A neural network assisted method for efficiently determining monomer reactivity ratios in copolymerization reactions

Abstract

The application belongs to the technical field of polymer synthesis, and particularly relates to a neural network assisted method for efficiently determining monomer reactivity ratio in copolymerization. The neural network assisted determination scheme provided by the application integrates the copolymer monomer consumption corresponding to each group of reactivity ratio values into a high-dimensional data matrix that can be processed by computer language, and describes the change relationship between the copolymer monomer conversion rate and the monomer feeding ratio and total conversion rate. By establishing a neural network model between the polymer conversion rate data matrix and the reactivity ratio, reliable reactivity ratio determination results can be obtained from a small amount of experimental data.

Description

Technical Field

[0001] This invention belongs to the field of polymer synthesis technology, specifically relating to a neural network-assisted method for efficiently determining the copolymerization rate of monomers in copolymerization reactions. Background Technology

[0002] Copolymerization plays a crucial role in polymer material development by introducing the properties of different monomers into the same polymer. By altering the composition and sequence structure of copolymers, different properties can be obtained while maintaining the same chemical composition. The reactivity ratio, the ratio of the self-growing rate constant of monomers to the cross-growing rate constant, can be used to calculate the instantaneous composition of copolymers with different monomer compositions, thus enabling the control of copolymer composition and sequence structure. Therefore, developing efficient and accurate reactivity ratio measurement methods is of paramount importance.

[0003] There are two traditional methods for determining the copolymerization reactivity ratio: one based on the differential equation of copolymer composition and the other on the integral equation. The former was first proposed by Mayo-Lewis (J. Am. Chem. Soc., 1944, 66, 1594), and later refined by Fineman-Ross and... An improved method (J. Polym. Sci., 1950, 5, 259; J. Polym. Sci., Polym. Chem. Ed., 1975, 13, 2277) primarily involves measuring the conversion rate of monomers at different feed ratios while keeping the comonomer conversion rate below 10%, and then using curve fitting to obtain the reactivity ratio. However, this method has problems with measurement difficulty and accuracy, including: 1) large analytical errors when the monomer conversion rate is within 10%; 2) low copolymer yield at low conversion rates, resulting in significant losses during purification and affecting the analysis of copolymer composition; 3) difficulty in maintaining all monomers within a low conversion range when the reactivity ratios of comonomers differ significantly (e.g., one is much greater than 1 while the other is much less than 1). In response, Meyer and Lowry proposed a method based on the integral equation of copolymer composition in 1965 (J. Polym. Sci., Part A: Gen. Pap., 1965, 3, 2843). This method measures the change in monomer conversion rate with the reaction process under the same set of copolymerization reactions, and uses this data to fit the reactivity ratio, thus overcoming the limitation of a conversion rate within 10%. However, this method was later shown to be significantly affected by the reaction feed ratio (Macromolecules, 2019, 52, 2277). Therefore, the above-mentioned traditional measurement methods still have certain limitations in terms of measurement efficiency and accuracy. Since copolymer development often involves the screening of multiple comonomers and reaction conditions, the number of reactivity ratios to be measured is enormous, leading to a huge workload and cost under traditional measurement methods.

[0004] Therefore, developing efficient and accurate methods for determining the reactivity ratio can provide precise guidance for controlling the composition and sequence structure of copolymers, and accelerate the research and development process of copolymer materials by rapidly establishing a reactivity ratio library. Summary of the Invention

[0005] In view of this, the purpose of this invention is to address the problems existing in the prior art by providing a neural network-assisted method for efficiently determining the copolymerization reactivity ratio of monomers in various polymerization systems, which is applicable to various polymerization systems and helps in the composition and sequence regulation of copolymers, as well as the research and development of copolymer materials.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A neural network-assisted method for efficiently determining the copolymerization reactivity ratio of monomers in a copolymerization reaction includes the following steps:

[0008] Step 1: Collect data on monomer conversion rate with feed ratio and total conversion rate under different reactivity ratios through Monte Carlo polymerization kinetics simulation, integrate them into a multi-dimensional data matrix, and establish a database in which the reactivity ratio and conversion rate data matrices correspond one-to-one.

[0009] Step 2: Using the established database, construct a neural network model to predict the reactivity ratio from copolymerization conversion data;

[0010] Step 3: Perform copolymerization reactions on any one or more groups of comonomers at any feed ratio, and collect monomer conversion rate data;

[0011] Step 4: Using an algorithm-based data matrix completion method, expand a small amount of conversion rate data into a complete data matrix, input it into the constructed algorithm model, and output the convergence rate.

[0012] In recent years, machine learning and neural network-assisted polymer synthesis methods have been frequently reported. Leveraging the advantages of machine learning and neural network technologies in handling multivariate analysis problems and exploring complex relationships, these methods can efficiently analyze existing data, discover potential patterns beyond intuition, and have already been implemented in areas such as automated living radical polymerization (Angew. Chem. Int. Ed., 2019, 58, 3183), self-optimized polyester preparation (Macromolecules, 2020, 53, 10847), and polymerization reverse analysis platforms (Sci. China Chem., 202164, 1039). However, existing research is still limited to optimizing polymer molecular weight, and studies related to the determination of polymerization reactivity ratio are rarely reported. This invention introduces neural network algorithms into the determination of polymerization reactivity ratio, establishing a correspondence between copolymerization reaction data and polymerization reactivity ratio. A correlation model is established with copolymerization reaction data as input and polymerization reactivity ratio as output, which can effectively contribute to the efficient and accurate determination of polymerization reactivity ratio.

[0013] It is worth noting that the neural network-assisted measurement scheme disclosed in this invention integrates the comonomer consumption corresponding to each set of polymerization reactivity ratio values ​​into a high-dimensional data matrix that can be processed by computer language, describing the relationship between comonomer conversion rate and monomer feed ratio and total conversion rate. By establishing a neural network model between the polymerization conversion rate data matrix and the polymerization reactivity ratio, reliable polymerization reactivity ratio measurement results can be obtained from a small amount of experimental data.

[0014] Furthermore, the Monte Carlo polymerization kinetics simulation in step 1 includes free radical copolymerization, anionic copolymerization, cationic copolymerization, or ring-opening copolymerization, with 2 to 4 comonomers, corresponding to binary, ternary, and quaternary copolymerization.

[0015] Furthermore, considering that different polymerization methods require different parameters, the parameter settings for the Monte Carlo polymerization kinetics simulation described in this invention include:

[0016] For free radical copolymerization, a fixed parameter is set as the initiator decomposition rate constant k. d The rate constant k of the reaction between primary free radicals and monomers i Chain growth rate constant k p Chain termination rate constant k t Total monomer concentration [M], initiator concentration [I]; set the varying parameters as polymerization reactivity ratio r, monomer feed ratio; or,

[0017] For anionic copolymerization, a fixed parameter is set as the initiator initiation rate constant k. i Chain growth rate constant k pThe total monomer concentration [M] and the initiator concentration [I] are set as the varying parameters: the reactivity ratio r and the monomer feed ratio, where the product of the reactivity ratios between the two monomers is fixed at 1, i.e., r1r2 = 1; or,

[0018] For cationic copolymerization, a fixed parameter is set as the initiator initiation rate constant k. i Chain growth rate constant k p Chain transfer rate constant k tr The total monomer concentration [M] and the initiator concentration [I] are set as the varying parameters: the reactivity ratio r and the monomer feed ratio, where the product of the reactivity ratios between the two monomers is fixed at 1, i.e., r1r2 = 1; or,

[0019] For ring-open copolymerization, a fixed parameter is set as the initiator initiation rate constant k. i Chain growth rate constant k p Chain termination rate constant k t The total monomer concentration [M] and the initiator concentration [I] are set as the variable parameters: the reactivity ratio r and the monomer feed ratio. The product of the reactivity ratios between the two monomers is fixed at 1, i.e., r1r2 = 1.

[0020] Furthermore, the form of the multidimensional data matrix in step 1 includes:

[0021] For binary copolymers, the first dimension is the feed ratio, the second dimension is the total monomer conversion rate, and the third dimension is the monomer type. The data in the matrix represents the monomer conversion rate; or...

[0022] For ternary copolymers, the first and second dimensions represent the feed ratio of two monomers, the third dimension represents the total monomer conversion rate, and the fourth dimension represents the monomer types. The data in the matrix represents the monomer conversion rate; or,

[0023] For quaternary copolymers, the first to third dimensions are the feed ratios of the three monomers, the fourth dimension is the total monomer conversion rate, and the fifth dimension is the monomer type. The data in the matrix are the monomer conversion rates.

[0024] Furthermore, the range of the polymerization competition rate in step 1 is r = 0.01 to 100; wherein, when simulating binary copolymerization, a set of polymerization competition rates contains 2 r; when simulating ternary copolymerization, a set of polymerization competition rates contains 6 r; and when simulating quaternary copolymerization, a set of polymerization competition rates contains 12 r.

[0025] Furthermore, the feeding ratio in step 1 is set within the range of f. n =0~1, f n The mole fraction of any comonomer in the monomer mixture; the total conversion rate in step 1 is set to a range of 0 to 1.

[0026] It is worth noting that the Monte Carlo polymerization kinetics method described in step 1 can quickly generate a large amount of copolymerization reaction data for training neural network models. The format of the multidimensional data matrix is ​​easily processed by computer languages, facilitating feature extraction by the neural network model.

[0027] Furthermore, the neural network model in step 2 includes:

[0028] Convolutional layers: The hyperparameters to be selected include the number of output channels, kernel size, and stride.

[0029] Pooling layer: The hyperparameters for filtering include the pooling function (max pooling or average pooling) and the filter size;

[0030] Activation layer: The activation functions selected include sigmoid, tanh, softmax, and ReLU functions;

[0031] Long Short Memory Recursive (LSTM) Layers: The hyperparameters for selection include the number of layers (num_layers), the hidden dimension (hidden_size), and whether it is bifunctional.

[0032] The training of the neural network model includes:

[0033] First, the dataset is split into a training set and a validation set (80 / 20). The training set is used to train the model and update the network parameters using the backpropagation algorithm. The validation set is used to evaluate the generalization ability of the neural network and check its overfitting degree. During training, the Adam optimizer is used to optimize the network model parameters, with the mean squared error function (MSELoss) as the loss function. The batch size and learning rate are the hyperparameters that need to be selected in the algorithm. The model with the smallest prediction error after training convergence is selected as the optimal model.

[0034] It is worth noting that the neural network structure described in step 2 gives the model greater flexibility, and the introduction of the LSTM layer helps capture the characteristics of the monomer conversion rate sequence data. Dataset splitting and the Adam optimizer during neural network training can select suitable hyperparameters, which helps improve the model's generalization ability and avoid overfitting.

[0035] Furthermore, the copolymerization reaction data in step 3 includes: a feed ratio set within the range of f. n =0~1, f n The molar fraction of each comonomer in the monomer mixture; the total conversion rate is set in the range of 0 to 1; the number of reaction data sets is 3 to 20.

[0036] Furthermore, the data matrix completion method based on neural networks in step 4 includes: convolutional neural networks, recurrent neural networks, generative adversarial networks, and variational autodecoders.

[0037] Furthermore, the method for establishing the data matrix completion method includes the following steps:

[0038] Step 1: Randomly mask the data matrix generated by the Monte Carlo method, leaving only a small number of data points in the data matrix, ranging from 3 to 100. This method generates an incomplete data matrix as a dataset for subsequent training of the neural network model.

[0039] Step 2: Employ different data matrix completion algorithms, setting the input as an incomplete data matrix and the output as a complete data matrix. Each incomplete data matrix corresponds to a complete original data matrix. Train the model, comparing the output data matrix with the original data matrix and calculating the error to evaluate model performance. Finally, select the optimal data matrix completion algorithm based on error performance.

[0040] It is worth noting that the data matrix completion method described in step 4 can be used to convert a small amount of data obtained through experiments in real-world situations into a complete data matrix required by a neural network model, thereby enhancing the availability of experimental data and enabling more accurate predictions of the convergence rate in the neural network model, thus reducing the experimental cost when acquiring data.

[0041] Compared with traditional methods for determining reactivity ratio, the neural network-assisted method for determining reactivity ratio proposed in this invention has the following advantages:

[0042] 1. There is no limitation on monomer conversion rate, which can effectively solve the cumbersome operation and measurement error under low conversion rate measurement and improve measurement efficiency;

[0043] 2. Applicable to polymerization data with any feed ratio, overcoming the problem that the accuracy of traditional methods for determining the reactivity ratio is affected by the feed ratio;

[0044] 3. By using ternary and quaternary copolymer data for measurement, multiple sets of polymerization reactivity ratio data can be obtained at one time, which can be used to efficiently establish a monomer polymerization reactivity ratio library. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0046] Figure 1 This is a flowchart of the neural network-assisted method for efficiently determining the copolymerization reactivity ratio of monomers in the copolymerization reaction according to the present invention.

[0047] Figure 2 This is a two-dimensional visualization of the conversion rate data matrix under binary copolymerization according to the present invention, wherein the left and right images correspond to the conversion rate data of monomers 1 and 2, respectively.

[0048] Figure 3 This invention presents a cohesive rate model structure constructed using a convolutional neural network under binary cohesion.

[0049] Figure 4 This is a visual comparison of the conversion rate data matrix in binary copolymerization and ternary copolymerization in this invention. Detailed Implementation

[0050] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] The term "embodiment" used herein, as an example, is not necessarily to be construed as superior to or better than other embodiments. Performance testing in the embodiments of this application, unless otherwise specified, employs conventional testing methods in the art. It should be understood that the terminology used in this application is merely for describing particular implementations and is not intended to limit the scope of this disclosure.

[0052] Unless otherwise stated, the technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; other experimental methods and technical means not specifically mentioned herein refer to experimental methods and technical means commonly used by one of ordinary skill in the art.

[0053] To better illustrate the content of this application, numerous specific details are provided in the following detailed embodiments. Those skilled in the art should understand that this application can be implemented even without certain specific details. In the embodiments, some methods, means, instruments, and devices well-known to those skilled in the art are not described in detail in order to highlight the main points of this application.

[0054] Without conflict, the technical features disclosed in the embodiments of this application can be combined arbitrarily, and the resulting technical solution belongs to the content disclosed in the embodiments of this application.

[0055] This invention provides a neural network-assisted measurement scheme that integrates the comonomer consumption corresponding to each set of polymerization reactivity ratio values ​​into a high-dimensional data matrix that can be processed by a computer language, describing the relationship between comonomer conversion rate and monomer feed ratio and total conversion rate. By establishing a neural network model between the polymerization conversion rate data matrix and the polymerization reactivity ratio, reliable polymerization reactivity ratio measurement results can be obtained from a small amount of experimental data.

[0056] To better understand the present invention, the following embodiments are provided for further detailed description of the present invention, but they should not be construed as limiting the present invention. Any non-essential improvements and adjustments made by those skilled in the art based on the above-described invention are also considered to fall within the protection scope of the present invention.

[0057] Example 1

[0058] Construction of polymerization conversion rate data matrix and establishment of neural network model under binary copolymerization:

[0059] First, the kinetics of binary radical polymerization was simulated using polymerization kinetics simulation with 2500 sets of reactivity ratios, establishing 2500 corresponding conversion rate data matrices, where the reactivity ratio ranged from 0 to 1000. Each data matrix contained data on the change of monomer conversion rate with total conversion rate for 50 monomer feed ratios (50 / 0, 48 / 2, ..., 0 / 50), specifically recording the conversion rate of the comonomer every 2% increase in total conversion rate. Finally, the simulation data from these 50 sets of reactions were integrated into a 50×50 data matrix for computer recognition and manipulation. These 2500 data matrices were used as a training set to establish different neural network algorithm models, including convolutional neural networks, recurrent neural networks, and long short-term memory recurrent neural networks. In addition to the training set, 500 sets of reactivity ratios were set, and corresponding data matrices were obtained using the same polymerization kinetics simulation, serving as a validation set to test the generalization ability of the neural network models. The screening results show that the Long Short-Term Memory (LSTM) recurrent neural network has the best generalization ability when establishing the pooling rate model. Its mean squared logarithmic error (MSO) on the training set is 0.02(r1) and 0.02(r2), and its mean squared logarithmic error on the test set is 0.04(r1) and 0.03(r2). The structure of this LSM recurrent neural network model includes two recurrent layers, one pooling layer, and two fully connected layers.

[0060] Example 2

[0061] Construction of the polymerization conversion rate data matrix and establishment of the neural network model for ternary copolymerization:

[0062] First, the ternary radical polymerization kinetics process was simulated using polymerization kinetics simulation methods for 230,340 groups of reactivity ratios, establishing 230,340 corresponding conversion rate data matrices, where each reactivity ratio ranges from 0 to 1000. Each data matrix contains data on the change of monomer conversion rate with total conversion rate under 1275 monomer feed ratios (each monomer's molar fraction in the monomer mixture varies at 2% intervals). Specifically, the conversion rate of the comonomer is recorded every 2% increase in the total conversion rate. Finally, the simulation data of these 1275 reaction groups were integrated into a 50×50×50 data matrix for computer recognition and manipulation. These 230,340 data matrices were used as a training set to establish different neural network algorithm models, including convolutional neural networks, recurrent neural networks, and long short-term memory recurrent neural networks. In addition to the training set, 1600 reactivity ratios were set up, and the corresponding data matrices were obtained using the same polymerization kinetics simulation, serving as a validation set to test the generalization ability of the neural network models. The screening results show that the Long Short-Term Memory (LSTM) recurrent neural network model has the best generalization ability when establishing the competition rate model. Its mean squared logarithm error (MSO) on the training set is 0.05 (the average of six competition rate errors), and its mean squared logarithm error on the test set is 0.07 (the average of six competition rate errors). The structure of this LSM recurrent neural network model includes three recurrent layers, two pooling layers, and three fully connected layers.

[0063] Example 3

[0064] Determination of the reactivity ratio of styrene and methyl acrylate under free radical copolymerization:

[0065] Five groups of free radical polymerization experiments of styrene and methyl acrylate were conducted at 60°C using azobisisobutyronitrile (AIBN) as the free radical initiator and N,N-dimethylformamide as the solvent, with a reaction time of 2 hours. The feed ratios in these five experiments ranged from 9 / 91 to 88 / 12, and the total monomer conversion rates after the reaction were stopped ranged from 36% to 62%. Based on the feed ratios and total conversion rates, the conversion rate data of the two monomers in the five reactions were filled into the corresponding positions in a 50×50 data matrix, with the remaining positions left empty as an incomplete data matrix. Subsequently, the data matrix was expanded to be complete using a generative adversarial network (GAN) algorithm and input into the neural network model established in Example 1, outputting the corresponding reactivity ratios, with results of r1 = 0.16 and r2 = 0.72. Meanwhile, we also used the Meyer-Lowry method to determine the reactivity ratios of styrene and methyl acrylate, and obtained the results r1 = 0.14 and r2 = 0.74, which are consistent with the results given by the neural network method, indicating that the neural network-assisted reactivity ratio determination method has good reliability.

[0066] Example 4

[0067] Determination of the reactivity ratio of styrene, butyl acrylate, and vinyl acetate in free radical terpolymerization:

[0068] Seven groups of free radical ternary polymerization experiments of styrene, butyl acrylate, and vinyl acetate were conducted at 60°C using azobisisobutyronitrile (AIBN) as the free radical initiator and N,N-dimethylformamide as the solvent, with a reaction time of 2 hours. In these seven groups of experiments, the proportion of the three monomers in the monomer mixture ranged from 12% to 89%, and the total monomer conversion rate after the reaction was stopped ranged from 54% to 77%. Based on the feed ratio and total conversion rate, the conversion rate data of the three monomers in the seven groups of reactions were filled into the corresponding positions in a 50×50×50 data matrix, with the remaining positions left empty as an incomplete data matrix. Subsequently, the data matrix was expanded to be complete using a generative adversarial network (GAN) algorithm and input into the neural network model established in Example 2, outputting the corresponding reactivity ratio, the result of which is r. 12 =0.79, r 21 =0.25, r 13 =18.81, r 31 =0.02, r 23 =3.48, r 32 =0.02, where 1, 2, and 3 refer to styrene, butyl acrylate, and vinyl acetate, respectively.

[0069] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A neural network-assisted method for efficiently determining the copolymerization reactivity ratio of monomers in a copolymerization reaction, characterized in that, Includes the following steps: Step 1: Collect data on monomer conversion rate with feed ratio and total conversion rate under different reactivity ratios through Monte Carlo polymerization kinetics simulation, integrate them into a multi-dimensional data matrix, and establish a database in which the reactivity ratio and conversion rate data matrices correspond one-to-one. Step 2: Using the established database, construct a neural network model to predict the reactivity ratio from copolymerization conversion data; Step 3: Perform copolymerization reactions on any one or more groups of comonomers at any feed ratio, and collect monomer conversion rate data; Step 4: Using an algorithm-based data matrix completion method, expand a small amount of conversion rate data into a complete data matrix, input it into the constructed algorithm model, and output the convergence rate; The multidimensional data matrix in step 1 takes the following form: for binary copolymers, the first dimension is the feed ratio, the second dimension is the total monomer conversion rate, and the third dimension is the monomer type; the data in the matrix represents the monomer conversion rate. Alternatively, For ternary copolymers, the first and second dimensions represent the feed ratio of two monomers, the third dimension represents the total monomer conversion rate, and the fourth dimension represents the monomer types. The data in the matrix represents the monomer conversion rate; or, For quaternary copolymers, the first to third dimensions are the feed ratios of the three monomers, the fourth dimension is the total monomer conversion rate, and the fifth dimension is the monomer type. The data in the matrix are the monomer conversion rates.

2. The method according to claim 1, characterized in that, The Monte Carlo polymerization kinetics simulation in step 1 includes free radical copolymerization, anionic copolymerization, cationic copolymerization, or ring-opening copolymerization, with 2 to 4 comonomers, corresponding to binary, ternary, and quaternary copolymerization.

3. The method according to claim 2, characterized in that, The parameter settings for the Monte Carlo polymerization kinetics simulation include: For free radical copolymerization, a fixed parameter is set as the initiator decomposition rate constant. k d Rate constant of the reaction between primary free radicals and monomers k i Chain growth rate constant k p Chain termination rate constant k t The total monomer concentration [M] and the initiator concentration [I] are set as the reactivity ratio. r The ratio of individual feed ingredients; or, For anionic copolymerization, a fixed parameter is set as the initiator initiation rate constant. k i Chain growth rate constant k p The total monomer concentration [M] and the initiator concentration [I] are set as the reactivity ratio. r The monomer feed ratio, wherein the product of the reactivity ratios between the two monomers is fixed at 1, i.e. r 1 r 2 = 1; or, For cationic copolymerization, a fixed parameter is set as the initiation rate constant of the initiator. k i Chain growth rate constant k p Chain transfer rate constant k tr The total monomer concentration [M] and the initiator concentration [I] are set as the reactivity ratio. r The monomer feed ratio, wherein the product of the reactivity ratios between the two monomers is fixed at 1, i.e. r 1 r 2 = 1; or, For ring-opening copolymerization, a fixed parameter is set as the initiator initiation rate constant. k i Chain growth rate constant k p Chain termination rate constant k t The total monomer concentration [M] and the initiator concentration [I] are set as the reactivity ratio. r The monomer feed ratio, wherein the product of the reactivity ratios between the two monomers is fixed at 1, i.e. r 1 r 2 = 1.

4. The method according to claim 2, characterized in that, The range of the reactivity ratio set in step 1 is: r =0.01 ~ 100; where, in simulating binary copolymerization, a set of reactivity ratios includes 2 r When simulating ternary copolymerization, a set of polymerization reactivity ratios contains 6... r When simulating quaternary copolymerization, a set of reactivity ratios contains 12... r .

5. The method according to claim 1, characterized in that, The feeding ratio setting range in step 1 is: f n = 0 ~ 1, f n The molar fraction of each comonomer in the monomer mixture; the total conversion rate in step 1 is set to a range of 0 to 1.

6. The method according to claim 1, characterized in that, The neural network model in step 2 includes: Convolutional layers: The selected hyperparameters include the number of output channels, kernel size, and stride. Pooling layer: The hyperparameters for filtering include the pooling function and the filter size; Activation layer: The activation functions selected include sigmoid, tanh, softmax, and ReLU. Long Short Memory Recursive Layer: The hyperparameters for selection include the number of layers, the dimension of hidden layers, and whether it is bidirectional.

7. The method according to claim 1, characterized in that, The copolymerization reaction data in step 3 includes: the feed ratio is set within a certain range. f n = 0 ~ 1, f n The molar fraction of any comonomer in the monomer mixture; the total conversion rate is set in the range of 0 to 1; the number of reaction data sets is 3 to 20.

8. The method according to claim 1, characterized in that, The data matrix completion method based on neural networks in step 4 includes: convolutional neural networks, recurrent neural networks, generative adversarial networks, and variational autodecoders.

9. The method according to claim 8, characterized in that, The method for establishing the data matrix completion method includes the following steps: Step 1: Randomly mask the data matrix generated by the Monte Carlo method, leaving only a small number of data points in the data matrix, ranging from 3 to 100. This method generates an incomplete data matrix as a dataset for subsequent training of the neural network model. Step 2: Using different data matrix completion algorithms, the input is set to an incomplete data matrix and the output is a complete data matrix. Each incomplete data matrix corresponds to a complete original data matrix. The model is trained, and the output data matrix is ​​compared with the original data matrix. The error between the two is calculated to evaluate the performance of the model. Finally, the optimal data matrix completion algorithm is selected based on the error performance.

Images