A Novel Detection Algorithm for Anomalies and Attacks on Electric Vehicle Charging Stations Using Distributed Data Privacy Protection

Electric vehicles with charging stations are being rapidly deployed. Traditional methods for detecting anomalies and attacks on charging stations merely employ local data, resulting in lower detection accuracy. Intuitively, centralized federated learning approaches impose heavy computational loads on the central node. When the central node experiences abnormalities or failures, the charging station network may encounter data isolation. This paper starts from individual charging stations and utilizes a transformer for anomaly and attack detection. Based on this, a distributed privacy-preserving data aggregation scheme is proposed. More specifically, the transformer effectively eliminates the influence of abnormal data fluctuations on detection accuracy. The proposed scheme dynamically aggregates data through a distributed federated learning framework without requiring any prior knowledge or central node deployment. In addition, considering the statistical characteristics of the model parameters, an effective model parameter updating method is proposed to reduce the communication bandwidth requirement in distributed federation learning and enhance the algorithm’s robustness. According to the results, the suggested system may safely detect anomalous activity or possible threats at EV charging stations while safeguarding user privacy.