High thermoelectric figure of merit material reverse design and system based on deep generative model

Abstract

The application relates to the fields of machine learning and new material prediction, and relates to a high-thermoelectric figure-of-merit material reverse design method and system based on a deep generative model, which comprises the following steps: obtaining a thermoelectric material dataset, performing temperature division operation on a plurality of target material data to obtain a low-temperature material data, a medium-temperature material data and a high-temperature material data, performing subscript interpolation operation on the low-temperature material data, the medium-temperature material data and the high-temperature material data to obtain Z expansion matrix groups, constructing a generative adversarial network by using a variational autoencoder and a true-false discrimination model, performing adversarial training on the generative adversarial network to obtain a target adversarial network, generating a plurality of target material chemical formulas by using a preset target zt value, a preset target temperature, the target adversarial network and a conditional discrimination model, and completing the reverse design of the high-thermoelectric figure-of-merit material. The application can improve the accuracy of the deep generative model in predicting the high-thermoelectric figure-of-merit material.

Description

TECHNICAL FIELD

[0001] The present application relates to the fields of machine learning and new material prediction, and in particular to a high-thermoelectric figure of merit material reverse design method and system based on a deep generative model, an electronic device and a computer readable storage medium. BACKGROUND

[0002] Based on the Seebeck effect and the Peltier effect, thermoelectric materials are materials that can directly convert heat energy into electrical energy. Compared with traditional energy conversion methods, thermoelectric materials have the advantages of high efficiency, environmental protection, reliability, and low maintenance cost.

[0003] At present, the method for predicting high-thermoelectric figure of merit thermoelectric materials usually uses a thermoelectric material database to train a generative adversarial network and a variational autoencoder, and then uses the trained generative adversarial network and variational autoencoder to predict high-thermoelectric figure of merit thermoelectric materials.

[0004] Although the existing generative adversarial network and variational autoencoder can achieve prediction of high-thermoelectric figure of merit materials, the amount of data in the thermoelectric material database is insufficient to support the training required by the model, and the number of thermoelectric materials in different temperature ranges is different. During the training of the model, the problem of overfitting often occurs. At the same time, the discriminant model in the generative adversarial network only judges the authenticity and falsity of the generated thermoelectric material chemical formula, and does not judge the thermoelectric figure of merit and the optimal temperature of the generated thermoelectric material, so that the generated thermoelectric material may not have the thermoelectric figure of merit and the optimal temperature required by the researchers. Therefore, how to expand the data in the thermoelectric material database and how to judge the thermoelectric figure of merit and the optimal temperature of the generated thermoelectric material have become problems to be solved. SUMMARY

[0005] The present application provides a high-thermoelectric figure of merit material reverse design method based on a deep generative model and a computer readable storage medium, which mainly aims to improve the accuracy of the deep generative model in predicting high-thermoelectric figure of merit materials.

[0006] To achieve the above-mentioned purpose, the present application provides a high-thermoelectric figure of merit material reverse design method based on a deep generative model, which comprises:

[0007] obtaining a thermoelectric material dataset, wherein the thermoelectric material dataset comprises a plurality of thermoelectric material data, and the thermoelectric material data comprises a material chemical formula, a maximum zt value, an optimal temperature and an index value range;

[0008] performing a chemical formula screening operation on the plurality of thermoelectric material data to obtain a plurality of target material data, performing a temperature division operation on the plurality of target material data to obtain a low-temperature material data, a medium-temperature material data and Data on high-temperature materials;

[0009] right Data on low-temperature materials, Data on medium-temperature materials and All high-temperature material data were subjected to subscript interpolation to obtain A low-temperature interpolated data, B medium-temperature interpolated data, and C high-temperature interpolated data;

[0010] Chemical formula quantization was performed on A low-temperature interpolation data, B medium-temperature interpolation data, and C high-temperature interpolation data to obtain... One extended matrix group;

[0011] use Each set of expanded matrices is used to train the pre-constructed first discrimination model to obtain the conditional discrimination model;

[0012] Utilizing pre-built variational autoencoders and A set of extended matrices is used to train the pre-constructed second discrimination model to obtain a true / false discrimination model, wherein the variational autoencoder includes an encoder and a decoder;

[0013] Generative Adversarial Networks (GANs) are constructed using variational autoencoders and a real / fake discrimination model. The GANs are then subjected to adversarial training to obtain a target adversarial network. The target adversarial network includes a target variational autoencoder, which in turn includes a target encoder and a target decoder.

[0014] By using preset target zt values, preset target temperatures, target adversarial networks, and conditional discrimination models, multiple target material chemical formulas are generated, thus completing the reverse design of high thermoelectric figure of merit materials.

[0015] Optionally, the chemical sieving operation on multiple thermoelectric material data to obtain multiple target material data includes:

[0016] For each thermoelectric material data point in the multiple thermoelectric material data sets, perform the following operation:

[0017] Identify the number of elements in the chemical formula of thermoelectric materials, and compare the number of elements with a preset first quantity threshold and a preset second quantity threshold, wherein the first quantity threshold is less than the second quantity threshold;

[0018] If the number of elements is greater than or equal to the first quantity threshold and the number of elements is less than or equal to the second quantity threshold, then the thermoelectric material data is recorded as the target material data;

[0019] By summarizing the target material data, multiple target material data sets are obtained.

[0020] Optionally, the temperature partitioning operation is performed on multiple target material data to obtain... Data on low-temperature materials, Data on medium-temperature materials and Data on high-temperature materials, including:

[0021] For each target material data point in a set of multiple target material data points, perform the following operation:

[0022] Compare the optimal temperature corresponding to the target material data with the preset first temperature and the preset second temperature, wherein the first temperature is lower than the second temperature;

[0023] If the optimal temperature is less than or equal to the first temperature, then the target material data corresponding to the optimal temperature is recorded as low-temperature material data;

[0024] If the optimal temperature is greater than the first temperature and less than the second temperature, then the target material data corresponding to the optimal temperature is recorded as medium-temperature material data.

[0025] If the optimal temperature is greater than or equal to the second temperature, then the target material data corresponding to the optimal temperature is recorded as high-temperature material data;

[0026] By summarizing the data for low-temperature materials, medium-temperature materials, and high-temperature materials respectively, we can obtain... Data on low-temperature materials, Data on medium-temperature materials and Data on high-temperature materials.

[0027] Optionally, the pair Data on low-temperature materials, Data on medium-temperature materials and All high-temperature material data were subjected to subscript interpolation, resulting in A sets of low-temperature interpolated data, B sets of medium-temperature interpolated data, and C sets of high-temperature interpolated data, including:

[0028] The low-temperature interpolation ratio, medium-temperature interpolation ratio, and high-temperature interpolation ratio are calculated based on the preset target data quantity. The calculation formulas are as follows:

[0029]

[0030] in, , and These represent the low-temperature interpolation ratio, the medium-temperature interpolation ratio, and the high-temperature interpolation ratio, respectively. For the target data quantity, This represents the rounding up operation;

[0031] Using low-temperature interpolation ratio Perform subscript interpolation on the low-temperature material data to obtain A low-temperature interpolated data, based on the medium-temperature interpolation ratio and... The medium-temperature material data acquisition obtains B medium-temperature interpolation data based on a high-temperature interpolation ratio and The high-temperature material data acquisition obtains C high-temperature interpolation data.

[0032] Optionally, the low-temperature interpolation ratio is used to perform an index interpolation operation on the A low-temperature material data to obtain A low-temperature interpolation data, including:

[0033] For each of the A low-temperature material data, the following operations are performed:

[0034] The low-temperature sampling interval is calculated according to the low-temperature interpolation ratio and the index value range corresponding to the low-temperature material data, and the calculation formula is as follows:

[0035]

[0036] wherein, is the low-temperature sampling interval, and are the maximum value and the minimum value in the index value range, respectively, is the low-temperature interpolation ratio;

[0037] The uniform sampling operation is performed on the index value range corresponding to the low-temperature material data based on the low-temperature sampling interval to obtain m index values, and for each of the m index values, the following operations are performed:

[0038] The index variable in the material chemical formula corresponding to the low-temperature material data is identified, and the index variable in the material chemical formula is updated to the index value to obtain an updated chemical formula;

[0039] The updated chemical formula and the maximum zt value corresponding to the low-temperature material data are integrated into a low-temperature interpolation data;

[0040] The low-temperature interpolation data is summarized to obtain A low-temperature interpolation data, wherein A = m x .

[0041] Optionally, the chemical formula quantization operation is performed on the A low-temperature interpolation data, the B medium-temperature interpolation data, and the C high-temperature interpolation data to obtain an extended matrix group, including:

[0042] For each of the A low-temperature interpolation data, the following operations are performed:

[0043] The low-temperature element matrix is constructed using the updated chemical formula in the low-temperature interpolation data, wherein the low-temperature element matrix is as follows:

[0044]

[0045] ​​wherein, is a low-temperature element matrix, , and are atomic numbers of elements ranked first, second and third in the update chemical formula respectively, , and are subscripts of elements ranked first, second and third in the update chemical formula respectively, is the number of elements corresponding to the update chemical formula;

[0046] A low-temperature condition matrix is constructed using the maximum zt value corresponding to the low-temperature interpolation data, and the low-temperature condition matrix is as follows:

[0047]

[0048] wherein, is a low-temperature condition matrix, is the maximum zt value corresponding to the low-temperature interpolation data;

[0049] The low-temperature element matrix and the low-temperature condition matrix are combined to obtain a low-temperature matrix group, and the low-temperature matrix groups are summarized to obtain A low-temperature matrix groups;

[0050] The following operations are performed on each of the B medium-temperature interpolation data:

[0051] A medium-temperature element matrix is obtained based on the update chemical formula in the medium-temperature interpolation data;

[0052] A medium-temperature condition matrix is constructed using the maximum zt value corresponding to the medium-temperature interpolation data, and the medium-temperature condition matrix is as follows:

[0053]

[0054] wherein, is a medium-temperature condition matrix, is the maximum zt value corresponding to the medium-temperature interpolation data;

[0055] The medium-temperature element matrix and the medium-temperature condition matrix are combined to obtain a medium-temperature matrix group, and the medium-temperature matrix groups are summarized to obtain B medium-temperature matrix groups;

[0056] The following operations are performed on each of the C high-temperature interpolation data:

[0057] A high-temperature element matrix is obtained based on the update chemical formula in the high-temperature interpolation data;

[0058] A high-temperature condition matrix is constructed using the maximum zt value corresponding to the high-temperature interpolation data, and the high-temperature condition matrix is as follows: ​​

[0059]

[0060] wherein, is a high-temperature condition matrix, is a maximum zt value corresponding to the high-temperature interpolation data;

[0061] combining the high-temperature element matrix and the high-temperature condition matrix to obtain a high-temperature matrix group, and aggregating the high-temperature matrix group to obtain C high-temperature matrix groups;

[0062] aggregating the A low-temperature matrix groups, the B medium-temperature matrix groups, and the C high-temperature matrix groups to obtain a plurality of extended matrix groups, wherein the extended matrix group is a low-temperature matrix group, a medium-temperature matrix group, or a high-temperature matrix group, and =A+B+C, the extended matrix group includes an extended element matrix and an extended condition matrix, the extended element matrix is a low-temperature element matrix, a medium-temperature element matrix, or a high-temperature element matrix, and the extended condition matrix is a low-temperature condition matrix, a medium-temperature condition matrix, or a high-temperature condition matrix.

[0063] Optionally, the training of the first discriminative model by using the extended matrix group to obtain a conditional discriminative model includes:

[0064] obtaining a training ratio and a verification ratio, wherein the sum of the training ratio and the verification ratio is 1;

[0065] calculating a training quantity based on the training ratio, and the calculation formula is as follows:

[0066]

[0067] wherein, is the training quantity, is the training ratio;

[0068] obtaining a verification quantity based on the verification ratio, and obtaining the extended matrix group based on the training quantity, the verification quantity, and a plurality of training matrix groups and a plurality of verification matrix groups, wherein, is the verification quantity;

[0069] randomly selecting training matrix groups from the training matrix groups, wherein, < ;

[0070] training the first discriminative model by using the training matrix groups to obtain an updated discriminative model;

[0071] to For each of the verification matrix groups, the following operation is performed:

[0072] Input the expanded element matrix from the validation matrix group into the updated discrimination model to obtain the conditional discrimination matrix. Calculate the conditional difference degree based on the conditional discrimination matrix and the expanded conditional matrix from the validation matrix group, as shown in the following formula:

[0073]

[0074] in, For conditional variability, and These are the elements in the first row and first column, and the elements in the first row and second column, of the expanded condition matrix in the verification matrix group, respectively. and These are the elements in the first row and first column of the conditional discrimination matrix, and the elements in the first row and second column, respectively. It is the hyperbolic tangent function;

[0075] Summarize the conditional differences to obtain multiple conditional differences, and calculate the mean difference based on the multiple conditional differences. The mean difference is the average of the multiple conditional differences.

[0076] Compare the mean difference with a preset difference threshold. If the mean difference is greater than the difference threshold, update the discrimination model as the first discrimination model and return to the previous state. Randomly selected from the training matrix groups The process involves training a set of matrices until the mean difference is greater than or equal to the difference threshold.

[0077] If the mean difference is less than or equal to the difference threshold, then the updated discrimination model is used as the conditional discrimination model.

[0078] Optionally, the method utilizing a pre-built variational autoencoder and Each expanded matrix group is used to train the pre-constructed second discrimination model to obtain a true / false discrimination model, including:

[0079] right Each of the extended matrix groups performs the following operation:

[0080] The encoder in the variational autoencoder is used to map the extended element matrix in the extended matrix group to the latent distribution;

[0081] Random sampling is performed on the latent distribution based on a preset number of samples to obtain y latent variables, where y is the number of samples;

[0082] The decoder in the variational autoencoder is used to decode the y latent variables to obtain y reconstruction matrices;

[0083] By summing the y reconstruction matrices, we obtain Y reconstruction matrices, where Y = ×y;

[0084] from Extract from each expanded matrix group Y expanded element matrices, and randomly extract Y elements from the Y reconstructed matrices. One reconstruction matrix;

[0085] right For each of the extended element matrices, the following operation is performed:

[0086] The expanded element matrix is ​​combined with the pre-constructed truth matrix to form a truth matrix group, where the truth matrix is: ;

[0087] By summing up the set of truth-determination matrices, we obtain A set of truth-determination matrices;

[0088] right For each of the reconstruction matrices, the following operation is performed:

[0089] The reconstructed matrix is ​​combined with the pre-constructed false positive matrix to form a false positive matrix group, where the false positive matrix is: ;

[0090] By summing up the false positive matrix set, we obtain A set of false detection matrices;

[0091] Summary Each set of truth-detection matrices and From a set of false-test matrices, we obtain N sets of true-false matrices, where each true-false matrix set is either a true-test matrix set or a false-test matrix set, and N = 2 × ;

[0092] Based on the training ratio, validation ratio, and N sets of true / false matrices. A set of real and fake training matrices and N sets of true / false verification matrices, where N = + ;

[0093] based on A set of true and false training matrices and a second discrimination model are used to obtain an updated discrimination model;

[0094] right For each of the true / false verification matrix groups, the following operation is performed:

[0095] inputting the extended element matrix or the reconstruction matrix in the true and false verification matrix group into the updated discrimination model to obtain a true and false identification matrix, obtaining a discrimination value based on the true and false identification matrix, obtaining an original value based on the true and false verification matrix in the true and false verification matrix group, and calculating a discrimination difference value according to the discrimination value and the original value, wherein the discrimination value is a value of an element in the first row and the first column of the true and false identification matrix, the original value is a value of an element in the first row and the first column of the true and false verification matrix, and the discrimination difference value is an absolute difference value of the discrimination value and the original value;

[0096]

[0097]

[0098]

[0099]

[0100]

[0101]

[0102]

[0103]

[0104] ​​​​​​​​​​​​If the target temperature is greater than or equal to the second temperature, a temperature index is recorded as 3, and a plurality of reference high-temperature data are obtained based on the C high-temperature interpolation data;

[0105] The following operations are performed on each of the plurality of reference low-temperature data, reference medium-temperature data, or reference high-temperature data:

[0106] The reference element matrix is obtained based on the reference low-temperature data, reference medium-temperature data, or reference high-temperature data;

[0107] The reference element matrix is obtained based on the reference low-temperature data, reference medium-temperature data, or reference high-temperature data;

[0108] The following operations are performed on each of the plurality of reference low-temperature data, reference medium-temperature data, or reference high-temperature data:

[0109] The y reference latent variables are obtained based on the sampling number and the reference latent distribution, and the y reference matrices are obtained by decoding the y reference latent variables using the target decoder of the target variational autoencoder in the target generative adversarial network;

[0110] The reference matrices are summarized to obtain a plurality of reference matrices, and the following operations are performed on each of the plurality of reference matrices:

[0111] The reference matrix is input into the conditional discrimination model to obtain a reference conditional matrix, and the reconstructed temperature value and the reconstructed zt value of the reference conditional matrix are identified, wherein the reconstructed zt value is the value of the element in the first row and the first column of the reference conditional matrix, and the reconstructed temperature value is the value of the element in the first row and the second column of the reference conditional matrix;

[0112] The temperature difference value is calculated according to the reconstructed temperature value and the temperature index corresponding to the reference matrix, and the reconstructed zt difference value is calculated according to the reconstructed zt value and the target zt value, wherein the temperature difference value is the absolute difference between the reconstructed temperature value and the reference matrix, and the reconstructed zt difference value is the absolute difference between the reconstructed zt value and the target zt value;

[0113] It is determined whether the temperature difference value is less than a preset temperature comparison threshold and whether the reconstructed zt difference value is less than a preset zt comparison threshold;

[0114] If the temperature difference value is less than the temperature comparison threshold and the reconstructed zt difference value is less than the zt comparison threshold, the reference matrix is recorded as a final matrix, and the target material chemical formula is restored according to the final matrix;

[0115] The target material chemical formula is summarized to obtain a plurality of target material chemical formulas.

[0116] To achieve the above objectives, the present invention also provides a reverse design system for high thermoelectric figure of merit materials based on a deep generative model, comprising:

[0117] The material data partitioning module is used to acquire a thermoelectric material dataset. This dataset includes multiple thermoelectric material data sets, each containing the material's chemical formula, maximum zt value, optimal temperature, and subscript value range. A chemical formula sieving operation is performed on these multiple thermoelectric material data sets to obtain multiple target material data sets. A temperature partitioning operation is then performed on these target material data sets to obtain... Data on low-temperature materials, Data on medium-temperature materials and Data on high-temperature materials;

[0118] The material data expansion module is used for... Data on low-temperature materials, Data on medium-temperature materials and All high-temperature material data underwent subscript interpolation, resulting in A sets of low-temperature interpolated data, B sets of medium-temperature interpolated data, and C sets of high-temperature interpolated data. Chemical formula quantization was then performed on the A sets of low-temperature interpolated data, B sets of medium-temperature interpolated data, and C sets of high-temperature interpolated data to obtain... One extended matrix group;

[0119] Generative model training module, used to utilize The pre-constructed first discrimination model is trained with an expanded matrix set to obtain a conditional discrimination model, which utilizes a pre-constructed variational autoencoder and An expanded matrix group trains a pre-constructed second discrimination model to obtain a true / false discrimination model. The variational autoencoder includes an encoder and a decoder. A generative adversarial network is constructed using the variational autoencoder and the true / false discrimination model. The generative adversarial network is then subjected to adversarial training to obtain a target adversarial network. The target adversarial network includes a target variational autoencoder, which in turn includes a target encoder and a target decoder.

[0120] The thermoelectric material generation module is used to generate multiple target material chemical formulas using preset target zt values, preset target temperatures, target adversarial networks, and conditional discrimination models, thereby completing the reverse design of high thermoelectric figure of merit materials.

[0121] To address the above problems, the present invention also provides an electronic device, the electronic device comprising:

[0122] Memory, storing at least one instruction; and

[0123] The processor executes the instructions stored in the memory to implement the reverse design method for high thermoelectric figure of merit materials based on deep generative models described above.

[0124] To solve the above problems, the application further provides a computer readable storage medium, wherein the computer readable storage medium stores at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the high-thermoelectric figure-of-merit material reverse design method based on a deep generation model.

[0125] To solve the problems in the background art, the application obtains a thermoelectric material dataset, wherein the thermoelectric material dataset includes a plurality of thermoelectric material data, and the thermoelectric material data includes a material chemical formula, a maximum zt value, an optimal temperature, and a subscript value range. It can be seen that, in the embodiment of the application, the chemical formula screening operation is performed on the plurality of thermoelectric material data to screen out the thermoelectric material data corresponding to the material chemical formula with too many or too few elements, so as to prevent the material chemical formula with too many or too few elements from affecting the subsequent model training, and then the temperature division operation and the subscript interpolation operation are performed on the plurality of target material data, so that the data amount of the low-temperature interpolation data, the medium-temperature interpolation data, and the high-temperature interpolation data is basically consistent, thereby preventing the condition discrimination model trained subsequently from overfitting the medium-temperature region with more data amount and ignoring the low-temperature region and the high-temperature region, and the chemical formula quantization operation is performed on the A low-temperature interpolation data, the B medium-temperature interpolation data, and the C high-temperature interpolation data to obtain A expansion matrix groups. It can be seen that, in the embodiment of the application, the A low-temperature interpolation data, the B medium-temperature interpolation data, and the C high-temperature interpolation data are converted into the form of matrixes, which facilitates the subsequent training of the first discrimination model and the second discrimination model, the first discrimination model is trained by using the A expansion matrix groups to obtain a condition discrimination model, and it can be seen that, in the embodiment of the application, the condition discrimination model is trained, which facilitates the subsequent auxiliary target generative adversarial network to discriminate the generated target material chemical formula and improves the accuracy of the deep generation model in predicting high-thermoelectric figure-of-merit materials. It can be seen that, in the embodiment of the application, the second discrimination model is trained by using the A expansion matrix groups and the pre-constructed variational autoencoder to obtain a true-false discrimination model, and the variational autoencoder includes an encoder and a decoder, and it can be seen that, in the embodiment of the application, the true-false discrimination model is trained, which facilitates the subsequent auxiliary target generative adversarial network to discriminate the generated target material chemical formula and improves the accuracy of the deep generation model in predicting high-thermoelectric figure-of-merit materials. ​​​​​​The expansion matrix set trains the true and false discrimination model, improves the accuracy of the true and false discrimination model in discriminating the material chemical formula generated by the generation model, and further improves the accuracy of the deep generation model in predicting high thermoelectric value materials. A generative adversarial network is constructed by using the variational autoencoder and the true and false discrimination model, and the generative adversarial network is adversarially trained to obtain a target adversarial network. The target adversarial network includes a target variational autoencoder, and the target variational autoencoder includes a target encoder and a target decoder. It can be seen that the embodiments of the present application train the target adversarial network by combining the variational autoencoder and the generative adversarial network, thereby improving the accuracy of the deep generation model in predicting high thermoelectric value materials. A plurality of target material chemical formulas are generated by using a preset target zt value, a preset target temperature, the target adversarial network and a conditional discrimination model, and the reverse design of high thermoelectric value materials is completed. It can be seen that the embodiments of the present application generate target material chemical formulas meeting the preset target zt value and the preset target temperature by using the conditional discrimination model to assist the target adversarial network, thereby improving the accuracy of the deep generation model in predicting high thermoelectric value materials. Therefore, the present application can improve the accuracy of the deep generation model in predicting high thermoelectric value materials. BRIEF DESCRIPTION OF DRAWINGS

[0126] Figure 1 A flowchart of a high thermoelectric value material reverse design method based on a deep generation model provided by an embodiment of the present application is shown in the figure.

[0127] Figure 2 A functional module diagram of a high thermoelectric value material reverse design system based on a deep generation model provided by an embodiment of the present application is shown in the figure.

[0128] Figure 3 A structural diagram of an electronic device for implementing the high thermoelectric value material reverse design method based on a deep generation model provided by an embodiment of the present application is shown in the figure.

[0129] Explanation of reference signs:

[0130] 1. Electronic device; 10. Processor; 11. Memory; 12. Bus.

[0131] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0132] It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0133] The embodiment of the present application provides a high thermoelectric figure of merit material reverse design method based on a deep generative model. The execution subject of the high thermoelectric figure of merit material reverse design method based on a deep generative model includes but is not limited to at least one of electronic devices such as a server, a terminal and the like which can be configured to execute the method provided by the embodiment of the present application. In other words, the high thermoelectric figure of merit material reverse design method based on a deep generative model can be executed by software or hardware installed in 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 and the like.

[0134] Referring to Figure 1 Fig. 1 is a flowchart of a high thermoelectric figure of merit material reverse design method based on a deep generative model provided by an embodiment of the present application. In the embodiment, the high thermoelectric figure of merit material reverse design method based on a deep generative model includes:

[0135] S1, obtaining a thermoelectric material dataset, wherein the thermoelectric material dataset includes a plurality of thermoelectric material data, and the thermoelectric material data includes a material chemical formula, a maximum zt value, an optimal temperature and an index value range.

[0136] In the embodiment of the present application, the thermoelectric material dataset is obtained from the MRL thermoelectric material database. The MRL thermoelectric material database (MRL Thermoelectric Database) is a public database developed by the Materials Research Laboratory (MRL) of the University of Illinois at Urbana-Champaign (UIUC). The database focuses on the collection and arrangement of thermoelectric material data, and provides researchers with a systematic resource for evaluating and developing efficient thermoelectric materials.

[0137] In the embodiment of the present application, the deep generative model refers to a deep learning model combining a variational autoencoder and a generative adversarial network.

[0138] It should be explained that the material chemical formula refers to the chemical formula of the thermoelectric material. Since the thermoelectric figure of merit of the thermoelectric material changes with temperature, the maximum zt value refers to the maximum thermoelectric figure of merit that can be reached by the thermoelectric material under the condition of temperature selection, and the optimal temperature is the temperature at which the thermoelectric material reaches the maximum zt value.

[0139] It should be understood that most of the thermoelectric material data in the MRL thermoelectric material database includes the material chemical formula of the thermoelectric material, the maximum zt value, the optimal temperature, and the subscript value range. In order to ensure the uniformity of the data in subsequent model training, the embodiment of the present application only selects the thermoelectric material data whose element symbol subscript in the material chemical formula contains an unknown number and whose value range of the unknown number is explicitly labeled in the MRL thermoelectric material database when obtaining the thermoelectric material data from the MRL thermoelectric material database. For example, the material chemical formula of a certain thermoelectric material is , and the MRL thermoelectric material database gives the value range of x in [0.1, 0.2], then the thermoelectric material data corresponding to the thermoelectric material is the thermoelectric material data required by the embodiment of the present application, and the value range of x is the subscript value range.

[0140] S2, performing chemical formula screening operation on the plurality of thermoelectric material data to obtain a plurality of target material data, performing temperature division operation on the plurality of target material data to obtain low-temperature material data, medium-temperature material data, and high-temperature material data.

[0141] In detail, the performing chemical formula screening operation on the plurality of thermoelectric material data to obtain a plurality of target material data comprises:

[0142] performing the following operation on each of the plurality of thermoelectric material data:

[0143] identifying the number of elements in the material chemical formula of the thermoelectric material data, and comparing the number of elements with a preset first number threshold and a preset second number threshold, wherein the first number threshold is less than the second number threshold;

[0144] if the number of elements is greater than or equal to the first number threshold and less than or equal to the second number threshold, the thermoelectric material data is recorded as a target material data;

[0145] summarizing the target material data to obtain a plurality of target material data.

[0146] It should be explained that the number of elements refers to the number of types of element symbols in the material chemical formula. For example, the material chemical formula is , because the material chemical formula contains four element symbols: , , , and , the number of elements is 4. Alternatively, the first number threshold is set to 3 and the second number threshold is set to 5.

[0147] It should be understood that the embodiments of the present application prevent the material formula with too many or too few elements from affecting the subsequent model training by screening out the thermoelectric material data corresponding to the material formula with too many or too few elements.

[0148] In detail, the temperature division operation is performed on the plurality of target material data to obtain

[0149] The following operation is performed on each of the plurality of target material data:

[0150] The optimal temperature corresponding to the target material data is compared with a preset first temperature and a preset second temperature, wherein the first temperature is less than the second temperature;

[0151] If the optimal temperature is less than or equal to the first temperature, the target material data corresponding to the optimal temperature is recorded as low-temperature material data;

[0152] If the optimal temperature is greater than the first temperature and less than the second temperature, the target material data corresponding to the optimal temperature is recorded as medium-temperature material data;

[0153] If the optimal temperature is greater than or equal to the second temperature, the target material data corresponding to the optimal temperature is recorded as high-temperature material data;

[0154] The low-temperature material data, the medium-temperature material data and the high-temperature material data are respectively summarized to obtain

[0155] Optionally, the first temperature is set to 500 degrees Celsius, and the second temperature is set to 1000 degrees Celsius.

[0156] It should be understood that the optimal temperature of most thermoelectric material data is in the medium-temperature zone (500-1000 degrees Celsius), and the corresponding thermoelectric material data in the low-temperature zone (below 500 degrees Celsius) and the high-temperature zone (above 1000 degrees Celsius) is less. Therefore, the embodiments of the present application divide the target material data into low-temperature material data, medium-temperature material data and high-temperature material data, and then perform the index interpolation operation on the low-temperature material data, the medium-temperature material data and the high-temperature material data, so that the data amount of the low-temperature interpolation data, the medium-temperature interpolation data and the high-temperature interpolation data is basically consistent, thereby preventing the subsequent trained conditional discrimination model from overfitting the medium-temperature zone with more data and ignoring the low-temperature zone and the high-temperature zone.

[0157] It should be explained that ​​​​​​a total number of low-temperature material data, a total number of medium-temperature material data, a total number of high-temperature material data.

[0158] S3, performing an index interpolation operation on a low-temperature material data, a medium-temperature material data, and a high-temperature material data, to obtain A low-temperature interpolation data, B medium-temperature interpolation data, and C high-temperature interpolation data.

[0159] In detail, the step S3 of performing an index interpolation operation on a low-temperature material data, a medium-temperature material data, and a high-temperature material data, to obtain A low-temperature interpolation data, B medium-temperature interpolation data, and C high-temperature interpolation data, comprises:

[0160] According to a preset target data quantity, a low-temperature interpolation ratio, a medium-temperature interpolation ratio, and a high-temperature interpolation ratio are calculated, and the calculation formula is as follows:

[0161]

[0162] wherein, , and are the low-temperature interpolation ratio, the medium-temperature interpolation ratio, and the high-temperature interpolation ratio respectively, is the target data quantity, represents a rounding-up operation;

[0163] The low-temperature interpolation ratio is used to perform an index interpolation operation on a low-temperature material data, to obtain A low-temperature interpolation data, B medium-temperature interpolation data is obtained based on the medium-temperature interpolation ratio and a medium-temperature material data, and C high-temperature interpolation data is obtained based on the high-temperature interpolation ratio and a high-temperature material data.

[0164] It should be explained that the target data quantity is related to the quantity of data required for training, and the target data quantity is greater than the total number of low-temperature material data, medium-temperature material data, and high-temperature material data. Optionally, the target data quantity is set to 2000.

[0165] It should be understood that the method of obtaining B medium-temperature interpolation data based on the medium-temperature interpolation ratio and a medium-temperature material data, and the method of obtaining C high-temperature interpolation data based on the high-temperature interpolation ratio and a high-temperature material data are the same as the method of obtaining A low-temperature interpolation data based on the low-temperature interpolation ratio and The method for obtaining the A low-temperature interpolation data is the same as that for obtaining the low-temperature material data, which will not be repeated here.

[0166] In detail, the low-temperature interpolation proportion is used to perform an index interpolation operation on the A low-temperature material data to obtain A low-temperature interpolation data, comprising:

[0167] perform the following operation on each of the A low-temperature material data:

[0168] According to the low-temperature interpolation proportion and the index value range corresponding to the low-temperature material data, the low-temperature sampling interval is calculated, and the calculation formula is as follows:

[0169]

[0170] wherein, is the low-temperature sampling interval, and are the maximum value and the minimum value in the index value range, respectively, is the low-temperature interpolation proportion;

[0171] Based on the low-temperature sampling interval, a uniform sampling operation is performed on the index value range corresponding to the low-temperature material data to obtain m index values, and the following operation is performed on each of the m index values:

[0172] The index variable in the material chemical formula corresponding to the low-temperature material data is identified, and the index variable in the material chemical formula is updated to the index value to obtain an updated chemical formula;

[0173] The updated chemical formula and the maximum zt value corresponding to the low-temperature material data are integrated into the low-temperature interpolation data;

[0174] The A low-temperature interpolation data are summarized to obtain A low-temperature interpolation data, wherein A = m x .

[0175] For example, if the low-temperature interpolation proportion is 6 and the index value range is [0.1, 0.2], the calculated low-temperature sampling interval is 0.02, and the uniform sampling operation is performed on the index value range to obtain 6 index values: 0.1, 0.12, 0.14, 0.16, 0.18, and 0.2. If the material chemical formula is , the index variable is x in , and after updating x to the index value 0.1, the updated chemical formula is .

[0176] S4, performing a chemical formula quantization operation on the A low-temperature interpolation data, the B medium-temperature interpolation data, and the C high-temperature interpolation data to obtain an extended matrix group.​

[0177] Specifically, the chemical formula quantization operation is performed on A low-temperature interpolation data, B medium-temperature interpolation data, and C high-temperature interpolation data to obtain... An expanded matrix group, including:

[0178] Perform the following operation on each of the A low-temperature interpolation data points:

[0179] A low-temperature element matrix is ​​constructed using the updated chemical formulas from the low-temperature interpolation data, as shown below:

[0180]

[0181] in, A low-temperature element matrix, , and These are the first, second, and third ranked chemical formulas updated respectively. The atomic number of the element. , and These are the first, second, and third ranked chemical formulas updated respectively. The index of the element. To update the number of elements corresponding to the chemical formula;

[0182] The low-temperature condition matrix is ​​constructed using the maximum zt value corresponding to the low-temperature interpolation data. The low-temperature condition matrix is ​​shown below:

[0183]

[0184] in, This is the low-temperature condition matrix. This represents the maximum zt value corresponding to the low-temperature interpolation data;

[0185] By combining the low-temperature element matrix and the low-temperature condition matrix, we obtain a low-temperature matrix group. By summing the low-temperature matrix groups, we obtain A low-temperature matrix groups.

[0186] For each of the B intermediate temperature interpolation data points, perform the following operation:

[0187] Obtain the mid-temperature element matrix based on the updated chemical formulas in the mid-temperature interpolation data;

[0188] The mid-temperature condition matrix is ​​constructed using the maximum zt value corresponding to the mid-temperature interpolation data. The mid-temperature condition matrix is ​​shown below:

[0189]

[0190] in, This is the medium temperature condition matrix. is the maximum zt value corresponding to the low temperature interpolation data;

[0191] combining the low temperature element matrix and the low temperature condition matrix to obtain a low temperature matrix group, and collecting the low temperature matrix groups to obtain A low temperature matrix groups;

[0192] performing the following operations on each of the C high temperature interpolation data:

[0193] obtaining a high temperature element matrix based on the updated chemical formula in the high temperature interpolation data;

[0194] constructing a high temperature condition matrix using the maximum zt value corresponding to the high temperature interpolation data, and the high temperature condition matrix is as follows:

[0195]

[0196] wherein, is the high temperature condition matrix, is the maximum zt value corresponding to the high temperature interpolation data;

[0197] combining the high temperature element matrix and the high temperature condition matrix to obtain a high temperature matrix group, and collecting the high temperature matrix groups to obtain C high temperature matrix groups;

[0198] collecting the A low temperature matrix groups, the B medium temperature matrix groups and the C high temperature matrix groups to obtain expansion matrix groups, wherein the expansion matrix group is a low temperature matrix group, a medium temperature matrix group or a high temperature matrix group, and =A+B+C, the expansion matrix group includes an expansion element matrix and an expansion condition matrix, the expansion element matrix is a low temperature element matrix, a medium temperature element matrix or a high temperature element matrix, and the expansion condition matrix is a low temperature condition matrix, a medium temperature condition matrix or a high temperature condition matrix.

[0199] It should be understood that when an element in the updated chemical formula does not have a subscript, the subscript of the element is 1 by default.

[0200] For example, if the updated chemical formula in the low temperature interpolation data is Since the atomic numbers of , , and are 30, 28, 50 and 51 respectively, the low temperature element matrix is:

[0201]

[0202] If the maximum zt value corresponding to the low temperature interpolation data is 1.3, the low temperature condition matrix is:

[0203]

[0204] The medium-temperature condition matrix and the high-temperature condition matrix are the same.

[0205] It can be understood that the method for obtaining the medium-temperature element matrix based on the updated chemical formula in the medium-temperature interpolation data and the method for obtaining the high-temperature element matrix based on the updated chemical formula in the high-temperature interpolation data are the same as the method for constructing the low-temperature element matrix by using the updated chemical formula in the low-temperature interpolation data, which will not be described here.

[0206] S5, training the pre-constructed first discrimination model by using the expansion matrix group, to obtain a conditional discrimination model.

[0207] In detail, the method for training the pre-constructed first discrimination model by using the expansion matrix group, to obtain a conditional discrimination model, comprises:

[0208] obtaining a training ratio and a verification ratio, wherein the sum of the training ratio and the verification ratio is 1;

[0209] calculating a training number based on the training ratio, and the calculation formula is as follows:

[0210]

[0211] wherein, the training number is the training ratio;

[0212] obtaining a verification number based on the verification ratio, and obtaining the expansion matrix group based on the training number, the verification number and training matrix groups and verification matrix groups, wherein, the verification number is

[0213] randomly selecting training matrix groups from the training matrix groups, wherein, < ;

[0214] training the first discrimination model by using the training matrix groups, to obtain an updated discrimination model;

[0215] performing the following operations on each of the verification matrix groups:

[0216] inputting the expansion element matrix in the verification matrix group into the updated discrimination model, to obtain a conditional discrimination matrix, calculating a conditional difference degree based on the conditional discrimination matrix and the expansion condition matrix in the verification matrix group, and the calculation formula is as follows:

[0217]

[0218] wherein, is a condition difference degree, and are respectively an element of a first row and a first column and an element of a first row and a second column in a first extended condition matrix in a verification matrix group, and are respectively an element of a first row and a first column and an element of a first row and a second column in a condition identification matrix, is a hyperbolic tangent function;

[0219] a plurality of condition difference degrees are obtained by summarizing the condition difference degrees, and a difference average is calculated according to the plurality of condition difference degrees, wherein the difference average is an average of the plurality of condition difference degrees;

[0220] the difference average is compared with a preset difference threshold value, if the difference average is greater than the difference threshold value, the updated identification model is taken as a first identification model, and the step of randomly selecting k training matrix groups from n training matrix groups is returned until the difference average is greater than or equal to the difference threshold value; the difference average is less than or equal to the difference threshold value, the updated identification model is taken as a condition identification model.

[0221]

[0222] It should be explained that the training proportion and the verification proportion are set by the trainer of the first identification model according to experience, and optionally, the training proportion is 70%, and the verification proportion is 30%.

[0223] It should be understood that the method of obtaining the verification quantity based on the verification proportion is the same as the method of calculating the training quantity based on the training proportion, which will not be repeated here.

[0224] For example, the training quantity is 700, the verification quantity is 300, and there are 1000 extended matrix groups, so 700 extended matrix groups are randomly extracted from the 1000 extended matrix groups as 700 training matrix groups, and the remaining 300 extended matrix groups after extraction of the 1000 extended matrix groups are taken as 300 verification matrix groups.

[0225] It should be understood that the first identification model is a fully connected neural network, and the training of the first identification model by using the k training matrix groups means that the k training matrix groups are sequentially input into the first identification model. ​​​The expanded element matrix of each training matrix group is input into the first discrimination model. The first discrimination model outputs an output matrix. If the output matrix is ​​not the expanded condition matrix in the training matrix group, the output matrix is ​​corrected using the expanded condition matrix. This allows the first discrimination model to gradually learn the mapping relationship between the expanded element matrix and the expanded condition matrix. The technique of training the first discrimination model with a set of training matrices is existing technology and will not be elaborated here. The updated discrimination model is then... The first discrimination model after training a set of training matrices.

[0226] Understandably, after the expanded element matrix is ​​input into the updated discrimination model, the updated discrimination model will automatically output a conditional discrimination matrix.

[0227] For example, the extended condition matrix is The conditional discrimination matrix is ,but and They are 1.3 and 3 respectively. and They are respectively and .

[0228] Understandably, the conditional dissimilarity reflects the degree of difference between the conditional discrimination matrix and the extended conditional matrix. A larger conditional dissimilarity indicates a greater difference between the conditional discrimination matrix and the extended conditional matrix, suggesting a worse training effect for the updated discrimination model. Therefore, it is necessary to return to the previous state. Randomly selected from the training matrix groups The process involves training a set of matrices and then retraining the updated discrimination model. Optionally, a difference threshold of 0.1 can be set.

[0229] S6, Utilizing pre-built variational autoencoders and A set of extended matrices is used to train a pre-constructed second discrimination model to obtain a true / false discrimination model. The variational autoencoder includes an encoder and a decoder.

[0230] In detail, the method utilizing a pre-built variational autoencoder and Each expanded matrix group is used to train the pre-constructed second discrimination model to obtain a true / false discrimination model, including:

[0231] right Each of the extended matrix groups performs the following operation:

[0232] The encoder in the variational autoencoder is used to map the extended element matrix in the extended matrix group to the latent distribution;

[0233] Randomly sampling in the potential distribution based on a preset sampling number, to obtain y latent variables, wherein y is the sampling number;

[0234] Decoding the y latent variables by using a decoder in the variational autoencoder, to obtain y reconstruction matrices;

[0235] Summarizing the y reconstruction matrices, to obtain Y reconstruction matrices, wherein Y= ×y;

[0236] Extracting x augmented element matrices from the x augmented matrix groups, and randomly extracting y reconstruction matrices from the Y reconstruction matrices;

[0237] Performing the following operations on each of the x augmented element matrices: Combining the augmented element matrix with a pre-constructed true matrix to obtain a true matrix group, wherein the true matrix is:

[0238] Summarizing the true matrix groups, to obtain x true matrix groups;

[0239] Performing the following operations on each of the y reconstruction matrices:

[0240] Combining the reconstruction matrix with a pre-constructed false matrix to obtain a false matrix group, wherein the false matrix is:

[0241] Summarizing the false matrix groups, to obtain y false matrix groups; Summarizing the x true matrix groups and the y false matrix groups, to obtain N true-false matrix groups, wherein the true-false matrix group is the true matrix group or the false matrix group, and N=2×

[0242] ; Based on the training ratio, the verification ratio and the N true-false matrix groups, obtaining N true-false training matrix groups and N true-false verification matrix groups, wherein N=

[0243] + ; Based on the N true-false training matrix groups and the second discrimination model, obtaining an updated discrimination model;

[0244] Performing the following operations on each of the N true-false training matrix groups:

[0245] Based on the N true-false training matrix groups and the second discrimination model, obtaining an updated discrimination model;

[0246] ​​​​​​​​​​​​​Each of the true and false verification matrix groups performs the following operations:

[0247] The augmented element matrix or the reconstruction matrix in the true and false verification matrix group is input into the updated discrimination model to obtain a true and false identification matrix, a discrimination value is obtained based on the true and false identification matrix, an original value is obtained based on the true or false judgment matrix in the true and false verification matrix group, and a discrimination difference value is calculated according to the discrimination value and the original value, wherein the discrimination value is the value of the first row and the first column element in the true and false identification matrix, the original value is the value of the first row and the first column element in the true or false judgment matrix, and the discrimination difference value is the absolute difference value of the discrimination value and the original value.

[0248] The discrimination difference values are summarized to obtain discrimination difference values, and discrimination average values are calculated according to the discrimination difference values, wherein the discrimination average value is the average value of the discrimination difference values.

[0249] The discrimination average value is compared with a preset discrimination threshold value, if the discrimination average value is greater than the discrimination threshold value, the updated discrimination model is taken as a second identification model, and the step of obtaining the updated discrimination model based on the true and false training matrix groups and the second identification model is returned until the discrimination average value is greater than or equal to the discrimination threshold value.

[0250] If the discrimination average value is less than or equal to the discrimination threshold value, the updated discrimination model is taken as a true and false identification model.

[0251] It should be explained that the variational autoencoder is a generative model, and the variational autoencoder includes an encoder and a decoder, both of which are fully connected neural networks, and the main function of the encoder is to map the augmented element matrix in the augmented matrix group to a latent distribution, and the main function of the decoder is to map the latent variable in the latent distribution back to the form of the matrix. The latent distribution is a normal distribution, and the latent variable is a value randomly sampled in the latent distribution. Random sampling in the latent distribution means that one or more values are randomly selected from the latent distribution, and the probability of random selection follows the probability density function of the latent distribution, for example, the randomly selected value is likely to be located near the mean value of the normal distribution corresponding to the latent distribution.

[0252] It should be understood that the decoding of the y latent variables by the decoder in the variational autoencoder refers to sequentially inputting the y latent variables into the decoder in the variational autoencoder, and the decoder in the variational autoencoder automatically maps the y latent variables into the form of a matrix, and the reconstructed matrix is the matrix generated by the decoder according to the latent variables. The techniques of mapping the augmented element matrix in the augmented matrix group to the latent distribution by the encoder in the variational autoencoder and decoding the y latent variables by the decoder in the variational autoencoder to obtain the y reconstructed matrices are prior art, which will not be described here.

[0253] It can be understood that the method of obtaining a true and false training matrix group and a true and false verification matrix group based on the training ratio, the verification ratio and the N augmented matrix groups is the same as the method of obtaining a training matrix group and a verification matrix group based on the training ratio, the verification ratio and the N augmented matrix groups, which will not be described here. The method of obtaining a true and false training matrix group and a second discriminant model to obtain an updated discriminant model based on the first discriminant model and the N augmented matrix groups is the same as the method of obtaining an updated discriminant model based on the first discriminant model and the N augmented matrix groups, which will not be described here. It can be understood that the method of obtaining an updated discriminant model based on the first discriminant model and the N augmented matrix groups is the same as the method of obtaining an updated discriminant model based on the first discriminant model and the N augmented matrix groups, which will not be described here.

[0254] For example, after the augmented element matrix in the true and false verification matrix group is input into the updated discriminant model, the updated discriminant model automatically outputs a true and false discriminant matrix, and if the true and false discriminant matrix is , and the true matrix in the true and false verification matrix group is , the discriminant value is 0.9, the original value is 1, and the discriminant difference value is 0.1. Alternatively, the discriminant threshold is set to 0.1.

[0255] S7, constructing a generative adversarial network by using the variational autoencoder and the true and false discriminant model, and performing adversarial training on the generative adversarial network to obtain a target generative adversarial network, wherein the target generative adversarial network comprises a target variational autoencoder, and the target variational autoencoder comprises a target encoder and a target decoder.

[0256] It should be explained that the generative adversarial network is a deep learning model composed of a generative model and a discriminant model, and in the embodiments of the present application, the variational autoencoder is used as the generative model of the generative adversarial network, and the true and false discriminant model is used as the discriminant model of the generative adversarial network.

[0257] ​Understandably, the adversarial training process for the generative adversarial network (GAN) is as follows: Multiple augmented element matrices are input into the generative model of the GAN. The encoder in the generative model generates multiple latent distributions based on these matrices. The decoder in the generative model randomly samples multiple latent variables from these latent distributions and generates multiple reconstruction matrices based on these latent variables. Then, multiple augmented element matrices are randomly sampled again from these matrices. Both the reconstruction matrices and the augmented element matrices are input into the discriminative model. The discriminative model determines which data in the matrices input to it are fake data generated by the generative model and which are real data from the multiple augmented element matrices, and outputs the discrimination result. If the discrimination result is not equal to the true result, the discriminative model will determine the true result based on the real data. The real results automatically correct the discriminative model, gradually enabling it to distinguish between fake data generated by the generative model and real data from multiple augmented element matrices. Simultaneously, the generative model also corrects itself based on the discriminative model's output, making the fake data it generates increasingly closer to the real data from the multiple augmented element matrices. Ultimately, through the adversarial interaction between the generative and discriminative models, both their generative and discriminative abilities become stronger. Finally, when both models converge (i.e., after correcting themselves using the real and discriminative results, their discriminative and generative abilities remain almost unchanged), the adversarial training of the generative adversarial network is complete. The target adversarial network is the generative adversarial network after adversarial training. The technique for adversarial training of the generative adversarial network is existing technology and will not be elaborated further here.

[0258] It should be explained that the target variational autoencoder is the variational autoencoder in the target adversarial network after adversarial training, while the target encoder and target decoder are the encoder and decoder in the target adversarial network after adversarial training, respectively.

[0259] S8. Using preset target zt values, preset target temperatures, target adversarial networks, and conditional discrimination models, generate multiple target material chemical formulas to complete the reverse design of high thermoelectric figure of merit materials.

[0260] In detail, the generation of multiple target material chemical formulas using preset target zt values, preset target temperatures, target adversarial networks, and conditional discrimination models includes:

[0261] The maximum reference zt value and the minimum reference zt value are determined based on the target zt value and the preset reference zt value. The maximum reference zt value is the sum of the target zt value and the reference zt value, and the minimum reference zt value is the absolute difference between the target zt value and the reference zt value.

[0262] Confirm a reference zt range based on a maximum reference zt value and a minimum reference zt value, wherein the maximum value of the reference zt range is the maximum reference zt value, and the minimum value of the reference zt range is the minimum reference zt value;

[0263] Compare the target temperature with the first temperature and the second temperature, if the target temperature is less than or equal to the first temperature, mark the temperature value as 1, and select a plurality of reference low-temperature data from the A low-temperature interpolation data, wherein the maximum zt value corresponding to the reference low-temperature data is located in the reference zt value range;

[0264] If the target temperature is greater than the first temperature and less than the second temperature, mark the temperature value as 2, and obtain a plurality of reference medium-temperature data based on the B medium-temperature interpolation data;

[0265] If the target temperature is greater than or equal to the second temperature, mark the temperature value as 3, and obtain a plurality of reference high-temperature data based on the C high-temperature interpolation data;

[0266] For each of the plurality of reference low-temperature data, reference medium-temperature data or reference high-temperature data, the following operations are performed:

[0267] Obtain a reference element matrix based on the reference low-temperature data, reference medium-temperature data or reference high-temperature data;

[0268] Summarize the reference element matrix to obtain a plurality of reference element matrices, input the plurality of reference element matrices into a target encoder of a target variational autoencoder in the target generative adversarial network, and obtain a plurality of reference latent distributions;

[0269] For each of the plurality of reference latent distributions, the following operations are performed:

[0270] Obtain y reference latent variables based on the sampling number and the reference latent distribution, and decode the y reference latent variables by using a target decoder of the target variational autoencoder in the target generative adversarial network to obtain y reference matrices;

[0271] Summarize the reference matrices to obtain a plurality of reference matrices, and perform the following operations on each of the plurality of reference matrices:

[0272] Input the reference matrix into the conditional discrimination model to obtain a reference conditional matrix, and identify a reconstructed temperature value and a reconstructed zt value of the reference conditional matrix, wherein the reconstructed zt value is the value of the element in the first row and the first column of the reference conditional matrix, and the reconstructed temperature value is the value of the element in the first row and the second column of the reference conditional matrix;

[0273] The temperature difference value is an absolute difference between the reconstructed temperature value and the temperature value corresponding to the reference matrix, and the reconstructed zt difference value is an absolute difference between the reconstructed zt value and the target zt value;

[0274] It is determined whether the temperature difference value is less than a preset temperature comparison threshold value and whether the reconstructed zt difference value is less than a preset zt comparison threshold value;

[0275] If the temperature difference value is less than the temperature comparison threshold value and the reconstructed zt difference value is less than the zt comparison threshold value, the reference matrix is recorded as a final matrix, and the target material chemical formula is restored according to the final matrix;

[0276] The target material chemical formulas are summarized to obtain a plurality of target material chemical formulas.

[0277] For example, Xiao Zhang is a researcher in a thermoelectric material research laboratory. Xiao Zhang hopes to predict a thermoelectric material with an optimal temperature of 600°C and a maximum zt value of 1.3. Therefore, 600°C is taken as the target temperature, and 1.3 is taken as the target zt value. Optionally, the reference zt value is set to 0.1.

[0278] It should be understood that, since each of the A low-temperature interpolation data corresponds to a maximum zt value, the method of screening a plurality of reference low-temperature data from the A low-temperature interpolation data means that a plurality of low-temperature interpolation data with a maximum zt value within the reference zt value range are screened from the A low-temperature interpolation data, and the plurality of low-temperature interpolation data are taken as the plurality of reference low-temperature data.

[0279] It can be understood that the method of obtaining a plurality of reference medium-temperature data based on the B medium-temperature interpolation data and the method of obtaining a plurality of reference high-temperature data based on the C high-temperature interpolation data are the same as the method of obtaining a plurality of reference low-temperature data based on the A low-temperature interpolation data, which will not be described herein. The method of obtaining a reference element matrix based on the reference low-temperature data, the reference medium-temperature data or the reference high-temperature data is the same as the method of constructing a low-temperature element matrix based on the updated chemical formula in the low-temperature interpolation data, which will not be described herein.

[0280] It should be understood that the inputting the plurality of reference element matrices into the target encoder of the target variational autoencoder in the target generative adversarial network to obtain the plurality of reference latent distributions refers to: sequentially inputting each of the plurality of reference element matrices into the target encoder of the target variational autoencoder in the target generative adversarial network, mapping the reference element matrix into a reference latent distribution by using the target encoder, and collecting the reference latent distributions to obtain the plurality of reference latent distributions. The method of mapping the reference element matrix into the reference latent distribution by using the target encoder is the same as the method of mapping the augmented element matrix in the augmented matrix group into the latent distribution by using the encoder in the variational autoencoder, which will not be described here.

[0281] It can be understood that the method of obtaining y reference latent variables based on the number of samples and the reference latent distribution is the same as the method of randomly sampling y latent variables based on the preset number of samples in the latent distribution, the method of decoding y reference latent variables by using the target decoder of the target variational autoencoder in the target generative adversarial network to obtain y reference matrices is the same as the method of decoding y latent variables by using the decoder in the variational autoencoder to obtain y reconstructed matrices, and the method of inputting the reference matrix into the conditional discrimination model to obtain a reference conditional matrix is the same as the method of inputting the augmented element matrix in the verification matrix group into the updated discrimination model to obtain a conditional discrimination matrix, which will not be described here.

[0282] Optionally, the temperature comparison threshold is set to 50°C, and the zt comparison threshold is set to 0.1.

[0283] For example, if the final matrix is Since the atomic number 12 corresponds to the element symbol Mg, the atomic number corresponds to the element symbol Sb, the atomic number corresponds to the element symbol Bi, and the atomic number corresponds to the element symbol Te, and 3.1 is the subscript of the element symbol Mg, is the subscript of the element symbol Sb, is the subscript of the element symbol Bi, is the subscript of the element symbol Te, then the chemical formula of the target material restored from the final matrix is , that is, is the thermoelectric material predicted by the target generative adversarial network and the conditional discrimination model required by the researcher.

[0284] To address the problems described in the background art, this invention obtains a thermoelectric material dataset, comprising multiple thermoelectric material data sets, each including: the material's chemical formula, maximum zt value, optimal temperature, and subscript value range. A chemical formula sieving operation is performed on the multiple thermoelectric material data sets to obtain multiple target material data sets. A temperature segmentation operation is then performed on the multiple target material data sets to obtain... Data on low-temperature materials, Data on medium-temperature materials and Data on high-temperature materials, for Data on low-temperature materials, Data on medium-temperature materials and Subscript interpolation was performed on all high-temperature material data to obtain A sets of low-temperature interpolated data, B sets of medium-temperature interpolated data, and C sets of high-temperature interpolated data. This embodiment of the invention performs chemical formula sieving on multiple thermoelectric material data, filtering out thermoelectric material data corresponding to chemical formulas with too many or too few elements. This prevents the chemical formulas with excessive or insufficient elements from affecting the training of subsequent models. Furthermore, temperature segmentation and subscript interpolation operations are performed on multiple target material data to ensure that the data volume of low-temperature, medium-temperature, and high-temperature interpolated data is basically consistent. This prevents the subsequently trained conditional discrimination model from overfitting the medium-temperature region with its larger data volume, while neglecting the low-temperature and high-temperature regions. Chemical formula quantization is then performed on A sets of low-temperature interpolated data, B sets of medium-temperature interpolated data, and C sets of high-temperature interpolated data to obtain... The extended matrix group, as seen in this embodiment of the invention, transforms A sets of low-temperature interpolation data, B sets of medium-temperature interpolation data, and C sets of high-temperature interpolation data into matrix form, facilitating subsequent training of the first and second discrimination models. The extended matrix group trains the pre-constructed first discrimination model to obtain the conditional discrimination model. It can be seen that this embodiment of the invention, by training the conditional discrimination model, facilitates the subsequent discrimination of the target material's chemical formula by the auxiliary target adversarial network, improving the accuracy of the deep generative model in predicting high thermoelectric figure of merit materials. This is achieved by utilizing the pre-constructed variational autoencoder and... A set of expanded matrices is used to train a pre-constructed second discrimination model to obtain a true / false discrimination model. The variational autoencoder includes an encoder and a decoder. It is evident that this embodiment of the invention... The true and false discrimination model is trained by the expanded matrix set, the accuracy of the true and false discrimination model in discriminating the material chemical formula generated by the generation model is improved, and the accuracy of the deep generation model in predicting high thermoelectric value materials is improved. The generative adversarial network is constructed by using the variational autoencoder and the true and false discrimination model, and the target adversarial network is obtained by adversarial training of the generative adversarial network. The target adversarial network includes a target variational autoencoder, and the target variational autoencoder includes a target encoder and a target decoder. It can be seen that the embodiment of the present application trains the target adversarial network by combining the variational autoencoder and the generative adversarial network, thereby improving the accuracy of the deep generation model in predicting high thermoelectric value materials. The target material chemical formula meeting the preset target zt value and the preset target temperature is generated by using the preset target zt value, the preset target temperature, the target adversarial network and the conditional discrimination model, and the reverse design of the high thermoelectric value material is completed. It can be seen that the embodiment of the present application generates the target material chemical formula meeting the preset target zt value and the preset target temperature by using the conditional discrimination model to assist the target adversarial network, thereby improving the accuracy of the deep generation model in predicting high thermoelectric value materials. Therefore, the present application can improve the accuracy of the deep generation model in predicting high thermoelectric value materials.

[0285] As shown in Figure 2 It is a functional module diagram of the high thermoelectric value material reverse design system based on a deep generation model provided by an embodiment of the present application.

[0286] The high thermoelectric value material reverse design system based on a deep generation model 100 can be installed in an electronic device. According to the functions implemented, the high thermoelectric value material reverse design system based on a deep generation model 100 can include a material data division module 101, a material data expansion module 102, a generation model training module 103 and a thermoelectric material generation module 104. The modules of the present application can also be referred to as units, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete a fixed function, and are stored in the memory of the electronic device.

[0287] The material data division module 101 is used to obtain a thermoelectric material data set, wherein the thermoelectric material data set includes a plurality of thermoelectric material data, and the thermoelectric material data includes a material chemical formula, a maximum zt value, an optimal temperature and an index value range. The chemical formula screening operation is performed on the plurality of thermoelectric material data to obtain a plurality of target material data, the temperature division operation is performed on the plurality of target material data to obtain a low-temperature material data, a medium-temperature material data and a high-temperature material data;

[0288] The material data expansion module 102 is used to expand a low-temperature material data, Data on medium-temperature materials and All high-temperature material data underwent subscript interpolation, resulting in A sets of low-temperature interpolated data, B sets of medium-temperature interpolated data, and C sets of high-temperature interpolated data. Chemical formula quantization was then performed on the A sets of low-temperature interpolated data, B sets of medium-temperature interpolated data, and C sets of high-temperature interpolated data to obtain... One extended matrix group;

[0289] The generative model training module 103 is used to utilize The pre-constructed first discrimination model is trained with an expanded matrix set to obtain a conditional discrimination model, which utilizes a pre-constructed variational autoencoder and An expanded matrix group trains a pre-constructed second discrimination model to obtain a true / false discrimination model. The variational autoencoder includes an encoder and a decoder. A generative adversarial network is constructed using the variational autoencoder and the true / false discrimination model. The generative adversarial network is then subjected to adversarial training to obtain a target adversarial network. The target adversarial network includes a target variational autoencoder, which in turn includes a target encoder and a target decoder.

[0290] The thermoelectric material generation module 104 is used to generate multiple target material chemical formulas using preset target zt values, preset target temperatures, target adversarial networks and conditional identification models, thereby completing the reverse design of high thermoelectric figure of merit materials.

[0291] In detail, the modules in the high thermoelectric figure of merit material reverse design system 100 based on a deep generative model described in this embodiment of the invention employ the same methods as described above. Figure 1 The reverse design method for high thermoelectric figure of merit materials based on deep generative models described herein uses the same technical means and can produce the same technical effects, so it will not be elaborated here.

[0292] like Figure 3 The diagram shown is a schematic representation of an electronic device that implements a reverse design method for high thermoelectric figure of merit materials based on a deep generative model, according to an embodiment of the present invention.

[0293] 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 capable of running on the processor 10, such as a reverse design method program for high thermoelectric figure of merit materials based on a deep generative model.

[0294] The memory 11 includes at least one type of readable storage medium, such as flash memory, mobile hard disk, multimedia card, card-type memory (e.g., SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 can also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 1. Further, the memory 11 includes both an internal storage unit and an external storage device of the electronic device 1. The memory 11 can be used not only to store application software and various data installed on the electronic device 1, such as the code of the high-thermoelectric figure-of-merit material reverse design method program based on the deep generation model, but also to temporarily store data that has been output or will be output.

[0295] The processor 10 can be composed of an integrated circuit in some embodiments, such as a single packaged integrated circuit or a plurality of packaged integrated circuits with the same or different functions, including one or more combinations of central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips, etc. The processor 10 is the control core of the electronic device, which connects various components of the entire electronic device through various interfaces and lines, executes programs or modules stored in the memory 11 (such as the high-thermoelectric figure-of-merit material reverse design method program based on the deep generation model, etc.), and calls data stored in the memory 11 to perform various functions and process data of the electronic device 1.

[0296] The bus 12 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 can be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0297] Figure 3 Only the electronic device with components is shown, and those skilled in the art can understand that, Figure 3The illustrated structure does not constitute a limitation on the electronic device 1, and can include fewer or more components than illustrated, or combine certain components, or different component arrangements.

[0298] For example, although not shown, the electronic device 1 can also include a power source (such as a battery) to power the various components, and preferably the power source can be logically connected to the at least one processor 10 through a power management device, so that the power management device can implement functions such as charge management, discharge management, and power consumption management. The power source can also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and any other components. The electronic device 1 can also include various sensors, Bluetooth modules, Wi-Fi modules, and the like, which are not described here.

[0299] Further, the electronic device 1 can also include a network interface, which can optionally include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), and is typically used to establish a communication connection between the electronic device 1 and other electronic devices.

[0300] Optionally, the electronic device 1 can also include a user interface, which can be a display (Display), an input unit (such as a keyboard (Keyboard)), and optionally a standard wired interface, a wireless interface. Optionally, in some embodiments, the display can be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch, etc. The display can also be appropriately referred to as a display screen or a display unit, and is used to display information processed in the electronic device 1 and to display a visualized user interface.

[0301] The high-thermoelectric figure-of-merit material reverse design method program based on a deep generative model stored in the memory 11 in the electronic device 1 is a combination of multiple instructions, which, when executed in the processor 10, can implement:

[0302] Obtaining a thermoelectric material dataset, wherein the thermoelectric material dataset includes a plurality of thermoelectric material data, and the thermoelectric material data includes a material chemical formula, a maximum zt value, an optimal temperature, and a range of index values;

[0303] Performing a chemical formula screening operation on the plurality of thermoelectric material data to obtain a plurality of target material data, performing a temperature division operation on the plurality of target material data to obtain a low-temperature material data, a medium-temperature material data, and a high-temperature material data;

[0304] right Data on low-temperature materials, Data on medium-temperature materials and All high-temperature material data were subjected to subscript interpolation to obtain A low-temperature interpolated data, B medium-temperature interpolated data, and C high-temperature interpolated data;

[0305] Chemical formula quantization was performed on A low-temperature interpolation data, B medium-temperature interpolation data, and C high-temperature interpolation data to obtain... One extended matrix group;

[0306] use Each set of expanded matrices is used to train the pre-constructed first discrimination model to obtain the conditional discrimination model;

[0307] Utilizing pre-built variational autoencoders and A set of extended matrices is used to train the pre-constructed second discrimination model to obtain a true / false discrimination model, wherein the variational autoencoder includes an encoder and a decoder;

[0308] Generative Adversarial Networks (GANs) are constructed using variational autoencoders and a real / fake discrimination model. The GANs are then subjected to adversarial training to obtain a target adversarial network. The target adversarial network includes a target variational autoencoder, which in turn includes a target encoder and a target decoder.

[0309] By using preset target zt values, preset target temperatures, target adversarial networks, and conditional discrimination models, multiple target material chemical formulas are generated, thus completing the reverse design of high thermoelectric figure of merit materials.

[0310] Specifically, the processor 10's implementation method for the above instructions can be found in [reference needed]. Figures 1 to 3 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0311] Furthermore, if the modules / units integrated in the electronic device 1 are implemented as software functional units and sold or used as independent products, they 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 may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0312] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following:

[0313] Obtain a thermoelectric material dataset, which includes multiple thermoelectric material data, and each thermoelectric material data includes: material chemical formula, maximum zt value, optimal temperature, and subscript value range;

[0314] Chemical sieving was performed on multiple thermoelectric material data to obtain multiple target material data. Temperature partitioning was then performed on these target material data to obtain... Data on low-temperature materials, Data on medium-temperature materials and Data on high-temperature materials;

[0315] right Data on low-temperature materials, Data on medium-temperature materials and All high-temperature material data were subjected to subscript interpolation to obtain A low-temperature interpolated data, B medium-temperature interpolated data, and C high-temperature interpolated data;

[0316] Chemical formula quantization was performed on A low-temperature interpolation data, B medium-temperature interpolation data, and C high-temperature interpolation data to obtain... One extended matrix group;

[0317] use Each set of expanded matrices is used to train the pre-constructed first discrimination model to obtain the conditional discrimination model;

[0318] Utilizing pre-built variational autoencoders and A set of extended matrices is used to train the pre-constructed second discrimination model to obtain a true / false discrimination model, wherein the variational autoencoder includes an encoder and a decoder;

[0319] Generative Adversarial Networks (GANs) are constructed using variational autoencoders and a real / fake discrimination model. The GANs are then subjected to adversarial training to obtain a target adversarial network. The target adversarial network includes a target variational autoencoder, which in turn includes a target encoder and a target decoder.

[0320] By using preset target zt values, preset target temperatures, target adversarial networks, and conditional discrimination models, multiple target material chemical formulas are generated, thus completing the reverse design of high thermoelectric figure of merit materials.

[0321] In the embodiments provided by this 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 merely illustrative, and actual implementations may have other classification methods.

[0322] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, i.e., may be located in one place, or may be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs.

[0323] In addition, each functional module in various embodiments of the present application can be integrated in one processing unit, or each unit can exist physically independently, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of hardware plus software functional modules.

[0324] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0325] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A reverse design method for high thermoelectric figure of merit materials based on a deep generative model, characterized in that, The method includes: Obtain a thermoelectric material dataset, which includes multiple thermoelectric material data, and each thermoelectric material data includes: material chemical formula, maximum zt value, optimal temperature, and subscript value range; Chemical sieving was performed on multiple thermoelectric material data to obtain multiple target material data. Temperature partitioning was then performed on these target material data to obtain... Data on low-temperature materials, Data on medium-temperature materials and Data on high-temperature materials; right Data on low-temperature materials, Data on medium-temperature materials and All high-temperature material data were subjected to subscript interpolation to obtain A low-temperature interpolated data, B medium-temperature interpolated data, and C high-temperature interpolated data; Chemical formula quantization was performed on A low-temperature interpolation data, B medium-temperature interpolation data, and C high-temperature interpolation data to obtain... One extended matrix group; use Each set of expanded matrices is used to train the pre-constructed first discrimination model to obtain the conditional discrimination model; Utilizing pre-built variational autoencoders and A set of extended matrices is used to train the pre-constructed second discrimination model to obtain a true / false discrimination model, wherein the variational autoencoder includes an encoder and a decoder; Generative Adversarial Networks (GANs) are constructed using variational autoencoders and a real / fake discrimination model. The GANs are then subjected to adversarial training to obtain a target adversarial network. The target adversarial network includes a target variational autoencoder, which in turn includes a target encoder and a target decoder. By using preset target zt values, preset target temperatures, target adversarial networks, and conditional discrimination models, multiple target material chemical formulas are generated, thus completing the reverse design of high thermoelectric figure of merit materials.

2. The reverse design method for high thermoelectric figure of merit materials based on a deep generative model as described in claim 1, characterized in that, The chemical sieving operation is performed on multiple thermoelectric material data to obtain multiple target material data, including: For each thermoelectric material data point in the multiple thermoelectric material data sets, perform the following operation: Identify the number of elements in the chemical formula of thermoelectric materials, and compare the number of elements with a preset first quantity threshold and a preset second quantity threshold, wherein the first quantity threshold is less than the second quantity threshold; If the number of elements is greater than or equal to the first quantity threshold and the number of elements is less than or equal to the second quantity threshold, then the thermoelectric material data is recorded as the target material data; By summarizing the target material data, multiple target material data sets are obtained.

3. The reverse design method for high thermoelectric figure of merit materials based on a deep generative model as described in claim 1, characterized in that, The temperature partitioning operation is performed on multiple target material data to obtain... Data on low-temperature materials, Data on medium-temperature materials and Data on high-temperature materials, including: For each target material data point in a set of multiple target material data points, perform the following operation: Compare the optimal temperature corresponding to the target material data with the preset first temperature and the preset second temperature, wherein the first temperature is lower than the second temperature; If the optimal temperature is less than or equal to the first temperature, then the target material data corresponding to the optimal temperature is recorded as low-temperature material data; If the optimal temperature is greater than the first temperature and less than the second temperature, then the target material data corresponding to the optimal temperature is recorded as medium-temperature material data. If the optimal temperature is greater than or equal to the second temperature, then the target material data corresponding to the optimal temperature is recorded as high-temperature material data; By summarizing the data for low-temperature materials, medium-temperature materials, and high-temperature materials respectively, we can obtain... Data on low-temperature materials, Data on medium-temperature materials and Data on high-temperature materials.

4. The reverse design method for high thermoelectric figure of merit materials based on a deep generative model as described in claim 1, characterized in that, The pair Data on low-temperature materials, Data on medium-temperature materials and All high-temperature material data were subjected to subscript interpolation, resulting in A sets of low-temperature interpolated data, B sets of medium-temperature interpolated data, and C sets of high-temperature interpolated data, including: The low-temperature interpolation ratio, medium-temperature interpolation ratio, and high-temperature interpolation ratio are calculated based on the preset target data quantity. The calculation formulas are as follows: ； in, , and These represent the low-temperature interpolation ratio, the medium-temperature interpolation ratio, and the high-temperature interpolation ratio, respectively. For the target data quantity, This represents the rounding up operation; Using low-temperature interpolation ratio Perform subscript interpolation on the low-temperature material data to obtain A low-temperature interpolated data, based on the medium-temperature interpolation ratio and... One set of medium-temperature material data was obtained, along with B sets of medium-temperature interpolation data, based on the high-temperature interpolation ratio and... C high-temperature interpolation data points were obtained from the high-temperature material data.

5. The reverse design method for high thermoelectric figure of merit materials based on a deep generative model as described in claim 4, characterized in that, The use of low-temperature interpolation ratio Perform subscript interpolation on the low-temperature material data to obtain A low-temperature interpolated data, including: right For each cryogenic material data point in the dataset, the following operation is performed: The low-temperature sampling interval is calculated based on the low-temperature interpolation ratio and the subscript value range corresponding to the low-temperature material data. The calculation formula is as follows: ； in, For low temperature sampling interval, and These are the maximum and minimum values ​​within the range of subscript values, respectively. This is the low-temperature interpolation ratio; Based on the low-temperature sampling interval, a uniform sampling operation is performed on the range of subscript values ​​corresponding to the low-temperature material data to obtain m subscript values. For each of the m subscript values, the following operation is performed: Identify the subscript variables in the chemical formula of the low-temperature material data, and update the subscript variables in the chemical formula to subscript values ​​to obtain the updated chemical formula; The maximum zt values ​​corresponding to the updated chemical formulas and low-temperature material data are integrated into low-temperature interpolation data. By summing the low-temperature interpolation data, we obtain A low-temperature interpolation data points, where A = m × .

6. The reverse design method for high thermoelectric figure of merit materials based on a deep generative model as described in claim 5, characterized in that, The chemical formula quantization operation was performed on A low-temperature interpolation data, B medium-temperature interpolation data, and C high-temperature interpolation data to obtain... An expanded matrix group, including: Perform the following operation on each of the A low-temperature interpolation data points: A low-temperature element matrix is ​​constructed using the updated chemical formulas from the low-temperature interpolation data, as shown below: ； in, A low-temperature element matrix, , and These are the first, second, and third ranked chemical formulas updated respectively. The atomic number of the element. , and These are the first, second, and third ranked chemical formulas updated respectively. The index of the element. To update the number of elements corresponding to the chemical formula; The low-temperature condition matrix is ​​constructed using the maximum zt value corresponding to the low-temperature interpolation data. The low-temperature condition matrix is ​​shown below: ； in, This is the low-temperature condition matrix. This represents the maximum zt value corresponding to the low-temperature interpolation data; By combining the low-temperature element matrix and the low-temperature condition matrix, we obtain a low-temperature matrix group. By summing the low-temperature matrix groups, we obtain A low-temperature matrix groups. For each of the B intermediate temperature interpolation data points, perform the following operation: Obtain the mid-temperature element matrix based on the updated chemical formulas in the mid-temperature interpolation data; The mid-temperature condition matrix is ​​constructed using the maximum zt value corresponding to the mid-temperature interpolation data. The mid-temperature condition matrix is ​​shown below: ； in, This is the medium temperature condition matrix. This represents the maximum zt value corresponding to the intermediate temperature interpolation data; By combining the intermediate temperature element matrix and the intermediate temperature condition matrix, we obtain a set of intermediate temperature matrices. By summing up the intermediate temperature matrix sets, we obtain B sets of intermediate temperature matrix sets. For each of the C high-temperature interpolation data points, perform the following operation: Obtain the high-temperature element matrix based on the updated chemical formulas in the high-temperature interpolation data; A high-temperature condition matrix is ​​constructed using the maximum zt value corresponding to the high-temperature interpolation data. The high-temperature condition matrix is ​​shown below: ； in, This is the high-temperature condition matrix. This represents the maximum zt value corresponding to the high-temperature interpolation data; By combining the high-temperature element matrix and the high-temperature condition matrix, we obtain a high-temperature matrix group. By summing the high-temperature matrix groups, we obtain C high-temperature matrix groups. Summarizing the low-temperature matrix group A, the medium-temperature matrix group B, and the high-temperature matrix group C, we obtain... An extended matrix group, wherein the extended matrix group is a low-temperature matrix group, a medium-temperature matrix group, or a high-temperature matrix group, and =A+B+C, the extended matrix group includes: an extended element matrix and an extended condition matrix. The extended element matrix is ​​a low temperature element matrix, a medium temperature element matrix, or a high temperature element matrix, and the extended condition matrix is ​​a low temperature condition matrix, a medium temperature condition matrix, or a high temperature condition matrix.

7. The reverse design method for high thermoelectric figure of merit materials based on a deep generative model as described in claim 6, characterized in that, The use of Each expanded matrix group is used to train the pre-constructed first discrimination model to obtain a conditional discrimination model, including: Obtain the training ratio and validation ratio, where the sum of the training ratio and validation ratio is 1; The number of training units is calculated based on the training ratio, using the following formula: ； in, For training quantity, For training ratio; The number of verifications is obtained based on the verification ratio, and is also based on the number of training samples and the number of verifications. Obtaining an expanded matrix group training matrix groups and There are 10 verification matrix groups, among which... To verify the quantity; from Randomly selected from the training matrix groups There are sets of training matrices, among which... < ; use The first discrimination model is trained using a set of training matrices to obtain the updated discrimination model; right For each of the verification matrix groups, the following operation is performed: Input the expanded element matrix from the validation matrix group into the updated discrimination model to obtain the conditional discrimination matrix. Calculate the conditional difference degree based on the conditional discrimination matrix and the expanded conditional matrix from the validation matrix group, as shown in the following formula: ； in, For conditional variability, and These are the elements in the first row and first column, and the elements in the first row and second column, of the expanded condition matrix in the verification matrix group, respectively. and These are the elements in the first row and first column of the conditional discrimination matrix, and the elements in the first row and second column, respectively. It is the hyperbolic tangent function; Summarize the conditional differences to obtain multiple conditional differences, and calculate the mean difference based on the multiple conditional differences. The mean difference is the average of the multiple conditional differences. Compare the mean difference with a preset difference threshold. If the mean difference is greater than the difference threshold, update the discrimination model as the first discrimination model and return to the previous state. Randomly selected from the training matrix groups The process involves training a set of matrices until the mean difference is greater than or equal to the difference threshold. If the mean difference is less than or equal to the difference threshold, then the updated discrimination model is used as the conditional discrimination model.

8. The reverse design method for high thermoelectric figure of merit materials based on a deep generative model as described in claim 7, characterized in that, The use of pre-built variational autoencoders and Each expanded matrix group is used to train the pre-constructed second discrimination model to obtain a true / false discrimination model, including: right Each of the extended matrix groups performs the following operation: The encoder in the variational autoencoder is used to map the extended element matrix in the extended matrix group to the latent distribution; Random sampling is performed on the latent distribution based on a preset number of samples to obtain y latent variables, where y is the number of samples; The decoder in the variational autoencoder is used to decode the y latent variables to obtain y reconstruction matrices; By summing the y reconstruction matrices, we obtain Y reconstruction matrices, where Y = ×y; from Extract from each expanded matrix group Y expanded element matrices, and randomly extract Y elements from the Y reconstructed matrices. One reconstruction matrix; right For each of the extended element matrices, the following operation is performed: The expanded element matrix is ​​combined with the pre-constructed truth matrix to form a truth matrix group, where the truth matrix is: ; By summing up the set of truth-determination matrices, we obtain A set of truth-determination matrices; right For each of the reconstruction matrices, the following operation is performed: The reconstructed matrix is ​​combined with the pre-constructed false positive matrix to form a false positive matrix group, where the false positive matrix is: ; By summing up the false positive matrix set, we obtain A set of false judgment matrices; Summary Each set of truth-detection matrices and From a set of false-test matrices, we obtain N sets of true-false matrices, where each true-false matrix set is either a true-test matrix set or a false-test matrix set, and N = 2 × ; Based on the training ratio, validation ratio, and N sets of true / false matrices. A set of real and fake training matrices and N sets of true / false verification matrices, where N = + ; based on A set of true and false training matrices and a second discrimination model are used to obtain an updated discrimination model; right For each of the true / false verification matrix groups, the following operation is performed: The expanded element matrix or reconstructed matrix in the true / false verification matrix group is input into the updated discrimination model to obtain the true / false discrimination matrix. The discrimination value is obtained based on the true / false discrimination matrix. The original value is obtained based on the true / false discrimination matrix or the false / true matrix in the true / false verification matrix group. The discrimination difference value is calculated based on the discrimination value and the original value. The discrimination value is the value of the first row and first column element in the true / false discrimination matrix. The original value is the value of the first row and first column element in the true / false discrimination matrix or the false / true matrix. The discrimination difference value is the absolute difference between the discrimination value and the original value. Summarize the discriminant values ​​to obtain Each discriminant difference value, based on The discriminant mean is calculated from the discriminant differences, where the discriminant mean is... The average of the discriminant means; The discrimination average is compared with a preset discrimination threshold. If the discrimination average is greater than the discrimination threshold, the updated discrimination model is used as the second discrimination model, and the process is returned to the previous state. The process of obtaining an updated discrimination model from a set of true and false training matrices and a second discrimination model continues until the average discrimination value is greater than or equal to the discrimination threshold. If the average discrimination value is less than or equal to the discrimination threshold, then the updated discrimination model is used as the true / false identification model.

9. The reverse design method for high thermoelectric figure of merit materials based on a deep generative model as described in claim 8, characterized in that, The process of generating multiple target material chemical formulas using preset target zt values, preset target temperatures, target adversarial networks, and conditional discrimination models includes: The maximum reference zt value and the minimum reference zt value are determined based on the target zt value and the preset reference zt value. The maximum reference zt value is the sum of the target zt value and the reference zt value, and the minimum reference zt value is the absolute difference between the target zt value and the reference zt value. The reference zt range is determined based on the maximum and minimum reference zt values, where the maximum value in the reference zt range is the maximum reference zt value, and the minimum value in the reference zt range is the minimum reference zt value. Compare the target temperature with the first temperature and the second temperature. If the target temperature is less than or equal to the first temperature, then record the temperature value as 1, and select multiple reference low temperature data from A low temperature interpolation data, wherein the maximum zt value corresponding to the reference low temperature data is within the range of reference zt values. If the target temperature is greater than the first temperature and less than the second temperature, then the temperature value is recorded as 2, and multiple reference intermediate temperature data are obtained based on B intermediate temperature interpolation data. If the target temperature is greater than or equal to the second temperature, then the temperature value is recorded as 3, and multiple reference high temperature data are obtained based on C high temperature interpolation data. For each of the multiple reference low temperature data points, multiple reference medium temperature data points, or multiple reference high temperature data points, perform the following operation: A reference element matrix is ​​obtained based on reference low temperature data, reference medium temperature data, or reference high temperature data; The reference element matrices are summarized to obtain multiple reference element matrices. These multiple reference element matrices are then input into the target encoder of the target variational autoencoder in the target adversarial network to obtain multiple reference latent distributions. Perform the following operation on each of the multiple reference latent distributions: Based on the number of samples and the reference latent distribution, y reference latent variables are obtained. The target decoder of the target variational autoencoder in the target adversarial network is used to decode the y reference latent variables to obtain y reference matrices. Summarize the reference matrices to obtain multiple reference matrices. Perform the following operation on each of the multiple reference matrices: Input the reference matrix into the conditional discrimination model to obtain the reference condition matrix. Identify the reconstructed temperature value and reconstructed zt value of the reference condition matrix. The reconstructed zt value is the value of the element in the first row and first column of the reference condition matrix, and the reconstructed temperature value is the value of the element in the first row and second column of the reference condition matrix. The temperature difference is calculated based on the reconstructed temperature value and the temperature scale corresponding to the reference matrix. The reconstructed zt difference is calculated based on the reconstructed zt value and the target zt value. The temperature difference is the absolute difference between the reconstructed temperature value and the reference matrix, and the reconstructed zt difference is the absolute difference between the reconstructed zt value and the target zt value. Determine whether the temperature difference is less than the preset temperature comparison threshold, and determine whether the reconstructed zt difference is less than the preset zt comparison threshold; If the temperature difference is less than the temperature comparison threshold and the reconstructed zt difference is less than the zt comparison threshold, then the reference matrix is ​​recorded as the final matrix, and the chemical formula of the target material is reconstructed based on the final matrix. By summarizing the chemical formulas of the target materials, multiple chemical formulas of the target materials are obtained.

10. A reverse design system for high thermoelectric figure of merit materials based on a deep generative model, characterized in that, The system includes: The material data partitioning module is used to acquire a thermoelectric material dataset. This dataset includes multiple thermoelectric material data sets, each containing the material's chemical formula, maximum zt value, optimal temperature, and subscript value range. A chemical formula sieving operation is performed on these multiple thermoelectric material data sets to obtain multiple target material data sets. A temperature partitioning operation is then performed on these target material data sets to obtain... Data on low-temperature materials, Data on medium-temperature materials and Data on high-temperature materials; The material data expansion module is used for... Data on low-temperature materials, Data on medium-temperature materials and All high-temperature material data underwent subscript interpolation, resulting in A sets of low-temperature interpolated data, B sets of medium-temperature interpolated data, and C sets of high-temperature interpolated data. Chemical formula quantization was then performed on the A sets of low-temperature interpolated data, B sets of medium-temperature interpolated data, and C sets of high-temperature interpolated data to obtain... One extended matrix group; Generative model training module, used to utilize The pre-constructed first discrimination model is trained with an expanded matrix set to obtain a conditional discrimination model, which utilizes a pre-constructed variational autoencoder and An expanded matrix group trains a pre-constructed second discrimination model to obtain a true / false discrimination model. The variational autoencoder includes an encoder and a decoder. A generative adversarial network is constructed using the variational autoencoder and the true / false discrimination model. The generative adversarial network is then subjected to adversarial training to obtain a target adversarial network. The target adversarial network includes a target variational autoencoder, which in turn includes a target encoder and a target decoder. The thermoelectric material generation module is used to generate multiple target material chemical formulas using preset target zt values, preset target temperatures, target adversarial networks, and conditional discrimination models, thereby completing the reverse design of high thermoelectric figure of merit materials.

Images