| Title |
A validated model for short-term prediction of cooling energy consumption in buildings - first step to forecast control of cooling |
| Authors |
Łokczewska, Wiktoria ; Cholewa, Tomasz ; Staszowska, Amelia ; Balaras, Constantinos A ; Fokaides, Paris A ; Deb, Chirag ; Mauro, Gerardo Maria ; Ascione, Fabrizio |
| DOI |
10.1016/j.jobe.2026.116361 |
| Full Text |
|
| Is Part of |
Journal of building engineering.. Amsterdam : Elsevier. 2026, vol. 127, art. no. 116361, p. 1-23.. eISSN 2352-7102 |
| Keywords [eng] |
Buildings ; Energy consumption ; Forecast control of cooling ; Modelling ; Outdoor impacts |
| Abstract [eng] |
Smart control of energy supply for cooling in buildings can significantly improve energy efficiency. However, existing modelling methods are often complex and rely mainly on artificial neural networks (ANN) and other machine learning techniques, posing various difficulties for integrating them in forecast control of cooling. Moreover, accurate cooling energy models are generally more demanding to develop than heating models. To address this research gap, this study proposes a novel, simple and physically based method for creating building cooling energy models that also take into account the characteristics of the existing cooling system. The approach uses measured cooling energy consumption and meteorological data-outdoor air temperature, wind speed and solar irradiance - to derive an equivalent outdoor temperature that represents the real thermal behaviour of the building during cooling operation. Proper selection of operating and weather data is essential to minimise the influence of unrelated factors. The method is demonstrated on an office building in Poland and a university building in Cyprus. For both case studies, the developed cooling energy models were validated, achieving for outdoor temperatures above 26 °C mean absolute percentage error (MAPE) values of 13.15% for the office building and 17.87% for the university building. To further assess robustness, Multilayer Perceptron (MLP) ANN models were trained using the same hourly inputs-outdoor temperature, wind speed and solar radiation. The ANN models did not significantly improve prediction accuracy, yielding higher MAPE values of 19.6–24.7% for the office building and 29.4–34.9% for the university building. The results highlight that buildings must be considered individually and show that the proposed method can provide a practical, transparent and accurate tool for estimating cooling energy performance. Future work will address occupant influence and integration into predictive control of air-conditioning systems. |
| Published |
Amsterdam : Elsevier |
| Type |
Journal article |
| Language |
English |
| Publication date |
2026 |
| CC license |
|
