Indoor occupancy estimation

• Main Author: Soh, Agnes Xiao Xi
• Other Authors: Soh Yeng Chai
• Format: Final Year Project
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/67728
• Institution: Nanyang Technological University

In most developed countries, the large amount of energy consumption and unnecessary wastage of energy are contributing to climate changes and high energy costs, hence being energy efficient has become increasingly important. In the past, motion sensors have been invented and were widely used for the detection of occupants’ presence until today. However there are limitations, one of it being that motion sensors are dependent on the motion of the occupants and motion sensors are not able to detect the definite number of occupants in a space. Knowing the actual number of occupants in a space is essential for optimizing the performance of Heating, Ventilation and Air Conditioning (HVAC) systems. One way to determine the occupancy levels is to measure the environmental parameters such as carbon dioxide (CO2), temperature, pressure and humidity levels. The environmental sensors used to measure these environmental parameters are not expensive and the privacy of the occupants would also not be intruded. From the data collected via these sensors, it is necessary to extract advantageous features and select the most pertinent features that will be able to return occupancy information. The filter-model-based feature selection approaches such as Relative Information Gain (RIG) and Symmetric Uncertainty (SU) used in previous works has shown promising results. In this work, Correlation Coefficient Clustering which is also a filter based feature selection method will be used for feature selection to select the most relevant features. To the best of the author’s knowledge, this kind of feature selection technique has not been implemented in an occupancy estimation problem. The most relevant features selected are then used to train and test three kinds of machine learning classifiers, namely the Support Vector Machines (SVMs), Decision Trees and Naive Bayes. A comparison among these three classifiers is also made. The occupancy estimation accuracy results of these three classifiers are comparably competitive. The Naive Bayes classifier has shown a less noisy and much smoother result giving the highest accuracy of 68.01% out of the three classifiers.