Wearable sensors applied to Personal Comfort Systems (PCS) can enable advanced and energy-efficient indoor environmental management solutions. Different physiological signals are commonly used to measure the human thermal comfort. Among them, the EEG is one of the most promising given its strong relationship with mental state, that is correlated to thermal perception. EEG brainwaves are used as features to detect human thermal comfort. However previous research did not turn out to identify a set of features with the same level of correlation with the thermal comfort. The reason of that relies on the fact that different experiments used different references, such as thermal comfort, thermal sensation, or environmental temperature. Moreover, brainwaves are influenced also by other human factors that lead to low accuracies. This paper presents a novel approach for EEG signal processing, coupled with a new labelling method of human thermal perception. A commercial 4-channels EEG wearable device has been used to collect the brain activity of 52 subjects exposed to different thermal conditions (cold, neutral and warm). The subjective thermal perception has been also collected using a common questionnaire based on 3 scores: thermal comfort, thermal sensation and thermal preference. After signal pre-processing, a short-time windowing technique has been applied to the cleaned EEG signal and a set of statistical, time-frequency, log-covariance and other features have been calculated to analyse the signal pattern. The dataset has been labelled using a parameter that is the result of a permutation and combination analysis applied to the 3 scores collected with the questionnaires. Finally, an artificial neural network (multilayer perceptron - MLP) has been trained to predict thermal label using the EEG signal as unique input. The model achieved an accuracy of 90% in the prediction of 5 different levels of thermal perception, demonstrating the high potential of the proposed approach.
Development and Application of EEG Signal Pattern Analysis and Artificial Neural Network for Indoor Comfort Measurement
Arnesano, Marco
2024-01-01
Abstract
Wearable sensors applied to Personal Comfort Systems (PCS) can enable advanced and energy-efficient indoor environmental management solutions. Different physiological signals are commonly used to measure the human thermal comfort. Among them, the EEG is one of the most promising given its strong relationship with mental state, that is correlated to thermal perception. EEG brainwaves are used as features to detect human thermal comfort. However previous research did not turn out to identify a set of features with the same level of correlation with the thermal comfort. The reason of that relies on the fact that different experiments used different references, such as thermal comfort, thermal sensation, or environmental temperature. Moreover, brainwaves are influenced also by other human factors that lead to low accuracies. This paper presents a novel approach for EEG signal processing, coupled with a new labelling method of human thermal perception. A commercial 4-channels EEG wearable device has been used to collect the brain activity of 52 subjects exposed to different thermal conditions (cold, neutral and warm). The subjective thermal perception has been also collected using a common questionnaire based on 3 scores: thermal comfort, thermal sensation and thermal preference. After signal pre-processing, a short-time windowing technique has been applied to the cleaned EEG signal and a set of statistical, time-frequency, log-covariance and other features have been calculated to analyse the signal pattern. The dataset has been labelled using a parameter that is the result of a permutation and combination analysis applied to the 3 scores collected with the questionnaires. Finally, an artificial neural network (multilayer perceptron - MLP) has been trained to predict thermal label using the EEG signal as unique input. The model achieved an accuracy of 90% in the prediction of 5 different levels of thermal perception, demonstrating the high potential of the proposed approach.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
