Combustible gas / steam multi-target burning and explosion characteristic prediction method and prediction system

Abstract

The invention discloses a combustible gas / steam multi-target burning and explosion characteristic prediction method and system, and the method comprises the following steps: collecting burning and explosion characteristic data of combustible gas / steam, forming a burning and explosion characteristic data set, and dividing the burning and explosion characteristic data set into a training set and a test set; carrying out optimization and frequency analysis on the molecular structure of the combustible gas / steam, characterizing the molecular structure of the combustible gas / steam by adopting quantum chemical descriptors, and screening the quantum chemical descriptors to obtain descriptors with linear correlation meeting a threshold requirement; on the basis of the training set, calculating corresponding screened quantum chemistry descriptor parameter values, and training and constructing a multi-target QSPR model of the combustion and explosion characteristics; testing and evaluating the model by adopting a test set, and verifying the prediction capability of the model; and through the constructed model, predicting the burning and explosion characteristics of the combustible gas / steam and analyzing the micromechanism influencing the burning and explosion characteristics. The method can accurately and quickly predict various burning and explosion characteristics of the combustible gas / steam.

Description

Technical Field

[0001] The present application relates to the technical field of gas explosion characteristics prediction, and in particular to a method and system for predicting multi-objective explosion characteristics of combustible gas / steam. Background Art

[0002] Due to the inherent characteristics of low flash temperature and wide explosion limit range of flammable gas / vapor, it is very easy to cause major production safety accidents such as fire and explosion during production, storage, use, transportation and waste disposal. The combustion and explosion characteristics of flammable gas / vapor mainly include upper explosion limit, lower explosion limit, flash point, autoignition point and heat of combustion. Mastering these combustion and explosion characteristics data plays an important guiding role in its safety. However, due to the wide variety of flammable gas / vapor, experimental measurement method is time-consuming and laborious.

[0003] The quantitative structure-property relationship (QSPR) method plays an important role in predicting its flammability and explosion characteristics. However, according to the applicant's understanding, the existing flammability and explosion characteristics prediction model can only achieve single-target prediction (Pan Y.; Jiang J. Flammability characteristics prediction using QSPR modeling. Machine learning in chemical safety health: fundamentals with applications, 2022: 47-80). For example, a genetic algorithm is used to construct a single flammability and explosion characteristics prediction method for organic compounds (Chinese patent CN101339181B); a support vector machine method is used to construct a single flammability and explosion characteristics prediction method for organic compounds and mixtures (Chinese patents CN1339180B and CN102608285B).

[0004] The explosion hazard of flammable gas / vapor needs to be comprehensively evaluated based on multiple parameters such as the lower explosion limit, upper explosion limit, and flash point. In molecular design, it is expected that the explosion concentration range of the substance is small and the flash point is high, that is, the lower explosion limit and flash point of the substance should be high, and the upper explosion limit should be low. However, since different single-objective QSPR models use different descriptors and modeling methods, a single lower explosion limit prediction model can only reflect the mechanism of action of its descriptor on the lower explosion limit, and cannot reveal the influence of the descriptor on other explosion characteristics such as the upper explosion limit and flash point. Therefore, it is difficult for a single-objective model to meet the requirements of multi-dimensional indicators in molecular design. In addition, the use of a single-objective model is cumbersome and requires a large amount of calculation. Summary of the invention

[0005] In view of the problem that the current single-objective prediction of the explosion characteristics of combustible gas / steam is difficult to meet the multi-dimensional index requirements during molecular design, the present application proposes a multi-objective prediction method and prediction system for the explosion characteristics of combustible gas / steam, which can accurately and quickly predict multiple explosion characteristic parameters through an improved QSPR model.

[0006] In a first aspect, the present application provides a method for predicting the multi-objective combustion and explosion characteristics of combustible gas / steam, including the following contents:

[0007] S1, collecting the combustion and explosion characteristic data of combustible gas / vapor to form a combustion and explosion characteristic data set, and dividing the combustion and explosion characteristic data set into a training set and a test set according to a proportion;

[0008] S2, optimizing and frequency analyzing the molecular structure of the combustible gas / vapor, characterizing the molecular structure of the combustible gas / vapor using quantum chemical descriptors, and screening the quantum chemical descriptors to obtain quantum chemical descriptors whose linear correlation meets the threshold requirements;

[0009] S3, based on the training set, calculating and screening the corresponding quantum chemical descriptor parameter values, constructing a multi-objective QSPR model of combustion and explosion characteristics, and using a test set to test and evaluate the multi-objective QSPR model of combustion and explosion characteristics to verify the prediction ability of the model;

[0010] S4, the explosion characteristics of combustible gas / steam are predicted through the constructed multi-objective QSPR model of explosion characteristics.

[0011] In a possible implementation of the first aspect, step S2 obtains the quantum chemical descriptors through Multiwfn calculation to form a descriptor set, and the descriptor set includes GIPF descriptor, dipole moment, highest occupied molecular orbital energy, lowest unoccupied molecular orbital energy and energy gap.

[0012] In a possible implementation manner of the first aspect, step S2 uses the Pearson correlation coefficient to screen the quantum chemical descriptors.

[0013] In a possible implementation of the first aspect, a multi-objective QSPR model of explosion characteristics based on a multi-layer perceptron network is constructed, and the construction process is as follows:

[0014] S300, selecting an activation function and a loss function;

[0015] S301, using the BatchNorm1d function of the data preprocessing layer to normalize the D descriptors obtained by screening;

[0016] S302, input the normalized D descriptors into hidden layer 1 and expand them to 3D parameters;

[0017] S303, converting the 3D parameters into D parameters through hidden layer 2;

[0018] S304, outputting the target explosion characteristics of the combustible gas / vapor after fitting through the fully connected layer;

[0019] Here, D represents the number of quantum chemical descriptors whose linear correlation meets the threshold requirement.

[0020] In another possible implementation of the first aspect, a multi-objective QSPR model of explosion characteristics based on a residual network is constructed, and the construction process is as follows:

[0021] S310, selecting an activation function and a loss function;

[0022] S311, using the BatchNorm1d function of the data preprocessing layer to normalize the selected D descriptors;

[0023] S312, using the one-dimensional convolution of the hidden layer 1 to perform convolution processing on the normalized descriptor, and again using the BatchNorm1d function to normalize the convolution processed descriptor;

[0024] S313, processed by the maximum pooling layer;

[0025] S314, after two convolutions and one addition process of the residual block, and then processed by the average pooling layer, all dimensions are flattened using the Flatten function to output 3D parameters;

[0026] S315, transform the parameters through hidden layer 2 and output D parameters;

[0027] S316, outputting the target characteristics of the combustible gas / vapor after fitting through the fully connected layer;

[0028] Here, D represents the number of quantum chemical descriptors whose linear correlation meets the threshold requirement.

[0029] In a possible implementation manner of the first aspect, the method further includes a step of analyzing the application domain of the multi-objective QSPR model of combustion and explosion characteristics.

[0030] In a possible implementation manner of the first aspect, the method further includes a step of analyzing the microscopic mechanism that affects the combustion and explosion characteristics.

[0031] In a second aspect, the present application provides a system for executing the combustible gas / steam multi-objective explosion characteristics prediction method, comprising:

[0032] An explosion characteristic data acquisition unit collects the explosion characteristic data of the combustible gas / vapor to form an explosion characteristic data set, and divides the explosion characteristic data set into a training set and a test set according to a proportion;

[0033] A descriptor calculation and screening unit is used to optimize and perform frequency analysis on the molecular structure of the combustible gas / vapor, characterize the molecular structure of the combustible gas / vapor using quantum chemical descriptors, and screen the quantum chemical descriptors to obtain quantum chemical descriptors whose linear correlation meets the threshold requirements;

[0034] A prediction model building unit calculates and screens the corresponding quantum chemical descriptor parameter values ​​based on the training set, builds a multi-objective QSPR model of combustion and explosion characteristics, and uses a test set to test and evaluate the multi-objective QSPR model of combustion and explosion characteristics to verify the prediction ability of the model;

[0035] The explosion characteristics prediction unit predicts the explosion characteristics of combustible gas / steam through the constructed explosion characteristics multi-objective QSPR model.

[0036] In a possible implementation manner of the second aspect, the system further includes:

[0037] The application domain analysis unit is used to analyze the application domain of the multi-objective QSPR model of combustion and explosion characteristics.

[0038] In a possible implementation manner of the second aspect, the system further includes:

[0039] The microscopic mechanism analysis unit is used to analyze the microscopic mechanism that affects the combustion and explosion characteristics.

[0040] Compared with the existing methods, this application has the following beneficial effects:

[0041] (1) This application can accurately and quickly predict various combustion and explosion characteristics of combustible gases and vapors based on their molecular structure, such as the lower explosion limit, upper explosion limit, flash point, etc.

[0042] (2) This application adopts a unified descriptor based on molecular wave function information to construct a multi-objective combustion and explosion characteristics prediction model, which enhances the interpretability of the model and can meet the requirements for various combustion and explosion characteristics indicators during molecular design.

[0043] (3) The prediction model of this application provides the necessary basic data for measuring the hazards of hazardous chemicals during production, use, storage and transportation, and provides new technical support for safety research. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a process flow of a combustible gas / vapor multi-objective explosion characteristic prediction method according to a preferred embodiment of the present application;

[0045] Figure 2 It is a comparison between the predicted value of the lower explosion limit of the multi-objective model of an embodiment of the present application and the experimental value;

[0046] Figure 3 It is a comparison between the predicted value of the explosion upper limit of the multi-objective model of an embodiment of the present application and the experimental value;

[0047] Figure 4 It is a comparison between the flash point prediction value of the multi-objective model of an embodiment of the present application and the experimental value;

[0048] Figure 5 is the average absolute contribution of each descriptor of an embodiment of the present application to the lower explosion limit of the ResNet model;

[0049] Figure 6 is the average absolute contribution of each descriptor of an embodiment of the present application to the explosion upper limit of the ResNet model;

[0050] Figure 7 It is the average absolute contribution of each descriptor to the flash point of the ResNet model in one embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings.

[0052] It should be clear that the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0053] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0054] In the first aspect, the present application proposes a method for predicting the multi-objective explosion characteristics of combustible gas / steam, and the specific steps are as follows:

[0055] S1, collect the combustion and explosion characteristics data of combustible gas / vapor to form a combustible gas and vapor combustion and explosion characteristics data set, and divide the combustible gas and vapor combustion and explosion characteristics data set into a training set and a test set according to the proportion; the combustion and explosion characteristics include but are not limited to the upper explosion limit, the lower explosion limit, the flash point, the autoignition point and the heat of combustion;

[0056] The main ways to collect explosion characteristics data include: experimental measurement, literature review, and database retrieval. Among them, experimental data must be obtained in strict accordance with the corresponding experimental standards. Data in the literature must be obtained through the original literature, not other secondary literature; if data is obtained through multiple literatures, it is necessary to ensure the consistency of a series of conditions such as experimental equipment and measurement methods; if there are repeated samples in different literatures, it is also necessary to test whether the data can meet the error requirements of the corresponding experimental standards; therefore, in order to ensure data consistency, the sample data is best from the same literature, and secondly from the same research group; otherwise, in order to ensure the quality of the sample, the sample must be analyzed and screened before modeling.

[0057] S2, in order to establish a multi-objective QSPR model of combustion and explosion characteristics, we must first obtain a set of descriptors that can accurately characterize multiple combustion and explosion characteristics at the same time. Descriptors obtained through quantum chemical calculations are often more accurate than other descriptors in characterizing the structure of molecules, so quantum chemical descriptors are selected to characterize the molecular structure. In addition, it is necessary to screen quantum chemical descriptors to obtain descriptors whose linear correlation meets the threshold requirements;

[0058] Optionally, quantum chemical descriptors are obtained by Multiwfn calculation to form a descriptor set, which includes GIPF (General interaction properties function) descriptors (molecular volume, estimated molecular density, molecular surface area, molecular surface area with positive electrostatic potential, molecular surface area with negative electrostatic potential, average molecular surface electrostatic potential, average molecular surface electrostatic potential of the positive part, average molecular surface electrostatic potential of the negative part, molecular surface electrostatic potential variance, molecular surface electrostatic potential variance of the positive part, molecular surface electrostatic potential variance of the negative part, charge balance, intrinsic charge separation, molecular polarity index, molecular non-polarizable surface area, molecular polarizable surface area), dipole moment, highest occupied molecular orbital energy, lowest unoccupied molecular orbital energy and energy gap, a total of 20 descriptors; then, the Pearson correlation coefficient (Pearson correlation coefficient coefficient) to screen 20 descriptors; specifically, calculate the Pearson correlation coefficient of each pair of descriptors in the descriptor set, and consider two descriptors whose Pearson correlation coefficient is greater than a preset threshold as being strongly linearly correlated, and remove the descriptors that are strongly linearly correlated with other descriptors from the descriptor set, and finally screen out D descriptors; the recommended value range of the preset threshold is 0.9 to 0.95;

[0059] It should be noted that before using quantum chemical descriptors to characterize the molecular structure of the target gas, the molecular structure of the target gas should be optimized and frequency analyzed; this can be achieved using software such as ORCA, Gaussian, and Materials Studio; when the maximum force, root mean square force, maximum displacement, and root mean square displacement of the molecule in the optimization task all meet the convergence criteria, the molecular structure is considered to be optimized to the minimum energy point. When the frequency analysis results are all positive values, it indicates that the molecular structure is optimized to the minimum energy point rather than the saddle point on the potential energy surface.

[0060] S3, for the training set, calculating the descriptors obtained in step S2 that meet the linear correlation threshold requirements, training and constructing a multi-objective QSPR model for explosion characteristics;

[0061] Assuming that the number of quantum chemical descriptors whose linear correlation meets the threshold requirement is D, the following modeling is performed using a multilayer perceptron (MLP) network and a residual network (ResNet) network as examples;

[0062] MLP: The MLP network includes a data preprocessing layer, hidden layer one and hidden layer two; the data preprocessing layer uses the BatchNorm1d function to normalize each descriptor; the hidden layer one contains 3D hidden neurons, the hidden layer two contains D hidden neurons, and finally the D parameters are fitted through the fully connected layer and output to obtain the target characteristics such as the lower explosion limit, lower explosion limit and flash point of the combustible gas / vapor.

[0063] ResNet: ResNet includes a data preprocessing layer, a hidden layer 1, a maximum pooling layer, a residual block, an average pooling layer, a hidden layer 2, and a fully connected layer; the data preprocessing layer uses the BatchNorm1d function to normalize each descriptor; the hidden layer 1 includes a one-dimensional convolution with an output of 3D, a stride of 1, a padding of 1, and a convolution kernel of 3, and uses BatchNorm1d to normalize the data. After the one-dimensional convolution, it is connected to a maximum pooling layer; after the maximum pooling layer, the residual block is connected. After the data is processed by two convolutions and one addition in the residual block, it is processed by the average pooling layer, and all dimensions are flattened to obtain 3D hidden neurons; the hidden layer 2 contains D hidden neurons, and the 3D hidden neurons are converted into D hidden neurons by the hidden layer 2; finally, the D parameters are fitted by the fully connected layer and output to obtain the target characteristics such as the lower explosion limit, lower explosion limit and flash point of the combustible gas / vapor.

[0064] After constructing and training the multi-objective QSPR model of combustion and explosion characteristics, the model should be evaluated and verified. An unverified QSPR model may not guarantee that it can make effective predictions. The verification of the QSPR model is divided into internal and external verification. The internal verification is mainly used to judge its fit and stability. This application divides the data set into a training set for internal verification and a test set for external verification. The training set is mainly involved in the construction of the model, and the external verification of the model uses the test set (not involved in modeling) to verify the external prediction ability and generalization ability of the model; specifically, there are three ways:

[0065] (1) Model fit evaluation: The statistical indicators involved include the fitting correlation coefficient (Determination coefficient, R 2 ), root mean square error (RMSE), mean absolute error (MAE), mean relative error (MRE), and standard error (SE).

[0066] (2) Model stability analysis: “leave-alone-out” cross-validation and residual plot analysis can be used.

[0067] (3) Evaluation of the predictive ability of the model: external validation was performed using a test set that was not involved in the modeling. The validation parameters involved included the fitting correlation coefficient R mentioned in point (1). 2 , root mean square error RMSE, mean absolute error MAE, mean relative error MRE, standard error SE and external cross-validation coefficient (Q 2 Ext ).

[0068] S4, the explosion characteristics of combustible gas / steam are predicted through the constructed multi-objective QSPR model of explosion characteristics.

[0069] In addition, the application domain of the model, that is, the scope of application of the model, should be defined as clearly as possible, and the mechanism of the QSPR model should be explained. Williams diagram can be used to determine the application domain of the built model and verify whether there are abnormal points in the test set; the difficulty of the mechanism explanation of the model depends on the choice of descriptor and modeling method. For MLP and ResNet models, Shapley Additive exPlanations (SHAP) can be used for importance analysis to analyze the mechanism.

[0070] In a second aspect, the present application provides a system for executing the combustible gas / steam multi-objective explosion characteristics prediction method, comprising:

[0071] An explosion characteristic data acquisition unit collects the explosion characteristic data of the combustible gas / vapor to form an explosion characteristic data set, and divides the explosion characteristic data set into a training set and a test set according to a proportion;

[0072] A descriptor calculation and screening unit is used to optimize and perform frequency analysis on the molecular structure of the combustible gas / vapor, characterize the molecular structure of the combustible gas / vapor using quantum chemical descriptors, and screen the quantum chemical descriptors to obtain quantum chemical descriptors whose linear correlation meets the threshold requirements;

[0073] A prediction model building unit calculates and screens the corresponding quantum chemical descriptor parameter values ​​based on the training set, builds a multi-objective QSPR model of combustion and explosion characteristics, and uses a test set to test and evaluate the multi-objective QSPR model of combustion and explosion characteristics to verify the prediction ability of the model;

[0074] The explosion characteristics prediction unit predicts the explosion characteristics of combustible gas / steam through the constructed explosion characteristics multi-objective QSPR model.

[0075] Furthermore, the system also includes an application domain analysis unit and / or a microscopic mechanism analysis unit; wherein the application domain analysis unit is used to determine the scope of application of the multi-objective QSPR model of combustion and explosion characteristics, and the microscopic mechanism analysis unit is used to explain the mechanism of the multi-objective QSPR model of combustion and explosion characteristics.

[0076] It should be understood that the division of the processing modules in the above system is only a division of logical functions, and in actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. In addition, the processing modules in the system can be implemented in the form of a processor calling software; for example, the system includes a processor, the processor is connected to a memory, and instructions are stored in the memory. The processor calls the instructions stored in the memory to implement any of the above methods or implement the functions of the processing units of the system, wherein the processor is a general-purpose processor, such as a central processing unit or a microprocessor, and the memory is a memory within the system or a memory outside the system.

[0077] The following is a specific application example and Figure 2-Figure 7 , the implementation methods of this application are explained and illustrated in detail.

[0078] Step 1: Collection of combustible gas / vapor explosion characteristics data set;

[0079] This example takes the three explosion characteristics of lower explosion limit, upper explosion limit and flash point as the research objects, and collects 289 substances from the Design Institute for Physical Properties (DIPPR) engineering 801 database, which can find the upper explosion limit, lower explosion limit and flash point experimental values ​​at the same time. These 298 substances are hydrocarbons, alcohols, ketones, aldehydes, ethers, esters, carboxylic acids, phenols and other combustible gases / vapors composed of C, H, and O. The data set is divided into 80% for training set and 20% for test set.

[0080] Step 2: Calculation and screening of descriptors;

[0081] Molecular structure optimization: ORCA software was used to perform structural optimization and frequency analysis on the data set (using 6-311G** basis set and M062X functional, zero-damped DFT-D dispersion correction (D3zero)).

[0082] Calculation of descriptors: The molecular wave function information after optimization analysis is input into Multiwfn, and the quantitative molecular surface analysis function of Multiwfn is used to obtain the GIPF descriptor, dipole moment, highest occupied molecular orbital energy, lowest unoccupied molecular orbital energy and energy gap.

[0083] Screening of descriptors: The Pearson correlation coefficient is used to screen the descriptors to avoid large linear correlations between descriptors. In this embodiment, the threshold is set to 0.9, and descriptors with a Pearson correlation coefficient greater than 0.9 are considered to be strongly correlated, and descriptors with stronger correlations with other descriptors are removed from the descriptor set, and finally 9 descriptors are screened (Table 1).

[0084] Table 1 Descriptors that meet the relevance requirements

[0085]

[0086] Step 3: Construction of multi-objective prediction model;

[0087] This embodiment uses a residual network to construct an improved multi-objective QSPR model of combustion and explosion characteristics. The construction process of the multi-objective QSPR model of combustion and explosion characteristics is as follows:

[0088] (1) Select ReLU as the activation function and MAE as the loss function;

[0089] (2) The 9 descriptors are normalized using the BatchNorm1d function of the data preprocessing layer;

[0090] (3) The normalized descriptor is convolved using the one-dimensional convolution of the hidden layer 1, and the convolved descriptor is normalized again using the BatchNorm1d function;

[0091] (4) After processing by the maximum pooling layer;

[0092] (5) After two convolutions and one addition process in the residual block, it is processed by the average pooling layer, and all dimensions are flattened using the Flatten function to output 27 parameters;

[0093] (6) The above 27 parameters are transformed by hidden layer 2 to output 9 parameters;

[0094] (7) Outputting the target characteristics of combustible gas / vapor after fitting through the fully connected layer;

[0095] Step 4: Evaluation and verification of the model;

[0096] This embodiment evaluates the accuracy of the model by calculating several degrees of fit of the model, such as Figure 2-Figure 4 The figures show the comparison between the predicted values ​​of the lower explosion limit, upper explosion limit and flash point based on the ResNet model and the experimental values, and the fitting degrees are all within the error range.

[0097] Step 5: Mechanistic explanation of the model:

[0098] like Figure 5-Figure 7 The following are the average absolute contributions of each descriptor obtained by the SHAP method to the lower explosion limit, upper explosion limit and flash point of the ResNet model. n It has a significant impact on the lower and upper explosion limits of combustible gases / vapors. n As the value of A increases, the lower and upper explosion limits of the flammable gas / vapor decrease. n The flash point of combustible gas / vapor is significantly affected by ρ and A. n As the temperature increases, the flash point of the combustible gas / vapor increases.

[0099] The above embodiments are only for illustrating the technical concept and features of the present application, and their purpose is to enable technicians familiar with the technical field to understand the content of the present application and implement it accordingly, and they cannot be used to limit the protection scope of the present application. For ordinary technicians in the technical field, they can make several improvements and modifications without departing from the principles of the present application, and these improvements and modifications should also be regarded as the protection scope of the present application.

Claims

1. A method for predicting the multi-objective explosion characteristics of combustible gas / steam, characterized in that: It includes the following: S1, collecting the combustion and explosion characteristic data of combustible gas / vapor to form a combustion and explosion characteristic data set, and dividing the combustion and explosion characteristic data set into a training set and a test set according to a proportion; S2, optimizing and frequency analyzing the molecular structure of the combustible gas / vapor, characterizing the molecular structure of the combustible gas / vapor using quantum chemical descriptors, and screening the quantum chemical descriptors to obtain quantum chemical descriptors whose linear correlation meets the threshold requirements; S3, based on the training set, calculating and screening the corresponding quantum chemical descriptor parameter values, constructing a multi-objective QSPR model of combustion and explosion characteristics, and using a test set to test and evaluate the multi-objective QSPR model of combustion and explosion characteristics to verify the prediction ability of the model; S4, the explosion characteristics of combustible gas / steam are predicted through the constructed multi-objective QSPR model of explosion characteristics.

2. The combustible gas / steam multi-objective explosion characteristics prediction method according to claim 1 is characterized in that: In step S2, the quantum chemical descriptors are obtained by Multiwfn calculation to form a descriptor set, which includes GIPF descriptor, dipole moment, highest occupied molecular orbital energy, lowest unoccupied molecular orbital energy and energy gap.

3. The combustible gas / steam multi-objective explosion characteristics prediction method according to claim 2 is characterized in that: In step S2, the quantum chemical descriptors are screened using the Pearson correlation coefficient.

4. The combustible gas / steam multi-objective explosion characteristics prediction method according to claim 1, characterized in that: Construct a multi-objective QSPR model of explosion characteristics based on a multi-layer perceptron network. The construction process is as follows: S300, selecting an activation function and a loss function; S301, using the BatchNorm1d function of the data preprocessing layer to normalize the D descriptors obtained by screening; S302, input the normalized D descriptors into the hidden layer 1, and expand them to 3D parameters; S303, converting the 3D parameters into D parameters through hidden layer 2; S304, outputting the target explosion characteristics of the combustible gas / vapor after fitting through the fully connected layer; Here, D represents the number of quantum chemical descriptors whose linear correlation meets the threshold requirement.

5. The combustible gas / steam multi-objective explosion characteristics prediction method according to claim 1 is characterized in that: A multi-objective QSPR model of combustion and explosion characteristics based on residual network is constructed. The construction process is as follows: S310, selecting an activation function and a loss function; S311, using the BatchNorm1d function of the data preprocessing layer to normalize the D descriptors obtained by screening; S312, using the one-dimensional convolution of the hidden layer 1 to perform convolution processing on the normalized descriptor, and again using the BatchNorm1d function to normalize the convolution processed descriptor; S313, processed by the maximum pooling layer; S314, after two convolutions and one addition process of the residual block, and then processed by the average pooling layer, all dimensions are flattened using the Flatten function to output 3D parameters; S315, transform the parameters through hidden layer 2 and output D parameters; S316, outputting the target characteristics of the combustible gas / vapor after fitting through the fully connected layer; Here, D represents the number of quantum chemical descriptors whose linear correlation meets the threshold requirement.

6. The combustible gas / steam multi-objective explosion characteristics prediction method according to claim 1, characterized in that: The method also includes a step of analyzing the application domain of the multi-objective QSPR model of combustion and explosion characteristics.

7. The combustible gas / steam multi-objective explosion characteristics prediction method according to claim 1, characterized in that: It also includes the step of analyzing the microscopic mechanism that affects the combustion and explosion characteristics.

8. A system for executing the multi-objective combustion and explosion characteristics prediction method of combustible gas / steam according to any one of claims 1 to 7, characterized in that: include: An explosion characteristic data acquisition unit collects the explosion characteristic data of the combustible gas / vapor to form an explosion characteristic data set, and divides the explosion characteristic data set into a training set and a test set according to a proportion; A descriptor calculation and screening unit is used to optimize and perform frequency analysis on the molecular structure of the combustible gas / vapor, characterize the molecular structure of the combustible gas / vapor using quantum chemical descriptors, and screen the quantum chemical descriptors to obtain quantum chemical descriptors whose linear correlation meets the threshold requirements; A prediction model building unit calculates and screens the corresponding quantum chemical descriptor parameter values ​​based on the training set, builds a multi-objective QSPR model of combustion and explosion characteristics, and uses a test set to test and evaluate the multi-objective QSPR model of combustion and explosion characteristics to verify the prediction ability of the model; The explosion characteristics prediction unit predicts the explosion characteristics of combustible gas / steam through the constructed explosion characteristics multi-objective QSPR model.

9. The system according to claim 8, characterized in that include: The application domain analysis unit is used to analyze the application domain of the multi-objective QSPR model of combustion and explosion characteristics.

10. The system according to claim 8, characterized in that include: The microscopic mechanism analysis unit is used to analyze the microscopic mechanism that affects the combustion and explosion characteristics.

Images