Buildings are one of the major sources of global energy consumption. In 2021, residential and commercial buildings were responsible for around 39% of total U.S. energy consumption and 74% of total U.S. electricity consumption. Consequently, research on the operation of building Heating Ventilation, and Air Conditioning (HVAC) systems can lead to significant energy savings and carbon emission reduction. 

In the post-COVID era, HVAC energy management has become even more challenging. High airflow rates can reduce the exposure of occupants to viral pathogens in indoor environments thus reducing the infection risks. Yet, this necessary increase of airflow rate can lead to higher energy consumption. In practice, many HVAC systems have been operating at maximum airflow rates in response to COVID-19. For instance, starting from the spring 2020, the Facilities Management (FM) at UC San Diego has implemented a policy of maximum fresh-air intake with minimal or no recirculation during office hours, which results in the building's energy consumption being 2-2.5 times higher than the nominal energy costs. While beneficial for air quality and reducing infection risks, these HVAC control policies are unsustainable for various reasons, including skyrocketing energy consumption and strain on the mechanical systems. 

Therefore, it is imperative to develop an integrated control framework that simultaneously ensures a comfortable and healthy indoor environment while minimizing the energy consumption.  

Our proposed framework is in the above diagram, where the system state includes airflow velocity, CO2 concentration (Note: we use CO2 concentrations as an indicator of air quality, and it allows for the inclusion of other aerosol particles, e.g., PM 2.5, PM 10 and airborne pathogens), and temperature. The airflow velocity field is modeled by the Navier-Stokes equations, and the dynamics of CO2 concentration and temperature are modeled with convection-diffusion PDEs. We then formulate both the learning and control of indoor dynamics as PDE-constrained optimization problems.

In the system learning task, the goal is to estimate the unknown building parameters that govern the fluid dynamics. In the control task, the goal is to minimize the energy consumption while ensuring thermal comfort and air quality, via optimizing the supply airflow rate and supply air temperature setpoints. 

Our approach achieves a significant reduction in energy consumption, compared to existing control methods such as maximum airflow policy, learning-based control with RL, and optimization-based control with ODE models. Compared to the maximum airflow policy, our method achieves a 52.6% reduction in energy consumption. Additionally, we see energy savings of 36.4% and 10.3% compared to RL and control with ODE models, respectively. While RL and control with ODE models occasionally violate the safety constraints, our approach successfully maintains comfortable and healthy environmental standards at all time.

One challenge of building research is the lack of a benchmark simulation environment for developing and evaluating different RL algorithms with realistic building models. In a recent work, we propose “BEAR” [1], a physics-principled Building Environment for Control and Reinforcement Learning. The platform allows researchers to benchmark both model-based and model-free controllers using a broad collection of standard building models in Python without co-simulation using external building simulators. We demonstrate the compatibility and performance of BEAR with different controllers, including both model predictive control (MPC) and several state-of-the-art RL methods with two case studies. BEAR is available at https://github.com/chz056/BEAR

Modern buildings encompass complex dynamics of multiple electrical, mechanical, and control systems. One of the biggest hurdles in applying conventional model-based optimization and control methods to building energy management is the huge cost and effort of capturing diverse and temporally correlated dynamics. 

Here we propose an data-driven predictive control framework for complex building systems and other unknown dynamical system control. By utilizing a high volume of data coming from advanced sensors, we train Input Convex Neural Networks (ICNN) for modeling complex system dynamics. In particular, we show that these ICNNs are universal approximators of all convex functions and provide efficient convex approximation of underlying physical dynamics. The optimal controllers can be achieved by solving a convex model predictive control problem.

Experiment results demonstrate the good potential of the proposed ICNN-based approach in a variety of control applications. In particular, we show that in the building HVAC control example, our method achieved up to 20% energy reduction compared with classic linear models; and for different robotic locomotion tasks, we could achieve over 10% higher performance using 5× less time compared with the state-of-the-art model-based reinforcement learning method.

References:

[1] Christopher Yeh, Victor Li, Rajeev Datta, Julio Arroyo, Nicolas Christianson, Chi Zhang, Yize Chen, Mohammad Mehdi Hosseini, Azarang Golmohammadi, Yuanyuan Shi, Yisong Yue, and Adam Wierman, "SustainGym: Reinforcement Learning Environments for Sustainable Energy Systems", in Thirty-seventh Conference on Neural Information Processing Systems (NeurIPS), 2023.

[2] Yuexin Bian, Xiaohan Fu, Bo Liu, Rohith Rachala, Rajesh K Gupta, and Yuanyuan Shi, "BEAR-Data: Analysis and Applications of an Open Multizone Building Dataset", ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (ACM BuildSys), 2023.

[3] Chi Zhang, Yuanyuan Shi, and Yize Chen, "BEAR: Physics-Principled Building Environment for Control and Reinforcement Learning", ACM International Conference on Future Energy Systems (ACM e-Energy), 2023.

[4] Yize Chen, Yuanyuan Shi, and Baosen Zhang, “Optimal Control Via Neural Networks: A Convex Approach,” In International Conference on Learning Representations (ICLR), 2019.

[5] Yize Chen, Yuanyuan Shi, and Baosen Zhang, “Modeling and optimization of complex building energy systems with deep neural networks”, Asilomar Conference on Signals, Systems, and Computers, 2017.