Predicting Domestic Hot Water Demand Using Machine Learning for Predictive Control Purposes

Abstract

An important part of a building energy consumption is related to the domestic hot water consumption of its occupants. Predictive controllers are often considered as having the potential to reduce the energy consumption of hot water systems. In this work, a recurrent neural network is trained from the measured domestic hot water consumption of a 40 unit residential building in Quebec City, Canada, to predict the future consumption. It is found that the water consumption profile of the building changes from day to day throughout the year and has an important noise component. A predicting model is developed in this work and is obtained by pairing a recurrent neural network to predict the filtered domestic hot water demand with a random forest to predict the noise signal. The evaluated performances indices for the prediction of the next demand are satisfying (i.e., RMSE of 142.02 L/h and R2 of 0.71). In addition, it is found that the predictions made over the following hour using the same predicting model are accurate and could likely be used in a predictive control context.