Effect of A Flat-Top Distortion on A Load Identification System with Feature Extraction Based on Fractional Fourier Transform

Nonintrusive load monitoring systems usually achieve load identification by extracting and processing appliance features from aggregated signals in a house. The identification performance depends on the selected feature set and it is expected that this performance remains under distorted conditions. This paper discusses the effect of a flat-top power supply distortion on a residential load identification system with feature extraction based on fractional Fourier transform (FRFT). Three study cases are developed: one case uses only non-distorted data in the training stage and the other two cases include data with distortion. Identification performance is high for the three cases and it enhances when data under distorted conditions are included in the design. The results are compared with a system with feature extraction based on Fourier.