基于时空图注意力网络的多变量时序数据异常检测方法

肖刚,卢大鹏,郑文博,程振波,张元鸣

Multivariable time series data anomaly detection method based on spatiotemporal graph attention network

Gang XIAO,Dapeng LU,Wenbo ZHENG,Zhenbo CHENG,Yuanming ZHANG

表 2 不同模型在4个数据集中的性能评价指标

Tab.2 Performance evaluation metrics of different models in four datasets %

模型 SWaT MSL SMD PSM P R F1 P R F1 P R F1 P R F1 DAGMM 89.92 57.84 70.40 89.60 63.93 74.62 67.30 49.89 57.30 93.49 70.03 80.08 LSTM 86.15 83.27 84.69 85.45 82.50 83.95 78.55 85.28 81.78 76.93 89.64 82.80 LSTM-VAE 76.00 89.50 82.20 73.71 88.54 80.44 75.76 90.07 82.30 73.62 89.92 80.96 BeatGAN 64.01 87.46 73.92 89.75 85.42 87.52 72.90 84.71 79.48 90.30 93.84 92.04 OminiAnomoly 81.42 84.30 82.83 89.75 85.40 87.54 83.68 86.82 85.22 88.39 74.46 80.83 THOC 83.93 86.36 85.13 86.45 88.92 87.66 79.76 90.95 84.99 88.14 90.99 89.54 MTAD-GAT 81.42 84.30 82.83 87.54 84.42 85.95 90.19 83.25 86.58 72.39 77.46 74.83 GDN 99.35 68.12 80.82 78.45 63.07 69.92 74.02 61.43 67.09 92.32 85.82 87.66 TimesNet 86.76 97.32 91.74 83.92 86.42 85.15 88.66 83.14 85.81 98.19 96.76 97.47 DLinear 81.51 95.63 87.96 86.44 88.12 87.34 87.62 83.81 85.72 97.28 95.46 96.32 STGAT 90.99 95.37 93.13 89.74 86.85 88.06 89.95 85.04 87.42 98.49 96.62 97.54