Optimising energy efficiency in Higher Education Institution buildings through denoise extreme learning machine occupancy estimation
Year: 2026
Type: dissertation
Abstract: Higher education institutions (HEIs) face significant challenges related to energy inefficiency during operational stage, with heating, ventilation, and air conditioning (HVAC) systems accounting 40% of total energy consumption. However, HVAC systems in HEI buildings often operating inefficiently. To investigate the underlying issues and identify potential solutions, this research employed an extensive literature review and a phenomenological study. The analysis revealed that energy inefficiencies in HVAC management are primarily caused by discrepancies between estimated and actual occupancy, resulting in suboptimal indoor conditions that impact health, comfort, and learning outcomes. The findings suggest that integrating occupancy-based control strategies, such as predictive control model (PCM), into HVAC systems can significantly reduce energy consumption. However, accurately estimating real-time occupancy in higher education buildings remains a challenge due to complex and dynamic patterns. Therefore, this research developed, tested, and validated a new DELM (denoise extreme learning machine) algorithm for accurate real-time occupancy estimation using only occupancy data and CO2 measurements. This algorithm aims to enhance student health, comfort, and performance while reducing energy consumption and the associated economic cost and CO2 emissions. Thus, this research is timely and relevant amidst current efforts to reduce energy consumption and improve well-being. DELM is a variation of the standard extreme learning machine (ELM) algorithm, which combines extreme learning machine (ELM) and denoising stacked autoencoder (DSAE) to effectively remove noise from CO2 data and achieve high estimation accuracy while maintaining computational efficiency. The DELM algorithm’s ability to accurately estimate occupancy was validated across four different scenarios of use and number of occupants, demonstrating its accuracy in those scenarios, achieving 91% accuracy (15% improvement compared to ELM) in 300-person conference rooms; 93% accuracy (18% improvement) in 100-person conference rooms; and 98% accuracy in a 25-person office. The DELM algorithm outperforms ELM in stability and accuracy. This study also confirms that using CO2 concentration data can result in accurate occupancy estimation.
Author: Sun, Zhuoqun
Institution: University of the West of England
Language: English
Subfield: Building and Construction
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204288454
