Machine learning based method for predicting secondary transition temperatures of polymeric materials

Abstract

The application discloses a polymer material secondary transition temperature prediction method based on machine learning, and comprises the following steps: obtaining a polymer material to be detected; constructing a secondary transition temperature prediction model; inputting the polymer material to be detected into the secondary transition temperature prediction model for prediction to obtain a prediction result; wherein the secondary transition temperature prediction model is obtained through training of a training set and testing of a test set. β The application can effectively capture the internal relationship between a polymer molecular structure and a secondary transition temperature (T β ) under the condition of less data, and reveal the rotation nature of the secondary transition temperature (T β ) of the polymer in the temperature change process.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence materials research and design technology, and in particular to a method for predicting the secondary transition temperature of polymer materials based on machine learning. Background Technology

[0002] Secondary transformation temperature (T) β Secondary transition temperature (T0) is an important thermodynamic property of polymer materials exhibited during temperature changes. It reflects the local mobility and flexibility of polymer chain segments and is of great significance for material processing, performance control, and application development. β Accurate prediction and understanding of secondary transition temperatures (T0) are crucial for the design and optimization of new materials. However, traditional experimental methods, such as differential scanning calorimetry (DSC) and dynamic mechanical analysis (DMA), have limitations such as cumbersome sample preparation, long testing cycles, and difficulties in data interpretation. Furthermore, the accuracy of secondary transition temperatures (T0) is limited. β The scarcity of experimental data for this substance severely limits in-depth understanding and widespread application of its physical mechanisms.

[0003] In recent years, machine learning (ML), as an emerging data-driven approach, has become an important tool in the field of materials science due to its powerful data analysis capabilities. Machine learning algorithms can be used to uncover the relationship between chemical structure and secondary transition temperature (T). β Based on existing experimental data, an efficient prediction model was established to identify the potential correlation between the two secondary transition temperatures (T0 and T1), significantly improving the accuracy and reliability of predictions. However, due to the limitations of existing secondary transition temperatures (T0), the prediction accuracy and reliability are still limited. β Due to the scarcity of data and the complex nonlinear relationship between data and molecular structure, the application of machine learning models in this field still faces certain challenges.

[0004] Therefore, how to fully utilize limited experimental data and combine it with machine learning methods to develop a secondary transition temperature (T) with high accuracy and high generalization ability? β Predictive models have become an important topic in current materials science research. Summary of the Invention

[0005] To address the technical problems existing in the prior art, this invention proposes a machine learning-based method for predicting the secondary transition temperature of polymer materials. This novel method can effectively compensate for the shortcomings of traditional experimental methods, provide a scientific basis for the design, processing and application of polymer materials, and promote the widespread application of polymer materials in high-performance fields.

[0006] To achieve the above objectives, this invention provides a machine learning-based method for predicting the secondary transition temperature of polymer materials, comprising:

[0007] Obtain the polymer material to be tested;

[0008] constructing a secondary transition temperature prediction model;

[0009] inputting the polymer material to be detected into the secondary transition temperature prediction model for prediction, and obtaining a prediction result;

[0010] The secondary transition temperature prediction model is obtained through training of a training set and testing of a test set.

[0011] Preferably, constructing the secondary transition temperature prediction model comprises:

[0012] collecting a chemical structural formula of a target polymer and corresponding secondary transition temperature data, dividing the secondary transition temperature data to obtain divided data;

[0013] converting the chemical structural formula of the target polymer into a SMILES string, and parsing the SMILES string to extract key structural features in the target polymer molecule;

[0014] establishing the secondary transition temperature prediction model based on the key structural features and the divided data.

[0015] Preferably, dividing the secondary transition temperature data comprises:

[0016] dividing the secondary transition temperature data into data affected by a main chain and data affected by a side chain through a rotation mechanism of a secondary transition temperature.

[0017] Preferably, converting the chemical structural formula of the target polymer into a SMILES string comprises:

[0018] drawing a molecular structure diagram of the target polymer through CHEMDRAW software to graphically represent a two-dimensional structure of the target polymer;

[0019] converting the molecular structure diagram into a simplified molecular input linear representation, i.e., the SMILES string, through CHEMDRAW software, and performing quality control on the SMILES string.

[0020] Preferably, parsing the SMILES string to extract key structural features in the polymer molecule comprises:

[0021] decomposing the given SMILES string into basic chemical structural units through a molecular fragment method, extracting heteroatoms, types of chemical bonds, and cyclic structure features in the basic chemical structural units, and calculating physical and chemical characteristics of different structural units, wherein the physical and chemical characteristics include but are not limited to rotational freedom, polarity, and flexibility;

[0022] The heteroatoms, types of chemical bonds and cyclic structure characteristics are standardized, and data cleaning and outlier removal are performed;

[0023] The extracted features are subjected to data conversion and processing according to different processing methods, a feature set corresponding to each method is generated, and different feature sets are integrated into comprehensive features of the target polymer, i.e., key structural features in the target polymer molecule.

[0024] Preferably, the key structural features in the target polymer molecule include:

[0025] Chemical descriptors: total number of high electronegativity, number of fused rings, number of six-membered rings, number of five-membered rings, number of carbonyl groups, total number of single bonds, number of single bonds on the main chain, number of hydrogen bonds;

[0026] Based on the molecular weight of the two ends of the single bond: the sum of the molecular weight difference of the side chain groups at the two ends of the single bond, the sum of the molecular weight of the side chain groups at the two ends of the single bond, the sum of the molecular weight difference of the side chain groups at the two ends of the single bond on the main chain, and the sum of the molecular weight of the side chain groups on the branch chain.

[0027] Preferably, the different processing methods include molecular fragment method, fused ring assignment method, benzene ring assignment method, and atom normalization plus fused ring / benzene ring assignment method.

[0028] Preferably, training the secondary transition temperature prediction model comprises:

[0029] Integrating the key structural features with the experimentally measured secondary transition temperature data to form an input data set and an output data set for model training, wherein the input data is the molecular features extracted by the molecular fragment method, and the output data is the secondary transition temperature value predicted by the model;

[0030] Selecting a machine learning algorithm, modeling according to the characteristics of the data, training the secondary transition temperature prediction model through the training set, testing through the test set, and evaluating the prediction accuracy of the model.

[0031] Preferably, the evaluation parameters used for testing the secondary transition temperature prediction model are:

[0032]

[0033] In the formula, R 2 is the coefficient of determination; y i is the true value, is the predicted value, is the average value of the true value, n is the total number of samples, and i is the sample index value.

[0034] Compared with the prior art, the present application has the following advantages and technical effects:

[0035] Compared with traditional secondary transition temperature (T β ) data sources, the present application can provide higher prediction accuracy in the case of data scarcity by extracting molecular structure features using the fragment method and combining machine learning models for training. By analyzing the chemical structure and physical and chemical properties of polymers, the present application method can capture the rotational nature of the secondary transition temperature, overcoming the limitations of traditional experimental methods. In addition, the present application has high flexibility, suitable for different types of polymers, and can be widely used in the design and optimization process of new materials, significantly improving the efficiency and accuracy of secondary transition temperature prediction. BRIEF DESCRIPTION OF DRAWINGS

[0036] The accompanying drawings, which form a part of this application, are intended to provide further understanding of the application and are incorporated herein in their entirety. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute undue limitations on the present application. In the drawings:

[0037] Figure 1 Flow chart of the machine learning-based polymer material secondary transition temperature prediction method of the embodiments of the present application;

[0038] Figure 2 Data distribution graph of the machine learning-based polymer material secondary transition temperature prediction method of the embodiments of the present application affected by the backbone;

[0039] Figure 3 Data distribution graph of the machine learning-based polymer material secondary transition temperature prediction method of the embodiments of the present application affected by the side chain. DETAILED DESCRIPTION

[0040] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0041] It should be noted that the steps shown in the flow chart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flow chart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0042] In the secondary transition temperature (T β) range, the mobility of polymer segments is significantly enhanced, ranging from local vibration and small bond rotation, followed by gradual transition to larger range of segment motion. The rotation mechanism of the main chain is mainly affected by the connecting groups (such as methyl, phenyl, acyl, etc.) connected to it, and the steric hindrance, charge distribution and polarity characteristics of these groups directly affect the rotation freedom and flexibility of the main chain. In this process, larger groups or polar groups usually weaken the rotation ability of the main chain, leading to the increase of the secondary transition temperature (T β ) of the polymer. At the same time, the rotation of the side chain is mainly dominated by the influence of the volume, polarity and high electronegativity atoms of the side groups. In particular, there is a close relationship between the molecular mobility of the side chain and the volume and polarity of the group, which further regulates the secondary transition temperature (T β ) by adjusting the rotation freedom of the single bond. The change of the secondary transition temperature (T β ) is essentially due to the rotation freedom of the single bond, so studying the behavior of the single bond is the key to understanding and predicting the secondary transition temperature (T β ). By deeply analyzing the influence of the rotation freedom of the single bond, the relationship between the mobility of the polymer segment and the secondary transition temperature (T β ) can be better revealed.

[0043] The present application proposes a machine learning-based polymer material secondary transition temperature prediction method, comprising:

[0044] Obtaining a polymer material to be detected;

[0045] Constructing a secondary transition temperature prediction model;

[0046] Inputting the polymer material to be detected into the secondary transition temperature prediction model for prediction to obtain a prediction result;

[0047] Wherein, the secondary transition temperature prediction model is obtained by training a training set and testing a test set;

[0048] In this embodiment, the secondary transition temperature prediction model can also be used to predict the secondary transition temperature value of unknown samples.

[0049] Further, constructing a secondary transition temperature prediction model includes (for example Figure 1 ):

[0050] Collecting the chemical structural formula of the target polymer and the corresponding secondary transition temperature data, dividing the secondary transition temperature data to obtain the divided data;

[0051] Converting the chemical structural formula of the target polymer into a SMILES string, and parsing the SMILES string to extract key structural features in the target polymer molecule;

[0052] establishing the secondary transition temperature prediction model based on the key structural features and the divided data.

[0053] The embodiment combines regression analysis method and fragmentation processing of polymer molecules to decompose the polymer molecules into a plurality of basic structural units, deeply studies the rotation nature of the secondary transition temperature (T β ) of the polymer material, and explores micro factors affecting the secondary transition temperature (T β ), such as the rotation freedom of single bond, the flexibility of molecular chain segment, and the intramolecular interaction. In order to effectively capture these influencing factors, the molecular fragmentation method, the condensed ring assignment method, the benzene ring assignment method, and the atomic normalization method are used respectively.

[0054] Specifically, the chemical structural formula and the corresponding secondary transition temperature (T β ) data of the target polymer are collected from literature or books, and the data are divided into two categories, i.e., the data affected by the main chain and the data affected by the side chain, according to the rotation mechanism of the secondary transition temperature (T β ). The distribution of the data affected by the main chain is shown in Figure 2 , and the distribution of the data affected by the side chain is shown in Figure 3 .

[0055] Further, the conversion of the chemical structural formula of the target polymer into the SMILES string includes:

[0056] The molecular structure diagram of the polymer is drawn using the CHEMDRAW software, the two-dimensional structure of the polymer is graphically represented, and the basis for subsequent data processing and analysis is provided;

[0057] The drawn molecular structure diagram is converted into a simplified molecular input linear representation (SMILES string) by the CHEMDRAW software, so as to facilitate subsequent calculation and analysis;

[0058] The generated SMILES string is subjected to quality control to ensure that the converted SMILES string is accurate, standard, and can reflect the real chemical structure of the polymer.

[0059] The chemical structural formula of the specific polymer is converted into the SMILES string, and the processing process includes:

[0060] The chemical structural formula of the polymer is converted into a standard SMILES string to ensure accurate representation of each atom and bond;

[0061] The “*” symbol is used to mark both ends of the polymer in the SMILES structure to represent the start and end of the polymer chain, and also to represent the repeated unit part;

[0062] For polymers with ring structures, ensure the accuracy of ring closure and size representation, avoid duplication or loss of ring labels; when processing polymers containing stereochemical information, add stereochemical symbols for cis / trans isomers appropriately;

[0063] Clean and standardize the structure formula, remove redundant or unnecessary duplicate information, and make it comply with the SMILES standard;

[0064] Further split the structure of the polymer, identify the main chain and side chain parts, and prepare for subsequent classification and analysis.

[0065] Further, the SMILES string is parsed to extract key structural features in the polymer molecule, including:

[0066] By molecular fragment method, the given SMILES string is decomposed into basic chemical structural units, and the heteroatoms, types of chemical bonds and cyclic structure features in the basic chemical structural units are extracted, and the physical and chemical characteristics of different structural units are calculated, including but not limited to rotation freedom, polarity, flexibility;

[0067] Standardize the heteroatoms, types of chemical bonds and cyclic structure features, and perform data cleaning and outlier removal;

[0068] According to different processing methods, the extracted features are data converted and processed to generate feature sets corresponding to each method, and different feature sets are integrated into the comprehensive features of the target polymer, i.e. the key structural features in the target polymer molecule.

[0069] During the assignment process, the influence of various structural units on the secondary transition temperature (T β ) will be considered, and these influencing factors will be converted into numerical features available for the model through quantitative analysis, to facilitate subsequent machine learning modeling and secondary transition temperature (T β ) prediction analysis.

[0070] Specifically, the feature values are extracted by the molecular fragment method, and the processing process specifically includes:

[0071] The polymer molecular structure is split into individual fragments or substructures (such as rings, heteroatoms, etc.);

[0072] Identify and classify each chemical fragment to ensure accurate labeling as single bond, double bond, ring structure or heteroatom, etc.

[0073] Calculate the physical and chemical characteristics of each fragment, including but not limited to rotation freedom, polarity, flexibility, etc.

[0074] Standardize the extracted fragments to ensure that each fragment complies with the rules of the molecular fragment method;

[0075] Data cleaning and outlier removal are performed to ensure the accuracy and consistency of the feature values;

[0076] The extracted features are converted and processed according to different processing methods to generate feature sets corresponding to each method. Finally, these processed feature sets are integrated into the comprehensive features of the polymer as input data for subsequent modeling and analysis.

[0077] Further, the data feature descriptors affected by the main chain include: the total number of high electronegativity, the number of fused rings, the number of six-membered rings, the number of five-membered rings, the number of carbonyl groups, the number of single bonds, the number of single bonds on the main chain, the number of hydrogen bonds, the sum of the molecular weight difference of the side chain groups at both ends of the single bond, the sum of the molecular weight of the side chain groups at both ends of the single bond, the sum of the molecular weight difference of the side chain groups at both ends of the single bond on the main chain, and the sum of the molecular weight of the side chain groups on the side chain. Among them, the high electronegativity atoms include O, N, F, Cl, and Br.

[0078] For the feature descriptors of the side chain data, the total number of high electronegativity, the number of carbonyl groups, the number of hydrogen bonds, the number of single bonds on the side chain, and the sum of the molecular weight on the side chain are included. Among them, the high electronegativity atoms include O, N, F, Cl, and Br.

[0079] Further, different processing methods include molecular fragment method, fused ring assignment method, benzene ring assignment method, and atomic normalization plus fused ring / benzene ring assignment method.

[0080] Specifically, the molecular fragment method, the feature descriptors of this method include: the sum of the molecular weight difference of the side chain groups at both ends of the single bond of the polymer, the sum of the molecular weight of the side chain groups at both ends of the single bond, the total number of high electronegativity, the number of carbonyl groups, the number of hydrogen bonds, and the number of single bonds. Among them, the high electronegativity atoms include O, N, F, Cl, and Br.

[0081] The fused ring assignment method, by identifying and assigning the fused ring structure in the polymer molecule, extracts its specific structural features, and quantifies according to the size, number and atomic composition characteristics of the ring, thereby providing key structural information for the property prediction and model establishment of the polymer. The feature descriptors of this method include: the total number of high electronegativity, the number of fused rings, the number of six-membered rings, the number of five-membered rings, the number of carbonyl groups, the number of single bonds, the number of single bonds on the main chain, the number of hydrogen bonds, the sum of the molecular weight difference of the groups at both ends of the single bond on the main chain, and the sum of the molecular weight on the side chain. Among them, the high electronegativity species are O, N, F, Cl, and Br.

[0082] Benzene ring assignment method, by identifying and assigning the benzene ring structure in the polymer molecule, extracting its specific structural characteristics, and quantifying the number, position and connection mode of the benzene ring and other groups, providing important structural information for the property prediction and model establishment of the polymer. The characteristic descriptors of this method include: the total number of high electronegativity, the number of five-membered rings, the number of non-five-membered rings, the number of carbonyl groups, the number of single bonds, the number of single bonds in the main chain, the number of hydrogen bonds, the sum of the molecular weight difference of the groups at both ends of the single bond in the main chain, and the sum of the molecular weight of the branch. Among them, the high electronegativity species are O, N, F, Cl, and Br.

[0083] Atomic normalization plus condensed ring / benzene ring assignment method, atomic normalization standardizes the characteristics of each atom in the polymer molecule, eliminates the influence of different atomic types or molecular structure size differences on the characteristic value, ensures the comparability of the contribution of each atom in the model, and improves the stability and prediction accuracy of the model. The characteristic descriptors of this method include: the total number of high electronegativity, the number of condensed rings, the number of five-membered rings, the number of non-five-membered rings, the number of carbonyl groups, the number of single bonds, the number of single bonds in the main chain, the number of hydrogen bonds, the sum of the molecular weight difference of the groups at both ends of the single bond in the main chain, and the sum of the molecular weight of the branch. Among them, the high electronegativity species are O, N, F, Cl, and Br.

[0084] Finally, all the extracted characteristic values are normalized and then expanded by 100 times, thereby providing consistent, comparable and amplified structural information for subsequent polymer property prediction and model establishment.

[0085] Based on the selection of characteristic values and the corresponding data processing, the data affected by the main chain and the side chain are shown in Table 1.

[0086] Table 1

[0087]

[0088]

[0089]

[0090] Further, the condensed ring assignment method uses the following formula for calculation:

[0091] For the number of C on the ring = 6, the corresponding assignment is a;

[0092] For the number of C on the ring < 6, the corresponding assignment is 0.5a;

[0093] For the number of C on the ring > 6, the corresponding assignment is 2a; where the value of a is 5.

[0094] In the atomic normalization plus condensed ring assignment method, normalization is calculated using the following formula:

[0095]

[0096] In the benzene ring valuation method, several different algorithms are introduced, including only assigning values to rings, hydrogen bonds to main chain single bonds, main chain single bonds without hydrogen bonds, single bonds with hydrogen bonds, and single bonds without hydrogen bonds. The following equation is used for calculation:

[0097]

[0098] where N i is the number of atoms in the repeating unit, ζ i is the effective mobility of each atom in the repeating unit. For single bonds, ζ i is 0.2a, for hydrogen bonds, ζ i is 0.4a, for single bonds, ζ i is 0.2a, for high electronegativity, ζ i is 0.6a, for carbonyl, ζ i is 0.7a, for five-membered rings, ζ i is 0.5a, for non-five-membered rings, ζ i is a + w a (n-1), where n is the number of rings, a is 5, and w a is 1.5.

[0099] The results show that the main factor affecting the secondary transition temperature (T β ) of the main chain is the ease of rotation of the single bond at both ends. This ease of rotation can be quantified by several key structural features, including high electronegative atoms, C=O, hydrogen bonds, molecular weight at both ends of the single bond, branched molecular weight, and other factors. Specifically, high electronegative atoms will attract the electron cloud in the covalent bond, causing uneven distribution of electron density at both ends of the single bond. Due to this asymmetric distribution of electron cloud, the polarity of the single bond is enhanced, resulting in stronger electrostatic interactions between molecules. These interactions increase the energy barrier of single bond rotation, making it more difficult for the single bond to rotate, thereby limiting the flexibility and rotational freedom of the chain segment.

[0100] The C=O group itself has strong polarity, making the single bond connected to it also exhibit strong polarity difference at both ends. This polarity difference increases the energy barrier of the single bond, making it more difficult for the single bond to rotate, thereby further limiting the rotational freedom of the single bond.

[0101] Hydrogen bonds increase the rigidity of the chain segment through intermolecular forces, strengthening the stability of the molecular structure and reducing the rotational freedom of the chain segment, thereby lowering the secondary transition temperature (T β) will increase. The molecular weight of the groups on both sides of the single bond is also a key factor affecting the ease of rotation. Larger groups usually have stronger intermolecular interactions, increasing the rigidity of the molecule, thereby making the single bond rotation of the main chain face greater resistance, thereby increasing the energy required to overcome the rotation barrier.

[0102] The distribution of branched chains also has a significant impact on the secondary transition temperature (T β ) of the polymer. The length, rigidity, distribution density and position of the branched chains all affect the flexibility and rigidity of the polymer by changing the relative mobility of the polymer segments and the intermolecular interaction force, thereby affecting the secondary transition temperature (T β ).

[0103] Table 2 is an analysis of the relationship between the side chain descriptor and R 2 .

[0104] Table 2

[0105]

[0106] The results show that: high electronegative atoms enhance the intermolecular interaction force through strong electronic effect, thereby limiting the movement of the chain segment, resulting in the increase of the secondary transition temperature (T β ). The carbonyl group on the main chain will reduce the rotation freedom of the side chain through the steric hindrance effect, so that the side chain cannot rotate or move freely as in the case without carbonyl group. In addition, the carbonyl group on the side chain further reduces the rotation freedom of the side chain due to its strong polarity, increasing the rigidity of the polymer.

[0107] The distribution of branched chains has a direct relationship with the secondary transition temperature. When there are multiple single bonds in the branched chain, the free rotation of these single bonds will increase the rotation freedom of the side chain, so that the polymer chain segment can start to transform or deform at a lower temperature. However, if the branched chain structure contains double bonds or ring structures, the rotation freedom of the single bond will be more restricted. Double bonds or ring structures will increase the rigidity of the polymer chain, inhibit its rotation freedom, and thus increase the secondary transition temperature (T β ).

[0108] Further, training the secondary transition temperature prediction model comprises:

[0109] Integrating the key structural features with the experimentally measured secondary transition temperature data to form an input data set and an output data set for model training, wherein the input data is the molecular features extracted by the molecular fragment method, and the output data is the secondary transition temperature value predicted by the model;

[0110] Selecting a machine learning algorithm, modeling according to the characteristics of the data, training the secondary transition temperature prediction model through the training set, and testing through the test set, and evaluating the prediction accuracy of the model, the training set and the test set are the feature values extracted by different methods and the corresponding secondary transition temperature data.

[0111] Specifically, in this embodiment, 80% of the preprocessed data is randomly selected as the training set, and 20% is selected as the test set.

[0112] Integrating the extracted molecular features with the experimentally measured secondary transition temperature (T β ) data to form an input-output data set for model training, where the input data is the molecular features extracted by the feature extraction technique, and the output data is the secondary transition temperature (T β ) value predicted by the model.

[0113] Selecting appropriate machine learning algorithms, such as regression analysis methods, decision trees, random forests, gradient boosting, etc., and selecting appropriate algorithms for modeling according to the characteristics of the data.

[0114] Through training, a prediction model is obtained, which can predict the secondary transition temperature (T β ) of the polymer according to the input molecular features, and can be used for predicting the secondary transition temperature (T β ) value of unknown samples.

[0115] Verify the model, verify the model prediction results according to the experimental data, evaluate the prediction accuracy of the model, and analyze the main features of the model to ensure the reliability and accuracy of the model.

[0116] The evaluation parameters used for testing the secondary transition temperature prediction model are:

[0117]

[0118] In the formula, R 2 is the determination coefficient; y i is the true value, is the predicted value, is the average value of the true value, n is the total number of samples, and i is the sample index value.

[0119] Further, according to the model prediction results, analyze and explain the rotational mechanism of the secondary transition temperature, including:

[0120] Analyze the relationship between the key features in the model and the secondary transition temperature (T β ), and identify the key features that affect the secondary transition temperature (T β) have significant influence on the glass transition temperature, such as the rotation freedom of single bond, the flexibility of molecular segment, the polarity of covalent bond, the size of ring and the influence of highly electronegative atom, etc.

[0121] In combination with the rotation mechanism of the secondary transition temperature, the model analysis results are compared with the existing literature on the secondary transition temperature (T β ) of the rotation mechanism, and how each structural feature regulates the secondary transition temperature (T β ) by affecting molecular motion, segment flexibility and intermolecular interaction is explained and verified.

[0122] According to the model analysis results, the micro factors affecting the secondary transition temperature (T β ) are further revealed, such as the mechanism of influence of main chain and side chain, and the specific influence path of intramolecular interaction on the secondary transition temperature (T β ), thereby providing theoretical guidance for optimizing the secondary transition temperature (T β ) of the polymer material.

[0123] The method of the present application analyzes the rotation nature of the secondary transition temperature (T β ) by using different method descriptors, and the prediction method developed by the present application can be applied to the design and optimization of polymer materials, especially in the case of data scarcity, the secondary transition temperature (T β ) of the polymer can be effectively predicted, and theoretical support is provided for the performance regulation of the material. At the same time, the method can also be widely used in the development of new polymer materials, the screening of functional polymers and the performance evaluation in engineering applications.

[0124] In view of the scarcity of research data on the secondary transition temperature (T β ) of the polymer, by combining regression model and various feature extraction techniques, a diversified solution is provided for the study of small amount of data, effectively overcoming the problem of insufficient data in traditional experimental methods. Through this innovative analysis framework, the present application not only improves the research efficiency, but also provides a new idea and theoretical support for the in-depth study of the secondary transition temperature (T β ). The present application opens up a new research path for exploring the behavior rule and application potential of the secondary transition temperature (T β ) in polymer materials, and promotes the technological innovation in the field of material science.

[0125] The above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for predicting secondary transition temperatures of polymer materials based on machine learning, characterized in that, The method comprises the following steps: acquiring a polymer material to be detected; constructing a secondary transition temperature prediction model; inputting the polymer material to be detected into the secondary transition temperature prediction model for prediction to obtain a prediction result; wherein the secondary transition temperature prediction model is obtained through training of a training set and testing of a test set; constructing the secondary transition temperature prediction model comprises: collecting a chemical structural formula of a target polymer and corresponding secondary transition temperature data, and dividing the secondary transition temperature data to obtain divided data; converting the chemical structural formula of the target polymer into a SMILES string, and parsing the SMILES string to extract key structural features in the target polymer molecule; establishing the secondary transition temperature prediction model based on the key structural features and the divided data; dividing the secondary transition temperature data comprises: dividing the secondary transition temperature data into data affected by a main chain and data affected by a side chain through a rotation mechanism of the secondary transition temperature; converting the chemical structural formula of the target polymer into a SMILES string comprises: drawing a molecular structure diagram of the target polymer through CHEMDRAW software to graphically represent a two-dimensional structure of the target polymer; converting the molecular structure diagram into a simplified molecular linear representation, i.e. the SMILES string, through CHEMDRAW software, and performing quality control on the SMILES string; parsing the SMILES string to extract key structural features in the polymer molecule comprises: decomposing the given SMILES string into basic chemical structural units through a molecular fragment method, extracting heteroatoms, types of chemical bonds and cyclic structure features in the basic chemical structural units, and calculating physical and chemical features of different structural units, wherein the physical and chemical features include but are not limited to rotational freedom, polarity and flexibility; standardizing the heteroatoms, types of chemical bonds and cyclic structure features, and performing data cleaning and outlier removal; performing data conversion and processing on the extracted features according to different processing methods to generate feature sets corresponding to each method, and integrating different feature sets into comprehensive features of the target polymer, i.e. key structural features in the target polymer molecule; the key structural features in the target polymer molecule comprise: chemical descriptors: total number of high electronegativity, number of condensed rings, number of six-membered rings, number of five-membered rings, number of carbonyl groups, total number of single bonds, number of single bonds on the main chain, and number of hydrogen bonds; based on the molecular weight of the two ends of a single bond: sum of the molecular weight difference of side chain groups at the two ends of a single bond, sum of the molecular weight of side chain groups at the two ends of a single bond, sum of the molecular weight difference of side chain groups at the two ends of a single bond on the main chain, and sum of the molecular weight of side chain groups on the side chain.

2. The machine learning based polymer material secondary transition temperature prediction method of claim 1, wherein, the different processing methods comprise a molecular fragment method, a condensed ring assignment method, a benzene ring assignment method, and an atomic normalization plus condensed ring / benzene ring assignment method.

3. The machine learning based polymer material secondary transition temperature prediction method of claim 1, wherein, training the secondary transition temperature prediction model comprises: Integrating the key structural features with the experimentally measured secondary transition temperature data to form an input data set and an output data set for model training, wherein the input data is the molecular features extracted by the molecular fragment method, and the output data is the secondary transition temperature value predicted by the model; Selecting a machine learning algorithm to model according to the characteristics of the data, training the secondary transition temperature prediction model through the training set, testing through the test set, and evaluating the prediction accuracy of the model.

4. The machine learning based polymer material secondary transition temperature prediction method of claim 3, wherein, The evaluation parameters used for testing the secondary transition temperature prediction model are: where R 2 is the coefficient of determination; y i is the true value, is the predicted value, is the mean of the true values, n is the total number of samples, and i is the sample index value.

Images