Research on indoor thermal sensation variation and cross-subject recognition based on electroencephalogram signals

• Published in: Journal of Building Engineering Vol. 76; p. 107305
• Main Authors: Zheng, Hanying, Pan, Liling, Li, Tingxun
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.10.2023
• Subjects: Deep learning; Electroencephalogram (EEG); Phase–amplitude coupling (PAC); Thermal sensation classification; Deep learning; Phase–amplitude coupling (PAC); Thermal sensation classification; Electroencephalogram (EEG)
• ISSN: 2352-7102; 2352-7102
• DOI: 10.1016/j.jobe.2023.107305

The accurate identification of the environmental assessment by occupants is an important factor in the smart home control system. To identify the neural characteristics associated with thermal sensation, this study conducted environments in five distinct thermal conditions based on predicted mean vote (PMV) values. When subjects perceived different thermal environments, the channels in alpha bands were obviously activated more than others, and Brodmann areas (BA) 9/21/22/44/45 were inferred to be involved in processing thermal sensation stimuli. Based on the neural signatures, the related brain activity mechanism was deduced. To improve the recognition accuracy of EEG and the applicability of deep learning frameworks, phase–amplitude coupling (PAC) was applied, and a new multi-layer PAC feature structure for cross-subject classification was proposed for the first time. The PAC features demonstrated significant differences under different thermal sensations, and the new feature structure, extracted from the significant area on the alpha band, achieved an accuracy of 92.0% when applied to LSTM networks, outperforming other feature structures. The findings offer valuable insights into the underlying mechanisms of thermal sensation by discussing the key frequency bands and brain regions that play a critical role, as well as the potential of PAC features. These advancements contribute to the practical application of EEG-based cross-subject thermal sensation classification in various domains, including but not limited to smart homes and healthcare. •Certain brain areas’ functions during thermal sensation generation were discussed.•Using phase–amplitude coupling to investigate thermal sensation is effective.•Multi-PAC features with deep learning frameworks improve cross-subject recognition.•PAC features from significant brain regions had better classification performance.