Processing of smart meters data based on random projections

Placed at the core of Smart Grids metering infrastructure are the Smart Meters, devices that not only measure electric energy consumption of a customer but also report data back to the utility. This process effectivelly sets a stage for a two-way communications link between meters and the utility, which may be used to improve not only the planning but also the operation of the grid. Data submitted goes through the Advanced Metering Infrastructure and is finally delivered to the utility at the edge of the topology. Once at the utility, the streamed data must be processed for billing and monitoring of the grid. Though, given the large volume of the data (e.g. billions of meters), usual processing may be costly both in time and resources, fact that motivates a search for lower complexity processing. In this paper we apply dimensionality reduction via random projection to obtain a reduced version (sketch) of Meters´ original data, thus increasing the processing throughput of the utility. Using real smart meters measurements, we show that processing using sketches sized 50% smaller than original data can achieve a 2% average relative error while presenting greater data rates.