Development of NetZero BEMS through AI-based HVAC System Control

Heating, Ventilation, and Air Conditioning (HVAC) systems contribute substantially to nearly half of total building energy consumption, leading to the building sector being a major energy consumer in many countries like Singapore and South Korea. To achieve net-zero energy buildings (NZEB), using optimal control/operation methodologies for energy management of HVAC systems is one of the prompt and cost-effective solutions that not only improve building energy efficiency but also reduce building energy consumption. 

Since traditional methodologies for controlling HVAC systems are limited by manual labour and expert knowledge, artificial intelligence (AI) technologies for energy management and optimisation of HVAC systems have been getting attention recently due to their capabilities of automation, self-learning, and predictive control. In this context, this research project aims to study AI-based energy management and optimization frameworks to enhance the energy efficiency of HVAC systems and solve coordinated control problems in HVAC sub-systems.

Lead PIs: Christopher Lee (Singapore) and Taesu Cheong (South Korea)

Co-PI

  • Yuen Chau (Singapore)

Host Institutions: Nanyang Technological University (Singapore) and Korea University (South Korea)

Industry Partners: Building System and Diagnostics Pte Ltd (Singapore) and CloudN Co.,Ltd (South Korea)

Group shot of the winning team with partner representatives and the AISG team at the Award Ceremony (Jul’23)

Problem Scope

Heating, Ventilation, and Air Conditioning (HVAC) systems contribute substantially near half of total building energy consumption leading to the building sector being a major energy consumer in many countries like Singapore and South Korea. To achieve net-zero energy building (NZEB), using optimal control/operation methodologies for energy management of HVAC systems is one of the prompt and cost-effective solutions that not only improve building energy efficiency but also reduces building energy consumption.  

Since traditional methodologies for controlling HVAC systems are limited by manual labour and expert knowledge, artificial intelligence (AI) technologies for energy management and optimization of HVAC systems have been getting attention recently due to their capability of automation, self-learning, and predictive control. In this context, this research project aims to study AI-based energy management and optimization frameworks to enhance the energy efficiency of HVAC systems and solve coordinated control problems in HVAC sub-systems.

Design Approach

From the engineering and algorithmic perspectives, the technological gaps between the current state-of-the-art research approaches and practical control methods for the optimal control/operation of AHU and chiller systems are summarized as follows.

  • Numerous contemporary studies focusing on AHU control have employed computational fluid dynamics (CFD) to forecast airflow patterns within buildings. However, the utilization of hydrodynamics calculations has posed a significant challenge. The primary objective of the predictive model presented in this study is to simulate air conditions using data without resorting to complex calculations.
  • The existing studies have mainly employed rule-based, model-based, and data-driven approaches for the development of optimal chiller control/operation. However, the rule-based approaches require expert knowledge and rules, that may be specific and customized since the chiller system in every building is unique, and generally fail to achieve optimal control performance. The performance of model-based and data-driven approaches relies on the high accuracy and quality of the system model and rich system measurements without uncertainty, and these requirements may be impractical in real systems.
  • Reinforcement learning-based control strategy has been demonstrated to be a potent strategy in prior studies aimed at reducing building energy consumption. However, the gap between RL-based control and practical control is still not eliminated. Most studies verified the performance of RL-based approaches through numerical simulation or small lab-based testbeds instead of engineering/experimental practice. Furthermore, to the best of our knowledge, none of these studies have achieved results on large scales as our testbeds.
  • Most of the studies have handled specific targets/objectives for example temperature, humidity, energy consumption, and particulate matters. However, this research project aims to integrate them.

Potential Impact/Benefits to Target Sector 

Our research outcome will ensure HVAC system energy management is more effective and robust, and the novelty and significance of the proposed approaches are summarized as follows.

  1. Data multicollinearity and uncertainty aware
    • The proposed transformer-based dual-stage attention approaches can address the data multicollinearity within multiple systems and reduce the complexity of system modelling.
    • The proposed deep neural network (DNN)-based approach can address and express the data uncertainty due to measurement error or unseen scenarios.
  2. Self-learning control strategy
    • The proposed deep reinforcement learning-based approaches are effective in real environments, where many factors are involved and correlated, by learning important features from complex environments.
    • The proposed approaches improve the robustness of control strategies
    • To our best knowledge, there is no research on training agents based on the proposed predictive models and then deploying the proposed control strategies in the real world at a large-scale system as our testbed.
  3. Optimal cooperation systems
    • The proposed transfer learning approaches can transfer the knowledge owned by an agent of individual space (or signal system) to another agent of similar space (or system).
    • The proposed transfer learning approaches not only facilitate the integration of optimal control/operation of multiple systems but also improve the transferability of the proposed control strategies crossing sub-systems in HVAC systems.