Research On Independent Thermal Comfort Discriminant Model For Intelligent Indoor Temperature Regulation

The individual thermal comfort discriminant model established based on electroencephalogram(EEG)signals can evaluate the changes of human thermal sensation in real time,based on it,an indoor thermal environment that accurately matches the individual occupants’ feelings can be established,which constitutes a potential brain-computer interaction smart home application scenario.However,this type of research is still in its infancy.Most of the models reported in the literature are based on the assumption that the use conditions are unchanged.Few studies have explored the applicability of the models under the changed use conditions,and there are also very few studies have conducted online experimental verification.Therefore,this article focuses on the generalization performance of the individual thermal comfort discrimination model based on EEG signals,and tries to establish a highly applicable thermal comfort discrimination model.First of all,this article analyzes the EEG signals of 22 subjects under the experimental conditions of overheating and thermal comfort.The typical methods of shallow learning and deep learning are used to establish non-individually dependent thermal comfort discrimination models,and the recognition rates are evaluated by cross-validation with a single subject as a unit.The results show that the highest accuracy of the models based on the classic shallow learning is only 60.65%,while the accuracy of the deep learning model based on the convolutional neural network(CNN)has been significantly improved(p<0.01),reaching 70.71%.The research shows that individual differences will have a more serious impact on the discriminant model,and CNN can more effectively extract the EEG features-related thermal sensation that have strong consistency between individuals,so that the thermal comfort discriminant model can be applied to non-modeled subjects.Furthermore,this article uses the established non-individual dependent thermal comfort discrimination model to build a set of intelligent room temperature adjustment system,which can automatically control the air conditioner by predicting the subjects’ thermal sensations in real time based on the EEG signals.Using different indoor heat radiation conditions to simulate normal and high heat summer environment respectively,the system was used to conduct online room temperature adjustment experiments with11 new subjects.The experimental results show that the system can automatically control the air conditioner to adjust the room temperature according to the subjects’ thermal sensations.During the two groups of experiments,the subjects’ average thermal comfort subjective score was significantly reduced from about 2 at the beginning to about 0.5 at the end(p<0.001),that is,from a moderate degree of "thermal uncomfortable" to close to "comfortable".The system can also adapt to changes of environmental.Under the experimental condition that simulate the high heat summer,the average single cooling time and total cooling time of the air conditioner are significantly higher than the experimental condition that simulate the normal summer(p<0.05).Finally,this article also tries to study the applicability of the thermal comfort discriminant model in sleep state.The overheating/thermal comfort EEG experiment was performed on 7 subjects in awake/sleeping state.The results show that the thermal comfort discriminant model established by CNN based on awake EEG signals can still maintain a certain degree of discriminative ability in sleep state,and there is a significant negative correlation between the model outputs and the changes of room temperature(p<0.05).The work of this article preliminarily confirmed the feasibility of establishing a highly applicable EEG thermal comfort discrimination model through deep learning and other methods,and provided a technical foundation for the further development of smart home technology based on brain-computer interaction.