Investigations On Neural Network Prediction Models For Vertical Transport Fluxes Of Polycyclic Aromatic Hydrocarbon Water Pollutants In Vadose Zones Under Typical Conditions

Surface water contaminated with organic pollutants can seriously threaten the groundwater quality safety through infiltration and pollutant transmission in the vadose zone.In order to efficiently control soil pollution and optimize the prevention and control of groundwater pollution,it is of great theoretical and practical value to investigate how to use non-massive numerical simulation data to establish a neural network substitution model for polycyclic aromatic hydrocarbon water pollutant fluxes in vadose zone.Based on the two typical conditions of quantitative infiltration of polycyclic aromatic hydrocarbons(PAHs)organic polluted water on the surface,namely,homogeneous porous medium vadose zone and layered heterogeneous vadose zone,this paper analyzes the different influencing factors of stable fluxes for PAHs,preliminarily identification the primary controlling factors,and obtain the stable pollutant fluxes data of the vadose zone with different parameters’ combination scenarios by CHEMFLO-2000.On this basis,this paper investigates how to establish practical and reliabe BP neural network substitution model for pollution fluxes in vadose zone under typical conditions.It is found that the cross-sectional depth,the saturated hydraulic conductivity of the vadose zone,the boundary flow rate of surface infiltration,the concentration of pollutants in the boundary sewage of surface infiltration,and the first-order decay coefficient of microorganic degradation are the five primary controlling factors determining the pollutant fluxes of the horizontal cross-section in homogeneous vadose zone.And for the the pollutant fluxes on the vadose zone-water table interface in layered heterogeneous vadose zone,the total thickness of the vadose zone,the proportion of the thickness of the first layer,the saturated hydraulic conductivity of the upper layer and of the lower layer,the boundary flow rate of surface infiltration,the concentration of pollutants in the boundary sewage of surface infiltration,and the first-order decay coefficient of microorganic degradation are the seven primary controlling factors.By obtaining non-massive simulation data and machine learning,a neural network substitution model of stable pollutant fluxes in the vadose zone can be established.In this study,the substitution model of horizontal section pollutant fluxes in the homogeneous vadose zone with 8 hidden layer nodes under 5 primary input factors can reach 95.48%,and the model of pollutant fluxes on the vadose zone-water table interface in layered heterogeneous vadose zone with 6 hidden layer node numbers under 7 primary input factors can reach 93.89%.The application of the established neural network substiton model under given typical conditions can effectively reduce the huge computational load and a lot of time cost caused by multiple numerical simulation calculations.This study provides feasibility and method reference for how to select suitable input primary controlling factors and use non-massive numerical simulation data to establish a practical and reliable artificial intelligence model of pollutant fluxes in vadose zone.This paper has 11 pictures,16 tables and 72 references.