Research On Real-time And Fast Monitoring System Of Vehicle Emissions Based On Ensemble Learning

Vehicle emissions are a major source of air pollution in China,seriously affecting daily living and human health.Since July 2019,we constantly implemented China Stage Ⅵ emission standards,which has more strict requirements on the measures of emission monitoring,in order to further improve the national environmental quality and develop the Green Intelligent Transportation.However,traditional gaseous emissions measurements exist problems such as too large and high-cost equipment、complicated process、low monitoring frequency and long processing time.Meanwhile,multi-sensor has been widely applied in the field of applied sciences.It has good sensitivity and specificity,providing rapid results,and allowing non-destructive sampling of odorants or analytes.Furthermore,multi-sensor generally is far less expensive than analytical systems,easier and cheaper to operate,and has greater potential for portability and field use compared with complex analytical laboratory instruments.Thus,this paper aimed to demonstrate an approach combined with multisensor and data algorithms,which designed a real-time,and rapid monitoring device system based on Adaboost-SVM of ensemble learning,to further advance the ability of real-time performance and data analysis in vehicle emissions measurements field.Firstly,we evaluated each sensors performance including cross-reactive,consistency and stability,and subsequently selected 12 kinds of sensors for the target gas of vehicle emissions.Then,by means of designing cavity and arranging sensors,the main multi-sensor array module was established.Next,the front part of the device was formed with flow pump、three-way valve and PTFE tube.Multi-sensor hardware design platform was established.The target gas sample experiment was further designed,and then it was verified that the multi-sensor system’s sample values were consistent with the gaseous emission by means of engine condition model.Then,according to requirements of the national standard,we established engine bench model and then conducted emission testing,obtaining the emission values that were subsequently verified by the multi-sensor system.Finally,with the aim of improving the ability of identification and classification of a multi-sensor system for vehicle emissions,PCA was used to extract the features of the data,and then the Adaboost-SVM model based on ensemble learning was established for pattern recognition and classification of vehicle emissions under different operating conditions,which further advanced the ability of real-time performance and data analysis in vehicle emissions measurements.Meanwhile,we proposed two SVM classification models and Back Propagation(BP)neural network which is the typical machine learning algorithm used to be the pattern recognition algorithm of multi-sensor,to compared with it.Consequently,the established Adaboost-SVM model based on ensemble learning,the best-performance one,was for the proposed multi-sensor under experimental vehicle emissions conditions,achieving the high precise detection accuracy and robustness.