Enhancing Cyber-Resilience of Power Systems' AGC Sensor Data by Time Series to Image Domain Encoding

The shift to wide area networks (WANs) in power systems has introduced security challenges. While research has focused on attack detection, adversaries can outsmart even advanced anomaly detectors. Further, existing techniques fail to ensure data confidentiality, exposing telemetered data to eavesdropping. To address these evolving threats, we propose a novel framework to enhance power system data communication security over WANs. This innovative approach combines time series to image transformation, digital watermarking, and machine learning. These elements collaboratively encode sensor data into an unintelligible format, maintaining operational impact. We begin by converting time-series data into images with the Gramian Angular Field algorithm. These images are subtly embedded with invisible watermarks. The framework introduces an additional security layer to detect image tampering during communication. Using an Autoencoder-Support Vector Regression model, we restore the watermarked images to the original time series. Notably, the watermarking method is system-independent, maintains sensor data confidentiality, and certifies data integrity. In time-critical power system operations, traditional encryption and authentication can introduce latency. The proposed approach minimizes computational overhead and ensures real-time responsiveness, effectively securing power system data communications. To demonstrate the proposed framework's effectiveness, experiments were conducted on the IEEE-39 bus system's Automatic Generation Control system.