Solvent explosion property efficient prediction method based on multi-dimensional molecule descriptor

Abstract

The invention discloses a solvent burning and explosion property efficient prediction method based on a multi-dimensional molecular descriptor, and relates to the technical field of burning and explosion safety engineering.The method comprises the steps that SMILES expression data of a target solvent and corresponding infrared spectrum data are obtained, the two kinds of data are processed to generate a molecular topological feature vector and a spectrum feature vector, and the molecular topological feature vector and the spectrum feature vector are used for predicting the burning and explosion property of the target solvent; splicing the molecular topological feature vector and the spectral feature vector to generate a mixed molecular descriptor; inputting the mixed molecule descriptor into a multi-task deep neural network to obtain a prediction result of the base value of the fire blast property; and constructing a collaborative correction mechanism, and adjusting the prediction base value by performing collaborative effect detection to obtain a final fire blast property prediction value. According to the method, multi-dimensional fusion molecular descriptor construction is carried out, a molecular topological feature vector and a spectral feature vector are combined and spliced to form a mixed molecular descriptor, the spectral features of a molecular structure and a functional group are captured at the same time, and the precision of describing the burning and explosion characteristics of the solvent is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of explosion safety engineering, and in particular to a method for efficiently predicting the explosion properties of solvents based on multi-dimensional molecular descriptors. Background Art

[0002] In the fields of chemistry, energy, and safety engineering, predicting the flammability and explosion properties of solvents is crucial for industrial production, hazardous chemical transportation, and safety assessments. These properties are typically determined by their molecular structure, chemical bond characteristics, functional group distribution, and environmental conditions (such as temperature and pressure). Traditional methods for predicting flammability and explosion properties rely primarily on experimental measurements or calculations based on empirical models. These methods have the following limitations:

[0003] Experimental determination method: Explosion and explosion tests are required, which is a complex and costly process, and has strict requirements on the experimental environment. Some hazardous solvents are even difficult to test directly.

[0004] Empirical model-based methods primarily rely on existing explosion data and make inferences through linear regression, qualitative rules, or chemometric methods. However, the accuracy of these methods is limited by data coverage, molecular structure diversity, and environmental factors, making them difficult to adapt to the prediction needs of new solvents or mixed solvent systems.

[0005] Therefore, an efficient prediction method for the explosive properties of solvents based on multidimensional molecular descriptors is urgently needed to solve the above problems. Summary of the Invention

[0006] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0007] In view of the above problems in the prior art, the present invention is proposed.

[0008] To solve the above technical problems, the present invention provides the following technical solution: an efficient prediction method for the explosion properties of solvents based on multidimensional molecular descriptors, the method comprising:

[0009] Obtaining SMILES expression data and corresponding infrared spectrum data of the target solvent, processing the two data separately to generate: a molecular topological feature vector and a spectral feature vector, and splicing the molecular topological feature vector and the spectral feature vector to generate a hybrid molecular descriptor;

[0010] Inputting the mixed molecular descriptor into a multi-task deep neural network to obtain a prediction result of the base value of the explosive property;

[0011] A synergistic correction mechanism is constructed to adjust the prediction base value through synergistic effect detection to obtain the final prediction value of the explosion properties.

[0012] As a preferred embodiment of the method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors described in the present invention, the synergistic correction mechanism includes:

[0013] S1: Calculate the cosine similarity of the mixed molecule descriptors of each molecule pair in the mixed solvent to generate a synergistic effect matrix;

[0014] S2: Determine whether the element value in the generated synergy matrix is greater than the similarity threshold;

[0015] S3: If it is greater than the threshold, it is determined to be a strong synergistic effect. At this time, the synergistic effect compensation algorithm is activated to generate a compensation coefficient G, and the compensation strength ΔP is determined according to the compensation coefficient;

[0016] S4: According to the calculated compensation intensity, the predicted base value is modified to obtain the final predicted value Q f =Q×(1+∑ΔP k );

[0017] Among them, Q represents the predicted base value, ΔP k Represents the compensated intensities calculated for molecular pairs that exceed the similarity threshold.

[0018] As a preferred embodiment of the method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors described in the present invention, the calculation formula of the compensation coefficient G is:

[0019] G=min(G max ,1+(SEM[i][j]-α) / δ);

[0020] Among them, G max represents the maximum gain upper limit, α represents the similarity threshold, δ represents the length of the linear gain interval, and SEM[i][j] represents the HMD cosine similarity between functional groups i and j;

[0021] The calculation formula of the compensation intensity ΔP is: ΔP=G×SEM[i][j]×Q. The calculation of the compensation intensity ΔP is used as the basis for the collaborative correction of the predicted base value.

[0022] As a preferred solution of the method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors according to the present invention, when the ambient temperature is higher than a first temperature threshold, the SEM[i][j] table needs to be adjusted by introducing a temperature correction factor;

[0023] That is, SEM[i][j] = HMD cosine similarity of functional groups i and j × β, and the temperature correction factor β is expressed as:

[0024] β=1+γ×(TT ref ) / ΔT;

[0025] Among them, γ represents the temperature influence coefficient, T represents the current ambient temperature, T ref represents the first temperature threshold, ΔT represents the normalized range parameter of temperature change;

[0026] When the ambient temperature is higher than the second temperature threshold, the similarity threshold needs to be adjusted by constructing a threshold adjustment formula, which is:

[0027] Similarity threshold after adjustment = similarity threshold before adjustment - γ×max(0.(TT th ) / ΔT);

[0028] Among them, T th Represents the second temperature threshold. The dynamic coupling relationship between the temperature parameter and the similarity threshold is constructed through this formula.

[0029] As a preferred solution of the method for efficiently predicting the explosive properties of solvents based on multi-dimensional molecular descriptors of the present invention, the generation process of the mixed molecular descriptors further includes: a concentration weight compensation mechanism;

[0030] Specifically: obtain the concentration ratio data of each component in the mixed solvent;

[0031] When the concentration of a component exceeds the set concentration threshold, a weighted calculation is performed on its mixed molecular descriptor. The calculation formula is:

[0032] Among them, HMD i represents the mixed molecule descriptor before adjustment, HMD' i represents the adjusted mixed molecule descriptor, represents the concentration gain coefficient, C represents the component concentration, and C th represents the component concentration threshold, and ΔC is used to convert the concentration limit range into a standardized parameter. Through the concentration gain coefficient and adjustment span, nonlinear enhancement of the characteristic vector of high-concentration components is achieved, breaking through the limitations of traditional linear superposition models.

[0033] As a preferred embodiment of the method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors described in the present invention, the SMILES expression data is processed by a molecular topology feature extraction model to generate a molecular topology feature vector. The molecular topology feature extraction model is a Tree-LSTM network, and its data processing process includes:

[0034] Convert SMILES expressions into directed acyclic graph (DAG) structure data;

[0035] Encode atomic nodes into vectors through word embedding layer;

[0036] The LSTM node state is propagated along the topological path of the directed acyclic graph, and a 256-dimensional molecular topological feature vector is obtained by aggregation.

[0037] As a preferred embodiment of the method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors described in the present invention, the infrared spectral data is processed by a spectral feature extraction model to generate a spectral feature vector. The spectral feature extraction model uses a 1D-CNN with frequency band enhancement, and its data processing includes:

[0038] For FTIR data, a 3-fold convolution kernel weight is set in the functional group characteristic frequency band;

[0039] After performing five layers of convolution operations, a 256-dimensional spectral feature vector is generated.

[0040] As a preferred solution of the method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors described in the present invention, the generation rule of the mixed molecular descriptor is: align the 256-dimensional molecular topological feature vector and the 256-dimensional spectral feature vector in dimensions and then splice them by dimension.

[0041] The present invention also discloses a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of the above-mentioned method for efficiently predicting the explosion properties of solvents based on multi-dimensional molecular descriptors.

[0042] The present invention also discloses a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned method for efficiently predicting the explosive properties of solvents based on multi-dimensional molecular descriptors are implemented.

[0043] Beneficial effects of the present invention:

[0044] 1. The present invention constructs a multidimensional fusion molecular descriptor (HMD): combining the molecular topological feature vector (obtained by processing the SMILES expression by the Tree-LSTM network) and the spectral feature vector (extracting FTIR spectrum information by the 1D-CNN network), and splicing the two to form a hybrid molecular descriptor, thereby simultaneously capturing the spectral characteristics of the molecular structure and functional groups, and improving the accuracy of describing the explosion characteristics of the solvent.

[0045] 2. The present invention calculates the cosine similarity matrix of molecular pairs for mixed solvent systems and uses a synergistic effect compensation algorithm to dynamically adjust the predicted values, thus avoiding the limitations of the traditional linear superposition model in complex mixed solvents.

[0046] 3. The present invention establishes a coupling relationship between temperature influence and molecular similarity through a temperature correction factor and a dynamic threshold adjustment mechanism, ensuring that the prediction model maintains high reliability under different environmental conditions. For situations where the concentration of specific components is high, a concentration gain coefficient is introduced to perform nonlinear enhancement of the HMD eigenvector, effectively solving the problem of insufficient contribution of high-concentration solvents in mixed systems and improving the adaptability of the model in situations with high component concentrations. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0048] Figure 1 This is a schematic diagram of an efficient prediction method for the explosive properties of solvents based on multi-dimensional molecular descriptors proposed in the present invention;

[0049] Figure 2 Data flow chart of a method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors proposed in the present invention. DETAILED DESCRIPTION

[0050] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0051] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0052] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0053] Example 1

[0054] Reference Figure 1-2, as one embodiment of the present invention, provides an efficient prediction method for the explosive properties of solvents based on multidimensional molecular descriptors, the method comprising the following steps:

[0055] Step 1: Obtain the SMILES expression data and corresponding infrared spectrum data of the target solvent, process the two data separately to generate: molecular (256-dimensional) topological feature vector and (256-dimensional) spectral feature vector, splice the molecular topological feature vector and the spectral feature vector to generate a hybrid molecular descriptor (HMD).

[0056] Specifically, the SMILES expression data is processed by the molecular topology feature extraction model to generate a molecular topology feature vector. The molecular topology feature extraction model is a Tree-LSTM network.

[0057] First of all, it should be noted that the SMILES (Simplified Molecular Input Line Entry System) mentioned above is a string representation method for representing molecular structures.

[0058] The Tree-LSTM network data processing process involves converting SMILES expressions into directed acyclic graph (DAG) structured data. This conversion primarily involves using atoms in a molecule as nodes and chemical bonds as edges to form a graph structure, thereby better capturing the topological characteristics of the molecule. For example, the SMILES expression for ethanol (C2H5OH) is CCO, and its DAG structure is as follows: CC-0, with single bonds between C (carbon atoms).

[0059] A single bond is formed between C and O (oxygen atom).

[0060] The word embedding layer is used to encode atomic nodes into vectors. In the DAG structure, each node (atom) needs to be converted into a vector form that can be processed by the model. Therefore, the word embedding layer is used to encode each atom into a vector representation of a fixed dimension.

[0061] For example: carbon (C) → vector representation [0.2, 0.8, ..., 0.1], oxygen (O) → vector representation [0.5, 0.4, ..., 0.3]. The vector representation is learned through training, so that atoms with similar chemical properties are closer in the vector space.

[0062] The LSTM node state is propagated along the topological path of the directed acyclic graph, and a 256-dimensional molecular topological feature vector is obtained by aggregation.

[0063] Specifically, Tree-LSTM transmits information along the topological path of the DAG structure. Each node receives information from its child nodes and is updated through the LSTM unit. This propagation method can effectively capture the hierarchical relationship of the molecular structure and avoid the information loss caused by the simple RNN structure.

[0064] Let’s take ethanol as an example: C1 first receives the information from C2 and combines it with its own state update. C1 then combines with O to finally aggregate the information of the entire molecule.

[0065] After Tree-LSTM processing, the entire DAG structure is compressed into a fixed-dimensional (256-dimensional) molecular topology feature vector, denoted as: HMD topo ∈R 256 , which can describe the overall topological characteristics of the molecule, including atom types, bond connections and structural information.

[0066] The infrared spectrum data (FTIR) is processed by the spectral feature extraction model to generate spectral feature vectors. The spectral feature extraction model uses a 1D-CNN with frequency band enhancement. Its data processing includes:

[0067] For FTIR data, a 3-fold convolution kernel weight is set in the functional group characteristic frequency band;

[0068] Certain frequency bands in FTIR data are associated with specific functional groups, for example:

[0069] CH stretching vibration (2800–3100 cm-1)

[0070] OH stretching vibration (3200–3600 cm-1)

[0071] C=O stretching vibration (1650–1750 cm-1), set the convolution kernel weight 3 times in these key frequency bands to enhance the model's ability to focus on functional groups.

[0072] After performing five layers of convolution operations, a 256-dimensional spectral feature vector is generated. That is, a one-dimensional convolutional neural network (1D-CNN) is used to gradually extract the high-level features of the spectral data through five layers of convolution operations.

[0073] That is, the first layer of convolution: extracts local frequency band features;

[0074] The second to fourth convolution layers: enhance the fusion of global information;

[0075] The fifth convolution layer forms the final 256-dimensional spectral feature vector.

[0076] After 1D-CNN processing, the FTIR data is converted into a fixed-length 256-dimensional spectral feature vector, denoted as: HMDspec ∈R 256 .

[0077] The generation rule of the hybrid molecular descriptor is: align the 256-dimensional molecular topological feature vector and the 256-dimensional spectral feature vector and then splice them by dimension, that is: The final HMD vector dimension is 512 (256+256), which can fully characterize the topological structure and spectral characteristics of the molecule.

[0078] Step 2: Input the mixed molecular descriptor into the multi-task deep neural network (MDNN) to obtain the prediction results of the base value of the explosive properties.

[0079] This example briefly describes the predictions of a multi-task deep neural network: MDNN is a neural network model that simultaneously learns multiple tasks. In predicting explosive properties, its basic approach is to first input characteristic molecular data, then extract key information from this data through a multi-layer neural network, and finally output predicted values for the explosive properties (such as the explosion limit).

[0080] Step 3: Construct a synergistic correction mechanism to adjust the predicted base value through synergistic effect detection to obtain the final prediction value of the explosion properties. The synergistic correction mechanism includes:

[0081] S1: Calculate the cosine similarity of the molecular descriptors for each pair of molecules in the mixed solvent to generate a synergistic effect matrix. This is because the interactions between different molecules in a mixed solvent system significantly affect its explosive properties. To this end, it is necessary to construct a synergistic effect matrix (SEM), which is used to measure the similarity between molecules in the mixed system to determine whether synergy exists.

[0082] The calculation formula is as follows:

[0083]

[0084] In the above formula, the numerator is the dot product operation, which measures the similarity between two vectors in the same direction. The denominator represents the vector modulus calculation. This modulus value represents the length of the descriptor of each molecule in the feature space. The present invention introduces a hybrid molecular descriptor as the basic data for calculation. It contains molecular information from multiple sources and can fully describe the properties of the molecule.

[0085] S2: Determine whether the element value in the generated synergy matrix is greater than the similarity threshold;

[0086] S3: If it is greater than the threshold, it is determined to be a strong synergistic effect. At this time, the synergistic effect compensation algorithm is activated to generate the compensation coefficient G. The calculation formula of the compensation coefficient G is:

[0087] G=min(G max,1+(SEM[i][j]-α) / δ);

[0088] Among them, G max represents the maximum gain upper limit (safety constraint to prevent overcompensation), α represents the similarity threshold, δ represents the length of the linear gain interval, and SEM[i][j] represents the HMD cosine similarity between functional groups i and j.

[0089] The compensation intensity ΔP is determined according to the compensation coefficient; the calculation formula of the compensation intensity ΔP is: ΔP = G × SEM[i][j] × Q, and the calculated compensation intensity ΔP is used as the basis for the collaborative correction of the predicted base value.

[0090] S4: According to the calculated compensation intensity, the predicted base value is modified to obtain the final predicted value Q f =Q×(1+∑ΔP k );

[0091] Among them, Q represents the predicted base value, ΔP k Represents the compensated intensities calculated for molecular pairs that exceed the similarity threshold.

[0092] Example 2

[0093] The difference from Example 1 is that the method of the present invention also constructs a temperature-similarity dynamic coupling rule and a concentration weight compensation rule.

[0094] When the ambient temperature is higher than the first temperature threshold, the SEM[i][j] table needs to be adjusted by introducing a temperature correction factor. That is, SEM[i][j] = HMD cosine similarity of functional groups i and j × β, and the temperature correction factor β is expressed as:

[0095] β=1+γ×(TT ref ) / ΔT;

[0096] Among them, γ represents the temperature influence coefficient, T represents the current ambient temperature, T ref represents the first temperature threshold, ΔT represents the normalized range parameter of temperature change;

[0097] When the ambient temperature is higher than the second temperature threshold (the second temperature threshold is greater than the first temperature threshold), the similarity threshold needs to be adjusted by constructing a threshold adjustment formula. The threshold adjustment formula is:

[0098] Similarity threshold after adjustment = similarity threshold before adjustment - γ×max(0.(TT th ) / ΔT);

[0099] Among them, T threpresents the second temperature threshold. This formula constructs a dynamic coupling relationship between the temperature parameter and the similarity threshold (expanding the detection range of synergistic effects in high temperature environments and identifying potential dangerous combinations earlier).

[0100] In concentration weight compensation, we first need to obtain the concentration ratio data of each component in the mixed solvent.

[0101] When the concentration of a component exceeds the set concentration threshold (such as 60%), a weighted calculation is performed on its mixed molecular descriptor. The calculation formula is:

[0102] Among them, HMD i represents the mixed molecule descriptor before adjustment, HMD' i represents the adjusted mixed molecule descriptor, represents the concentration gain coefficient, C represents the component concentration, and C th represents the component concentration threshold, and ΔC is used to convert the concentration limit range into a standardized parameter. Through the concentration gain coefficient and adjustment span, nonlinear enhancement of the characteristic vector of high-concentration components is achieved, breaking through the limitations of traditional linear superposition models.

[0103] This embodiment also provides a computer device, which is suitable for a method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors as proposed in the above embodiment.

[0104] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0105] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for efficiently predicting the explosive properties of solvents based on multi-dimensional molecular descriptors as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.

Claims

1. An efficient prediction method for the explosive properties of solvents based on multidimensional molecular descriptors, characterized by: The method includes: Obtaining SMILES expression data and corresponding infrared spectrum data of the target solvent, processing the two data separately to generate: a molecular topological feature vector and a spectral feature vector, and splicing the molecular topological feature vector and the spectral feature vector to generate a hybrid molecular descriptor; Inputting the mixed molecular descriptor into a multi-task deep neural network to obtain a prediction result of the base value of the explosive property; A synergistic correction mechanism is constructed to adjust the prediction base value through synergistic effect detection to obtain the final prediction value of the explosion properties.

2. The method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors according to claim 1, characterized in that: The collaborative correction mechanism includes: S1: Calculate the cosine similarity of the mixed molecule descriptors of each molecule pair in the mixed solvent to generate a synergistic effect matrix; S2: Determine whether the element value in the generated synergy matrix is greater than the similarity threshold; S3: If it is greater than the threshold, it is determined to be a strong synergistic effect. At this time, the synergistic effect compensation algorithm is activated to generate a compensation coefficient G, and the compensation strength ΔP is determined according to the compensation coefficient; S4: According to the calculated compensation intensity, the predicted base value is modified to obtain the final predicted value Q f =Q×(1+∑ΔP k ); Among them, Q represents the predicted base value, ΔP k Represents the compensated intensities calculated for molecular pairs that exceed the similarity threshold.

3. The method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors according to claim 2, characterized in that: The calculation formula of the compensation coefficient G is: G=min(G max ,1+(SEM[i][j]-a) / d); Among them, G max represents the maximum gain upper limit, α represents the similarity threshold, δ represents the length of the linear gain interval, and SEM[i][j] represents the HMD cosine similarity between functional groups i and j; The calculation formula of the compensation intensity ΔP is: ΔP=G×SEM[i][j]×Q. The calculation of the compensation intensity ΔP is used as the basis for the collaborative correction of the predicted base value.

4. The method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors according to claim 3, characterized in that: When the ambient temperature is higher than the first temperature threshold, the SEM[i][j] needs to be adjusted by introducing a temperature correction factor; That is, SEM[i][j] = HMD cosine similarity of functional groups i and j × β, and the temperature correction factor β is expressed as: β=1+γ×(T-T ref ) / ΔT; Among them, γ represents the temperature influence coefficient, T represents the current ambient temperature, T ref represents the first temperature threshold, ΔT represents the normalized range parameter of temperature change; When the ambient temperature is higher than the second temperature threshold, the similarity threshold needs to be adjusted by constructing a threshold adjustment formula, which is: Similarity threshold after adjustment = similarity threshold before adjustment - γ×max(0.(TT th ) / ΔT); Among them, T th Represents the second temperature threshold. The dynamic coupling relationship between the temperature parameter and the similarity threshold is constructed through this formula.

5. The method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors according to claim 1, characterized in that: The generation process of the hybrid molecular descriptor also includes: a concentration weight compensation mechanism; Specifically: obtain the concentration ratio data of each component in the mixed solvent; When the concentration of a component exceeds the set concentration threshold, a weighted calculation is performed on its mixed molecular descriptor. The calculation formula is: HMD' i =HMD i ×[1+θ×(CC th ) / ΔC] Among them, HMD i represents the mixed molecule descriptor before adjustment, HMD' i represents the adjusted mixed molecular descriptor, θ represents the concentration gain coefficient, C represents the component concentration, and C th represents the component concentration threshold, and ΔC is used to convert the concentration limit range into a standardized parameter.

6. The method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors according to claim 1, characterized in that: The SMILES expression data is processed by a molecular topology feature extraction model to generate a molecular topology feature vector. The molecular topology feature extraction model is a Tree-LSTM network, and its data processing process includes: Convert SMILES expression into directed acyclic graph structure data; Encode atomic nodes into vectors through word embedding layer; The LSTM node states are propagated along the topological path of the directed acyclic graph, and a 256-dimensional molecular topological feature vector is obtained by aggregation.

7. The method for efficiently predicting the explosive properties of solvents based on multidimensional molecular descriptors according to claim 6, characterized in that: The infrared spectrum data is processed by a spectrum feature extraction model to generate a spectrum feature vector. The spectrum feature extraction model uses a 1D-CNN with frequency band enhancement. The data processing includes: For FTIR data, a 3-fold convolution kernel weight is set in the functional group characteristic frequency band; After performing five layers of convolution operations, a 256-dimensional spectral feature vector is generated.

8. The method for efficiently predicting the explosion properties of solvents based on multidimensional molecular descriptors according to claim 7, characterized in that: The generation rule of the hybrid molecular descriptor is: aligning the dimensions of the 256-dimensional molecular topological feature vector and the 256-dimensional spectral feature vector and then splicing them according to the dimensions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for efficiently predicting the explosion properties of solvents based on multidimensional molecular descriptors according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for efficiently predicting the explosion properties of solvents based on multidimensional molecular descriptors according to any one of claims 1 to 8 are implemented.

Images