Data centers have been water guzzlers owing to heavy-duty cooling and energy requirements. Globally, very limited efforts have been made in water conservation and water utilization effectiveness (WUE) in data centers (DC). The industry standard for WUE is 1.8 L/kWh. This novel research is based on water conservation principle by improvising the WUE in DC. To achieve this objective this work proposes an efficient smart water cooling modeling framework that leverages the simulation software EnergyPlus (Open source platform) coupled with Sinergym (Open source framework) based on Deep Reinforcement Learning (DRL).

In this study, a realistic DC has been leveraged to fully test the simulated solution for summer design days. The smart water cooling model optimizer has been trained using a Deep reinforcement learning (DRL) that minimizes the total water consumption. A comparison of the study (with or without DRL) infers that merely deploying the simulation software does not efficiently assist in improvising the WUE. It’s the DRL approach using DDPG algorithm that has improvised the WUE industry standard by 20.64%. With this integrated solution, being tested in the real time DC, the resultant WUE has outperformed the industry standard.