Energy-saving potentials in high-temperature data centers: A spatio-temporal analysis

Because of the rapid growth of digital transformation, data center energy consumption has risen significantly, with cooling systems accounting for approximately 40% of total energy use. This study investigates high-temperature data centers, where operating temperatures are moderately increased to reduce cooling energy consumption while maintaining safe hardware operation. A comprehensive spatio-temporal analysis of sensor data under varying temperature and power levels (50%, 75%, and 100%) revealed strong correlations among sensors (r>0.8), confirming the reliability of measurements and enabling predictive temperature management. In particular, power consumption showed negligible correlation (r<0.3) with sensor readings at constant temperatures, indicating that temperature is the primary driver of sensor behavior. Spatial analysis highlighted a unified thermal response across inlet sensors (r>0.95) and air-conditioning systems, supporting targeted cooling strategies, whereas outlet sensors revealed localized hotspots, emphasizing the need for rack-level monitoring. Temporal analysis confirmed stationarity and strong autocorrelation (lag−1>0.9), validating the suitability of predictive models such as ARIMA, LSTM, and hybrid ARIMA–LSTM for proactive thermal management. Together, these findings establish a solid foundation for optimizing cooling strategies, enhancing energy efficiency, and improving the operational reliability of high-temperature data centers.