Abstract
This paper presents a reliable, scalable, and transferable framework to predict occupancy in a building utilizing diverse, multi-modal information. We propose a new methodology for learning-driven occupancy detection built on the concepts of probabilistic graphical modeling and observable Markov chain modeling. To capture the relationship between multi-sensor data and occupancy, we propose this Occ-STPN framework that is flexible to support both multivariate and univariate formulations. While the multivariate Occ-STPN performs feature-level fusion of multiple predictors and occupancy time-series data, the univariate Occ-STPN involves decision fusion of occupancy predictions using individual predictors based on a mutual information weighted fusion scheme. We also propose a new metric to evaluate the performance of occupancy prediction algorithms. Two popular datasets are used to validate our approach and demonstrate that our framework is scalable in terms of the number of information sources (e.g., sensors) as well as it is possible to transfer trained models from one building to another without significant reduction in performance. Reliability of the algorithm is also tested by injecting noise into the datasets.
Original language | American English |
---|---|
Pages | 5884-5889 |
Number of pages | 6 |
State | Published - 2019 |
Event | 2019 American Control Conference (ACC) - Philadelphia, Pennsylvania Duration: 10 Jul 2019 → 12 Jul 2019 |
Conference
Conference | 2019 American Control Conference (ACC) |
---|---|
City | Philadelphia, Pennsylvania |
Period | 10/07/19 → 12/07/19 |
Bibliographical note
See NREL/CP-5000-73359 for preprintNREL Publication Number
- NREL/CP-5D00-75055
Keywords
- building occupancy
- Occ-STPN framework
- spatiotemporal