Industrial park perfluorinated-polyfluorinated compound emission prediction method based on deep learning

Abstract

The invention relates to the field of emission of perfluorinated-polyfluorinated compounds, in particular to an industrial park perfluorinated-polyfluorinated compound emission prediction method based on deep learning. Comprising the following steps: S1, determining and collecting data required by industrial park perfluorinated-polyfluorinated compound emission prediction accounting; s2, calculating the emission of perfluorinated-polyfluorinated compounds in the production, processing, use and disposal processes of the industrial park by adopting a material flow method; s3, establishing a perfluorinated-polyfluorinated compound emission prediction framework in the percolate based on deep learning; and S4, in combination with the convolutional neural network model, obtaining the emission characteristics of the industrial park perfluorinated compounds under the production system through matrix calculation. According to the method, based on the roll machine neural network model, through the coupling graph data and the deep learning algorithm, the multi-branch space-time network is fused, and multiple driving factors influencing the emission concentration of the perfluorinated-polyfluorinated compound are analyzed, so that the space-time change characteristics of the perfluorinated-polyfluorinated compound are comprehensively reflected, and the prediction precision and effect are improved.

Description

Technical Field

[0001] The present invention relates to the field of perfluoro-polyfluoro compound emissions, and in particular to a method for predicting perfluoro-polyfluoro compound emissions in industrial parks based on deep learning. Background Art

[0002] Perfluoro-polyfluoro compounds (PFAs) are a class of synthetic chemicals that are highly durable and stable due to their water- and oil-repellent properties and are widely used in industries such as papermaking, electroplating, and textiles. PFAs are often referred to as "permanent chemicals" that are difficult to metabolize or biodegrade in the human body. In addition, PFAs are associated with endocrine disrupting effects and may have adverse effects on reproductive health and fetal development. However, due to the diverse types and wide range of uses of PFAs, as well as the incompleteness of relevant regulations, a large amount of PFAs can still be detected in the environment. Therefore, tracking the source emission characteristics of PFAs is of great significance for providing reasonable source reduction recommendations.

[0003] Previous studies on the prediction of PFAS emissions were mainly based on the assessment of actual site concentrations in the receiving environment, which can preliminarily calculate the distribution characteristics of PFAS in the receiving environment medium. However, PFAS data is an unstructured irregular graph data, and traditional data-driven methods are difficult to achieve high-precision and strong stability predictions based on the integration of the spatiotemporal correlation and influencing factor characteristics of multiple potential emission sources, which limits the accurate prediction of PFAS distribution in the short term. Summary of the invention

[0004] In response to the above technical problems, the present invention provides a method for predicting the emission of perfluoro-polyfluoro compounds in industrial parks based on deep learning. Based on the convolution neural network model, by coupling graph data and deep learning algorithm, a multi-branch spatiotemporal network is integrated to analyze the multiple driving factors affecting the emission concentration of perfluoro-polyfluoro compounds, thereby comprehensively reflecting the spatiotemporal variation characteristics of perfluoro-polyfluoro compounds and improving the prediction accuracy and effect.

[0005] The method for predicting the emission of perfluorinated and polyfluorinated compounds in industrial parks based on deep learning includes the following steps:

[0006] S1. Determine and collect the data required for the forecast and accounting of PFAS emissions in industrial parks;

[0007] Data list collection:

[0008] The first part is the collection of production and processing volumes of chemical enterprises in the industrial park and basic data of economic sectors, the use of raw and auxiliary materials, the use of substances in the production links involving perfluorinated and polyfluorinated compounds, and the concentration, detection rate and substance category of perfluorinated and polyfluorinated compounds in each production link;

[0009] The second part is the meteorological data of the industrial park during the same period (temperature, humidity, rainfall, altitude);

[0010] The production and processing volumes, basic data of economic sectors, usage of raw and auxiliary materials, and usage of substances in the production links involving perfluoro-polyfluoro compounds in the first part are derived from specific survey statistical data of enterprises in the park, while the concentrations, detection rates, and substance categories of perfluoro-polyfluoro compounds in each production link are derived from actual sampling site measurement data. The data in the second part are all from the National Climate Data Center. The time range of the meteorological monitoring site data is determined based on the survey and actual sampling time, with a time resolution of 1h.

[0011] S2, using the material flow approach to account for PFAS emissions from production, processing, use and disposal in industrial parks;

[0012] Quantitative analysis of the occurrence characteristics of PFAS in industrial processes, including four elements: substances, processes, inventories and flows. Emphasis is placed on tracking the migration paths of PFAS and analyzing their environmental impacts.

[0013] (1) Analytical framework

[0014] According to the division of production processes of different enterprises into multiple subsystems, a full-process perfluorinated-polyfluorinated compound material flow analysis framework is established, with the complete production process as the geographical boundary and the time set as one production day.

[0015] (2) Acquisition of enterprise production process data

[0016] Collect basic information of industry enterprises in the park, such as pollution discharge permits, environmental impact assessment information, etc., and comprehensively understand the information on raw and auxiliary materials, production processes, three wastes generation and treatment facilities, emission routes and surrounding receiving environment of industry enterprises, and summarize the development profile of the industry in the autonomous region in recent years.

[0017] (3) Determination of perfluorinated and polyfluorinated compounds

[0018] Take 0.2-0.5L or 0.5g of wastewater, waste residue and other samples collected from the process flow and add 50μL of internal standard solution. Install the solid phase extraction column on the solid phase extraction device, add 4mL of 0.1% ammonia / methanol solution, 4mL and 4mL of pure water in turn, and keep the column head moist during the addition of solvent. Connect the universal adapter to the top of the solid phase extraction column, absorb the sample under negative pressure through the pipeline, and control the flow rate at about 1 drop / second. Elute the solid phase extraction column with 4mL, 25mmol / L ammonium acetate aqueous solution. At this time, various perfluoro-polyfluoro compounds are fixed on the solid phase extraction column, and the previous samples and eluents are discarded. After the solid phase extraction column is dried at low temperature with a vacuum freeze dryer (other effective methods such as nitrogen blow dryer and vacuum can also be used), elute with 4mL of methanol and 4mL of ammonia / methanol solution in turn, and collect the eluent with a 15mL polypropylene centrifuge tube. The collected eluent was concentrated with nitrogen on a nitrogen blow dryer, made up to 1 mL with methanol, passed through a 0.22 μm filter membrane and placed in a 1.5 mL brown injection bottle, stored at 4°C, and analyzed using HPLC-MS / MS.

[0019] (4) Material balance calculation method

[0020] From the material flow analysis framework, it can be seen that the perfluorinated and polyfluorinated compounds entering the production system mainly leave the system in three forms: perfluorinated and polyfluorinated compounds leaving with the effluent, perfluorinated and polyfluorinated compounds carried away by the waste residue, and perfluorinated and polyfluorinated compounds converted into other forms through material action. The equilibrium percentage equation of perfluorinated and polyfluorinated compounds is obtained as follows:

[0021]

[0022] Where M inf,PFAS 、M eff,PFAS 、M S,PFAS 、M denit,PFAS They are the mass of perfluorinated and polyfluorinated compounds entering production, ending the process, entering waste slag, and consumed in material conversion, respectively.

[0023] S3, establish a deep learning-based prediction framework for PFAS emissions in leachate;

[0024] (1) Data preparation: Collect and organize the input data and target values ​​(i.e., labels) required for regression prediction. Preprocess the data, such as normalizing or standardizing, to ensure that the data is within the same range to facilitate network training.

[0025] (2) Model construction: Construct a convolutional neural network model, which usually includes an input layer, several convolutional layers, a pooling layer, a fully connected layer, and an output layer. The convolutional layer is used to extract features from the data, and the pooling layer is used to reduce the size of the feature map, thereby reducing computational complexity and preventing overfitting.

[0026] (3) Model training: The preprocessed data is input into the constructed CNN model. The difference between the model prediction value and the actual value is evaluated through the defined loss function (such as mean square error). The model parameters are adjusted using an optimization algorithm (such as gradient descent) so that the value of the loss function gradually decreases, thereby improving the prediction accuracy of the model.

[0027] (4) Model evaluation: Use the validation data set to evaluate the trained model to test its prediction performance and generalization ability. Use some evaluation indicators (such as mean square error and mean absolute error) to quantify the prediction effect of the model.

[0028] (5) Model optimization: Based on the evaluation results, adjust and optimize the model, which may require adjusting the network structure, optimization algorithm or hyperparameters. Repeat training and evaluation until the model performance reaches the expected standard.

[0029] S4, combined with the convolutional neural network model (CNN) through matrix calculation, the emission characteristics of perfluorinated compounds in the industrial park under the production system were obtained.

[0030] Use the trained CNN model to perform regression prediction on new data and output the predicted value. Post-process or apply the prediction results as needed.

[0031] The technical effects of the present invention are as follows:

[0032] (1) The present invention uses a convolutional neural network model, couples graph data with a deep learning algorithm, integrates a multi-branch spatiotemporal network, analyzes the multiple driving factors that affect the emission concentration of perfluoro-polyfluoro compounds, and comprehensively reflects the spatiotemporal variation characteristics of perfluoro-polyfluoro compounds.

[0033] (2) Compared with the traditional method of predicting the distribution characteristics of perfluoro-polyfluoro compounds in receiving environmental media, the perfluoro-polyfluoro compound emission prediction model constructed by this technology can carry out high-precision and strong stability prediction of perfluoro-polyfluoro compound emission concentrations integrating multi-factor spatiotemporal correlations. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a flow chart of step S3 of the present invention;

[0035] Figure 2 Comparison between the actual value and the predicted value of the training set of the embodiment;

[0036] Figure 3 The actual value and predicted value results of the test set of the embodiment are compared. DETAILED DESCRIPTION

[0037] The specific technical solution of the present invention is explained in conjunction with embodiments.

[0038] In a chemical park in Xinjiang Uygur Autonomous Region, we collected basic information about the enterprises in the industry in the park, such as pollution discharge permits and environmental impact assessment materials, to comprehensively understand the raw and auxiliary materials, production processes, three wastes generation and treatment facilities, emission pathways, and surrounding receiving environment of the enterprises, and summarize the development of the industry in the autonomous region in recent years. We completed the collection of production and processing volumes of chemical enterprises in the industrial park and basic data of economic departments, as well as the use of raw and auxiliary materials and materials used in the production links of perfluoro-polyfluoro compounds. At the same time, we determined the time range of meteorological monitoring station data based on the survey and actual sampling time, with a time resolution of 1h, and obtained meteorological data of the industrial park in the same period (including temperature, humidity, rainfall, and altitude). We collected wastewater and waste residue samples in each production link according to the production process of the enterprises in the industry in the park.

[0039] After sample collection, water samples were stored in brown glass sampling bottles, adjusted to pH = 2-4, and sealed; waste residue samples were stored in aluminum foil bags and transported back to the laboratory at 4°C and protected from light. After sample collection, sample analysis was completed within 7 days to obtain the concentration, detection rate, and substance category data of perfluoro-polyfluoro compounds in each production link. The test data are shown in Table 1 below:

[0040] Table 1 Detection data

[0041]

[0042] According to the material flow analysis framework, the material flow of perfluorinated and polyfluorinated compounds entering the production system is calculated, and the main characteristic influencing factors affecting perfluorinated and polyfluorinated compounds are clarified. The collected characteristic values ​​and perfluorinated and polyfluorinated compound emission data are input into MATLAB R2023b software, and the data is preprocessed and analyzed using the model, divided into training sets (70% data set) and test sets (30% data set), and the data is normalized at the same time. The data is flattened into 1D data, the network structure is constructed, and the parameters are set to train the model. Figure 1 As shown. The accuracy of the prediction model is judged based on the data of R2 and RMSE. Finally, the following is obtained: Figure 2 and Figure 3 Comparison of model prediction results shown.

[0043] Table 2

[0044]

[0045]

[0046] The emission of perfluorinated and polyfluorinated compounds can be accurately predicted based on the characteristic values, as shown in Table 2.

Claims

1. A method for predicting the emission of perfluorinated and polyfluorinated compounds in industrial parks based on deep learning, characterized in that: The following steps are involved: S1. Determine and collect the data required for the forecast and accounting of PFAS emissions in industrial parks; S2, using the material flow approach to account for PFAS emissions from production, processing, use and disposal in industrial parks; Quantitative analysis of the occurrence characteristics of PFAS in industrial processes, including four elements: substances, processes, inventories and flows; Focus on tracking the migration pathways of perfluorinated and polyfluorinated compounds and analyzing their environmental impacts; S3, establish a deep learning-based prediction framework for PFAS emissions in leachate; S4, combined with the convolutional neural network model CNN, the emission characteristics of perfluorinated compounds in the industrial park under the production system were obtained through matrix calculation; Use the trained CNN model to perform regression prediction on new data and output the predicted value; Post-process or apply the prediction results according to needs.

2. The method for predicting emissions of perfluoro-polyfluoro compounds in industrial parks based on deep learning according to claim 1, characterized in that: S1 specifically includes: Data list collection: The first part is the collection of production and processing volumes of chemical enterprises in the industrial park and basic data of economic sectors, the use of raw and auxiliary materials, the use of substances in the production links involving perfluorinated and polyfluorinated compounds, and the concentration, detection rate and substance category of perfluorinated and polyfluorinated compounds in each production link; The second part is the meteorological data of the industrial park during the same period, including temperature, humidity, rainfall, and altitude; The production and processing volumes, basic data of economic sectors, usage of raw and auxiliary materials, and usage of substances in the production links involving perfluoro-polyfluoro compounds in the first part are derived from specific survey statistical data of enterprises in the park, while the concentrations, detection rates, and substance categories of perfluoro-polyfluoro compounds in each production link are derived from actual sampling site measurement data. The data in the second part are all from the National Climate Data Center. The time range of the meteorological monitoring site data is determined based on the survey and actual sampling time, with a time resolution of 1h.

3. The method for predicting emissions of perfluoro-polyfluoro compounds in industrial parks based on deep learning according to claim 1, characterized in that: S2 specifically includes: (1) Analytical framework; According to the production process of different enterprises, it is divided into multiple subsystems, and a material flow analysis framework of perfluorinated and polyfluorinated compounds in the whole process is established, with the complete production process as the geographical boundary and the time set as one production day; (2) Acquisition of enterprise production process data; Collect basic information about the industry enterprises in the park, and comprehensively understand the raw and auxiliary materials, production processes, three wastes generation and treatment facilities, emission pathways and surrounding receiving environment information of the industry enterprises, and summarize the development of the industry in the autonomous region in recent years; (3) Determination of perfluorinated and polyfluorinated compounds Take 0.2-0.5L or 0.5g of wastewater and waste residue samples collected from the process flow and add 50μL of internal standard solution; install the solid phase extraction column on the solid phase extraction device, and add 4mL 0.1% ammonia / methanol solution, 4mL and 4mL pure water, keep the column head moist during the addition of solvent; connect the universal adapter to the top of the solid phase extraction column, absorb the sample under negative pressure through the pipeline, and control the flow rate at about 1 drop / second; elute the solid phase extraction column with 4mL, 25mmol / L ammonium acetate aqueous solution; at this time, various perfluoro-polyfluoro compounds are fixed on the solid phase extraction column, and the previous samples and eluents are discarded; after the solid phase extraction column is dried at low temperature, it is sequentially elute with 4mL methanol and 4mL ammonia / methanol solution, and the eluent is collected in a 15mL polypropylene centrifuge tube; the collected eluent is concentrated with nitrogen on a nitrogen blow dryer, fixed to 1mL with methanol, passed through a 0.22μm filter membrane and placed in a 1.5mL brown injection bottle, stored at 4℃, and analyzed using HPLC-MS / MS; (4) Material balance calculation method From the material flow analysis framework, it can be seen that the perfluorinated-polyfluorinated compounds entering the production system leave the system mainly in three forms: the perfluorinated-polyfluorinated compounds leaving with the effluent, the perfluorinated-polyfluorinated compounds carried away by the waste residue, and the perfluorinated-polyfluorinated compounds converted into other forms through material action; the equilibrium percentage equation of the perfluorinated-polyfluorinated compounds is obtained as follows: Where M inf,PFAS 、M eff,PFAS 、M S,PFAS 、M denit,PFAS They are the mass of perfluorinated and polyfluorinated compounds entering production, ending the process, entering waste slag, and consumed in material conversion, respectively.

4. The method for predicting emissions of perfluoro-polyfluoro compounds in industrial parks based on deep learning according to claim 1, characterized in that: S3 specifically includes: (1) Data preparation: Collect and organize the input data and target values ​​(i.e., labels) required for regression prediction; preprocess the data to ensure that the data is in the same range to facilitate network training; (2) Model construction: Construct a convolutional neural network model, including an input layer, several convolutional layers, a pooling layer, a fully connected layer, and an output layer; the convolutional layer is used to extract features from the data, and the pooling layer is used to reduce the size of the feature map, thereby reducing computational complexity and preventing overfitting; (3) Model training: input the preprocessed data into the constructed CNN model; evaluate the difference between the model prediction value and the actual value through the defined loss function; use the optimization algorithm to adjust the model parameters so that the value of the loss function gradually decreases, thereby improving the prediction accuracy of the model; (4) Model evaluation: Use the validation data set to evaluate the trained model to test its prediction performance and generalization ability; use some evaluation indicators to quantify the prediction effect of the model; (5) Model optimization: Based on the evaluation results, adjust and optimize the model, which may require adjusting the network structure, optimization algorithm or hyperparameters; repeatedly train and evaluate until the model performance reaches the expected standard.

Images