基于决策树模型的居住建筑人员热舒适预测

Human thermal comfort prediction in residential buildings based on decision tree model

传统热舒适模型由于确定的变量引入降低了其预测精度,而神经网络等模型由于其复杂的运算规则限制了对模型的解释能力。通过引入分类应用中的经典C&RT决策树方法,对夏热冬冷地区6个城市自由运行居住建筑中的人员热舒适及影响因素和权重进行了预测分析,建立了以城市为第一分类特征的热感觉预测决策树模型,解释了这些因素之间的内部逻辑,并较好地识别了影响热感觉的关键因素,从而为热舒适的预测评价和热环境营造提供了新的方法参考。

The established variables reduce the prediction accuracy of traditional thermal comfort models, while the models such as artificial neural network limit to be explained well due to their complicated operation rules. Presents a typical classification tree--C＆RT model to analyse the occupants＇ thermal comfort and the related influencing factors. Based on the field survey of free-running residential buildings in six cities （Chengdu, Chongqing, Wuhan, Nanjing, Changsha, Hangzhou） in hot summer and cold winter zone, develops a decision tree model of thermal sensation by choosing the city as the first classification feature. The results show that the obtained decision tree model has good performance to identify the key factors affecting on thermal comfort and explain the internal logic between these factors, which provides a new approach for predicting thermal comfort and designing thermal environments.