Improving the Recognition Performance of NIALM Algorithms through Technical Labeling

Abstract

A myriad of different electrical devices populate a typical household nowadays. Non-intrusive appliance load monitoring (NIALM) is an approach to find out how much energy each of them consumes in order to take measures to improve the overall energy efficiency. This article describes the ongoing research on improving electric loads recognition performed by NIALM algorithms within the context of smart homes and intelligent environments. The recognition performance can be significantly improved by decreasing the number of categories to be analyzed. The authors studied several labeling methods to categorize and group loads in order to increase the overall recognition rate. 31 different devices have been measured and labeled in different device states. Their input curves have been compared with 5 different machine learning algorithms. The best results could be reached by dividing all the loads into groups with small divergence in their normalized current curve. This approach has significantly increased the performance of NIALM recognition algorithms.