基于大规模用户数据的中国燃油私家车出行碳排放测算方法

Method for estimating carbon emissions from fuel-powered private cars in China based on large-scale user data

【目的】私家车行驶碳排放是中国居民日常出行碳排放的主要来源,现有测算方法局限于宏观层面研究,本文以车主大数据和机器学习方法为基础,旨在构建燃油私家车出行碳排放量测算方法,可为道路交通碳排放量核算提供科学依据,助力实现碳达峰、碳中和。【方法】本文以2014—2023年中国13万余燃油车主披露的多维度数据为基础,运用线性回归、决策树、神经网络等方法,提出车辆(单辆车)累积出行碳排放量、相同行驶天数车辆群的平均累积出行碳排放量计算模型及预测模型,构建车辆年均出行碳排放量计算模型。【结果】①在车辆累积出行碳排放量预测方面,相比仅基于车龄的预测模型,由车型、制造商、区域、城市级别、价格区间等5个类别以及车龄构建的包含25个变量的模型可将预测精度(R2)由0.666提高到0.821,其中车龄对车辆累积出行碳排放量预测的影响程度仍然最高;②在相同行驶天数车辆群的平均累积出行碳排放量预测方面,基于神经网络方法的预测结果显示,全国层面的车辆平均累积出行碳排放量的预测精度达到0.915,车型、制造商、区域、城市级别、价格区间等5个类别的车辆平均累积出行碳排放量的预测精度处于0.875~0.925之间;③在车辆年均出行碳排放量测算方面,全国层面的车辆年均出行碳排放量为2.99 t,进口制造商、价格区间在(50,100]万、价格高于100万等3个类别下的车辆年均出行碳排放量高于4 t,MPV、SUV车型、华南区域、超大城市级别、价格区间在(20,30]万、价格区间在(30,50]万等6个类别下的车辆年均出行碳排放量处于3.00~3.99 t之间,其余15个子类别划分下车辆年均出行碳排放量均低于全国层面的平均水平,处于2.15~2.99 t之间。【结论】具有车龄、车型、制造商、区域、城市级别、价格区间等多维特征的大规模车主数据以及神经网络模型有助于提高私家车使用周期碳排放量测算的科学性、全面性,不同区域的道路交通碳排放量测算应考虑上述特征差异。本文可为新能源私家车出行碳排放量测算研究提供参考。

[Objective]The objective of this study was to develop a precise calculation method for fuel-powered private car travel carbon emissions,utilizing big data from car owners and employing machine learning techniques.This method aims to address the limitations of existing macro-level studies and provide a scientific basis for accurately estimating road transport carbon emissions.It will also contribute to achieving peak carbon dioxide emissions and promoting carbon neutrality in China.[Methods]Based on multidimensional data disclosed by over 130000 fuel vehicle owners in China from 2014 to 2023,this article utilizes methods such as linear regression,decision trees,and neural networks to propose a calculation model and prediction model for the average cumulative travel carbon emissions of vehicle groups with the same number of days traveled as well as individual vehicles.Subsequently,a calculation model for annual average travel carbon emissions of vehicles is constructed.[Results](1)In terms of predicting the cumulative travel carbon emissions of vehicles,compared to the vehicle cumulative travel carbon emission prediction model based solely on vehicle age,the prediction accuracy(R2)of the model with 25 variables in five categories(vehicle type,manufacturer,region,city level,price range)and vehicle age increased from 0.666 to 0.821.Age was found to have the highest impact on predicting cumulative travel carbon emissions for vehicles.(2)In terms of predicting the average cumulative carbon emissions from vehicles in a group with the same number of days traveled,the results based on neural network methods show that the prediction accuracy for the average cumulative travel carbon emissions at a national level reaches 0.915.The prediction accuracy for average cumulative travel carbon emissions for vehicles in five categories including vehicle type,manufacturer,region,city level,and price range ranges from 0.875 to 0.925.(3)In terms of calculating the average annual carbon emissions from vehicle travel,the national level has an average of 2.99 t.Vehicles in three categories—imported manufacturers,prices between(500,1000]thousand CNY,and prices over 1000 thousand CNY—have higher average annual carbon emissions exceeding 4 t.Vehicles in six categories—MPV and SUV models,South China region,super large cities,price range between(200,300]thousand CNY and price range between(300,500]thousand CNY—have average annual carbon emissions ranging from 3.00 to 3.99 t.The remaining fifteen subcategories have lower average annual carbon emissions than the national level's average level and fall within the range of 2.15 to 2.99 t.[Conclusion]The large-scale vehicle owner data of multiple dimensions such as vehicle age,vehicle type,manufacturer,region,city level,and price range,along with the neural network model,contribute to improving the scientific and comprehensive measurement of carbon emissions from private car usage cycles.The calculation of road traffic carbon emissions in different regions should take into account the aforementioned differences in features.This study provides a reference for calculating carbon emissions from new energy private car travel.