Unsupervised anomaly detection algorithm for sensor data streams based on GAT-GAN

Detecting data flow anomalies within Wireless Sensor Networks (WSNs) and reaching the server can promptly identify faults and issue maintenance alerts, thereby ensuring safe and reliable operation of the system. However existing methods can utilize single variables in data streams to perform most anomaly detection tasks, they often ignore the correlation between variables, resulting in reduced detection performance. Therefore, our approach constructs a generative adversarial network model based on Graph Attention Network (GAT) and employs a dual autoencoding structure in both the generator and discriminator to learn the latent representation of the entire time series. The results of ablation experiments and comparison experiments on two real datasets with baseline models show that GAT-GAN outperforms the former in detecting data anomalies