基于合理遗忘选择训练样本的网络流量预测模型

NETWORK TRAFFIC PREDICTION MODEL BASED ON SELECTING TRAINING SAMPLES WITH REASONABLE FORGETTING

为了提高网络流量的预测精度,针对极限学习机的训练样本选择问题,提出一种基于合理遗忘选择训练样本的网络流量预测模型(SF-ELM)。首先通过引入遗忘因子减弱旧训练样本对预测结果的影响,合理对训练样本进行在线更新;然后以泛化能力作为评价准则,选择性更新极限学习机输出权值;最后进行仿真分析。结果表明,SF-ELM的网络流量学习速度快于对比模型,获得了更加理想的网络流量预测效果,更适于实时性要求高的网络流量在线预测。

In order to improve prediction accuracy of traffic network, we propose a novel network traffic prediction model to solve the problem of training samples selection in extreme learning machine, it is based on selecting training samples with reasonable forgetting. First, we weaken the influence of old training samples on prediction results by introducing forgetting factor, and reasonably make online update on training samples; then we take the generalisation performance as the evaluation criterion to selectively update the output weight of extreme learning machine; finally, we carry out simulation analysis. Results show that the proposed model has faster network traffic learning speed than the contrast models, acquires more reasonable network traffic prediction effect, and is more suitable for online network traffic prediction with higher real-time requirement.