徐晗， 杨岐元， 史国良， 冯银厂

1.南开大学 环境科学与工程学院， 生态环境部城市空气颗粒物污染防治重点实验室， 天津 300350；2.中国气象局-南开大学大气环境与健康研究联合实验室， 天津 300074

引用格式：

徐晗， 杨岐元， 史国良， 等. 从传统溯源模型到人工智能：大气颗粒物来源解析进展［J］. 中国粉体技术， 2027， 33（2）： 24-37.

Xu Han， Yang Qiyuan， Shi Guoliang， et al. From traditional source apportionment methods to artificial intelligence： advances in atmospheric particulate matter source apportionment［J］. China Powder Science and Technology， 2027， 33（2）： 24-37.

DOI：10.13732/j.issn.1008-5548.2027.02.003

收稿日期： 2026-07-17， 修回日期： 2026-08-17， 上线日期： 2026-09-22。

基金项目： 国家自然科学基金项目，编号：42477105。

第一作者： 徐晗（1999—），男，博士生，研究方向为基于机器学习的大气污染成因分析。E-mail：nkxuhan@mail.nankai.edu.cn。

通信作者： 史国良（1981—），男，研究员，博士，博士生导师，国家生态环境保护专业技术青年拔尖人才，研究方向为大气的污染防治。E-mail：nksgl@nankai.edu.cn。

摘要： 【目的】 系统梳理大气颗粒物来源解析方法的发展与应用，为高精度与智能化的源解析工作提供技术参考。 【研究现状】 颗粒物源解析方法的发展历程表明，排放清单方法可定量表征污染源及其排放强度，为源解析提供基础数据；化学传输模型通过模拟大气物理化学全过程，实现区域尺度的源贡献解析；受体模型则能够基于观测点位的受体数据反推污染源贡献。近年来，机器学习方法在源解析领域得到广泛应用，凭借在非线性建模、高维特征提取及多源数据融合方面的优势，角色已从传统溯源模型的辅助工具，发展为主导性的源贡献解析手段，并进一步与物理化学机制深度融合，为揭示二次反应机制、提升源解析时空分辨率提供新路径。 【结论与展望】 未来研究应进一步加强人工智能方法与物理化学机制的深度融合，以提升溯源结果的物理意义；构建兼具物理一致性、预测精度与计算效率的智能源解析框架，实现经典统计方法向机制约束与数据驱动协同发展的转变，为大气污染防控及区域协同治理提供科学支撑。

关键词： 大气颗粒物； 来源解析； 机器学习

Abstract

Significance Atmospheric particulate matter （PM） is a major air pollutant that exerts substantial impacts on human health， climate， atmospheric visibility， and ecosystems. Accurate identification of emission sources and quantitative estimation of their contributions are fundamental for understanding particulate pollution formation and developing effective emission control strategies. Over the past several decades， particulate matter source apportionment has evolved from conventional emission-based and statistical approaches toward integrated frameworks combining emission inventories， chemical transport models （CTMs）， receptor models， high-time-resolution observations， and artificial intelligence （AI）. In particular， rapid advances in machine learning， high-performance computing， online monitoring， and multi-source environmental data are reshaping the methodological framework of particulate matter source apportionment. This review systematically summarizes the development， characteristics， representative applications， advantages， and limitations of major particulate matter source apportionment methods， with particular emphasis on the transition from traditional approaches to machine learning-driven and mechanism–data-integrated methods. It aims to provide a methodological reference for achieving high-precision and intelligent source apportionment.

Progress Traditional particulate matter source apportionment methods can generally be divided into two categories： source-oriented approaches and receptor-oriented approaches. Emission inventories represent one of the earliest source-oriented approaches and quantitatively characterize emissions from anthropogenic and natural sources based on activity data， emission factors， statistical information， and observational constraints. Their evolution from coarse-resolution statistical inventories toward high-resolution， multi-pollutant， and multi-sector datasets has provided essential inputs for atmospheric modeling and source attribution. With advances in the understanding of atmospheric transport and chemical transformation， CTMs have gradually evolved from simple dispersion models into three-dimensional modeling systems capable of representing emissions， transport， diffusion， chemical reactions， secondary aerosol formation， and deposition. Source-oriented techniques， including brute-force emission perturbation， sensitivity analysis， and source-tagging methods， further enable CTMs to quantify contributions from different regions and emission sectors.

Conclusions and Prospects Overall， particulate matter source apportionment has evolved from emission statistics and linear receptor analysis toward dynamic process simulation， high-time-resolution observations， nonlinear data-driven modeling， and mechanism–data integration. Traditional emission inventories， CTMs， CMB， and PMF remain indispensable due to their clear physical or statistical foundations. However， limitations related to computational cost， parameter uncertainty， source profile dependence， and spatiotemporal resolution remain. Machine learning provides complementary advantages in nonlinear modeling， automatic feature extraction， multi-source data fusion， and computational efficiency， while purely data-driven methods still face challenges in physical interpretability， uncertainty quantification， robustness， and cross-regional generalization. Future research should therefore focus on deeper integration of AI with receptor models， CTMs， and atmospheric physical and chemical mechanisms. Incorporating physical laws， source profile knowledge， chemical transformation mechanisms， and mass conservation relationships into model architectures and training processes， together with the integration of ground observations， online measurements， satellite remote sensing， emission inventories， meteorological data， and socioeconomic information， will facilitate the development of intelligent source apportionment frameworks with high accuracy， efficiency， generalization ability， and interpretability. Such advances are expected to promote particulate matter source apportionment toward an intelligent and mechanism-informed stage， thereby providing stronger scientific support for precision air quality management and coordinated regional pollution control.

Keywords： atmospheric particulate matter； source apportionment； machine learning

参考文献（References）

［1］Pope III C A， Dockery D W. Health effects of fine particulate air pollution： lines that connect［J］. Journal of the Air & Waste Management Association， 2006， 56（6）： 709-742.

［2］Sangkham S， Phairuang W， Sherchan S P， et al. An update on adverse health effects from exposure to PM_2.5［J］. Environmental Advances， 2024， 18： 100603.

［3］Guo Juanmei， Chai Guorong， Song Xuping， et al. Long-term exposure to particulate matter on cardiovascular and respiratorydiseases in low- and middle-income countries： a systematic review and meta-analysis［J］. Frontiers in Public Health， 2023， 11： 1134341.

［4］Kalisoras A， Georgoulias A K， Akritidis D， et al. Decomposing the effective radiative forcing of anthropogenic aerosols based on CMIP6 Earth system models［J］. Atmospheric Chemistry and Physics， 2024， 24（13）： 7837-7872.

［5］Orellano P， Kasdagli M I， Pérez Velasco R， et al. Long-term exposure to particulate matter and mortality： an update of the WHO Global Air Quality Guidelines systematic review and meta-analysis［J］. International Journal of Public Health， 2024， 69： 1607683.

［6］Materano-Escalona L S. Particulate matter sampling to assess air pollution［J］. Nature Reviews Earth & Environment， 2024， 5： 230.

［7］Wang An， Paul S， DeSouza P， et al. Key themes， trends， and drivers of mobile ambient air quality monitoring： a systematic review and meta-analysis［J］. Environmental Science & Technology， 2023， 57（26）： 9427-9444.

［8］Zhang Xinran， Yin Shasha， Lu Xuan， et al. Establish of air pollutants and greenhouse gases emission inventory and cobenefits of their reduction of transportation sector in Central China［J］. Journal of Environmental Sciences， 2025， 150： 604-621.

［9］Li Dongpu， Liu Hongli， Duan Guo. High-resolution anthropogenic emission inventory for China （2015-2024）： Spatiotemporal changes and environmental application［J］. Atmospheric Environment， 2025， 361： 121495.

［10］Gao Zhaoqi， Zhou Xuehua. A review of the CAMx， CMAQ， WRF-Chem and NAQPMS models： application， evaluation and uncertainty factors［J］. Environmental Pollution， 2024， 343： 123183.

［11］戴树桂， 朱坦， 白志鹏. 受体模型在大气颗粒物源解析中的应用和进展［J］. 中国环境科学， 1995， 15（4）： 252-257.

Dai Shugui， Zhu Tan， Bai Zhipeng. Application and development of receptor models for the source apportionment of airborne particulate matter［J］. China Environmental Science， 1995， 15（4）： 252-257.

［12］Liu Jie， Ma Fangjingxin， Chen T L， et al. High-time resolution PM_2.5 source apportionment assisted by spectrum-based characteristics analysis［J］. Science of the Total Environment， 2024， 912： 169055.

［13］Zhang Yong， Tian Jie， Wang Qiyuan， et al. High-time-resolution chemical composition and source apportionment of PM_2.5 in northern Chinese cities： implications for policy［J］. Atmospheric Chemistry and Physics， 2023， 23（16）： 9455-9471.

［14］Fu Xiao， Geng Guannan， Hopke P K， et al. Advances in emissions， modeling， and source apportionment［J］. Environmental Science & Technology， 2026， 60（8）： 5927-5933.

［15］Tian Jie， Wang Qiyuan， Zhang Yong， et al. A review of receptor model and new advances in source apportionment techniques for air pollution research［J/OL］. Aerosol Science and Engineering， 2025.https：//link.springer.com/article/10.1007/s41810-025-00328-x.

［16］Crippa M， Guizzardi D， Pagani F， et al. Insights into the spatial distribution of global， national， and subnational greenhouse gas emissions in the Emissions Database for Global Atmospheric Research （EDGAR v8.0）［J］. Earth System Science Data， 2024， 16（6）： 2811-2830.

［17］Hoesly R M， Smith S J， Feng Leyang， et al. Historical （1750–2014） anthropogenic emissions of reactive gases and aerosols from the Community Emissions Data System （CEDS）［J］. Geoscientific Model Development， 2018， 11（1）： 369-408.

［18］Wu Nana， Geng Guannan， Xu Ruochong， et al. Development of a high-resolution integrated emission inventory of air pollutants for China［J］. Earth System Science Data， 2024， 16（6）： 2893-2915.

［19］Li Shengyue， Wang Shuxiao， Wu Qingru， et al. Emission trends of air pollutants and CO₂ in China from 2005 to 2021［J］. Earth System Science Data， 2023， 15（6）： 2279-2294.

［20］Zhou Yaduan， Zhao Yu， Mao Pan， et al. Development of a high-resolution emission inventory and its evaluation and application through air quality modeling for Jiangsu Province， China［J］. Atmospheric Chemistry and Physics， 2017， 17（1）： 211-233.

［21］Huang Zhijiong， Zhong Zhuangmin， Sha Qinge， et al. An updated model-ready emission inventory for Guangdong Province by incorporating big data and mapping onto multiple chemical mechanisms［J］. Science of the Total Environment， 2021， 769： 144535.

［22］Huang Xin， Song Yu， Li Mengmeng， et al. A high-resolution ammonia emission inventory in China［J］. Global Biogeochemical Cycles， 2012， 26： B1030.

［23］Liu Huan， Fu Mingliang， Jin Xinxin， et al. Health and climate impacts of ocean-going vessels in East Asia［J］. Nature Climate Change， 2016， 6（11）： 1037-1041.

［24］Song Lilai， Li Zhen， Zhang Jinqiu， et al. An hourly and localized optimization method for soil fugitive dust emission inventory based on machine learning［J］. Journal of Environmental Sciences， 2025， 158： 1-12.

［25］Li Jinxi， Li Yijie， Wang Zifa， et al. Next generation air quality models： dynamical mesh， new insights into mechanism， datasets and applications［J］. Current Pollution Reports， 2025， 11： 25.

［26］Badawy H. Recent trends in spatial modeling studies of air quality （2010-2023）［J］. Earth Science Informatics， 2025， 18： 193.

［27］Wang Zifa， Li Xinwei， Gao Meng， et al. Advances and perspectives in atmospheric environment modeling in China［J］. Advances in Atmospheric Sciences， 2026， 43： 1-25.

［28］EPICC Model Working Group. Description and evaluation of the emission and atmospheric processes integrated and coupled community （EPICC） model version 1.0［J］. Advances in Atmospheric Sciences， 2026， 43（4）： 671-694.

［29］Zhang Lei， Gong Sunling， Zhao Tianliang， et al. Development of WRF/CUACE v1.0 model and its preliminary application in simulating air quality in China［J］. Geoscientific Model Development， 2021， 14（2）： 703-718.

［30］罗干， 徐北瑶， 王体健， 等. 区域大气环境模拟系统RegAEMS的研究进展［J］. 中国环境科学， 2022， 42（10）： 4571-4580.

Luo Gan， Xu Beiyao， Wang Tijian， et al. Development of regional atmospheric environment modeling system （RegAEMS）［J］. China Environmental Science， 2022， 42（10）： 4571-4580.

［31］Hu Feng， Xie Pinhua， Tian Xin， et al. Integrated analysis of the transport process and source attribution of an extreme ozone pollution event in Hefei at different vertical heights： a case of study［J］. Science of the Total Environment， 2024， 906： 167237.

［32］Kwok R H F， Napelenok S L， Baker K R. Implementation and evaluation of PM_2.5 source contribution analysis in a photochemical model［J］. Atmospheric Environment， 2013， 80： 398-407.

［33］Wagstrom K M， Pandis S N， Yarwood G， et al. Development and application of a computationally efficient particulate matter apportionment algorithm in a three-dimensional chemical transport model［J］. Atmospheric Environment， 2008， 42（22）： 5650-5659.

［34］Miller M S， Friedlander S K， Hidy G M. A chemical element balance for the Pasadena aerosol［J］. Journal of Colloid and Interface Science， 1972， 39（1）： 165-176.

［35］Hopke P K. Review of receptor modeling methods for source apportionment［J］. Journal of the Air & Waste Management Association， 2016， 66（3）： 237-259.

［36］Cooper J A， Watson J G Jr. Receptor oriented methods of air particulate source apportionment［J］. Journal of the Air Pollution Control Association， 1980， 30（10）： 1116-1125.

［37］Tian Yingze， Wang Xiaoning， Zhao Peng， et al. PM_2.5 source apportionment using organic marker-based chemical mass balance modeling： Influence of inorganic markers and sensitivity to source profiles［J］. Atmospheric Environment， 2023， 294： 119477.

［38］Bai Wenyu， Zhao Xueyan， Yin Baohui， et al. Characteristics of PM_2.5 in an industrial city of Northern China： mass concentrations， chemical composition， source apportionment， and health risk assessment［J］. International Journal of Envionmental Research and Public Health， 2022， 19（9）： 5443.

［39］Kaur K， Krall J R， Ivey C， et al. Long-term PM_2.5 source apportionment at 12 sites in the western United States from 2000 to 2020［J］. Atmospheric Pollution Research， 2026， 17（2）： 102779.

［40］Bi Xiaohui， Feng Yinchang， Wu Jianhui， et al. Source apportionment of PM₁₀ in six cities of northern China［J］. Atmospheric Environment， 2007， 41（5）： 903-912.

［41］Wang Qiaoqiao， Shao Min， Zhang Yuanxun， et al. Source apportionment of fine organic aerosols in Beijing［J］. Atmospheric Chemistry and Physics， 2009， 9（21）： 8573-8585.

［42］赵宏， 白志鹏， 郭光焕， 等. 源解析NKCMB最优拟合法及其实现［J］. 南开大学学报（自然科学版）， 2010， 43（1）： 71-75.

Zhao Hong， Bai Zhipeng， Guo Guanghuan， et al. The designing and implementation of NKCMB best fitting algorithm of source apportionment［J］. Acta Scientiarum Naturalium Universitatis Nankaiensis （Natural Science Edition）， 2010， 43（1）：71-75.

［43］Shi Guoliang， Tian Yingze， Zhang Yufen， et al. Estimation of the concentrations of primary and secondary organic carbon in ambient particulate matter： application of the CMB-Iteration method［J］. Atmospheric Environment， 2011， 45（32）： 5692-5698.

［44］Dinh V N T， Uzu G， Dominutti P， et al. Toolbox for accurate estimation and validation of positive matrix factorization solutions in particulate matter source apportionment［J］. Atmospheric Measurement Techniques， 2025， 18（22）： 6817-6833.

［45］Fossum K N， Lin C， O'Sullivan N， et al. Two distinct ship emission profiles for organic-sulfate source apportionment of PM in sulfur emission control areas［J］. Atmospheric Chemistry and Physics， 2024， 24（18）： 10815-10831.

［46］Gao Jie， Dong Shihao， Yu Haofei， et al. Source apportionment for online dataset at a megacity in China using a new PTT-PMF model［J］. Atmospheric Environment， 2020， 229： 117457.

［47］Pekney N J， Davidson C I， Zhou Liming， et al. Application of PSCF and CPF to PMF-modeled sources of PM_2.5 in Pittsburgh［J］. Aerosol Science and Technology， 2006， 40（10）： 952-961.

［48］Dai Qili， Liu Baoshuang， Bi Xiaohui， et al. Dispersion normalized PMF provides insights into the significant changes in source contributions to PM_2.5 after the COVID-19 outbreak［J］. Environmental Science & Technology， 2020， 54（16）： 9917-9927.

［49］Bhattu D， Tripathi S N， Bhowmik H S， et al. Local incomplete combustion emissions define the PM_2.5 oxidative potential in Northern India［J］. Nature Communications， 2024， 15： 3517.

［50］Peng Zezhi， Zhang Bin， Wang Diwei， et al. Application of machine learning in atmospheric pollution research： a state-of-art review［J］. Science of the Total Environment， 2024， 910： 168588.

［51］Alkhodaidi A， Attiah A， Mhawish A， et al. The role of machine learning in enhancing particulate matter estimation： a systematic literature review［J］. Technologies， 2024， 12（10）： 198.

［52］Song Xinhua， Hopke P K. Solving the chemical mass balance problem using an artificial neural network［J］. Environmental Science & Technology， 1996， 30（2）： 531-535.

［53］Dinh V N T， Jaffrezo J L， Hough I， et al. Unveiling the optimal regression model for source apportionment of the oxidative potential of PM₁₀［J］. Atmospheric Chemistry and Physics， 2024， 24（12）： 7261-7282.

［54］Peng Xing， Ma Haonan， He Lingyan， et al. Machine-learning source apportionment of particulate pollution aids urban emission regulations［J］. Environmental Science & Technology， 2026， 60（15）： 11564-11576.

［55］Xu Yongjiang， Wang Zaihua， Pei Chenglei， et al. Single particle mass spectral signatures from on-road and non-road vehicle exhaust particles and their application in refined source apportionment using deep learning［J］. Science of the Total Environment， 2024， 930： 172822.

［56］连鸿运， 周围， 马景金， 等. 基于机器学习的大气颗粒物来源解析方法综述［J］. 环境化学， 2025， 44（12）： 4633-4643.

Lian Hongyun， Zhou Wei， Ma Jingjin， et al. Review of machine learning-based source apportionment methods for atmospheric particulate matter［J］. Environmental Chemistry， 2025， 44（12）： 4633-4643.

［57］Kim Y， Yi S M， Heo J， et al. Is replacing missing values of PM_2.5 constituents with estimates using machine learning better for source apportionment than exclusion or Median replacement？［J］. Environmental Pollution， 2024， 354： 124165.

［58］Dai Tianjiao， Dai Qili， Ding Jing， et al. Measuring the emission changes and meteorological dependence of source-specific BC aerosol using factor analysis coupled with machine learning［J］. Journal of Geophysical Research： Atmospheres， 2023， 128（15）： e2023JD038696.

［59］Xu Han， Huang Jiahao， Zhao Dongheng， et al. Co-benefits between PM_2.5 control and carbon reduction revealed by interpretable machine learning methods［J］. Geophysical Research Letters， 2025， 52（12）： e2025GL115654.

［60］Xu Bo， Huang Junbo， Ge Yi， et al. Automatic identification of factor profiles can be achieved by improved machine learning model［J］. Atmospheric Environment， 2024， 323： 120407.

［61］Liao Dan， Hong Youwei， Huang Huabin， et al. Machine learning exploring the chemical compositions characteristics and sources of PM_2.5 from reduced on-road activity［J］. Atmospheric Pollution Research， 2024， 15（11）： 102265.

［62］Wang Shuai， Zhang Mengyuan， Gao Yueqi， et al. Diagnosing drivers of PM_2.5 simulation biases in China from meteorology，chemical composition， and emission sources using an efficient machine learning method［J］. Geoscientific Model Development， 2024， 17（9）： 3617-3629.

［63］Letaïef S， Camps P， Carvallo C. A supervised machine learning approach to classify traffic-derived PM sources based on their magnetic properties［J］. Environmental Research， 2023， 231： 116006.

［64］Wang Wei， Xu Weiman， Deng Shuai， et al. Self-feedback LSTM regression model for real-time particle source apportionment［J］. Journal of Environmental Sciences， 2022， 114： 10-20.

［65］Xu Han， Yu Haofei， Xu Bo， et al. Machine learning coupled structure mining method visualizes the impact of multiple drivers on ambient ozone［J］. Communications Earth & Environment， 2023， 4： 265.

［66］Wang Guanzhong， Ruser H， Schade J， et al. Machine learning approaches for automatic classification of single-particle mass spectrometry data［J］. Atmospheric Measurement Techniques， 2024， 17（1）： 299-313.

［67］Li Lei， Li Mei， Huang Zhengxu， et al. Ambient particle characterization by single particle aerosol mass spectrometry in an urban area of Beijing［J］. Atmospheric Environment， 2014， 94： 323-331.

［68］Zhao Peng， Zhao Pusheng，Zhang Wei， et al. Source identification of atmospheric particles via low-voltage electron microscopy image recognition： a case study of submicrometer particles［J］. Environmental Science & Technology Letters， 2026， 13（3）： 360-366.

［69］Dadheech N， He Tailong， Turner A J. High-resolution greenhouse gas flux inversions using a machine learning surrogate model for atmospheric transport［J］. Atmospheric Chemistry and Physics， 2025， 25（10）： 5159-5174.

［70］Lyu Baolei， Huang Ran， Wang Xinlu， et al. FastCTM （v1.0）： atmospheric chemical transport modelling with a principle-informed neural network for air quality simulations［J］. Geoscientific Model Development， 2025， 18（18）： 6295-6312.

［71］Xing Jia， Zheng Shuxin， Ding Dian， et al. Deep learning for prediction of the air quality response to emission changes［J］. Environmental Science & Technology， 2020， 54（14）： 8589-8600.

［72］Gui Ke， Zhang Xutao， Che Huizheng， et al. Advancing operational global aerosol forecasting with machine learning［J］. Nature， 2026， 651（8106）： 658-665.

［73］Uddin Z， Sai Ganga， Asthana R， et al. Wavelets based physics informed neural networks to solve non-linear differential equations［J］. Scientific Reports， 2023， 13： 2882.

［74］Pandey H， Singh A， Behera R. An efficient wavelet-based physics-informed neural network for multiscale problems［J］. Neural Networks， 2026， 200： 108860.

［75］Xu Bo， Xu Han， Zhao Huan， et al. Source apportionment of fine particulate matter at a megacity in China， using an improved regularization supervised PMF model［J］. Science of the Total Environment， 2023， 879： 163198.

［76］Xu Bo， Yu Haofei， Shi Zongbo， et al. Knowledge-guided machine learning reveals pivotal drivers for gas-to-particle conversion of atmospheric nitrate［J］. Environmental Science and Ecotechnology， 2024， 19： 100333.

［77］Naderi F， Perez-Raya I， Yadav S， et al. Towards chemical source tracking and characterization using physics-informed neural networks［J］. Atmospheric Environment， 2024， 334： 120679.

［78］Sharma H， Shrivastava M， Singh B. Physics informed deep neural network embedded in a chemical transport model for the Amazon rainforest［J］. Nature Partner Journals Climate and Atmospheric Science， 2023， 6： 28.