Regression versus probabilistic approach for operational data – Heat demand of buildings to be connected to a district heating system

• Published in: Energy and buildings 2023-09, Vol.294, p.113209, Article 113209
• Main Authors: Hachez, Jonathan, Chicherin, Stanislav, Chaudhry, Afraz Mehmood, Bram, Svend
• Format: Article
• Language: English
• Subjects: Design; Deterministic; Normal distribution; Prosumer; Regression; Stochastic
• ISSN: 0378-7788
• DOI: 10.1016/j.enbuild.2023.113209

The robust design of 5th generation district heating and cooling (5GDHC) systems requires accurate and sufficient operational data on the energy consumption and production of the connected prosumers. Often these data requirements are not met. Historical energy-related data for buildings is sometimes no longer accessible, and if available, often limited in time span and temporal resolution. Also, technical details on the utilization of the energy within the heating, ventillation and air-conditioning (HVAC) system of a building and its relationship with building occupation and outside weather conditions are often difficult to obtain. This lack of data requires careful pre-processing of this data, to ensure a robust 5GDHC system design. To cope with this lack of data, we propose a two-step model. First, techniques of linear regression are used to detect relations with contextual key factors, such as occupation, day of the week, hour of the day and outside temperature. In a second step, a piece-wise stochastic approach is used to generate artificial data for any combination of the key factors, allowing for the detection and regeneration of intermittent behavior that is typical present in the obtained operational data. The advantage of this method is that it takes into consideration the unknowns and randomness of the system’s behavior. It allows for variance within the artificial time series (peaks and gaps), which makes it better suited for the robust design of 5GDHC systems. We applied this model on operational data from 23 buildings in Flanders, Belgium and produced artificial data for these buildings. The model can separate HVAC energy data related to the outdoor weather conditions from non-HVAC related energy data. It can also forecast energy data for other weather conditions and building occupation.