Gas Emission Prediction Of Large And Small Buses Based On Filter-Wrapper Method And GA?BP Neural Network

In view of the lack of research on the prediction of exhaust emissions of large and small buses,this paper takes large-scale diesel buses and small-scale gasoline buses as the research object.In order to build a relatively high-precision prediction model of bus gaseous emissions,this paper has carried out relevant research.The main research contents are as follows:(1)In view of the fact that the emission data of large and small buses are insufficient in the actual driving process,this paper uses the detection technology of portable vehicle emission test system.After making a reasonable plan and route,the actual road tests were carried out on the large and small buses under different road conditions in Changsha,Zhuzhou and Xiangtan of Hunan Province.Then,the collection of data for reasonable screening and optimization,that is,after the completion of the data preprocessing process,the preliminary establishment of a large and small bus gaseous emissions database.(2)The test results of large and small buses are analyzed.First of all,combining with the actual situation of road test,this paper deduces the formula of specific power for small bus and large bus respectively.By calculating their corresponding specific power values,the database of gaseous emissions of large and small passenger cars is further improved.Then,according to the specific power value of the large and small buses,the specific power distribution characteristics of the small buses and large buses under different road conditions are analyzed by cluster analysis.Finally,with the help of cluster analysis method,the law of CO₂,CO and NO_x emission factors changing with the specific power range of large and small buses under different road conditions is explored.(3)Based on the Filter-Wrapper method,the relevant parameters of gas emission of large and small buses are extracted and selected.Firstly,the speed,acceleration,specific power and fuel consumption in this database are first-order differentiated,second-order differentiated and first-order mobile differentiated respectively.The feature parameter set including original value,first-order difference,second-order difference and first-order moving difference is obtained.Then,the filter method based on Pearson correlation coefficient is used to quantify the correlation between different characteristic parameters and the target value.Then,for the established characteristic parameter sets of speed,acceleration,specific power and fuel consumption of large and small buses,the first two items with higher correlation coefficient are extracted respectively,and a new characteristic parameter set is established.Then,based on the encapsulation method of random strategy,the main four kinds of feature parameters in the new feature parameter set are arranged and combined.Then,a three-layer BP neural network prediction model including four hidden layer nodes is built.Finally,MSE is taken as the evaluation index.The best combination of characteristic parameters corresponding to the emission factors of CO₂,CO and NO_x of minibus is selected.(4)A comprehensive comparison is made between the emission prediction model of large and small buses based on genetic optimization BP neural network and the emission prediction model of large and small buses based on support vector regression.First of all,according to the best feature parameter input combination corresponding to CO₂,CO and NO_x emission factors of large and small passenger cars,the BP neural network model of large and small passenger car exhaust prediction optimized by genetic algorithm is built respectively,and verified by some test data.Then,according to the best combination of feature parameters one by one,the parameters of SVR are optimized by grid search method based on cross validation.Based on the support vector regression,the prediction model of the large and small bus's gas state is established and verified by the same test data.Finally,taking MSE,RMSE and Pearson correlation coefficient as evaluation indexes,the prediction resu lts of the two models are compared and analyzed.The results show that the prediction model based on GA-BP neural network has more advantages in the instantaneous prediction level than that based on support vector regression algorithm.In the overall error level,the best relative errors of CO₂,CO and NO_x emission factors of minibus are 0.00579%,4.427%and 32.276%respectively.On the whole error level,the best relative errors of CO₂,CO and NO_x emission factors are 0.02029%,0.0317%and 0.01987%,respectively.The method described in this paper has a good effect on the overall prediction of gaseous emissions of large and small buses.It provides some reference and guidance for the prediction of vehicle gaseous emissions,and has certain theoretical research significance and engineering application value.