REINFORCEMENT LEARNING METHOD FOR DYNAMIC ALLOCATION OF PRIORITIES IN TSN

Abstract

This article explores the possibility of using reinforcement learning to dynamically assign priorities in time-sensitive networks. The presented approach purposefully optimizes the process of managing time constraints in the network. Using reinforcement learning techniques, the system independently adjusts priorities depending on the requirements of network traffic. To achieve this goal, two configu ration schemes based on TSN standards are proposed: centralized and distributed. Having considered these schemes, we will identify their limitations necessary in meeting requirements close to real time and ensuring strict quality of service guarantees, taking into account the restrictions applied to a time-sensitive environment. The work also reveals the need to use additional equipment, a centralized controller, to reallocate priorities.