With the rapid adoption of electric cars globally, particularly in regions like China where the China EV market is expanding, new safety challenges have emerged. The quiet operation of electric cars at low speeds increases the risk of pedestrian collisions, as traditional auditory cues from internal combustion engines are absent. This study addresses this issue by proposing a collaborative optimization framework that integrates the Acoustic Vehicle Alerting System (AVAS) with passive protection devices. Through multi-objective optimization methods, we aim to harmonize active warning and passive impact mitigation functions, thereby enhancing road safety for pedestrians. The research establishes a human-vehicle-road collaboration model, optimizes system response strategies and structural designs, and validates the approach via simulations and real-world tests. Our findings demonstrate that this synergistic approach reduces pedestrian injury risks while minimizing false alarm rates, offering a theoretical foundation for designing intelligent safety systems in electric cars.
The proliferation of electric cars, especially in the China EV sector, has highlighted inherent safety concerns due to their silent operation. Unlike conventional vehicles, electric cars produce minimal noise at low speeds, leading to a higher incidence of pedestrian accidents. Existing safety systems often focus on isolated improvements in either active or passive components, resulting in response conflicts or redundant protections. In this paper, we introduce a dynamic coupling mechanism between AVAS and passive protection devices, leveraging multi-sensor data fusion and adaptive control strategies. By addressing complex traffic scenarios, our work provides a novel paradigm for intelligent safety technology in electric cars, with potential applications across the evolving China EV landscape.
The core of our methodology lies in the dynamic coupling mechanism between AVAS and passive protection systems. For the acoustic warning signals to be effective, they must precisely align with collision timelines to avoid premature or delayed triggers. We developed a matching model that incorporates dynamic time-to-collision (TTC) prediction based on vehicle speed, acceleration, and pedestrian relative position. For instance, when an electric car moves at 10 km/h, the warning activates within 3–5 meters of a pedestrian, allowing 1–2 seconds for reaction. Machine learning models, such as Long Short-Term Memory (LSTM) neural networks, are employed to predict pedestrian motion intent and adjust warning thresholds dynamically. Acoustic signal parameters are optimized to fall within the human-sensitive frequency range of 200–5000 Hz, with sound pressure levels adhering to regulations like EU UN R138. Environmental noise sensors enable real-time adjustments, ensuring perceptibility in 60–80 dB noise environments. The timing strategy ensures that if a pedestrian does not evade and TTC falls below a threshold, passive devices activate immediately, with timing errors controlled within ±50 ms. This is critical for electric cars operating in urban China EV environments, where unpredictable pedestrian behavior is common.
To enhance passive protection, we integrated a Bayesian network model for predicting pedestrian trajectories. Using a fusion of YOLOv7 and LSTM algorithms, the system forecasts pedestrian movements up to 1 second ahead with a position error of ≤0.3 meters. The collision probability \( P_c \) is calculated as:
$$ P_c = \phi \left( \frac{d_{\text{min}}}{\sigma_d} \right) $$
where \( d_{\text{min}} \) is the minimum distance and \( \sigma_d \) represents prediction uncertainty. If \( P_c > 0.8 \), pre-tensioning of seatbelts and reduction of brake hydraulic delays are initiated. The triggering threshold for passive devices, such as active hood systems, is dynamically adjusted based on collision energy \( E_k \):
$$ E_k = \frac{1}{2} m_p v_{\text{impact}}^2 $$
When \( E_k > 150 \, \text{J} \), a two-stage buffer mechanism is activated: the first stage uses shape memory alloys (SMA) to rapidly lift the hood by 50 mm, and the second stage employs aluminum honeycomb structures to absorb energy, reducing the Head Injury Criterion (HIC) below 800. This adaptive approach accounts for varying pedestrian postures, such as children (height < 1.2 m) and adults (height ≥ 1.5 m), by using millimeter-wave radar to identify key skeletal points and adjust hood lift heights (30–70 mm) and bumper crush rates, ensuring compliance with Wrap Around Distance (WAD1000) standards. This is particularly relevant for electric cars in dense China EV urban settings, where diverse pedestrian demographics exist.
Multi-sensor information fusion is pivotal for collaborative decision-making. Our architecture integrates heterogeneous sensors—cameras (30 fps), millimeter-wave radar (100 Hz), and lidar (10 Hz)—using an extended Kalman filter (EKF) for synchronization, with timestamp errors < 10 ms. Spatial registration employs the Iterative Closest Point (ICP) algorithm, achieving point cloud matching accuracy of ±2 cm. To resolve conflicts in sensor data, we apply Dempster-Shafer (D-S) evidence theory, assigning credibility weights: lidar (0.6), millimeter-wave radar (0.3), and camera (0.1). The combination rule is defined as:
$$ m(A) = \frac{\sum_{B \cap C = A} m_1(B) m_2(C)}{1 – K} $$
where \( K \) is the conflict factor; if \( K < 0.2 \), the fused result is output; otherwise, redundant sensors are activated. A three-level response mechanism is implemented: Level 1 (\( P_c < 0.3 \)) involves AVAS operation at 55 dB; Level 2 (\( 0.3 \leq P_c < 0.6 \)) enhances AVAS to 65 dB and pre-pressurizes the brake system; and Level 3 (\( P_c \geq 0.6 \)) triggers passive protection and automatic emergency braking (AEB). This is managed via ROS2 middleware, ensuring decision cycles ≤ 20 ms and meeting ASIL-B functional safety requirements, which is essential for the reliability of electric cars in the China EV market.
In the multi-objective parameter design, we focus on Pareto frontier analysis to balance warning intensity and bumper stiffness. These parameters exhibit a trade-off: excessive warning strength may cause pedestrian panic, while insufficient warnings fail to alert; similarly, high bumper stiffness increases impact force, whereas low stiffness reduces energy absorption. Through simulation experiments, we established an optimization model targeting pedestrian injury metrics (e.g., HIC) and false alarm rates. Using genetic algorithms, we derived a set of non-dominated solutions, as summarized in Table 1, which illustrates the Pareto optimal combinations for electric cars. For example, when warning intensity is low, bumper stiffness is increased to maintain protection, and vice versa. This approach allows designers to select parameters that suit specific constraints in China EV applications, ensuring both safety and efficiency.
| Warning Intensity (dB) | Bumper Stiffness (MPa) | HIC Value | False Alarm Rate (%) |
|---|---|---|---|
| 55 | 150 | 850 | 12 |
| 60 | 120 | 780 | 9 |
| 65 | 100 | 720 | 7 |
| 70 | 80 | 690 | 5 |
To address uncertainties in pedestrian posture, we developed a robust optimization model based on a Gaussian Mixture Model (GMM). This models key motion parameters—height \( h \), speed \( v_p \), and relative angle \( \theta \)—as a combination of Gaussian distributions. The probability density function is:
$$ p(x) = \sum_{k=1}^{K} \pi_k \mathcal{N}(x | \mu_k, \Sigma_k) $$
where \( x = [h, v_p, \theta]^T \) is the pedestrian state vector, \( K \) is the number of mixture components, \( \pi_k \) is the weight, and \( \mu_k \) and \( \Sigma_k \) are the mean and covariance. The robust optimization problem is formulated as minimizing the worst-case scenario:
$$ \min_{x} \max_{\theta \in \Theta} f(x, \theta) $$
where \( \theta \) represents uncertainty parameters, such as collision angles within ±15°. We solved this using stochastic gradient descent (SGD), ensuring HIC < 950 with 80% confidence. An adaptive robust control strategy employs a state observer and parameter adaptation law. For instance, bumper stiffness \( K_{\text{bumper}} \) is adjusted in real-time based on sensor data:
$$ K_{\text{bumper}} = K_0 (1 + \alpha \Pi_{\text{adult}} \Pi_{\text{upright}}) $$
where \( K_0 \) is the baseline stiffness, \( \alpha = 0.1 \) is an adjustment coefficient, and \( \Pi_{\text{adult}} \) and \( \Pi_{\text{upright}} \) are indicator functions for pedestrian state. This enhances responsiveness in variable China EV environments, where pedestrian behavior can be unpredictable.
Lightweight constraints are integrated through material-structure协同设计准则. We evaluated materials like aluminum alloys and carbon fiber composites for strength, stiffness, and energy absorption. Topology and shape optimization methods were applied to structural designs, ensuring weight reduction without compromising safety. For example, energy-absorbing regions use high-performance materials, while non-critical areas employ lighter alternatives. This is crucial for electric cars, as lightweighting improves energy efficiency and range—a key selling point in the China EV market. The协同设计准则 ensure that passive protection devices, such as bumpers and hoods, maintain optimal performance while adhering to weight limits.
System efficacy was validated through complex scenarios based on Euro NCAP standards. We constructed a test matrix encompassing four scenarios—adult frontal collision, child side collision, low-light nighttime conditions, and complex acoustic environments—with 12 evaluation metrics. Using VTD/PreScan co-simulation platforms, 1,000 Monte Carlo experiments confirmed system robustness. Key results are shown in Table 2, where the collaborative system significantly outperforms independent systems. For instance, in standard adult collision tests at 40 km/h, dynamic adjustments to hood lift height (70 mm) and bumper crush (30 mm) reduced HIC15 by 23.2%. For child targets, skeletal recognition algorithms improved WAD1000 compliance by 15.8 percentage points. In noisy environments (85 dB background), the adaptive AVAS maintained an 83.7% effective warning rate. These outcomes underscore the viability of our approach for electric cars, especially in China EV contexts with diverse operational conditions.
| Test Scenario | Evaluation Metric | Baseline (Independent System) | Optimized (Collaborative System) | Weight Coefficient |
|---|---|---|---|---|
| Adult Frontal Collision | HIC15 | 823 ± 32 | 632 ± 28** | 0.25 |
| Child Side Collision | WAD1000 Compliance Rate (%) | 76.5 | 92.3** | 0.20 |
| Nighttime Low Light | Warning Recognition Delay (ms) | 420 | 285** | 0.15 |
| Complex Acoustic Environment | Effective Warning Rate (%) | 68.2 | 83.7** | 0.10 |
| Emergency Lane Change | AEB Trigger Accuracy (%) | 79.1 | 91.4** | 0.30 |
Note: ** indicates p < 0.01 for significant differences.
Extreme工况下的失效模式与容错机制 were analyzed using fault tree analysis (FTA). Key failure modes include sensor malfunctions and actuator delays. For sensor failures, Kalman filter-based compensation algorithms limit positioning errors to ±0.5 m; if multiple sensors fail, V2X roadside unit coordination is activated, extending response time to 120 ms. For hydraulic system leaks, a backup electronic mechanical brake (EMB) system engages within 150 ms. In heavy rain (>50 mm/h), where millimeter-wave radar signal-to-noise ratio drops by 40%, optical flow features from cameras maintain over 80% pedestrian detection rates. Table 3 summarizes test results under extreme conditions, demonstrating the system’s resilience for electric cars in challenging China EV operational environments.
| Failure Type | Occurrence Probability (/1000 h) | System Degradation Level | Recovery Time (ms) |
|---|---|---|---|
| Camera Failure | 2.3 | Level 1 → Level 2 | 200 |
| Radar False Alarm | 1.7 | Warning Delay 15% | 180 |
| Hydraulic System Leak | 0.5 | Braking Distance +1.2 m | 350 |
Human-machine interaction (HMI) significantly influences system efficacy. Through driving simulator experiments, we found that a 67 dB sound pressure level with intermittent warning tones (200 ms on/100 ms off) in the 500–2000 Hz range reduced driver braking reaction time to 1.2 seconds, 0.3 seconds faster than continuous tones. Multi-modal feedback—combining tactile vibrations (50 Hz on the steering wheel) with visual heads-up display (HUD) prompts—lowered misoperation rates by 17%. User studies revealed that drivers expect warning lead times of 1.5–2.5 seconds, earlier than regulatory requirements, and tolerance for false alarms drops sharply beyond 5 per 100 km, causing a 37% decline in satisfaction. This highlights the need for HMI designs that align with cognitive models, ensuring that electric car safety systems in the China EV market are both effective and user-acceptable.
In conclusion, our collaborative optimization framework for electric cars successfully integrates AVAS and passive protection devices through multi-objective methods, dynamic coupling, and robust design. Simulations and real-world tests confirm reductions in pedestrian injury risks and false alarms, validating the approach for intelligent safety systems. The proposed models—including timing匹配, posture adaptation, and sensor fusion—offer a new范式 for addressing the silent operation challenges of electric cars. Future work could incorporate V2X technology to enhance responses in complex scenarios, further advancing the safety of electric cars, particularly in the rapidly growing China EV industry. This research contributes to the ongoing evolution of automotive safety, ensuring that electric cars not only reduce environmental impact but also prioritize pedestrian protection.