Computer Science · Computer Networks and Communications
CLSDRL:A routing optimization method for traffic feature extraction
Hong Xia, Jianguo Li, Yanping Chen, Ning Lv, Zhongmin Wang, Qingyi Dong
See on the mapAbstract
In order to solve the impact of the temporal and spatial characteristics of traffic on network routing optimization, this paper proposes convolution long-short memory neural network deep reinforcement learning (CLSDRL) model for routing optimization. The CLSDRL model consists of deep deterministic policy gradients (DDPG) deep couple with convolution neural network (CNN) and long-short memory neural network (LSTM). After extracting the spatial and temporal characteristics of network traffic with CNN and LSTM, routing decisions are made with DDPG algorithm. Experiments are conducted under different load intensities, and the network performance is evaluated by the average network delay and packet loss rate, experimental results show that this method can improve significantly network performance.