Building Mixed-Label Detection Based on Load Profile Clutering

Building Mixed-Label Detection Based on Load Profile Clutering

As part of my work with SinBerBEST2 I worked with Dr. Pandarasamy Arjunan on electrical meter data of numerous buildings. Our work was fous mainly on the Primary-Space-Usage (PSU) labels defined by domain experts. Such labels are often unrepresentative of the modern built environment since most of current infrastructure cannot be completely classified as one category.

We propose a modular methodology for systematic validation of the man-made building classification. This methodology leverages its modularity for homogenizing power meter readings based on different temporal contexts and aggregation functions. These PSU labels might not be an accurate representation for most building with diverse uses and loads but still carry enough information for detailed and grey-box simulations