Evaluation of Feature Selection Methods for Non-Intrusive Load Monitoring

Residential and commercial buildings consume a major portion of electrical energy. Efficient and smart usage of electrical energy requires a detailed analysis of energy usage across all appliances in residential and commercial buildings. Non-intrusive load monitoring (NILM) is a method that disaggregates the total power into individual appliance’s consumption. In this paper, according to the IEEE standard 1459, different power quantities are found by using the voltage and current signals of different appliance events (ON/OFF). Then three feature selection algorithms i.e., neighborhood component analysis (NCA), minimum redundancy maximum relevance (MRMR) and Chi-squared tests (CST) are used to select the most prominent set of features. The selected features are then used as input to the classifier for the classification of appliances into pre-defined classes. The proposed method has reduced the computational time and improved classification results.