基于机器学习的热舒适投票(TCV)预测  

Thermal Comfort Voting(TCV)Prediction Based on Machine Learning

作　　者：王孔迪 鹿世化[1] WANG Kongdi;LU Shihua

机构地区：[1]南京师范大学能源与机械工程学院,江苏南京210046

出　　处：《煤气与热力》2024年第6期V0006-V0011,共6页Gas & Heat

摘　　要：以某大型商场作为调研场所,对商场内空气状态参数、空气品质参数、光环境参数、声环境参数进行实测,对受试者个人参数进行实测调研。运用机器学习算法(决策树、逻辑回归、支持向量机、随机森林)建立热舒适投票预测模型,对单输入参数、多输入参数对预测模型预测性能影响进行分析。将预期平均评价模型参数作为单输入参数时,4种预测模型的预测精度均不理想,预测结果决定系数变化范围为0.43~0.54,波动比较大。最好成绩是决策树预测模型在空气相对湿度作为单输入参数时。与其他单输入参数相比,空气温度、空气相对湿度作为单输入参数时,4种预测模型的预测性能比较好。当空气流速、活动代谢率作为单输入参数时,4种预测模型的预测性能比较差。与单输入参数相比,以预期平均评价模型参数作为多输入参数时,预测模型的预测性能并未得到有效提升。将输入参数扩大至所有调研参数后,4种预测模型的预测性能均有所提高。空气状态参数、个人参数、空气品质参数的输入参数组合可以提升预测模型对热舒适投票的预测性能。Taking a large shopping mall as the research site,the air state parameters,air quality parameters,light environment parameters and sound environment parameters inside the mall were measured,and the personal parameters of the participants were measured and surveyed.Machine learning algorithms(decision tree,logistic regression,support vector machine and random forest)are used to establish a thermal comfort voting prediction model,and the impact of single input parameters and multiple input parameters on the prediction performance of the prediction model is analyzed.When the expected average evaluation model parameters are used as single input parameters,the prediction accuracy of the four prediction models is not ideal,and the determination coefficient of the prediction results varies from 0.43 to 0.54,with significant fluctuations.The best performance is when the decision tree prediction model has air humidity as a single input parameter.Compared with other single input parameters,when air temperature and air relative humidity are used as single input parameters,the prediction performance of the four prediction models is better.When air velocity and active metabolic rate are used as single input parameters,the prediction performance of the four prediction models is poor.Compared with the single input parameters,the prediction performance of the prediction model is not effectively improved when the expected average evaluation model parameters are used as the multiple input parameters.After expanding the input parameters to all survey parameters,the prediction performance of all four prediction models is improved.The combination of input parameters,such as air state parameters,personal parameters and air quality parameters can improve the prediction performance of the prediction model for thermal comfort voting.

关 键 词：机器学习 热舒适投票 预测 输入参数 

分 类 号：R126.8[医药卫生—环境卫生学]