TY - GEN
T1 - UGAS
T2 - 10th IEEE International Conference on Smart Internet of Things, SmartIoT 2026
AU - Guo, Mingkun
AU - Ma, Jiuwu
AU - Gao, Zhiqiao
AU - Ma, Teng
N1 - Publisher Copyright:
© 2026 IEEE.
PY - 2026
Y1 - 2026
N2 - Hazardous gas leak source localization is a core requirement for safety monitoring in industrial Internet of Things (IoT) systems. Existing solutions rely on either densely deployed static sensor networks or mobile robot-based search strategies, yet both paradigms suffer from inherent limitations. Specifically, static sensor networks incur drastic degradation in localization accuracy under sparse deployment, alongside limited scalability, while mobile robot search exhibits low search efficiency. To address these bottlenecks, we propose a static-dynamic collaborative source localization framework for IoT gas sensing networks. A deep learning algorithm is designed to produce uncertainty estimates and convert them into active decision signals that drive mobile robots, establishing a collaborative paradigm of "static coarse localization → uncertainty-guided dynamic supplementation → closed-loop iteration."To handle the dynamic sensor topology and irregular temporal sampling in this collaborative setting, the algorithm uses permutation-invariant attention for spatial encoding and a continuous-time model for temporal encoding. In a 10×10×10 m room, the configuration of 3×3×3 static sensor network with 1 mobile platform achieves a source localization accuracy of 0.14 m, outperforming the 4×4×4 static sensor network and delivers a 12-fold accuracy improvement over traditional baseline methods.
AB - Hazardous gas leak source localization is a core requirement for safety monitoring in industrial Internet of Things (IoT) systems. Existing solutions rely on either densely deployed static sensor networks or mobile robot-based search strategies, yet both paradigms suffer from inherent limitations. Specifically, static sensor networks incur drastic degradation in localization accuracy under sparse deployment, alongside limited scalability, while mobile robot search exhibits low search efficiency. To address these bottlenecks, we propose a static-dynamic collaborative source localization framework for IoT gas sensing networks. A deep learning algorithm is designed to produce uncertainty estimates and convert them into active decision signals that drive mobile robots, establishing a collaborative paradigm of "static coarse localization → uncertainty-guided dynamic supplementation → closed-loop iteration."To handle the dynamic sensor topology and irregular temporal sampling in this collaborative setting, the algorithm uses permutation-invariant attention for spatial encoding and a continuous-time model for temporal encoding. In a 10×10×10 m room, the configuration of 3×3×3 static sensor network with 1 mobile platform achieves a source localization accuracy of 0.14 m, outperforming the 4×4×4 static sensor network and delivers a 12-fold accuracy improvement over traditional baseline methods.
KW - Evidential Deep Learning
KW - Gas Source Localization
KW - Hybrid Gas Sensor Network
KW - IoT Sensing
UR - https://www.scopus.com/pages/publications/105051326549
U2 - 10.1109/SmartIoT70864.2026.00009
DO - 10.1109/SmartIoT70864.2026.00009
M3 - Conference contribution
AN - SCOPUS:105051326549
T3 - Proceedings - 2026 IEEE International Conference on Smart Internet of Things, SmartIoT 2026
SP - 1
EP - 8
BT - Proceedings - 2026 IEEE International Conference on Smart Internet of Things, SmartIoT 2026
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 21 August 2026 through 24 August 2026
ER -