Doped thermoelectric material performance optimization research method and system based on machine learning

Abstract

The invention relates to the technical field of material data processing, in particular to a doped thermoelectric material performance optimization research method and system based on machine learning, and the method comprises the steps: constructing a doped thermoelectric material data set, obtaining a thermoelectric material feature group based on the doped thermoelectric material data set, and carrying out the key value pairing of the thermoelectric material feature group and a doped thermoelectric figure of merit, carrying out low-correlation feature elimination on the complete material feature set to obtain a high-correlation feature set, training a deep learning model group by utilizing the high-correlation feature set to obtain a thermoelectric figure of merit prediction model group and a prediction deviation coefficient group, and carrying out thermoelectric figure of merit prediction on the basis of the prediction deviation coefficient group. And carrying out prediction weight distribution on the thermoelectric figure of merit prediction model group to obtain an integrated thermoelectric figure of merit prediction model. According to the invention, the efficiency and the precision of performance optimization research of the doped thermoelectric material can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of material data processing, and in particular to a method and system for researching performance optimization of doped thermoelectric materials based on machine learning. Background Art

[0002] In today's world, with the intensification of the energy crisis and the increasing severity of environmental pollution problems, it has become particularly important to find new and efficient energy conversion and utilization technologies. Thermoelectric materials, as a material that can directly convert thermal energy into electrical energy, have attracted widespread attention due to their potential applications in energy conservation and emission reduction. However, the performance optimization of thermoelectric materials is a complex process, among which the thermoelectric figure of merit (ZT value) is a key indicator for measuring the performance of thermoelectric materials.

[0003] Traditional methods for optimizing the performance of thermoelectric materials mainly rely on a single machine learning model to predict the thermoelectric figure of merit of thermoelectric materials. However, the results obtained by a single machine learning model have large errors, and the traditional method does not include a step for cleaning the training data set, which further reduces the accuracy and efficiency of subsequent machine learning. Summary of the invention

[0004] The present invention provides a method and system for researching performance optimization of doped thermoelectric materials based on machine learning, the main purpose of which is to improve the efficiency and accuracy of researching performance optimization of doped thermoelectric materials.

[0005] To achieve the above-mentioned purpose, the present invention provides a method for optimizing the performance of doped thermoelectric materials based on machine learning, comprising:

[0006] Acquire an original thermoelectric material data set, wherein the original thermoelectric material data includes: an original material chemical formula and an original thermoelectric figure of merit;

[0007] Identifying a data set to be cleaned in the original thermoelectric material data set, and removing the data set to be cleaned from the original thermoelectric material data set to obtain a doped thermoelectric material data set;

[0008] Extracting doped thermoelectric material data in the doped thermoelectric material data set in turn, identifying the doped material chemical formula and the doped thermoelectric figure of merit in the doped thermoelectric material data, and constructing features for the doped material chemical formula to obtain a thermoelectric material feature group, and performing key value pairing on the thermoelectric material feature group and the doped thermoelectric figure of merit to obtain a complete material feature group;

[0009] Summarizing the complete material feature groups of each doping material chemical formula to obtain a complete material feature group set, and removing low-correlation features from the complete material feature group set to obtain a high-correlation feature group set;

[0010] Dividing the high correlation feature group sets into groups to obtain high correlation training feature group sets and high correlation verification feature group sets, wherein each high correlation feature group contains a chemical formula of a doping material;

[0011] Using a highly correlated training feature group set to train the pre-acquired deep learning model group, a thermoelectric figure of merit prediction model group is obtained, wherein the thermoelectric figure of merit prediction models in the thermoelectric figure of merit prediction model group correspond one-to-one to the deep learning models in the deep learning model group;

[0012] The thermoelectric figure of merit prediction model group is verified by using a highly correlated verification feature group set to obtain a prediction deviation coefficient group, and based on the prediction deviation coefficient group, prediction weights are assigned to the thermoelectric figure of merit prediction model group to obtain an integrated thermoelectric figure of merit prediction model;

[0013] Based on the integrated thermoelectric figure of merit prediction model, performance optimization research of doped thermoelectric materials based on machine learning is realized.

[0014] Optionally, the step of identifying a data set to be cleaned in the original thermoelectric material data set includes:

[0015] Extracting original thermoelectric material data in the original thermoelectric material data set in sequence;

[0016] Determine whether the original material chemical formula in the original thermoelectric material data is a valid chemical formula;

[0017] If the original material chemical formula in the original thermoelectric material data is not a valid chemical formula, the original thermoelectric material data is recorded as material data to be cleaned;

[0018] If the original material chemical formula in the original thermoelectric material data is a valid chemical formula, determining whether the original thermoelectric figure of merit in the original thermoelectric material data is within a preset valid value range;

[0019] If the original thermoelectric figure of merit in the original thermoelectric material data is not within the valid value range, the original thermoelectric material data is recorded as the material data to be cleaned;

[0020] If the original thermoelectric figure of merit in the original thermoelectric material data is within the valid value range, determining whether the original material chemical formula in the original thermoelectric material data is a preset doping material chemical formula;

[0021] If the original material chemical formula in the original thermoelectric material data is not the doped material chemical formula, the original thermoelectric material data is recorded as the material data to be cleaned;

[0022] The data of the materials to be cleaned are aggregated to obtain a data set of materials to be cleaned.

[0023] Optionally, the determining whether the original material chemical formula in the original thermoelectric material data is a preset doping material chemical formula includes:

[0024] Get the minimum component content in the chemical formula of the original material;

[0025] Determining whether the minimum component content is greater than a preset doping element content;

[0026] If the minimum component content is not greater than the doping element content, the original material chemical formula is recorded as the doping material chemical formula.

[0027] Optionally, the feature construction of the chemical formula of the doping material to obtain a thermoelectric material feature group includes:

[0028] Obtaining a basic element group of the doped thermoelectric material data set, wherein the basic element group is a combination of all elements appearing in the doped thermoelectric material data set, and there is no repeated element in the basic element group;

[0029] Identifying a component element group in a chemical formula of a doping material, and counting the component proportion of each component element in the component element group in the chemical formula of the doping material to obtain a component proportion group;

[0030] Setting a vacancy component vector according to the basic element group, wherein the length of the vacancy component vector is the same as the number of basic elements in the basic element group;

[0031] Based on the component element group, fill the empty component vector with the component proportion group at the same position to obtain a material component vector, wherein the length of the material component vector is the same as the length of the empty component vector, and the material component vector and the empty component vector represent the same elements at the same position;

[0032] According to the component element group, the chemical formula of the doping material is analyzed for constituent elements to obtain an element property vector, and the material component vector and the element property vector are combined to obtain a thermoelectric material characteristic group.

[0033] Optionally, the step of performing composition element analysis on the chemical formula of the doping material according to the component element group to obtain the element property vector includes:

[0034] Setting an element attribute category group, and constructing a vacancy attribute vector according to the element attribute category group, wherein the length of the vacancy attribute vector is the same as the number of element attribute categories in the element attribute category group;

[0035] Analyze the element properties of the component element groups to obtain the target element attribute groups;

[0036] The target element attribute group is used to fill the empty attribute vector in the same position to obtain an element property vector, wherein the vector length of the element property vector is the same as the vector length of the empty attribute vector.

[0037] Optionally, performing element property analysis on the component element group to obtain the target element attribute group includes:

[0038] Based on the element attribute category group, an element attribute query is performed on each component element in the component element group to obtain an element attribute group set, wherein the element attribute is the physical and chemical property of the component element, and the element attributes in the element attribute group correspond one to one with the element attribute categories in the element attribute category group;

[0039] Extracting element attribute categories in the element attribute category group in sequence, and extracting similar element attribute groups in the element attribute group set based on the element attribute categories, wherein similar element attributes in the similar element attribute groups come from different element attribute groups in the element attribute group set respectively;

[0040] According to the component proportion group and the similar element attribute group, the target element attribute is calculated, wherein the target element attribute is expressed as:

[0041]

[0042] Among them, P represents the target element attribute, n represents the number of component proportions in the component proportion group, and A i represents the proportion of the i-th component in the component proportion group, S i Indicates the i-th similar element attribute in the similar element attribute group;

[0043] The target element attributes of each element attribute category are summarized to obtain a target element attribute group.

[0044] Optionally, removing low-correlation features from the complete material feature set to obtain a high-correlation feature set includes:

[0045] Based on the complete material feature group set, a complete feature category group is set;

[0046] Extracting complete feature categories in the complete feature category group in sequence, and removing the complete feature categories from the complete feature category group to obtain a local feature category group;

[0047] According to the local feature category group, the complete material feature group set is screened to obtain a local material feature group set, wherein the number of local material features in the local material feature group is the same as the number of local feature categories in the local feature category group;

[0048] Based on the local material feature set, eigenvalue association analysis is performed to obtain category-associated eigenvalues;

[0049] Summarizing the category-associated feature values ​​corresponding to each complete feature category to obtain a category-associated feature value set, and identifying a low-associated feature value group in the category-associated feature value set based on a preset number of eliminated features, wherein the number of low-associated feature values ​​in the low-associated feature value group is the same as the number of eliminated features;

[0050] Identifying a low-correlation feature category group corresponding to a low-correlation feature value group in the complete feature category group, and removing the low-correlation feature category group from the complete feature category group to obtain a removed feature category group;

[0051] The complete feature category group is updated using the eliminated feature category group, and the step of sequentially extracting complete feature categories from the complete feature category group is returned until the number of eliminated feature categories in the eliminated feature category group is equal to the preset number of standard features, and the eliminated feature category group at this time is recorded as a high-correlation feature category group;

[0052] Based on the high correlation feature category group, a high correlation feature group set is extracted from the complete material feature group set, wherein the number of features in the high correlation feature group is the same as the number of correlation feature categories in the high correlation feature category group.

[0053] Optionally, performing feature value association analysis based on the local material feature set to obtain category-associated feature values ​​includes:

[0054] Based on a preset training division ratio, the local material feature set is divided to obtain a local training feature set and a local verification feature set;

[0055] The preset random forest model is trained using the local training feature set to obtain a simulated thermoelectric figure of merit model;

[0056] Extracting local thermoelectric merit sets and local feature set sets from the local verification feature set sets;

[0057] Inputting the local feature groups in the local feature group set into the simulated thermoelectric figure of merit model in sequence to obtain a simulated thermoelectric figure of merit set, wherein one simulated thermoelectric figure of merit corresponds to one local feature group;

[0058] According to the local thermoelectric merit set and the simulated thermoelectric merit set, the feature-associated eigenvalue set is calculated, where the feature-associated eigenvalue set is expressed as:

[0059] Y={Y1,…,Y j ,…Y m}, 1 <j<m

[0060]

[0061] Among them, Y represents the feature association feature value set, Y1 represents the first feature association value in the feature association feature value set, and Y j represents the jth feature association value, Y m represents the mth feature association value, m represents the number of feature association values ​​in the feature association feature value set, represents the jth local thermoelectric figure of merit in the local thermoelectric figure of merit set, represents the jth simulated thermoelectric figure of merit in the simulated thermoelectric figure of merit set, and |*| represents the absolute value symbol;

[0062] The average associated feature value of the feature associated feature value set is calculated, and the average associated feature value is recorded as the category associated feature value.

[0063] Optionally, the thermoelectric figure of merit prediction model group is verified by using the high correlation verification feature group set to obtain a prediction deviation coefficient group, including:

[0064] Perform the following processing on each thermoelectric figure of merit prediction model in the thermoelectric figure of merit prediction model group;

[0065] Inputting the highly correlated verification feature groups in the highly correlated verification feature group set into the thermoelectric figure of merit prediction model in sequence to obtain a predicted thermoelectric figure of merit group;

[0066] Extracting the high correlation thermoelectric figure of merit of each high correlation verification feature group in the high correlation verification feature group set to obtain a high correlation thermoelectric figure of merit group;

[0067] According to the predicted thermoelectric merit group and the highly correlated thermoelectric merit group, the prediction deviation factor is calculated using the following formula:

[0068]

[0069] Where B represents the prediction deviation factor, p represents the number of predicted thermoelectric merits in the predicted thermoelectric merit group, represents the ath predicted thermoelectric figure of merit in the predicted thermoelectric figure of merit group, represents the ath high-correlation thermoelectric figure of merit in the high-correlation thermoelectric figure of merit group;

[0070] The prediction deviation factors of each thermoelectric figure of merit prediction model are summarized to obtain a prediction deviation factor group, and a prediction deviation coefficient group is calculated according to the prediction deviation factor group, wherein the prediction deviation coefficient group is expressed as:

[0071] R={R1,…,R e ,…R s}, 1 <e<s

[0072]

[0073] Among them, R represents the prediction deviation coefficient group, R1 represents the first prediction deviation coefficient in the prediction deviation coefficient group, and R e represents the e-th prediction deviation coefficient, R s represents the sth prediction deviation coefficient, s represents the number of prediction deviation coefficients in the prediction deviation coefficient group or the number of prediction deviation factors in the prediction deviation factor group, B b Represents the bth prediction deviation factor in the prediction deviation factor group.

[0074] To achieve the above objectives, the present invention also provides a doped thermoelectric material performance optimization research system based on machine learning, comprising:

[0075] A doping data acquisition module is used to acquire an original thermoelectric material data set, wherein the original thermoelectric material data includes: an original material chemical formula and an original thermoelectric figure of merit, identify a data set to be cleaned in the original thermoelectric material data set, and remove the data set to be cleaned from the original thermoelectric material data set to obtain a doped thermoelectric material data set;

[0076] A correlation feature extraction module is used to extract doped thermoelectric material data in the doped thermoelectric material data set in turn, identify the doped material chemical formula and the doped thermoelectric figure of merit in the doped thermoelectric material data, and perform feature construction on the doped material chemical formula to obtain a thermoelectric material feature group, perform key value pairing on the thermoelectric material feature group and the doped thermoelectric figure of merit to obtain a complete material feature group, summarize the complete material feature group of each doped material chemical formula to obtain a complete material feature group set, and perform low correlation feature elimination on the complete material feature group set to obtain a high correlation feature group set;

[0077] A prediction model training module is used to divide the high correlation feature group sets into groups to obtain high correlation training feature group sets and high correlation verification feature group sets, wherein each high correlation feature group contains a doping material chemical formula, and the pre-acquired deep learning model group is trained using the high correlation training feature group sets to obtain a thermoelectric figure of merit prediction model group, wherein the thermoelectric figure of merit prediction models in the thermoelectric figure of merit prediction model group correspond one-to-one to the deep learning models in the deep learning model group;

[0078] The integrated model building module is used to verify the thermoelectric figure of merit prediction model group using a high-correlation verification feature group set to obtain a prediction deviation coefficient group, and based on the prediction deviation coefficient group, to assign prediction weights to the thermoelectric figure of merit prediction model group to obtain an integrated thermoelectric figure of merit prediction model.

[0079] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising:

[0080] a memory storing at least one instruction; and

[0081] A processor executes instructions stored in the memory to implement the above-mentioned method for optimizing the performance of doped thermoelectric materials based on machine learning.

[0082] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned machine learning-based doped thermoelectric material performance optimization research method.

[0083] In order to solve the problem described in the background technology, the present invention first identifies the data set to be cleaned in the original thermoelectric material data set, and removes the data set to be cleaned from the original thermoelectric material data set to obtain a doped thermoelectric material data set. This step can reduce the impact of noise on model training by identifying and removing abnormal values ​​or inaccurate data in the data set, thereby improving the accuracy and reliability of the model. Then, low-correlation features are removed from the complete material feature set to obtain a high-correlation feature set. This step can reduce redundant information in the model training process by removing low-correlation features in the complete material feature set, thereby improving the efficiency of model training and the accuracy of model prediction. Since only one deep learning model is used to predict the thermoelectric figure of merit, a large error will be generated. Therefore, the deep learning model group is trained using the high-correlation training feature set to obtain a thermoelectric figure of merit prediction model. The thermoelectric merit prediction model is constructed in this step, which greatly improves the accuracy of subsequent thermoelectric merit prediction. Since different models have different prediction deviations for thermoelectric merit, in order to make the prediction values ​​of different thermoelectric merit prediction models in the thermoelectric merit prediction model group correlated, the thermoelectric merit prediction model group can be verified to obtain a prediction deviation coefficient group, and based on the prediction deviation coefficient group, the thermoelectric merit prediction model group is predicted weighted to obtain an integrated thermoelectric merit prediction model. The establishment of the integrated thermoelectric merit prediction model makes different thermoelectric merit prediction models highly correlated, further improving the accuracy of thermoelectric merit prediction. Finally, performance optimization research is realized based on the integrated thermoelectric merit prediction model. This step can more accurately guide the design and optimization of doped thermoelectric materials through the prediction of the integrated model, thereby improving the thermoelectric performance of the materials. Therefore, the present invention can improve the efficiency and accuracy of performance optimization research of doped thermoelectric materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 A schematic diagram of a process for optimizing the performance of doped thermoelectric materials based on machine learning provided in one embodiment of the present invention;

[0085] Figure 2 A functional module diagram of a doped thermoelectric material performance optimization research system based on machine learning provided in one embodiment of the present invention;

[0086] Figure 3 A schematic diagram of the structure of an electronic device for implementing the method for researching performance optimization of doped thermoelectric materials based on machine learning provided in one embodiment of the present invention.

[0087] Description of reference numerals:

[0088] 1. Electronic device; 10. Processor; 11. Memory; 12. Bus.

[0089] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0090] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0091] The embodiment of the present application provides a method for optimizing the performance of doped thermoelectric materials based on machine learning. The execution subject of the method for optimizing the performance of doped thermoelectric materials based on machine learning includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided in the embodiment of the present application. In other words, the method for optimizing the performance of doped thermoelectric materials based on machine learning can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0092] Reference Figure 1 FIG. 1 is a flow chart of a method for optimizing the performance of doped thermoelectric materials based on machine learning according to an embodiment of the present invention. In this embodiment, the method for optimizing the performance of doped thermoelectric materials based on machine learning includes:

[0093] S1. Obtaining an original thermoelectric material data set, wherein the original thermoelectric material data includes: an original material chemical formula and an original thermoelectric figure of merit.

[0094] It can be understood that the original thermoelectric material data set includes a large number of thermoelectric material chemical formulas and thermoelectric figures of merit corresponding to the thermoelectric material chemical formulas. The original thermoelectric material data set can be obtained from online databases and related articles.

[0095] For example, Xiao Zhang is a materials researcher. One day, Xiao Zhang needs to optimize the performance of doped thermoelectric materials, so Xiao Zhang retrieves thermoelectric material data from the UCSB database and related thermoelectric material literature, obtains multiple thermoelectric material chemical formulas, and queries the corresponding thermoelectric figures of merit corresponding to multiple thermoelectric material chemical formulas in the corresponding database or literature, and summarizes these thermoelectric figures of merit with the thermoelectric material chemical formula to obtain the original thermoelectric material data.

[0096] S2. Identify the data set to be cleaned in the original thermoelectric material data set, and remove the data set to be cleaned from the original thermoelectric material data set to obtain the doped thermoelectric material data set.

[0097] It can be understood that the doped thermoelectric material data set refers to the original thermoelectric material data set after data cleaning, and the data cleaning includes chemical formula determination, numerical determination and doped material determination, wherein chemical formula determination refers to determining whether the chemical formula of the original material in the original thermoelectric material data set is written correctly and whether it is a harmful radioactive substance, numerical determination refers to determining whether the original thermoelectric material in the original thermoelectric material data set is within a normal numerical range, and doped material determination refers to determining whether the original thermoelectric material belongs to a doped thermoelectric material.

[0098] In detail, the step of identifying the data set to be cleaned in the original thermoelectric material data set includes:

[0099] Extracting original thermoelectric material data in the original thermoelectric material data set in sequence;

[0100] Determine whether the original material chemical formula in the original thermoelectric material data is a valid chemical formula;

[0101] If the original material chemical formula in the original thermoelectric material data is not a valid chemical formula, the original thermoelectric material data is recorded as material data to be cleaned;

[0102] If the original material chemical formula in the original thermoelectric material data is a valid chemical formula, determining whether the original thermoelectric figure of merit in the original thermoelectric material data is within a preset valid value range;

[0103] If the original thermoelectric figure of merit in the original thermoelectric material data is not within the valid value range, the original thermoelectric material data is recorded as the material data to be cleaned;

[0104] If the original thermoelectric figure of merit in the original thermoelectric material data is within the valid value range, determining whether the original material chemical formula in the original thermoelectric material data is a preset doping material chemical formula;

[0105] If the original material chemical formula in the original thermoelectric material data is not the doped material chemical formula, the original thermoelectric material data is recorded as the material data to be cleaned;

[0106] The data of the materials to be cleaned are aggregated to obtain a data set of materials to be cleaned.

[0107] It should be explained that the valid chemical formula refers to a chemical formula that is written correctly and is not a chemical formula of harmful radioactive substances, and the step of judging whether the original material chemical formula in the original thermoelectric material data is a valid chemical formula is manually judged.

[0108] For example, a raw material has the chemical formula Y o975 Yb 0.025 Ba2Cu3O Z After the judgment of the relevant personnel Xiao Zhang, the chemical formula of the original material is invalid. The chemical formula of a certain original material is Pr 1.85 Th 0.15 Cu1O 3.98 , Xiao Zhang judged that the chemical formula of the original material was a radioactive substance.

[0109] Furthermore, the valid numerical range refers to a constant range set artificially, for example, 0 to 5. When the value of the original thermoelectric figure of merit is not within the range of 0 to 5, it can be determined that the original thermoelectric figure of merit exceeds the normal data range and is abnormal data.

[0110] In detail, the determining whether the original material chemical formula in the original thermoelectric material data is a preset doping material chemical formula includes:

[0111] Get the minimum component content in the chemical formula of the original material;

[0112] Determining whether the minimum component content is greater than a preset doping element content;

[0113] If the minimum component content is not greater than the doping element content, the original material chemical formula is recorded as the doping material chemical formula.

[0114] It can be understood that the minimum component content refers to the content corresponding to the element with the least content in the chemical formula of the original material. For example, for the chemical formula of water, H2O, its constituent elements are H and O, where the contents of H and O are 2 / 3 and 1 / 3 respectively. The minimum content of 1 / 3 can be artificially taken as the minimum component content.

[0115] Furthermore, the doping element content refers to an artificially set content constant, for example: 2%.

[0116] S3. Extract the doped thermoelectric material data in the doped thermoelectric material data set in turn, identify the doped material chemical formula and the doped thermoelectric figure of merit in the doped thermoelectric material data, and perform feature construction on the doped material chemical formula to obtain a thermoelectric material feature group, perform key-value pairing on the thermoelectric material feature group and the doped thermoelectric figure of merit to obtain a complete material feature group.

[0117] It can be understood that the thermoelectric material characteristics refer to parameters used to represent the physicochemical properties of the chemical formula of the doped material, and the complete material characteristic group refers to a combination of the thermoelectric material characteristic group and the doping thermoelectric figure of merit.

[0118] In detail, the characteristic construction of the chemical formula of the doping material to obtain the characteristic group of the thermoelectric material includes:

[0119] Obtaining a basic element group of the doped thermoelectric material data set, wherein the basic element group is a combination of all elements appearing in the doped thermoelectric material data set, and there is no repeated element in the basic element group;

[0120] Identifying a component element group in a chemical formula of a doping material, and counting the component proportion of each component element in the component element group in the chemical formula of the doping material to obtain a component proportion group;

[0121] Setting a vacancy component vector according to the basic element group, wherein the length of the vacancy component vector is the same as the number of basic elements in the basic element group;

[0122] Based on the component element group, fill the empty component vector with the component proportion group at the same position to obtain a material component vector, wherein the length of the material component vector is the same as the length of the empty component vector, and the material component vector and the empty component vector represent the same elements at the same position;

[0123] According to the component element group, the chemical formula of the doping material is analyzed for constituent elements to obtain an element property vector, and the material component vector and the element property vector are combined to obtain a thermoelectric material characteristic group.

[0124] It can be understood that the basic elements refer to the combination of all elements appearing in the target thermoelectric material data set, and there are no repeated elements in the basic element group. The component elements refer to the combination of elements appearing in the chemical formula of the doping material, so the component element group should be a subset of the basic element group, and the component proportion refers to the composition ratio of the component elements in the component element group in the chemical formula of the doping material.

[0125] For example, for the doping material chemical formula: Y1Ba2Cu 2.99 Zn 0.01 O7, its component elements are: {Y, Ba, Cu, Zn, O}, and the component proportions of these component elements in the chemical formula of the doping material are: {1, 2, 2.99, 0.01, 7}.

[0126] It can be understood that the vacancy component vector refers to a vector used to represent the position information of the basic element group. The length of the vacancy component vector is the same as the number of basic elements in the basic element group, and each position in the vector is 0, but the basic element represented by each position of the vector is fixed. The material component vector refers to the vacancy component vector after the same position is filled, and the material component vector and the vacancy component vector represent the same element at the same position.

[0127] For example, a basic element group is: {Na, Eu, Cu, Sn, Pb, Sr, Te}, and the chemical formula of the doping material is: NaEu 0.03 Sn 0.02 Pb0.95 Te, the component element group and component proportion group of the chemical formula of the doping material are: {Na, Eu, Sn, Pb, Te} and (1, 0.03, 0.02, 0.95, 1), respectively. Then the component proportion group is used to fill the vacant component vector in the same position to obtain the material component vector: (1 0.03 00.02 0.95 0 1).

[0128] It should be explained that the element property vector refers to the parameters used to represent the physical and chemical properties of the chemical formula of the doping material. The specific method for obtaining the element property vector will be given in the subsequent steps.

[0129] In detail, the chemical formula of the doping material is analyzed for constituent elements according to the component element group to obtain the element property vector, including:

[0130] Setting an element attribute category group, and constructing a vacancy attribute vector according to the element attribute category group, wherein the length of the vacancy attribute vector is the same as the number of element attribute categories in the element attribute category group;

[0131] Analyze the element properties of the component element groups to obtain the target element attribute groups;

[0132] The target element attribute group is used to fill the empty attribute vector in the same position to obtain an element property vector, wherein the vector length of the element property vector is the same as the vector length of the empty attribute vector.

[0133] It can be understood that the element attribute category group refers to a combination of artificially set element attributes, wherein the element attribute refers to the physical and chemical properties of the element, such as electronegativity, atomic radius, ion ionization energy, atomic number, the number of electrons filled outside the nucleus, etc. These element attributes can be considered as important parameters affecting the thermoelectric figure of merit of the target thermoelectric material. It should be noted that the element attribute category group only contains the name of the element attribute, and does not contain any numerical information.

[0134] It should be explained that the vacant attribute vector refers to a vector used to represent the position information of the element attribute category, the target element attribute group refers to a combination of specific attribute values ​​of all element attribute categories of the component elements in the element attribute category group, and the detailed method for obtaining the target element attribute group will be given in the subsequent steps. The element property vector refers to the vacant attribute vector after filling in the same position, and the element property vector and the vacant attribute vector represent the same element attribute category at the same position.

[0135] In detail, the element property analysis of the component element group to obtain the target element attribute group includes:

[0136] Based on the element attribute category group, an element attribute query is performed on each component element in the component element group to obtain an element attribute group set, wherein the element attribute is the physical and chemical property of the component element, and the element attributes in the element attribute group correspond one to one with the element attribute categories in the element attribute category group;

[0137] Extracting element attribute categories in the element attribute category group in sequence, and extracting similar element attribute groups in the element attribute group set based on the element attribute categories, wherein similar element attributes in the similar element attribute groups come from different element attribute groups in the element attribute group set respectively;

[0138] According to the component proportion group and the similar element attribute group, the target element attribute is calculated, wherein the target element attribute is expressed as:

[0139]

[0140] Among them, P represents the target element attribute, n represents the number of component proportions in the component proportion group, and A i represents the proportion of the i-th component in the component proportion group, S i Indicates the i-th similar element attribute in the similar element attribute group;

[0141] The target element attributes of each element attribute category are summarized to obtain a target element attribute group.

[0142] It can be understood that the element attribute group set refers to a combination of specific attribute values ​​of all element attribute categories of the component element in the element attribute category group.

[0143] For example, an element property category group is: {electronegativity, atomic radius, ion ionization energy, atomic number, number of electrons filled outside the nucleus}, and a component element is Na. After searching relevant websites and literature, the specific values ​​of Na for each element property category in the element property category group are obtained: electronegativity is 0.93, atomic radius is 186 picometers, first ionization energy is 495.8 kilojoules per mole, atomic number is 11, and the number of electrons filled outside the nucleus is 1. Then the element property group of Na is: {0.93, 186, 495.8, 11, 1}.

[0144] It can be understood that the same type of element attribute group refers to a set of values ​​representing the same element attribute category in each element attribute group.

[0145] For example, the chemical formula of a doping material is: NaEu 0.03 Sn 0.02 Pb 0.95Te, then its component element group is: {Na, Eu, Sn, Pb, Te}. Then Xiao Zhang obtains the element attribute groups of these component elements. There are 5 element attribute groups in total. Then, an element attribute category is extracted from the element attribute category group, which is atomic radius. The values ​​of atomic radius in these 5 element attribute groups are: 186, 203, 141, 146 and 135, respectively. Their units are all picometers. Then the same type of element attribute groups are: (186, 203, 141, 146, 135).

[0146] It should be explained that the target element attribute refers to the value calculated by weighting the attributes of similar elements by component proportion.

[0147] S4. Summarize the complete material feature groups of each doping material chemical formula to obtain a complete material feature group set, and remove low-correlation features from the complete material feature group set to obtain a high-correlation feature group set.

[0148] It can be understood that the highly correlated feature set refers to features that are highly correlated with the thermoelectric figure of merit. Since there are a large number of features in the complete material feature set that are irrelevant to the thermoelectric figure of merit or have a low correlation with the thermoelectric figure of merit, in order to reduce the complexity of subsequent machine learning, these irrelevant or low-correlated features can be eliminated to improve the efficiency of subsequent data processing.

[0149] Furthermore, the method for eliminating low-correlation features is to use a random forest model to train the complete material feature set, and compare the deviation between the trained results and the thermoelectric figure of merit in the complete material feature set, and finally use the features with smaller deviations as high-correlation material features.

[0150] In detail, the process of eliminating low-correlation features from the complete material feature set to obtain a high-correlation feature set includes:

[0151] Based on the complete material feature group set, a complete feature category group is set;

[0152] Extracting complete feature categories in the complete feature category group in sequence, and removing the complete feature categories from the complete feature category group to obtain a local feature category group;

[0153] According to the local feature category group, the complete material feature group set is screened to obtain a local material feature group set, wherein the number of local material features in the local material feature group is the same as the number of local feature categories in the local feature category group;

[0154] Based on the local material feature set, eigenvalue association analysis is performed to obtain category-associated eigenvalues;

[0155] Summarizing the category-associated feature values ​​corresponding to each complete feature category to obtain a category-associated feature value set, and identifying a low-associated feature value group in the category-associated feature value set based on a preset number of eliminated features, wherein the number of low-associated feature values ​​in the low-associated feature value group is the same as the number of eliminated features;

[0156] Identifying a low-correlation feature category group corresponding to a low-correlation feature value group in the complete feature category group, and removing the low-correlation feature category group from the complete feature category group to obtain a removed feature category group;

[0157] The complete feature category group is updated using the eliminated feature category group, and the step of sequentially extracting complete feature categories from the complete feature category group is returned until the number of eliminated feature categories in the eliminated feature category group is equal to the preset number of standard features, and the eliminated feature category group at this time is recorded as a high-correlation feature category group;

[0158] Based on the high correlation feature category group, a high correlation feature group set is extracted from the complete material feature group set, wherein the number of features in the high correlation feature group is the same as the number of correlation feature categories in the high correlation feature category group.

[0159] It can be understood that the complete feature category group refers to the feature categories that appear in the complete material feature group set, and each complete material feature group in the complete material feature group set contains all the complete feature categories in the complete feature category group, the local feature category group refers to the complete feature category group after the complete feature categories are eliminated, and the local material feature group set refers to the complete material feature group set that only contains each local feature category in the local feature category group.

[0160] For example, a complete material feature set is: {B, C}, where B and C are two complete material feature sets in the complete material feature set, B is (zTB: (K1b, K2b)), C is (zTC: (K1c, K2c)), where zTB and zTC are the thermoelectric merits in group B and group C, respectively, K1b and K2b refer to the characteristic values ​​in group B, and their categories are: K1 and K2, respectively. Similarly, K1c and K2c refer to the thermoelectric merits in group B and group C, respectively. The eigenvalues ​​in C are of categories K1 and K2 respectively, so the complete feature category group is (K1, K2). Then, one of the complete feature categories K1 is eliminated to obtain the local feature category group (K2). The eigenvalues ​​corresponding to K2 in {B, C} are K2b and K2c, so the local material feature group set is {B', C'}, where B' is (zTB: (K2b)) and C' is (zTC: (K2c)).

[0161] It needs to be explained that in order to obtain feature categories that are highly correlated with the thermoelectric figure of merit, it is necessary to eliminate the feature categories one by one, and perform deep learning model simulation on the local material feature groups corresponding to the remaining feature categories, and compare the simulated thermoelectric figure of merit with the original thermoelectric figure of merit. When the simulated thermoelectric figure of merit has a lower correlation value with the original thermoelectric figure of merit, that is, the category-associated eigenvalue, it means that the feature category eliminated at this time has a higher correlation with the thermoelectric figure of merit. The detailed calculation process of the category-associated eigenvalue will be given later.

[0162] Furthermore, the number of eliminated features refers to an artificially set constant, which represents the number of complete feature category groups that are eliminated from the complete feature category group at a single time; the number of standard features refers to the minimum number of eliminated feature categories in the artificially set eliminated feature category group; the low-correlation feature value group refers to a combination of the f category-correlation feature values ​​with the lowest values ​​in the category-correlation feature value set, wherein f represents the number of eliminated features.

[0163] In detail, the characteristic value association analysis is performed based on the local material characteristic set to obtain the category-associated characteristic value, including:

[0164] Based on a preset training division ratio, the local material feature set is divided to obtain a local training feature set and a local verification feature set;

[0165] The preset random forest model is trained using the local training feature set to obtain a simulated thermoelectric figure of merit model;

[0166] Extracting local thermoelectric merit sets and local feature set sets from the local verification feature set sets;

[0167] Inputting the local feature groups in the local feature group set into the simulated thermoelectric figure of merit model in sequence to obtain a simulated thermoelectric figure of merit set, wherein one simulated thermoelectric figure of merit corresponds to one local feature group;

[0168] According to the local thermoelectric merit set and the simulated thermoelectric merit set, the feature-associated eigenvalue set is calculated, where the feature-associated eigenvalue set is expressed as:

[0169] Y={Y1,…,Y j ,…Y m}, 1 <j<m

[0170]

[0171] Among them, Y represents the feature association feature value set, Y1 represents the first feature association value in the feature association feature value set, and Y j represents the jth feature association value, Y mrepresents the mth feature association value, m represents the number of feature association values ​​in the feature association feature value set, represents the jth local thermoelectric figure of merit in the local thermoelectric figure of merit set, represents the jth simulated thermoelectric figure of merit in the simulated thermoelectric figure of merit set, and |*| represents the absolute value symbol;

[0172] The average associated feature value of the feature associated feature value set is calculated, and the average associated feature value is recorded as the category associated feature value.

[0173] It can be understood that the training division ratio refers to an artificially set proportional constant, which is used to represent the ratio of the data set used for model training and model verification. The local training feature set refers to the local material feature set used to train the model in the subsequent machine learning step. The local verification feature set refers to the local material feature set used to verify the model in the subsequent machine learning step. The simulated thermoelectric figure of merit model refers to a model that simulates the thermoelectric figure of merit of the input feature set. The characteristic value is input into the simulated thermoelectric figure of merit model, and the model will output a simulated thermoelectric figure of merit.

[0174] It should be explained that the local thermoelectric merit set refers to the set of thermoelectric merits in each local verification feature group in the local verification feature group set, the local feature group set refers to the set of feature groups in each local verification feature group in the local verification feature group set, the simulated thermoelectric merit set refers to the thermoelectric merit output by the simulated thermoelectric merit model, and the average associated eigenvalue refers to the average value of all feature associated eigenvalues ​​in the feature associated eigenvalue set.

[0175] S5. Divide the high correlation feature groups into groups to obtain high correlation training feature groups and high correlation verification feature groups, wherein each high correlation feature group includes a chemical formula of a doping material.

[0176] It can be understood that the highly correlated training feature set refers to a data set used to train the selected deep learning model in the subsequent machine learning step, and the highly correlated verification feature set refers to a data set used to verify the trained deep learning model in the subsequent machine learning step, wherein the ratio of group division is set manually.

[0177] S6. Use a highly correlated training feature set to train the pre-acquired deep learning model group to obtain a thermoelectric figure of merit prediction model group, wherein the thermoelectric figure of merit prediction models in the thermoelectric figure of merit prediction model group correspond one-to-one to the deep learning models in the deep learning model group.

[0178] It should be explained that the deep learning model group refers to a combination of multiple deep learning models, wherein the deep learning includes: LGB model, XGB model, RF model, DT model, etc., and the thermoelectric figure of merit prediction model refers to the deep learning model in the trained deep learning model group, and each thermoelectric figure of merit prediction model in the thermoelectric figure of merit prediction model group can perform thermoelectric figure of merit prediction independently.

[0179] Furthermore, using only one thermoelectric figure of merit prediction model to predict the thermoelectric figure of merit will result in a large error. Therefore, multiple thermoelectric figure of merit prediction models can be used to predict the thermoelectric figure of merit, and the weight relationship between the multiple thermoelectric figure of merit prediction models can be adjusted in combination with the multiple prediction results obtained, so that the obtained integrated thermoelectric figure of merit prediction model has extremely high accuracy.

[0180] S7. Use a highly correlated verification feature group set to verify the thermoelectric figure of merit prediction model group to obtain a prediction deviation coefficient group, and based on the prediction deviation coefficient group, assign prediction weights to the thermoelectric figure of merit prediction model group to obtain an integrated thermoelectric figure of merit prediction model.

[0181] It can be understood that the prediction deviation coefficient refers to a numerical value used to represent the degree of deviation between the predicted result and the actual result when the thermoelectric figure of merit prediction model predicts the thermoelectric figure of merit, and the integrated thermoelectric figure of merit prediction model refers to the weighted combination of the individual thermoelectric figure of merit prediction models in the thermoelectric figure of merit prediction model group.

[0182] Exemplarily, an integrated thermoelectric figure of merit prediction model is composed of three thermoelectric figure of merit prediction models, namely U1, U2 and U3, and the prediction deviation coefficients of these three models are T1, T2 and T3 respectively. After the feature group is input into the integrated thermoelectric figure of merit prediction model, U1, U2 and U3 will perform data processing on the feature group respectively to obtain three predicted thermoelectric figures of merit. Finally, the three predicted deviation coefficients are used to perform weighted summation on the three predicted thermoelectric figures of merit to obtain the final prediction result, wherein the weighted summation formula is: T1×R1+T2×R2+T3×R3, wherein R1, R2 and R3 are the above three predicted thermoelectric figures of merit respectively.

[0183] In detail, the thermoelectric figure of merit prediction model group is verified by using the high correlation verification feature group set to obtain a prediction deviation coefficient group, including:

[0184] Perform the following processing on each thermoelectric figure of merit prediction model in the thermoelectric figure of merit prediction model group;

[0185] Inputting the highly correlated verification feature groups in the highly correlated verification feature group set into the thermoelectric figure of merit prediction model in sequence to obtain a predicted thermoelectric figure of merit group;

[0186] Extracting the high correlation thermoelectric figure of merit of each high correlation verification feature group in the high correlation verification feature group set to obtain a high correlation thermoelectric figure of merit group;

[0187] According to the predicted thermoelectric merit group and the highly correlated thermoelectric merit group, the prediction deviation factor is calculated using the following formula:

[0188]

[0189] Where B represents the prediction deviation factor, p represents the number of predicted thermoelectric merits in the predicted thermoelectric merit group, represents the ath predicted thermoelectric figure of merit in the predicted thermoelectric figure of merit group, represents the ath high-correlation thermoelectric figure of merit in the high-correlation thermoelectric figure of merit group;

[0190] The prediction deviation factors of each thermoelectric figure of merit prediction model are summarized to obtain a prediction deviation factor group, and a prediction deviation coefficient group is calculated according to the prediction deviation factor group, wherein the prediction deviation coefficient group is expressed as:

[0191] R={R1,…,R e ,…R s}, 1 <e<s

[0192]

[0193] Among them, R represents the prediction deviation coefficient group, R1 represents the first prediction deviation coefficient in the prediction deviation coefficient group, and R e represents the e-th prediction deviation coefficient, R s represents the sth prediction deviation coefficient, s represents the number of prediction deviation coefficients in the prediction deviation coefficient group or the number of prediction deviation factors in the prediction deviation factor group, B b Represents the bth prediction deviation factor in the prediction deviation factor group.

[0194] It can be understood that the predicted thermoelectric figure of merit refers to the thermoelectric figure of merit output by the thermoelectric figure of merit prediction model, the high-correlation thermoelectric figure of merit refers to the thermoelectric figure of merit in the high-correlation verification feature group, and the prediction deviation factor refers to a numerical value used to represent the degree of prediction deviation between the prediction result of the thermoelectric figure of merit prediction model and the actual high-correlation thermoelectric figure of merit.

[0195] S8. Based on the integrated thermoelectric figure of merit prediction model, a machine learning-based performance optimization study of doped thermoelectric materials is implemented.

[0196] It should be explained that when the chemical formula of a thermoelectric material is known, the corresponding high-correlation feature group can be obtained based on the chemical formula of the thermoelectric material, wherein the method for obtaining the high-correlation feature group has been given in detail above, and then the high-correlation feature group is input into the integrated thermoelectric figure of merit prediction model, which will output the predicted thermoelectric figure of merit of the thermoelectric material.

[0197] In order to solve the problem described in the background technology, the present invention first identifies the data set to be cleaned in the original thermoelectric material data set, and removes the data set to be cleaned from the original thermoelectric material data set to obtain a doped thermoelectric material data set. This step can reduce the impact of noise on model training by identifying and removing abnormal values ​​or inaccurate data in the data set, thereby improving the accuracy and reliability of the model. Then, low-correlation features are removed from the complete material feature set to obtain a high-correlation feature set. This step can reduce redundant information in the model training process by removing low-correlation features in the complete material feature set, thereby improving the efficiency of model training and the accuracy of model prediction. Since only one deep learning model is used to predict the thermoelectric figure of merit, a large error will be generated. Therefore, the deep learning model group is trained using the high-correlation training feature set to obtain a thermoelectric figure of merit prediction model. The thermoelectric merit prediction model is constructed in this step, which greatly improves the accuracy of subsequent thermoelectric merit prediction. Since different models have different prediction deviations for thermoelectric merit, in order to make the prediction values ​​of different thermoelectric merit prediction models in the thermoelectric merit prediction model group correlated, the thermoelectric merit prediction model group can be verified to obtain a prediction deviation coefficient group, and based on the prediction deviation coefficient group, the thermoelectric merit prediction model group is predicted weighted to obtain an integrated thermoelectric merit prediction model. The establishment of the integrated thermoelectric merit prediction model makes different thermoelectric merit prediction models highly correlated, further improving the accuracy of thermoelectric merit prediction. Finally, performance optimization research is realized based on the integrated thermoelectric merit prediction model. This step can more accurately guide the design and optimization of doped thermoelectric materials through the prediction of the integrated model, thereby improving the thermoelectric performance of the materials. Therefore, the present invention can improve the efficiency and accuracy of performance optimization research of doped thermoelectric materials.

[0198] like Figure 2 , is a functional module diagram of a doped thermoelectric material performance optimization research system based on machine learning provided by one embodiment of the present invention.

[0199] The doped thermoelectric material performance optimization research system 100 based on machine learning of the present invention can be installed in an electronic device. According to the functions implemented, the doped thermoelectric material performance optimization research system 100 based on machine learning can include a doping data acquisition module 101, an associated feature extraction module 102, a prediction model training module 103 and an integrated model construction module 104. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0200] The doping data acquisition module 101 is used to acquire an original thermoelectric material data set, wherein the original thermoelectric material data includes: an original material chemical formula and an original thermoelectric figure of merit, identify a data set to be cleaned in the original thermoelectric material data set, and remove the data set to be cleaned from the original thermoelectric material data set to obtain a doped thermoelectric material data set;

[0201] The associated feature extraction module 102 is used to extract doped thermoelectric material data in the doped thermoelectric material data set in turn, identify the doped material chemical formula and the doped thermoelectric figure of merit in the doped thermoelectric material data, and perform feature construction on the doped material chemical formula to obtain a thermoelectric material feature group, perform key value pairing on the thermoelectric material feature group and the doped thermoelectric figure of merit to obtain a complete material feature group, summarize the complete material feature group of each doped material chemical formula to obtain a complete material feature group set, and perform low-associated feature elimination on the complete material feature group set to obtain a high-associated feature group set;

[0202] The prediction model training module 103 is used to divide the high correlation feature group set into groups to obtain a high correlation training feature group set and a high correlation verification feature group set, wherein each high correlation feature group includes a doping material chemical formula, and the pre-acquired deep learning model group is trained using the high correlation training feature group set to obtain a thermoelectric figure of merit prediction model group, wherein the thermoelectric figure of merit prediction models in the thermoelectric figure of merit prediction model group correspond one-to-one to the deep learning models in the deep learning model group;

[0203] The integrated model construction module 104 is used to verify the thermoelectric figure of merit prediction model group using a high correlation verification feature group set to obtain a prediction deviation coefficient group, and based on the prediction deviation coefficient group, perform prediction weight allocation on the thermoelectric figure of merit prediction model group to obtain an integrated thermoelectric figure of merit prediction model.

[0204] In detail, each module in the doped thermoelectric material performance optimization research system 100 based on machine learning in the embodiment of the present invention is used in the same manner as described above. Figure 1 The same technical means are used as the research method for performance optimization of doped thermoelectric materials based on machine learning described in the previous section, and can produce the same technical effects, so they will not be repeated here.

[0205] like Figure 3 , is a schematic diagram of the structure of an electronic device for implementing a method for researching performance optimization of doped thermoelectric materials based on machine learning provided by one embodiment of the present invention.

[0206] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a method program for optimizing the performance of doped thermoelectric materials based on machine learning.

[0207] Among them, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. The memory 11 can be an internal storage unit of the electronic device 1 in some embodiments, such as a mobile hard disk of the electronic device 1. The memory 11 can also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 11 also includes an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can not only be used to store application software and various types of data installed in the electronic device 1, such as the code of the program of the method for optimizing the performance of doped thermoelectric materials based on machine learning, but also can be used to temporarily store data that has been output or is to be output.

[0208] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (such as a doped thermoelectric material performance optimization research method program based on machine learning, etc.), and calls data stored in the memory 11 to execute various functions of the electronic device 1 and process data.

[0209] The bus 12 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 may be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0210] Figure 3 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 3The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0211] For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management system, so that the power management system can realize functions such as charging management, discharging management, and power consumption management. The power source may also include any components such as one or more DC or AC power sources, recharging systems, power failure detection circuits, power converters or inverters, and power status indicators. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0212] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0213] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0214] The program of the method for researching performance optimization of doped thermoelectric materials based on machine learning stored in the memory 11 in the electronic device 1 is a combination of multiple instructions. When running in the processor 10, the following can be achieved:

[0215] Acquire an original thermoelectric material data set, wherein the original thermoelectric material data includes: an original material chemical formula and an original thermoelectric figure of merit;

[0216] Identifying a data set to be cleaned in the original thermoelectric material data set, and removing the data set to be cleaned from the original thermoelectric material data set to obtain a doped thermoelectric material data set;

[0217] Extracting doped thermoelectric material data in the doped thermoelectric material data set in turn, identifying the doped material chemical formula and the doped thermoelectric figure of merit in the doped thermoelectric material data, and constructing features for the doped material chemical formula to obtain a thermoelectric material feature group, and performing key value pairing on the thermoelectric material feature group and the doped thermoelectric figure of merit to obtain a complete material feature group;

[0218] Summarizing the complete material feature groups of each doping material chemical formula to obtain a complete material feature group set, and removing low-correlation features from the complete material feature group set to obtain a high-correlation feature group set;

[0219] Dividing the high correlation feature group sets into groups to obtain high correlation training feature group sets and high correlation verification feature group sets, wherein each high correlation feature group contains a chemical formula of a doping material;

[0220] Using a highly correlated training feature group set to train the pre-acquired deep learning model group, a thermoelectric figure of merit prediction model group is obtained, wherein the thermoelectric figure of merit prediction models in the thermoelectric figure of merit prediction model group correspond one-to-one to the deep learning models in the deep learning model group;

[0221] The thermoelectric figure of merit prediction model group is verified by using a highly correlated verification feature group set to obtain a prediction deviation coefficient group, and based on the prediction deviation coefficient group, prediction weights are assigned to the thermoelectric figure of merit prediction model group to obtain an integrated thermoelectric figure of merit prediction model;

[0222] Based on the integrated thermoelectric figure of merit prediction model, performance optimization research of doped thermoelectric materials based on machine learning is realized.

[0223] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 3 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0224] Furthermore, if the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or system capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0225] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement:

[0226] Acquire an original thermoelectric material data set, wherein the original thermoelectric material data includes: an original material chemical formula and an original thermoelectric figure of merit;

[0227] Identifying a data set to be cleaned in the original thermoelectric material data set, and removing the data set to be cleaned from the original thermoelectric material data set to obtain a doped thermoelectric material data set;

[0228] Extracting doped thermoelectric material data in the doped thermoelectric material data set in turn, identifying the doped material chemical formula and the doped thermoelectric figure of merit in the doped thermoelectric material data, and constructing features for the doped material chemical formula to obtain a thermoelectric material feature group, and performing key value pairing on the thermoelectric material feature group and the doped thermoelectric figure of merit to obtain a complete material feature group;

[0229] Summarizing the complete material feature groups of each doping material chemical formula to obtain a complete material feature group set, and removing low-correlation features from the complete material feature group set to obtain a high-correlation feature group set;

[0230] Dividing the high correlation feature group sets into groups to obtain high correlation training feature group sets and high correlation verification feature group sets, wherein each high correlation feature group contains a chemical formula of a doping material;

[0231] Using a highly correlated training feature group set to train the pre-acquired deep learning model group, a thermoelectric figure of merit prediction model group is obtained, wherein the thermoelectric figure of merit prediction models in the thermoelectric figure of merit prediction model group correspond one-to-one to the deep learning models in the deep learning model group;

[0232] The thermoelectric figure of merit prediction model group is verified by using a highly correlated verification feature group set to obtain a prediction deviation coefficient group, and based on the prediction deviation coefficient group, prediction weights are assigned to the thermoelectric figure of merit prediction model group to obtain an integrated thermoelectric figure of merit prediction model;

[0233] Based on the integrated thermoelectric figure of merit prediction model, performance optimization research of doped thermoelectric materials based on machine learning is realized.

[0234] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative, and actual implementation may have other division methods.

[0235] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0236] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0237] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0238] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for optimizing the performance of doped thermoelectric materials based on machine learning, characterized in that: The method comprises: Acquire an original thermoelectric material data set, wherein the original thermoelectric material data includes: an original material chemical formula and an original thermoelectric figure of merit; Identifying a data set to be cleaned in the original thermoelectric material data set, and removing the data set to be cleaned from the original thermoelectric material data set to obtain a doped thermoelectric material data set; Extracting doped thermoelectric material data in the doped thermoelectric material data set in turn, identifying the doped material chemical formula and the doped thermoelectric figure of merit in the doped thermoelectric material data, and constructing features for the doped material chemical formula to obtain a thermoelectric material feature group, and performing key value pairing on the thermoelectric material feature group and the doped thermoelectric figure of merit to obtain a complete material feature group; Summarizing the complete material feature groups of each doping material chemical formula to obtain a complete material feature group set, and removing low-correlation features from the complete material feature group set to obtain a high-correlation feature group set; Dividing the high correlation feature group sets into groups to obtain high correlation training feature group sets and high correlation verification feature group sets, wherein each high correlation feature group contains a chemical formula of a doping material; Using a highly correlated training feature group set to train the pre-acquired deep learning model group, a thermoelectric figure of merit prediction model group is obtained, wherein the thermoelectric figure of merit prediction models in the thermoelectric figure of merit prediction model group correspond one-to-one to the deep learning models in the deep learning model group; The thermoelectric figure of merit prediction model group is verified by using a highly correlated verification feature group set to obtain a prediction deviation coefficient group, and based on the prediction deviation coefficient group, prediction weights are assigned to the thermoelectric figure of merit prediction model group to obtain an integrated thermoelectric figure of merit prediction model; Based on the integrated thermoelectric figure of merit prediction model, performance optimization research of doped thermoelectric materials based on machine learning is realized.

2. The method for optimizing performance of doped thermoelectric materials based on machine learning according to claim 1, characterized in that: The step of identifying a data set to be cleaned in the original thermoelectric material data set comprises: Extracting original thermoelectric material data in the original thermoelectric material data set in sequence; Determine whether the original material chemical formula in the original thermoelectric material data is a valid chemical formula; If the original material chemical formula in the original thermoelectric material data is not a valid chemical formula, the original thermoelectric material data is recorded as material data to be cleaned; If the original material chemical formula in the original thermoelectric material data is a valid chemical formula, determining whether the original thermoelectric figure of merit in the original thermoelectric material data is within a preset valid value range; If the original thermoelectric figure of merit in the original thermoelectric material data is not within the valid value range, the original thermoelectric material data is recorded as the material data to be cleaned; If the original thermoelectric figure of merit in the original thermoelectric material data is within the valid value range, determining whether the original material chemical formula in the original thermoelectric material data is a preset doping material chemical formula; If the original material chemical formula in the original thermoelectric material data is not the doped material chemical formula, the original thermoelectric material data is recorded as the material data to be cleaned; The data of the materials to be cleaned are aggregated to obtain a data set of materials to be cleaned.

3. The method for optimizing the performance of doped thermoelectric materials based on machine learning according to claim 2, characterized in that: The determining whether the original material chemical formula in the original thermoelectric material data is a preset doping material chemical formula includes: Get the minimum component content in the chemical formula of the original material; Determining whether the minimum component content is greater than a preset doping element content; If the minimum component content is not greater than the doping element content, the original material chemical formula is recorded as the doping material chemical formula.

4. The method for optimizing performance of doped thermoelectric materials based on machine learning according to claim 3, characterized in that: The characteristic construction of the chemical formula of the doping material to obtain a thermoelectric material characteristic group includes: Obtaining a basic element group of the doped thermoelectric material data set, wherein the basic element group is a combination of all elements appearing in the doped thermoelectric material data set, and there is no repeated element in the basic element group; Identifying a component element group in a chemical formula of a doping material, and counting the component proportion of each component element in the component element group in the chemical formula of the doping material to obtain a component proportion group; Setting a vacancy component vector according to the basic element group, wherein the length of the vacancy component vector is the same as the number of basic elements in the basic element group; Based on the component element group, fill the empty component vector with the component proportion group at the same position to obtain a material component vector, wherein the length of the material component vector is the same as the length of the empty component vector, and the material component vector and the empty component vector represent the same elements at the same position; According to the component element group, the chemical formula of the doping material is analyzed for constituent elements to obtain an element property vector, and the material component vector and the element property vector are combined to obtain a thermoelectric material characteristic group.

5. The method for optimizing performance of doped thermoelectric materials based on machine learning according to claim 4, characterized in that: The step of performing element analysis on the chemical formula of the doping material according to the component element group to obtain the element property vector includes: Setting an element attribute category group, and constructing a vacancy attribute vector according to the element attribute category group, wherein the length of the vacancy attribute vector is the same as the number of element attribute categories in the element attribute category group; Analyze the element properties of the component element groups to obtain the target element attribute groups; The target element attribute group is used to fill the empty attribute vector in the same position to obtain an element property vector, wherein the vector length of the element property vector is the same as the vector length of the empty attribute vector.

6. The method for optimizing performance of doped thermoelectric materials based on machine learning according to claim 5, characterized in that: The element property analysis of the component element group to obtain the target element attribute group includes: Based on the element attribute category group, an element attribute query is performed on each component element in the component element group to obtain an element attribute group set, wherein the element attribute is the physical and chemical property of the component element, and the element attributes in the element attribute group correspond one to one with the element attribute categories in the element attribute category group; Extracting element attribute categories in the element attribute category group in sequence, and extracting similar element attribute groups in the element attribute group set based on the element attribute categories, wherein similar element attributes in the similar element attribute groups come from different element attribute groups in the element attribute group set respectively; According to the component proportion group and the similar element attribute group, the target element attribute is calculated, wherein the target element attribute is expressed as: Among them, P represents the target element attribute, n represents the number of component proportions in the component proportion group, and A i represents the proportion of the i-th component in the component proportion group, S i Represents the i-th similar element attribute in the similar element attribute group; The target element attributes of each element attribute category are summarized to obtain a target element attribute group.

7. The method for optimizing performance of doped thermoelectric materials based on machine learning according to claim 6, characterized in that: The step of removing low-correlation features from the complete material feature set to obtain a high-correlation feature set includes: Based on the complete material feature group set, a complete feature category group is set; Extracting complete feature categories in the complete feature category group in sequence, and removing the complete feature categories from the complete feature category group to obtain a local feature category group; According to the local feature category group, the complete material feature group set is screened to obtain a local material feature group set, wherein the number of local material features in the local material feature group is the same as the number of local feature categories in the local feature category group; Based on the local material feature set, eigenvalue association analysis is performed to obtain category-associated eigenvalues; Summarizing the category-associated feature values ​​corresponding to each complete feature category to obtain a category-associated feature value set, and identifying a low-associated feature value group in the category-associated feature value set based on a preset number of eliminated features, wherein the number of low-associated feature values ​​in the low-associated feature value group is the same as the number of eliminated features; Identifying a low-correlation feature category group corresponding to a low-correlation feature value group in the complete feature category group, and removing the low-correlation feature category group from the complete feature category group to obtain a removed feature category group; The complete feature category group is updated using the eliminated feature category group, and the step of sequentially extracting complete feature categories from the complete feature category group is returned until the number of eliminated feature categories in the eliminated feature category group is equal to the preset number of standard features, and the eliminated feature category group at this time is recorded as a high-correlation feature category group; Based on the high correlation feature category group, a high correlation feature group set is extracted from the complete material feature group set, wherein the number of features in the high correlation feature group is the same as the number of correlation feature categories in the high correlation feature category group.

8. The method for optimizing performance of doped thermoelectric materials based on machine learning according to claim 7, characterized in that: The method of performing feature value association analysis based on the local material feature set to obtain category-associated feature values ​​includes: Based on a preset training division ratio, the local material feature set is divided to obtain a local training feature set and a local verification feature set; The preset random forest model is trained using the local training feature set to obtain a simulated thermoelectric figure of merit model; Extracting local thermoelectric merit sets and local feature set sets from the local verification feature set sets; Inputting the local feature groups in the local feature group set into the simulated thermoelectric figure of merit model in sequence to obtain a simulated thermoelectric figure of merit set, wherein one simulated thermoelectric figure of merit corresponds to one local feature group; According to the local thermoelectric merit set and the simulated thermoelectric merit set, the feature-associated eigenvalue set is calculated, where the feature-associated eigenvalue set is expressed as: Y={Y1,…,Y j ,…Y m },1 <j<m Among them, Y represents the feature association feature value set, Y1 represents the first feature association value in the feature association feature value set, and Y j represents the jth feature association value, Y m represents the mth feature association value, m represents the number of feature association values ​​in the feature association feature value set, represents the jth local thermoelectric figure of merit in the local thermoelectric figure of merit set, represents the jth simulated thermoelectric figure of merit in the simulated thermoelectric figure of merit set, and |*| represents the absolute value symbol; The average associated feature value of the feature associated feature value set is calculated, and the average associated feature value is recorded as the category associated feature value.

9. The method for optimizing performance of doped thermoelectric materials based on machine learning according to claim 8, characterized in that: The method of using the highly correlated verification feature group set to verify the thermoelectric figure of merit prediction model group to obtain a prediction deviation coefficient group includes: Perform the following processing on each thermoelectric figure of merit prediction model in the thermoelectric figure of merit prediction model group; Inputting the highly correlated verification feature groups in the highly correlated verification feature group set into the thermoelectric figure of merit prediction model in sequence to obtain a predicted thermoelectric figure of merit group; Extracting the high correlation thermoelectric figure of merit of each high correlation verification feature group in the high correlation verification feature group set to obtain a high correlation thermoelectric figure of merit group; According to the predicted thermoelectric merit group and the highly correlated thermoelectric merit group, the prediction deviation factor is calculated using the following formula: Where B represents the prediction deviation factor, p represents the number of predicted thermoelectric merits in the predicted thermoelectric merit group, represents the ath predicted thermoelectric figure of merit in the predicted thermoelectric figure of merit group, represents the ath high-correlation thermoelectric figure of merit in the high-correlation thermoelectric figure of merit group; The prediction deviation factors of each thermoelectric figure of merit prediction model are summarized to obtain a prediction deviation factor group, and a prediction deviation coefficient group is calculated according to the prediction deviation factor group, wherein the prediction deviation coefficient group is expressed as: R={R1,…,R e ,…R s },1<e<s Among them, R represents the prediction deviation coefficient group, R1 represents the first prediction deviation coefficient in the prediction deviation coefficient group, and R e represents the e-th prediction deviation coefficient, R s represents the sth prediction deviation coefficient, s represents the number of prediction deviation coefficients in the prediction deviation coefficient group or the number of prediction deviation factors in the prediction deviation factor group, B b Represents the bth prediction deviation factor in the prediction deviation factor group.

10. A doped thermoelectric material performance optimization research system based on machine learning, characterized in that: The system comprises: A doping data acquisition module is used to acquire an original thermoelectric material data set, wherein the original thermoelectric material data includes: an original material chemical formula and an original thermoelectric figure of merit, identify a data set to be cleaned in the original thermoelectric material data set, and remove the data set to be cleaned from the original thermoelectric material data set to obtain a doped thermoelectric material data set; A correlation feature extraction module is used to extract doped thermoelectric material data in the doped thermoelectric material data set in turn, identify the doped material chemical formula and the doped thermoelectric figure of merit in the doped thermoelectric material data, and perform feature construction on the doped material chemical formula to obtain a thermoelectric material feature group, perform key value pairing on the thermoelectric material feature group and the doped thermoelectric figure of merit to obtain a complete material feature group, summarize the complete material feature group of each doped material chemical formula to obtain a complete material feature group set, and perform low correlation feature elimination on the complete material feature group set to obtain a high correlation feature group set; A prediction model training module is used to divide the high correlation feature group sets into groups to obtain high correlation training feature group sets and high correlation verification feature group sets, wherein each high correlation feature group contains a doping material chemical formula, and the pre-acquired deep learning model group is trained using the high correlation training feature group sets to obtain a thermoelectric figure of merit prediction model group, wherein the thermoelectric figure of merit prediction models in the thermoelectric figure of merit prediction model group correspond one-to-one to the deep learning models in the deep learning model group; The integrated model building module is used to verify the thermoelectric figure of merit prediction model group using a high-correlation verification feature group set to obtain a prediction deviation coefficient group, and based on the prediction deviation coefficient group, to assign prediction weights to the thermoelectric figure of merit prediction model group to obtain an integrated thermoelectric figure of merit prediction model.

Images