Anomaly Detection Method Based on Gaussian Mixture Model and Orthogonal Generative Adversarial Network

In the analysis of power grid data, a critical task is to identify samples that significantly deviate from others, which constitutes anomaly detection. Presently, anomaly detection methods based on deep generative models face challenges related to complex network structures and difficult training processes. Addressing these issues, this paper proposes a lightweight anomaly detection method for power grid data based on Gaussian Mixture Model and Orthogonal Generative Adversarial Network. This approach combines traditional statistical techniques with advanced deep learning methods. In comparison to anomaly detection methods relying solely on deep generative models, this method features a more concise network structure, reducing both the model's complexity and training time, while retaining a certain level of expressive power.