Power Consumption Data Compression via Shift-Invariant Dictionary Learning

Prevalent smart meters are recording a vast amount of power consumption data. The escalation of data volume renders storing and transmitting costly. Traditional data compression techniques were effective solutions. However, they often neglect the pattern-rich and time-invariant properties of the power consumption data. To address this challenge, we propose a two-module pipeline, combined with shift-invariant dictionary learning (SIDL) and k-means clustering. Experimental results prove that the proposed pipeline attains the best effect in general cases. The results also imply that the pattern-rich and time-invariant properties are the key to efficiently handle the power consumption data.