Weight Optimization Algorithm for Multi-Link State Information in Intelligent Routing

Aiming at the problems of existing routing algorithms combining software-defined networks and reinforcement learning that use only limited link state information and the existence of subjective weight assignments, a weight optimization algorithm for multi-link state information in intelligent routing (NPIR) is proposed. The algorithm comprehensively considers three link state information, namely delay, bandwidth and packet loss rate, and uses normal distribution normalisation and optimisation weights for processing, which can make the routing algorithm more accurately reflect the actual condition of the network, thus obtaining a better routing strategy among the switches. Through simulation experiments, it is verified that the NPIR algorithm reduces the average value of delay on all selected paths by $\mathbf{5 5. 2 \%}$ and 9.2 %, and the average value of packet loss rate on all selected paths by 44.8 % and 12.6 %, respectively, compared with the reinforcement learning-based Widest Path Routing (Q-WPRA) algorithm and the reinforcement learning-based Intelligent Routing for Software-Defined Networks (RSIR) algorithm.