Abstract: Intelligent and connected vehicles driving in platoon can significantly improve road capacity,but it also causes new road traffic safety risks.Aiming at the collision risk caused by random vehicular lane-change and cut-in,this paper proposes a novel distributed control method for the mixed vehicular platoon based on deep reinforcement learning.Firstly,according to the characteristics of vehicle lane-change and cut-in scenarios,the deep reinforcement learning reward and agent network structure were designed with multiple control objectives,and a hierarchical control model for mixed vehicular platoon was designed.Secondly,considering the evaluation indexes of road safety,the stability of the control system was analyzed by using the complex frequency domain method,and the final safety constraints for vehicular platoon can be obtained.Finally,compared with traditional linear control and model predictive control method based on Prescan/Simulink co-simulation experiments,it was found that under the proposed method in the same scenarios,the spacing descend has been reduced by 56.9% and 52.1%,the speed descend has been reduced by 58.72% and 51.19%,the maximum standard deviation of speed has been reduced by 54.8% and 49%,and the minimum time-to-collision has been improved by 420.7% and 265%.It is demonstrated that the proposed method applied in mixed vehicular platoon control can significantly improve road safety.