结合数据校正的模型预测控制在LHTES功率调控中的应用：数据中心部署与案例研究

Abstract Latent Heat Thermal Energy Storage (LHTES) can effectively reduce cooling energy consumption in data centers through renewable energy utilization and peak load management. However, the lack of practical discharging power control methods for real-world engineering applications has hindered their widespread adoption. To address this challenge, this study used the model predictive control (MPC) strategy incorporating data correction to solve the power control challenges of LHTES, and validated in a data center cooling system retrofit project. Specifically, an efficient LHTES unit was first designed, with a series of charging/discharging experiments conducted to characterize its thermal storage properties. Based on the unit's structure, a temperature field model was established, which achieved a prediction error below 5 % within the flow rate range of 0.5–1.5 m 3 /h through composite parameter identification using experimental data on heat transfer fluid and phase change material (PCM) temperatures. To mitigate model divergence caused by abrupt operating condition changes and accumulated errors, an assimilation method based on extended Kalman filtering was employed for real-time model correction. Building on this model, MPC-based LHTES power control was implemented, achieving a control error as low as 3 %. To verify the feasibility of the proposed model in practical engineering applications, validation was conducted in a small-scale data center in China. Results showed that integrating the LHTES system with optimized operation strategies reduced the relative error in global power control to 1.52 %, while achieving 21.5 % energy savings and 60.3 % operational cost reduction. This strategy addresses the control challenge in LHTES applications, providing reliable technical support for the implementation of LHTES technology in data centers.