1. Introduction
As the adoption of electric vehicles (EVs) continues to rise globally, understanding their driving behavior in complex traffic scenarios has become increasingly critical. Interchange merging areas, in particular, pose significant safety challenges due to high-frequency vehicle interactions and maneuvering demands. In this study, we aimed to evaluate the safety of driving behavior for electric vehicles in these areas using a combination of driving simulation and data envelopment analysis (DEA).
Recent studies have highlighted that interchange merging areas account for approximately 70% of traffic accidents in highway interchanges . The unique dynamic characteristics of EVs, such as distinct acceleration and braking behaviors, may exacerbate risks in these zones . Unlike conventional fuel vehicles, EVs require drivers to adapt to new control dynamics, potentially leading to more aggressive maneuvers and conflict situations .
To address this gap, we conducted a comprehensive driving simulation experiment involving 40 drivers (15 experienced and 25 novice) operating electric vehicles. The study leveraged the SBM-DEA (Slack-Based Measure Data Envelopment Analysis) model to quantify safety performance, integrating multiple input and output indicators related to vehicle dynamics and collision risks .
2. Methodology and Experimental Design
2.1 Experimental Setup
The experiment was executed using a six-degree-of-freedom driving simulator at the Shanghai Research Center for Smart Mobility and Road Safety. The simulator featured a high-fidelity cockpit, panoramic LED displays, and acoustic systems to replicate real-world driving conditions . We adjusted the vehicle dynamics model based on the Roewe Ei5 to ensure the simulated EV mirrored real EV acceleration, deceleration, and braking behaviors .
2.2 Scenario Design
We recreated a real-world merging area from the Shanghai Shanghai Ring Expressway interchange, incorporating five vehicle interaction scenarios derived from field observations:
- Passive cut-in
- Lateral interaction
- Longitudinal interaction
- Active front cut-in (right)
- Active rear cut-in (right)
Each scenario included moderate-density ambient traffic to simulate typical merging conditions. Table 1 summarizes the interaction patterns and objectives of each scenario.
| Scenario ID | Interaction Pattern | Key Driver Task |
|---|---|---|
| 1 | Passive cut-in | Respond to a vehicle merging from the ramp |
| 2 | Lateral interaction | Adjust to a vehicle changing lanes adjacent to the main vehicle |
| 3 | Longitudinal interaction | Maintain safe following distance during 前方急刹车 (front vehicle emergency braking) |
| 4 | Active front cut-in (right) | Merge into the main road from the ramp ahead of adjacent vehicles |
| 5 | Active rear cut-in (right) | Merge into the main road from the ramp behind adjacent vehicles |
Table 1. Summary of Experimental Scenarios for Electric Vehicle Merging Behavior
2.3 Participant Recruitment
Forty licensed drivers (30 male, 10 female) were recruited, with an average age of 29.2 years. The sample included:
- 15 experienced drivers (≥4 years driving experience, >10,000 km accumulated mileage)
- 25 novice drivers (≤2 years driving experience, <1,000 km accumulated mileage)
All participants had no reported physical or mental health conditions that could affect driving performance .
2.4 Data Collection Protocol
The simulation platform collected vehicle operation data at 100 Hz, including:
- Longitudinal and lateral acceleration
- Yaw rate
- Steering wheel angle and its variation rate
- Driving speed
- Time to Collision (TTC)
- Vehicle position (x, y, z coordinates)
We defined interaction segments as periods where TTC ≤ 20 seconds, yielding 115 valid data segments for analysis .
3. SBM-DEA Model for Safety Evaluation
3.1 Model Formulation
The SBM-DEA model was chosen for its capability to handle multi-input and multi-output systems without requiring pre-defined weights, making it suitable for evaluating driving behavior safety . For a system with n decision-making units (DMUs), m input indicators, and s output indicators, the input-oriented SBM-DEA model is defined as:
\(\min \rho = 1 – \frac{1}{m} \sum_{i=1}^{m} \frac{s_i^{-}}{x_{ik}}\)\(\text{s.t.} \left\{ \begin{array}{l} X\lambda + s^{-} = x_k \\ Y\lambda \geq y_k \\ e\lambda = 1 \\ \lambda, s \geq 0 \end{array} \right.\)
Where:
- \(\rho\) = comprehensive efficiency score (\(\rho = 1\) indicates strong efficiency)
- \(s_i^{-}\) = slack variables for input indicators
- X, Y = input and output indicator matrices
- \(\lambda\) = linear programming solution vector
- e = a small positive constant to ensure solution feasibility
3.2 Indicator Selection
Input indicators were selected to reflect vehicle dynamics and driver control efforts:
- Longitudinal acceleration
- Lateral acceleration
- Yaw rate variation
- Steering wheel angle variation rate
Output indicators focused on safety and driving effectiveness:
- Driving speed (expected output)
- Time to Collision (TTC) (expected output)
- Severe conflict exposure time (TTC < 2.8 s) (unexpected output)
Table 2 summarizes the model parameters and indicators.
| Model Component | Setting | Indicators |
|---|---|---|
| Model orientation | Input-oriented | – Longitudinal acceleration – Lateral acceleration – Yaw rate variation – Steering wheel angle variation |
| Scale return | Variable | |
| Expected outputs | – Driving speed – Time to Collision (TTC) | |
| Unexpected outputs | – Severe conflict exposure time (TTC < 2.8 s) |
Table 2. SBM-DEA Model Parameters for Electric Vehicle Driving Safety Evaluation
4. Results and Analysis
4.1 Model Stability Verification
To validate the SBM-DEA model’s reliability, we conducted a stability analysis on the 115 DMUs. The results showed that over 85% of DMUs had ranking standard deviations ≤4, indicating strong model stability . This stability confirms the model’s suitability for evaluating driving behavior safety in electric vehicles.
4.2 Safety Score Calculation
The driving behavior safety score (\(Q_k\)) for each driver was calculated using a weighted aggregation of scenario-specific efficiency scores:
\(Q_k = \sum_{i=1}^{5} w_i \rho_{ik} \times 100\)
Where:
- \(w_i\) = weight of scenario i (proportional to its risk degree)
- \(\rho_{ik}\) = efficiency score of driver k in scenario i
4.3 Comparative Analysis of Novice vs. Experienced Drivers
Table 3 presents the safety score statistics for novice and experienced drivers across all scenarios.
| Scenario | Driver Type | Mean Score | Standard Deviation |
|---|---|---|---|
| 1 (Passive cut-in) | Novice | 62.84 | 29.83 |
| Experienced | 65.15 | 29.81 | |
| 2 (Lateral interaction) | Novice | 55.41 | 30.01 |
| Experienced | 82.48 | 26.66 | |
| 3 (Longitudinal interaction) | Novice | 75.19 | 22.04 |
| Experienced | 91.17 | 12.79 | |
| 4 (Active front cut-in) | Novice | 69.57 | 39.12 |
| Experienced | 59.33 | 30.67 | |
| 5 (Active rear cut-in) | Novice | 75.69 | 35.84 |
| Experienced | 98.63 | 3.35 | |
| Overall | Novice | 61.30 | 19.49 |
| Experienced | 76.66 | 18.38 |
Table 3. Safety Score Statistics for Novice and Experienced Drivers in Electric Vehicle Merging Scenarios
Key findings include:
- Experience effect: Experienced drivers showed significantly higher mean safety scores (76.66 vs. 61.30) and lower score variability, indicating more consistent safe behavior .
- Scenario impact: Novice drivers performed better in expected merging scenarios (Scenarios 4 and 5) than in unexpected interaction scenarios (Scenarios 1 and 2), with score differences of 7.33 and 10.78 points, respectively .
- Conflict type sensitivity: In lateral conflict scenarios (Scenarios 1 and 2), experienced drivers outperformed novices by 19.64 and 27.07 points, highlighting superior hazard response skills .
4.4 Efficiency Analysis by Interaction Pattern
Figure 1 illustrates the distribution of safety scores for all drivers. Experienced drivers consistently clustered at higher score ranges, while novice scores showed broader dispersion, especially in non-expected scenarios.
5. Discussion
5.1 Implications for Electric Vehicle Safety
The results highlight critical differences in how novice and experienced drivers handle EVs in merging areas. The higher safety scores among experienced drivers suggest that familiarity with EV dynamics and traffic interaction patterns significantly enhances safety . For novice drivers, the improved performance in expected scenarios indicates that structured merging tasks (e.g., intentional ramp merging) are safer than reactive responses to sudden interactions .
5.2 Model Applications
The SBM-DEA model proved effective in quantifying driving behavior safety for electric vehicles, offering a data-driven approach to evaluate driver performance without subjective weighting . This model could be integrated into EV driver training programs to identify high-risk behaviors and optimize training curricula.
5.3 Limitations and Future Work
Our study focused on vehicle dynamics and control inputs, neglecting driver visual cognition and physiological factors, which are critical in real-world driving . Future research should:
- Incorporate eye-tracking and biometric data to enhance evaluation comprehensiveness.
- Compare driving behaviors between EVs and internal combustion engine vehicles to identify EV-specific safety challenges.
- Develop scenario-specific safety indicators to address the heterogeneity of merging area risks.
6. Conclusion
In this study, we evaluated the safety of electric vehicle driving behavior in interchange merging areas using a combination of driving simulation and the SBM-DEA model. Key conclusions include:
- The SBM-DEA model demonstrated high stability, with over 85% of decision units showing consistent ranking results .
- Experienced drivers exhibited significantly higher safety scores and greater behavioral consistency than novices across all scenarios .
- Novice drivers performed safer driving in expected merging scenarios, highlighting the importance of task predictability in EV safety .
These findings provide valuable insights for improving EV safety in complex traffic environments, particularly in guiding driver training and intersection design for electric vehicle users.