What question did this study set out to answer?

The aim is to enhance building energy management by improving occupancy detection methods using a multimodal framework.

March 21, 2026Open Access

Building energy management through smart occupancy based control

Key Points

The aim is to enhance building energy management by improving occupancy detection methods using a multimodal framework.
Developed a privacy-preserving occupancy detection framework integrating direct and ambient sensing.
Utilized a Random Forest classifier for occupancy identification based on multimodal data fusion.
Analyzed statistical relationships between occupancy and sensor inputs using parametric classification techniques.
Achieved high predictive performance with accuracy of 90.6%, precision of 1.00, and F1-score of 0.88.
Model stability improved under variable occupancy due to enhanced lighting data integration.
Results support energy savings and adaptive control of HVAC and lighting systems.

Abstract

Building energy consumption is dominated by lighting and heating, ventilation, and air conditioning (HVAC) systems, whose operation is often inefficient due to inadequate control strategies and unreliable occupancy detection. As occupancy patterns become increasingly dynamic, existing occupancy-based control models suffer from reduced robustness, particularly under variable occupant numbers and privacy constraints. This study addresses these limitations by proposing a privacy-preserving, multimodal occupancy detection framework that integrates direct sensing, ambient environmental sensing, and lighting-related parameters. An intelligent occupancy identification system based on a Random Forest (RF) classifier is developed to investigate how multimodal data fusion improves occupancy prediction accuracy and energy management performance. Parametric classification techniques are employed to analyze statistically significant relationships between occupancy states and sensor inputs. The optimized (RF), model achieves strong predictive performance, with accuracy, precision, recall, and F1-score values of 0.99, 1.00, 0.86, and 0.88, respectively. The optimized Random Forest model achieved an average testing accuracy of 90.6%, with recall, precision, and F1-score values of 0.86, 0.88, and 0.87, respectively, demonstrating robust predictive performance under variable occupancy conditions. Incorporating lighting data enhances model stability under varying occupancy conditions and enables adaptive control of HVAC and lighting systems, resulting in measurable energy savings. Overall, the results demonstrate that combining multimodal sensing with machine-learning-based modeling significantly improves occupancy prediction and supports intelligent building energy management. The proposed framework is applicable to real-time smart building systems and aligns with policy objectives aimed at reducing building energy consumption. Future work will extend the model to larger, more diverse datasets and multi-zone environments.

KI fragen

Bookmark

View Full Paper