Spatial-Temporal Correlation-Concerned Measurement Manipulation Detection Based on Gramian Angular Summation Field and Convolutional Neural Networks

The growing integration of information and communication technology exposes more exploitable cyber-attack vectors to electric power systems. Measurement manipulation is an impactive type of cyber-attacks that can be launched to cause massive damages by interfering decision-making of control centers. Existing data-driven measurement manipulation detection models often bear criticisms due to the deficient mining of spatial-temporal correlations. In this paper, based on the Gramian Angular Summation Field (GASF) algorithm, various types of phasor time-series measured at different locations are transformed into a series of generalized multichannel images. From these generalized images, Convolutional Neural Networks (CNNs) are trained to discover suspicious patterns. GASF can visualize the temporal features of a single time-series and facilitate collaborative CNNs to figure out spatial-temporal correlations between different measurements. The improvements of the proposed framework in the detection performance are demonstrated through case studies.