|
چکیده
|
Amid the swift advancements in smart power systems, the traffic data within power system servers exhibit pronounced complexity and multimodality, rendering traditional anomaly detection methods insufficient for these sophisticated challenges. To mitigate these challenges, this study introduces a G-VAE framework, which synergistically integrates the Gaussian Mixture Model (GMM) with a Variational Autoencoder (VAE) for anomaly detection in power system server traffic. The framework discerns potential normal and anomalous patterns by employing the GMM to cluster and analyze the traffic data, while the VAE conducts latent space modeling and feature reconstruction to achieve precise anomaly detection in complex data sets. The GMM adeptly captures the multimodal distribution characteristics of the data, while the VAE enhances the model's representational and generalization capabilities in detecting anomalies. In empirical tests utilizing a real-world power system server traffic dataset, the G-VAE framework demonstrates superior performance in metrics such as accuracy and false alarm rate, significantly surpassing traditional machine learning and deep learning methodologies. The experimental outcomes reveal that G-VAE not only enhances the accuracy and resilience of anomalous traffic detection but also paves a novel technical pathway for power system security monitoring. These research findings offer robust support for the security of intelligent power systems, provide a scientific foundation for power operation management and decision-making, and foster the advancement of power system security technologies.
|