Pollutant concentration prediction method based on environment multi-medium spatial differentiation model

Abstract

The invention belongs to the technical field of ecological risk evaluation, and relates to a new pollutant environmental behavior prediction method, in particular to a pollutant concentration prediction method based on an environmental multi-medium spatial differentiation model, which is used for predicting spatial distribution of new pollutants in different media (such as air, water, soil and the like). The method comprises the steps of building an environment behavior parameter prediction model, building an environment parameter database, building an emission rate prediction model and building an environment multi-medium space differentiation model. According to the method, molecular fingerprint coding is combined with a multi-task learning strategy, multiple key environment behavior parameters are predicted at the same time, the parameter prediction efficiency and accuracy are greatly improved, and key input is provided for a subsequent multi-medium environment model. According to the method, deep combination of the machine learning model and the fugacity model is realized, parameter prediction and pollutant migration fate simulation are integrated, and a more comprehensive technical framework is provided for environment exposure prediction.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of ecological risk assessment, and relates to a new pollutant environmental behavior prediction method, in particular to a pollutant concentration prediction method based on an environmental multi-medium spatial differentiation model, which is used to predict the spatial distribution of new pollutants in different media (such as air, water, soil, etc.). BACKGROUND

[0002] Determining the environmental concentration of new pollutants is the key to the management of new pollutants. However, the emission of new pollutants is affected by regional economic structure, industrial layout and environmental conditions, and the emission types and levels show spatial differentiation characteristics. Due to the variety of new pollutants and the fact that the environmental concentration is usually at trace and trace level, this poses challenges to the spatial representativeness of environmental monitoring sites, the sensitivity of detection technology, the coverage of new pollutant types and the detection efficiency. In addition, the collection and analysis of new pollutant environmental samples is time-consuming and costly. Therefore, it is urgent to develop an efficient and feasible new pollutant environmental concentration spatial differentiation prediction technology, which is based on the structural characteristics of new pollutants, regional emission data, and environmental factors to predict the concentration level of new pollutants in each region and each environmental medium, in order to address the above challenges.

[0003] Based on the new pollutant environmental concentration spatial differentiation prediction technology, not only can the key pollution areas be quickly screened out for targeted management actions, but also the implementation scenarios of different management policies can be simulated to estimate the impact of the policies on pollutant concentration, which is helpful to achieve scientific and precise management. At present, the prediction of new pollutant concentration based on environmental multi-medium model is one of the important methods in this field. Traditional environmental multi-medium models are mostly based on fugacity theory and use homogenization assumption to simplify the structure of environmental system, so as to efficiently simulate the migration and distribution process of pollutants in different media (such as air, water, soil, etc.). Although this kind of model can predict the overall distribution trend of new pollutants and their environmental fate, it still has certain limitations in spatial resolution and dynamic concentration simulation due to the high simplification of environmental media.

[0004] In order to improve prediction accuracy and meet diverse environmental needs, researchers have gradually developed multi-media models with spatial differentiation capabilities in recent years. Based on regional zoning and spatial heterogeneity, these models divide the environmental region into multiple sub-regions and connect them through processes such as water flow and atmospheric exchange, which can effectively describe the dynamic migration of pollutants in spatial and temporal dimensions. Internationally, a variety of spatial differentiation models have been widely used in the field of environmental management and ecological risk assessment. For example, the paper "Environmental Science & Technology, 2015, 49 (1): 212-222" used Globo-POP to explore the environmental fate of trichlorodiphenylamine from 2000 to 2012 at a global resolution of 1°×1° and evaluated its long-range migration ability. For example, the paper "Environmental Toxicology and Chemistry, 2002, 21 (8): 1628-1637" simulated the environmental fate of toxaphene in North America based on the BETR-North America model and calculated the residues of toxaphene in various environmental media. The paper "Environmental Science & Technology, 2005, 39(4):1119-1128" used the Impact 2002 model to assess the environmental concentration distribution of 2,3,4,7,8-pentachlorodibenzofuran in Western Europe, the levels of contaminants in related foods, and calculated the proportion of contaminants ingested by humans. These models cover multi-level environmental systems, from local regions to global scales.

[0005] However, most existing models focus on small and medium scales, making it difficult to fully characterize the distribution dynamics and fate characteristics of new pollutants in large-scale spatial environmental systems. Although significant progress has been made in the research of environmental multi-media models, their development still faces several key challenges: the parameters required to construct environmental multi-media spatial differentiation models include pollutant environmental behavior parameters, environmental parameters, and emission rates. There are great difficulties in obtaining the environmental behavior parameters of pollutants, which directly restricts the development and application scope of the model; the current parameter prediction model system has significant heterogeneity, which seriously affects the convenience and universality of its application. The lack of emission rate data leads to incomplete model input parameters, affecting the reliability of the prediction results; at the same time, existing models can usually only make predictions for a single or a few compounds, and the prediction ability for multiple types of substances is still limited. The existence of these problems limits the application effect of the model in actual environmental management. Summary of the Invention

[0006] To solve the above problems, the present application provides a simple, fast and efficient method for predicting the concentration of pollutants, establishes an environmental multi-medium spatial differentiation model for various types of new pollutant concentration prediction, can simulate the environmental behavior of new pollutants, and obtain the spatial multi-medium concentration distribution characteristics, and provides the necessary basic data for chemical risk assessment and management.

[0007] The technical scheme of the present application is as follows:

[0008] The pollutant concentration prediction method based on the environmental multi-medium spatial differentiation model has the following steps:

[0009] Step (1) establishing an environmental behavior parameter prediction model

[0010] Collect ten key behavior parameter data of multiple pollutants in environmental media, including: vapor pressure (Vapor Pressure, VP), Henry's law constant (Henry's Law Constant, HL), bioconcentration factor (Bioconcentration Factor, BCF), octanol / air partition coefficient (Octanol / Air Partition Coefficient, K OA ), soil sediment adsorption coefficient (Soil Organic Carbon-Water Partition Coefficient, K OC ), octanol / water partition coefficient (Octanol / Water Partition Coefficient, K OW ), half-life of air, water, soil and sediment (t 1 / 2air ,t 1 / 2water ,t 1 / 2soil ,t 1 / 2sed ). The collected data is preprocessed, and the preprocessed data set is divided into a training set and a test set, wherein the training set accounts for 80%, and the test set accounts for 20%. Calculate the extended connectivity molecular fingerprint (ECFP) of the pollutants as the feature representation of the molecular structure.

[0011] An artificial neural network (ANN) algorithm was used to construct an environmental behavior parameter prediction model. Specifically, a multi-branch neural network model was constructed, including a shared input layer and ten independent parallel branch output layers. The shared input layer receives the ECFP molecular fingerprints of multiple pollutants as input. Each independent branch network outputs an environmental behavior parameter, realizing ten environmental behavior parameters. The structure of each branch network includes: an input layer, multiple hidden layers, and an output layer. The hidden layer uses the LeakyReLU activation function. The model training uses the Adam optimizer, and the initial learning rate and learning rate decay are set. The mean square error (MSE) is used as the loss function for model training.

[0012] Step (2) Establishing an environmental parameter database

[0013] When constructing an environmental multi-media spatial differentiation model, the present invention collects environmental attribute parameters related to pollutant migration, including scale parameters, temperature parameters, organic carbon fraction, environmental phase density, migration rate parameters, subphase volume fraction, clearance ratio parameters, and flow matrices. The parameters used in the model include both single-valued parameters and parameters with spatial variation. Based on the required spatial resolution of the model, the spatial analysis module of the ArcGIS platform is used for interpolation processing to generate a matching spatial distribution dataset.

[0014] Among them, the present invention adopts the flow matrix method to simulate the horizontal transmission process between grids of different media (atmosphere, fresh water, seawater). For the atmospheric advection module, the air flow rate is calculated based on the annual average wind speed data, combined with the grid interface height and length parameters; the freshwater migration module estimates the freshwater flow rate based on the runoff direction and flow data; the ocean advection module calculates the ocean flow rate by integrating parameters such as ocean current velocity, water depth and interface scale, and finally realizes the water balance verification of each medium interface through the verification of the runoff into the sea. A hierarchical method is adopted for the processing of river data: for the main channels of first-level rivers, their flow direction and flow characteristics are accurately analyzed based on hydrological observation data; the remaining grid units realize the processing of freshwater flow matrix through the four-way diffusion algorithm (average of the upper, lower, left and right neighbors of the grid).

[0015] Step (3) Establishing an emission rate prediction model

[0016] The present invention obtains pollutant emission rate data by the following method:

[0017] Based on the total production and use data of target pollutants, spatial allocation is performed using GDP share, and spatial processing is performed using ArcGIS software. In terms of emission factor estimation, both the ERCs method and the AB table method can be used simultaneously. Through emission factor calculation, the emission amount of pollutants in the region to the atmosphere, water, and soil is obtained, and the emission rate is calculated based on the emission amount. The calculation formula is as follows:

[0018] M i = M x EF i,j (1)

[0019] M i,total =∑M i (2)

[0020] F i = U x EF i,j (3)

[0021] F i,total =∑F i (4)

[0022] U i = (U - F i ) x EF i,j (5)

[0023] U i,total =∑U i (6)

[0024]

[0025] Where: EF i,j is the emission factor of the new pollutant in different environmental media in the life cycle (dimensionless), i is the environmental medium, which is air, water or soil, and j is the production, processing or use process; M i is the amount of pollutant in environmental medium i in the production (kg); M is the production amount of pollutant (kg); F i is the amount of pollutant in environmental medium i in the manufacturing (kg); U is the use amount of pollutant (kg); U i is the amount of pollutant in environmental medium i in the use (kg); M i,total is the total amount of pollutant discharged into environmental medium i in the production process (kg), F i,total is the total amount of pollutant discharged into environmental medium i in the manufacturing process (kg); U i,total is the total amount of pollutant discharged into environmental medium i in the use process (kg); E i,total is the emission rate of the corresponding pollutant in a certain environmental medium (kg·h -1 ), and h is the number of hours of the calculation period.

[0026] Step (4) establishes the environmental multi-medium spatial differentiation model

[0027] After obtaining the environmental behavior parameters, environmental parameters and emission rates of the chemical pollutant, the environmental multi-medium spatial differentiation model of the new pollutant is established. The model will finally predict the concentrations of the new pollutant in seven environmental phases, namely air, vegetation, soil, freshwater, seawater, freshwater sediment and seawater sediment, respectively.

[0028] Because there is still a big gap between the real environment system and the simulated environment system, a series of assumptions are needed to simplify the environment system before building the model, which are as follows:

[0029] ①The environment system is composed of multiple subsystems, and each subsystem includes a main environment phase and several sub-phases.

[0030] ②The pollutants in the main environment phase are uniformly distributed at any time, and the fugacity of each sub-phase is equal at the same time.

[0031] ③The fugacity values of the main environment phases are different.

[0032] ④The chemical reactions in each environment phase are first-order reactions.

[0033] Based on the above assumptions, the material migration between units in the present invention only occurs between the same environment phases, and only through advection transmission without considering the influence of diffusion.

[0034] Under the condition of a certain non-equilibrium and steady state, modeling is carried out based on the III-level fugacity model, and the mass balance equation is used to calculate the concentration of new pollutants in each medium. The specific environment multi-medium spatial differentiation model is as follows:

[0035] For the atmospheric phase: f(1)·D T (1) = E(1) + f(2)·D(2,1) + f(3)·D(3,1) + f(4)·D(4,1) + f(5)·D(5,1)

[0036] For the vegetation phase: f(2)·D T (2) = E(2) + f(1)·D(1,2) + f(5)·D(5,2)

[0037] For the soil phase: f(3)·D T (3) = E(3) + f(1)·D(1,3) + f(2)·D(2,3)

[0038] For the freshwater phase: f(4)·D T (4) = E(4) + f(1)·D(1,4) + f(3)·D(3,4) + f(5)·D(5,3)

[0039] For the seawater phase: f(5)·D T (5) = E(5) + f(1)·D(1,5) + f(4)·D(4,5) + f(7)·D(7,5)

[0040] For the freshwater sediment phase: f(6)·D T (6) = E(6) + f(4)·D(4,6)

[0041] For seawater sediment phase: f(7) D T (7) = E(7) + f(5) D(5,7)

[0042] Wherein, f(x) is fugacity (Pa) of pollutants in environmental phase x, x = 1-7, 1-7 represent atmosphere, vegetation, soil, fresh water, seawater, fresh water sediment, seawater sediment respectively; E(x) is the emission rate of pollutants in environmental phase x (mol / h); D T (x) is the total D value of pollutants removed from the environmental phase (mol / Pa h -1 ); D(x, y) is the D value of interphase migration of pollutants from environmental phase x to environmental phase y (mol / Pa h -1 ).

[0043] In the calculation of the environmental multi-medium space differentiation model, first, based on the environmental process, the corresponding environmental phase and environmental parameters are determined; according to the environmental behavior parameters of pollutants solved by the neural network model and the emission rate calculated by the emission rate model, the fugacity capacity Z i and the migration coefficient D of pollutants in the environmental migration process in each link phase are calculated, and then the fugacity f i is solved by establishing a mass balance equation group. The previous calculation results are taken as new input parameters into the model, so as to obtain the updated environmental concentration distribution of pollutants. The process realizes system balance through iterative calculation, and finally the environmental distribution of pollutants under steady state condition is obtained.

[0044] In the initial iterative calculation, the input of new pollutants in the environmental phase only considers local emission, and the fugacity value of each environmental phase is calculated. From the second iteration, the input of new pollutants needs to consider not only local emission but also the input flux brought by the air phase and water phase of other regions. The stopping condition of simulation calculation is: when the relative deviation of the results of two consecutive iterations is less than the preset threshold value, the model operation can be terminated. At this time, the simulation results output represent the spatial distribution characteristics and occurrence state of new pollutants in the environmental medium of the target research area when reaching dynamic balance.

[0045] The beneficial effects of the present application are as follows:

[0046] 1. Using molecular fingerprint coding combined with multi-task learning strategy, a plurality of key environmental behavior parameters are predicted at the same time, which greatly improves the parameter prediction efficiency and accuracy, and provides key input for subsequent multi-medium environmental model.

[0047] 2. Deep combination of machine learning model and fugacity model is realized, and parameter prediction and pollutant migration and simulation are integrated, which provides a more comprehensive technical framework for environmental exposure prediction. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 is an environmental multi-medium space differentiation model grid environment process schematic diagram;

[0049] Figure 2 is a neural network model schematic diagram;

[0050] Figure 3 is a neural network model prediction effect diagram of Henry constant. DETAILED DESCRIPTION

[0051] The specific embodiments of the present application are further illustrated below in combination with the drawings and technical solutions.

[0052] This example takes di(2-ethylhexyl) phthalate as an example to predict its concentration in seven environmental phases, and the specific steps are as follows:

[0053] Step 1, prediction of ten environmental behavior parameters

[0054] The relevant data of environmental behavior parameters were obtained by collecting EPI suite database and related literature to constitute the data set of the model. The data was preprocessed, including data screening and normalization. The data set was divided, and the training set ratio was 80% and the test set was 20%. The model used ECFP2(1, 2048) as the chemical molecular feature input, selected Adam optimizer, and the initial learning rate was set to 1x10 -3 , the learning rate decay rate was 5x10 -5 , and the model training epoch was 180. Each branch network of the model was composed of an input layer (2048 neurons), a first hidden layer (512 neurons), a second hidden layer (256 neurons), a third hidden layer (128 neurons), and an output layer (1 neuron), and the model structure was as shown in Figure 2 .

[0055] The prediction effect of the neural network model on the environmental behavior parameters is that the determination coefficients R 2 of the model for the training set and test set data of the Henry constant parameter are 0.966 and 0.723, respectively, as shown in Figure 3The model prediction results are similar to the true result distribution, and the relative error is basically less than 1, of which the maximum relative error is less than 4. The determination coefficients of the model for the training set and test set data of the octanol-air distribution coefficient parameter are 0.953 and 0.878 respectively, and the maximum relative error is not more than 1.5. The determination coefficients of the model for the training set and test set data of the organic carbon distribution coefficient parameter are 0.928 and 0.766 respectively, and the distribution trend of the training set and test set data is the same, and the relative error is less than 0.5 except for individual outliers. The determination coefficients of the model for the training set and test set data of the octanol-water distribution coefficient parameter are 0.926 and 0.774 respectively, and the data distribution of the training set and test set is consistent. The determination coefficients of the model on the training set and test set of vapor pressure parameter are 0.921 and 0.768 respectively. The determination coefficients R 2 of the model for the training set and test set of biological enrichment constant are 0.919 and 0.822 respectively, the prediction results of the data set are the same as the true result distribution, and the relative error is less than 1. The determination coefficients R 1 / 2air of the model for the training set of air, water, soil and sediment half-life (t 1 / 2water , t 1 / 2soil , t 1 / 2sed ) are 0.936, 0.914, 0.949 and 0.926 respectively. The determination coefficients R 2 of the test set are 0.616, 0.711, 0.832 and 0.827 respectively. 2

[0056] Step 2, prediction of the emission rate of di(2-ethylhexyl) phthalate

[0057] Based on the production and use data of target pollutants, combined with the characteristics of positive correlation between pollutant emissions and GDP, the total emissions are allocated to each calculation unit by the proportion of GDP of each grid. The GDP data of each calculation unit comes from the 2023 national GDP distribution data set processed by ArcGIS software. In the calculation of emissions, the emission factors are estimated by ERCs method and A-B table method.

[0058] ​The gridded emission rate of the compound di(2-ethylhexyl) phthalate (DEHP) in the study area was calculated. The production and use of DEHP is about 1.2 million tons per year. Its emission factors are: during the production process, the emission factor to the atmosphere is 0.05, the emission factor to water is 0.06, and the emission factor to soil is 0.0001; during the processing process, the emission factor to the atmosphere is 0.3, the emission factor to water is 0.002, and the emission factor to soil is 0.0001; during the use process, the emission factor to the atmosphere is 0.0005, the emission factor to water is 0.032, and the emission factor to soil is 0.032. After calculation, the gridded emission rate of DEHP in the study area is: 0 to 3.32×10 3 mol·h -1 , with an average value of 2.74×10 mol·h -1 , the median is 9.98×10 -2 mol·h -1 ; Discharge to water bodies 0~7.41×10 2 mol·h -1 , with an average value of 6.12 mol·h -1 , the median is 2.23×10 -2 mol·h -1 ; Emission to soil 0~3.32×10 3 mol·h -1 , with an average value of 1.34 mol·h -1 , the median is 4.86×10 -3 mol·h -1 .

[0059] Step 3: Prediction of the environmental concentration of di(2-ethylhexyl) phthalate

[0060] For the compound di(2-ethylhexyl) phthalate (DEHP), three parameters (environmental behavior parameter, environmental parameter and emission rate parameter) were obtained as input to calculate the environmental concentration of DEHP in various environmental media in the study area.

[0061] The structural information of DEHP was input into the trained multi-task model to simulate the Henry's constant, n-octanol / air partition coefficient, organic carbon / water partition coefficient, n-octanol / water partition coefficient, vapor pressure, bioconcentration coefficient, and the logarithmic values ​​of the half-life in the atmosphere, water, soil and sediment, which were -5.79, 8.47, 4.71, 4.49, -6.58, 3.00, 1.81, 2.80, 4.02 and 3.46, respectively.

[0062] The single environmental parameter mainly contains the volume fraction of each medium, organic carbon content, organic carbon content, molecular diffusion rate, etc. The values are obtained from various literatures, and the environmental parameter differentiation value data is from the Resource Environment Science and Data Center (RESDC).

[0063] The annual production and use of DEHP are obtained, and the emission rate is calculated according to the emission factor.

[0064] The obtained parameters are input into the environmental multi-medium space differentiation model (as shown in Figure 1 ), and the concentration value distribution of DEHP in each medium in the study area is calculated. The model predicts that the concentration range of DEHP is: the concentration in the atmospheric phase is: the maximum value is 1.84×10 5 ng·m -3 , and the average value is 1.16×10 3 ng·m -3 ; the concentration in the plant phase is: the maximum value is 1.19×10 3 ng·g -1 , and the average value is 1.33×10ng·g -1 ; the concentration in the soil phase is: the maximum value is 4.34×10 6 ng·g -1 , and the average value is 3.06×10 4 ng·g -1 ; the concentration in the freshwater phase is: the maximum value is 2.90×10 7 ng·L -1 , and the average value is 1.32×10 5 ng·L -1 ; the concentration in the seawater phase is: the maximum value is 7.30×10 5 ng·L -1 , and the average value is 9.19×10 2 ng·L -1 ; the concentration in the freshwater sediment phase is: the maximum value is 1.56×10 2 ng·g -1 , and the average value is 1.56×10 2 ng·g -1 ; the concentration in the seawater sediment phase is: the maximum value is 9.98×10 3 ng·g -1 , and the average value is 1.14×10ng·g -1 . The error between the model calculation result and the measured value is less than one order of magnitude, the simulated concentration situation is consistent with the true situation, and the model has good reliability.

[0065] This invention explores and optimizes the prediction of pollutant environmental behavior parameters and the construction and application of multi-media environmental models, and proposes a comprehensive method that can be used to predict the environmental fate of new pollutants. This method not only improves the prediction efficiency of pollutant environmental fate, but also provides a more intelligent and convenient technical means for environmental exposure assessment.

Claims

1. A pollutant concentration prediction method based on an environmental multi-media spatial differentiation model is characterized by: Here are the steps: Step (1) Establishing an environmental behavior parameter prediction model Collect data on ten key behavioral parameters of multiple pollutants in environmental media, including: vapor pressure VP, Henry constant HL, bioconcentration factor BCF, and n-octanol / air partition coefficient K OA , soil sediment adsorption coefficient K OC , n-octanol / water partition coefficient K OW , half-life of atmosphere, water, soil and sediment 1 / 2air ,t 1 / 2water ,t 1 / 2soil ,t 1 / 2sed ; Preprocess the collected data and divide the preprocessed data set into training set and test set; calculate the extended connectivity molecular fingerprint ECFP of the pollutants as a characteristic representation of the molecular structure; An artificial neural network (ANN) algorithm is used to construct an environmental behavior parameter prediction model. Specifically, a multi-branch neural network model is constructed, including a shared input layer and ten independent parallel branch output layers. The shared input layer receives the ECFP molecular fingerprints of multiple pollutants as input. Each independent branch network outputs one environmental behavior parameter, realizing ten environmental behavior parameters. The structure of each branch network includes: an input layer, multiple hidden layers, and an output layer. Step (2) Establishing an environmental parameter database Collect environmental attribute parameters related to pollutant migration, including scale parameters, temperature parameters, organic carbon fraction, ambient phase density, migration rate parameters, subphase volume fraction, removal ratio parameters, and flow matrix; Step (3) Establish an emission rate prediction model The emission factor is calculated to obtain the amount of pollutants emitted to the atmosphere, water, and soil in the region, and the emission rate is calculated based on the emission amount. The calculation formula is as follows: M i = M×EF i,j (1) M i,total =∑M i (2) F i =U × EF i,j (3) F i,total =∑F i (4) IN i =(UF i )×EF i,j (5) IN i,total =∑U i (6) Among them: EF i,j M is the emission factor of the new pollutant under study in different environmental media during its life cycle, i is the environmental medium, which can be air, water or soil, and j is the production, processing and use process; i is the amount of pollutants in the environmental medium i under production conditions; M is the production amount of pollutants; F i is the amount of pollutants in the environmental medium i under manufacturing conditions; U is the amount of pollutants used; U i M is the amount of pollutants in the environmental medium i under the condition of use; i,total is the total amount of pollutants discharged into the environmental medium i during the production process, F i,total is the total amount of pollutants discharged into the environmental medium i during the manufacturing process; U i,total is the total amount of pollutants discharged into the environmental medium i during the manufacturing process; E i,total is the emission rate of the corresponding pollutant in a certain environmental medium, and h is the number of hours in the calculation period; Step (4) Constructing an environmental multi-media spatial differentiation model The model predicts the concentrations of new pollutants in seven environmental phases, namely, atmosphere, vegetation, soil, freshwater, seawater, freshwater sediments, and seawater sediments; Modeling is performed based on the Level III fugacity model, and the mass balance equation is used to calculate the concentration of new pollutants in each medium. The details of the environmental multi-media spatial differentiation model are as follows: For atmospheric phase: f(1)·D T (1)=E(1)+f(2)·D(2,1)+f(3)·D(3,1)+f(4)·D(4,1)+f(5)·D(5,1)For vegetation phase: f(2)·D T (2)=E(2)+f(1)·D(1,2)+f(5)·D(5,2) For soil phase: f(3)·D T (3)=E(3)+f(1)·D(1,3)+f(2)·D(2,3) For freshwater phase: f(4)·D T (4)=E(4)+f(1)·D(1,4)+f(3)·D(3,4)+f(5)·D(5,3) For seawater phase: f(5)·D T (5)=E(5)+f(1)·D(1,5)+f(4)·D(4,5)+f(7)·D(7,5) For freshwater sediment phase: f(6)·D T (6) = E(6) + f(4)·D(4,6) For seawater sediment phase: f(7)·D T (7) = E(7) + f(5)·D(5,7) Where f(x) is the fugacity of pollutants in environmental phase x, x = 1 to 7, 1 to 7 represent the atmosphere, vegetation, soil, fresh water, sea water, fresh water sediment, and sea water sediment respectively; E(x) is the emission rate of pollutants in environmental phase x; D T (x) is the total D value of the pollutants removed from the environmental phase; D(x,y) is the D value of the interphase migration of pollutants from environmental phase x to environmental phase y.

2. The pollutant concentration prediction method based on the environmental multi-media spatial differentiation model according to claim 1 is characterized in that: The hidden layer in the network model in step (1) uses the LeakyReLU activation function; the model training uses the Adam optimizer, and sets the initial learning rate and learning rate decay; the model training uses the mean square error as the loss function.

3. The pollutant concentration prediction method based on the environmental multi-media spatial differentiation model according to claim 1 is characterized in that: In step (2), the flow matrix method is used to simulate the horizontal transmission process between different media such as atmosphere, fresh water and seawater grids; for the atmospheric advection module, the air flow rate is calculated based on the annual average wind speed data and the grid interface height and length parameters; the freshwater migration module estimates the freshwater flow rate based on the runoff direction and flow data; the ocean advection module calculates the ocean flow rate by integrating parameters such as ocean current velocity, water depth and interface scale, and finally realizes the water balance verification of each medium interface through the verification of the runoff into the sea; a hierarchical method is used for the processing of river data: for the main channel of the first-level river, its flow direction and flow characteristics are accurately analyzed based on the hydrological observation data; the remaining grid units realize the processing of the freshwater flow matrix through the four-way diffusion algorithm.

4. The pollutant concentration prediction method based on the environmental multi-media spatial differentiation model according to claim 1 is characterized in that: In step (4), before building the model, a series of assumptions need to be made on the environmental system to simplify it, as follows: ① The environmental system is composed of multiple subsystems, each of which includes the main environmental phase and several sub-phases; ② The pollutants in the main environmental phase are evenly distributed at any time, and the fugacity of each sub-phase is equal at the same time; ③ The fugacity values ​​are different between the main environmental phases; ④The chemical reactions in each environmental phase are all first-order reactions.

5. The pollutant concentration prediction method based on the environmental multi-media spatial differentiation model according to claim 1 is characterized in that: In step (4), when calculating the environmental multi-media spatial differentiation model, first determine the corresponding environmental phase and environmental parameters based on the environmental process; then calculate the fugacity capacity Z in each link phase based on the pollutant environmental behavior parameters solved by the neural network model and the emission rate calculated by the emission rate model. i and the migration coefficient D of pollutants in the environmental migration process, and then solve the fugacity f by establishing a mass balance equation group i ; Substitute the previous calculation results as new input parameters into the model to obtain the updated pollutant environmental concentration distribution; This process achieves system balance through iterative calculation and finally obtains the pollutant environmental distribution under steady-state conditions; During the initial iterative calculation, the input of new pollutants in the environmental phase only considers local emissions, and calculates the fugacity values ​​of each environmental phase; from the second iteration onwards, the input of new pollutants needs to consider both local emissions and the input flux brought by air and water phase advection in other regions; the condition for stopping the simulation calculation is: when the relative deviation of the calculation results of two consecutive iterative calculations is less than the preset threshold, the model operation can be terminated; the simulation results output at this time represent the spatial distribution characteristics and occurrence state of the new pollutants when they reach dynamic equilibrium in the environmental medium of the target study area.

Images