Polyimide material glass transition temperature prediction method based on machine learning

Abstract

The invention relates to the field of research, development and design of artificial intelligence materials, and provides a polyimide material glass transition temperature prediction method based on machine learning, and the method comprises the steps: collecting experimental data including the structure and glass transition temperature of a polyimide polymer; preprocessing the obtained data, and then dividing a training set and a test set based on the feature data and the label data; constructing a prediction model based on gradient lifting, and continuously adjusting parameters until optimal parameters are found; evaluating the training model by using evaluation parameters such as a decision coefficient and a mean square error, and finally confirming a prediction model; and inputting related parameters to the model based on the trained prediction model, and outputting the glass transition temperature of the corresponding polyimide. By adopting the gradient lifting model, the scientific research cost is reduced, the scientific research efficiency is improved, and a new path is opened up for the development of the high glass transition temperature polymer.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence material research and development design technology, and in particular to a method for predicting the glass transition temperature of a polyimide material based on machine learning. Background Art

[0002] Polyimide (PI) is a polymer at the top of the polymer pyramid. Due to its outstanding comprehensive performance, good heat resistance and strong mechanical properties, it is widely used in aerospace, mechanical chemical industry, electronics, photoresist, fiber and national defense military industry. Glass transition temperature (T g ) is one of the most important characteristics of polyimide's good performance and wide application, and most application scenarios require polyimide to have high thermal stability. Because the unique structure of polyimide gives it a high glass transition temperature, it can be used in many high-end fields. In particular, polyimide has received more and more attention as a flexible substrate material in the display field.

[0003] Currently, traditional methods for studying polymer properties, such as trial and error, DFT and MD, rely on researchers' experimental experience and scientific intuition, and consume a lot of experimental resources and time. Therefore, it is difficult to explore the structures of all polyimides to find the optimal structure. Summary of the invention

[0004] In recent years, with the rapid development of machine learning and deep learning technology, researchers have used machine learning to predict unknown chemical structures and properties from a large amount of experimental data, reducing experimental costs and increasing experimental efficiency, so that more and more fields have used machine learning, such as the structure and properties of polymers. Therefore, the present invention provides a method for predicting the glass transition temperature of polyimide materials based on machine learning, so as to achieve the prediction of high polyimide structures, reduce scientific research costs, and improve scientific research efficiency.

[0005] The purpose of the present invention is to solve the existing technical problems, overcome the drawbacks of low efficiency and high cost in finding a polyimide structure with a high glass transition temperature, and provide a method for predicting the glass transition temperature of a polyimide material based on machine learning, the method comprising:

[0006] Step S1, collecting historical experimental data, wherein the historical experimental data is the SMILES structure and glass transition temperature of the polyimide polymer;

[0007] Step S2, preprocessing the historical experimental data to obtain preprocessed historical experimental data;

[0008] Step S3, constructing a transition temperature prediction model based on XGBoost, and using the preprocessed historical experimental data to train the transition temperature prediction model to obtain a trained transition temperature prediction model;

[0009] Step S4: Use the trained transition temperature prediction model to predict the glass transition temperature of the polyimide material for the test data to obtain a prediction result.

[0010] Optionally, in step S2, the preprocessing process specifically includes:

[0011] Cleaning the SMILES structure of the polyimide polymer to obtain first data without abnormal SMILES values;

[0012] Acquire a descriptor based on the SMILES value in the first data to obtain original descriptor data, and form an original descriptor data set based on the original descriptor data;

[0013] The original descriptor data set is normalized to obtain preprocessed historical experimental data.

[0014] Optionally, the SMILES value acquisition descriptor in the first data specifically includes:

[0015] Electronegativity, number of fused rings, total number of five- and six-membered rings, number of carbonyl groups, number of nitrogen atoms, number of single bonds, number of oxygen atoms, wherein the electronegativity is the total sum of F, Cl, and Br.

[0016] Optionally, normalizing the glass transition temperature to obtain standardized data specifically includes:

[0017]

[0018] Among them, x' is the standardized data, x is the original data of the descriptor, μ is the mean of the original data set of the descriptor, and σ is the standard deviation of the original data set of the descriptor.

[0019] Optionally, in step S3, the parameters of XGBoost specifically include n_estimators, random_state, learning_rate, max_depth, and subsample.

[0020] Optionally, in step S3, obtaining the trained transition temperature prediction model specifically includes:

[0021] Using the pre-processed historical experimental data to train the transition temperature prediction model to obtain an initially trained model;

[0022] The initially trained model is evaluated using evaluation parameters, and the initially trained model is optimized based on the evaluation results to obtain a trained transition temperature prediction model.

[0023] Optionally, the evaluation parameters are specifically:

[0024]

[0025] Among them, R 2 is the coefficient of determination, indicating the degree of fit of the model, ranging from [0,1]; 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;

[0026]

[0027] Among them, MAE is the mean absolute error, which represents the average value of the absolute error between the predicted value and the actual value, y i is the true value, is the predicted value.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] The method of the present invention adopts a descriptor method to introduce the influence of the polyimide structure on the performance. The prediction method developed by the present invention can be used to predict other properties of other polymers. Compared with the traditional polyimide testing method, the method of the present invention has the following advantages: the method has high accuracy, and the prediction of the polyimide glass transition temperature is achieved by disassembling the polyimide structure, training the gradient boosting model, and using data-driven and machine learning to achieve the reasonable design of the polyimide structure; the gradient boosting model used in the present invention reduces the cost of scientific research, improves the efficiency of scientific research, and opens up a new path for the development of high glass transition temperature polymers. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0031] Figure 1 A method step diagram of a method for predicting the glass transition temperature of a polyimide material based on machine learning according to an embodiment of the present invention;

[0032] Figure 2 is a data distribution diagram of the glass transition temperature of the polyimide of the embodiment of the present invention;

[0033] Figure 3 Schematic diagram of the accuracy of the glass transition temperature prediction model of polyimide according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0035] Example

[0036] The prediction method of glass transition temperature of polyimide materials based on machine learning, such as Figure 1 As shown, the method includes:

[0037] Step S1, collecting historical experimental data, wherein the historical experimental data is the SMILES structure and glass transition temperature of polyimide polymer. Collect relevant data from public literature libraries and books to construct a data set;

[0038] The data set includes the structure of polyimide, the SMILES structure of the polyimide monomer and the glass transition temperature of the polyimide polymer, and is stored in a one-to-one correspondence between SMILES and the glass transition temperature.

[0039] Step S2: preprocess the historical experimental data to obtain preprocessed historical experimental data.

[0040] The pretreatment process specifically includes: cleaning the SMILES structure of the polyimide polymer to obtain first data with abnormal SMILES values ​​removed; obtaining descriptors based on the SMILES values ​​in the first data; and normalizing the glass transition temperature to obtain standardized data.

[0041] The SMILES value acquisition descriptors in the first data specifically include: electronegativity, the number of fused rings, the total number of five- and six-membered rings, the number of carbonyl groups, the number of nitrogen, the number of single bonds, and the number of oxygen, wherein the electronegativity is the total sum of F, Cl, and Br.

[0042] Based on this embodiment, the RDKit toolkit is used to clean and standardize the SMILES data of polyimide. RDKit is an open source chemical informatics toolkit that provides functions such as molecular structure recognition and processing, and calculation of descriptors. SMILES is standardized by RDKit, and abnormal SMILES strings with duplicates and missing values ​​are deleted; the cleaned data set contains 367 glass transition temperature data of polyimide materials. The distribution of the data set is as follows Figure 2 As shown, Figure 2The data distribution diagram of the glass transition temperature of polyimide materials. Feature selection was performed on the data set. Specifically, SMILES was parsed by RDKit to obtain descriptors, and 7 descriptors were obtained as input data, including electronegativity (the total number of F, Cl, and Br), the number of fused rings, the total number of five- and six-membered rings, the number of carbonyl groups, the number of nitrogen, the number of single bonds, and the number of oxygen.

[0043] The glass transition temperature is normalized to obtain standardized data, specifically including:

[0044]

[0045] Among them, x' is the standardized data, x is the original data of the descriptor, μ is the mean of the original data set of the descriptor, and σ is the standard deviation of the original data set of the descriptor.

[0046] 80% of the preprocessed data is randomly selected as the training set and 20% as the test set.

[0047] Step S3, constructing a transition temperature prediction model based on XGBoost, and using the training set in the preprocessed historical experimental data to train the transition temperature prediction model to obtain a trained transition temperature prediction model.

[0048] In step S3, the parameters of XGBoost specifically include n_estimators, random_state, learning_rate, max_depth, and subsample.

[0049] n_estimators is the number of trees, which is also the number of iterations. Increasing this parameter can improve the model's fitting ability;

[0050] random_state is a random seed, which is used to ensure the repeatability of model training. Different random seeds will lead to different results.

[0051] learning_rate is the learning rate. A low learning rate can make the model stable, while a high learning rate can easily lead to overfitting.

[0052] max_depth is the maximum depth of the tree. A deeper value is helpful to capture more feature relationships and improve the fit of the model.

[0053] Subsample is the sample ratio of each tree. A higher sample ratio can produce a better model.

[0054] In step S3, obtaining the trained transition temperature prediction model specifically includes:

[0055] Using the pre-processed historical experimental data to train the transition temperature prediction model to obtain an initially trained model;

[0056] The initially trained model is evaluated using evaluation parameters, and the initially trained model is optimized based on the evaluation results to obtain a trained transition temperature prediction model.

[0057] The evaluation parameters are as follows:

[0058]

[0059] Among them, R 2 is the determination coefficient, which indicates the degree of fit of the model. Its range is [0,1]. The closer the value of the determination coefficient is to 1, the better the degree of fit of the model. i is the true value, is the predicted value, is the average value of the true value, n is the total number of samples, used to travel through all samples, and i is the sample index value;

[0060]

[0061] Among them, MAE is the mean absolute error, which represents the average value of the absolute error between the predicted value and the actual value. The smaller the MSE value, the better. i is the true value, is the predicted value.

[0062] In this embodiment, a prediction model based on the XGBoost algorithm is constructed based on the above-mentioned cleaned data set and descriptors. XGBoost is an efficient and flexible gradient boosting decision tree algorithm, which is widely used in regression tasks and integrates multiple decision trees for prediction. The main idea is to gradually train multiple weak learners, each time reducing the residual of the previous step. Figure 1 The workflow diagram for constructing the polyimide glass transition temperature prediction model for this embodiment includes:

[0063] (1) Divide the dataset into training set and test set in a ratio of 8:2;

[0064] (2) Add polynomial features (degree = 2) to enhance the model’s capabilities;

[0065] (3) Using the XGBoost model architecture, the model hyperparameters were optimized by random search, including n_estimators, learning_rate, max_depth, subsample, and colsample_bytree. The range of random search was 100-300 for n_estimators, 0.1-1 for learning_rate, 6-10 for max_depth, 0.5-1 for subsample, and 0.5-1 for colsample_bytree. g The prediction model has hyperparameters of n_estimators = 200, learning_rate = 0.5, max_depth = 6, subsample = 0.8, and colsample_bytree = 0.8.

[0066] (4) Using the coefficient of determination (R 2 ), mean absolute error (MAE) was used to evaluate the accuracy of the machine learning, and the polyimide T g High-precision prediction model. Figure 3 is the polyimide material T in this embodiment g The accuracy of the prediction model. Figure 3 As shown, the coefficient of determination (R 2 ) is 0.8 and the mean absolute error (MAE) is 22.8.

[0067] Step S4: Use the trained transition temperature prediction model to predict the glass transition temperature of the polyimide material for the test data to obtain a prediction result.

[0068] The collected SMILES of polyimide monomers are input into the trained transition temperature prediction model to obtain the prediction result of the glass transition temperature of the polyimide material.

[0069] The method of the present invention adopts a descriptor method to introduce the influence of the polyimide structure on the performance. The prediction method developed by the present invention can be used to predict other properties of other polymers.

[0070] Compared with the traditional polyimide testing method, the method of the present invention has the following advantages: the method has high accuracy, and the polyimide glass transition temperature is predicted by disassembling the polyimide structure, training the gradient boosting model, and utilizing data-driven and machine learning, so as to reasonably design the polyimide structure; the gradient boosting model adopted in the present invention reduces the scientific research cost, improves the efficiency of scientific research, and opens up a new path for the development of high glass transition temperature polymers.

[0071] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for predicting the glass transition temperature of polyimide materials based on machine learning, characterized in that: Methods include: Step S1, collecting historical experimental data, wherein the historical experimental data is the SMILES structure and glass transition temperature of the polyimide polymer; Step S2, preprocessing the historical experimental data to obtain preprocessed historical experimental data; Step S3, constructing a transition temperature prediction model based on XGBoost, and using the preprocessed historical experimental data to train the transition temperature prediction model to obtain a trained transition temperature prediction model; Step S4: Use the trained transition temperature prediction model to predict the glass transition temperature of the polyimide material for the test data to obtain a prediction result.

2. The method for predicting the glass transition temperature of polyimide materials based on machine learning according to claim 1, characterized in that: In step S2, the pre-processing process specifically includes: Cleaning the SMILES structure of the polyimide polymer to obtain first data without abnormal SMILES values; Acquire a descriptor based on the SMILES value in the first data to obtain original descriptor data, and form an original descriptor data set based on the original descriptor data; The original descriptor data set is normalized to obtain preprocessed historical experimental data.

3. The method for predicting the glass transition temperature of polyimide materials based on machine learning according to claim 2, characterized in that: The SMILES value acquisition descriptor in the first data specifically includes: Electronegativity, number of fused rings, total number of five- and six-membered rings, number of carbonyl groups, number of nitrogen atoms, number of single bonds, number of oxygen atoms, wherein the electronegativity is the total sum of F, Cl, and Br.

4. The method for predicting the glass transition temperature of polyimide materials based on machine learning according to claim 2, characterized in that: The glass transition temperature is normalized to obtain standardized data, specifically including: Among them, x' is the standardized data, x is the original data of the descriptor, μ is the mean of the original data set of the descriptor, and σ is the standard deviation of the original data set of the descriptor.

5. The method for predicting the glass transition temperature of polyimide materials based on machine learning according to claim 1, characterized in that: In step S3, the parameters of XGBoost specifically include n_estimators, random_state, learning_rate, max_depth, and subsample.

6. The method for predicting the glass transition temperature of polyimide materials based on machine learning according to claim 1, characterized in that: In step S3, obtaining the trained transition temperature prediction model specifically includes: Using the pre-processed historical experimental data to train the transition temperature prediction model to obtain an initially trained model; The initially trained model is evaluated using evaluation parameters, and the initially trained model is optimized based on the evaluation results to obtain a trained transition temperature prediction model.

7. The method for predicting the glass transition temperature of polyimide materials based on machine learning according to claim 6, characterized in that: The evaluation parameters are specifically: Among them, R 2 is the coefficient of determination, indicating the degree of fit of the model, ranging from [0,1]; 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; Among them, MAE is the mean absolute error, which represents the average value of the absolute error between the predicted value and the actual value, y i is the true value, is the predicted value.

Images