Method, system, and storage medium for predicting microstructure of impact polypropylene
By training and predicting using neural network models, the accuracy of predicting the microstructure of impact-resistant polypropylene was solved, improving product quality and performance and enhancing the adaptability of process parameters.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- EAST CHINA UNIV OF SCI & TECH
- Filing Date
- 2024-08-05
- Publication Date
- 2026-07-24
AI Technical Summary
Existing technologies struggle to accurately predict the microstructure of impact-resistant polypropylene, resulting in insufficient adaptability to process parameter adjustments and impacting product quality and performance.
A neural network-based approach was adopted, which used a large amount of historical data to train a BP neural network model to predict the molecular weight characteristics of impact-resistant polypropylene, and then adjusted the process parameters to achieve the target value.
It improves the quality and performance of impact-resistant polypropylene products, enhances adaptability to various process parameters, and enables more accurate adjustment of process parameters.
Smart Images
Figure CN118983036B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer chemical engineering technology, and in particular to a method for predicting the microstructure of impact-resistant polypropylene, a system for predicting the microstructure of impact-resistant polypropylene, and a computer-readable storage medium. Background Technology
[0002] Impact-resistant polypropylene (PP) is a complex multi-component, multiphase polymer alloy system. Its matrix is polypropylene, and by introducing appropriate impact modifiers into the PP matrix, its impact resistance and low-temperature toughness can be significantly improved while maintaining the excellent properties of PP. It has found wide applications in the automotive, home appliance, and furniture industries. With increasingly stringent quality requirements, high-performance impact-resistant PP products have become a target for manufacturers. Researchers have continuously studied the process parameters and mechanical properties of impact-resistant PP, discovering that different polymerization process parameters result in different molecular weight characteristics, thus affecting the strength, toughness, and other mechanical properties of impact-resistant PP. However, traditional preparation methods, such as those based on mathematical models or using Aspen software for chemical process simulation, have limited applicability. This is especially true for polymers like impact-resistant PP, which have multi-component and complex phase structures, making performance prediction using mechanistic models challenging. Furthermore, the various interactions and molecular arrangements involved in the polymerization process often make it difficult for differential equation systems to fully capture these complexities. The complexity and computational burden of the model further increase when considering factors such as polymer chain orientation, copolymer composition, and the influence of additives.
[0003] In order to overcome the above-mentioned defects in the existing technology, there is an urgent need in the field for an improved method for predicting the microstructure of impact-resistant polypropylene, which can be used to more accurately adjust the process parameters for processing impact-resistant polypropylene and improve the adaptability to various process parameters, thereby improving the quality and performance of impact-resistant polypropylene products. Summary of the Invention
[0004] The following provides a brief overview of one or more aspects to offer a basic understanding of them. This overview is not an exhaustive summary of all conceived aspects, nor is it intended to identify key or decisive elements of all aspects, nor to define the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed descriptions that follow.
[0005] To overcome the aforementioned deficiencies in the existing technology, this invention provides a method for predicting the microstructure of impact-resistant polypropylene, a system for predicting the microstructure of impact-resistant polypropylene, and a computer-readable storage medium. This method utilizes a neural network model, trained and predicted using a large amount of historical data, to more accurately adjust the process parameters for processing impact-resistant polypropylene and improve its adaptability to various process parameters, thereby enhancing the quality and performance of impact-resistant polypropylene products.
[0006] Specifically, the method for predicting the microstructure of impact-resistant polypropylene according to the first aspect of the present invention includes the following steps: obtaining process parameters for processing impact-resistant polypropylene. The process parameters include the concentration and homopolymerization time of at least one first raw material in the homopolymerization stage, and the concentration and copolymerization time of at least one second raw material in the copolymerization stage; determining predicted values of the molecular weight characteristics of the impact-resistant polypropylene product based on the process parameters; and adjusting the process parameters based on the difference between the target value and the predicted value of the molecular weight characteristics to obtain an impact-resistant polypropylene product that meets the target value.
[0007] Furthermore, in some embodiments of the present invention, the at least one first raw material includes propylene and hydrogen. The at least one second raw material includes propylene, ethylene, and hydrogen. The target and predicted values of the molecular weight characteristics correspond to at least one of the following: first number-average molecular weight Mn_PP, first weight-average molecular weight Mw_PP, first molecular weight distribution index PDI_PP, first mass percentage PP_per; second number-average molecular weight Mn_iPP, second weight-average molecular weight Mw_iPP, second molecular weight distribution index PDI_iPP; third number-average molecular weight Mn_aPP, third weight-average molecular weight Mw_aPP, third molecular weight distribution index PDI_aPP; fourth number-average molecular weight Mn_EPR, fourth weight-average molecular weight Mw_EPR, fourth molecular weight distribution index PDI_EPR, fourth mass percentage EPR_per; first ethylene molar content E_mole_EPR; fifth number-average molecular weight Mn_EPS, fifth weight-average molecular weight Mw_EPS, fifth molecular weight distribution index PDI_EPS, fifth mass percentage EPS_per; second ethylene molar content E_mole_EPS; and third ethylene molar content E_in_IPC in the polymer.
[0008] Furthermore, in some embodiments of the present invention, the step of determining the predicted value of the molecular weight characteristics of the impact-resistant polypropylene product based on the process parameters includes: inputting the process parameters into a pre-trained BP neural network model to determine the predicted value of the molecular weight characteristics of the impact-resistant polypropylene product via the BP neural network model.
[0009] Furthermore, in some embodiments of the present invention, the step of training the BP neural network model includes: acquiring multiple sets of historical sample data for processing impact-resistant polypropylene to construct a historical sample dataset. Each set of historical sample data includes the process parameters and true values of molecular weight characteristics of the corresponding historical sample; constructing a BP neural network model to be trained. The BP neural network model consists of an input layer, two hidden layers, and an output layer; and sequentially inputting each set of training sample data from the historical sample dataset into the BP neural network model to obtain the corresponding predicted values of molecular weight characteristics, and adjusting the learning parameters of the BP neural network model according to the difference between the predicted values and the true values of molecular weight characteristics of the corresponding training samples, thereby training the BP neural network model.
[0010] Furthermore, in some embodiments of the present invention, before sequentially inputting each training sample from the historical sample dataset into the BP neural network model, the prediction method further includes the following step: normalizing each historical sample data in the historical sample dataset to normalize its value to between -1 and 1:
[0011]
[0012] Among them, X reduced，p，i X is the normalized variable of the p-th historical sample data of the i-th neuron. p，i X is the original variable of the p-th historical sample data of the i-th neuron. min，i X is the minimum value of the variable of the i-th neuron in the historical sample dataset. max，i It is the maximum value of the variable of the i-th neuron in the historical sample dataset.
[0013] Furthermore, in some embodiments of the present invention, the step of sequentially inputting each training sample data in the historical sample dataset into the BP neural network model to obtain the corresponding molecular weight characteristic prediction value includes: randomly dividing the historical sample dataset into a training set and a test set according to a preset ratio; defining the historical sample data in the training set as training sample data; and sequentially inputting each group of training sample data in the training set into the BP neural network model to obtain the corresponding molecular weight characteristic prediction value.
[0014] Furthermore, in some embodiments of the present invention, the following steps are also included: defining historical sample data in the test set as test sample data; in response to completing the training of the BP neural network model, sequentially inputting each group of test sample data in the test set into the BP neural network model to obtain the corresponding molecular weight characteristic test values; and verifying the effectiveness of the BP neural network model based on the degree of agreement between the regression curve of each test value and the true value of the corresponding test sample.
[0015] Furthermore, in some embodiments of the present invention, the number N of neurons in the input layer is... i The number of neurons in the output layer (No) is determined based on the input dimension of the predicted molecular weight characteristics. The number of neurons in the hidden layer is determined based on the number of neurons in the input layer (Ni), the number of neurons in the output layer (No), and the number of training sample data groups (N). s Sure:
[0016]
[0017] The value of α depends on the type of molecular weight property to be predicted, and ranges from 4 to 10.
[0018] Furthermore, in some embodiments of the present invention, the step of sequentially inputting each group of training sample data from the historical sample dataset into the BP neural network model to obtain the corresponding molecular weight characteristic prediction value includes: adjusting the weights w connecting neurons i and j in the BP neural network model. j，i Initialize as a normal distribution:
[0019]
[0020] Where U is a function of the normal distribution, n in n is the row number in the training sample data matrix. out The number of columns in the training sample data matrix; the threshold B of each neuron i in the BP neural network model. i Initialize to 0; calculate the net input variable of each neural network layer based on the weights of the connections between the neurons and the thresholds of the neurons in each layer:
[0021]
[0022] in, It is the net input variable of the p-th data point of the j-th neuron in the (k+1)-th layer of the neural network. It is the threshold of the j-th neuron in the (k+1)-th layer of the neural network. It represents the weight of the connection between the j-th neuron in the (k+1)-th layer of the neural network and the i-th neuron in the k-th layer of the neural network. It is the output variable of the p-th data point of the i-th neuron in the k-th layer of the neural network. The output variable of the j-th neuron in the (k+1)-th layer of the neural network is calculated by passing the net input variable of the j-th neuron in the (k+1)-th layer of the neural network according to the user-defined activation function.
[0023]
[0024] in, Let p be the net input variable of the p-th data point of the j-th neuron in the (k+1)-th layer of the neural network. Let α1 be the output variable of the p-th data point of the j-th neuron in the (k+1)-th layer of the neural network, with a value of 1.5.
[0025] Furthermore, in some embodiments of the present invention, the step of adjusting the learning parameters of the BP neural network model based on the difference between the predicted and true values of the molecular weight characteristics of the corresponding training samples to train the BP neural network model includes: calculating the mean square error between the predicted and true values of the molecular weight characteristics of each training sample.
[0026]
[0027] in, It is the true value of the p-th training sample of the j-th neuron. It is the predicted value of the j-th neuron and the p-th training sample; and the backpropagation of the BP neural network model based on the mean square error is performed using the Adam optimizer until the training of the BP neural network model is completed.
[0028] Furthermore, in some embodiments of the present invention, the step of using the Adam optimizer to perform backpropagation on the BP neural network model based on the mean squared error until the training of the BP neural network model is completed includes: setting the maximum number of iterations t for training the BP neural network model. max The parameters include the indicator parameter θ1 and the patience parameter to prevent overfitting during training. The value of the patience parameter is determined according to the formula patience. θ =max(input_n, output_n) / 2. input_n is the number of input nodes. output_n is the number of output nodes. The initial value of the indicator parameter θ1 is 0; this is in response to the current iteration number t not reaching the maximum iteration number t. max And if the indication parameter θ1 is less than or equal to the patience parameter, the first moment estimate m of the Adam optimizer will be... t and second-order moment estimate v tThese are combined to update the weights of the connections between neurons and the thresholds of each neuron in the BP neural network model:
[0029]
[0030] Among them, w t b t These are the weights and thresholds after the t-th round update. t-1 b t-1 α is the weight and threshold in round t-1. α2 is the learning rate. θ2 is the zero-prevention constant. The first-moment estimate m of the Adam optimizer is... t The expression is:
[0031] m t =β1m t-1 +(1-β1)h t
[0032] Where, m t-1 and m t represents the first-moment estimates before and after the iteration, respectively, where β1 is the exponentially weighted average parameter. t This represents the gradient of the weight ω or the threshold b.
[0033] The second-order moment estimate v of the Adam optimizer t The expression is:
[0034] v t =β2v t-1 +(1-β2)h t 2
[0035] Among them, v t-1 and v t These are the second-order moment estimates before and after the iteration, respectively, with β2 being the exponentially weighted average parameter. t 2 The gradient squared represents the weight ω or threshold b; it determines whether the mean square error of the current round has increased; if it has increased, it increments the indicator parameter θ1 by 1, otherwise it sets the indicator parameter θ1 to zero; and it responds to the current round number t reaching the maximum iteration round number t. max If the indicated parameter θ1 is greater than the patience parameter, the training of the BP neural network model is determined to be complete.
[0036] Furthermore, the above-described impact-resistant polypropylene microstructure prediction system according to a second aspect of the present invention includes a memory and a processor. The memory stores computer instructions. The processor is connected to the memory and configured to execute the computer instructions stored in the memory to implement the impact-resistant polypropylene microstructure prediction method as provided in the first aspect of the present invention.
[0037] Furthermore, the computer-readable storage medium provided according to the third aspect of the present invention stores computer instructions thereon. When the computer instructions are executed by a processor, the method for predicting the microstructure of impact-resistant polypropylene as provided in the first aspect of the present invention is implemented. Attached Figure Description
[0038] The above-described features and advantages of the present invention will be better understood after reading the following detailed description of embodiments of the present disclosure in conjunction with the accompanying drawings. In the drawings, components are not necessarily drawn to scale, and components having similar related characteristics or features may have the same or similar reference numerals.
[0039] Figure 1 A schematic diagram of the structure of a BP neural network model provided according to some embodiments of the present invention is shown.
[0040] Figure 2A A regression curve of the second number-average molecular weight Mn iPP provided according to some embodiments of the present invention is shown.
[0041] Figure 2B A regression curve of the second weight-average molecular weight Mw iPP provided according to some embodiments of the present invention is shown.
[0042] Figure 3A The regression curve of the fourth number-average molecular weight MnEPR of ethylene-propylene random copolymer provided according to some embodiments of the present invention is shown.
[0043] Figure 3B The regression curve of the fifth number-average molecular weight Mn EPS of ethylene-propylene multiblock copolymer provided according to some embodiments of the present invention is shown.
[0044] Figure 4 A flowchart illustrating a method for predicting the microstructure of impact-resistant polypropylene according to some embodiments of the present invention is shown. Detailed Implementation
[0045] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Although the description of the present invention is presented in conjunction with preferred embodiments, this does not mean that the features of the invention are limited to these embodiments. On the contrary, the purpose of describing the invention in conjunction with embodiments is to cover other options or modifications that may be derived based on the claims of the present invention. To provide a thorough understanding of the invention, many specific details will be included in the following description. The invention may also be implemented without using these details. Furthermore, to avoid confusion or obscuring the focus of the invention, some specific details will be omitted in the description.
[0046] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0047] Furthermore, the terms "upper," "lower," "left," "right," "top," "bottom," "horizontal," and "vertical" used in the following description should be understood as the orientations shown in the relevant paragraphs and accompanying drawings. These relative terms are for illustrative purposes only and do not imply that the described apparatus must be manufactured or operated in a specific orientation, and therefore should not be construed as limiting the invention.
[0048] It is understood that although terms such as "first," "second," and "third" may be used herein to describe various components, regions, layers, and / or parts, these components, regions, layers, and / or parts should not be limited by these terms, and these terms are only used to distinguish different components, regions, layers, and / or parts. Therefore, the first components, regions, layers, and / or parts discussed below may be referred to as second components, regions, layers, and / or parts without departing from some embodiments of the present invention.
[0049] As mentioned above, traditional preparation methods, such as those based on mathematical models or using Aspen software for chemical process simulation, have limited applicability. This is especially true for impact-resistant polypropylene, a polymer with multi-component and complex phase structures, where predicting its performance using mechanistic models is challenging. Furthermore, the various interactions and molecular arrangements involved in the polymerization process often make it difficult for differential equation systems to fully capture these complexities. The complexity and computational burden of the model increase further, particularly when considering factors such as polymer chain orientation, copolymer composition, and the influence of additives.
[0050] To overcome the aforementioned deficiencies in the existing technology, this invention provides a method for predicting the microstructure of impact-resistant polypropylene, a system for predicting the microstructure of impact-resistant polypropylene, and a computer-readable storage medium. Based on a neural network model and trained and predicted using a large amount of historical data, this method can more accurately adjust the process parameters for processing impact-resistant polypropylene and improve its adaptability to various process parameters, thereby enhancing the quality and performance of impact-resistant polypropylene products.
[0051] In some non-limiting embodiments, the method for predicting the microstructure of impact-resistant polypropylene provided in the first aspect of the present invention can be implemented based on the system for predicting the microstructure of impact-resistant polypropylene provided in the second aspect of the present invention. Specifically, the system for predicting the microstructure of impact-resistant polypropylene can be configured with a memory and a processor. The memory includes, but is not limited to, the computer-readable storage medium provided in the third aspect of the present invention, on which computer instructions are stored. The processor is connected to the memory and configured to execute the computer instructions stored in the memory to implement the method for predicting the microstructure of impact-resistant polypropylene provided in the first aspect of the present invention.
[0052] Specifically, the processor can pre-train the BP neural network model offline. The processor can first acquire multiple sets of historical sample data from the processing of impact-resistant polypropylene to construct a historical sample dataset. Each historical sample dataset includes the actual values of the corresponding historical sample's process parameters and molecular weight characteristics.
[0053] Please refer to Figure 1 . Figure 1 A schematic diagram of the structure of a BP neural network model provided according to some embodiments of the present invention is shown.
[0054] like Figure 1 As shown, the processor can build a BP neural network model to be trained. Here, the BP neural network model consists of an input layer 11, two hidden layers 12, and an output layer 13.
[0055] Subsequently, in some preferred embodiments, the processor may first normalize the historical sample data in the historical sample dataset to normalize its values to between -1 and 1:
[0056]
[0057] Among them, X reduced，p，i X is the normalized variable of the p-th historical sample data of the i-th neuron. p，i X is the original variable of the p-th historical sample data of the i-th neuron. min，i X is the minimum value of the variable of the i-th neuron in the historical sample dataset. max，i It is the maximum value of the variable of the i-th neuron in the historical sample dataset.
[0058] Next, the processor can randomly divide the historical sample dataset into a training set and a test set according to a preset ratio (e.g., 8:2). Then, the processor can define the historical sample data in the training set as training sample data, and input each group of training sample data in the training set into the BP neural network model in sequence to obtain the corresponding molecular weight characteristic prediction values.
[0059] Furthermore, in determining the structure of the aforementioned BP neural network model, the number of neurons N in the input layer... i The number of neurons N in the output layer is determined based on the input dimension of each group of training sample data. o The number of neurons in the hidden layer is determined by the output dimension based on the predicted value of molecular weight characteristics, and the number of neurons in the hidden layer is based on the number of neurons N in the input layer. i The number of neurons N in the output layer o and the number of training sample data groups N s Sure:
[0060]
[0061] The value of α depends on the type of molecular weight property to be predicted, and ranges from 4 to 10. Here, when predicting the molecular weight properties of both homopolymers and copolymers simultaneously, α can be set to 4; when predicting the molecular weight properties of homopolymers, α can be set to 8; and when predicting the molecular weight properties of copolymers, α can be set to 10.
[0062] Furthermore, after sequentially inputting the training sample data from each group in the training set into the BP neural network model, the processor can adjust the weights w of the connections between neurons i and j in the BP neural network model. j，i Initialize as a normal distribution:
[0063]
[0064] Where U is the normal distribution function, n in n is the row number in the training sample data matrix. out is the number of columns in the training sample data matrix.
[0065] Then, the processor can first set the threshold B of each neuron i in the BP neural network model. i Initialize to 0, and then calculate the net input variables of the BP neural network model based on the weights of the connections between neurons and the thresholds of neurons in each layer of the neural network:
[0066]
[0067] in, It is the net input variable of the p-th data point of the j-th neuron in the (k+1)-th layer of the neural network. It is the threshold of the j-th neuron in the (k+1)-th layer of the neural network. It represents the weight of the connection between the j-th neuron in the (k+1)-th layer of the neural network and the i-th neuron in the k-th layer of the neural network. is the output variable of the p-th data point of the i-th neuron in the k-th layer of the neural network. The output variable of the j-th neuron in the (k+1)-th layer of the neural network is calculated by passing the net input variable of the j-th neuron in the (k+1)-th layer of the neural network according to a user-defined activation function.
[0068]
[0069] in, Let p be the net input variable of the p-th data point of the j-th neuron in the (k+1)-th layer of the neural network. Let α1 be the output variable of the p-th data point of the j-th neuron in the (k+1)-th layer of the neural network, with a value of 1.5.
[0070] In this way, the processor can further process the variables in the hidden layer and the output layer using the same formula described above, thereby obtaining the output variables of the BP neural network model.
[0071] Then, the processor can adjust the learning parameters of the BP neural network model based on the difference between its molecular weight characteristics and the true values of the corresponding training samples, in order to train the BP neural network model.
[0072] Specifically, the processor can calculate the mean squared error between the predicted and true values of the molecular weight characteristics of each training sample:
[0073]
[0074] in, It is the true value of the p-th training sample of the j-th neuron. It is the predicted value of the j-th neuron for the p-th training sample.
[0075] Then, the processor can use the Adam optimizer to perform backpropagation on the BP neural network model based on mean squared error until the training of the BP neural network model is complete.
[0076] Specifically, during the backpropagation process of a BP neural network, the processor can first set the maximum number of iterations tm for training the BP neural network model. ax、 The parameter θ1 is an indicator parameter, and the patience parameter is a parameter to prevent overfitting during training. Here, the value of the patience parameter is determined according to the formula patience. θ =max(input_n, output_n) / 2, where input_n is the number of input nodes and output_n is the number of output nodes. The initial value of the indicator parameter θ1 is 0. Here, in the process of predicting the molecular weight characteristics of homopolymers, the patience parameter... θ It can be set to 5. The patience parameter is used in predicting the molecular weight characteristics of copolymers.θ It can be set to 6.
[0077] Subsequently, in response to the current round number t not reaching the maximum iteration round number t max And if the indicator parameter θ1 is less than or equal to the patience parameter, the processor can estimate the first moment m of the Adam optimizer. t and second-order moment estimate v t These are combined to update the weights of connections between neurons and the thresholds of each neuron in the BP neural network model:
[0078]
[0079] Among them, w t b t These are the weights and thresholds after the t-th round update, w t-1 b t-1 These are the weights and threshold for the (t-1)th round. α2 is the learning rate, with a value of 0.0025. θ2 is the zero-prevention constant, with a value of 10. -8 .
[0080] Here, the first-order moment estimate m of the Adam optimizer t The expression is:
[0081] m t =β1m t-1 +(1-β1)h t
[0082] Where, m t-1 and m t These represent the first-moment estimates before and after the iteration, respectively, where β1 is the exponentially weighted average parameter, and h... t This represents the gradient of the weight ω or the threshold b.
[0083] Here, the second-order moment estimate v of the Adam optimizer t The expression is:
[0084] v t =β2v t-1 +(1-β2)h t 2
[0085] Among them, v t-1 and v t These are the second-order moment estimates before and after the iteration, respectively, where β2 is the exponentially weighted average parameter, and h... t 2 This represents the squared gradient of the weight ω or the threshold b.
[0086] Then, the processor can determine whether the mean square error of the current round has increased. If it has increased, the indicator parameter θ1 is incremented by 1; otherwise, the indicator parameter θ1 is set to zero.
[0087] Then, in response to the current round number t, the maximum number of iteration rounds t is reached. max If the indicator parameter θ1 is greater than the patience parameter, the processor can determine that the training of the BP neural network model is complete.
[0088] Furthermore, in some preferred embodiments, the processor can also define historical sample data in the test set as test sample data. In response to completing the training of the BP neural network model, the processor can sequentially input each set of test sample data from the test set into the BP neural network model to obtain the corresponding molecular weight characteristic test values. Afterwards, the processor can verify the effectiveness of the BP neural network model based on the degree of agreement between the regression curve of each test value and the true value of the corresponding test sample.
[0089] Specifically, the processor can use mean squared error (MSE), mean absolute error (MAE), and goodness of fit (r). 2 Three metrics are used for performance evaluation:
[0090]
[0091] Among them, y i This represents the true value of the test sample. This represents the average of the true values of the test sample. This represents the true value of the test sample.
[0092] Here, the target and predicted values of the aforementioned molecular weight characteristics correspond to at least one of the following: first number-average molecular weight Mn_PP, first weight-average molecular weight Mw_PP, first molecular weight distribution index PDI_PP, first mass percentage PP_per; second number-average molecular weight Mn_iPP, second weight-average molecular weight Mw_iPP, second molecular weight distribution index PDI_iPP; third number-average molecular weight Mn_aPP, third weight-average molecular weight Mw_aPP, third molecular weight distribution index PDI_aPP; fourth number-average molecular weight Mn_EPR, fourth weight-average molecular weight Mw_EPR, fourth molecular weight distribution index PDI_EPR, fourth mass percentage EPR_per; first ethylene molar content E_mole_EPR; fifth number-average molecular weight Mn_EPS, fifth weight-average molecular weight Mw_EPS, fifth molecular weight distribution index PDI_EPS, fifth mass percentage EPS_per; second ethylene molar content E_mole_EPS; and third ethylene molar content E_in_IPC in the polymer.
[0093] Please refer to Table 1 for further details. Figure 2A and Figure 2B Table 1 shows the performance indicators of the molecular weight properties of homopolymers provided according to some embodiments of the present invention. Figure 2A A regression curve of the second number-average molecular weight Mn_iPP provided according to some embodiments of the present invention is shown. Figure 2B A regression curve of the second weight-average molecular weight Mw_iPP provided according to some embodiments of the present invention is shown.
[0094] Table 1 Performance Indicators of Homopolymer Molecular Weight Characteristics
[0095] 0.00034036 0.01100553 0.9966
[0096] As shown in Table 1, when the value of α is 8, the BP neural network model predicts the molecular weight characteristics of the homopolymer. The mean squared error of the above BP neural network model is 0.00034036, the mean absolute error is 0.01100553, and the goodness of fit is 0.9966. Figure 2A and Figure 2B As shown, the BP neural network model has low error and high accuracy in determining the molecular weight characteristics of homopolymers.
[0097] Please also refer to Table 2. Figure 3A and Figure 3B Table 2 shows the performance indicators of the molecular weight properties of the copolymers provided according to some embodiments of the present invention. Figure 3A A regression curve of the fourth number-average molecular weight Mn_EPR of ethylene-propylene random copolymer provided according to some embodiments of the present invention is shown. Figure 3B The regression curve of the fifth number-average molecular weight Mn_EPS of ethylene-propylene multiblock copolymer provided according to some embodiments of the present invention is shown.
[0098] Table 2 Performance Indicators of Copolymer Molecular Weight Characteristics
[0099] 0.00064456 0.01680668 0.9601
[0100] As shown in Table 2, when the value of α is 10, the BP neural network model predicts the molecular weight characteristics of the copolymer. The mean squared error of the above BP neural network model is 0.00064456, the mean absolute error is 0.01680668, and the goodness of fit is 0.9601. Figure 3A and Figure 3B As shown, the BP neural network model has low error and high accuracy in determining the molecular weight characteristics of copolymers.
[0101] Please refer to the details. Figure 4 . Figure 4 A flowchart illustrating a method for predicting the microstructure of impact-resistant polypropylene according to some embodiments of the present invention is shown.
[0102] like Figure 4 As shown, during the online processing of impact polypropylene, the processor can first acquire the process parameters for processing the impact polypropylene. These process parameters include the concentration and homopolymerization time of at least one first raw material in the homopolymerization stage, and the concentration and copolymerization time of at least one second raw material in the copolymerization stage.
[0103] Furthermore, in some embodiments, the at least one first raw material includes propylene and hydrogen, and the at least one second raw material includes propylene, ethylene and hydrogen.
[0104] The processor can then determine the predicted values of the molecular weight characteristics of the impact-resistant polypropylene product based on the above process parameters.
[0105] Specifically, the processor can input process parameters into a pre-trained BP neural network model to determine the predicted values of the molecular weight characteristics of impact-resistant polypropylene products via the BP neural network model.
[0106] Then, the processor can adjust the process parameters based on the difference between the target value and the predicted value of the molecular weight characteristics to obtain an impact-resistant polypropylene product with the target load value.
[0107] In summary, the method, system, and computer-readable storage medium for predicting the microstructure of impact-resistant polypropylene provided by this invention can all be used to more accurately adjust the process parameters for processing impact-resistant polypropylene and improve the adaptability to various process parameters by using a neural network model for training and prediction with a large amount of historical data, thereby improving the quality and performance of impact-resistant polypropylene products.
[0108] Although the methods described above are illustrated and depicted as a series of actions for the sake of simplicity, it should be understood and appreciated that these methods are not limited by the order of the actions, as some actions may occur in a different order and / or concurrently with other actions from the illustrations and descriptions herein or not illustrated and described herein but which may be understood by those skilled in the art, according to one or more embodiments.
[0109] The prior description of this disclosure is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not intended to be limited to the examples and designs described herein, but should be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for predicting the microstructure of impact-resistant polypropylene, characterized in that, Includes the following steps: Obtain process parameters for processing impact-resistant polypropylene, wherein the process parameters include the first raw material concentration and homopolymerization time of at least one first raw material in the homopolymerization stage, and the second raw material concentration and copolymerization time of at least one second raw material in the copolymerization stage. The process parameters are input into a pre-trained BP neural network model to determine the predicted values of the molecular weight characteristics of the impact-resistant polypropylene product. The step of training the BP neural network model includes: acquiring multiple sets of historical sample data from the processing of impact-resistant polypropylene to construct a historical sample dataset, wherein each historical sample data includes the corresponding process parameters and the true values of the molecular weight characteristics; constructing a BP neural network model to be trained, wherein the BP neural network model consists of an input layer, two hidden layers, and an output layer; and configuring each neuron in the BP neural network model... i , j Weights of inter-connections Initialize as a normal distribution: in, It is a normal distribution function. The number of rows in the training sample data matrix. The column number in the training sample data matrix; the number of neurons in the BP neural network model. i threshold Initialize to 0; calculate the net input variable of each neural network layer based on the weights of the connections between the neurons and the thresholds of the neurons in each layer: in, It is the first k +1 layer neural network j The first neuron p Net input variables of the data, It is the first k +1 layer neural network j Threshold of a neuron It is the first k +1 layer neural network j The first neuron and the second k The first layer of a neural network i The weights of connections between neurons It is the first k The first layer of a neural network i The first neuron p The output variable of the data, the first k +1 layer neural network j The output variable of the nth neuron is determined by the nth neuron. k +1 layer neural network j The net input variable of each neuron is calculated based on the custom activation function: in, For the (k+1)th layer of the neural network, the th j The first neuron p Net input variables of the data, For the (k+1)th layer of the neural network, the th j The first neuron p Output variables of the data, The value is set to 1.5; and the learning parameters of the BP neural network model are adjusted according to the difference between the BP neural network model and the true value of the molecular weight characteristics of the corresponding training samples, so as to train the BP neural network model; and Based on the difference between the target value and the predicted value of the molecular weight characteristic, the process parameters are adjusted to obtain an impact-resistant polypropylene product that meets the target value.
2. The prediction method as described in claim 1, characterized in that, The at least one first raw material includes propylene and hydrogen, and the at least one second raw material includes propylene, ethylene, and hydrogen. The target and predicted values of the molecular weight characteristics correspond to at least one of the following: first number-average molecular weight Mn_PP, first weight-average molecular weight Mw_PP, first molecular weight distribution index PDI_PP, first mass percentage PP_per; second number-average molecular weight Mn_iPP, second weight-average molecular weight Mw_iPP, second molecular weight distribution index PDI_iPP; third number-average molecular weight Mn_aPP, third weight-average molecular weight Mw_aPP, third molecular weight distribution index PDI_aPP; fourth number-average molecular weight Mn_EPR, fourth weight-average molecular weight Mw_EPR, fourth molecular weight distribution index PDI_EPR, fourth mass percentage EPR_per; first ethylene molar content E_mole_EPR; fifth number-average molecular weight Mn_EPS, fifth weight-average molecular weight Mw_EPS, fifth molecular weight distribution index PDI_EPS, fifth mass percentage EPS_per; second ethylene molar content E_mole_EPS; and third ethylene molar content E_in_IPC in the polymer.
3. The prediction method as described in claim 1, characterized in that, Before sequentially inputting each training sample from the historical sample dataset into the BP neural network model, the prediction method further includes the following steps: The historical sample data in the aforementioned historical sample dataset are normalized to normalize their values to between -1 and 1: in, It is the first i The first neuron p Normalized variables of historical sample data, It is the first i The first neuron p The original variables of historical sample data, It is the first in the historical sample dataset. i The minimum value of the variable for each neuron. It is the first in the historical sample dataset. i The maximum value of the variable for each neuron.
4. The prediction method as described in claim 1, characterized in that, The step of sequentially inputting each training sample data from the historical sample dataset into the BP neural network model to obtain the corresponding molecular weight characteristic prediction value includes: The historical sample dataset is randomly divided into a training set and a test set according to a preset ratio; The historical sample data in the training set is defined as the training sample data; and The training sample data of each group in the training set are sequentially input into the BP neural network model to obtain the corresponding molecular weight characteristic prediction values.
5. The prediction method as described in claim 4, characterized in that, It also includes the following steps: The historical sample data in the test set is defined as test sample data; In response to the completion of training the BP neural network model, the test sample data of each group in the test set are sequentially input into the BP neural network model to obtain the corresponding molecular weight characteristic test values; and The effectiveness of the BP neural network model is verified by comparing the regression curves of each test value with the true values of the corresponding test samples.
6. The prediction method as described in claim 1, characterized in that, The number of neurons in the input layer The number of neurons in the output layer is determined based on the input dimension of the training sample data in each group. The number of neurons in the hidden layer is determined based on the output dimension of the predicted molecular weight characteristics, and the number of neurons in the hidden layer is based on the number of neurons in the input layer. The number of neurons in the output layer and the number of groups of the training sample data Sure: in, The value depends on the type of molecular weight characteristic to be predicted, and ranges from 4 to 10.
7. The prediction method as described in claim 1, characterized in that, The step of adjusting the learning parameters of the BP neural network model based on the difference between its molecular weight characteristics and the true values of the corresponding training samples to train the BP neural network model includes: Calculate the mean square error between the predicted and true values of the molecular weight characteristics of each training sample: in, It is the first j The first neuron p The true value of each training sample. It is the first j The first neuron p The predicted values of each training sample; and The Adam optimizer is used to perform backpropagation on the BP neural network model based on the mean squared error until the training of the BP neural network model is completed.
8. The prediction method as described in claim 7, characterized in that, The step of using the Adam optimizer to perform backpropagation on the BP neural network model based on the mean squared error until the training of the BP neural network model is completed includes: Set the maximum number of iterations for training the BP neural network model. Indicator parameters And a patience parameter to prevent overfitting during training, wherein the value of the patience parameter is determined according to the formula... Sure, The number of nodes is the input. The indicator parameter represents the number of nodes to be output. The initial value is 0; Response to the current round number The maximum number of iterations was not reached. And the indicated parameters If the first moment estimate of the Adam optimizer is less than or equal to the patience parameter, then... and second-order moment estimation These are combined to update the weights of the connections between neurons and the thresholds of each neuron in the BP neural network model: in, , It is the first The updated weights and thresholds in each round. , It is the first The weights and thresholds of the wheel, It's the learning rate. It is a zero-prevention constant, the first-order moment estimate of the Adam optimizer. The expression is: in, and These represent the first-order moment estimates before and after the iteration, respectively. The parameter is the exponentially weighted average parameter. Represents weight or threshold gradient, The second-order moment estimation of the Adam optimizer The expression is: in, and These are the second-order moment estimates before and after the iteration, respectively. The parameter is the exponentially weighted average parameter. Representative weight or threshold The squared gradient; Determine whether the mean square error of the current round has increased; if it has, adjust the indicated parameter. Increment by 1, otherwise set the indicated parameter to zero. ;as well as Response to the current round number Reaching the maximum number of iterations , or the indicated parameter If the value is greater than the patience parameter, the training of the BP neural network model is considered complete.
9. A system for predicting the microstructure of impact-resistant polypropylene, characterized in that, include: Memory, on which computer instructions are stored; as well as A processor, connected to the memory, and configured to execute computer instructions stored in the memory to implement the method for predicting the microstructure of impact-resistant polypropylene as described in any one of claims 1 to 8.
10. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instructions are executed by the processor, the method for predicting the microstructure of impact-resistant polypropylene as described in any one of claims 1 to 8 is implemented.