基于生成式对抗网络的自监督多元时间序列异常检测方法
周业瀚, 沈子钰, 周清, 李云

Self⁃supervised multivariate time series anomaly detection based on GAN
Yehan Zhou, Ziyu Shen, Qing Zhou, Yun Li
表4 CPCGAN与其他五种对比方法异常段检测的评价指标情况
Table 4 Performance of CPCGAN and five baseline approaches on anomalous segment detection
SWaT (segment)WADI (segment)
PRF1PRF1
CPCGAN0.81420.80100.80750.76910.78270.7758
AE0.75130.73340.74220.54230.57370.5575
MAD⁃GAN0.72250.68660.70400.60220.67140.6349
LSTM⁃VAE0.74680.79180.76860.76210.70010.7297
DAGMM0.62210.75100.68050.63660.87820.7381
TadGAN0.73920.85810.79420.77820.70750.7411