Leveraging Machine Learning for Identifying Occupancy Patterns from Power Data with a Moving Window Feature Extraction Method

Accurate and timely occupancy detection is pivotal for achieving energy efficiency in buildings, as it enables the fine-tuning of heating, cooling, lighting, and other systems, ultimately reducing energy waste and environmental impact while enhancing user comfort and cost savings. Leveraging power consumption data for occupancy detection presents a compelling advantage, as it offers a non-intrusive, cost-effective, and readily available solution, eliminating the need for extensive sensor or camera installations and enabling a more scalable and practical approach to building management. This paper presents an innovative approach for identifying occupancy patterns through utilizing machine learning (ML) techniques applied to power consumption data. We analyze two datasets, each representing diverse real-world scenarios, to uncover hidden occupancy patterns and their temporal variations. To enhance the discriminative power of the model, we incorporate a moving window methodology, which allows us to extract valuable features such as the average, standard deviation (SD), and range over specific time intervals. The motivation behind employing ML approaches in this context stems from the growing need for energy-efficient building management and the potential for ML to capture intricate temporal relationships within power consumption data. The results of our study offer valuable insights for applications in intelligent buildings, energy conservation, and occupancy-aware systems.