Dynamic adaptive modeling for non-intrusive load monitoring with unknown loads

Abstract

Non-Intrusive Load Monitoring (NILM) technology has emerged as a promising approach to promote sustainable development. However, the complexity of real-world scenarios poses challenges to applying traditional NILM methods. These methods are typically designed for static load spaces, where a fixed number of known appliances are considered, limiting their ability to identify unknown loads accurately. To address this limitation, we propose an unsupervised method named the Dynamic Adaptive Factorial Hidden Semi-Markov Model (DA-FHSMM) that effectively distinguishes unknown loads in real-world scenarios. By considering the rate of change in the power consumption curves, the second-order derivative is utilized to provide additional insights into the dynamics of the load. It is then employed to measure the correlation between unknown and known loads. Via adaptively selecting highly similar models, we disaggregate the unknown load and improve the recognition accuracy compared to traditional methods. We validate the proposed model through case studies on two publicly available datasets, showing its superior performance in terms of MAE, F1-score, and Accuracy compared to existing models, consequently demonstrating its applicability to the scenario including unknown loads.