Occupancy modelling using data driven models

• Main Author: Hu, Yun Nan
• Other Authors: Soh Yeng Chai
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering
• Online Access: https://hdl.handle.net/10356/176475
• Institution: Nanyang Technological University

The Final Year Project aims to improve building energy efficiency by developing and applying data-driven models for precise occupancy modeling. The project uses multiple types of machine learning algorithms to predict occupancy levels under multi-occupant single-zone (MOSZ) and multi-occupant multi-zone (MOMZ) scenarios. The goal is to improve energy consumption in HVAC (heating, ventilation, and air conditioning) systems and lighting in commercial buildings. A dataset obtained from the School of Electrical and Electronic Engineering office at Nanyang Technological University was used in the study. Decision Trees, Support Vector Machines, Principal Component Analysis, Convolutional Neural Networks, and Long Short-Term Memory (LSTM) networks were among the algorithms tested. The combination of LSTM with Principal Component Analysis was the most significant performance; it surpasses other models by obtaining the best accuracy, emphasizing the significance of dimensionality reduction in improving prediction performance. This research demonstrates how machine learning can be used to improve building energy management systems. It also sets a standard for further research in the same area. The project highlights the importance of having sufficient datasets and investigating different machine learning approaches.