An Advanced Data Center Multi-Chiller Dynamic Load Distribution Methodology and Engineering Practice
Yiming Luo, Xiang Sean Zhou, Jianchao Cao, Biao Li, Yahui Zhao, Tangbo Jing, Lifei Zhang, Nishi Ahuja, Jun Zhang, Yuyang Xia · 2018
Data Center energy efficiency is one of most popular subjects in CSP (Cloud Service Provider) infrastructure research. This paper introduces one methodology as well as engineering practice from Baidu data center on dynamic load distribution of multi-chiller system based on machine learning theory. This methodology is to establish a dynamic multi-chiller modeling strategy. With theoretical analysis and engineering practice, rRMSE (relative Root Mean-Squared Error) has been used and result shows less than 5%. Furthermore, a pilot-deployed engineering case with Baidu multiple objects framework including dynamic multi-chiller modeling, real-time IT load, next 30 minutes IT load as well as outdoor environmental conditions is introduced to demonstrate overall energy efficiency which has been proven to be an excellent case of Artificial Intelligence usage in data center energy saving. Energy efficiency outcome, TCO (Total Cost Ownership) benefits and future plan are summarized in the end.