Tin-based flame-retardant polymer composite material based on artificial intelligence and design method thereof

Abstract

The invention discloses a tin-based flame-retardant polymer composite material based on artificial intelligence and a design method thereof. The method comprises the following steps: S1, constructing a flame-retardant property data set of the flame-retardant polymer composite material; s2, constructing a random forest regression model; s3, screening a flame-retardant polymer composite material; and S4, verifying the performance of the candidate flame-retardant polymer composite material. The tin-based flame-retardant polymer composite material with excellent flame retardant property is designed by adopting an artificial intelligence design strategy, the flame-retardant polymer composite material with specific LOI and TS properties at the same time can be obtained, and the flame-retardant polymer composite material with multiple synchronous and excellent properties can be rapidly, efficiently and accurately designed; and the requirements of industrialization development of the tin-based flame-retardant polymer composite material are met.

Description

Technical Field

[0001] The present invention relates to the technical field of flame-retardant polymer composite materials, and in particular to an artificial intelligence-based tin-containing flame-retardant polymer composite material and a design method thereof. Background Art

[0002] At present, flame-retardant polymer composites have excellent flame retardancy and can suppress smoke. They can reduce fire risks and safety hazards to a certain extent, so they have certain application potential in construction, transportation, aerospace and many other fields.

[0003] Tin-based flame-retardant polymer composites are a new type of material that combines tin-based flame retardants with polymer matrices and has excellent flame retardant properties and broad application prospects. Since tin-based flame-retardant polymer composites combine the excellent flame retardant properties of tin-based flame retardants with the light weight and high strength of polymer-based composites, they have broad application prospects. In particular, such materials will play an important role in fields with strict requirements on flame retardant properties, such as construction, automobiles, aerospace, and the electronics industry.

[0004] The flame retardant properties of tin-based flame-retardant polymer composites are generally determined by two aspects at the same time. The first is the flame retardancy of the polymer matrix itself, and the other is the flame retardancy of the added filler. The flame retardancy of the matrix itself is difficult to change. Therefore, in order to improve the flame retardant properties of flame retardants, adding flame retardant fillers to the matrix is ​​a common method to improve the flame retardancy of flame retardants. However, the addition of a large amount of fillers, especially the amount of hydroxide fillers, will greatly affect the mechanical properties of flame retardant materials. However, in some scenarios such as construction, aerospace and other fields, it is necessary to pay attention to both flame retardant properties and mechanical properties. Therefore, due to the huge chemical space of flame retardant polymer composites, it is facing great challenges to design flame retardant polymer composites with excellent flame retardant properties by relying solely on traditional trial and error methods. On the one hand, it requires a lot of resources to consume, and on the other hand, it takes a long time.

[0005] The current general artificial intelligence technology and commonly used decision-making support algorithms can handle tasks that have never been seen before, understand abstract concepts, and learn and adapt in new environments. For example, it can handle complex scientific concepts, create creative works at a human level, and transfer knowledge and skills between different fields. However, in the field of flame-retardant polymer composite material design technology, due to the large variety of raw materials and finished products of composite materials, large differences in properties, and the large number of process conditions in the preparation process, there are still many technical difficulties in the application of general artificial intelligence technology, and it is difficult to screen flame-retardant polymer composite materials with specific LOI and TS properties. Summary of the invention

[0006] In view of the deficiencies of the above-mentioned prior art, the object of the present invention is to provide a tin-containing flame-retardant polymer composite material based on artificial intelligence and a design method thereof, and to design a flame-retardant polymer composite material with excellent flame retardant properties based on an artificial intelligence design strategy of a machine learning method, so as to obtain a flame-retardant polymer composite material having both specific LOI and TS properties, and at the same time overcome the deficiencies of general artificial intelligence and traditional trial and error methods, and is simple, fast, low-cost and reliable in results, which can meet the needs of industrial development of tin-containing flame-retardant polymer composite materials.

[0007] To achieve the above object, the technical solution adopted by the present invention is:

[0008] A design method of a tin-based flame-retardant polymer composite material based on artificial intelligence, characterized in that it comprises the following steps:

[0009] S1: Constructing a flame retardant performance dataset for flame retardant polymer composites

[0010] First, collect flame retardant performance-mechanical performance data with corresponding relationships from the existing experimental data or test data of tin-based flame retardant polymer composite materials;

[0011] Then the matrix, content of flame retardant and process parameters for preparing flame retardant polymer composites are used as modeling features, and data standardization is performed on the features;

[0012] S2: Building a random forest regression model

[0013] First, a two-step method is used to screen feature subsets. Specifically, the variable autocorrelation method in the first step and the maximum correlation minimum redundancy combined with feature recursive elimination method and genetic algorithm in the second step are used to screen the random forest regression model features that correspond to the two targets synchronously.

[0014] Then, based on the multi-objective optimization strategy, feature subsets were selected to establish random forest regression models of LOI and TS, and the grid search method was used to optimize the random forest regression models of LOI and TS;

[0015] S3: Screening of flame retardant polymer composites

[0016] The LOI and TS random forest regression models were applied to the Pareto optimization process based on the NSGA-II algorithm. In this process, the combination of LOI and TS models was used as the fitness function in the genetic algorithm. When the maximum iteration number was reached, the genetic iteration was stopped to screen out candidate flame retardant polymer composites with excellent LOI and TS properties.

[0017] S4: Verify the performance of candidate flame retardant polymer composites

[0018] The experimental synthesis step S3 screens candidate flame retardant polymer composites, and characterizes the LOI and TS properties of the candidate flame retardant polymer composites to verify their conformity and reliability with the screening performance in step S3.

[0019] A tin-containing flame-retardant polymer composite material is prepared by using the aforementioned artificial intelligence-based design method for tin-containing flame-retardant polymer composite materials.

[0020] Compared with the existing technology, the present invention has the following beneficial effects:

[0021] 1. The artificial intelligence-based tin-containing flame-retardant polymer composite material and its design method provided by the present invention focus on using the artificial intelligence design strategy of the machine learning method to design a flame-retardant polymer composite material with excellent flame retardant properties, so as to obtain a flame-retardant polymer composite material with both specific LOI and TS properties, and at the same time overcome the shortcomings of general artificial intelligence and traditional trial and error methods. It is simple, fast, low-cost and reliable in results, and can quickly, efficiently and accurately design flame-retardant polymer composite materials with multiple excellent properties, which can meet the needs of industrial development of tin-containing flame-retardant polymer composite materials.

[0022] 2. The design method and material provided by the present invention overcome the defects of the existing trial-and-error technology in designing flame-retardant polymer composite materials with excellent flame retardant properties and mechanical properties, which are high cost and time-consuming, and provide a simple, fast, low-cost and reliable artificial intelligence design method for quickly designing flame-retardant polymer composite materials based on the Pareto algorithm; at the same time, it also overcomes the shortcomings of general artificial intelligence technology such as lack of specificity, inability to control intermediate processes and reliability of output results.

[0023] 3. The artificial intelligence design strategy based on the machine learning method of the present invention, in particular, needs to solve multi-objective optimization problems in the artificial intelligence design strategy to achieve the solution of specific technical problems and output reliable results. The present invention selects the Pareto optimization method based on the NSGA-II algorithm to carry out the intelligent design of flame-retardant polymer composite materials. At the same time, combined with the random forest algorithm, the results of different decision tree models are finally averaged as the final result to obtain reliable output results.

[0024] 4. The design method and materials provided by the present invention utilize Pareto optimization based on NSGA-II to quickly design a flame-retardant polymer composite material with two excellent target properties, and then conduct experimental synthesis and performance characterization on the designed flame-retardant polymer composite material. The cost is lower than that of the traditional trial and error method in terms of time, manpower and material resources. After actual testing, the cost of developing a new material according to the present invention can be controlled within 1 / 10 of that of the traditional method.

[0025] 5. The design method and material provided by the present invention find the Pareto front of two objectives through the Pareto optimization algorithm based on the NSGA-II algorithm. This NSGA-II algorithm can find the Pareto front where the two objectives are balanced. Therefore, this method can select candidate flame-retardant polymer composite materials with excellent LOI and TS. This method can effectively realize multi-objective optimization and solve the problem that both trial and error method and general artificial intelligence technology cannot realize multi-objective optimization.

[0026] The above is an overview of the technical solution of the invention. The present invention will be further described below in conjunction with specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a comparison chart of the leave-one-out cross-validation experimental value and the predicted value of the LOI random forest regression model in an embodiment of the present invention.

[0028] Figure 2 1 is a comparison chart of the test set experimental value and the predicted value of the LOI random forest regression model in an embodiment of the present invention.

[0029] Figure 3 This is a comparison chart of the leave-one-out cross-validation experimental value and the predicted value of the TS random forest regression model in an embodiment of the present invention.

[0030] Figure 4 This is a comparison chart of the test set experimental value and the predicted value of the TS random forest regression model in an embodiment of the present invention.

[0031] Figure 5 It is a Pareto front diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The technical solution of the present invention is further described in detail below in conjunction with the embodiments, but the protection scope of the present invention is not limited thereto. For researchers in this field, several changes and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the patent of the present invention.

[0033] Basic Example

[0034] The embodiment of the present invention provides a method for designing a tin-based flame-retardant polymer composite material based on artificial intelligence, comprising the following steps:

[0035] S1: Constructing a flame retardant performance dataset for flame retardant polymer composites

[0036] First, collect flame retardant performance-mechanical performance data with corresponding relationships from the existing experimental data or test data of tin-based flame retardant polymer composite materials;

[0037] Then the matrix, content of flame retardant and process parameters for preparing flame retardant polymer composites are used as modeling features, and data standardization is performed on the features;

[0038] S2: Building a random forest regression model

[0039] First, a two-step method is used to screen feature subsets. Specifically, the variable autocorrelation method in the first step and the maximum correlation minimum redundancy combined with feature recursive elimination method and genetic algorithm in the second step are used to screen the random forest regression model features that correspond to the two targets synchronously.

[0040] Then, based on the multi-objective optimization strategy, feature subsets were selected to establish random forest regression models of LOI and TS, and the grid search method was used to optimize the random forest regression models of LOI and TS;

[0041] S3: Screening of flame retardant polymer composites

[0042] The LOI and TS random forest regression models were applied to the Pareto optimization process based on the NSGA-II algorithm. In this process, the combination of LOI and TS models was used as the fitness function in the genetic algorithm. When the maximum iteration number was reached, the genetic iteration was stopped to screen out candidate flame retardant polymer composites with excellent LOI and TS properties.

[0043] S4: Verify the performance of candidate flame retardant polymer composites

[0044] The experimental synthesis step S3 screens candidate flame retardant polymer composites, and characterizes the LOI and TS properties of the candidate flame retardant polymer composites to verify their conformity and reliability with the screening performance in step S3.

[0045] Wherein, step S1 specifically includes the following steps:

[0046] S1-1 Collect flame retardant performance-mechanical performance data with corresponding relationships from existing experimental data or test data of tin-based flame retardant polymer composite materials; the specific flame retardant performance is the limiting oxygen index LOI, and the mechanical performance is the tensile strength TS;

[0047] S1-2 uses the matrix, content of flame retardant and process parameters of the flame retardant polymer composite as modeling features and performs data standardization on the features;

[0048] S1-3 randomly divides the dataset into training and test sets in a ratio of 4:1.

[0049] In step S2, a two-step method is used to select feature subsets, including the first step of using the variable autocorrelation method to remove redundant features, and the second step of using the maximum correlation minimum redundancy combined with feature recursive elimination method and genetic algorithm to select the random forest regression model features corresponding to the two targets synchronously. The evaluation index is the determination coefficient R of the leave-one-out cross-validation method. 2 , R 2 The larger the value, the better the feature set. The specific steps include:

[0050] The variable autocorrelation method in the first step of S2-1 is as follows:

[0051] First, the features are sorted according to mutual information and correlation and according to the maximum correlation and minimum redundancy. The performance of the model when multiple features are used as feature sets is considered, and the optimal feature subset is selected according to the model performance.

[0052] In feature engineering, the first step is to set the threshold of the Pearson correlation coefficient to 0.9 to reduce the number of features;

[0053] S2-2 The second step of feature screening is as follows:

[0054] Different methods are used for different targets. For LOI targets, the method of combining mRMR with RFE to screen features is used. For TS targets, the method of genetic algorithm to screen features is used, and the genetic algorithm is used to simulate the crossover and mutation optimization process in nature.

[0055] The maximum relevance minimum redundancy combined with the feature recursive elimination method and the genetic algorithm in step S2-2 is to respectively select the random forest regression model features corresponding to the two targets synchronously. The specific process of the maximum relevance minimum redundancy method for calculating the feature ranking is as follows:

[0056] First, calculate the mutual information between each feature f and the target value c. The mutual information calculation formula is as follows:

[0057]

[0058] Where p(f,c) is the joint probability density, p(f) and p(c) are the marginal probability densities;

[0059] Sm represents the feature subset consisting of the m features with the largest mutual information with c, and Sn represents the n features to be selected;

[0060] Let f' represent the feature in Sn, and the correlation D between the feature f' and c in Sn is as follows:

[0061] D=I(f',c)

[0062] The redundancy R of the features in Sm and Sn is as follows:

[0063]

[0064] The mRMR evaluation function evaluates the maximum correlation between the feature set to be selected and the target value and its minimum redundancy with the selected feature set, that is, the following formula 4:

[0065] max[DR]

[0066] At this point, the final evaluated and ranked feature set S is obtained as follows:

[0067] S={f”1,f”2,…,f” h ,…,f N ”}.

[0068] The step S3 of screening the flame retardant polymer composite material is to apply the LOI and TS random forest regression model of step S2 to the Pareto optimization process based on the NSGA-II algorithm according to the multi-objective optimization strategy of the flame retardant polymer composite material. The optimization process specifically includes the following steps:

[0069] S3-1 Find solutions in non-dominated sorting space

[0070] f1, f2 represent two targets to be optimized. i For example, if there is another solution x j f1 and f2 are not worse than x i f1 and f2, and x j At least one target is better than x i , then x j Dominate x i ; The non-dominated sorting calculates the number of dominated solutions of each solution in the space, and sorts the levels according to the number of dominated solutions. The solutions with 0 dominated solutions are placed in the first layer, and the solutions in the first layer are removed, and the second layer is calculated. This process is repeated until all solutions in the population are assigned to different non-dominated layers; the non-dominated solutions of each layer constitute a subset. The earlier the layer, the better the solution, and the first layer is the Pareto front;

[0071] S3-2 Calculate the crowding distance to measure the density of the solution

[0072] NSGA-II is used to calculate the crowding distance to measure the density of solutions to ensure the diversity of solutions. The crowding distance is d i It is defined as the sum of the distances between each solution and its adjacent solutions, and the calculation formula is as follows:

[0073]

[0074] where f k is the kth objective function, x i +1 and xi -1 are the solutions x i Neighboring solutions on this objective.

[0075] S3-3 Screening of candidate flame retardant polymer composites

[0076] The GA algorithm is used to generate new candidate solutions through genetic algorithm crossover mutation, and the above steps S3-1 and S3-2 are repeated. Finally, NSGA-II finds the Pareto frontier. In this process, the combination of LOI and TS models is used as the fitness function in the genetic algorithm. When the maximum iteration generation is reached, the genetic iteration is stopped, and the candidate flame retardant polymer composites with excellent LOI and TS performance are screened out, and their reliability is verified.

[0077] A tin-based flame-retardant polymer composite material is prepared by adopting the artificial intelligence-based tin-based flame-retardant polymer composite material design method.

[0078] The Pareto optimization method used in the embodiment of the present invention is a commonly used method for solving multi-objective optimization. It optimizes one of the objectives without losing one of the objectives, and thus becomes a reliable method for solving the common "seesaw" problem in the field of materials. It obtains the Pareto frontier by finding a non-dominated solution, and the points on the frontier can balance the two objectives. Since the NSGA-II algorithm is used as one of the commonly used Pareto optimization algorithms based on a genetic algorithm with global search capabilities, the embodiment of the present invention selects a Pareto optimization method based on the NSGA-II algorithm to carry out intelligent design of flame-retardant polymer composite materials.

[0079] Random Forest (RF) is an ensemble learning algorithm proposed by Leo Breima in 2001. It integrates decision trees together and integrates the results of these decision tree algorithms in an average manner. This ensemble learning algorithm reduces the risk of overfitting of a single tree. First, it uses the Bagging method to randomly sample and generate multiple subsets, trains a decision tree based on each subset, and constructs different decision tree models by randomly selecting features as nodes of different trees, and finally averages the results of different decision tree models as the final result. The embodiment of the present invention combines Pareto optimization with the random forest algorithm to solve various technical problems and make the final output result reliable.

[0080] The following is a detailed description with reference to a number of specific embodiments.

[0081] Example 1

[0082] The artificial intelligence-based tin-containing flame-retardant polymer composite material and the design method thereof provided in the embodiments of the present invention are specific applications based on the basic embodiments. An artificial intelligence-based design method is constructed by using a Pareto optimal multi-objective optimization method to quickly design a flame-retardant polymer composite material containing a tin-based flame retardant, and the composite material has both high flame retardant properties and mechanical properties.

[0083] Specifically, a random forest regression model was established based on 139 flame-retardant polymer composite experimental LOI data and 10 features in the training set randomly divided from the experimental data set, and the leave-one-out method was used to cross-validate the output results.

[0084] A tin-based flame-retardant polymer composite material based on artificial intelligence and a design method thereof, which specifically comprises the following steps:

[0085] 1. Get the initial data set

[0086] A total of 174 flame retardant (LOI) data and mechanical properties (TS) data of flame retardant polymer composites were collated from existing experimental data or test data. The two are interrelated synchronous performance data. In the original data, LOI is mostly concentrated below 28%, and LOI data exceeding 28% only account for 13% of the original data set. The TS data in the original data are evenly distributed, concentrated in 16MPa to 22MPa;

[0087] 2. The quality of 18 formulations of flame-retardant polymer composites and two process parameters were used as the initial feature input. In order to reduce the impact of dimensional differences on the model, the features were standardized using the standardization method in scikit-learn.

[0088] 3. The standardized data set is randomly divided into training set and test set according to the ratio of 4:1.

[0089] 4. In feature engineering, the first step is to set the threshold of the Pearson correlation coefficient to 0.9. The number of features is reduced from 20 to 17. In the second step of feature screening, different methods are used for different targets. For the LOI target, the method of combining mRMR with RFE for feature screening is selected. First, the features are sorted according to mutual information and correlation and according to maximum correlation and minimum redundancy. The performance of the model when 4 to 17 features are considered as the feature set is selected, and the optimal feature subset is selected according to the model performance. In the second step of feature screening of the LOI model, when the number of features is 10, the R of the LOI model is 2The highest, so a feature subset consisting of 10 features is selected as the optimal feature subset of the LOI model. For the TS target, the method of selecting features by genetic algorithm is used. The genetic algorithm simulates the process of crossover and mutation optimization in nature. In this optimization process, the minimum number of features selected is limited to be greater than 2, and the fitness function is the RMSE of the model. The feature subset with the best modeling effect is selected through the optimization process of minimizing the fitness function, and finally 7 features are selected. Among these features, PP represents polypropylene, EBS is lubricant, DBDPE is Decabromodiphenyl Ethane, DOPO is 9,10-dihydro-9-oxa-10-phosphaphenanthrene-10-oxide, ZS is zinc stannate, and MPP is Methylphosphonic Polymers. TCA is a triazine carbonizing agent, and the selected features are shown in Table 1.

[0090] The specific process of mRMR calculation feature ranking is as follows:

[0091] First, calculate the mutual information between each feature f and the target value c. The mutual information calculation formula is:

[0092]

[0093] Where p(f,c) is the joint probability density, p(f) and p(c) are the marginal probability densities; Sm represents the feature subset consisting of the m features with the largest mutual information with c, and Sn represents the n features to be selected. Let f' represent the feature in Sn, and the correlation D between feature f' and c in Sn is:

[0094] D=I(f',c)

[0095] The redundancy R of the features in Sm and Sn is:

[0096]

[0097] The mRMR evaluation function evaluates the maximum correlation between the feature set to be selected and the target value and its minimum redundancy with the selected feature set, that is:

[0098] max[DR]

[0099] At this point, the final evaluated and sorted feature set S is obtained:

[0100] S={f”1,f”2,…,f” h ,…,f N ”}

[0101] Table 1 shows the features selected by feature engineering for establishing the LOI random forest regression model and the TS random forest regression model.

[0102] Table 1

[0103]

[0104]

[0105] 5. Based on the above selected features, LOI and TS random forest regression models are established, and the grid search method is used to optimize the random forest regression models respectively. When the parameter n_estimators of the LOI random forest regression model is 96, its model performance is the best, and when the parameter n_estimators of the TS random forest regression model is 114, its model performance is the best.

[0106] 6. In the Pareto optimization process, a virtual space is constructed according to the range of the characteristics of the original data set, the initial population is set to 200, the maximum number of evolution generations is set to 20,000, the mutation probability is set to 0.2, the crossover probability is set to 0.7, and the combination of two random forest regression models is used as the fitness function to design a flame-retardant polymer composite material with excellent LOI and TS and verify its reliability.

[0107] The steps of NSGA-II algorithm to solve Pareto front are as follows:

[0108] a) Solutions in the non-dominated sorting space. f1 and f2 represent two objectives to be optimized. For a solution x i For example, if there is another solution x j f1 and f2 are not worse than x i f1 and f2, and x j At least one target is better than x i , then x j Dominate x i . Non-dominated sorting calculates the number of dominated solutions for each solution in the space, and sorts the layers according to the number of dominated solutions. Solutions with 0 dominated solutions are placed in the first layer, and the solutions in the first layer are removed, and the second layer is calculated. This process is repeated until all solutions in the population are assigned to different non-dominated layers. The non-dominated solutions of each layer constitute a subset. The earlier the layer, the better the solution. The first layer is the Pareto front.

[0109] b) To ensure the diversity of solutions, NSGA-II calculates the crowding distance to measure the density of solutions.

[0110] The crowding distance is d i It is defined as the sum of the distances between each solution and its neighboring solutions, and the calculation formula is:

[0111]

[0112] where f k is the kth objective function, x i +1 and x i -1 are the solutions x i Neighboring solutions on this objective.

[0113] The GA algorithm generates new candidate solutions through genetic algorithm crossover mutation and repeats the above steps. Finally, NSGA-II finds the Pareto frontier.

[0114] 7. Experimentally synthesize candidate flame-retardant polymer composites and characterize their flame-retardant and mechanical properties, and verify the consistency and reliability of experimental values ​​with predicted values

[0115] In this embodiment, based on the output results of the random forest regression model established based on the 139 flame retardant polymer composite experimental LOI data and 10 features in the training set randomly divided from the experimental data set, the results of cross-validation (comparison of experimental values ​​with predicted values) using the leave-one-out method are shown in Figure 2. Figure 1 shown.

[0116] Depend on Figure 1 It can be seen that the random forest regression algorithm was used to perform leave-one-out cross-validation on the LOI data of 139 flame-retardant polymer composites in the training set divided from the original data set to evaluate the model effect. The Pearson correlation coefficient (R) of the leave-one-out cross-validation of the training set was 0.919, and the determination coefficient R 2 The value is 0.844, and the root mean square error RMSE is 1.369, which shows that the model has good robustness, high conformity and high reliability.

[0117] Example 2

[0118] The artificial intelligence-based tin-based flame-retardant polymer composite material and the design method thereof provided in the embodiment of the present invention are substantially the same as those in Example 1, except that:

[0119] The test set output results of the model established based on the LOI data of 35 flame-retardant polymer composites and 10 features in the test set randomly divided from the experimental data set are compared with the experimental values ​​of the test set. Figure 2 shown.

[0120] Depend on Figure 2 It can be seen that the result of the test set is R=0.923, R 2 The value is 0.850 and the RMSE is 1.282. This result shows that the generalization ability of the model is good.

[0121] Example 3

[0122] The artificial intelligence-based tin-based flame-retardant polymer composite material and the design method thereof provided in the embodiments of the present invention are substantially the same as those in Embodiment 1 and Embodiment 2, except that:

[0123] The results of the leave-one-out cross-validation of the random forest regression model established based on 139 flame-retardant polymer composite experimental TS data and 7 features in the training set randomly divided from the experimental data set (comparison of experimental values ​​with predicted values), such as Figure 3 shown.

[0124] Depend on Figure 3 It can be seen that the random forest regression algorithm was used to perform leave-one-out cross-validation on the TS data of 139 flame-retardant polymer composites in the training set divided from the original data set. This was used to evaluate the model effect. The Pearson correlation coefficient (R) of the training set leave-one-out cross-validation was 0.924, and the determination coefficient R 2 The root mean square error (RMSE) is 0.853 and 0.706, indicating that the model has good robustness.

[0125] Example 4

[0126] The artificial intelligence-based tin-based flame-retardant polymer composite material and the design method thereof provided in the embodiment of the present invention are substantially the same as those in Embodiments 1 to 3, except that:

[0127] The test set results of the model established based on TS data of 35 flame-retardant polymer composites and 7 features in the test set randomly divided from the experimental data set (comparison chart of test set experimental values ​​and predicted values), as shown in Figure 4 shown.

[0128] Depend on Figure 4 It can be seen that the result of the test set is R=0.916, R 2 The value of the regression coefficient is 0.825 and the RMSE is 0.826. This result shows that the generalization ability of the model is good.

[0129] Table 2 shows the experimental values ​​of the virtual flame-retardant polymer composites designed by Pareto optimization.

[0130] Table 2

[0131]

[0132] Example 5

[0133] The artificial intelligence-based tin-based flame-retardant polymer composite material and the design method thereof provided in the embodiments of the present invention are substantially the same as those in Embodiments 1 to 4, except that:

[0134] The Pareto optimization based on NSGA-II algorithm designed virtual flame retardant polymer composites with better performance in both objectives, as shown in Table 2. Figure 5 As shown, the red points represent the Pareto frontier. The selected materials were prepared experimentally and tested. The test results are shown in Table 2. The results in Table 2 show that the intelligent design method achieves the design of flame-retardant polymer composites with better performance in both objectives.

[0135] In summary, the above embodiments of the present invention focus on preparing flame-retardant polymer composite materials with high flame retardant properties and mechanical properties based on multi-objective optimization prediction results and actual experiments; the multi-strategy design based on artificial intelligence is efficient, which solves the shortcomings of general artificial intelligence technology that lacks specificity and unreliable output results, greatly reduces costs compared to trial and error methods, and can meet the industry's needs for composite material development.

[0136] It should be particularly noted that, with reference to the specific models, algorithms, data, formulas, parameter value ranges, and specific conditions of the preparation process recorded in the above embodiments, in other embodiments, similar models, components, ratios, and process conditions can be specifically selected within the scope recorded in the present invention, and the technical effects of the present invention can all be achieved, and the embodiments of the present invention will no longer list them one by one.

[0137] The above embodiments have elaborated the technical solutions of the present invention in detail. It should be understood that the purpose is to enable people familiar with this technology to understand the contents of the present invention and implement them accordingly, and it cannot be used to limit the protection scope of the present invention. Technical solutions that are the same or similar to the above embodiments of the present invention are all within the protection scope of the present invention.

Claims

1. A design method for a tin-based flame-retardant polymer composite material based on artificial intelligence, characterized in that: It includes the following steps: S1: Constructing a flame retardant performance dataset for flame retardant polymer composites First, collect flame retardant performance-mechanical performance data with corresponding relationships from the existing experimental data or test data of tin-based flame retardant polymer composite materials; Then the matrix, content of flame retardant and process parameters for preparing flame retardant polymer composites are used as modeling features, and data standardization is performed on the features; S2: Building a random forest regression model First, a two-step method is used to screen feature subsets. Specifically, the variable autocorrelation method in the first step and the maximum correlation minimum redundancy combined with feature recursive elimination method and genetic algorithm in the second step are used to screen the random forest regression model features that correspond to the two targets synchronously. Then, based on the multi-objective optimization strategy, feature subsets were selected to establish random forest regression models of LOI and TS, and the grid search method was used to optimize the random forest regression models of LOI and TS; S3: Screening of flame retardant polymer composites The LOI and TS random forest regression models were applied to the Pareto optimization process based on the NSGA-II algorithm. In this process, the combination of LOI and TS models was used as the fitness function in the genetic algorithm. When the maximum iteration number was reached, the genetic iteration was stopped to screen out candidate flame retardant polymer composites with excellent LOI and TS properties. S4: Verify the performance of candidate flame retardant polymer composites The experimental synthesis step S3 screens candidate flame retardant polymer composites, and characterizes the LOI and TS properties of the candidate flame retardant polymer composites to verify their conformity and reliability with the screening performance in step S3.

2. The design method of the tin-based flame-retardant polymer composite material based on artificial intelligence according to claim 1, characterized in that: Step S1 specifically includes the following steps: S1-1 Collect flame retardant performance-mechanical performance data with corresponding relationships from existing experimental data or test data of tin-based flame retardant polymer composite materials; the specific flame retardant performance is the limiting oxygen index LOI, and the mechanical performance is the tensile strength TS; S1-2 uses the matrix, content of flame retardant and process parameters of the flame retardant polymer composite as modeling features and performs data standardization on the features; S1-3 randomly divides the dataset into training and test sets in a ratio of 4:

1.

3. The design method of the tin-based flame-retardant polymer composite material based on artificial intelligence according to claim 1, characterized in that: Step S2 uses a two-step method to select feature subsets, including the first step of using the variable autocorrelation method to remove redundant features, and the second step of using the maximum correlation minimum redundancy combined with feature recursive elimination method and genetic algorithm to select the random forest regression model features corresponding to the two targets synchronously. The evaluation index is the determination coefficient R of the leave-one-out cross-validation method. 2 , R 2 The larger the value, the better the feature set. The specific steps include: The variable autocorrelation method in the first step of S2-1 is as follows: First, the features are sorted according to mutual information and correlation and according to the maximum correlation and minimum redundancy. The performance of the model when multiple features are used as feature sets is considered, and the optimal feature subset is selected according to the model performance. In feature engineering, the first step is to set the threshold of the Pearson correlation coefficient to 0.9 to reduce the number of features; S2-2 The second step of feature screening is as follows: Different methods are used for different targets. For LOI targets, the method of combining mRMR with RFE to screen features is used. For TS targets, the method of genetic algorithm to screen features is used, and the genetic algorithm is used to simulate the crossover and mutation optimization process in nature.

4. The design method of the tin-based flame-retardant polymer composite material based on artificial intelligence according to claim 3 is characterized in that: The maximum relevance minimum redundancy combined with the feature recursive elimination method and the genetic algorithm in step S2-2 is to respectively select the random forest regression model features corresponding to the two targets synchronously. The specific process of the maximum relevance minimum redundancy method for calculating the feature ranking is as follows: First, calculate the mutual information between each feature f and the target value c. The mutual information calculation formula is as follows: Where p(f,c) is the joint probability density, p(f) and p(c) are the marginal probability densities; Sm represents the feature subset consisting of the m features with the largest mutual information with c, and Sn represents the n features to be selected; Let f' represent the feature in Sn, and the correlation D between the feature f' and c in Sn is as follows: D=I(f′,c) The redundancy R of the features in Sm and Sn is as follows: The mRMR evaluation function evaluates the maximum correlation between the feature set to be selected and the target value and its minimum redundancy with the selected feature set, that is, the following formula 4: max[DR] At this point, the final evaluated and ranked feature set S is obtained as follows: S={f″1,f″2,…,f″ h ,…,f N ″}。 5. The design method of the tin-based flame-retardant polymer composite material based on artificial intelligence according to claim 1, characterized in that: The step S3 of screening the flame retardant polymer composite material is to apply the LOI and TS random forest regression model of step S2 to the Pareto optimization process based on the NSGA-II algorithm according to the multi-objective optimization strategy of the flame retardant polymer composite material. The optimization process specifically includes the following steps: S3-1 Find solutions in non-dominated sorting space f1, f2 represent two targets to be optimized. i For example, if there is another solution x j f1 and f2 are not worse than x i f1 and f2, and x j At least one target is better than x i , then x j Dominate x i ; The non-dominated sorting calculates the number of dominated solutions of each solution in the space, and sorts the levels according to the number of dominated solutions. The solutions with 0 dominated solutions are placed in the first layer, and the solutions in the first layer are removed, and the second layer is calculated. This process is repeated until all solutions in the population are assigned to different non-dominated layers; the non-dominated solutions of each layer constitute a subset. The earlier the layer, the better the solution, and the first layer is the Pareto front; S3-2 Calculate the crowding distance to measure the density of the solution NSGA-II is used to calculate the crowding distance to measure the density of solutions to ensure the diversity of solutions. The crowding distance is d i It is defined as the sum of the distances between each solution and its neighboring solutions, and is calculated as follows: where f k is the kth objective function, x i +1 and x i -1 are the solutions x i Neighboring solutions on this objective. S3-3 Screening of candidate flame retardant polymer composites The GA algorithm is used to generate new candidate solutions through genetic algorithm crossover mutation, and the above steps S3-1 and S3-2 are repeated. Finally, NSGA-II finds the Pareto frontier. In this process, the combination of LOI and TS models is used as the fitness function in the genetic algorithm. When the maximum iteration number is reached, the genetic iteration is stopped to screen out candidate flame retardant polymer composites with excellent LOI and TS properties.

6. A tin-based flame-retardant polymer composite material, characterized in that: The composite material is prepared according to the artificial intelligence-based design method for tin-based flame-retardant polymer composite materials according to any one of claims 1 to 5.

Images