Thermal Comfort Prediction Of Passenger Compartment Based On Deep Forest Method

The thermal comfort of the passenger compartment will affect the riding comfort of the passengers.In order to achieve satisfactory thermal comfort of the passengers,it is necessary to understand the thermal response state of the passengers in the vehicle to the thermal environment in the passenger compartment,and establish the thermal comfort prediction model of the passenger compartment.Based on the experimental data of the passenger compartment in summer and winter,the data set that conforms to the thermal environment of the passenger compartment was constructed through data sorting and feature engineering.Based on the deep forest method,the prediction model of the objective environmental parameters of the passenger compartment on the subjective score of human thermal sensation and thermal comfort was established.The main research contents are as follows:(1)The original experimental data were subjected to data standardization processing,such as coverage analysis,outlier detection and data smoothing analysis.Based on the response characteristics and statistical characteristics of the thermal environment of the passenger compartment,the input characteristics in line with the characteristics of the thermal environment of the passenger compartment and the experimental records were constructed;The sensitivity of input and output features and the correlation of observed data sets are analyzed based on grey correlation method,and the dimension of highly correlated features of data sets is reduced based on principal component analysis clustering.46 input features of data sets are integrated into 27 features,which simplifies the training difficulty of machine learning.(2)The decision tree and deep forest methods were used to build the thermal comfort prediction model of the passenger compartment;The accuracy,recall rate and kappa coefficient are used to evaluate the model for the two output characteristics of thermal sensation and thermal comfort.The results show that the classification accuracy of the decision algorithm for the output characteristics is less than 65%,and the output characteristics cannot be classified well;The classification accuracy of the deep forest algorithm for the output features is about 85%,which can better classify the output features.The deep forest method is selected as the machine learning method of the model;The performance of the deep forest method in the test set was observed,and it is found that the performance of the test set in summer is better than that in winter.(3)The performance of the universal deep forest was improved,in which the multi granularity scanning module adopts the head to tail splicing input feature method to strengthen the situation that there are many feature groups in the middle but weak on both sides after scanning;Xgboost,random forest and limit forest are used to reconstruct the linked forest module,and the confidence channel is established.When the confidence of the sample is greater than the confidence threshold,the jump is made,and the running time of the improved algorithm is reduced from 92 s to 59 s,which improves the performance of the deep forest model.(4)Establish a deep forest thermal comfort prediction model to assist CFD simulation.Through the training of CFD simulation data and deep forest prediction model,the accuracy of the prediction model is verified,and the accurate prediction of the thermal comfort of passengers in the vehicle under different thermal environment parameters is realized,which can provide a reference for the thermal environment parameter setting in the passenger compartment,and can effectively reduce the cost of real vehicle test and CFD simulation.