Abstract: Considering the problem of complex topology in the rapid movement of vehicles in the network of vehicles, a dynamic network slicing resource deployment algorithm based on deep enhanced learning (DRL) was designed. In the communication scenario of vehicle to infrastructure (V2I), for the changing vehicle topology and business requests, the environment is modeling as an observed Markov decision model, and the software definition network (SDN) controller is used to monitor the network status in real time. The parameters are updated in real time according to the actions value of the distribution ratio of slicing resources, and the introduction of priority experience back and placement strategies to speed up the convergence speed and provide sufficient communication resources for each business request to interact the speed and location information of the vehicle. The results of the simulation experiments show that compared with other algorithms, the algorithm on the end-to-end throughput, end-to-end delay, slice package rate, and vehicle business request acceptance rate all show better performance.

Key words: network slicing, V2I communication, deep reinforcement learning, Markov decision process