Ai-based model predictive control of chiller plant with thermal energy storage combining a water-side economizer system in data centers

초록

The thermal energy storage (TES) system in data centers can be applied to cooling energy reduction by utilizing natural cooling sources through free cooling operation when integrated with a water-side economizer (WSE) system. Furthermore, the use of model predictive control (MPC), which has been widely applied to enhance the charging and discharging performance of TES, can further improve cooling energy reduction by refining the free cooling–based charging operation. However, the high nonlinearity of the integrated system, with multiple devices interacting dynamically, reduces prediction accuracy, thereby limiting the effectiveness of MPC. To address this issue, this study developed and applied an MPC strategy combined with the XGBoost machine learning model, known for its superior predictive performance, to a WSE system integrated with TES. The grid-search method was incorporated into the MPC strategy to precisely control the free cooling operation of the integrated system by adjusting the chilled water set-point temperature and TES operating states. Comparative analysis between the proposed AI-based MPC and a conventional rule-based control demonstrated that, under stable indoor temperature conditions during the intermediate season, the chilled water set-point temperature increased by approximately 3–4 °C. As a result, the proposed method achieved an 11% reduction in chiller energy consumption and an overall 4% reduction in total cooling energy use during the intermediate and winter seasons. Copyright © 2026. Published by Elsevier B.V.