Economic Optimization of Large-Scale Commercial Building Heating, Ventilation, and Air Conditioning Systems

Nishith Patel

Heating, ventilation, and air conditioning (HVAC) systems of commercial buildings have received significant attention recently due to the large amounts of energy consumed in these applications. Commercial buildings represent about 20% of U.S. energy consumption corresponding to almost 18 quads of energy. Almost all heating and cooling systems in commercial buildings and educational facilities nowadays rely on temperature controllers whose only goal is to converge to the desired temperature set point and stay there, within a small tolerance. The current gold standard for this control is a method known as model predictive control (MPC), where the aim is to reach the set point over a specified time horizon with minimum controller effort. A much better goal is to minimize total energy (i.e. total cost), using the recent advancements in economic MPC.

Substantial energy cost savings are possible for HVAC systems in utility markets with time-varying price structures through load shifting by using energy storage resources. While model predictive control has been recommended as a strategy to attain these savings, wide-scale implementation has not yet occurred. In large-scale commercial applications with hundreds of zones, solving a single comprehensive mixed-integer optimization problem online for MPC in real time is not feasible. In this work, a hierarchical decomposition of the overall problem is proposed to address this gap. The proposed hierarchical decomposition uses aggregate models of the airside and waterside problems to reduce complexity, and iterations are not required between the subsystem controllers. Energy cost savings between 5% and 40% can be achieved with this control system based on the amount of energy storage and the economic incentives for load shifting.

The decomposition framework is also extended to handled systems with embedded battery units. Working towards the long-term objective of implementation, methods for model development are also areas of interest. Models and data are developed for a large-scale case study loosely based on the Stanford University central HVAC plant. The parameters from the case study are made publicly available for researchers in the HVAC community who would like to design other control architectures and evaluate their performance. The case study is also used to demonstrate that the proposed decomposition can be solved in real time.

This project is in collaboration with Johnson Controls, Inc.