Data-driven Forecasting of Air Handling Unit Energy Consumption in Data Centers


Aras A. N., Tepe İ. F., Kül S., Irmak E.

IEEE 6th International Conference on Smart Information Systems and Technologies, Astana, Kazakistan, 13 - 15 Mayıs 2026, ss.1-6, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/sist61674.2026.11596446
  • Basıldığı Şehir: Astana
  • Basıldığı Ülke: Kazakistan
  • Sayfa Sayıları: ss.1-6
  • Gazi Üniversitesi Adresli: Evet

Özet

The rapid advancement in cloud computing and artificial intelligence (AI) technologies has significantly increased energy demand in data centers, where operating costs are highly dependent on cooling systems. Non-linear thermal behavior arising from power densities exceeding traditional limits requires predictive energy management strategies rather than reactive controls. This study presents a new approach to short-term forecasting of Air Handling Unit (AHU) energy consumption, using 4,320 hours of anonymized real-world operating data from a typical largescale data center. Linear Regression (LR), Random Forest (RF), and a hybrid CNN-LSTM architecture were considered in the study. Findings reveal that non-linear models significantly outperform traditional statistical methods, with the Random Forest model accurately forecasting changes in energy consumption patterns. The high forecasting accuracy provides a robust decisionsupport mechanism for control strategies to optimize Power Usage Effectiveness (PUE) and enhance energy efficiency while minimizing critical thermal risks.