Occupancy modelling using data driven models

Abstract

Knowing the occupancy profile is useful in the efficient control of Heating, Ventilation and Air-conditioning systems, allowing significant energy savings. The non-intrusive aspect of environmental sensors makes them popular, and they are ubiquitous in modern buildings. From the original time-domain CO2 dataset, different feature engineering steps are applied. As the accuracy of occupancy estimation can be improved by using effective feature engineering methods. Various statistical and different domain features will be computed then concatenated for machine learning. Visualizing the different features obtained to assess its value and usefulness in telling different occupancy levels apart. The approach using other domain representations of the signal is popular in signal processing research, but less so in the field of occupancy estimation and modelling. Different datadriven machine learning algorithms will be used to train and then test on the partitioned feature set. Two separate experiments called Test 1 and Test 2 will be done. With the Test 1 feature set using the original CO2 data. Unlike Test 1, Test 2 will use a mean smoothed CO2 data and add another feature analysis method. The models with the best accuracies will be presented and discussed.