A Lightweight CNN-based Mechanism to Efficiently Identify Electric Meters under Stealthy False Data Injection Attacks in Smart Power Grid

In the era of the power grid digitization, the smart electric meters faces the stealthy false data injection attack (SFDI), which compromises the state estimation (SE) in the supervisory control and data acquisition system (SCADA). Especially, the state-of-the-art identification mechanism in the intrusion detection systems (IDSes) cannot meet the timely response demand, despite its high accuracy as it adopts a deep CNN to capture the injections via fully exploiting the wide-area measurement statistics. To solve the above challenge, we propose an efficient and accurate identification mechanism (EACI) designed in the following aspects. To improve the efficiency, EACI adopts a lightweight CNN with reduced neural layers to avoid the unnecessary exploitation. To maintain the accuracy, EACI extracts the major correlation features in the adjacent statistics, and employs the CNN to accurately discriminate the injections. We evaluate EACI under extensive SFDI scenarios, in both the low and medium voltages power grids constructed according to the IEEE 30-bus and 57-bus datasets, respectively. The numerical results demonstrate that EACI reduces the response time to hundreds of microseconds, while keeping a high detection accuracy.