Self-feedback LSTM regression model for real-time particle source apportionment

Atmospheric particulate matter pollution has attracted much wider attention globally.In recent years,the development of atmospheric particle collection techniques has put forwards new demands on the real-time source apportionments techniques.Such demands are summarized,in this paper,as how to set up new restraints in apportionment and how to develop a non-linear regression model to process complicated circumstances,such as the existence of secondary source and similar source.In this study,we firstly analyze the possible and potential restraints in single particle source apportionment,then propose a novel three-step self-feedback long short-term memory(SF-LSTM)network for approximating the source contribution.The proposed deep learning neural network includes three modules,as generation,scoring and refining,and regeneration modules.Benefited from the scoring modules,SF-LSTM implants four loss functions representing four restraints to be followed in the apportionment,meanwhile,the regeneration module calculates the source contribution in a non-linear way.The results show that the model outperforms the conventional regression methods in the overall performance of the four evaluation indicators(residual sum of squares,stability,sparsity,negativity)for the restraints.Additionally,in short time-resolution analyzing,SF-LSTM provides better results under the restraint of stability.

Machine learning and theoretical analysis release the non-linear relationship among ozone,secondary organic aerosol and volatile organic compounds

Fine particulate matter(PM_(2.5))and ozone(O_(3))pollutions are prevalent air quality issues in China.Volatile organic compounds(VOCs)have significant impact on the formation of O_(3)and secondary organic aerosols(SOA)contributing PM_(2.5).Herein,we investigated 54 VOCs,O_(3)and SOA in Tianjin from June 2017 to May 2019 to explore the non-linear relationship among O_(3),SOA and VOCs.The monthly patterns of VOCs and SOA concentrations were characterized by peak values during October to March and reached a minimum from April to September,but the observed O_(3)was exactly the opposite.Machine learning methods resolved the importance of individual VOCs on O_(3)and SOA that alkenes(mainly ethylene,propylene,and isoprene)have the highest importance to O_(3)formation;alkanes(C_(n),n≥6)and aromatics were the main source of SOA formation.Machine learning methods revealed and emphasized the importance of photochemical consumptions of VOCs to O_(3)and SOA formation.Ozone formation potential(OFP)and secondary organic aerosol formation potential(SOAFP)calculated by consumed VOCs quantitatively indicated that more than 80%of the consumed VOCs were alkenes which dominated the O_(3)formation,and the importance of consumed aromatics and alkenes to SOAFP were 40.84%and 56.65%,respectively.Therein,isoprene contributed the most to OFP at 41.45%regardless of the season,while aromatics(58.27%)contributed the most to SOAFP in winter.Collectively,our findings can provide scientific evidence on policymaking for VOCs controls on seasonal scales to achieve effective reduction in both SOA and O_(3).