Energy Consumption of Methods for Pattern Recognition using Microcontrollers

Abstract

This paper presents the study of energy consumption of the methods for recognizing patterns/anomalies in numerical series, namely, the light sensor values in a smart home system. Methods for analyzing time series, identifying anomalous zones, and testing anomaly recognition algorithms are presented, and the smart system is prototyped. The energy consumption of correlation, comparison, and recognition methods using NNs is measured and analyzed. The case study has confirmed that the most resistant to signal changes and interference is the correlation analysis method. A methodology for applying recognition algorithms for different strategies for using optimal energy consumption is presented.