Abstract: With the increase of the scale of Internet, network traffic prediction on Overlay has become a research focus gradually. Compared with traditional networks, the features of Overlay network mean that traditional predictions are out of its demand. It proposes a new method that is based on neural network using particle swarm-based simulated annealing, and applies reverse calculation, starts from the ideal optimal value and finds the optimal solution through the shortest path. This method increases the probability of success to find the optimal solution and cut off the running time. Through the simulation it deduces that the proposed method is better than the traditional one obviously.

Key words: Overlay network, network traffic, prediction, Back Propagation（BP） neural network

摘要： 随着网络规模的增长，Overlay网络流量预测已经日渐成为研究热点。与传统网络相比，Overlay网络本身的特性决定了传统的预测方法已不能适应它的要求。提出一种基于模拟退火的粒子群神经网络来预测Overlay网络的流量，运用反向计算方法，从理想最优值出发，近距离寻找最优解，缩短了求解时间并加大了找到最优解的几率。通过实验仿真可以看出，改进的BP神经网络方法的预测效果要明显好于传统的BP神经网络。

关键词: 覆盖网络, 网络流量, 预测, 反向传播（BP）神经网络