Method and system for predicting secondary reaction rate constant of active nitrogen free radicals and new pollutants

Abstract

The invention discloses a method and system for predicting a secondary reaction rate constant of an active nitrogen free radical and a new pollutant, and the method comprises the steps: converting the new pollutant into a feature vector through a Morgan molecular fingerprint characterization method and an MACCS fingerprint characterization method; respectively inputting the two reaction rate constant prediction results into the first prediction model and the second prediction model to obtain two reaction rate constant prediction results, and carrying out weighted summation on the two prediction results to obtain a final reaction rate constant prediction result. According to the method, the k value of the reaction of the active nitrogen free radicals and the new pollutants can be accurately predicted, the cost required by an experimental method is remarkably reduced, and reliable key kinetic parameter support is provided for engineering practice of removing the new pollutants in the environment by the active nitrogen free radicals.

Description

Technical Field

[0001] The present invention relates to machine learning technology, and in particular to a method and system for predicting the secondary reaction rate constant of active nitrogen free radicals with new pollutants. Background Art

[0002] Emerging pollutants (ECs) have attracted widespread attention due to their potential harm to the ecological environment and human health. Advanced oxidation processes (AOPs) can produce strong oxidative free radicals and have the potential to efficiently degrade ECs. They have become one of the most promising technologies for removing ECs. According to the reaction kinetics, the reaction rate of RNS with ECs is positively correlated with the RNS concentration, pollutant concentration and the secondary reaction rate constant (k). The secondary reaction rate constant of compounds such as caffeic acid, sinapinic acid, and 3-methyl-1-butyne with RNS reaches 10 8 –10 9 M -1 s -1 This numerical range fully demonstrates the high reactivity of reactive nitrogen radicals toward certain compounds. Furthermore, nitrates, a key RNS precursor, are ubiquitous in environmental media, with an average concentration as high as 15.35 mg / L. These nitrates can generate large quantities of reactive nitrogen radicals through photochemical and electrochemical pathways. Based on these reaction kinetic data and the abundance of precursors, reactive nitrogen radicals are expected to demonstrate significant advantages in the treatment of emerging pollutants.

[0003] As one of the key parameters for evaluating the reactivity of free radicals and compounds, the secondary reaction rate constant can be measured experimentally. Currently, the PubChem database contains over 120 million compounds, and experimental determination of k values ​​is time-consuming and labor-intensive. Many researchers use quantitative structure-activity relationships (QSARs) to fit the mathematical relationship between EC structures and k values ​​to predict k values. However, existing QSAR models have two shortcomings: 1. The relationship between many EC structures and k values ​​is no longer a simple linear relationship, making it difficult for QSAR to meet the needs of high-precision predictions; 2. QSAR models are based on the researcher's own physical and chemical knowledge, and the subjectivity involved affects the model fitting effect.

[0004] Machine learning (ML) has powerful data processing capabilities and can effectively capture high-dimensional, nonlinear relationships between data. Researchers have developed ECs k-value prediction models based on ML. For example, the patent application number CN202211540751.3 established a new pollutant reactivity prediction method based on machine learning. The process is to use the collected SO4 ·- and CO3 ·- The k value is used to train three models: random forest, neural network and support vector machine, and the best model is selected to predict SO4 ·- and CO3 ·-The k value of ECs. Disadvantages of this method: 1. It does not consider the impact of environmental variables on k, which may affect the model prediction results; 2. After setting the initial fingerprint length to 1024, no further optimization of the fingerprint length is performed, and the model prediction performance needs to be improved. Summary of the Invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a method and system for predicting the secondary reaction rate constant of active nitrogen radicals with new pollutants, so as to solve the problems of low efficiency in determining k value and low prediction accuracy in the prior art.

[0006] Technical solution: The method for predicting the secondary reaction rate constant of active nitrogen radicals with new pollutants according to the present invention comprises the following steps:

[0007] collecting the secondary reaction rate constant, reaction temperature, and pH value of the active nitrogen free radical and the compound and establishing a feature vector, establishing a data set, and using the data set to train a first prediction model and a second prediction model;

[0008] The method for establishing the characteristic vector is as follows: the compound is converted into a first vector and a second vector by the Morgan molecular fingerprint characterization method and the MACCS fingerprint characterization method respectively; the type of the reactive nitrogen radical is represented by 0 or 1, the reactive nitrogen radical type that reacts with the compound is 1, and the type that does not react is 0; the first vector and the second vector are respectively combined with the reaction temperature, pH value and the type of the reactive nitrogen radical to form a first characteristic vector and a second characteristic vector; the first characteristic vector is input into the first prediction model to predict the first and second-order reaction rate constants, and the second characteristic vector is input into the second prediction model to predict the second and second-order reaction rate constants; the first and second-order reaction rate constants are weightedly summed to obtain the final predicted second-order reaction rate constant;

[0009] For the new pollutant to be tested, the first eigenvector and the second eigenvector are established, and the first and second secondary reaction rate constants and the second secondary reaction rate constants are predicted using the trained first prediction model and the second prediction model respectively; the first and second secondary reaction rate constants are weightedly summed to obtain the final predicted secondary reaction rate constant.

[0010] Furthermore, the first prediction model is a GBR model, and the second prediction model is an XGBoost model.

[0011] Furthermore, using the data set to train the first prediction model and the second prediction model also includes dividing the data set into a training set and a test set, calculating the structural similarity of the compounds in the training set and the test set, setting a structural similarity threshold, and determining the application domain of the first prediction model and the second prediction model.

[0012] Furthermore, outliers whose second-order reaction rate constants exceed the upper and lower data limits are deleted from the data set.

[0013] Furthermore, a weighted coefficient of the weighted sum of the first and second second-order reaction rate constants is calculated using an optimal linear proportional algorithm.

[0014] Furthermore, when the first prediction model and the second prediction model are trained using the data set, hyperparameter optimization is performed using a Bayesian method.

[0015] The prediction system for the secondary reaction rate constant of active nitrogen radicals with new pollutants of the present invention comprises:

[0016] A model building and training unit, for collecting the secondary reaction rate constant, reaction temperature and pH value of the active nitrogen free radical and the compound and establishing a feature vector, establishing a data set, and using the data set to train the first prediction model and the second prediction model;

[0017] The method for establishing the characteristic vector is as follows: the compound is converted into a first vector and a second vector by the Morgan molecular fingerprint characterization method and the MACCS fingerprint characterization method respectively; the type of the reactive nitrogen radical is represented by 0 or 1, the reactive nitrogen radical type that reacts with the compound is 1, and the type that does not react is 0; the first vector and the second vector are respectively combined with the reaction temperature, pH value and the type of the reactive nitrogen radical to form a first characteristic vector and a second characteristic vector; the first characteristic vector is input into the first prediction model to predict the first and second-order reaction rate constants, and the second characteristic vector is input into the second prediction model to predict the second and second-order reaction rate constants; the first and second-order reaction rate constants are weightedly summed to obtain the final predicted second-order reaction rate constant;

[0018] The secondary reaction rate constant prediction unit is used to establish a first eigenvector and a second eigenvector for a new pollutant to be tested, and respectively predict the first and second secondary reaction rate constants and the second secondary reaction rate constants using the trained first prediction model and the second prediction model; and weightedly sum the first and second secondary reaction rate constants to obtain the final predicted secondary reaction rate constant.

[0019] Furthermore, the first prediction model is a GBR model, and the second prediction model is an XGBoost model.

[0020] The electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the method for predicting the secondary reaction rate constant of the active nitrogen radicals and the new pollutants is implemented.

[0021] The computer-readable storage medium of the present invention stores a computer program, and when the computer program is executed by a processor, the method for predicting the secondary reaction rate constant of the active nitrogen radical and the new pollutant is implemented.

[0022] Beneficial effect: Compared with the prior art, the advantages of the present invention are: the present invention constructs a gradient boosting regression model based on Morgan molecular fingerprints and an XGBoost model based on MACCS fingerprints based on the secondary reaction rate constant k value and reaction conditions of the active nitrogen free radicals and compounds, and further integrates the two best models into a Linear-hybrid mixed model through the optimal linear ratio, that is, the prediction results of the two best models are linearly combined to obtain the final prediction result. The present invention takes into account the influence of environmental variables on the reaction activity, and optimizes the length of the Morgan molecular fingerprint, which can accurately predict the k value of the reaction between active nitrogen free radicals and ECs. Compared with the existing experimental method for determining the k value, the present invention has low cost and accurate prediction, which significantly reduces the cost required for the experimental method, provides reliable key kinetic parameter support for the engineering practice of removing new pollutants in the environment by active nitrogen free radicals, effectively guides the reactor design and operation parameter optimization of the advanced oxidation process, and is of great significance to improving the removal efficiency of new pollutants in the water treatment process. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Flow chart of the method of the present invention.

[0024] Figure 2 This is a comparison chart of model optimization results after hyperparameter optimization in an embodiment of the present invention.

[0025] Figure 3 This is a comparison chart of the actual values ​​and predicted values ​​of XGBoost-MACCS, GBR-MF, and Linear-hybrid in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] Emerging pollutants pose a significant threat to human health and water quality. Research shows that emerging pollutants can accumulate in the human body through the food chain and respiratory processes, disrupting the normal functioning of the endocrine system and inducing health risks such as cancer and malformations. Effectively removing emerging pollutants is a key research topic of widespread concern among experts and scholars both domestically and internationally.

[0027] Current treatment technologies for emerging pollutants primarily include precipitation, adsorption, filtration, and biodegradation. However, these conventional treatment techniques have limitations in practice. First, due to the structural differences in pollutants, conventional technologies often suffer from high degradation selectivity but poor broad-spectrum performance. Second, the treatment process may produce byproducts that are potentially harmful to human health, such as trihalomethanes and halogenated acetic acids. Toxicological studies have shown that these substances are highly cytotoxic and are closely linked to cancer and genetic mutations, posing a serious threat to public health.

[0028] Advanced oxidation processes can produce highly reactive free radicals that can effectively destroy the structure of new pollutants and achieve efficient degradation of new pollutants. Among them, reactive nitrogen radicals have a high precursor concentration and a high secondary reaction rate constant when reacting with certain compounds. They are expected to play an important role in the degradation of new pollutants and environmental risk control. However, with the development of industry, the number of compounds included in the PubChem database has exceeded 120 million, and the number of pollutant types is still increasing. It is time-consuming and labor-intensive to use experimental methods to determine the reaction rate constant of each pollutant with reactive nitrogen radicals one by one, and the cost is extremely high. QSAR and machine learning methods can significantly reduce experimental costs, but a high-precision model that can predict the secondary reaction rate constant of reactive nitrogen radicals with new pollutants has not yet been developed.

[0029] In order to solve the above problems, the present invention proposes a method for predicting the secondary reaction rate constant of active nitrogen free radicals with new pollutants. This method is based on the QSAR principle and uses the structural information and reaction conditions of the compound. It can not only achieve accurate prediction of its k value, but also has good generalization ability and can be naturally extended to the prediction of k value of the reaction between new pollutants and active nitrogen free radicals. The technical solution of the present invention is further explained below with reference to the accompanying drawings. Figure 1 As shown, the method of the present invention comprises the following steps:

[0030] Step 1: Collect the secondary reaction rate constants of active nitrogen radicals and compounds through published papers and databases.

[0031] Step 2: Use molecular characterization methods to process compound information into a binary vector. After representing the type of reactive nitrogen radical that reacts with the compound as 0 / 1 and normalizing the reaction temperature and pH, merge it with the binary vector to form a feature vector, and then integrate it with the secondary reaction rate constant (k value) to form a data set. The specific steps are:

[0032] Step 2.1, converting the compound structure into a binary vector by molecular characterization method to obtain the compound structural characteristics;

[0033] Step 2.2, obtaining the active nitrogen radical characteristics by representing the type of active nitrogen radical that reacts with the compound by 0 / 1;

[0034] Step 2.3, obtaining reaction condition characteristics by normalizing the reaction temperature and pH;

[0035] Step 2.4, obtaining a feature vector of the compound by combining the compound structure features, active nitrogen radical features, and reaction condition features;

[0036] In step 2.5, the complete data set is obtained by combining the second-order reaction rate constants with the eigenvectors.

[0037] When extracting compound structural information, molecular characterization methods can convert the compound's SMILES into a binary vector that can be recognized and processed by computers, providing standardized input for subsequent machine learning model construction and optimization.

[0038] The Morgan molecular fingerprint is a topological fingerprint that captures local molecular features by extending atomic neighborhood information and mapping this information into a binary vector of customizable length. It is suitable for feature extraction and high-precision modeling of molecules with unknown structures. The MACCS fingerprint automatically extracts molecular features based on 166 predefined substructures. Each site corresponds to a specific substructure, providing high interpretability. The MACCS fingerprint is also low-cost to store and transmit, requiring minimal computational resources. This allows for rapid molecular fingerprint extraction, making it crucial for applications requiring high interpretability and rapid initial screening.

[0039] Step 3: After removing outliers from the dataset using statistical methods, the dataset is divided into a training set and a test set in a ratio of 9:1. The specific steps are:

[0040] Step 3.1, identify outliers in the data set using statistical methods;

[0041] Step 3.2: Using statistical methods to remove outliers from the data set, a data set was obtained for model training and testing. The statistical method specifically refers to the interquartile range (IQR) method, which removes data points that exceed the upper and lower limits of the log-second-order reaction rate constant.

[0042] Step 3.3: Divide the dataset for model training and testing into a ratio of 9:1 to obtain the training set and test set.

[0043] In step 3.4, the machine learning model is trained using two datasets.

[0044] In the process of machine learning model development, the training set is used to train the model. The training set can be further divided into a training set and a validation set, and the model is trained together through cross-validation.

[0045] As an enhanced algorithm based on the gradient boosting framework, XGBoost introduces a regularization term on top of traditional gradient boosting to prevent model overfitting, and uses the Hessian matrix to optimize the objective function, significantly improving convergence speed. XGBoost also supports multi-threaded parallel computing and can automatically handle missing values, performing exceptionally well in scenarios with large-scale data and high-dimensional features. GBR is a classic implementation of the gradient boosting algorithm. Its core idea is to correct the prediction error through continuous iterations, gradually approaching the true target value, and using the weighted superposition of the prediction results of the weak learners as the final prediction value. GBR not only achieves progressive optimization of the residual, but also supports multiple loss functions, is highly interpretable, and has a high degree of transparency in the decision-making process, performing exceptionally well in small and medium-sized data.

[0046] Step 4: Evaluate the model results using evaluation indicators, including root mean square error (RMSE) and coefficient of determination (R 2 ).

[0047] Step 5: Optimize the model's hyperparameters to obtain the best prediction model.

[0048] Step 6: Further improve the model accuracy by setting the application domain for the best prediction model.

[0049] Step 7: Combine the best models into a hybrid model using the optimal linear scaling. After hyperparameter optimization, two optimal models were obtained: the gradient boosting regression model based on Morgan molecular fingerprint (GBR-MF) and the XGBoost model based on MACCS fingerprint (XGBoost-MACCS). The prediction results of these two models were combined using the optimal linear scaling to obtain the final prediction result.

[0050] The method of the present invention is verified by specific experiments below.

[0051] Step (1) A total of 285 secondary reaction rate constants (k) of nitrogen dioxide radicals (NO2·) and amino radicals (NH2·) with compounds under different reaction conditions were collected from published papers and public databases. The compounds were one or more of the following organic compounds: personal care products, endocrine disruptors, antibiotics, phenols, aldehydes, alcohols, ketones, etc. The structural information of the compounds was processed into binary vectors using two molecular characterization methods: Morgan molecular fingerprints with a length of 1024 and MACCS fingerprints with a length of 166. The reaction temperature and pH were normalized to floating point numbers in the range of 0 to 1 and then merged with the binary vector representing the structural information to form a new feature vector. The secondary reaction rate constants were processed into logarithmic form as label values, i.e., the model prediction objects. Based on the statistical method (IQR), 6 outliers in the logarithmic secondary reaction rate constants that exceeded the upper and lower limits of the data were identified and removed. The data set with a data volume of 279 was divided into a training set and a test set with a ratio of 9:1, with 251 training set data and 28 test set data.

[0052] Step (2) 11 commonly used machine learning models were trained based on the training sets of the two molecular characterization methods, Morgan molecular fingerprint and MACCS fingerprint, including support vector machine, K nearest neighbor, bagging, AdaBoost, decision tree, super tree, XGBoost, ARD regression, Bayesian ridge regression, random forest and gradient boosting regression, with root mean square error and determination coefficient as evaluation indicators of model prediction accuracy.

[0053] Step (3) Bayesian hyperparameter optimization was performed on nine models (including support vector machine, bagging, AdaBoost, hypertree, XGBoost, ARD regression, Bayesian ridge regression, random forest, and gradient boosting regression) whose training set and test set determination coefficients were greater than or equal to 0.40 in the machine learning model based on Morgan molecular fingerprint. The determination coefficient of the test set was used as the optimization target, and 5-fold cross validation was used to determine the optimal hyperparameter combination for each model. By comparing the prediction accuracy of the nine models under the optimal hyperparameter combination, it was found that the gradient boosting regression model based on Morgan molecular fingerprint (GBR-MF) and the XGBoost model based on MACCS fingerprint (XGBoost-MACCS) had the best prediction performance.

[0054] Table 1 and Table 2 show the 9 models under Morgan molecular fingerprint and MACCS fingerprint characterization respectively. As can be seen from the table, the model prediction performance of the two molecular characterization methods is good. 2 train Reaching 0.7280~0.9995, R 2 testThe results reached 0.6872~0.8701, among which the GBR model showed the best prediction performance; the R 2 train Between 0.7199 and 0.9767, R 2 test The values ​​are between 0.6349 and 0.8871, among which XGBoost shows the best prediction performance.

[0055] Table 1. Prediction performance of nine optimized models under Morgan molecular fingerprint characterization

[0056]

[0057] Table 2. Prediction performance of nine optimized models under MACCS fingerprint representation

[0058]

[0059] Figure 2 Figures (a) and (b) are comparisons of the optimization results of the nine models under MF and MACCS characterization after hyperparameter optimization. As can be seen from the figure, when the molecular characterization method is Morgan molecular fingerprint, the gradient boosting regression model has the best prediction performance on the training set and the test set, and its R 2 train =0.9995, R 2 test =0.8701. When the molecular characterization method is MACCS fingerprint, the XGBoost model has the best prediction performance, and its R 2 train =0.9140, R 2 test =0.8871. Figure 2 As shown in the figure, different models under different molecular characterization methods have different prediction performances after optimization. Among them, the gradient boosting regression model based on Morgan molecular fingerprint (GBR-MF) and the XGBoost model based on MACCS fingerprint (XGBoost-MACCS) showed the best goodness of fit on the training set and test set.

[0060] Step 4: Calculate the structural similarity between the compounds in the training set and the test set, and determine the application domain of GBR-MF and XGBoost-MACCS by setting the similarity threshold.

[0061] Step 5: Input the feature vector into the GBR-MF model to obtain the prediction result based on GBR-MF;

[0062] Step 6: Input the feature vector into the XGBoost-MACCS model to obtain the prediction results based on XGBoost-MACCS;

[0063] Step 7: Linearly combine GBR-MF and XGBoost-MACCS using the undetermined coefficients w1 and w2. Use the best linear scaling algorithm to obtain the optimal ratio of XGBoost-MACCS:GBR-MF = 0.2259:0.7741.

[0064] Step 8: The hybrid model Linear-hybrid is obtained by using the ratio of XGBoost-MACCS:GBR-MF = 0.2259:0.7741. The output of the hybrid model is the final predicted second-order reaction rate constant.

[0065] Figure 3 (a), (b), and (c) are the comparison diagrams of the true value and predicted value of the three models XGBoost-MACCS, GBR-MF, and Linear-hybrid in this experiment. In the figure, the horizontal axis is the true value and the vertical axis is the predicted value of the model. Figure 3 It can be seen that both the XGBoost-MACCS model and the GBR-MF model show good prediction performance, and their data point distribution is close to the straight line y=x, that is, the predicted value of the model is close to the actual value. The performance difference of the XGBoost-MACCS model on the training set and the test set is not significant (R 2 train =0.9140, R 2 test =0.8871), reflecting its good generalization ability, but its determination coefficient on the training set is low and the training accuracy is insufficient; the GBR-MF model has high prediction accuracy on the training set (R 2 train =0.9995), but its generalization ability is poor. In contrast, the Linear-hybrid model performs better in predicting the secondary reaction rate constant of active nitrogen radicals and new pollutants (R 2 train =0.9952, R 2 test =0.8879), which not only makes up for the defect of insufficient training accuracy of the XGBoost-MACCS model, but also reduces the overfitting risk of the GBR-MF model. This model shows excellent prediction performance on both the training set and the test set.

[0066] The prediction system for the secondary reaction rate constant of active nitrogen radicals with new pollutants of the present invention comprises:

[0067] A model building and training unit, for collecting the secondary reaction rate constant, reaction temperature and pH value of the active nitrogen free radical and the compound and establishing a feature vector, establishing a data set, and using the data set to train the first prediction model and the second prediction model;

[0068] The method for establishing the characteristic vector is as follows: the compound is converted into a first vector and a second vector by the Morgan molecular fingerprint characterization method and the MACCS fingerprint characterization method respectively; the type of the reactive nitrogen radical is represented by 0 or 1, the reactive nitrogen radical that reacts with the compound is 1, and the reactive nitrogen radical that does not react is 0; the first vector and the second vector are respectively combined with the reaction temperature, pH value and the type of the reactive nitrogen radical to form a first characteristic vector and a second characteristic vector; the first characteristic vector is input into the first prediction model to predict the first and second-order reaction rate constants, and the second characteristic vector is input into the second prediction model to predict the second and second-order reaction rate constants; the first and second-order reaction rate constants are weightedly summed to obtain the final predicted second-order reaction rate constant;

[0069] The secondary reaction rate constant prediction unit is used to establish a first eigenvector and a second eigenvector for a new pollutant to be tested, and respectively predict the first and second secondary reaction rate constants and the second secondary reaction rate constants using the trained first prediction model and the second prediction model; and weightedly sum the first and second secondary reaction rate constants to obtain the final predicted secondary reaction rate constant.

[0070] The electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the method for predicting the secondary reaction rate constant of the active nitrogen radicals and the new pollutants is implemented.

[0071] The computer-readable storage medium of the present invention stores a computer program, and when the computer program is executed by a processor, the method for predicting the secondary reaction rate constant of the active nitrogen radical and the new pollutant is implemented.

[0072] The computer-readable storage media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, flash memory, or any other medium that can be used to store program code in the form of instructions or data structures and that can be accessed by a computer.

[0073] The processor is configured to execute the computer program stored in the memory to implement the various steps in the method involved in the above embodiment.

Claims

1. A method for predicting the secondary reaction rate constant of active nitrogen radicals with new pollutants, characterized in that: The steps include: collecting the secondary reaction rate constant, reaction temperature, and pH value of the active nitrogen free radical and the compound and establishing a feature vector, establishing a data set, and using the data set to train a first prediction model and a second prediction model; The method for establishing the characteristic vector is as follows: the compound is converted into a first vector and a second vector by the Morgan molecular fingerprint characterization method and the MACCS fingerprint characterization method respectively; the type of the reactive nitrogen radical is represented by 0 or 1, the reactive nitrogen radical type that reacts with the compound is 1, and the type that does not react is 0; the first vector and the second vector are respectively combined with the reaction temperature, pH value and the type of the reactive nitrogen radical to form a first characteristic vector and a second characteristic vector; the first characteristic vector is input into the first prediction model to predict the first and second-order reaction rate constants, and the second characteristic vector is input into the second prediction model to predict the second and second-order reaction rate constants; the first and second-order reaction rate constants are weightedly summed to obtain the final predicted second-order reaction rate constant; For the new pollutant to be tested, the first eigenvector and the second eigenvector are established, and the first and second secondary reaction rate constants and the second secondary reaction rate constants are predicted using the trained first prediction model and the second prediction model respectively; the first and second secondary reaction rate constants are weightedly summed to obtain the final predicted secondary reaction rate constant.

2. The method for predicting the secondary reaction rate constant of active nitrogen radicals with new pollutants according to claim 1, characterized in that: The first prediction model is a GBR model, and the second prediction model is an XGBoost model.

3. The method for predicting the secondary reaction rate constant of active nitrogen radicals with new pollutants according to claim 1, characterized in that: Using the data set to train the first prediction model and the second prediction model also includes dividing the data set into a training set and a test set, calculating the structural similarity of the compounds in the training set and the test set, setting a structural similarity threshold, and determining the application domain of the first prediction model and the second prediction model.

4. The method for predicting the secondary reaction rate constant of active nitrogen radicals with new pollutants according to claim 1, characterized in that: Outliers whose second-order reaction rate constants exceeded the upper and lower data limits were deleted from the data set.

5. The method for predicting the secondary reaction rate constant of active nitrogen radicals with new pollutants according to claim 1, characterized in that: The weighting coefficient of the weighted sum of the first and second second-order reaction rate constants is calculated by the best linear proportional algorithm.

6. The method for predicting the secondary reaction rate constant of active nitrogen radicals with new pollutants according to claim 1, characterized in that: When the first prediction model and the second prediction model are trained using the data set, hyperparameter optimization is performed using a Bayesian method.

7. A prediction system for the secondary reaction rate constant of active nitrogen radicals with new pollutants based on the method of claim 1, characterized in that: include: A model building and training unit, for collecting the secondary reaction rate constant, reaction temperature and pH value of the active nitrogen free radical and the compound and establishing a feature vector, establishing a data set, and using the data set to train the first prediction model and the second prediction model; The method for establishing the characteristic vector is as follows: the compound is converted into a first vector and a second vector by the Morgan molecular fingerprint characterization method and the MACCS fingerprint characterization method respectively; the type of the reactive nitrogen radical is represented by 0 or 1, the reactive nitrogen radical type that reacts with the compound is 1, and the type that does not react is 0; the first vector and the second vector are respectively combined with the reaction temperature, pH value and the type of the reactive nitrogen radical to form a first characteristic vector and a second characteristic vector; the first characteristic vector is input into the first prediction model to predict the first and second-order reaction rate constants, and the second characteristic vector is input into the second prediction model to predict the second and second-order reaction rate constants; the first and second-order reaction rate constants are weightedly summed to obtain the final predicted second-order reaction rate constant; The secondary reaction rate constant prediction unit is used to establish a first eigenvector and a second eigenvector for a new pollutant to be tested, and respectively predict the first and second secondary reaction rate constants and the second secondary reaction rate constants using the trained first prediction model and the second prediction model; and weightedly sum the first and second secondary reaction rate constants to obtain the final predicted secondary reaction rate constant.

8. The prediction system for the secondary reaction rate constant of active nitrogen radicals with new pollutants according to claim 7, characterized in that: The first prediction model is a GBR model, and the second prediction model is an XGBoost model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is loaded into a processor, the method for predicting the secondary reaction rate constant of active nitrogen radicals with new pollutants according to any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for predicting the secondary reaction rate constant of active nitrogen radicals with new pollutants according to any one of claims 1 to 6 is implemented.

Images