SCAAE: Using Self-Supervised Contrastive Learning in Adversarial AutoEncoder for Anomaly Detection of Multivariate Time Series in Cyber Physics Systems

The network security protection of Cyber- Physical Systems (CPS), represented by power monitoring systems, is a significant research area. Anomaly detection of multivariate time series generated during CPS operation plays a crucial role in preventing industrial accidents and ensuring system security. However, the field of industrial multivariate time series anomaly detection still faces two main challenges: (1) the lack of more efficient representation methods for multivariate time series data; (2) the difficulty in acquiring a sufficient amount of anomaly data in practical industrial processes. This paper introduces a novel unsupervised anomaly detection method that combines the self-supervised learning mechanism with multivariate time series anomaly detection. The proposed method is named Self-Contrastive learning and Adversarial Autoencoder-based Multivariate Time Series Anomaly Detection (SCAAE). The autoencoder structure of SCAAE enables it to learn feature representations in an unsupervised manner. Moreover, the introduction of data augmentation techniques and contrastive learning during the training process allows the model to better utilize data characteristics, leading to improved performance even in the presence of data noise.Finally, the effectiveness and characteristics of the proposed method are validated through experimental studies on public datasets.

I. Introduction

With the development of information technologies such as control, communication, and computing, the traditional power system has extensively incorporated sensing, networking, and computing technologies. Gradually, it has evolved into a Cyber-Physical System (CPS) characterized by the deep integration of information and physical systems, enabling extensive interactions. The security status of the cyber-physical system critically relates to the secure operation of the power grid and electrical supply, establishing it as a vital and central component of industrial network security safeguarding.