Review of advances in scaling non-intrusive load monitoring for real-world applications

Sustainable and safe electricity usage is essential to making electrification more environmentally and human-friendly. The first step is to make fine-grained electricity consumption data readily available. Recently, non-intrusive load monitoring (NILM) has gained attention for estimating appliance-level electricity usage only from aggregated measurements, offering a cost-effective solution for large-scale electricity monitoring. However, as NILM scales from the lab to real-world applications, it faces not only methodological difficulties from diverse electricity consumption characteristics under complex disturbances, but also operational difficulties in managing numerous devices and decentralized data across cloud and edge with limited computing resources. Given the critical need to enhance NILM's practicality for widespread adoption, research interest in this field has surged significantly over the past six years. Therefore, we analyze the above difficulties following a brief review of NILM's fundamentals. Then, we highlight the advances across four key aspects of NILM's practicality: robustness, adaptability, collaboration, and deployability. In addition, we discuss the limitations that hinder real-world NILM applications, aiming to inspire further research. Finally, we provide an outlook on the developments in data ecosystem, implementation guidance, application scenarios, and related services