Impacts of raw data temporal resolution using selected clustering methods on residential electricity load profiles

There is growing interest in discerning behaviors of electricity users in both the residential and commercial sectors. With the advent of high-resolution time-series power demand data through advanced metering, mining this data could be costly from the computational viewpoint. One of the popular techniques is clustering, but depending on the algorithm the resolution of the data can have an important influence on the resulting clusters. This paper shows how temporal resolution of power demand profiles affects the quality of the clustering process, the consistency of cluster membership (profiles exhibiting similar behavior), and the efficiency of the clustering process. This work uses both raw data from household consumption data and synthetic profiles. The motivation for this work is to improve the clustering of electricity load profiles to help distinguish user types for tariff design and switching, fault and fraud detection, demand-side management, and energy efficiency measures. The key criterion for mining very large data sets is how little information needs to be used to get a reliable result, while maintaining privacy and security