A unified probabilistic model for predicting occupancy, domestic hot water use and electricity use in residential buildings

Abstract

A novel strategy that combines separate probabilistic models developed by other researchers into a unified model for generating schedules of active occupancy, domestic hot water (DHW) use, and non-HVAC electricity use in multiple residences with a 10-min resolution for every day of the year is described. A variety of new model functions are introduced in order to generate stochastic predictions for each of numerous residences at once, to enforce appropriate variability of behaviors between dwellings and to ensure that domestic hot water and electricity use are coincident with occupancy. The separate models used in this paper were previously developed for the US and the UK; in the unified model, scaling factors were added to these models to adjust the predictions so as to better agree with national aggregated data for Canada. The unified model was validated with measurements of domestic hot water use and electricity consumption from the 40 residential units of a social housing building in Quebec City, Canada. The behavior of occupants in the case study building was simulated 100 times in order to validate the outputs of the unified model. Goodness-of-fit tests applied to each of these simulations showed that the fit between simulated and measured dwelling-per-dwelling distributions was acceptable for 97% of the DHW consumption profiles and for 92% of the electricity consumption profiles. However, there remain discrepancies between simulations and measurements, such as an overestimation of the DHW and electricity consumption in the morning.