Research Article
Research on DUAL-ADGAN Model for Anomaly Detection Method in Time-Series Data
Algorithm 1
DUAL-ADGAN training network model pseudocode.
| (1) | Function training network ; | | | Input: Time series data sliced by sliding window x, Noise vector z | | | Output: The trained WGAN generator , Fence-GAN discriminator , and predictor P | | (2) | If model is WGAN: | | (3) | For epochs do | | (4) | Feed the noise vector z into the generator to generate the data | | (5) | Feed the generated data and the real data x into the discriminator | | (6) | Training and with WGAN-GP loss function | | (7) | Return | | (8) | If model is Fence-GAN: | | (9) | For epochs do | | (10) | Feed the noise vector z into the generator to generate the data | | (11) | Feed the generated data and the real data x into the discriminator | | (12) | Training and with Fence-GAN loss function | | (13) | Return | | (14) | If model is Predictor: | | (15) | For epochs do | | (16) | Preprocess the training data x into that matches the RNN input | | (17) | Train the RNN network | | (18) | Return P | | (19) | end |
|