Vehicle Lane-Changing scenario generation using time-series generative adversarial networks with an Adaptative parameter optimization strategy.

Connected and automated vehicles (CAVs) hold promise for enhancing transportation safety and efficiency. However, their large-scale deployment necessitates rigorous testing across diverse driving scenarios to ensure safety performance. In order to address two challenges of test scenario diversity and comprehensive evaluation, this study proposes a vehicle lane-changing scenario generation method based on a time-series generative adversarial network (TimeGAN) with an adaptive parameter optimization strategy (APOS). With just 13.3% of parameter combinations tested, we successfully trained a satisfactory TimeGAN and generate a substantial number of lane-changing scenarios. Then, the generated scenarios were evaluated for diversity, fidelity, and utility, demonstrating their effectiveness in capturing a wide range of driving situations. Furthermore, we employed a Lane-Changing Risk Index (LCRI) to identify the rare adversarial cases in scenarios. Compared to real scenarios, our approach generates 27 times more adversarial cases with 1.8 times higher average risk, highlighting its potential for uncovering critical safety vulnerabilities. This study paves the way for more comprehensive and effective CAV testing, ultimately contributing to safer and more reliable autonomous driving technologies.