长程相关网络通信量的预测

Forecast of network traffic with long rang dependence

针对长程突发通信量提出了两种基于α-平稳信息的预测方法:根据α-平稳过程的协变概念,推导出双曲线偏差渐近意义下的FARIMA(fractionally autoregressive integrated moving average)预测,采用自回归神经网络模拟ARMA过程,并利用遗传算法的全局优化能力与人工免疫算法的多种群快速局部收敛能力对神经网络权值进行准确估计,从而实现对通信量的FARIMA预测.这两种预测方法均能在无限方差准则下实现偏差最小,合并这两种预测值以获得最后的预测结果.对实际踪迹的预测结果证实了两种独立的预测方法有效准确,最后的混合预测能进一步提高最后的预测精度.

Two distinctive predictors based on α-stable innovation are presented for the long range bursty traffic. The first FARIMA （fractionally autoregressive integrated moving average） predictor under the meaning of hyperbolic deviation asymptote is concision and computational quickness because of the injection of the covariation. The second FARIMA predictor in which autoregressive moving-average （ARMA） is simulated by recurrent neural network （RNN） was achieved. The estimation of the weights in RNN was finished exactly by genetic algorithm with the global optimization ability and artificial immune algorithm with the quick local convergence ability based on multi-population. The two predictors can minimize the dispersion according to the criteria with infinite variance. The final predicted values are obtained by combining the previous two individual predicted values. The predicted results of actual traces show that the two individual predictors are effective and accurate, what is more, the last compound predictors can enhance the final predicted precision.