Large-Scale Cooling System Modeling and Control Optimization of a Real-Life Building

Model predictive controls (MPC) for cooling systems in buildings have been widely acknowledged as an effective approach for minimizing energy use without disturbing the required cooling loads. It can search for optimal control variables over a time horizon, considering the building’s heating/cooling demands and its interaction with environmental conditions. However, many challenges such as formulating an optimization problem, modeling a target system, and implementing MPC in real-time need to be overcome for it to function as expected. In this study, a large-scale cooling system was targeted, considering these obstacles.

The key to the success of this large-scale optimization is as to how to balance the number of operating cooling towers and chillers according to the given cooling load and outdoor wet-bulb temperature. This interaction was also influenced by the cooling water temperature and volumetric flow rate, which subsequently determined the number of cooling water pumps. The pumps are connected to a supply header that maintains a constant supply pressure. Accordingly, the optimization problem of our interest must solve nonlinear interwoven dynamic relationships between the chillers, cooling towers, pumps, as well as the time-varying building’s cooling demand influenced by outdoor weather condition. In addition, it also should take into account part-load ratios of the operating devices and their operating efficiencies.

The building’s operational data were collected by a facility team for several years. A set of simulation models were developed in a hybrid fashion where physics-based and data-driven models were combined. Chillers and cooling towers were individually modeled based on their respective data. These models were grouped at the system level, and subsequently a set of the group models determined which devices should be operated in terms of a global optimization.

For the real-life implementation of the developed MPC, DeepONet was introduced for intuitive understanding of the optimization results and enhancing the efficiency and effectiveness of cooling system control. By leveraging DeepONet, complex relationships between various system parameters can be captured and processed in real-time. This not only optimizes energy consumption but also improves system reliability and performance. Furthermore, the use of DeepONet facilitates easier integration and scalability of advanced control algorithms in existing cooling infrastructure, providing a robust solution for maintaining optimal operating conditions.

By combining the strengths of MPC and DeepONet, this paper addresses the complexities and challenges inherent in managing large-scale cooling systems, offering a pathway to more sustainable and efficient building management practices.