Descriptor screening method and device for multifunctional material development and medium

Abstract

The invention discloses a descriptor screening method and device for multifunctional material development and a medium, and relates to the field of material science, and the method comprises the steps: creating a plurality of physical descriptors for the structure component information of a target material in a material data set, and carrying out the dimension reduction optimization; respectively inputting the optimized descriptor sets into a plurality of target performance prediction models, and explaining the models by adopting an SHAP algorithm to obtain a plurality of optimized descriptor sorting coefficient sets; the target performance prediction model is a pre-trained machine learning model; on the basis of the multiple optimization descriptor sorting coefficient sets, feature sensitivity difference analysis is conducted on the optimization descriptor sets, multiple important descriptors are obtained through screening, and the important descriptors are key physical factors for multi-performance decoupling of the target material. According to the method, the numerical value levels of the different key physical factors corresponding to the material can be comprehensively evaluated so as to guide efficient development of the multifunctional material which meets the requirements of multiple target performance indexes at the same time.

Description

Technical Field

[0001] The present application relates to the field of materials science, and in particular to a descriptor screening method, device and medium for developing multifunctional materials. Background Art

[0002] In actual engineering applications, the functional requirements for materials are often diverse, involving optical, electrical, thermal and mechanical properties. The close connection between structure and performance leads to deep coupling between various properties. Most of the current material design research focuses only on the optimization of a single specific performance, but while the target performance parameters are improved, other key performance indicators may unexpectedly decrease. In addition, in existing material design research, due to the limitations of the machine learning "black box" model, most of the explicit relationships between material structure and performance are difficult to reveal.

[0003] For example, for the development of multifunctional materials, the mainstream method is to use multi-objective optimization algorithms, including but not limited to multi-objective genetic algorithms and multi-objective Bayesian optimization. However, on the one hand, these algorithms require experimental / calculated performance verification of the newly recommended structure in each round of optimization iteration, which is an expensive and time-consuming process. On the other hand, such as genetic algorithms, their optimization convergence performance is greatly affected by the initial samples; the performance of Bayesian optimization algorithms mainly depends on kernel functions, typically Gaussian process regression, and the training of kernel functions requires tedious hyperparameter optimization. In other words, existing material designs are limited by machine learning models, and the performance testing of materials needs to be repeated during the model optimization iteration process, which has the disadvantages of being time-consuming, inefficient, and costly. Summary of the invention

[0004] The purpose of this application is to provide a descriptor screening method, device and medium for the development of multifunctional materials, which can realize the efficient development of multifunctional materials that simultaneously meet different performance index requirements.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a descriptor screening method for the development of multifunctional materials, comprising:

[0007] Acquire a material data set; the material data set includes structural composition data of a target material and a plurality of target performance indicators; the target performance indicator is used to reflect one of the optical property, electrical property, thermal property and mechanical property of the target material;

[0008] For any of the target performance indicators, multiple physical descriptors are created based on the structural composition data of the target material; the physical descriptors are used to characterize the physical indicators derived from the component characteristics, atomic characteristics, electronic characteristics and topological characteristics of the corresponding target material;

[0009] Performing dimensionality reduction optimization on multiple physical descriptors corresponding to all the target performance indicators to obtain an optimized descriptor set;

[0010] The optimization descriptor set is respectively input into a plurality of target performance prediction models, and the target performance prediction models are interpreted using the SHAP algorithm to obtain a plurality of optimization descriptor ranking coefficient sets; wherein one target performance prediction model corresponds to one target performance indicator and one optimization descriptor ranking coefficient set; the target performance prediction model is a pre-trained machine learning model, which is used to predict the value of the target performance indicator according to the optimization descriptor;

[0011] Based on the multiple optimization descriptor sorting coefficient sets, the optimization descriptor sets are subjected to feature sensitivity difference analysis and screened to obtain multiple important descriptors; the important descriptors are key physical factors for decoupling multiple properties of the target material, and are used as a basis for evaluating whether the material has multifunctional properties.

[0012] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a descriptor screening method for the development of multifunctional materials.

[0013] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a descriptor screening method for developing multifunctional materials.

[0014] According to the specific embodiments provided by the present application, the present application discloses the following technical effects: the present application provides a descriptor screening method, device and medium for the development of multifunctional materials, which can determine multiple target performance indicators of the target material as needed, thereby constructing a multi-performance requirement that reflects one or more of the optical properties, electrical properties, thermal properties and mechanical properties of the target material, and providing a multi-performance basis for subsequent step processing. Then, multiple physical descriptors are created based on the structural composition data of the target material. The setting of the physical descriptors enables each descriptor to have a physical indicator derived from the component characteristics, atomic characteristics, electronic characteristics and topological characteristics of the corresponding target material. The obtained multiple physical descriptors are subsequently screened, and dimension reduction optimization processing and SHAP (SHapley Additive exPlanation, machine learning model interpretation tool) algorithm are used in the screening process to obtain the final multiple important descriptors, which are the key physical factors for decoupling the multiple performances of the target material. It should be noted that in the screening process of the above-mentioned important descriptors, multi-objective genetic algorithms or Bayesian optimization algorithms are not used, but only a machine learning model is constructed based on the publicly available data, and numerical sorting and comparison are performed, which is low in cost, short in time and more efficient. In addition, the important descriptors obtained in this application have actual physical meanings and can be used as a basis for evaluating whether a material has multifunctional properties. The determination of important descriptors also reveals the explicit relationship between material structure and performance and can be applied. Therefore, after obtaining multiple important descriptors, they can also be used to screen more advanced materials with multifunctional properties in the chemical space with only structural information based on the same target performance indicators. That is, from the design process of multiple target materials, this application scheme can also further improve design efficiency and reduce design costs. In summary, this application determines the decoupling key physical factors (structurally derived physical descriptors) between different properties of materials, which can achieve the purpose of efficient development of multifunctional materials that meet different performance index requirements at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 This is an application environment diagram of a descriptor screening method for developing multifunctional materials in an example of this application;

[0017] Figure 2 A schematic diagram of a process for a descriptor screening method for developing multifunctional materials in an example of the present application;

[0018] Figure 3 This is a comparison chart of model performances corresponding to three target performance indicators predicted by random forests in another example of the present application;

[0019] Figure 4 A comparison chart of SHAP analysis results corresponding to three target performance indicators in another example of the present application;

[0020] Figure 5 This is a schematic diagram of the dimensionality reduction result of the sorting matrix in another example of the present application;

[0021] Figure 6 A radar chart showing normalized values ​​of important descriptors of five typical polymers containing only structural component information in another example of the present application;

[0022] Figure 7 Schematic diagram of the distribution relationship between SHAP values ​​and feature values ​​of three important descriptors, namely HybRatio, BCUTZ-1h and AATS0Z, in another example of the present application;

[0023] Figure 8 A schematic diagram of the structure of a computer device provided as an example of the present application. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0025] This application constructs a mapping relationship between physical descriptors derived from material structures and different properties, and provides a method for decoupling multiple material properties based on sensitivity difference analysis of different physical descriptors, so as to achieve the goal of efficient development of multifunctional materials that simultaneously meet different performance index requirements.

[0026] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0027] The descriptor screening method for multifunctional material development provided in the embodiments of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, or it can be integrated on the server 104, or it can be placed on the cloud or other servers. The terminal 102 can send the structural component data of the target material and a variety of target performance indicators to the server 104. After the server 104 receives the structural component data of the target material and a variety of target performance indicators, it can perform descriptor screening through a preset algorithm program or using the algorithm program and molecular descriptor calculation software (including but not limited to Mordred, Dragon and RadonPy, etc.) loaded in the server, thereby obtaining a plurality of important descriptors. The server 104 can feedback the obtained multiple important descriptors to the terminal 102.

[0028] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, and tablet computers. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0029] In an exemplary embodiment, Figure 2 As shown, a descriptor screening method for the development of multifunctional materials is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or can be executed by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used for explanation, and the steps include the following steps 201 to 205.

[0030] Step 201, obtaining a material data set; the material data set includes structural composition data of the target material and multiple target performance indicators; the target performance indicator is used to reflect one of the optical properties, electrical properties, thermal properties and mechanical properties of the target material, such as thermal conductivity, refractive index, etc.; the structural composition data of the target material includes chemical element components and chemical formula information of the target material. The target material is an inorganic crystal material, a metal material, a thermoelectric material or an organic polymer material.

[0031] Step 202, for any of the target performance indicators, create multiple physical descriptors based on the structural composition data of the target material; the physical descriptors are used to characterize physical indicators derived from the component characteristics, atomic characteristics, electronic characteristics and topological characteristics of the corresponding target material.

[0032] Specifically, when the target material is an inorganic crystal, the corresponding physical descriptor can be collected from the existing database. When the target material is an organic molecular material, the corresponding physical descriptor can be calculated by molecular descriptor calculation software (including but not limited to Mordred, Dragon and RadonPy, etc.). Generally speaking, a physical descriptor consists of a 1×X dimensional vector, in which the X sites have independent physical meanings.

[0033] Step 203, dimensionality reduction optimization is performed on multiple physical descriptors corresponding to all the target performance indicators to obtain an optimized descriptor set. In an exemplary embodiment, the physical descriptors obtained through step 202 usually contain redundant information, resulting in overfitting of the machine learning model and low prediction accuracy. Therefore, for each target performance indicator, standardized feature engineering is required to perform feature dimensionality reduction. Step 203 of this application specifically includes:

[0034] Step 31) For any of the target performance indicators, the multiple physical descriptors corresponding to the target performance indicators are sequentially subjected to variance removal feature processing, statistical coefficient (including but not limited to Pearson correlation coefficient, Spearman correlation coefficient, etc.) screening processing and recursive feature elimination processing to obtain multiple processed physical descriptors; it should be noted that after the above three processes, a set of processed physical descriptors that match each target performance indicator (the trained machine learning model has a high accuracy) can be obtained.

[0035] Among them, the variance feature removal process: the variance index is used to evaluate the numerical fluctuation of each physical descriptor itself, and the features with lower variance values ​​are removed. For each physical descriptor, the variance can be expressed as:

[0036]

[0037] Where NUM is the number of target materials in the material data set, x i is the value of the physical descriptor, is the average value of the physical descriptor. Physical descriptors with variance values ​​less than a preset threshold are deleted.

[0038] Statistical coefficient screening and processing: Pearson correlation coefficient is selected to consider the linear relationship between the two groups of variables, and Spearman correlation coefficient is used to measure the monotonic relationship between the two groups of variables; among them, one group of variables in the two groups is the physical descriptor, represented by x; the other group of variables is the target performance indicator, represented by y.

[0039] Pearson correlation coefficient Pea x,y The calculation formula is:

[0040]

[0041] Among them, μ x , μ y , S x and S y are the mean and standard deviation of variables x and y, respectively. num represents the number of physical descriptors after the variance feature is removed. i represents the i-th physical descriptor, y i represents the target performance indicator corresponding to the i-th physical descriptor.

[0042] The calculation formula of Spearman correlation coefficient Spa is:

[0043]

[0044] Among them, d i It represents the difference between the rankings of variables x and y, that is, the difference in the rank between the two variables (one is the descriptor and the other is the target performance) after they are ranked separately.

[0045] Recursive feature elimination: The best feature subset (a set of physical descriptors) is selected by repeatedly building a model using the random forest algorithm and removing the least important features (based on the weight values ​​of the random forest algorithm). In short, the model is built using the random forest algorithm to obtain the values ​​of the target performance indicators corresponding to different combinations of physical descriptors; the obtained values ​​are compared, and then the corresponding combination of physical descriptors with the best performance is inferred, thereby eliminating unimportant features from all physical descriptors.

[0046] Step 32) Merge and remove duplicates of the multiple processed physical descriptors corresponding to all the target performance indicators to obtain an optimized descriptor set D M (M is the number of optimized descriptors).

[0047] Step 204, input the optimization descriptor set into multiple target performance prediction models respectively, and use the SHAP algorithm to interpret the target performance prediction model to obtain multiple optimization descriptor ranking coefficient sets; wherein, one target performance prediction model corresponds to one target performance indicator and one optimization descriptor ranking coefficient set; the target performance prediction model is a pre-trained machine learning model, which is used to predict the value of the target performance indicator based on the optimization descriptor.

[0048] There are many machine learning models available, including but not limited to random forest, support vector machine, and artificial neural network regression algorithms. When faced with multiple available regression algorithms, we can use the optimized descriptor set to build machine learning regression models for predicting each target performance indicator, and comprehensively calculate the R square (R2 ) and root mean square error (RMSE) evaluation indicators to select the machine learning model with the best prediction performance. Among them, R square is a value not greater than 1. The closer it is to 1, the better the model performance. The smaller the RMSE value, the better the model performance.

[0049] In an exemplary embodiment, after a plurality of target performance prediction models (W target performance prediction models corresponding to W target performance indicators) have been determined, step 204 includes:

[0050] Step 41) For any target performance indicator, obtain a target performance prediction model corresponding to the target performance indicator.

[0051] Step 42) Input the set of optimization descriptors into the target performance prediction model, and interpret the target performance prediction model using the SHAP algorithm to obtain the SHAP value corresponding to each optimization descriptor. SHAP is a game theory-based method that links the input optimal credit allocation of the target performance prediction model with local interpretation. For each independent target performance indicator, for each optimization descriptor, by multiplying all marginal contributions by the corresponding weights and summing them, the SHAP value of the input optimization descriptor q can be obtained. The calculation formula is:

[0052]

[0053] in, represents the SHAP value of the optimization descriptor q, M represents the number of optimization descriptors in the optimization descriptor set, and D represents all possible permutations and combinations of M optimization descriptors. represents a subset of all optimization descriptors in the permutation and combination S of optimization descriptors that do not contain the optimization descriptor q; f x (Z) denotes the conditional expectation of the optimized descriptor subset Z.

[0054] Step 43) Based on all the optimization descriptors and the corresponding SHAP values, the ranking coefficients of the multiple optimization descriptors corresponding to the target performance index are calculated to obtain a set of optimization descriptor ranking coefficients. The calculation formula for the ranking coefficient of each optimization descriptor is:

[0055]

[0056] Among them, R q represents the ranking coefficient of the optimization descriptor q, NUM is the number of target materials in the material data set, p is the sequence number of each target material, is the SHAP value corresponding to the optimized descriptor q for the pth target material. qThe larger the value of , the greater the role played by the optimization descriptor in the model.

[0057] Step 205, based on the plurality of optimization descriptor sorting coefficient sets, perform feature sensitivity difference analysis on the optimization descriptor set and screen to obtain a plurality of important descriptors; the important descriptors are key physical factors for decoupling multiple properties of the target material, and are used as a basis for evaluating whether the material has multifunctional properties.

[0058] In an exemplary embodiment, step 205 includes:

[0059] Step 51) The plurality of optimization descriptor ranking coefficient sets are combined according to the corresponding target performance indicators to obtain a multidimensional ranking matrix; the number of the target performance indicators is consistent with the column dimension value of the multidimensional ranking matrix. It should be noted that a single optimization descriptor plays a different role in the prediction model of each target performance indicator, so the optimization descriptor ranking ordinal numbers corresponding to each target performance indicator are combined to form a ranking matrix rank, which is an M×W matrix.

[0060]

[0061] Among them, r M,W is the ranking coefficient of the Mth optimization descriptor corresponding to the Wth target performance indicator.

[0062] Step 52) Based on the projection method, the multi-dimensional sorting matrix is ​​subjected to dimensionality reduction processing to obtain a two-dimensional optimized descriptor matrix; in a specific practical application, the principal component analysis method can be used for dimensionality reduction processing, and the processing process is as follows:

[0063] First, the sorting matrix rank is standardized to obtain the following matrix Rank:

[0064]

[0065] in, and S W They are the mean and standard deviation of the ranking coefficients of all optimized descriptors corresponding to the target performance indicator w:

[0066]

[0067] Furthermore, the covariance matrix Cov of the matrix Rank is calculated as follows:

[0068]

[0069] Furthermore, the eigenvalues ​​and eigenvectors of the covariance matrix Cov are calculated, and the first two eigenvectors A1 and A2 are taken to form a new matrix:

[0070]

[0071] Furthermore, we take the first two dimensions of the Rank matrix that contain the largest amount of information:

[0072]

[0073] Step 53) Based on the two-dimensional optimization descriptor matrix, the sorting distance coefficient corresponding to each optimization descriptor is calculated, and the calculation formula is as follows:

[0074]

[0075] When optimizing the descriptor d m The larger the value, the higher the importance level of the optimization descriptor. m Several important optimized descriptors are selected as the basis for judging multifunctional materials.

[0076] Step 54) Based on a preset distance threshold, the optimized descriptor set is screened according to the sorting distance coefficient to obtain a plurality of important descriptors. In a specific application example, a plurality of important descriptors can also be used as a basis for efficient preliminary screening / design of multifunctional materials. One design method is to first obtain materials with unknown properties, and calculate and analyze the values ​​of important descriptors for the materials with unknown properties, which specifically includes the following steps:

[0077] (1) Obtaining a set of preset performance threshold ranges corresponding to the target material; a preset performance threshold range in the set of preset performance threshold ranges corresponds to one of the target performance indicators.

[0078] (2) Calculating the values ​​of the plurality of important descriptors corresponding to the material of unknown attribute, and performing a multifunctional evaluation on the material of unknown attribute according to the values ​​of the plurality of important descriptors to obtain an evaluation result; the evaluation result is that the material of unknown attribute has potential multifunctional attributes or the material of unknown attribute does not have potential multifunctional attributes.

[0079] (3) When the evaluation result is that the material of unknown property has potential multifunctional properties, the material of unknown property is subjected to experimental analysis to obtain a corresponding set of actual performance values.

[0080] (4) When the actual performance values ​​in the actual performance value set all meet the corresponding preset performance threshold range, the unknown property material is marked as a material that meets the multi-functional requirements.

[0081] Another design method is to first determine the value of the important descriptor, and then reversely design the material of the corresponding structure based on the value, which specifically includes the following steps:

[0082] (1) Obtaining a set of preset performance threshold ranges corresponding to the target material; a preset performance threshold range in the set of preset performance threshold ranges corresponds to one of the target performance indicators.

[0083] (2) After obtaining the plurality of important descriptors, based on the plurality of target performance indicators, the values ​​of the plurality of important descriptors are designed and assigned to obtain a plurality of candidate material structures.

[0084] (3) For any of the candidate material structures, perform experimental analysis to obtain a corresponding set of actual performance values.

[0085] (4) When the actual performance values ​​in the actual performance value set all meet the corresponding preset performance threshold range, the material candidate structure is marked as a structure that meets the material multifunctional requirements.

[0086] In another exemplary embodiment of the present application, the target material is a multifunctional polymer material, and the process of designing the material is as follows:

[0087] (1) Corresponding to the multiple target performance indicators mentioned in step 201 above: collect a benchmark data set containing 1075 polymers for use in establishing a machine learning model that reflects the structure-performance relationship in subsequent steps. The benchmark data set includes three important properties of polymers: refractive index RI, glass transition temperature Tg, and thermal conductivity TC. Among them, RI reflects the optical properties of the polymer and is closely related to the electrical and magnetic transport properties; high RI polymers are urgently needed organic materials for advanced display devices, image sensors, and organic light-emitting diodes. The other two are thermal properties. Tg affects the thermal stability and high-temperature processing rheology of the polymer, and TC reflects the heat dissipation performance of the polymer.

[0088] As shown in Table 1 below, the numerical distribution of the three target performance indicators for the 1075 polymers in the benchmark data set is shown; in Table 1, RI is calculated by density functional theory (DFT), with a value range of 1.27 to 1.85, of which 7.72% of the polymers are greater than 1.70; Tg is measured by experiment (Exp.), ranging from -77.71℃ to 405.65℃, of which only 1.77% of the data are greater than 380.0℃. TC is calculated by molecular dynamics (MD), with a distribution range of 0.08~1.07Wm -1 K -1 , of which 8.19% of the structures have a TC value exceeding 0.35 Wm -1 K -1 Based on the distribution of different target performance indicators, the performance optimization targets of the material are designed to be RI>1.70, Tg>380.00℃ and TC>0.35Wm-1 K -1 .

[0089] Table 1

[0090] nature unit Minimum Maximum Standard Deviation method RI / 1.27 1.85 0.31 DFT Tg ℃ -77.71 405.65 10.93 Exp. TC <![CDATA[Wm -1 K -1 ]]> 0.08 1.07 0.31 MD

[0091] (2) Corresponding to step 202 above, the corresponding physical descriptors were calculated by molecular descriptor calculation software Mordred and RadonPy, totaling 323. Then, feature optimization was performed independently for the three target performance indicators that needed to be optimized, and the thresholds for removing low-variance features, Pearson correlation coefficient, and Spearman correlation coefficient were set respectively, and the recursive feature elimination (RFE) method was used for feature optimization, as shown in Table 2 below.

[0092] Table 2

[0093]

[0094] Then, the processed physical descriptors corresponding to the three target performance indicators to be optimized are merged and deduplicated to form a new unified descriptor set containing 45 optimized descriptors. Please note that the final 22+12+25 in Table 2 above is not equal to 45 because the optimized descriptors obtained for different target performance indicators may overlap. The 45 optimized descriptors obtained here are the number after removing duplicate descriptors.

[0095] (3) Corresponding to the construction of the machine learning model in step 204 above: For the three target performance indicators that need to be optimized, the random forest (RF) algorithm is used to build the machine learning model. For each target performance indicator, the RF model is repeatedly trained 20 times, and before each training, the entire data set is randomly split according to the training / test set ratio of 90% / 10%. The results are shown in the figure. Figure 3 As shown in Figure 2, the accuracy of the trained RF model is R in terms of the three performance predictions of RI, Tg and TC. 2 They are 0.954, 0.860 and 0.721 respectively.

[0096] (4) The structure-activity relationship of the polymer determines the intrinsic connection between various properties. Different physical characteristics (physical descriptors) are used to describe the electronic structure, atomic properties, bonding and other information of the polymer system. Moreover, the sensitivity of each optimization target performance to different descriptors is inconsistent. In fact, even the same descriptor will have different influences on different optimization performances. Understanding the contribution of different physical descriptors to each optimization target helps to grasp the decoupling factors between different properties, thereby efficiently designing / discovering multifunctional polymers. On this basis, corresponding to the application of the SAHP algorithm in step 204 above: the SHAP method is used to interpret the trained machine learning model, and the resulting SHAP value is distributed with the eigenvalue as follows: Figure 4 As shown (for each optimization target, the top 12 features ranked by importance are shown). Then, for each target performance indicator, the ranking coefficient of each optimized descriptor is calculated in turn. The importance ranking of the descriptors is determined based on the average of the SHAP absolute values ​​corresponding to each descriptor, which are extracted from the RF model trained with the optimized descriptors.

[0097] (5) Corresponding to step 51) to step 53) in step 205 above: sort the ranking coefficients of each optimized descriptor to obtain a ranking matrix rank with a dimension of 45×3, and reduce the dimension of the matrix. The result is as follows Figure 5 shown. Figure 5 The contribution of each optimized descriptor to the three properties (RI, Tg and TC) is revealed. Each optimized descriptor corresponds to a vector containing three components, and each component is the average value of the SHAP corresponding to different properties. Therefore, the original data is a three-dimensional array, represented by three mutually orthogonal coordinate axes corresponding to RI, Tg and TC (gray arrows). The principal component analysis technology used for dimensionality reduction extracts two principal component factors PC1 and PC2 with the largest differences and orthogonal to each other from these data, and uses these two factors as coordinate axes on the two-dimensional plane. Therefore, the comprehensive contribution of each optimized descriptor to the three properties can be judged by the distance between the corresponding data point and the origin. The larger the distance, the higher the degree of comprehensive contribution. The top nine important descriptors marked in the figure are consistent with the descriptors listed in Table 3 below.

[0098] Table 3

[0099] Descriptors introduce HybRatio The number of SP3 hybridized carbon atoms accounts for the sum of the number of SP2 and SP3 hybridized carbon atoms K_dih_ave The average value of the dihedral force constant ETA_dBeta Extended topological atomic Δβ coefficient Xch-6d Sixth-order ordered Chi chain weighted by σ electrons AATS2d Second-order delayed Moreau-Broto averaged autocorrelation function weighted by σ electrons Theta0_ave Average value of bond angle structural parameters under thermodynamic equilibrium K_bond_min Minimum value of bond force constant BCUTZ-1h Largest eigenvalue of a Burden matrix weighted by atomic number AATS0Z Zero-lag Moreau-Broto averaged autocorrelation function weighted by atomic number

[0100] (6) To further explore the relationship between the nine important descriptors and the three properties, this application selected five typical polymers, namely P130063, P164417, P362001, P373845, and P462066. Figure 6The normalized values ​​of nine important descriptors (based on 1075 benchmark data) are statistically analyzed through radar charts. At the same time, the different target performances of the five structures obtained by the experiment are listed in Table 4. Among them, SMILES stands for Simplified molecular input line entry system, which is a linear symbol used to input and represent molecular reactions. And, by Figure 7 It can be seen that the SHAP values ​​of the samples corresponding to the three important descriptors HybRatio, BCUTZ-1h and AATS0Z are inversely proportional to the feature values ​​(i.e., the values ​​of the corresponding target performance indicators), indicating that these three important descriptors are negatively correlated with the optimization target (i.e., the target performance indicator). When the normalized values ​​of these three important descriptors are close to 0, it is beneficial to the improvement of the three target performance indicators.

[0101] Table 4

[0102]

[0103] Taking the biphenyl-3,3',4,4'-tetraamine-naphthalene-2,6-dicarboxylic acid alternating copolymer (P164417) shown in the radar chart as an example, its three target performance indicators meet the target thresholds (i.e., RI>1.70, Tg>380.00℃ and TC>0.35Wm -1 K -1 ). For each important descriptor, except for the three important descriptors HybRatio, BCUTZ-1h and AATS0Z, the values ​​of the other important descriptors are at a relatively balanced level. At the same time, through the SHAP analysis, it can be seen that among the nine important descriptors identified, eight have made positive contributions to the performance improvement of P164417. This verifies that the present application is economical and efficient in the design of multifunctional materials.

[0104] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 8As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a descriptor screening method for the development of multifunctional materials is implemented.

[0105] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0106] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0107] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0108] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0109] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0110] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0111] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A descriptor screening method for the development of multifunctional materials, characterized in that: The descriptor screening method for multifunctional material development includes: Acquire a material data set; the material data set includes structural composition data of a target material and a plurality of target performance indicators; the target performance indicator is used to reflect one of the optical property, electrical property, thermal property and mechanical property of the target material; For any of the target performance indicators, multiple physical descriptors are created based on the structural composition data of the target material; the physical descriptors are used to characterize the physical indicators derived from the component characteristics, atomic characteristics, electronic characteristics and topological characteristics of the corresponding target material; Performing dimensionality reduction optimization on multiple physical descriptors corresponding to all the target performance indicators to obtain an optimized descriptor set; The optimization descriptor set is respectively input into a plurality of target performance prediction models, and the target performance prediction models are interpreted using the SHAP algorithm to obtain a plurality of optimization descriptor ranking coefficient sets; wherein one target performance prediction model corresponds to one target performance indicator and one optimization descriptor ranking coefficient set; the target performance prediction model is a pre-trained machine learning model, which is used to predict the value of the target performance indicator according to the optimization descriptor; Based on the multiple optimization descriptor sorting coefficient sets, the optimization descriptor sets are subjected to feature sensitivity difference analysis and screened to obtain multiple important descriptors; the important descriptors are key physical factors for decoupling multiple properties of the target material, and are used as a basis for evaluating whether the material has multifunctional properties.

2. The descriptor screening method for multifunctional material development according to claim 1, characterized in that: The optimization descriptor sets are respectively input into a plurality of target performance prediction models, and the target performance prediction models are interpreted using the SHAP algorithm to obtain a plurality of optimization descriptor ranking coefficient sets, including: For any target performance indicator, obtaining a target performance prediction model corresponding to the target performance indicator; Inputting the optimization descriptor set into the target performance prediction model, and interpreting the target performance prediction model using the SHAP algorithm to obtain the SHAP value corresponding to each optimization descriptor; Based on all the optimization descriptors and the corresponding SHAP values, the ranking coefficients of the multiple optimization descriptors corresponding to the target performance indicator are calculated to obtain a set of optimization descriptor ranking coefficients.

3. The descriptor screening method for multifunctional material development according to claim 2, characterized in that: The calculation formula for the SHAP value corresponding to each optimized descriptor is: in, represents the SHAP value of the optimization descriptor q, M represents the number of optimization descriptors in the optimization descriptor set, and D represents all possible permutations and combinations of M optimization descriptors. represents a subset of all optimization descriptors in the permutation and combination S of optimization descriptors that do not contain the optimization descriptor q; f x (Z) represents the conditional expectation of the optimized descriptor subset Z; The ranking coefficient of each optimization descriptor is calculated as: Among them, R q represents the ranking coefficient of the optimization descriptor q, NUM is the number of target materials in the material data set, p is the sequence number of each target material, SHAP value corresponding to the optimized descriptor q for the p-th target material.

4. The descriptor screening method for multifunctional material development according to claim 1, characterized in that: Based on the plurality of optimization descriptor ranking coefficient sets, feature sensitivity difference analysis is performed on the optimization descriptor set and a plurality of important descriptors are screened to obtain, including: Combining the plurality of optimization descriptor sorting coefficient sets according to corresponding target performance indicators to obtain a multidimensional sorting matrix; the number of the target performance indicators is consistent with the column dimension value of the multidimensional sorting matrix; Performing dimensionality reduction processing on the multi-dimensional sorting matrix based on a projection method to obtain a two-dimensional optimized descriptor matrix; Based on the two-dimensional optimized descriptor matrix, calculating the sorting distance coefficient corresponding to each optimized descriptor; Based on a preset distance threshold, descriptor screening is performed on the optimized descriptor set according to the sorting distance coefficient to obtain a plurality of important descriptors.

5. The descriptor screening method for multifunctional material development according to claim 1, characterized in that: Performing dimension reduction optimization on multiple physical descriptors corresponding to all the target performance indicators to obtain an optimized descriptor set, including: For any of the target performance indicators, a plurality of physical descriptors corresponding to the target performance indicator are sequentially subjected to variance removal feature processing, statistical coefficient screening processing and recursive feature elimination processing to obtain a plurality of processed physical descriptors; The multiple processed physical descriptors corresponding to all the target performance indicators are merged and deduplicated to obtain an optimized descriptor set.

6. The descriptor screening method for multifunctional material development according to claim 5, characterized in that: When performing statistical coefficient screening, the Pearson correlation coefficient is used to consider the linear relationship between the two groups of variables, and the Spearman correlation coefficient is used to measure the monotonic relationship between the two groups of variables; among them, one of the two groups of variables is the physical descriptor, represented by x; the other group of variables is the target performance indicator, represented by y; Pearson correlation coefficient Pea x,y The calculation formula is: Among them, μ x , μ y , S x and S y are the mean and standard deviation of variables x and y, respectively. num represents the number of physical descriptors after the variance feature is removed. i represents the i-th physical descriptor, y i represents the target performance indicator corresponding to the i-th physical descriptor; The calculation formula of Spearman correlation coefficient Spa is: Among them, d i Represents the difference in rank between variable x and variable y.

7. The descriptor screening method for multifunctional material development according to claim 1, characterized in that: The structural composition data of the target material includes chemical element components and chemical formula information of the target material; The target material is an inorganic crystal material, a metal material, a thermoelectric material or an organic polymer material.

8. The descriptor screening method for multifunctional material development according to claim 1, characterized in that: The descriptor screening method for multifunctional material development also includes: Acquire a preset performance threshold range set corresponding to the target material; a preset performance threshold range in the preset performance threshold range set corresponds to one of the target performance indicators; Calculating the values ​​of the plurality of important descriptors corresponding to the material of unknown attribute, and performing a multifunctional evaluation on the material of unknown attribute according to the values ​​of the plurality of important descriptors to obtain an evaluation result; the evaluation result is that the material of unknown attribute has a potential multifunctional attribute or the material of unknown attribute does not have a potential multifunctional attribute; When the evaluation result is that the material of unknown attribute has potential multifunctional attributes, performing test analysis on the material of unknown attribute to obtain a corresponding set of actual performance values; When the actual performance values ​​in the actual performance value set all meet the corresponding preset performance threshold range, the unknown property material is marked as a material that meets the multi-functional requirements.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the descriptor screening method for multifunctional material development according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the descriptor screening method for developing multifunctional materials according to any one of claims 1 to 8 is implemented.

Images