Research On Influence Factors Of Taxi Emission Factors Based On Random Forest Regression

Due to the rapid development of vehicle,the on-road mobile sources has become one of the primary local pollution sources in China.As an indispensable part of urban passenger transport system,the average daily mileage as well as the driving time of taxis are significantly higher than that of ordinary passenger cars,which may cause more urban pollution.Vehicle emissions reduction will be of great significance to the improvement urban air environment and human health.On-road Remote Sensing(RS)is an effective and economic tool to monitor and control vehicle emissions.Establishing the real-world traffic flow-emission big data testing and real-time analysis technology is a hotspot in recent years,which has been applied in emission regulation and environmental governance.Random forest?a machine learning method?abandoned the modeling approach of making assumptions before estimating parameters,and exploring the inherent quantitative features from the data itself.The explanatory machine learning theory provides technical support for the evaluation and analysis of the algorithm model.The applications,major findings and shortcomings of remote sensing in previews studies were reviewed to guide its research.Then,27737 RS data were collected,three major pollutant emission factors(EFs)were calculated,and outliers of them were detected by z-score.According to the cross-section data of RS,Bootstrap sampling is performed first.And on this basis,with vehicle information,driving patterns,meteorological conditions and test time as explanatory variables,Random Forest Regression model was established for three EFs respectively and the Variable Importance Measure was achieved.After that,Partial Dependence Plot and Accumulated Local Effects Plot were used to analyze the influence of variables on the model response.It was revealed that vehicle age was the most important factor for vehicle emission.Due to the different generation mechanisms of different pollutants,the impact of different driving conditions on them is diverse.Avoiding frequent start,acceleration/deceleration and long-time low-speed operation is desirable.At present,the RS data quality is not quite well,and the QA/QC needs to be strengthened in the future.This study provides a feasible scheme for the practical application of emission monitoring big data in the field of vehicle supervision.