基于深度学习与热成像技术的热舒适度预测方法研究OA
Research on thermal comfort prediction method based on deep learning and thermal imaging technology
随着电气化和智能化的发展,对以人为中心的空调系统提出了更高要求.基于图像识别技术,构建一个以ResNet50为骨干网络,并融入空间注意力模块(Attention-ResNet50)的深度学习模型,用于以人为中心的空调系统.通过采集不同热舒适条件下的红外图像数据并分类为冷、舒适、热三类进行训练与测试,该方法显著提升了预测精度.试验结果表明,引入空间注意力机制后模型整体准确率提升约5%,其中对男性受试者的最佳模型准确率超过94%,对女性受试者的最佳模型准确率超过96%.研究证实了注意力机制在特征提取中的有效性,为个性化热舒适调控提供了可靠且普适的技术路径.
With the advancement of electrification and intelligent technologies,higher demands have been placed on human-centric air conditioning systems.Based on image recognition technology,a deep learning model using ResNet50 as the backbone network and incorporating a spatial attention module(Attention-ResNet50)was constructed for human-centric air conditioning systems.By collecting infrared image data under different thermal comfort conditions and classifying them into three categories,cold,comfort,and hot,for training and testing,this method significantly improved prediction accuracy.Experimental results showed that the introduction of the spatial attention mechanism increased the overall model accuracy by approximately 5%.The best model accuracy exceeded 94%for male subjects and surpassed 96%for female subjects.The study confirms the effectiveness of the attention mechanism in feature extraction,providing a reliable and universal technical pathway for personalized thermal comfort control.
严锦君;胡家云;王佳韵;李康
中国三峡新能源(集团)股份有限公司,北京 101100上海理工大学 新能源车辆热管理综合评估实验室,上海 200093||上海理工大学 能源与动力工程学院,上海 200093上海理工大学 新能源车辆热管理综合评估实验室,上海 200093||上海理工大学 能源与动力工程学院,上海 200093上海理工大学 新能源车辆热管理综合评估实验室,上海 200093||上海理工大学 能源与动力工程学院,上海 200093
信息技术与安全科学
深度学习热成像注意力集中机制个性化热舒适性模型
deep learningthermal imagingattention mechanismpersonalized thermal comfort model
《流体机械》 2026 (2)
100-108,9
国家自然科学基金面上项目(52376203)
评论