Energy use intensity disaggregation in institutional buildings – A data analytics approach

Abstract

This work addresses the issue of extracting device-level Energy Use Intensity (EUI) information for institutional buildings (IB) through an analytical non-intrusive load monitoring (NILM) where no sub-meter is employed at device-level. The proposed model urges the use of similar meteorological conditions in the building classification. We studied load profiles as energy footprint of six major load categories, and in accordance with some embodiments of peripheral parameters of ambient temperature, workday, time of day, daylight length, intensity of sunlight, number of occupants, and humidity. Among those major load categories, initially we disintegrated thermal loads through finding an equilibrium temperature from fitted cooling and heating functions. The contradiction of data and building occupancy renders disaggregation problem particularly challenging, so abiding to the definition of time equivalent occupancy (TEO), we deployed a k-means approach to categorize the daily load profile, where particular disaggregation approaches was selected for each individual load profile cluster to calculate the common and occupancy load categories. By imposing the daylight difference for equinox and solstice periods for non-working days, we estimated and extracted the exterior lighting. The remained accumulated energy casts alluded interior lighting with part of common and occupancy load categories, so a day light factor is proposed to offset the interior lighting from common and occupancy loads. We delineated an implementation of this framework using two case studies. The efficiency of our approach is demonstrated using a real data, where maximum EUI discrepancy errors of 4.84% in lighting, 3.85% for thermal, 1.6% in common and 3.24% in occupancy load categories were found between the proposed method and the sub-metered data.