Impact of stochastic modelling of windows operation on residential building energy simulation and rating

Window operation has a significant impact on residential building energy consumption, and realistic representation in building performance simulations remains a challenging task. This paper describes and evaluates the impact of integrating seven data-driven stochastic window operation models within an Australian residential energy rating tool. These models, which represent the aggregate behaviour of a cohort of buildings, include Gaussian distribution, logistic regression, classification tree, Markov-logit hybrid, random forest and artificial neural network approaches. Two sets of models were trained, using operational data from 1) eight apartments in an Australian temperate climate and 2) 41 homes in an Australian subtropical climate. The models were then coupled in a co-simulation environment, with a building performance simulation software to analyse the performance of the aforementioned eight apartments. Results showed that models trained with local climate-matched data reproduced observed window opening patterns more accurately (average Jaccard similarity > 0.68) than those trained on data from different climates (>0.45), while both models gave more realistic window operation in comparison with the energy rating software’s default rule-based model. Benchmarked against the energy rating software’s default rule-based logic, the stochastic models predicted substantially higher cooling energy use (71–480% increase), leading to star rating reductions of 0.5–3 out of 10 under the local rating scheme. These findings highlight the critical influence of occupant behaviour modelling on energy ratings and demonstrate, for the first time, the timestep-level impact of co-simulated stochastic window operation models in residential Building Performance Simulations.