Probabilistic Bottom-up Modelling of Occupancy and Activities to Predict Electricity Demand in Residential Buildings

In Switzerland and most other developed countries, the building sector is one of the most important sources of energy consumption and its related adverse environmental impacts. As the reduction of greenhouse gas emissions and the abandonment of nuclear power have become key policy objectives, considerable e ort is undertaken to reduce energy demand, particularly in buildings, and to expand the use of renewable energies. For this reason, it is crucial to develop strategies and solutions, which optimally exhaust the possibilities to improve the e ciency of these systems. Dynamic simulation models are increasingly used to gain a more precise understanding of the underlying processes of energy ows in buildings. However, many of the fundamental phenomena are still not su ciently understood, resulting in potentially signi cant errors. The description of occupants' behaviour still leaves substantial room for improvement in the simulations; in particular, there is a lack of comprehensive and validated stochastic models predicting residential occupancy and activities, as well as their variations between individuals and households. However, the stochastic nature of residential behaviour is central regarding uncertainties in residential buildings' energy demand. This thesis develops adequate bottom-up models to predict time-dependent residential occupancy and activities, as well as household appliance ownership as a function of individual characteristics, and further proposes an innovative approach to relate the use of electrical appliances to the activities performed. The models are calibrated with detailed survey statistics of individuals' time use, as well as households' appliance ownership and appliances' power consumption. The approaches to predict presence are based on time-inhomogeneous rstand higher-order Markov processes, where sub-population-speci c behaviour is represented by corresponding parameters in the time-dependent transition probabilities and duration distributions. These explanatory variables have been rigorously selected on the basis of statistical signi cance using backward elimination. The higher-order approach is validated by establishing the relationship to the rst-order model, and by comparing the results. The variation of the model predictions for di erent sub-populations is illustrated and discussed. Similar approaches have been applied to simulate time-dependent residential activities, based on multinomial logit models predicting starting probabilities for