation, highlighting their proficiency in managing rear interactions during merges—a common situation for electric cars in China’s busy highways.
Further analysis using regression models revealed that inputs like longitudinal acceleration and yaw rate had significant negative correlations with safety scores, meaning that higher values of these variables (indicating aggressive driving) reduced efficiency. For example, a unit increase in longitudinal acceleration decreased the safety score by approximately 0.15 points on average, emphasizing the need for smooth acceleration in electric cars to enhance safety. The output indicators, particularly TTC, showed that longer time to collision correlated with higher scores, validating its use as a safety metric. These findings align with the characteristics of electric cars, where instant torque can lead to rapid acceleration if not managed properly, posing risks in merging areas. The results underscore the potential for targeted training programs for drivers of China EV models, focusing on moderating control inputs to improve overall safety.
Discussion: Implications for Electric Car Safety and Future Research
The findings of this study have significant implications for the safety of electric cars, especially in the context of China’s rapidly evolving EV ecosystem. The higher safety scores and consistency among experienced drivers suggest that as drivers accumulate mileage in electric cars, they adapt to the vehicles’ dynamics, such as the immediate power delivery and energy recovery systems. This adaptation is crucial for reducing accidents in interchange merging areas, where the interplay between multiple vehicles requires precise control. For novice drivers, the lower scores indicate a learning curve, which could be addressed through specialized training modules that simulate merging scenarios in electric cars. Given the growing prevalence of China EV in global markets, automakers and policymakers could incorporate these insights into driver assistance systems, such as adaptive cruise control or lane-keeping aids, tailored to the unique behavior of electric cars.
Another key observation is the scenario-dependent performance, particularly the better scores for novices in expected merging tasks (Scenarios 4 and 5) compared to unexpected interactions (Scenarios 1 and 2). This suggests that predictability plays a role in driving behavior safety for electric cars. In expected scenarios, drivers can anticipate actions and adjust their inputs more effectively, whereas in unexpected situations, the rapid response of electric cars might lead to overcompensation. This highlights the importance of infrastructure design, such as clear signage and lane markings, to enhance predictability in interchange areas for electric car drivers. Moreover, the use of the SBM-DEA model proved effective in capturing these nuances, as it accounts for slack in inputs and outputs, providing a holistic view of safety efficiency. Future research could expand this model to include additional variables, such as driver eye-tracking data, to offer a more comprehensive assessment.
However, this study has limitations, such as the relatively small sample size and the focus on a specific type of electric car. As the China EV market diversifies with models ranging from compact cars to SUVs, further studies should compare different electric car types to generalize findings. Additionally, the driving simulation, while high-fidelity, may not fully replicate real-world conditions, such as weather variations or traffic congestion. Future work could integrate real-world data from connected electric cars to validate the simulation results. Despite these limitations, this research contributes to the broader goal of enhancing road safety for electric vehicles, emphasizing the need for continuous evaluation and adaptation in the era of sustainable transportation.
Conclusion
In conclusion, this study demonstrates the utility of the SBM-DEA model in evaluating the safety of driving behavior for electric cars in interchange merging areas, with a focus on the Chinese EV context. The results confirm that experienced drivers exhibit higher and more consistent safety scores, underscoring the role of driving proficiency in managing the unique characteristics of electric cars. The model’s stability and the scenario-based analysis provide valuable insights for improving driver training and vehicle design. As electric cars continue to gain traction in China and globally, such evaluations will be essential for mitigating risks and promoting safe integration into existing traffic systems. Future research should explore multi-dimensional assessments, including visual and cognitive factors, to further enhance the safety of electric vehicles in complex environments.