Sensitivity analysis of probabilistic occupancy prediction model using big data

Abstract

The sustainable development of the building sector relies on having more accurate data pertinent to the different energy aspects of building design and operation phases. One of the accepted methods for estimating building energy consumption is simulation, which requires building parameters and occupancy information as inputs to model the energy performance of buildings. Building energy simulation tools are mature in terms of incorporating proper building parameters in the energy analysis. Some shortcomings are, however, observed regarding occupancy data, which cause large discrepancies in the energy usage even between similar buildings with the same characteristics. In order to improve the performance of energy simulation models, the sources of errors regarding the occupancy input data should be investigated. To this aim, the sensitivity of the occupancy prediction models, which are widely used to represent the occupancy information in energy models, to their input occupancy data needs to be evaluated. Occupancy prediction models exploit real data pertinent to the occupants’ locations and behavior to predict the probability of an event and generate the occupancy probabilistic profiles. The data collection period and the resolution level used to analyze the collected data are two crucial factors for developing accurate occupancy prediction models. This study aims to perform a comprehensive sensitivity analysis on these parameters affecting the performance of probabilistic occupancy prediction models. The outcomes of this research are the optimum settings of occupancy prediction models, which result in the generation of the most reliable occupancy information.