Modeling and simulation of heterogeneous traffic flow in urban expressway merging areas

The issue of traffic congestion in urban expressway merging areas has garnered significant research attention. However, while improving traffic flow efficiency, ensuring the safety and comfort of driving is essential. This study leveraged deep reinforcement learning algorithms to simultaneously consider speed, safety and comfort, enhance the reward function and train neural network models to control multiple intelligent connected and automated vehicles (CAVs) in heterogeneous traffic flows. K-means and genetic algorithms were applied to calibrate the operational model of human-driven vehicles (HDVs) with a simulation environment built in Python to replicate heterogeneous traffic flows in urban expressway merging areas. The effectiveness of the proposed strategy was validated by comparing its speed, acceleration and lane change frequency. The results indicated that as the CAV penetration rate controlled by the proposed strategy increased, the average speed of heterogeneous traffic flow improved significantly and the frequency of lane changes decreased, creating a safer and more comfortable driving environment in urban expressway merging areas.