Smart building with intelligent indoor lighting control system using reinforcement learning simulations

• Main Author: Chan, Keno Jia Nuo
• Other Authors: En-Hua Yang
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Environmental engineering
• Online Access: https://hdl.handle.net/10356/150183
• Institution: Nanyang Technological University

In most conventional indoor environments today, such as households or office buildings, the lighting control systems deployed incorporates digitalization through motion sensors or wireless communication technology to operate or control the lighting’s operation and brightness. Such utilization often leads to low user satisfaction rates as energy minimalization is more optimized to reduce energy wastage without consideration of users’ lighting preferences. Although the presence of wireless controls allows users to manually adjust the lighting intensity based on their preferences, this becomes a problem when there is more than one occupant in the environment due differences in lighting preferences. Hence, the most preferrable lighting control system is one that can consider the lighting preferences of all the occupants in the room and to some extent, minimize the energy wastage for lighting. This study aims to propose a reinforcement learning (RL) control system that can obtain a balance between lighting preferences of occupants and energy efficiency within the environment. Through this control system, the RL agents will optimize the lighting comfort based on the lighting preferences profiles of all occupants within the environment while the negotiator will maximize the lighting comfort between different occupants with different lighting preferences while minimizing energy consumption. The control agents will be trained by Q-learning, which is a model-free reinforcement learning algorithm, and simulated with three different lighting preference profiles from three different occupants. To obtain the ideal learning parameter of the proposed control system, different learning parameters, such as ε – greedy value, max steps per episode, learning rate α, and discount rate γ, will be used to test the ideal learning conditions for this study. In addition, to test the adaptability of the proposed control system, changes will be made to the environment, such as a change in the number of occupants in the environment, different starting lighting state of the environment, and accounting for energy efficiency. From the results, the proposed control system was able to reach the optimum lighting comfort after 116 simulation runs. This shows that the proposed control system was able to achieve good lighting comfort optimization performance and has a relatively efficient learning speed even with the introduction of environmental changes.