Energy demand model (FAIRiCUBE Use Case "Stock modeling of buildings")

Result of a role-based energy demand model (EPISCOPE)

Energy demand for space heating and domestic hot water was estimated based on key building characteristics—such as height, adjacent wall area, age, residential type, and location—along with annual average daily temperature. The analysis considered two scenarios: "as built" and "medium renovation", with assumptions derived from the EPISCOPE project. These assumptions were informed by expert insights and survey data, defining parameters such as the thermal transmittance of materials, internal heat sources, ventilation systems, and solar irradiation. Additionally, the number of heating days was determined based on the annual average daily temperature.

The energy demand estimates account for heat transfer losses through walls, roofs, windows, and floors, using U-values and material-specific transmission coefficients. Ventilation losses were also factored in by considering air exchange rates and conditioned air volume. In contrast, solar and internal heat gains helped offset heating demand, where solar gains were influenced by building orientation, shading, and window properties, while internal gains stemmed from occupant activities and household appliances. The model was applied to Oslo, Barcelona, and Rennes, and the results were validated against observed energy performance.

The scatter plots in the following figures illustrate the model’s ability to estimate residential energy demand across Oslo, Rennes, and Barcelona. The predictive performance was assessed using Root Mean Square Error (RMSE) and R², offering insights into its reliability across different years.

Oslo: The model achieved moderate accuracy, with an average RMSE of 78.87 and an R² of 0.63. However, the prediction quality varied across years, with R² values ranging from 0.47 to 0.70, suggesting potential inconsistencies in data representation or modeling assumptions. Notably, 2015 showed weaker performance (R² = 0.47, RMSE = 92.19), but overall, the model remained reasonably reliable.

Rennes: A similar level of performance was observed, with an average RMSE of 65.72 and R² of 0.60. The France OpenGov dataset showed a strong correlation with FAIRiCUBE predictions, with R² values consistently between 0.58 and 0.63. While the RMSE values indicate some prediction errors, the consistency in R² suggests the model effectively captures energy demand trends.

Barcelona: The model struggled to establish a meaningful relationship between Energia Barcelona data and FAIRiCUBE predictions. Despite a relatively low RMSE (55.48), the average R² was just 0.06, indicating a poor fit. In some years, R² values were near zero or even negative, suggesting that the model failed to explain variations in energy consumption.