A New Set of Steady-State and Transient Features for Power Signature Analysis Based on V-I Trajectory

Among consumers, there is a growing concern with the monitoring of the consumption of each individual equipment or appliance, rather than the consumption measurement of the whole unit. One way of individualizing the consumption is NonIntrusive Load Monitoring (NILM) based on power signature analysis. The main challenges with NILM approaches are the selection of relevant signature features for load identification (classification). Several approaches have been proposed in the literature using steady-state and transient characteristics. Among these methods, Voltage-Current (V-I) trajectories can be highlighted, mainly in steady-state conditions. In this work, we extend the features extracted from V-I trajectories, including new steady-state and transient features. We compare our approach to two other methods based on V-I trajectories recently presented in literature, using three different machine learning classifiers for load identification and a publicly available dataset as a benchmark (COOLL - Controlled ON-OFF Loads Library). Our results show that the proposed approach presents comparable results to those presented in the literature, achieving more than 99% of load recognition accuracy while significantly reducing the number of features. Most of the new proposed features - including transient-related - were selected in a feature selection procedure, demonstrating their relevance in the load identification and classification.