DATA-DRIVEN OCCUPANCY DETECTION BY EMPLOYING DEEP AND TRANSFER LEARNING APPROACHES

Abstract

Building occupancy information could aid energy preservation while maintaining the end-user comfort level. Energy conservation becomes essential since energy resources are scarce and human dependency on appliances is only exponentially increasing. While the usage of intrusive sensors (i.e., cameras and microphones) can raise privacy concerns, this thesis presents an innovative non-intrusive occupancy detection approach using environmental sensor data (e.g., temperature, humidity, Carbon Dioxide (CO2) and light sensors). The proposed scheme transforms multivariate time-series data into images for better encoding and extracting relevant features. The utilized image transformation method is based on data normalization and matrix conversion. By representing time-series in 2D space, an encoding kernel can move in two directions while moving only in one direction when applied to a 1D signal. Moreover, Machine Learning (ML) and Deep Learning (DL) techniques were utilized to classify occupancy patterns. Several simulations are used to evaluate the approach; mainly, we investigated pre-trained and custom Convolutional Neural Network (CNN) models. The custom CNN attained an accuracy of 98.98%. Additionally, pixel data are extracted from the generated images and subjected to traditional ML methods. Throughout the numerous comparison settings, it was observed that the latter strategy provided the optimal balance of 99.42% accuracy performance and minimal training time across the occupancy datasets. Further, the thesis proposes a real-time occupancy detection method using non intrusive ambient data and a DL model. An environmental sensing board gathered temperature, humidity, pressure, light, motion, sound, and CO2 data. The detection approachwas deployed on an edge device, enabling low-cost computing while increasing data security. The system was set up at a university office as the primary case study testing location. We analyzed two CNN models to confirm the optimum alternative for edge deployment. A 2D-CNN technique was used for one day to identify occupancy in real-time. The model proved robust and reliable, with a 99.75% real-time prediction accuracy.