Inverse Modeling and Part-Load Performance Optimization of Data Center Air-Cooled Chillers Using Differential Evolution Algorithm

As data centers expand rapidly, cooling systems have become a major contributor to their overall energy consumption, typically accounting for 30-40% of the total facility power usage. This study presents a physics-based white-box thermodynamic model for air-cooled chillers in data center cooling applications. The model, implemented in Python using the CoolProp library, incorporates an Enhanced Power Law compressor efficiency formulation to better capture part-load behavior. Key model parameters were identified through inverse analysis using a Differential Evolution (DE) algorithm, and the calibrated model showed a mean absolute error (MAE) of 0.089, with relative errors within 5%, over an operating range of part-load ratio (PLR) 0.25-1.00 and outdoor air temperatures of 15-35℃. With the R-22 model as the baseline, five alternative refrigerants (R-32, R-134a, R-410a, R-1234yf, and R-1233zd) were compared by means of contour maps of the change in the coefficient of performance (ΔCOP) across the entire operating domain. Annual 8,760-hour simulations were conducted for six climatically distinct U.S. cities representing ASHRAE climate zones 1A, 2A, 4A, 4C, 5A, and 6B. Of these, San Francisco (marine climate, ASHRAE 4C) and Cincinnati, Kentucky (continental climate, ASHRAE 4A) were selected as representative cities for a detailed evaluation of refrigerant-specific annual energy performance. Among the candidate refrigerants, R-1233zd showed the lowest annual energy consumption in both climates, whereas R-410a recorded the lowest COP under most operating conditions.