关键词: 以太网无源光网络（EPON）, 动态带宽分配, 公平性, 人工神经网络, 微粒群

Abstract: Dynamic Bandwidth Allocation（DBA） is a key technology of EPON. The paper proposes an improved polling algorithm, adjusts the order of ONU authorization time slot according to the data traffic flow burst characteristics, constructs a neural network prediction model in Particle Swarm Optimization（PSO）, enhances the ONU added data prediction precision in the polling cycle, and thus guarantees bandwidth allocation fairness. The simulation proves that the algorithm performs better in the improvement of bandwidth utilization and has a smaller average packet delay compared with traditional DBA.

Key words: Ethernet Passive Optical Network（EPON）, Dynamic Bandwidth Allocation（DBA）, fairness, Artificial Neural Network（ANN）, Particle Swarm Optimization（PSO）