Abstract
The European Commission has consistently emphasised the importance of decarbonising the building sector by introducing various building and energy-related directives. The research community shares this sentiment and has continuously put effort into optimising energy management in buildings. One key requirement for intelligent energy management approaches, such as model predictive control (MPC), is the availability of accurate load predictions. This allows the controller to anticipate the system’s behaviour, facilitating proactive decision-making to reduce energy consumption and emissions. In this context, this paper presents a multi-seasonal autoregressive integrated moving average (mSARIMA) approach for forecasting the domestic hot water (DHW) load. mSARIMA is a novel extension of the traditional SARIMA approach and can handle multiple seasonalities in a time-series dataset. Furthermore, the work presents a well-structured dynamic system formulation of the mSARIMA approach using a state-space representation, allowing efficient forecasting and easier integration into real-time decision-making systems. With its adaptive nature, the mSARIMA model is retrained at regular intervals using newly available measurements and parameters are re-estimated. For parameter estimation, the maximum likelihood method was employed, minimising the negative log-likelihood of the joint distribution of residuals. The model delivers reliable predictions, with an average root mean squared error and average percentage absolute bias of 2.1 kW and 8.2 %, respectively, over the simulation period with a 24-h prediction horizon. Finally, the model is implemented as software in the control hardware of a pilot site and undergoes real-time testing and validation. The model’s training process is quick, while the prediction process is instantaneous, with sufficiently accurate predictions.