Anomaly Detection for Charging Piles Based on Conditional Variational Autoencoder

The dependability and robustness of electric vehicle (EV) charging infrastructure, which functions as a pivotal nexus between consumers and the electrical network, play a crucial role in enhancing the efficiency of EV usage and ensuring the safe management of the power grid. Therefore, it is highly urgent to develop an effective and robust anomaly detection system to provide early warnings of potential risks and address the issue of imbalanced data distribution. In this paper, a conditional variational autoencoder (CVAE) is employed to construct an anomaly detection model for charging pile data. In contrast to other anomaly detection methodologies, the method proposed in this study demonstrates more desirable performance. Furthermore, this paper extends the investigation by modifying the architecture of the CVAE to facilitate supervised learning. The reconfigured network structure yields enhanced detection accuracy, obtaining better anomaly detection performance when evaluated on the charging pile dataset.