A Nonintrusive Load Identification Method Based on Improved Gramian Angular Field and ResNet18

• DOI: 10.3390/electronics12112540
• Publication Date: 2023-06-05T06:18:29Z
• Authors: Jingqin Wang; Yufeng Wu; Liang Shu
• Keywords: Identification; Feature (linguistics)

Image classification methods based on deep learning have been widely used in the study of nonintrusive load identification. However, process encoding electrical signals into images, how to fully retain features raw data and thus increase recognizability loads carried with very similar current are still challenging, loss will cause overall accuracy identification decrease. To deal this problem, paper proposes a method improved Gramian angular field (iGAF) ResNet18. In proposed method, fast Fourier transform is calculate amplitude spectrum phase reconstruct pixel matrices B channel, G R channel generated GAF images so that color image fused by three channels contains more information. This improvement enables feature usually missed general image. ResNet18 trained iGAF for Experiments conducted two private datasets, ESEAD EMCAD, public PLAID WHITED. Experimental results suggest performs well both achieving accuracies 99.545%, 99.375%, 98.964%, 100% four respectively. particular, demonstrates significant effects waveforms datasets.