文献类型：期刊文献

英文题名：Overcoming selectivity challenges in single gas sensor leveraging temperature modulation and multi-feature fusion

作者：Mei, Haixia[1,3];Peng, Jingyi[1];Wang, Tao[2];Wang, Lijie[3];Zhang, Bowei[2];Xuan, Fuzhen[2]

机构：[1]Changchun Univ, Key Lab Intelligent Rehabil & Barrier Free Disable, Minist Educ, Changchun 130022, Peoples R China;[2]East China Univ Sci & Technol, Sch Mech & Power Engn, Shanghai Key Lab Intelligent Sensing & Detect Tech, Shanghai 200237, Peoples R China;[3]Jilin Univ, Coll Elect Sci & Engn, State Key Lab Integrated Optoelect, Changchun 130012, Peoples R China

年份：2025

卷号：522

外文期刊名：CHEMICAL ENGINEERING JOURNAL

收录：;EI(收录号：20253519089678);WOS:【SCI-EXPANDED(收录号:WOS:001568244600002)】；

基金：This work was supported by the Jilin Provincial Science and Tech-nology Development Program Project (YDZJ202501ZYTS591) , National Natural Science Foundation of China (62301314) , and the Science Fund for Creative Research Groups of the National Natural Science Foundation of China (52321002) . This work also received support from the In Situ Devices Center of East China Normal University.

语种：英文

外文关键词：Electronic nose; Temperature modulation; Machine learning; Feature fusion; Anti-interference

摘要：Cross-sensitivity in complex atmospheric environments has been a shortcoming of metal oxide gas sensors. The pulsed driven mode enables rapid temperature modulation for MEMS gas sensors, which offers a novel approach to overcome the sensitivity challenges. In this study, the capacity of a single MEMS gas sensor is extended by leveraging three temperature modulations to identify ten volatile organic compounds with concentrations ranging from 10 to 50 ppm. A multi-task neural network based on multi-feature fusion (MTFNet) is developed to integrate the temporal, envelope, and detail features of the signal. The results demonstrated that the pulsed heating mode combined with the MTFNet achieved the best gas recognition performance, with a classification accuracy of 99.19 % and an R2 value of 94.08 %. Furthermore, the gas recognition accuracy remained as high as 96.60 % even under various interference conditions including four kinds of abnormal samples based on ethanol gas are drift, missing value, spikes and intrusion. Model interpretability analysis was also conducted to reveal the feature dependencies and enhance the transparency of the recognition process. Valuable insights are provided for developing high-performance, interference-resistant, and low-power intelligent electronic nose systems.