Two-Stage Event-Driven NILM Utilizing Odd Harmonic Distortion

In this paper, the harmonic distortion (HD) of the 3
^rd
, 5
^th
and 7
^th
harmonic currents is utilized along with other common electrical features that commercially available smart electrical energy meters could extract from the main feeding panel of a household's electrical installation. A two-stage event-driven end-cloud Support Vector Machine (SVM) NILM framework is proposed in order to successfully apply the load identification task. This approach is compared to the one-stage time-driven end-cloud SVM NILM architecture that was developed in a previous work. The comparison results indicate that the load identification performance remains unchanged while at the same time the total requests for identification in the cloud-side server are remarkably reduced. In this way the cloud-side is being relieved from the high computational complexity and the high communication network burden for end users. In the first stage of the proposed framework an event detection algorithm indicates an event occurrence and activates the second stage for the load identification process in the cloud-side. The contribution of the % HD of the 3
^rd
total harmonic current in the event detection is highlighted. Regarding the SVM model utilization in the cloud-side, an already developed SVM algorithm is utilized in order to facilitate the comparison task. For the same reason, the utilized dataset is exactly the same as the one which was developed in the previous work.