CNN-Driven Real-time Intrusion Detection and Mitigation Scheme for Solid-State Power Substation

This paper proposes a resilient control scheme for a Solid-State Power Transformer (SST) based power substation (SSPS). The proposed control scheme consists of a centrally located secondary aggregator and primary controller that govern the operation of corresponding SST cells. A cross-correlation-based Intrusion Detection (CC-IDS) mechanism estimates the cross-correlation coefficient between the global voltage feedback and a reference signal. As the estimated cross-correlation drops below the preset threshold, an intrusion on the SST’s global voltage feedback is detected, and a Convolutional Neural Network (CNN) enabled attack mitigation mechanism is triggered. The CNN estimator utilizes the SSPS’s attacked global input voltage feedback to predict the desired global input voltage. Estimated global input reference is then substituted with noisy global input voltage feedback in the SST cell’s primary controllers to mitigate the impacts of the intrusion. To ensure high prediction accuracy, the CNN estimator is trained using datasets encompassing various side-channel noise intrusion (SNI) attack scenarios. Several scenarios are emulated in MATLAB Simulink environment to validate the CNN intrusion mitigation approach under the influence of different attack vectors.