Real-time model for unit-level heating and cooling energy prediction in multi-family residential housing

In this paper, we introduce a real-time modelling approach to predict the heating and cooling energy consumption of each housing unit in multi-family residential buildings. We first present measured yearly heating and cooling energy use data from an actual building and introduce the eco-feedback design and associated modelling challenges. Subsequently, we present a real-time parameter learning-based modelling approach. The model has a state-space structure while state filtering and parameter estimation are simultaneously executed through particle filter with sequential Bayesian update. The housing unit-level model is coupled with a probabilistic model of the heating and cooling system by using thermostat, power metre, and mechanical system catalogue data through a Bayesian approach. The results show that the median power prediction of the model deviates less than 3.1% from measurements while the model learns seasonal parameters such as the cooling efficiency coefficient through sequential Bayesian update.