Research On Thermal Comfort Prediction Of Elderly Based On Improved Random Forest

With the deepening of China's aging population,the demand for elderly buildings is increasing,and the elderly are also constantly pursuing high-quality old-age living environment.Compared with other age groups,the physiological function and the body immunity of the elderly are gradually declining,and the metabolism is slowing down,which makes the elderly exhibit higher requirements for the living environment.At present,there are relatively few researches on the indoor thermal environment of elderly buildings and the thermal comfort of the elderly in China,and the factors involved in thermal comfort prediction for the elderly are far more than those involved in classical indoor thermal comfort evaluation.Therefore,there is a significant deviation in using the current indoor thermal environment evaluation standard to assess and predict the thermal comfort of the elderly.How to multi-dimensionally and accurately evaluate the thermal environment of the elderly room and predict the thermal comfort of the elderly has become an urgent problem to be solved to create a pleasant thermal environment for the elderly building.For this reason,the thesis starts from finding a more accurate method to predict the thermal comfort of the elderly,and proposes a theoretical model and method for thermal comfort of the elderly based on improved random forest algorithm.In view of the shortcomings of the random forest algorithm,it is mainly improved from two aspects: First,the research improves the node splitting algorithm and proposes a hybrid algorithm that combines the CART algorithm and the C4.5 algorithm node splitting rules to form a linear function,and assigns the weight of the decision tree as the adaptive coefficient of the hybrid algorithm based on the prediction accuracy of the two algorithms.Second,the classification voting method is enhanced,that is,to give the corresponding weight to the voting ability when classifying the decision tree in the forest,so that the decision tree has the voting ability corresponding to its growth effect.Through the improvement of the random forest algorithm to improve the applicability and prediction objectivity of the algorithm applied to the prediction of thermal comfort for the elderly,and provide a thermal comfort prediction method for elderly people with multi-dimensional information.Taking Hangzhou as an example,this thesis used the prediction method for thermal comfort of the elderly to conduct research,and preliminary analysis of the factors influencing the thermal comfort of the elderly.On-site testing and subjective questionnaire surveys were applied to obtain relevant data on the thermal comfort of the elderly in elderly buildings in Hangzhou,and to build a prediction database for the thermal comfort of the elderly in Hangzhou.The constructed database is preprocessed,including continuous data discretization,discrete feature assignment,class imbalance processing,etc.,and through correlation analysis and feature importance analysis,refine and form an index system for predicting the thermal comfort of the elderly and determine the quantitative weight of each index.The adaptability coefficient of the node hybrid splitting algorithm was determined through cross experiments.Then through multiple experiments,the optimal decision tree number(ntree)and the best random characteristic variable value(mtry)of the model were obtained by using the OOB error rate as the evaluation criterion.Class imbalance processing was performed on the data to enhance the classification performance of the model.The performance of the model employed in this thesis was compared with the prediction performance of the PMV model,decision tree algorithm,and traditional random forest algorithm.The comparison results revealed that the model of this paper displayed significant advantages in confusion matrix analysis,prediction accuracy,accuracy,and recall rate,etc.The research in this thesis could provide theoretical and methodological references for thermal comfort prediction of the elderly and furnish specific guidance for thermal environment design of the elderly.