An Advanced 2-Output DNN Model for Impulse Noise Mitigation in NOMA-Enabled Smart Energy Meters

The next-generation power grid enables information exchange between consumersand suppliersthrough advanced metering infrastructure. However, the performance of the smart meter degrades due to impulse noise present in the power system. Besides conventional thresholding techniques, deep learning has been proposed in the literature for detecting noise in NOMA-enabled smart energy meters. This research introduces a novel deep neural network (DNN) capable of simultaneously detecting and classifying impulse noise as either high or low impulse. Combining the analysis of detectednoise and its class has proven to be more effective in mitigating noise compared to previously proposed methods. The input feature vector to DNN is chosen based on its characteristics to detect impulse noise and its level in the data and includesROAD characteristics, median differences,and probability of impulse arrival. The performance evaluation shows that the Bit Error Rate (BER) of the proposed DNN is lower than the BER of single output DNN which is proposed in the literature for mitigation only. It is also shown that besides simultaneous detection and mitigation, the second output of the proposed DNN i.e. classification of IN validates the first output which is IN identification.