Cyber-Physical Data Fusion & Threat Detection with LSTM-Based Autoencoders in the Grid

The power grid, traditionally perceived as an independent physical network has undergone a significant transformation in recent years due to its integration with cyber communication networks and modern digital components. Cyber situations, including cyber-attacks and network anomalies, can directly affect the physical operation of the grid; therefore, studying this intricate relationship between the physical and cyber systems is pivotal for enhancing the resilience and security of modern power systems. In this digest, a novel Long Short-Term Memory (LSTM)-based Autoencoder (AE) model for cyber-physical data fusion and threat detection is proposed. The scenario under consideration includes the effective detection of a physical disturbance and a Denial-of-Service (DoS) attack, which obstructs control commands during the physical disturbance in the power grid. Detailed analysis and quantitative results regarding the LSTM-based AE model's training and evaluation phases is provided, which highlight its key operation features and benefits for guaranteeing security and resilience in the power grid.