Research on traffic flow generation model based on improved ACGAN and its evaluation system

Testing and verification of intelligent vehicles pose significant challenges in the field of intelligent driving. Given the complexity of current autonomous driving systems, it is necessary to expand the testing scenario library and test boundaries for intelligent vehicles. The data-driven scenario construction method enables the generation of a large number of previously unknown test scenarios and is considered the most effective approach. In response to this, this paper proposes a method for constructing a traffic flow data generation model. The method employs an improved implementation of the Auxiliary Classifier Generative Adversarial Network (ACGAN). The U-Net network serves as the generator’s network structure within the ACGAN, while the deep residual network (ResNet) serves as the discriminator’s network structure. Additionally, the model’s loss function has been enhanced. Furthermore, this paper introduces a method for evaluating the quality of the samples generated by the traffic flow generation model. The key elements of the evaluation method proposed in this paper are the consistency index and the diversity comparison method. The evaluation method confirms that the improved traffic flow generation model outperforms the original model.