A method for rapid detection of dioxin concentration based on chlorobenzene as indicator

Abstract

The application provides a solid-phase dioxin concentration rapid detection method based on chlorobenzene as an indicator, and relates to the technical field of dioxin concentration detection. The method comprises the following steps: S1, collecting a fly ash sample; S2, determining the chlorobenzene concentration and the dioxin concentration in the fly ash sample, and constructing sample data; S3, establishing a dioxin prediction model; S4, training the dioxin prediction model by using the sample data; and S4, measuring the chlorobenzene concentration of a fly ash to be detected, inputting the chlorobenzene concentration into the dioxin prediction model, and obtaining the dioxin concentration of the fly ash to be detected. The application provides a method which is high in accuracy, low in cost and easy to operate, and indirectly detects the dioxin concentration in fly ash by measuring the indicator concentration.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of dioxin concentration detection, and in particular to a solid-phase dioxin concentration rapid detection method based on chlorobenzene as an indicator. BACKGROUND

[0002] In 2023, the national urban domestic waste disposal volume reached 250 million tons, of which the incineration proportion exceeded 82.5%, and the environmental protection disposal problem of fly ash generated by domestic waste incineration has also become one of the main challenges for the sustainable development of the domestic waste incineration power generation industry. According to statistics, 30-50 kg of fly ash is generated for every 1 t of domestic waste incineration, and the annual fly ash generated by domestic waste incineration in China is 6.0x106-7.0x106 t. The fly ash contains a highly toxic substance, dioxin (PCDD / Fs), which has strong toxicity, stability, acid resistance and low solubility, and has the hazards of teratogenicity, carcinogenicity and mutagenicity to humans and other animals. In order to standardize the treatment and disposal of fly ash, promote the resource utilization of fly ash, improve the technical level of fly ash treatment and disposal, standardize and guide the environmental management of fly ash, in April 2019, the Solid Waste and Chemicals Division of the Ministry of Ecology and Environment initiated the preparation of HJ 1134-2020 “Technical Code for Pollution Control of Domestic Waste Incineration Fly Ash (Trial)”, which stipulates that the total amount of dioxin residues in the fly ash treatment product should not exceed 50 ng-TEQ / kg, and it was implemented on August 27, 2020.

[0003] Offline detection is currently the mainstream method for detecting PCDD / Fs in fly ash, and pretreatment of fly ash samples is required before entering the machine for detection. The results of laboratory offline detection are often lagging behind, and the pretreatment process is complex, the detection period is long, and the detection cost is high. SUMMARY

[0004] The application provides a solid-phase dioxin concentration rapid detection method based on chlorobenzene as an indicator to at least solve the above technical problems in the prior art.

[0005] According to a first aspect of the application, a solid-phase dioxin concentration rapid detection method based on chlorobenzene as an indicator is provided, comprising the following steps:

[0006] S1, collecting a fly ash sample;

[0007] S2, determining the chlorobenzene concentration and the dioxin concentration in the fly ash sample, and constructing sample data;

[0008] S3, establishing a dioxin prediction model;

[0009] S4, training the dioxin prediction model using the sample data;

[0010] S4, measuring the chlorobenzene concentration of the fly ash to be detected, inputting the chlorobenzene concentration into a dioxin prediction model to obtain the dioxin concentration of the fly ash to be detected.

[0011] In some embodiments of the first aspect of the application, in S1, the collected fly ash samples are sieved and divided.

[0012] In some embodiments of the first aspect of the application, in S2, the sample data is subjected to data cleaning, including missing value processing and outlier removal, and the data is subjected to normalization and / or standardization processing.

[0013] Normalization: ;

[0014] Standardization: ;

[0015] wherein, is the original data, is the normalized data, is the standardized data, and are the minimum value and the maximum value in the data set, respectively, is the mean value, is the standard deviation.

[0016] In some embodiments of the first aspect of the application, the method of missing value processing is as follows: if the sample missing rate is <5%, multiple imputation or KNN interpolation is used; if the missing rate is >5%, direct rejection or special working conditions are marked.

[0017] In some embodiments of the first aspect of the application, in S2, the sample data is subjected to small sample data enhancement based on a Monte Carlo model, including the following steps:

[0018] Step 21: Probability manifold modeling

[0019] Sub-step 21.1: Diffusion probability modeling is performed on the original small sample set to construct a stochastic differential equation:

[0020]

[0021] wherein, is a vector function, referred to as a drift factor, is a real number function, referred to as a diffusion factor, represents a standard Brownian motion, is an infinitesimal white noise;

[0022] wherein the drift term f learns the data manifold structure through a neural network;

[0023] Sub-step 21.2: Estimate the gradient field ∇x log pt(x) using score matching to capture the characteristics of multimodal distribution;

[0024] Step 22: Adaptive importance sampling

[0025] Sub-step 22.1: Build a hierarchical sampling framework, sample each region according to Mahalanobis distance

[0026] Sub-step 22.2: Dynamically adjust the sampling weight of each region;

[0027]

[0028] Where the effective sample size is the current weight, is the new weight;

[0029] Step 23: Tensor core acceleration

[0030] Sub-step 23.1: Split the high-dimensional data into d / 32 tensor blocks, and assign each tensor block to a GPU streaming multiprocessor;

[0031] Sub-step 23.2: Design a hybrid precision sampling algorithm, use FP16 for key calculation nodes, and use FP32 for probability accumulation.

[0032] In some embodiments of the first aspect of the application, the dioxin prediction model is a neural network model, and the network architecture of the neural network model comprises:

[0033] Input layer: the number of nodes is equal to the number of features;

[0034] Hidden layer: 2-3 layers, the number of neurons in each layer decreases according to the pyramid rule;

[0035] Output layer: single node, activation function is linear.

[0036] In some embodiments of the first aspect of the application, the loss function of the neural network model is mean square error or smooth L1 loss.

[0037] In some embodiments of the first aspect of the application, the neural network model is verified, including performance evaluation indicators, explainability analysis, and sensitivity analysis.

[0038] The performance evaluation indicators are as follows:

[0039] Regression indicators: MSE, RMSE, MAE, R²;

[0040]

[0041]

[0042] wherein n is the total amount of test set data, is the true value, is the predicted value, is the sample mean;

[0043] Residual analysis: residual-QQ plot is drawn to verify normality;

[0044] Residual-predicted value scatter plot is drawn to check heteroscedasticity;

[0045] The interpretability analysis is as follows:

[0046] Local interpretation: LIME is used to show the contribution of chlorobenzene to a single sample;

[0047] Global interpretation: Partial Dependence Plot is calculated to quantify the marginal effect of chlorobenzene concentration on dioxin concentration;

[0048] SHAP Summary Plot: feature importance and influence direction are visualized;

[0049] The sensitivity analysis is as follows:

[0050] Single variable perturbation: other variables are fixed, and the chlorobenzene concentration is gradually increased to observe the change gradient of the predicted value of dioxin concentration;

[0051] Adversarial sample test: noise is injected to evaluate the robustness of the model.

[0052] In some embodiments of the first aspect of the application, the dioxin prediction model is a linear regression model; the linear regression model uses the Linear Regression class of scikit-learn to initialize the linear regression model, divides the training set and the test set according to a ratio of 7:3, uses the test set to evaluate the model, and calculates the R² and MSE indicators.

[0053] In some embodiments of the first aspect of the application, the chlorobenzene concentration is the concentration of 1,3,5-trichlorobenzene or 1,2,3,5-tetrachlorobenzene.

[0054] Compared with the prior art, the application has the following beneficial effects:

[0055] The present application uses chlorobenzene (CBz) as an indicator to achieve rapid detection of dioxin concentration. Chlorobenzene (CBz) is a key component in catalytic reactions that generate dioxins, and its structure is similar to the benzene ring structure of PCDD / Fs, which can further form PCDD / Fs in the presence of high temperature or metal catalysts. Scientific research shows that the emission concentration of chlorobenzene (CBz) is 3 orders of magnitude higher than that of trace organic pollutants PCDD / Fs, and has a high correlation with PCDD / Fs. Therefore, when conducting rapid monitoring of PCDD / Fs, the present application selects CBz to indicate the associated PCDD / Fs I-TEQ concentration, which has the following advantages:

[0056] Cost-effective: using low-cost detection equipment and simplified operation process;

[0057] High efficiency: achieving rapid monitoring and instant feedback;

[0058] High accuracy: verified prediction model ensures the reliability of the results;

[0059] Safe and environmentally friendly: reducing the demand for and exposure risk of high-risk chemicals.

[0060] It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of the present application, nor to limit the scope of the present application. Other features of the present application will become readily apparent through the following description. BRIEF DESCRIPTION OF DRAWINGS

[0061] The above and other objects, features and advantages of the exemplary embodiments of the present application will be more apparent from the following detailed description read in conjunction with the accompanying drawings, in which:

[0062] In the drawings, identical or corresponding reference numerals indicate identical or corresponding parts.

[0063] Figure 1 A block diagram of the virtual sample generation algorithm of the present application based on a statistical model is shown.

[0064] Figure 2 A small sample data augmentation process based on a Monte Carlo model of the present application is shown.

[0065] Figure 3 A Pearson correlation coefficient thermogram between fly ash components and each chlorobenzene and dioxin of the present application is shown.

[0066] Figure 4 A correlation thermogram between each chlorobenzene and dioxin TEQ of the present application is shown.

[0067] Figure 5A heat map showing the correlation between the isomers of chlorobenzene and dioxin TEQ in this application is shown.

[0068] Figure 6 A linear model result of the total concentration of chlorobenzene and the concentration of dioxin TEQ in this application is shown.

[0069] Figure 7 A neural network model result one of the total concentration of chlorobenzene and the concentration of dioxin TEQ in this application is shown.

[0070] Figure 8 A neural network model result two of the total concentration of chlorobenzene and the concentration of dioxin TEQ in this application is shown. DETAILED DESCRIPTION

[0071] In order to make the purpose, features and advantages of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0072] The present embodiment provides a rapid detection method for solid-phase dioxin concentration based on chlorobenzene as an indicator, comprising the following steps:

[0073] S1, collect fly ash samples.

[0074] In the present embodiment, 6 municipal solid waste incineration fly ash samples are collected, and the results show that the dioxin TEQ concentrations of the 6 fly ash samples have significant differences, ranging from 230 to 1403 ng-TEQ / kg, which may be related to factors such as temperature, pressure and combustion conditions during the formation process. These differences reveal that the composition of different samples will affect their toxicity, providing a basis for subsequent dioxin concentration prediction and best indicator search.

[0075] In order to explore the best indicator, the above-mentioned fly ash samples are sieved and evenly divided, and then the chemical components of the fly ash samples are detected by XRF characterization method. The main elements of fly ash are O, Ca, Mg, Al, Cl, Na, Si, P, S, K, etc.

[0076] Then the fly ash samples are pretreated according to the method of China HJ77.3 2008, and the chlorobenzene concentration and dioxin (TEQ) concentration are detected by gas chromatograph and high-resolution mass spectrometer.

[0077] Finally, the dioxin (TEQ) concentrations, XRF spectrum information and chlorobenzene concentrations of the 6 samples are summarized, as shown in Table 1 below.

[0078] Table 1 Sample correlation analysis dataset [CBz concentration (ng / g), fly ash component (%), PCDD / Fs I-TEQ concentration (ng-TEQ / kg)]

[0079]

[0080] Correlation analysis was performed on each of the above characteristics, i.e. chemical components and CBz concentration, and distribution fitting was performed on the data of each characteristic. Common distributions include normal distribution, lognormal distribution, and Poisson distribution, etc. Statistical methods (such as Pearson correlation coefficient) were used to determine potential indicators with significant correlation with dioxins. The results are shown in Table 2. Figure 3 It is worth mentioning that input1-input16 are different characteristics in the first column of Table 1 in turn, and Output is the dioxin TEQ concentration. Analysis shows that the correlation of input7 to the result can be ignored, that is, there is no significant correlation between O element in XRF and other characteristics. The dioxin TEQ concentration as the output, Input9-Input12 and it are positively correlated, that is, Mg, Al, Si, P elements and dioxin TEQ have a higher positive correlation. Input8, Input13, Input14 and it are negatively correlated, that is, Na, S, Cl elements and dioxin TEQ concentration have a higher negative correlation.

[0081] Correlation analysis was performed on each homolog (X1-X6 are monochlorobenzene to hexachlorobenzene) in CBz and dioxin TEQ concentration (Y), and it was found that they were positively correlated, as shown in Table 3. Figure 4 Among them, the correlation between tetrachlorobenzene and dioxin TEQ concentration is the highest, with a correlation coefficient of 0.88; followed by trichlorobenzene, with a correlation coefficient of 0.78; the correlation between pentachlorobenzene and dioxin TEQ concentration is the worst, with a correlation coefficient of only 0.17.

[0082] S2, determine the CBz concentration and dioxin concentration in the fly ash sample, and construct the sample data;

[0083] According to the aforementioned correlation analysis, the CBz concentration and the dioxin concentration are positively correlated, so the CBz concentration and the dioxin concentration are used to construct the sample data here.

[0084] Further, the single substance correlation analysis of dioxin TEQ concentration was performed on the isomers in dichlorobenzene to tetrachlorobenzene with high correlation, as shown in Table 4. Figure 5 It was found that the correlation of 1,3,5-trichlorobenzene in trichlorobenzene was 0.89, and the correlation of 1,2,3,5-tetrachlorobenzene in tetrachlorobenzene was 0.90, indicating that further correlation research on single isomers will have higher accuracy of the model.

[0085] Therefore, the concentration of chlorobenzene is preferably the concentration of 1,3,5-trichlorobenzene or 1,2,3,5-tetrachlorobenzene.

[0086] It is worth mentioning that the sample data is also data cleaning, including missing value processing and outlier removal, and the data is normalized and / or standardized;

[0087] Normalization: Scale the data to the [0, 1] interval.

[0088] Standardization: Make the data conform to the distribution with mean 0 and standard deviation 1.

[0089] where, is the original data, is the normalized data, is the standardized data, and are the minimum and maximum values in the data set, is the mean, is the standard deviation.

[0090] The method of missing value processing is as follows: if the sample missing rate is <5%, use multiple imputation or KNN interpolation; if the missing rate is >5%, directly eliminate or mark as special working conditions.

[0091] Here, the sample data is also small sample data augmentation based on the Monte Carlo model,

[0092] Establish a Monte Carlo model data augmentation sample to improve the generalization and robustness of the small sample data model. Please refer to Figure 1 , a Monte Carlo simulation method is used to generate multiple possible dioxin concentration data by random sampling. First, based on the known fly ash characteristics and dioxin data, determine the key parameters that affect the dioxin concentration, and set reasonable ranges and distributions (such as normal distribution or uniform distribution) for these parameters. Then, through Monte Carlo simulation, randomly extract values from these parameter distributions, and input them into the dioxin prediction model (such as regression model or other prediction model). Finally, according to the parameter combination of each simulation, the corresponding dioxin concentration is calculated, and the sample containing data augmentation is generated for further analysis.

[0093] Please refer to Figure 2 , which includes the following steps:

[0094] Step 21: Probability manifold modeling

[0095] Sub-step 21.1: Diffusion probability modeling on the original small sample set X ∈ R n×d , construct a stochastic differential equation:

[0096] where, is a vector function, called drift factor, is a real function, called diffusion coefficient, denotes a standard Brownian motion, is an infinitesimal white noise.

[0097] Sub-step 21.2: Estimate the gradient field ∇x log pt(x) using Score Matching to capture the multi-modal distribution characteristics;

[0098] Step 22: Adaptive Importance Sampling (AIS)

[0099] Sub-step 22.1: Construct a hierarchical sampling framework, divide the feature space into core area (D M ≤ 2), transition area (2 < D M ≤ 4), and edge area (D M > 4) according to Mahalanobis distance;

[0100] The calculation method of Mahalanobis distance D M is as follows:

[0101] D M (x)=

[0102] where, represents the difference vector between a single data point and the center point, represents the mean vector of each variable of the data set, and the inverse covariance matrix is introduced to "transform" the possibly tilted and stretched data distribution in the original space to a standardized space.

[0103] Sub-step 22.2: Dynamically adjust the sampling weights of each area;

[0104]

[0105] where, the effective sample size is the current weight, is the new weight.

[0106] Step 23: Tensor Kernel Acceleration (TKA)

[0107] Sub-step 23.1: Split the high-dimensional data into ⌈d / 32⌉ tensor blocks, and assign each tensor block to a GPU Streaming Multiprocessor (SM);

[0108] Sub-step 23.2: Design a mixed-precision sampling algorithm, use FP16 for key calculation nodes and FP32 for probability accumulation.

[0109] S3, establishing a dioxin prediction model;

[0110] The dioxin prediction model is preferably a neural network model or a linear regression model.

[0111] I. The network architecture of the neural network model comprises:

[0112] Input layer: the number of nodes is equal to the number of features (e.g. chlorobenzene concentration + 3 derived features → 4 nodes);

[0113] Hidden layer: 2-3 layers, the number of neurons in each layer decreases according to the "pyramid rule" (e.g. 16→8→4);

[0114] Activation function: ReLU (hidden layer), Sigmoid / Tanh (if the output needs to be normalized);

[0115] ReLU function:

[0116] Output layer: single node, activation function is linear.

[0117] Regularization and optimization

[0118] Anti-overfitting strategy: L2 regularization (λ=0.01), Dropout (ratio=0.2);

[0119] L2 regularization:

[0120] Early stopping method (Early Stopping): monitor the validation set Loss, patience=10 epochs;

[0121] Optimizer configuration: Adam optimizer, initial learning rate=3e-4, dynamic adjustment (ReduceLROnPlateau);

[0122] The loss function of the neural network model is mean square error (MSE) or smooth L1 loss (Huber Loss, robust to outliers).

[0123] The neural network model is verified, including performance evaluation indicators, explainability analysis, and sensitivity analysis;

[0124] The performance evaluation indicators are as follows:

[0125] Regression indicators: MSE, R² (need >0.7);

[0126]

[0127] Where n is the total amount of test set data, is the true value, is the predicted value;

[0128]

[0129] where n is the total amount of test set data, is the true value, is the predicted value, is the sample mean;

[0130] Residual analysis: draw residual-QQ plot to verify normality (p>0.05, K-S test);

[0131] Residual-predicted value scatter plot to check heteroscedasticity;

[0132] The interpretability analysis is as follows:

[0133] Local interpretation: use LIME (Local Interpretable Model-agnostic Explanations) to show the contribution of chlorobenzene to individual samples;

[0134] Global interpretation: calculate Partial Dependence Plot to quantify the marginal effect of chlorobenzene concentration on dioxin concentration;

[0135] SHAP Summary Plot: visualize feature importance and impact direction;

[0136] The sensitivity analysis is as follows:

[0137] Univariate perturbation: fix other variables, gradually increase chlorobenzene concentration, and observe the change gradient of dioxin concentration predicted value;

[0138] Adversarial sample test: inject noise to evaluate the robustness of the model.

[0139] II. The linear regression model uses the Linear Regression class of scikit-learn to initialize the linear regression model, divides the training set and test set according to the ratio of 7:3, uses the test set to evaluate the model, and calculates R², MSE index.

[0140] And through this sample to establish the total chlorobenzene concentration and dioxin TEQ concentration between the linear mathematical relationship and neural network model, the linear mathematical relationship is: y = 0.558*x-410.775, R 2 =0.8728, see Figure 6 . Compared with the MLP neural network model, there is a certain gap between the prediction and the original value, please refer to Figure 7 and Figure 8 , the neural network model shows its R 2= 0.7847. Thus, it can be seen that the linear model performs better in small sample discrete data.

[0141] S4, training a dioxin prediction model by using the sample data;

[0142] S4, measuring the chlorobenzene concentration of the fly ash to be detected, inputting the chlorobenzene concentration into the dioxin prediction model, and obtaining the dioxin concentration of the fly ash to be detected.

[0143] It should be understood that the various forms of flow shown above can be used to reorder, add or delete steps. For example, each step described in the present application can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical solutions of the present application can be achieved, which is not limited herein.

[0144] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specifically limited.

[0145] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for rapid detection of dioxin concentration based on chlorobenzene as an indicator, characterized in that, The method comprises the following steps: S1, collecting fly ash samples; S2, determining the concentration of chlorobenzene and dioxin in the fly ash samples, and constructing sample data; S3, establishing a dioxin prediction model; S4, training the dioxin prediction model using the sample data; S4, measuring the chlorobenzene concentration of the fly ash to be detected, inputting the chlorobenzene concentration into the dioxin prediction model, and obtaining the dioxin concentration of the fly ash to be detected; In the S2, the sample data is subjected to small sample data enhancement based on a Monte Carlo model, comprising the following steps: Step 21: probability manifold modeling Substep 21.1: diffusion probability modeling of the original small sample set, and constructing a stochastic differential equation: wherein, is a vector function, called the drift factor, is a real function, called the diffusion factor, denotes a standard Brownian motion, is an infinitesimal white noise; Wherein, the drift term f learns the data manifold structure through a neural network; Sub-step 21.2: Estimate the gradient field with score matching x log pt(x), capturing the characteristics of multimodal distributions; Step 22: adaptive importance sampling Substep 22.1: constructing a hierarchical sampling framework, and dividing the feature space into a core area, a transition area and an edge area according to the Mahalanobis distance; Substep 22.2: dynamically adjusting the sampling weights of each area; wherein the effective sample size is the current weight, is the new weight; Step 23: tensor kernel acceleration Substep 23.1: splitting the high-dimensional data into d / 32 tensor blocks, and assigning each tensor block to a GPU streaming multiprocessor; Substep 23.2: designing a mixed precision sampling algorithm, using FP16 for key calculation nodes and using FP32 for probability accumulation.

2. The method for rapid detection of dioxin concentration based on chlorobenzene as indicator according to claim 1, characterized in that, In the S1, the collected fly ash samples are subjected to screening and equal division.

3. The method for rapid detection of dioxin concentration based on chlorobenzene as indicator according to claim 1, characterized in that, In the S2, the sample data is subjected to data cleaning, including missing value processing and abnormal value removal, and the data is subjected to normalization and / or standardization processing; Normalization: ; standardization: ; where, is the original data, is the normalized data, is the standardized data, and are the minimum and maximum values in the dataset, respectively, is the mean, is the standard deviation.

4. The method for rapid detection of dioxin concentration based on chlorobenzene as indicator according to claim 3, characterized in that, The method of the missing value processing is as follows: if the sample missing rate is less than 5%, multiple imputation or KNN interpolation is used; if the missing rate is greater than 5%, direct elimination or special working condition marking is used.

5. The method for rapid detection of dioxin concentration based on chlorobenzene as indicator according to claim 1, characterized in that, The dioxin prediction model is a neural network model, and the network architecture of the neural network model comprises: Input layer: the number of nodes is equal to the number of features; Hidden layer: 2 to 3 layers, and the number of neurons decreases according to the pyramid rule; Output layer: single node, and the activation function is linear.

6. The method for rapid detection of dioxin concentration based on chlorobenzene as indicator according to claim 5, characterized in that, The loss function of the neural network model is mean square error or smooth L1 loss.

7. The method for rapid detection of dioxin concentration based on chlorobenzene as indicator according to claim 5, characterized in that, The neural network model is verified, including performance evaluation index, explainability analysis and sensitivity analysis; The performance evaluation index is as follows: Regression index: MSE, RMSE, MAE, R²; Wherein, n is the total amount of input test set data, is the true value, is the predicted value, is the sample mean; Residual analysis: residual-QQ plot is drawn to verify normality; Residual-predicted value scatter plot to check heteroscedasticity; The explainability analysis is as follows: Local explanation: LIME is used to show the chlorobenzene contribution of a single sample; Global explanation: Partial Dependence Plot is calculated to quantify the marginal effect of chlorobenzene concentration on dioxin concentration; SHAP Summary Plot: visualizes feature importance and influence direction; The sensitivity analysis is as follows: Single variable disturbance: fixing other variables, gradually increasing the chlorobenzene concentration, and observing the change gradient of the predicted value of the dioxin concentration; Adversarial sample test: injecting noise to evaluate the robustness of the model.

8. The method for rapid detection of dioxin concentration based on chlorobenzene as indicator according to claim 1, characterized in that, The dioxin prediction model is a linear regression model; the linear regression model initializes the linear regression model using the Linear Regression class of scikit-learn, divides the training set and the test set according to the ratio of 7:3, evaluates the model using the test set, and calculates the R² and MSE indicators.

9. The method for rapid detection of dioxin concentration based on chlorobenzene as indicator according to claim 1, characterized in that, The chlorobenzene concentration is the concentration of 1,3,5-trichlorobenzene or the concentration of 1,2,3,5-tetrachlorobenzene.

Images