We classify Radio frequency (RF) signals recorded in a high-voltage power substation via a supervised Neural Network (NN) with two hidden layers. The signals are densely digitized RF pulses detected by a network of wide-band antennae controlled by a Field Programmable Gate Array. We first decompose the complex RF wave-field into few sources emitting repetitive radiation by grouping the pulses in clusters sharing a similar waveform. Then we describe each pulse with a set of features; with these inputs we train the NN by labeling the pulses with the corresponding cluster indices. NN shows an accuracy of about 95% in the classification of unlabeled pulses when they are described by a part of the fully sampled waveform or by its under-sampled envelope combined with other features like the AC power phase.
Detection, Features Extraction and Classification of Radio-Frequency Pulses in a High-Voltage Power Substation: Results from a Measurement Campaign
Ogliari, Emanuele;
2020-01-01
Abstract
We classify Radio frequency (RF) signals recorded in a high-voltage power substation via a supervised Neural Network (NN) with two hidden layers. The signals are densely digitized RF pulses detected by a network of wide-band antennae controlled by a Field Programmable Gate Array. We first decompose the complex RF wave-field into few sources emitting repetitive radiation by grouping the pulses in clusters sharing a similar waveform. Then we describe each pulse with a set of features; with these inputs we train the NN by labeling the pulses with the corresponding cluster indices. NN shows an accuracy of about 95% in the classification of unlabeled pulses when they are described by a part of the fully sampled waveform or by its under-sampled envelope combined with other features like the AC power phase.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
