ENHANCING INDOOR OCCUPANCY ESTIMATION WITH HYPERPARAMETER OPTIMISATION OF ARTIFICIAL NEURAL NETWORKS

Abstract

Buildings consume a significant amount of energy. Occupancy and occupant behaviour in buildings have a significant impact on energy consumption. Energy consumed in buildings can be saved using occupancy-based control of devices. Occupancy estimation can be done automatically using video or images or using data from sensors. The latter method is more feasible economically and applicable due to its non-intrusive nature. Machine learning techniques and Artificial neural networks (ANNs) are commonly used in occupancy estimation. This research was focused on enhancing the indoor occupancy estimation accuracy with hyperparameter optimisation in ANNs. A method and tool developed for finding the best performing ANN with the most accurate input hyperparameters is utilised to enhance the performance of the ANNs for the indoor occupancy estimation problem. The results demonstrated that high accuracy rates of 99.8% can be achieved with this method. © 2023, Scibulcom Ltd.. All rights reserved.