Research on intelligent stochastic traffic flow model based on deep reinforcement learning

In order to quickly form richer simulation scenarios, this paper proposes a method for training a stochastic traffic flow model based on deep reinforcement learning DQN (DeepQ-Network), which can integrate the generated model into a simulation loop for testing intelligent driving algorithms. Firstly, a training environment was constructed based on a two-degree-of-freedom dynamic model, a random rigid object moving target, and a three-lane road. Then, through different reward function designs, three intelligent stochastic traffic models, namely cautious, balanced, and aggressive, were trained. Finally, multiple types of traffic agents were integrated in the CarMaker simulation software to form a stochastic traffic flow, which verified the interactivity and effectiveness of the model. The results show that the model trained using deep reinforcement learning algorithm can effectively generate random scenarios quickly, greatly improving the coverage of simulation test scenarios.