中文摘要：

针对小时间尺度网络流量预测中的复杂性、非线性和高度自相似性等问题,提出使用一种改进模拟退火法优化的相关向量机（PSA-RVM）来解决网络流量预测问题。对网络流量时间序列进行相空间重构形成训练样本集,通过改进模拟退火法优化相关向量机的超参数,并构建网络流量预测模型。此外,通过实例进一步分析超参数对于相关向量机预测性能的影响。实例表明,PSA-RVM预测模型的预测精度、稳定性都优于RVM模型和PSO-SVR模型。

英文摘要：

This paper constructed an improved simulated annealing algorithm and relevance vector machine（ PSA-RVM） network traffic prediction model based on features of small time scale network traffic（ NT） such as the complexity,nonlinearity and highly self-similarity. Meanwhile,it reconstructed NT time sequences into a new training sample set by means of an improved simulated annealing algorithm to optimize the ultra parameters of RVM and built the prediction model. In addition,it gave an analysis of the influence of ultra parameters on the performance of this model. Experiments show that PSA-RVM prediction model is of better prediction accuracy and stability than RVM and PSO-SVR models.