Artificial Intelligence Enabled Online Non-intrusive Load Monitoring Embedded in Smart Plugs

As an Internet-of-Things (IoT) device for smart homes, smart plugs have been pervasive in households, which enable users to monitor and control their electrical appliances remotely and automatically. It is promising that, the networks of smart plugs in the power system will enable autonomous demand response for optimal grid operation. This benefits power systems from several aspects, e.g., enhancing renewable penetration and reducing the peak load. In order to facilitate energy management and minimize the impact of load shedding on the customers, it is meaningful to know the type of appliances connected to the smart plugs in a real-time and non-intrusive manner. Conventionally, the online non-intrusive load monitoring (NILM) is conducted in a central server, and it requires a large number of measurements transmitted to the cloud, which can impose a huge communication burden. In this paper, using the edge-computing capability of the smart plugs, some lightweight artificial intelligence-based NILM algorithms are developed and implemented inside the smart plugs. These practical algorithms are validated using massive hardware experiments. Case studies indicate that high accuracy can be achieved for NILM with limited measurements and limited storage.