Vehicle Emission Estimation Algorithms Based On Remote Sensing Data

With the development of modern technology and economy,the number of motor vehicles in China has continued to increase,and the environmental problems caused by a large number of vehicle emissions have become increasingly serious.Therefore,it is very important to accurately detect or estimate the real emission level of vehicle and reasonably assessing the overall emissions of vehicle fleet in a certain range of narrow areas for air quality monitoring and formulation of emission standards.The vehicle emission data,which is detected by remote sensing method,is greatly affected by environmental factors such as wind speed and temperature.The detection results are unstable,and the remote detection emission data cannot be directly used to evaluate the overall emission level of the vehicle fleet in the monitoring area.In this thesis,based on large amounts of remote sensing emission data which is collected from the Urban Road Network Motor Vehicle Emission Monitoring System,and some vehicle inspection emission data which is based on the chassis and engine dynamometer testing measure,we study the emission estimation algorithm of a single motor vehicle at the micro level and the overall vehicle fleet in a narrow area at the macro level,respectively.We make accurate estimates of vehicle pollutant emissions at different levels.In this thesis,the emission data analysis and processing methods,and emission estimation algorithms are explored and analyzed.The main work are as follows:Firstly,an improved method,which is based on semi-supervised cooperative training regression,for approximate compensation of emission data is proposed.The incomplete data and the abnormal data in the raw emission data records are approximated,while the distribution of raw data is not changed.Secondly,in view of the problem that the remote sensing vehicle emission test results are not completely accurate and are greatly affected by meteorological conditions.Based on the priori corrected emission data,we study the emission CO/HC/NO concentration inversion and estimation algorithm based on XGBoost and the over-standard emission determination algorithm based on multi-task learning neural network,and accurately estimate the emission of single vehicle at the micro level.Besides,based on the improved generative adversarial network,the consistency mapping algorithm between the remote sensing emission data and the vehicle inspection emission is performed.Thirdly,considering the fact that it is difficult to quantify and estimate the overall vehicle fleet emission level in the macroscopical narrow area,we propose a new method for constructing the emission timing samples and a new vehicle fleet emission estimation and prediction algorithm based on wavelet-LSTM,and based on the prediction algorithm we accurately estimates the emission CO/HC/HC of the overall fleet in a narrow area in the short-term.In this thesis,we also study the average emission factor estimation method based on the COPERT model,and combine with the long-term traffic statistical information to quantitatively and accurately estimate the long-term emission level of vehicle fleet in the narrow area.In addition,this thesis proves the validity and great accuracy of the proposed micro-and macro-level emission estimation algorithm by lots of comparative experiments.