Residential smart meter data compression and pattern extraction via non-negative K-SVD

Smart meter plays a vital role in the management of smart grid. Massive electricity consumption data are being collected with the popularity of smart meters, which pushes the electricity demand side into a big data world and poses great challenges to data communication and storage. The residential electricity consumption behaviors vary with different lifestyles and family configurations. It's assumed that each daily load profile is essentially a combination of several certain hidden usage patterns. On this basis, a K-SVD based data compression technique is proposed in this paper to decompose each overall load profile into a linear combination of several meaningful partial patterns and minimize the reconstruction error using sparse and redundant representations. Comparisons with discrete Wavelet transform (DWT) and principal component analysis (PCA) are conducted. The results show that the proposed method can achieve better compression quality and identify meaningful hidden usage patterns.