Research on substation intrusion event identification method based on MTF and CNN

Abstract

This research provides a detection approach based on Markov transition field (MTF) and convolutional neural network (CNN) for substation perimeter intrusion event recognition. Because of the complexity and variety of external signals, which makes sensor detection more challenging, determining and analyzing the intrusion behavior of vibration signals induced by intrusion has become critical to improving the identification rate of intrusion-like occurrences. The obtained one-dimensional signals are mapped into two-dimensional pictures using MTF to reflect better the properties of intrusion-like signals, which can yield deeper signal details than the usual feature extraction approach. The CNN-vgg19 model’s excellent image feature identification capabilities are utilized to detect and categorize the acquired 2D feature pictures. The experimental results show that the average recognition rate of the six intrusion events is 96.7%, and the average recognition rate of the noise events is 99%, which can effectively identify the noise events and reduce the false-positive rate. The study’s findings are valuable for substation and peripheral security intrusion products.