Search

According to the description of 50 Gbit/s-Ethernet Passive Optical Network (EPON) in IEEE 802.3ca new standard, multi-wavelength dynamic bandwidth allocation is needed in network uplink. In this paper, long short-term memory neural network is used to predict the network traffic in order to improve the efficiency of bandwidth allocation. At the same time, considering the different requirements of network delay and bandwidth for different types of services in the network, different allocation schemes are adopted for different priority services. The simulation results show that the neural network prediction method can effectively reduce the delay of the network. When different allocation schemes are adopted for different services, the latency of high-priority services decreases, and the latency varies with the increase of network load. The algorithm proposed in this paper meets the requirements of high-priority service. Although the latency of low-priority service is increased compared with the previous method, it has little impact on the overall service quality of the network considering that low-priority service does not have high latency requirements.