Method for predicting properties of grafted polypropylene for high-voltage power cable

Abstract

The present invention relates to the field of organic polymers. Specifically disclosed is a method for predicting the properties of grafted polypropylene for a high-voltage power cable. In the method of the present invention, a property prediction model for grafted polypropylene is constructed on the basis of a multi-layer perceptron model and a random forest model. The model solves the defect that a conventional machine learning model mainly relies on a large amount of data and is not suitable for a small data set. Experimental data is used to quantitatively establish a relationship between a grafted structure and mechanical properties, the amount of data required is small, the data is easy to obtain, and the prediction accuracy is high, thereby providing a reliable grafted group for a grafted polypropylene material used for a high-voltage power cable.

Description

Method for predicting properties of graft polypropylene for high-voltage power cable TECHNICAL FIELD

[0001] The present application belongs to the field of organic polymers, and particularly relates to a method for predicting properties of graft polypropylene for high-voltage power cable. BACKGROUND

[0002] The commonly used insulation material in high-voltage power cable (cable with voltage level greater than or equal to 6 kV and less than 500 kV in power system) is cross-linked polyethylene, and a large amount of by-products produced in the production process of cross-linked polyethylene will reduce its electrical properties. Compared with cross-linked polyethylene, polypropylene has high temperature resistance and excellent electrical properties, and is gradually replacing cross-linked polyethylene and being applied to high-voltage power cable due to its recyclable and environmentally friendly characteristics. However, in terms of mechanical properties, the flexibility of polypropylene material is not as good as that of polyethylene material due to the introduction of methyl side groups in polypropylene. In the case where the material needs to be bent, such as the position passing through the grounding interface, the cable of polypropylene material is more difficult to bend. Traditionally, the incorporation of elastomers in the polypropylene matrix will make the composite material softer, but the introduction of elastomers will increase the risk of interface separation and lead to the decline of electrical properties. The chemical grafting method can change the molecular structure and space free volume by introducing polar side groups on the side chain of polypropylene, thereby improving its mechanical properties. At the same time, the polar side chain groups can adsorb carriers to form deep traps at the interface between polypropylene and grafting groups, thereby improving the electrical properties of polypropylene material.

[0003] Chemically grafted modified polypropylene is still in its infancy, and it is difficult to obtain large-scale mechanical property data of grafted polypropylene material in the short term due to the actual cost, experimental conditions and the complexity of chemical grafting groups. Studies have shown that some reliable data can be obtained quickly and at low cost by referring to machine models to guide actual research and application. The machine model built is the key to prediction calculation, but the traditional machine model built has the defects of large amount of required data and low reliability of prediction results, and cannot be directly applied to the mechanical property research of chemically grafted polypropylene modification. SUMMARY

[0004] In view of the defects of the above machine model that the amount of required data is large and the reliability of prediction results is low, the present application provides a method for predicting properties of graft polypropylene for high-voltage power cable.

[0005] To achieve the above purpose, the following technical solutions are specifically included:

[0006] The method for predicting properties of graft polypropylene for high-voltage power cable comprises the following steps:

[0007] (1) obtaining a dataset, the dataset comprising a plurality of sample data, each sample data comprising at least one of molecular property information, atomic property information, and performance parameter of the graft polypropylene;

[0008] (2) based on the dataset, first using the glass transition temperature as the sample data to input a multilayer perception model for pre-training, after the pre-training is completed, freezing all the parameters of the hidden layers in the multilayer perception model, replacing the last layer of the multilayer perception model with a random forest model, and constructing a prediction model of the mechanical property of the graft polypropylene; and then based on the dataset, using the prediction model of the mechanical property of the graft polypropylene for transfer training to obtain a predicted mechanical property value of the graft polypropylene;

[0009] In the multilayer perception model, a ReLU activation function, a square loss function for optimization, and a Dropout function for avoiding overfitting are used.

[0010] In an embodiment, the graft polypropylene is prepared by initiating a grafting reaction of a monomer and polypropylene under heating conditions using an initiator, and the monomer includes at least one of styrene, vinyl triethoxysilane, glycidyl methacrylate, vinyl pyridine, ethylene carbazole, styrene-maleic anhydride, methyl methacrylate, N-vinyl pyrrolidone, methyl acrylate-acrylic acid, vinyl acetate, vinyl triethoxysilane, maleic anhydride, and vinyl imidazole.

[0011] In an embodiment, the monomer split chemical structure fragment includes at least one of the following chemical structural formulae:

[0012] In an embodiment, the multilayer perception model includes a neural network input layer, an output layer, a hidden layer, and a number of hidden layer neurons, the number of layers of the hidden layer is 4, and the dimensions are 300, 50, 20, and 4, respectively.

[0013] In an embodiment, the performance of the graft polypropylene includes at least one of a heat distortion temperature, a bending modulus, a room temperature impact strength, and a bending strength.

[0014] In an embodiment, in the pre-training and the transfer training, 80% of the dataset is used as a training set, and 20% is used as a test set.

[0015] In an embodiment, in the transfer training, a k-fold method is used to increase the number of training sets, wherein k of the k-fold method is 5.

[0016] In an embodiment, the sample data comprises dielectric constant, band gap, CBM energy level, VBM energy level, mass density, glass transition temperature, melting temperature, molecular or atomic descriptors; the language used by the molecular or atomic descriptors is Python.

[0017] In an embodiment, the molecular or atomic descriptors comprise at least one of atomic mean mass, heavy atom count, density of NH functional groups, density of OH functional groups, hydrogen bond acceptor density, hydrogen bond donor density, heteroatom density, valence electron density, density of amide bonds, ring count, density of ring structures, density of rotatable bonds, density of SP3 hybridized carbon atoms, density of SP2 hybridized carbon atoms, density of SP hybridized carbon atoms, density of bridgehead atoms, density of spiro atoms, Labut ASA descriptors for polymer monomers, MOE-like descriptors described using partial charges and surface area, MOE-like descriptors described using MR and surface area, MOE-like descriptors described using LogP and surface area, hybrid EState-VSA descriptors, molecular quantum numbers for polymer monomers, Chi indices for polymer monomers, Hall Kier Alpha descriptors describing molecular topological information, topological molecular polar surface area, MolLogP descriptors for polymer monomer molecules, Kappa shape index for polymer monomer molecules, Balaban J descriptors for polymer monomer molecules, a topological index quantifying the “complexity” of a monomer molecule, Ipc descriptors describing monomer molecule topological information, descriptors describing monomer molecule flexibility, ratio of number of backbone atoms in a polymer to total number of atoms in the polymer, ratio of shortest distance between polymer branches to length of backbone, density of structural fragments in a polymer and in a backbone from the MACC library, density of structural fragments in a polymer and in a backbone from the Morgan library, density of structural fragments in a polymer and in a backbone, density of different atoms in a polymer backbone, branches, and monomers.

[0018] In an embodiment, the Shapley coefficient is used to evaluate the contribution of the functional groups in the grafted polypropylene to the flexural modulus in the grafted polypropylene, with the density of the functional groups in the grafted polypropylene as the input of the prediction model of the mechanical properties of the grafted polypropylene, and the flexural modulus of the grafted polypropylene as the output of the prediction model of the mechanical properties of the grafted polypropylene.

[0019] In an embodiment, the sample data of the CBM, VBM energy level, and charge traps of the grafted polypropylene and pure polypropylene are used as a data set, input into the prediction model of the mechanical properties of the grafted polypropylene, output the flexural modulus, and determine the correlation between the depth of the charge traps and the flexural modulus, and the chemical structure fragment fragments of the grafted polypropylene.

[0020] In an embodiment, the chemical structure fragments of the grafted polypropylene include at least one of the following chemical structural formulas:

[0021] Compared with the prior art, the present application has the following beneficial effects:

[0022] (1) The traditional machine learning model mainly relies on a large amount of data, which is not suitable for small data sets;

[0023] (2) The relationship between the graft structure and the mechanical properties is quantitatively established from the experimental data, the transfer learning model is applied to analyze the mechanical properties of the grafted polypropylene insulation material system, the required data amount is small, the data is easy to obtain, the prediction accuracy is high, and a reliable graft group is provided for the polypropylene material in the high-voltage cable;

[0024] (3) The correlation between the glass transition temperature and the mechanical properties of polypropylene is established, and the existence of traps is considered separately, the influence of trap depth is combined, and the graft structure which is soft and has a deep trap level in the limited chemical space is screened by using the prediction model of the present application; the graft side chain structure is divided into fragment blocks. BRIEF DESCRIPTION OF DRAWINGS

[0025] Fig. 1 is a flowchart of a prediction method for the properties of grafted polypropylene.

[0026] Fig. 2 is a schematic diagram of the grafting reaction principle of grafted polypropylene.

[0027] Fig. 3 is a distribution of the bending modulus of five grafted polypropylenes.

[0028] Fig. 4 is a schematic diagram of an MLP model.

[0029] Fig. 5 is a schematic diagram of pre-training of an MLP model.

[0030] Fig. 6 is a schematic diagram of the relationship between the predicted glass transition temperature and the measured bending modulus.

[0031] Fig. 7 is a schematic diagram of a transfer learning prediction model of the present application.

[0032] Fig. 8 is a prediction result of the glass transition temperature Tg, and the correlation coefficient R of the test set is 0.834.

[0033] Fig. 9 is a prediction result of (a) heat distortion temperature value, (b) bending modulus value, (c) room temperature impact strength, and (d) bending strength.

[0034] Fig. 10 is nine structure fragment fragments split in monomers.

[0035] Fig. 11 is three structure fragment fragments with low bending modulus and high charge trap depth value which are preferably selected. DETAILED DESCRIPTION

[0036] In order to better illustrate the purposes, technical solutions and advantages of the present application, the present application will be further described below through specific examples. The test methods used in the examples and / or comparative examples are all conventional methods unless otherwise specified; the materials, reagents, etc. used are all commercially available unless otherwise specified.

[0037] Example 1

[0038] A method for predicting the properties of grafted polypropylene for high-voltage power cables, a flowchart is shown in Figure 1, comprising the following steps:

[0039] (1) The experimental test data set is obtained as follows:

[0040] The grafted polypropylene is prepared by grafting reaction of a certain monomer and polypropylene under the action of initiator benzoyl peroxide at a temperature of 100°C.

[0041] The monomers include styrene, vinyl triethoxysilane, glycidyl methacrylate, vinyl pyridine, ethylene carbazole, styrene-maleic anhydride, methyl methacrylate, N-vinyl pyrrolidone, methyl acrylate-acrylic acid, vinyl acetate, vinyl triethoxysilane, maleic anhydride, and vinyl imidazole, a total of 11 kinds; different monomers contain different functional groups and functional group densities, molecular structures, and molecular chain backbone atom arrangements and branched chain atom arrangements. The chemical structures of each monomer are shown in Table 1.

[0042] Table 1

[0043] Each monomer is reacted with polypropylene and an initiator, and each monomer is used to prepare grafted polypropylene with a monomer volume fraction of 2% and 5%, respectively, a total of 22 kinds of grafted polypropylene are prepared;

[0044] The principle of the grafted polypropylene reaction is as follows: in the grafting reaction, the initiator benzoyl peroxide (BPO) is first decomposed by heat to form primary free radicals, which have strong chemical activity and can capture tertiary hydrogen atoms on the molecular chain of polypropylene (PP); the BPO initiator forms active sites on the PP backbone, producing PP macromolecular radicals; the carbon-carbon double bonds in the grafting monomer react with the PP macromolecular radicals, and are grafted onto the PP molecular chain, as shown in Figure 2, wherein R in the figure is a monomer group.

[0045] The 23 kinds of grafted polypropylenes and pure polypropylenes prepared above were tested for heat distortion temperature, flexural modulus, room temperature impact strength, and flexural strength performance, and 23 sets of test data were collected as the experimental test data set of the present example. The test conditions were the same during testing of different materials. The MTS CMT4304 microcomputer-controlled electronic universal testing machine was used for mechanical property testing. The punching machine was used to obtain dumbbell-shaped tensile test samples. Tensile testing was performed at room temperature at a tensile rate of 20 mm / min until the sample broke, and the flexural strength and elongation at break of the sample were obtained. Due to the dispersion of the tensile test results, five tests were performed for each sample, and the average value was calculated based on these tests.

[0046] Figure 3 shows the flexural modulus of five polypropylene graft structures. It can be seen that the flexural modulus of the system changes significantly with the introduction of graft groups. Among them, the polypropylene material grafted with vinyl triethoxysilane has the best flexibility, and the flexibility of other structures with polar graft groups all shows a downward trend.

[0047] Due to the long and rotatable bond of the siloxane bond (-Si-O-), compared with the pure polypropylene system, the grafting of vinyl triethoxysilane will reduce the flexural modulus of the structure. At the same time, the polar groups of the side chains such as styrene, glycidyl methacrylate, vinylpyridine, and vinyl acetate increase the internal rotation activation energy and intermolecular forces. The increase in the hindrance of internal rotation leads to an increase in the hardness of the grafted polypropylene structure, making it more difficult to bend.

[0048] However, it should be noted that the grafted polypropylene polymer is essentially a mixture, not a perfect grafted monomer structure, and the grafted segments will exist in copolymerization. Therefore, compared with pure polymers, the influence of graft groups on the mechanical properties of grafted polypropylene systems is smaller. However, since all samples are in the grafted polypropylene system, the data can be compared.

[0049] At the same time, since the side group structure of the grafted polypropylene used in the model training of the present example (i.e. obtained by grafting monomers) only contains a small number of structural fragments, it is only a "drop in the ocean" among the common functional groups in the entire chemical space. The small number of structural fragment types makes it difficult to directly apply the established rapid prediction model to a larger chemical space. In order to avoid blindly generalizing the results / experience trained on a small-scale data set to a larger chemical space and to avoid existing prediction pitfalls, the grafted structures in the test data set are split into 9 common structural fragment pieces (Figure 10). Figure 10 shows the 9 structural fragment pieces. Since these chemical structure fragments come from the small-scale data set used to obtain experience, these structure fragments are spliced to form the chemical space structure of the grafted polypropylene finally used for prediction.

[0050] (2) Constructing a prediction model of the mechanical properties of the grafted polypropylene, the prediction model of the mechanical properties of the grafted polypropylene including a multilayer perception (MLP) model and a random forest model, based on the experimental test data set of step (1) above and the data set collected in (2-1), as the data set of the prediction model, converting the data set into a language that can be operated, inputting into the prediction model for pre-training and migration training, and outputting the mechanical property value of the grafted polypropylene with a certain specific chemical structure.

[0051] Specifically comprising the following steps:

[0052] (2-1) Collecting data set:

[0053] The collected data set is shown in Table 2, wherein the four parameters of dielectric constant, band gap, CBM energy level and VBM energy level related to the grafted polypropylene are calculated parameters, and the data are from Khazana database and high-throughput calculation. The three performance data of mass density, glass transition temperature and melting temperature represented by the laboratory are mainly from PolyInfo database. In the process of selecting experimental performance data in the database, the standard for selecting different configuration performance data of polypropylene materials (such as isotactic, syndiotactic, atactic and other configurations of polypropylene materials) is to select the more common and more widely used structure performance parameters as much as possible. At the same time, there are various measurement methods to obtain the glass transition temperature and melting temperature performance data of polypropylene materials, and the data selected in the data set are measured based on conventional measurement methods. At the same time, some obviously very irregular parameter performance values are also artificially excluded from the data set.

[0054] Table 2 Data set type and quantity

[0055] Descriptors of grafted polypropylene: The fingerprint database of grafted polypropylene contains two parts of descriptors. One part of descriptors: This part of descriptors is evolved from the traditional small organic molecule fingerprint descriptors of polypropylene, which contains the density of different structural unit fragments in polypropylene, and the structural unit fragments are mainly collected from the public MACCS public key and Morgan; this part also contains some traditional molecular fingerprint characteristic descriptors, such as BalabanJ, TPSA, etc. Another part of descriptors, this part of descriptors represents the fingerprint characteristics mainly aiming at the unique characteristics of polypropylene, such as the length of the main chain, the length of the longest branch, the maximum distance between the branches, the density of the structural unit fragments in the main chain, etc. The two parts of descriptors together constitute the fingerprint characteristic library of polypropylene structure, a total of 812 descriptors, part of the fingerprint characteristic descriptors are shown in Table 3. According to the characteristics of grafted polypropylene, the chemical structure of various types of grafted polypropylene is compiled into symbols that can be operated to form the fingerprint atlas data set of grafted polypropylene. In this embodiment, the language used by the descriptor extraction program of the fingerprint characteristics is Python, and most of the fingerprint characteristic descriptors use the Python extension package RDkit for extraction.

[0056] Based on the above collected grafted polypropylene fingerprint database (chemical structure information) and parameter performance, the correlation between the chemical structure of grafted polypropylene and the dielectric constant, band gap, CBM energy level and VBM energy level related to grafted polypropylene can be constructed. When a certain chemical structure of grafted polypropylene or its structural fragments is determined, its specific parameter performance can be determined.

[0057] Chemical structure, related dielectric constant, band gap, CBM energy level and VBM energy level are all characteristic parameters of the macro mechanical properties of materials, and based on the above characteristic parameters, the mechanical properties of a chemical structure of a substance can be predicted.

[0058] Table 3

[0059] (2-2) Pre-training before transfer learning:

[0060] In the machine learning model, transfer learning is to improve the learning in the new task by transferring the knowledge from the completed related task, which can be trained on the basis of the completed training model to achieve the purpose of rapid convergence, therefore, in order to construct the transfer learning mode of the present application, the performance parameters related to the mechanical properties of grafted polypropylene need to be selected as the pre-training performance.

[0061] The glass transition refers to the transition between the glass state and the high-elastic state of an amorphous polymer, and the glass transition also occurs in the amorphous region of a crystalline or semi-crystalline polymer, which has a great influence on the performance of the polymer, especially the mechanical properties. The temperature at which the glass transition occurs is called the glass transition temperature, which is a characteristic temperature of the polymer. The glass transition temperature is greatly related to the flexibility of the polymer segment. Generally speaking, the softer the polymer, the lower the glass transition temperature; the greater the rigidity of the segment, the higher the glass transition temperature. Therefore, based on the relationship between the glass transition temperature and the mechanical properties, the glass transition temperature performance (with a large amount of data) is selected as a performance parameter related to the mechanical properties of the grafted polypropylene in this embodiment, pre-training is performed in the migration learning, and the mechanical properties (with a small amount of data) are migrated and trained based on the trained glass transition temperature prediction model, which can reflect the reliability of the selected trained glass transition temperature model.

[0062] Model for pre-training glass transition temperature performance is constructed:

[0063] The MLP neural network model is used as the model for pre-training the glass transition temperature performance, the MLP includes an input layer, an output layer, a hidden layer and a number of hidden layer neurons, the hidden layer includes 4 layers, and the dimensions are 300, 50, 20 and 4 respectively; the ReLU activation function, the square loss function (Loss i ) is used for optimization, and the Dropout function is used to avoid overfitting:

[0064] (a) MLP model

[0065] The MLP model can be considered as a generalized linear model, and the schematic diagram is shown in FIG. 4. For the linear model,

[0066] For the MLP, x and y represent the input layer and the output layer respectively, w is the weight, b is the bias, and a set of h modules is added between x and y, which is called the hidden layer;

[0067] The relationship between y, h and x is:

[0068] h[1] = f(w[1,1]x[1] + w[2,1]x[2] + w[3,1]x[3] + b[0]) (Formula 2);

[0069] h[2] = f(w[1,2[x[1] + w[2,2]x[2] + w[3,2]x[3] + b[1]) (Formula 3);

[0070] y = v[1]h[1] + v[2]h[2] + s (Formula 4);

[0071] where f(·) is some non-linear function, called activation function;

[0072] In this model, the ReLU activation function is used: where f(x) is the ReLU function, and x is the input value of the hidden layer neuron;

[0073] In this embodiment, the input layer, output layer, hidden layer and the number of hidden layer neurons of the MLP are determined, and the hidden layer is 4 layers, and the number of neurons of each hidden layer is 300, 50, 20 and 4 respectively;

[0074] (2) Constructing a square loss function:

[0075] Let the prediction error of the MLP for the sample be the smallest, and for any one sample, the error can be defined as: The above formula becomes

[0076] Loss=[y-(v1f(w 11 x1+w 21 x2+w 31 x3+b0)+v2f(w 12 x1+w 22 x2+w 32 x3+b1)+s)] 2 (Formula 7);

[0077] If there are N samples in the training set, the errors of each sample need to be added and minimized: In the above formula, the variables to be optimized are v, w, b and s, and it is an unconstrained problem, so it can be solved by gradient-based algorithms. The classical gradient-based algorithms can refer to the principles of gradient-based algorithms: steepest descent method, Newton method and quasi-Newton method. In order to improve the training efficiency, people generally choose the steepest descent method which only needs the first-order gradient as the basic algorithm.

[0078] Before calculation, we first need to clarify that the gradient here refers to the gradient of the loss function Loss with respect to the optimization variables v, w, b and s. Take the gradient of Loss with respect to w as an example to explain the solving process. Still use the above MLP example, and write the expression in matrix form:

[0079] According to the chain rule of differentiation:

[0080] This is what we usually call: error back propagation, which can be understood as the process of gradient calculation is derived from back to front.

[0081] The above gives the gradient solving process for a sample. To get accurate gradient values, we need to go through the above process for each sample. If our training sample number is large, it will cause the gradient calculation process to be very slow. If we change our thinking to process, we only calculate the gradient value of one sample each time, and then take the value as the final gradient value, then the result is that the gradient calculation process is fast but the value is not very accurate. Therefore, a part of the sample is selected each time, and the gradient value is calculated as the gradient value of the whole. By adjusting the number of samples, the efficiency and accuracy of the calculation can be considered.

[0082] With the gradient, we review the basic iteration formula of the gradient class algorithm:

[0083] θ k+1 =θ k -η k ·d k (Formula 13);

[0084] where θ represents the optimization variable, η k and d k are the iteration step (MLP is more referred to as "learning rate") and iteration direction of the kth time, respectively.

[0085] In the steepest descent method, the basic formula is:

[0086] (3) Dropout function formula:

[0087] where z represents the input vector; l represents the layer; y represents the output vector; W represents the weight; b represents the bias; f represents the activation function; and r represents the value after the Bernoulli distribution.

[0088] (b) Pre-training:

[0089] Based on the glass transition temperature sample data in the data set of step (2-1) above and the grafted polypropylene fingerprint database (chemical structure fragments) capable of determining chemical structure information as the data set, the data set fingerprint recognition is converted into a language input capable of calculation. The pre-trained MLP model is constructed, and the output value is the glass transition temperature of a certain grafted polypropylene with a certain chemical structure. 80% of the data set is used as the training set, and 20% is used as the test set. Based on the results of pre-training, the glass transition temperature of a certain grafted polypropylene composed of chemical structure fragments can be determined. Therefore, based on the results of pre-training, the predicted glass transition temperature corresponding to the grafted polypropylene structure after the pure PP is grafted with a monomer is obtained, and the predicted glass transition temperature is taken as the x-axis data. The flexural modulus of the grafted polypropylene measured in step (1) above is taken as the y-axis data, and the result is shown in FIG. 6.

[0090] For linear polar grafting groups (such as VP, VAC, GMA, etc. structure), as the predicted glass transition temperature increases, the flexural modulus of the grafted polypropylene structure also increases, which means that the hardness of the grafting system is higher. At the same time, the ring structure in the grafted structure (such as VK, NVP, etc.) causes the glass transition temperature to increase, while the flexural modulus decreases, and the system becomes soft. This may be related to the increase in free volume caused by the cyclic structure. Therefore, it can be known that the mechanical properties, glass transition temperature and chemical structure of the grafted polypropylene are related.

[0091] (c) Training small amount of mechanical property data:

[0092] After the above pre-training is completed, the parameters in the hidden layer are frozen (all frozen), and the last layer of the above model is replaced with a random forest layer. Among them, the MLP includes an input layer, a hidden layer and an output layer, and the last layer is the output layer, so the output layer is replaced with a random forest layer, and the final output layer is the output result of the random forest model as the output layer, as the model of the transfer learning training.

[0093] Based on the above collected experimental test data set (mechanical properties) and the data set collected in step (2-1) (chemical structure information) as the overall transfer training data set, the data set fingerprint is identified as a language that can be operated, input into the transfer learning training model for transfer learning training, and the output value is the mechanical property of the grafted polypropylene of a certain determined chemical structure. Among them, 80% of the data set is used as the training set, and 20% is used as the test set. In the training process, the k-fold method (k=5) is adopted to increase the number of training sets. In addition, a 10% random removal process parameter is also adopted in the training process to avoid overfitting phenomenon.

[0094] The pre-training and transfer learning training process model is shown in Figures 5 and 7.

[0095] Based on the results of the transfer training, the mechanical properties of a certain grafted polypropylene composed of chemical structure fragments can be determined.

[0096] Regarding the relationship between k-fold method and random forest model: Machine learning methods often cannot model the data directly because they learn specific features of the training set that do not exist in the test set. Because the fair performance of the model is related to its role, simply dividing the data into a training set and a test set cannot truly understand the performance of the model. So the general idea is to divide the data set into k parts, k-1 for training, and one for validation / testing. In this way, we can divide the data set into 5 folds, so 4 folds will be used to train the model, and the remaining fold will be used to evaluate the performance of the model. If such partitioning is performed, the fold position used to evaluate the model needs to be changed each time and repeated 5 times.

[0097] The final output result of the random forest (RF) prediction model: f(x) is the output prediction value, M is the number of trees of the random forest (RF) prediction model, fm(x) is the prediction result of the mth decision tree, and in the prediction model of the present application, different input quantities correspond to different output quantities. f(x) can be one of the heat distortion temperature, bending modulus, room temperature impact strength, and bending strength.

[0098] (d) Mechanical property prediction results of the grafted polypropylene system:

[0099] After training, the data points of the glass transition temperature are shown in Figure 8. The correlation coefficient of the test set is 0.834, and it can be seen that the training result has good consistency. The mechanical property parameters involved in the prediction include heat distortion temperature, bending modulus, impact strength at room temperature, and bending strength, and the results are shown in Figure 9. In the prediction results, the correlation coefficients of the test set of the four performance data are 0.78, 0.89, 0.57, and 0.87, respectively. Except for the poor prediction result of the impact strength at room temperature, the prediction accuracy of the other three mechanical property parameters is high.

[0100] In order to quantitatively study the influence of different structural fragments on the bending modulus performance, on the basis of the completed training of the transfer learning model, the Shapley coefficients of different chemical structure components were calculated by taking the density of each component functional group as input, so as to analyze the contribution of different functional group structures as side groups to the bending modulus. Among them, the test results show that the Shapley coefficients of siloxane bond functional group and pyrrolidone functional group are -0.28 and -0.11, which are the lowest in the selected structural fragments, which means that the siloxane bond functional group and the pyrrolidone functional group can make the whole become softer. The highest is the pyridine functional group and the ester bond, and the Shapley coefficients of the two are 0.17 and 0.09. This is consistent with the above experimental test results, and also indirectly proves the accuracy of the prediction model.

[0101] Wherein, the Shapley coefficient is calculated by using the Shapley value method, including:

[0102] Data i has (|S|-1)! kinds of orders when participating in set S, |S| represents the number of data contained in set S, and the order of the remaining (n-|S|) data has (n-|S|)! kinds, and the different order combinations of data i participating in are divided by the random order combination of n data, which is the weight of the benefit that data i should get for the whole set, recorded as [(|S|-1)!(n-|S|)!] / (n!);

[0103] Data i participates in different set S for its own participation in data optimization, recorded as [v(S)-v(S\{i})], then the benefit that data i gets from the optimization of the whole v(N) is:

[0104] In the formula, S\{i} represents a set obtained by deleting element i from set S, is the Shapley coefficient;

[0105] Wherein, when the functional group density is taken as the input and the bending modulus is taken as the output, according to the Shapley value method, the Shapley coefficients of the siloxane bond and the pyrrolidone functional group are-0.28 and-0.11 respectively, and the Shapley coefficients of the pyridine functional group and the ester bond are 0.17 and 0.09 respectively, which indicates that the siloxane bond and the pyrrolidone functional group can improve the flexibility of the grafted polypropylene, and the pyridine functional group and the ester bond can reduce the flexibility of the grafted polypropylene;

[0106] In the above-mentioned constructed predicted chemical space structure of the grafted polypropylene (Figure 10), the CBM (conduction band minimum) / VBM energy level (electron / hole carrier trap) dataset of the grafted polypropylene material and the pure polypropylene material is selected, combined with the dataset of the charge trap, and input into the above-mentioned prediction model for prediction, and 3 groups of structure fragment fragments with soft (low bending modulus) and high charge trap depth are obtained (Figure 11). It can be seen that the structure of the siloxane bond provides the basis for flexibility, and at the same time, the pyrrolidone structure also provides a large amount of trap density to effectively capture electrons, so that the prediction model of the application can also accurately predict the ideal chemical space structure of the grafted polypropylene, which is used for subsequent synthesis of the grafted polypropylene.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit the protection scope of the present application, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the essence and scope of the technical solutions of the present application.

Claims

1. A method for predicting the properties of a grafted polypropylene for high voltage power cables, characterized in that, The method comprises the following steps: (1) obtaining a data set, the data set comprising a plurality of sample data, each sample data comprising at least one of molecular characteristic information, atomic characteristic information and performance parameters of the graft polypropylene; (2) based on the data set, first using the glass transition temperature as the sample data to input a multilayer perception machine model for pre-training, after the pre-training is completed, freezing all the hidden layers of the multilayer perception machine model, replacing the last layer of the multilayer perception machine model with a random forest model, and constructing a prediction model of the mechanical properties of the graft polypropylene; and then based on the data set, using the prediction model of the mechanical properties of the graft polypropylene for transfer training to obtain the predicted mechanical properties of the graft polypropylene; In the multilayer perception machine model, a ReLU activation function, a square loss function for optimization and a Dropout function for avoiding overfitting are used.

2. The method of predicting the properties of a graft polypropylene for high voltage power cables according to claim 1, characterized in that, The graft polypropylene is prepared by grafting reaction of monomers and polypropylene under the action of an initiator under heating conditions, and the monomers include at least one of styrene, vinyl triethoxysilane, glycidyl methacrylate, vinyl pyridine, ethylene carbazole, styrene-maleic anhydride, methyl methacrylate, N-vinyl pyrrolidone, methyl acrylate-acrylic acid, vinyl acetate, vinyl triethoxysilane, maleic anhydride and vinyl imidazole.

3. The method of predicting the properties of a graft polypropylene for high voltage power cables according to claim 2, characterized in that, The monomer split chemical structure fragments include at least one of the following chemical structures:

4. The method of predicting the properties of grafted polypropylene for high voltage power cables according to claim 1, characterized in that, The multilayer perception machine model comprises a neural network input layer, an output layer, a hidden layer and a number of hidden layer neurons, the number of layers of the hidden layer is 4, and the dimensions are 300, 50, 20 and 4 respectively.

5. The method of predicting the properties of grafted polypropylene for high voltage power cables according to claim 1, characterized in that, The performance of the graft polypropylene includes at least one of heat distortion temperature, bending modulus, room temperature impact strength and bending strength.

6. The method of predicting the properties of a grafted polypropylene for high voltage power cables according to claim 1, characterized in that, In the pre-training and the transfer training, 80% of the data set is used as a training set, and 20% is used as a test set.

7. The method of predicting the properties of a grafted polypropylene for high voltage power cables according to claim 1, characterized in that, In the transfer training, a k-fold method is used to increase the number of training sets, wherein k of the k-fold method is 5.

8. The method of predicting the properties of a grafted polypropylene for high voltage power cables according to claim 1, characterized in that, The sample data includes dielectric constant, band gap, CBM energy level, VBM energy level, mass density, glass transition temperature, melting temperature, molecular or atomic descriptor; the language used by the molecular or atomic descriptor is Python.

9. The method of predicting the properties of grafted polypropylene for high voltage power cables according to claim 7, characterized in that, The molecular or atomic descriptors include at least one of: atomic mean mass, heavy atom count, NH functionality density, OH functionality density, hydrogen bond acceptor density, hydrogen bond donor density, heteroatom density, valence electron density, amide bond density, ring count, ring structure density, rotatable bond count, SP3 hybridized carbon atom count, SP2 hybridized carbon atom count, SP hybridized carbon atom count, bridgehead atom count, spiro atom count, polymer monomer LabutASA descriptors, MOE-like descriptors described using partial charge and surface area, MOE-like descriptors described using MR and surface area, MOE-like descriptors described using LogP and surface area, hybrid EState-VSA descriptors, polymer monomer molecular quantum numbers, polymer monomer Chi indices, Hall Kier Alpha descriptors describing molecular topological information, topological polar surface area, polymer monomer molecular MolLogP descriptors, polymer monomer molecular Kappa shape index, polymer monomer molecular Balaban J descriptors, a topological index quantifying the "complexity" of the monomer molecule, Ipc descriptors describing monomer molecule topological information, descriptors describing monomer molecule flexibility, ratio of backbone atoms to total atoms in the polymer, ratio of the shortest distance between polymer branches to the length of the backbone, density of structural fragments in the MACC library within the polymer and within the backbone, density of structural fragments in the Morgan library within the polymer and within the backbone, density of structural fragments within the polymer and within the backbone, density of different atoms within the polymer backbone, branch, and monomer.

10. The method of predicting the properties of a grafted polypropylene for high voltage power cables according to claim 1, characterized in that, The method comprises at least one of the following A-B: A, using the density of the functional groups of the grafted polypropylene as the input of the prediction model of the mechanical properties of the grafted polypropylene, and the bending modulus of the grafted polypropylene as the output of the prediction model of the mechanical properties of the grafted polypropylene, and then using the Shapley coefficient to evaluate the contribution of the functional groups in the grafted polypropylene to the bending modulus in the grafted polypropylene; B, using the sample data of the CBM, VBM energy level, and charge trap of the grafted polypropylene and pure polypropylene as the data set, inputting the prediction model of the mechanical properties of the grafted polypropylene, outputting the bending modulus, and determining the correlation between the charge trap depth and the bending modulus and the chemical structure fragments of the grafted polypropylene.

Images