Deep Reinforcement Learning Based Defense System for Electric Vehicle Charging Stations

The widespread adoption of Electric Vehicles (EVs) requires an efficient charging infrastructure. Using data connection-based charging equipment, smart charging, and vehicle-to-grid (V2G) charging technologies allow electric vehicles to connect to the electrical grid. This enables the exchange of information and instructions. However, this ecosystem is susceptible to physical or cyberattacks, just like any other cyber-physical system. In this paper, we investigate load-altering (LA) attacks that impact the functionality of the smart grid. We propose a two-phase strategy to avoid, detect, and mitigate attacks. Our first step is scheduling charging station operations to normalize system utilization to avert future simultaneous attacks. Therefore, a distributed Deep Reinforcement Learning (DRL) model is used to establish the ON/OFF status of each charging station. To maintain the system's power stability during the charging process, the system uses an event-based method to respond to the abrupt change in charging/discharging behavior. To detect attacks, a second multi-agent deep-reinforcement learning model is created. This enables the system to recognize and neutralize the effects of LA attacks on the power grid. After the compromised entities are located, they are isolated and the demands of certain backup charging stations either completely or partially replace the canceled power demands. The performance study shows that the suggested strategy minimizes the impact of LA attacks and delivers good results in terms of detection accuracy.