S-KDGAN: Series-Knowledge Distillation With GANs for Anomaly Detection of Sensor Time-Series Data in Smart IoT

Nowadays, smart Internet of Things (IoT) technology has emerged as a new paradigm, widely utilized across various fields of our lives. Sensors in smart IoT systems generate massive amounts of high-dimensional time-series data, where anomalies often occur. Detecting these anomalies is crucial for ensuring the stability, security, and reliability of IoT systems. However, traditional anomaly detection methods typically involve transmitting data to the cloud or a centralized server for processing, which can increase latency and make them unsuitable for real-time applications. Conversely, deploying anomaly detection directly on edge devices enhances detection speed and efficiency. To address this, we propose a lightweight time-series anomaly detection model that combines attention anomaly detection generative adversarial networks (Att-ADGANs) and knowledge distillation (KD). Initially, we train a generative adversarial network (GAN) with encoder and decoder architectures to capture the features of normal data distributions. Subsequently, the GAN is compressed and distilled into a lightweight model using KD. Furthermore, we introduce a series-knowledge distillation GAN (S-KDGAN) framework, which leverages progressive distillation to enhance distillation efficiency and minimize performance degradation. The experimental results demonstrate that S-KDGAN achieves compression ratios of 2.37:1, 48.85:1, and 15.93:1 on the SWMRU, KDDCUP99, and HomeC datasets, respectively, with performance degradation rates of only 1.38%, 2.27%, and 1.61%.