Method for determining copolymerization reactivity rate based on artificial neural network, reactivity rate database and application

Abstract

The invention provides a method for determining a copolymerization reactivity rate based on an artificial neural network, a reactivity rate database and application, a collected monomer structure is converted into a molecular fingerprint as a molecular feature, the molecular fingerprint is combined with reactivity rate data r1 and r2 of a corresponding monomer pair, the molecular fingerprint is used as an input feature, and the reactivity rate data r1 and r2 of the monomer pair are obtained. According to the method, the reactance rates r1 and r2 are directly and accurately predicted by using the artificial neural network model, so that the sequence structure of a macromolecular chain is better designed, regulated and controlled, the performance of a copolymer is optimized, the method has a good application prospect in the development of a macromolecular material, the design precision of the copolymer can be improved, the synthesis and processing processes can be guided, and the development period of a new material can be accelerated.

Description

Technical Field

[0001] The invention belongs to the technical field of polymer design, and in particular relates to a method for determining copolymerization reactivity ratio based on an artificial neural network, a reactivity ratio database and an application thereof. Background Art

[0002] Copolymerization refers to the interaction of two or more different monomers to form copolymer chains with random, alternating or blocky distribution. The reactivity ratio is a quantitative indicator to measure the reaction tendency of these monomers, which is defined as the ratio of the rate at which one monomer reacts with itself to the rate at which it reacts with another monomer. The reactivity ratio not only affects the microstructure of the copolymer molecular chain, but also determines the macroscopic properties of the final material. Therefore, accurate understanding and prediction of the reactivity ratio is crucial for the design and optimization of copolymers with specific properties. For example, in the process of preparing polyacrylonitrile (PAN) precursor, in order to improve the spinnability of its solution and control the concentrated heat release in the pre-oxidation stage, copolymer components such as itaconic acid and acrylic acid derivatives are usually introduced. The addition of these components not only improves the processing properties of the precursor, but also helps to regulate the thermal stability and mechanical properties of the copolymer.

[0003] At present, the most common method for calculating the reactivity ratio is to use the Alfrey-Price Qe formula, which correlates the reaction rate constant between free radicals and monomers with the conjugation effect and polarity effect. In the Qe formula, Q1 and Q2 represent the conjugation effect measure of the activity of monomer 1 and monomer 2, respectively, and e1 and e2 represent the polarity of monomer 1 and its free radical, monomer 2 and its free radical, respectively, and the reactivity ratios r1 and r2 can be obtained:

[0004] Summary of the invention

[0005] The first object of the present invention is to provide a method for determining copolymerization reactivity ratio based on artificial neural network.

[0006] To this end, the above-mentioned purpose of the present invention is achieved through the following technical solutions:

[0007] A method for determining copolymerization reactivity ratio based on artificial neural network comprises the following steps:

[0008] S1, collecting a large amount of copolymerization reaction data, wherein the copolymerization reaction data includes monomer structure and known Qe value, and further constructing a large amount of reactivity ratio data r1 and r2 of monomer pairs using Alfrey-Price Qe formula;

[0009] S2, filtering the obtained reactivity ratio data r1 and r2 of the monomer pair, and retaining the data within a reasonable range;

[0010] S3, converting the reactivity rate data r1 and r2 of the monomer pair into natural logarithmic form, converting the collected monomer structures into molecular fingerprints as molecular features, and combining the molecular fingerprints with the reactivity rate data r1 and r2 of the corresponding monomer pair to create a comprehensive database for testing and training the artificial neural network learning model;

[0011] S4, randomly divide the data in the comprehensive database into training set and test set;

[0012] Optimize the hyperparameters of artificial neural network learning models;

[0013] S5. Use the optimized hyperparameter combination to create a new artificial neural network learning model, and use the data in the training set to train the artificial neural network learning model to obtain the optimal model parameters;

[0014] The new artificial neural network learning model includes an input layer, multiple hidden layers and an output layer; the molecular fingerprint is input into the input layer, and the competitive aggregation rate is output in the output layer;

[0015] S6. Use the data in the test set to evaluate the predictive performance of the trained artificial neural network learning model;

[0016] S7. According to the prediction performance of the evaluated artificial neural network learning model, iteratively optimize the artificial neural network learning model until the artificial neural network learning model meets the accuracy requirements.

[0017] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions:

[0018] As a preferred technical solution of the present invention: in step S2, the specific process of filtering includes: removing the reactivity ratio data r1 and r2 less than 0.001 and greater than 100.

[0019] As a preferred technical solution of the present invention: in step S4, 90% of the data is used as a training set, and 10% of the data is used as a test set.

[0020] As a preferred technical solution of the present invention: in step S4, the hyperparameters include optimizer, number of neurons, batch size and number of training cycles.

[0021] As a preferred technical solution of the present invention: in step S6, the performance of the trained artificial neural network learning model is evaluated by using the loss function and the determination coefficient index.

[0022] The second object of the present invention is to provide a reactivity ratio database obtained by the method for determining the reactivity ratio of copolymerization reaction based on artificial neural network as described above.

[0023] Another object of the present invention is to provide an application of the reactivity ratio database obtained by the method for determining the reactivity ratio of copolymerization reaction based on artificial neural network as described above in the development of polymer materials.

[0024] The present invention provides a method for determining the copolymerization reactivity ratio based on an artificial neural network, a reactivity ratio database and an application thereof. The collected monomer structures are converted into molecular fingerprints as molecular features, and the molecular fingerprints are combined with reactivity ratio data r1 and r2 of corresponding monomer pairs. The molecular fingerprints are used as input features, and an artificial neural network model is used to directly and accurately predict the reactivity ratios r1 and r2, so as to better design and regulate the sequence structure of the polymer chain, optimize the performance of the copolymer, have good application prospects in the development of polymer materials, can not only improve the accuracy of copolymer design, guide the synthesis and processing processes, but also accelerate the development cycle of new materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The present invention provides a flowchart of a method for determining copolymerization reactivity ratio based on an artificial neural network.

[0026] Figure 2 is the loss function loss and the determination coefficient R 2 Plot of changes during training.

[0027] Figure 3 This is a comparison chart of the predicted competitive aggregation rate and the actual competitive aggregation rate of the training set.

[0028] Figure 4 This is a comparison chart of the predicted competitive aggregation rate and the actual competitive aggregation rate of the test set. DETAILED DESCRIPTION

[0029] The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.

[0030] A method for determining copolymerization reactivity ratio based on artificial neural network comprises the following steps:

[0031] S1, collecting a large amount of copolymerization reaction data, wherein the copolymerization reaction data includes monomer structure and known Qe value, and further constructing a large amount of reactivity ratio data r1 and r2 of monomer pairs using Alfrey-Price Qe formula;

[0032] S2, filtering the obtained reactivity ratio data r1 and r2 of the monomer pair, and retaining the data within a reasonable range;

[0033] S3, converting the reactivity rate data r1 and r2 of the monomer pair into natural logarithmic form, converting the collected monomer structures into molecular fingerprints as molecular features, and combining the molecular fingerprints with the reactivity rate data r1 and r2 of the corresponding monomer pair to create a comprehensive database for testing and training the artificial neural network learning model;

[0034] S4, randomly divide the data in the comprehensive database into training set and test set;

[0035] Optimize the hyperparameters of artificial neural network learning models;

[0036] S5. Use the optimized hyperparameter combination to create a new artificial neural network learning model, and use the data in the training set to train the artificial neural network learning model to obtain the optimal model parameters;

[0037] The new artificial neural network learning model includes an input layer, multiple hidden layers and an output layer; the molecular fingerprint is input into the input layer, and the competitive aggregation rate is output in the output layer;

[0038] S6. Use the data in the test set to evaluate the predictive performance of the trained artificial neural network learning model;

[0039] S7. According to the prediction performance of the artificial neural network learning model after evaluation, iteratively optimize the artificial neural network learning model until the artificial neural network learning model meets the accuracy requirements. The iterative optimization methods used include stochastic gradient descent method, Adam and its extension method, and root mean square propagation method.

[0040] In step S2, the specific process of filtering includes: removing the reactivity ratio data r1 and r2 less than 0.001 and greater than 100.

[0041] In step S4, 90% of the data is used as a training set and 10% of the data is used as a test set.

[0042] In step S4, the hyperparameters include optimizer, number of neurons, batch size and number of training cycles.

[0043] In step S6, the performance of the trained artificial neural network learning model is evaluated by using the loss function and the determination coefficient index.

[0044] The present invention also provides a reactivity ratio database obtained by the method for determining the copolymerization reactivity ratio based on an artificial neural network as described above.

[0045] The present invention also provides the application of the reactivity ratio database obtained by the method for determining the reactivity ratio of copolymerization reaction based on artificial neural network as described above in the development of polymer materials. In the actual development of polymer materials, it can not only improve the accuracy of copolymer design, guide the synthesis and processing process, but also accelerate the development cycle of new materials.

[0046] 1. Establishment and application of artificial neural network model

[0047] The process of establishing the artificial neural network model is as follows Figure 1 As shown, the process mainly includes three parts: (1) construction of learning sample database, (2) establishment, learning, optimization and evaluation of learning model, and (3) application of learning model.

[0048] (1) Construction of learning sample database: First, a large amount of copolymerization reaction data, including monomer structures and known Qe values, were collected from published literature, polymer handbooks, and polymer textbooks. The Alfrey-Price Qe formula was used to construct a large number of monomer pair reactivity data. Then, the data was filtered and preprocessed according to certain rules to retain only data within a reasonable range and delete abnormally large or abnormally small data. In addition, the reactivity ratio r1 and r2 data pairs were converted to natural logarithm form. The natural logarithm transformation can help process skewed distributed data and make it closer to the normal distribution, which is more suitable for model training. Here, we collected and generated a total of about 27,000 r1 and r2 data pairs, and about 15,000 data remained after preprocessing. After that, the collected monomer structures were converted into molecular fingerprints as molecular features. These fingerprints can accurately describe the chemical properties and structural information of the monomers. Finally, the molecular fingerprints and the corresponding reactivity ratios were combined to create a comprehensive database for training and testing artificial neural network learning models.

[0049] (2) Learning model establishment, learning, optimization and evaluation: First, the data is randomly divided into two parts: training set and test set. The training set is used to train the learning model, while the test set is used to evaluate the performance of the model. Here, we use 90% of the data as the training set and the remaining 10% of the data as the test set. This division ensures that the model has enough data for training, while also leaving out a portion of the data to independently evaluate the performance of the model. Then, the grid search method is used to optimize the hyperparameters. The optimized hyperparameters include optimizer, number of neurons, batch size and number of training cycles. These parameters are all key factors affecting the performance of the model. Through grid search, the optimal hyperparameter combination can be found. After that, a new model is created using the best hyperparameter combination and trained using the training data to obtain the optimal model parameters. Finally, the prediction performance of the model is evaluated using the test set data to ensure accuracy and reliability. The trained model parameters are saved so that the trained model can be used for prediction. In the architecture of our artificial neural network learning model, there is an input layer (molecular fingerprint), multiple hidden layers and an output layer (competitive rate).

[0050] (3) Application of learning model: The trained artificial neural network learning model is used for the newly designed molecular structure to predict its reactivity ratio. The design of the copolymer can be adjusted according to the prediction results to achieve the desired copolymer performance. In addition, the reactivity ratio measured experimentally can be used for comparative analysis and model verification to continuously improve the accuracy and reliability of the model.

[0051] Through these three steps, an accurate and reliable prediction system can be established. In the actual development of polymer materials, it can not only improve the accuracy of copolymer design, guide the synthesis and processing processes, but also accelerate the development cycle of new materials.

[0052] 2. Verification of the reactivity ratio calculation model

[0053] In the process of learning model training, the loss function (loss) and the coefficient of determination (R 2 ) indicators to evaluate the performance of the model. These indicators are calculated on both the training set and the test set to fully understand the learning effect and prediction ability of the model. Calculating these indicators on the training set can evaluate the learning ability of the model; while calculating them on the test set reflects the generalization ability of the model on unseen data.

[0054] like Figure 2As shown in the figure, after 100 Epochs of training, the loss function tends to be unchanged, and the loss functions of the training set and the test set are 0.072 and 0.085 respectively, indicating that the model has good performance on both data sets and small errors. The coefficient of determination is close to 1, and the coefficients of determination of the training set and the test set are 0.990 and 0.986 respectively, indicating that the model can explain the variance of the target variable very well, and the predicted value is very close to the actual value. At this point, the model training is completed. The mean absolute error (MAE) of the training set and the test set is small, 0.08 and 0.09 respectively, further confirming the high accuracy of the model.

[0055] 450 and 50 data sets were randomly selected from the training set and the test set, respectively, and the reactivity ratio predicted by the model was compared with the actual reactivity ratio. The results are as follows: Figure 3 , 4 As shown in the figure, it can be seen that the difference between the competitive aggregation rate predicted by the model and the actual competitive aggregation rate in the training set and the test set is very small, which shows that the model has high accuracy and reliability. All the determination coefficients are close to 1, indicating that the model can capture the changes in the data well.

[0056] In addition, further, a detailed comparison between the predicted values ​​and experimental values ​​of the reactivity ratios of some systems is given as an example, as shown in Table 1. The results show that the reactivity ratios predicted by the model are very close to the experimental values. Although the relative errors of individual systems are slightly larger, the overall error range is small, and the relative error is usually within 5%. This shows that the new artificial neural network learning model has a high prediction accuracy when dealing with both known and unknown monomer combinations. Through this comparison, the effectiveness of the artificial neural network model is further verified.

[0057] Table 1 Comparison of predicted and experimental values ​​of reactivity ratios of some systems

[0058]

[0059] The above-mentioned specific implementation methods are used to explain the present invention and are only preferred embodiments of the present invention, rather than limiting the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A method for determining copolymerization reactivity ratio based on artificial neural network, characterized in that: The method comprises the following steps: S1, collecting a large amount of copolymerization reaction data, wherein the copolymerization reaction data includes monomer structure and known Qe value, and further constructing a large amount of reactivity ratio data r1 and r2 of monomer pairs using Alfrey-Price Qe formula; S2, filtering the obtained reactivity ratio data r1 and r2 of the monomer pair, and retaining the data within a reasonable range; S3, converting the reactivity rate data r1 and r2 of the monomer pair into natural logarithmic form, converting the collected monomer structures into molecular fingerprints as molecular features, and combining the molecular fingerprints with the reactivity rate data r1 and r2 of the corresponding monomer pair to create a comprehensive database for testing and training the artificial neural network learning model; S4, randomly divide the data in the comprehensive database into training set and test set; Optimize the hyperparameters of artificial neural network learning models; S5. Use the optimized hyperparameter combination to create a new artificial neural network learning model, and use the data in the training set to train the artificial neural network learning model to obtain the optimal model parameters; The new artificial neural network learning model includes an input layer, multiple hidden layers and an output layer; the molecular fingerprint is input into the input layer, and the competitive aggregation rate is output in the output layer; S6. Use the data in the test set to evaluate the predictive performance of the trained artificial neural network learning model; S7. According to the prediction performance of the evaluated artificial neural network learning model, iteratively optimize the artificial neural network learning model until the artificial neural network learning model meets the accuracy requirements.

2. The method for determining copolymerization reactivity ratio based on artificial neural network according to claim 1, characterized in that: In step S2, the specific process of filtering includes: removing the reactivity ratio data r1 and r2 less than 0.001 and greater than 100.

3. The method for determining copolymerization reactivity ratio based on artificial neural network according to claim 1, characterized in that: In step S4, 90% of the data is used as a training set and 10% of the data is used as a test set.

4. The method for determining copolymerization reactivity ratio based on artificial neural network according to claim 1, characterized in that: In step S4, the hyperparameters include optimizer, number of neurons, batch size and number of training cycles.

5. The method for determining copolymerization reactivity ratio based on artificial neural network according to claim 1, characterized in that: In step S6, the performance of the trained artificial neural network learning model is evaluated by using the loss function and the determination coefficient index.

6. A reactivity ratio database obtained by the method for determining copolymerization reactivity ratio based on artificial neural network according to any one of claims 1 to 5.

7. Application of the reactivity ratio database obtained by the method for determining the reactivity ratio of copolymerization reaction based on artificial neural network according to claim 6 in the development of polymer materials.

Images