Improvement of forecasting and classification in smart metering systems using a neural compute stick

Abstract

Analyzing data on smart meters is a trend increasingly used by utility companies as it allows a better understanding of data directly from the source of origin. New distributed computing architectures like edge computing have given advance to improve data analytics. Generally, the capacity of such devices, including smart meters, is quite limited, so the use of specialized auxiliary hardware has begun to be used in these devices. The present work shows the results of using a neural stick compute for forecasting and data classification processes within smart metering systems. The results show that the processing times can be remarkably improved with the use of stick computers having a suitable model for artificial neural networks.