Kalman reinforcement learning-based provably secured smart grid false data intrusion detection and resilience enhancement

Smart grid intrusion is now increasing due to increased cyberattacks on intelligent devices. Cyberthreats like false data injection attack (FDIA) can bypass conventional security mechanisms. To defend against smart grid intrusion, in this research work, recurrent neural network with a Kalman filter is proposed to detect smart grid fault, normal, and FDIA events for a multi-sourced smart grid system. By using the stacking method, a novel parallel reinforcement learning with adaptive feature boosting is utilized to extract deterministic features. In the proposed feature extraction process, firstly Kalman filters are used to reduce feature dimension. Secondly, the resilient defence was constructed to improve the stable operation of the smart grid. The performance of the proposed Kalman filter reinforced neural network (KFRNN) is demonstrated by the presence of deterministic critical features under FDIA and without FDIA on a smart grid multi-sources data. The proposed KFRNN is evaluated by standard WUSTIL-2021 and real-time hardware-in-loop (HIL) test bed case study with FDIA. The obtained result shows that the proposed KFRNN provides resilient operation for smart grid by achieving a high classification accuracy of 97.3%, increased F1-score, increased receiver operating characteristic, and high detection probability than conventional schemes. Finally, a comprehensive simulation is performed on the IEEE 118 bus New England System to validate the effectiveness of the proposed KFRNN. From the obtained performance indexes, it is observed that the proposed intrusion detection scheme has high accuracy with enhanced resilient operation.