基于SA-WGAN的网络流量异常检测方法

Network traffic anomaly detection method based on SA-WGAN

针对网络流量数据的海量、复杂、多维、不平衡、低价值密度等特点,提出了一种引入了自注意力机制的WGAN异常检测方法。该方法将轻量化的自注意力机制嵌入到WGAN中,充分挖掘了流量数据中的潜在关联性,利用生成误差和重构误差评估了综合异常得分,再利用自适应窗口技术进行异常初判和异常裁剪。实验结果表明:该方法在精确率、召回率和F1值等指标的检测性能上,较传统的生成式异常检测方法有明显提升。

In view of the characteristics of massive,complex,multi-dimensional,unbalanced and low value density of network traffic data,a WGAN anomaly detection method based on self-attention mechanism was proposed.Lightweight self-attention mechanism was embedded into WGAN,by fully exploit the potential correlation in traffic data and comprehensively using the anomaly scores of generation error and reconstruction error,the abnormal score was evaluated.Adaptive window technology was used for anomaly judgment and anomaly clipping.Experimental results show that the proposed method is obviously superior to the traditional generative anomaly detection method in terms of precision,recall and F1-measure.