机器学习方法在预测废水中PFAS剖面和归宿中的应用.pdf

1、Applications of Machine Learning Approaches to Predict PFAS Profiles and Fate in Wastewater1Christopher I.Olivares chris.olivaresuci.eduDepartment of Civil and Environmental EngineeringUniversity of California,Irvine Figure adapted from AgriLife Today(TAMU)Field SamplingLow PFAS cross-contamination

2、LC-MS/MSSkilled data interpretation2PFAS concerns Sampling and measurement of PFAS in WWTP Observed increase in quantifiable PFAS in effluents Biosolids heterogenous matrixData Availability3Machine Learning Approaches to predict PFAS profilesin WWTPsObjective Statewide WWTP database(PFAS,wastewater

3、characteristics,treatment)Machine learning models to predict PFAS in effluent and biosolids Identify features that influence PFAS composition in effluents,biosolids4Data sources5California Water BoardsGeotracker-PFASWater Quality WWTP Effluent Self-Monitoring Report DataWastewater User Charge Data-S

4、ewageCOVID-19 Wastewater Surveillance dataGeotracker-QuestionnaireDemographics by City,Town,CountyUS Census Bureau,other agenciesCounty Gross Domestic ProductetcPFAS data validationWe scrutinized data in scenarios:PFASs were detected in both influent and effluent but not in biosolids No PFASs were d

5、etected in any of the liquid,solid matrices PFASs were detected in biosolids but not in influent or effluent Outliers in PFASs concentration in any of these phases.PFAS Sampling locations&WWTP locations 6The compiled dataset includes:213 WWTPs from California 2020-2023 39 PFASs concentrations in inf

6、luent,effluent,and biosolids Features:Wastewater sourceWWTP information(e.g.,location,sample date,WWTP size,treatment process,water quality indicators)Socioeconomic factors 380 columns,931 rowsGeotracker map showing sampled WWTPsResults Statistical Summary7PFAS occurrence and profile in California W