A time-variant load model based on smart meter data mining

This paper proposes a novel time-variant load model based on data-mining of a historical smart meter database. As part of the ongoing smart grid transformation, smart meters have been widely installed producing massive amount of data and information yet unexplored. One of the critical needs for distribution system operations and planning applications is enhanced modeling of the load, in particular, its dependence on the voltage. Under the typical smart meter recording resolution (15-minutes), the load´s P-V and Q-V properties are buried in the spontaneous load changes caused by random customer behaviors. To overcome this, the concept of load condition is introduced and data mining techniques such as Kullback-Leibler divergence and K-subspace are implemented. The proposed model is tested on a large database for the Georgia Tech campus, and the results demonstrate that the new model captures the time-variant property of the load on the building level without additional investment.