<p>In this study, a system that recovered waste heat from data center’s centralized water-cooling system to preheat centralized heating systems was proposed, and the optimal design parameters were investigated, resolving the mismatch between waste heat and building demand. Firstly, parametric analyses were conducted under the design condition of centralized heating. Effects of mass flow rate and temperature after heat recovery of the data center’s chilled water on performances of the system were investigated. The results showed that, there existed an optimal temperature after heat recovery minimizing primary energy consumption of the system for each mass flow rate, and the values of optimal temperature after heat recovery for different mass flow rate were quite different. Then, the optimal mass flow rate and temperature after heat recovery were determined for different heat capacity ratio between heat supply from data center and heat demand of the centralized heating system. With the data center’s heat density being 500, 800, and 1100 W/m², and the ratio varying from 0.25 to 2, the optimal parameters were summarized. Based on the optimized values, a random forest-based parameter estimation model for machine learning tool was yielded to determine the optimal parameters. Percentage errors of the model were within ±15%. With the system being designed following the optimization values, primary energy saving ratio was from 58.7% to 78.2% compared with the traditional ones. This work provides a replicable tool a to assist optimal design and equipment selection of the heat recovery system.</p>