In power systems, generation must be maintained in constant equilibrium with consumption. A key indicator for this balance is the frequency of the power grid. The load frequency control (LFC) system is responsible for maintaining the frequency close to its nominal value and the power deviation of tie-lines at their scheduled levels. However, the remote communication system of LFC exposes it to several cyber threats. A successful cyberattack against LFC attempts to affect the field measurements that are transferred though its remote control loop. In this work, a data-driven, attack recovery method is proposed against denial of service and false data injection attacks, called DAR-LFC. For this purpose, a deep neural network is developed that generates estimations of the area control error (ACE) signal. When a cyberattack against the LFC occurs, the proposed estimator can temporarily compute and replace the affected ACE, mitigating the effects of the cyberattacks. The effectiveness and the scalability of the DAR-LFC is verified on a single and a two area LFC simulations in MATLAB/Simulink.
