Dynamic Bandwidth Allocation Based on Ensemble Learning Predicition for Optical Access Networks

In order to make the existing Ethernet passive optical network can cope with the increasing demand of network traffic in the future and adapt to the low delay transmission characteristics required by new services in 5G environment, In this paper, we propose a DBA algorithm based on bandwidth prediction. Using Xgboost algorithm in ensemble learning to predict both the additional request bandwidth of ONUs during each polling period and the bandwidth that will be requested by ONUs in the next polling period. Based on the predicted bandwidth value, OLT can grant ONUs uplink transmission window in advance, which reduces the packet blocking time and control frame overhead caused by request granting mechanism in EPON system. Extensive simulation experiments show that the algorithm can improve the transmission delay and bandwidth utilization of EPON system.