Comprehensive Performance Comparison of Supervised Machine Learning Algorithms in Non-Intrusive Load Monitoring

Recent developments in the field of smart grid have led to renewed interest in load monitoring strategies for achieving effective energy management schemes. There are vast amount of published studies describing the role of non-intrusive load monitoring (NILM) system based on various learning algorithms. It is widely known that the accuracy of load identification depends strongly on utilized methods and its features. Thus, the main aim of this study is to investigate the comparative accuracy of machine learning algorithms which have the same training data with different feature subsets. Afterwards, a low-cost data acquisition system for NILM using bagged tree ensemble algorithm is developed and demonstrated in detail. The proposed structure is tested on the ThingSpeak IoT platform to reveal the effectiveness of the evaluated concept.