APPLICATION OF REGRESSION ANALYSIS FOR BUILDING BASELINE MODELLING WITH OCCUPANCY DATA

Abstract

In occupant centric metrics, electricity consumption is usually disaggregated into end uses load for a more accurate analysis on the occupant’s impact. Even so, not 100% of the end uses consumption is attributed to occupants. The concept of the linear regression equation, y = mx+c (where x is the number of occupants and y is the total electricity consumption), is applied to separate the baseline electricity consumption (y- intercept, c), from the total electricity consumption, to obtain only the Non-Baseline Electricity Consumption (NBEC). Hence, this study first aims to identify the best fitting parametric regression model that can effectively quantify NBEC, and thereafter determine if using NBEC will have a significant impact on occupant centric metrics.

6 different types of NUS buildings: Ventus, AS2, Central Library (CLB), Temasek Laboratories (T-Lab), E1 and University Health Center (UHC)’s datasets comprising of hourly Wi-Fi counts and electricity consumption are used for univariate regression analysis. Different types of parametric regressions are performed to determine the best fit model for the datasets. Results shows that Piecewise Linear Regression (PLR) provides the best fit for the datasets.

The finalized PLR models for each building are used to quantify and analyze NBEC. Results shows that the highest percentage of NBEC are during the pre and post operational periods, which is in line with literature review, implying that PLR can effectively quantify NBEC. The NBEC are then applied on occupant centric metrics, and results from the Wilcoxon Signed Rank Test (WSRT) shows that the overall metrics results using NBEC is significantly lower than the metrics results using the conventional electricity consumption, with a low p-value of 2.98e-07. Thus, it is concluded that NBEC have a significant impact on occupant centric metrics.