A bottom-up stochastic model to predict building occupants' time-dependent activities

A bottom-up modelling approach together with a set of calibration methodologies is presented to predict residential building occupants' time-dependent activities, for use in dynamic building simulations. The stochastic model to predict activity chains is calibrated using French time-use survey data (of 1998/1999), based on three types of time-dependent quantities: (i) the probability to be at home, (ii) the conditional probability to start an activity whilst being at home, and (iii) the probability distribution function for the duration of that activity. The behaviour of the individual agents in the model is first calibrated using a generic approach, where every individual is assumed to behave the same. A refinement is then presented to account for variations in the behaviours of sub-populations, having specific individual characteristics. Furthermore, a statistical approach is introduced for the modelling of transitions between two successive activity types as a Markov process. The models are then validated using a cross-validation technique, and their predictive performance is compared at an individual level, as well as for aggregated (sub-)populations.

► Residential activities can be reliably predicted time-dependently.
► Starting probabilities/duration distributions of activities are key models implied.
► Global activity shares are satisfactorily predicted using a generic approach.
► The predictivity is improved using an individual-dependent approach.
► Including transitions between activity types also improves predictivity.