| The automotive industry is an important pillar of the national economy.The impact of automobile emissions on air quality is becoming increasingly apparent.Conducting research on automotive emission performance models is of great significance for reducing carbon emissions and protecting the environment.In order to study emission performance models,this thesis designs a gasoline emission performance modeling method based on the Py Torch framework.A model for emissions from a certain turbocharged PFI engine+CVT gearbox equipped vehicle is established,and the optimal model topology structure and hyperparameter configuration applicable to this model are obtained through experiments.Additionally,a system for processing and modeling emission data has been developed.The specific contents are as follows:(1)This thesis designed and developed a software program for processing and modeling emission data using Python.Based on the induction of the mechanism of machine learning algorithm to establish regression models,the software program includes modules such as data processing,correlation analysis,dataset partitioning,and emission prediction models.The program can easily adjust critical parameters in data processing and model building,supporting experimental and research work.(2)This study investigated modeling and model evaluation methods for the emission performance of a certain vehicle model.Based on modeling requirements,a certain amount of WLTC and RDE emission data were designed and collected.After dividing them into cold and hot machine operation conditions,Spearman’s method,which is insensitive to non-linear relationships and outliers,was selected to perform correlation analysis on the hot machine operation condition data.Eight parameters including pedal opening,vehicle speed,engine speed,intake VVT,exhaust VVT,ignition angle,catalytic oxygen storage capacity,and catalytic converter temperature were selected as input variables for the model based on strong correlation,controllability,low coupling,calculation complexity,and ease of acquisition.Through experimentation in the modeling process,the optimal topology and hyperparameter combination for the BP neural network model were obtained.A BP neural network with three hidden layers,each containing 40 neurons,and Re LU activation function can achieve relatively good prediction accuracy.(3)This study investigated peak error correction strategies to improve the emission model.In order to solve the problem of low accuracy in CO prediction and large differences between predicted and actual values at pollutant peaks,this thesis proposed a peak error correction strategy by introducing a peak correction model on top of the basic emission prediction model.The predicted R~2 value of the corrected CO emission model increased from 0.071 to 0.709,and there was also an improvement in the predicted R~2 values for NOx and PN emission models.This demonstrates that the peak error correction strategy has practicality and feasibility,and can effectively improve the accuracy of pollutant emission predictions. |
