Towards greener, smarter and healthier office building management via cognitive digital twin

Abstract

In this thesis, we propose an AI-driven digital twin solution to empower and drive greener, smarter and healthier buildings in the tropical environment (e.g., Singapore). Our solution aims to leverage the existing off-the-shelf hardware (e.g., sensors, wearables, etc.) and synergistically combine an emerging digital twin paradigm with the AIoT technologies, for digitalizing and optimizing the HVAC control strategy via a deep learning-based approach. We plan to optimize the HVAC system of buildings from the following three perspectives, energy efficiency, occupant comfort and behaviour study.

The first challenge is the implementation of the HVAC control policy because the changes of the thermal states in a building environment are influenced by various factors. The relationships among these influencing factors are hard to model and are always different in different building environments. To address this challenge, we propose a deep reinforcement learning based framework, DeepComfort, for thermal comfort control in buildings. We validate the performance of DeepComfort in simulation environment and the experimental results show that our approaches can improve the performance of thermal comfort prediction by 14.5% and reduce the energy consumption of HVAC by 4.31% while improving the occupants' thermal comfort by 13.6%.

The second challenge we meet with is the modeling of human behaviour. We want to derive a novel human behaviour influence mechanism to motivate occupants' behavioural changes to form sustainable energy-saving habits. Hence, we conduct national surveys, human behavioural experiments, and AI-based deep analytics to build the online personalized behaviour influence models for energy-saving behaviour studies. The national survey helps us better understand people's orientations and perceptions of energy-saving behaviour, and we get insights in both social and psychological ways from the survey. We upgrade the HVAC system of the experiment lab, and design the new experiment procedure for data collection and building influence model. We also design the problem formulation of our behaviour influence model. However, due to the explosion of COVID-19, our original plan was delayed because large gatherings are strictly not allowed. As a result, the experiment starts in December 2020 and it is not finished yet. In the mean time, we extend our research of smart building in the area of health. We propose a novel method for respiratory symptoms detection by using the built-in sensors of smart device, and we verify the method getting a quite accurate result on a small scale dataset we collected ourselves in office environment.