Tc > 140K superconductor reverse design method and system based on gradient guidance

Abstract

The invention relates to the technical field of superconductors, in particular to a Tc > 140K superconductor reverse design method and system based on gradient guidance, and the method comprises the steps: obtaining superconductive data, constructing a superconductive data set, obtaining a preliminary data set, carrying out the data cleaning, obtaining a cleaned data set, constructing a superconductive representation matrix, carrying out the training to obtain a training generation model and a training diffusion model, and carrying out the mapping. Performing random disturbance addition for multiple times to obtain a disturbance addition vector, calculating disturbance gradient estimation, updating the low-dimensional potential vector, obtaining an updated potential vector, obtaining a reconstruction sample, performing optimization to obtain a reconstruction optimization sample, performing critical temperature prediction, obtaining a reconstruction temperature and an updated temperature, and calculating a temperature absolute difference value; and obtaining an update temperature sequence, extracting an initial update temperature, determining a target update temperature, and determining a target temperature in the update temperature sequence. According to the method, the design and discovery of the high-critical-temperature superconducting material can be accelerated.

Description

Technical Field

[0001] The present invention relates to the technical field of superconductors, and in particular to a reverse design method and system for a Tc>140K superconductor based on gradient guidance. Background Art

[0002] With the continuous progress in the fields of materials science and physics, the research and application of superconductors have gradually become an important part of modern technology. Superconductors have the characteristic of being completely resistance-free at low temperatures and are widely used in energy transmission, quantum computing, magnetic resonance imaging and other fields. The performance of superconductors is closely related to their critical temperature.

[0003] The traditional design method of superconductors is a forward prediction method based on machine learning, that is, a prediction-screening mode, which requires calculation and evaluation of a large number of candidate materials. The calculation is large and the efficiency is low. The traditional forward prediction method is difficult to c The current generation modeling methods for superconductors have the following problems: they rely on random sampling, have difficulty generating superconductors beyond the training data range, have difficulty controlling the specific properties of the generated materials, and have a low critical temperature. Therefore, how to accelerate the design and discovery of high-critical-temperature superconducting materials is an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides a gradient-guided reverse design method and system for superconductors with a Tc>140K, the main purpose of which is to accelerate the design and discovery of high critical temperature superconducting materials.

[0005] To achieve the above-mentioned object, the present invention provides a gradient-guided reverse design method for a Tc>140K superconductor, comprising:

[0006] Acquire superconducting data, construct a superconducting data set based on the superconducting data, detect unrecorded critical data and repeated critical data according to the superconducting data set, remove the unrecorded critical data and repeated critical data from the superconducting data set to obtain a preliminary data set, and perform data cleaning on the preliminary data set to obtain a cleaned data set, wherein the cleaned data includes: superconductor components;

[0007] Constructing a superconducting representation matrix, and training a pre-constructed superconducting generation model and a pre-constructed superconducting diffusion model based on the superconducting representation matrix and the cleaned data set to obtain a training generation model and a training diffusion model, wherein the superconducting representation matrix is ​​a three-dimensional matrix, and the training generation model includes: a training encoding unit and a training decoding unit;

[0008] performing a low-dimensional latent space mapping operation on the superconductor component using the training coding unit to obtain a low-dimensional latent vector;

[0009] Perform multiple random perturbations on the low-dimensional latent vector to obtain a perturbation-added vector, calculate a perturbation gradient estimate based on the perturbation-added vector, and use the perturbation gradient estimate to update the low-dimensional latent vector to obtain an updated latent vector;

[0010] Using the trained decoding unit to perform a high-dimensional output space mapping operation on the low-dimensional latent vector to obtain a reconstructed sample, using the trained diffusion model to optimize the reconstructed sample to obtain a reconstructed optimized sample, and using the pre-trained critical temperature prediction model to predict the critical temperature of the reconstructed optimized sample to obtain a reconstructed temperature;

[0011] Performing a critical temperature prediction operation on the updated latent vector using a critical temperature prediction model to obtain an updated temperature, and calculating the absolute temperature difference based on the reconstructed temperature and the updated temperature;

[0012] Comparing the absolute temperature difference with a preset absolute difference threshold;

[0013] If it is confirmed that the absolute temperature difference is greater than the absolute difference threshold, the low-dimensional latent vector is updated using the updated latent vector, and the updated low-dimensional latent vector is used to return to the above step of adding multiple random perturbations to the low-dimensional latent vector until it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold;

[0014] If it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold, return to the above step of adding multiple random perturbations to the low-dimensional latent vector to obtain an updated temperature sequence, and the updated temperature sequence includes 6 updated temperatures, and the absolute temperature differences corresponding to the updated temperatures are all less than or equal to the absolute difference threshold;

[0015] Extract initial update temperatures from the update temperature sequence in sequence, and perform the following operations on the extracted initial update temperatures:

[0016] According to the initial update temperature, a target update temperature is determined in the update temperature sequence, wherein the target update temperature is adjacent to the initial update temperature in the update temperature sequence and lags behind the initial update temperature;

[0017] After confirming that the initial update temperature is greater than or equal to the target update temperature, the target update temperature is used as the initial update temperature, and the process returns to the step of confirming the target update temperature in the update temperature sequence based on the initial update temperature. If the initial update temperatures are all greater than or equal to the target update temperature, the target temperature is confirmed in the update temperature sequence, wherein the target temperature is the last update temperature in the update temperature sequence, and the target temperature is greater than 140K.

[0018] Optionally, constructing a superconducting representation matrix includes:

[0019] Obtaining a conventional element set, wherein the conventional element set consists of 86 common elements found in superconductors;

[0020] Set the matrix ordinal number, where the matrix ordinal number consists of ten numbers from 0 to 9;

[0021] Set a ten-dimensional matrix and a matrix channel set, wherein the ten-dimensional matrix is ​​a one-hot encoding representation of the stoichiometric coefficients, and the matrix channel set consists of a first channel, a second channel, and a third channel;

[0022] A superconducting representation matrix is ​​constructed based on the matrix ordinal number, the ten dimensions of the matrix, the first channel, the second channel, and the third channel, wherein the shape of the superconducting representation matrix is ​​3×86×10.

[0023] Optionally, the training of a pre-built superconducting generation model and a pre-built superconducting diffusion model based on the superconducting representation matrix and the cleaned data set to obtain a trained generation model and a trained diffusion model includes:

[0024] Using a superconducting representation matrix to represent the cleaned data in the cleaned data set to obtain superconducting representation data, inputting the superconducting representation data into a superconducting generation model, and using the superconducting generation model to generate training space data;

[0025] The training space data is transformed using a superconducting generative model to obtain reconstructed representation data;

[0026] Calculate the reconstruction loss and KL divergence based on the training space data and the reconstructed representation data respectively;

[0027] Set the batch size, learning rate, and number of training rounds. The batch size is 128, the learning rate is 1e-4, and the number of training rounds is 2000.

[0028] The superconducting generative model is adjusted based on a preset optimizer, a preset generation loss threshold, a preset divergence threshold, a batch size, a learning rate, a number of training rounds, a reconstruction loss, and a KL divergence to obtain a trained generative model, wherein the optimizer is an Adam optimizer;

[0029] The superconducting representation data is input into a training generation model to obtain a training sample, and the superconducting diffusion model is trained using the training sample to obtain a training diffusion model.

[0030] Optionally, the training of the superconducting diffusion model using the training samples to obtain the trained diffusion model includes:

[0031] Calculating diffusion loss based on the training samples;

[0032] A probability restoration function is constructed, and the superconducting diffusion model is trained according to the diffusion loss, optimizer, batch size, learning rate, number of training rounds, probability restoration function and preset diffusion loss threshold to obtain a trained diffusion model.

[0033] Optionally, constructing a probability restoration function includes:

[0034] Obtain the noise level and noise superconductor components, and construct a probability recovery function based on the noise level and noise superconductor components:

[0035] p(X0|X s )=∫p(X s-1 |X s )p(X0|X s-1 )dX s-1

[0036] Among them, p(X0|X s ) refers to the probability of reducing the noise superconductor component with noise level s to the noise superconductor component with noise level 0, X0 refers to the noise superconductor component with noise level 0, and X s Refers to the noise superconductor component with noise level s, p(X s-1 |X s ) refers to the probability of reducing a noise superconductor component with a noise level of s to a noise superconductor component with a noise level of s-1, p(X0|X s-1 ) refers to the probability of reducing a noise superconductor component with a noise level of s-1 to a noise superconductor component with a noise level of 0, X s-1 Refers to a noisy superconductor component with a noise level of s-1.

[0037] Optionally, performing multiple random perturbation additions on the low-dimensional latent vector to obtain a perturbation-added vector includes:

[0038] Get the random perturbation vector and calculate the perturbation addition vector based on the preset amplitude parameter, random perturbation vector and low-dimensional potential vector:

[0039] Z=γ×δ+H

[0040] Here, Z refers to the perturbation addition vector, γ refers to the amplitude parameter, δ refers to the random perturbation vector, and H refers to the low-dimensional latent vector.

[0041] Optionally, calculating a disturbance gradient estimate based on the disturbance addition vector includes:

[0042] Decoding the disturbance adding vector by using the training decoding unit to obtain a disturbance adding sample;

[0043] Optimizing the disturbance addition samples using the training diffusion model to obtain optimized addition samples;

[0044] Obtain the number of disturbances, and calculate the disturbance gradient estimate based on the optimized added samples, reconstructed optimized samples, the number of disturbances, the critical temperature prediction model, and the random disturbance vector:

[0045]

[0046] Among them, Grad(H) refers to the perturbation gradient estimation, Q refers to the number of perturbations, q refers to the preset ergodic parameter, f(*) refers to the critical temperature prediction model, Dec(Z) refers to the optimized added samples, Dec(H) refers to the reconstructed optimized samples, δ q Refers to the random perturbation vector when the ergodic parameter is q.

[0047] Optionally, the updating the low-dimensional potential vector by using the perturbation gradient estimation to obtain the updated potential vector includes:

[0048] Set learning parameters and update the potential vector based on the low-dimensional potential vector, perturbation gradient estimation calculation and learning parameter calculation:

[0049] Z1=H+∈Grad(H)

[0050] Here, Z1 refers to the updated latent vector and ∈ refers to the learning parameters.

[0051] Optionally, performing a critical temperature prediction operation on the updated potential vector using a critical temperature prediction model to obtain an updated temperature includes:

[0052] Performing a decoding operation on the updated latent vector using the training decoding unit to obtain an updated decoded sample;

[0053] Optimizing the updated decoding samples according to the training diffusion model to obtain updated optimized samples;

[0054] The critical temperature prediction model is used to perform a critical temperature prediction operation on the updated optimized sample to obtain an updated temperature.

[0055] To achieve the above object, the present invention further provides a gradient-guided reverse design system for a Tc>140K superconductor, comprising:

[0056] a model training module, configured to acquire superconducting data, construct a superconducting data set based on the superconducting data, detect unrecorded critical data and repeated critical data according to the superconducting data set, remove the unrecorded critical data and repeated critical data from the superconducting data set to obtain a preliminary data set, and perform data cleaning on the preliminary data set to obtain a cleaned data set, wherein the cleaned data includes: superconductor components;

[0057] Constructing a superconducting representation matrix, and training a pre-constructed superconducting generation model and a pre-constructed superconducting diffusion model based on the superconducting representation matrix and the cleaned data set to obtain a training generation model and a training diffusion model, wherein the superconducting representation matrix is ​​a three-dimensional matrix, and the training generation model includes: a training encoding unit and a training decoding unit;

[0058] a vector updating module, configured to perform a low-dimensional latent space mapping operation on the superconductor component using the training coding unit to obtain a low-dimensional latent vector;

[0059] Perform multiple random perturbations on the low-dimensional latent vector to obtain a perturbation-added vector, calculate a perturbation gradient estimate based on the perturbation-added vector, and use the perturbation gradient estimate to update the low-dimensional latent vector to obtain an updated latent vector;

[0060] a difference calculation module for performing a high-dimensional output space mapping operation on the low-dimensional latent vector using the trained decoding unit to obtain a reconstructed sample, optimizing the reconstructed sample using the trained diffusion model to obtain a reconstructed optimized sample, and predicting the critical temperature of the reconstructed optimized sample using the pre-trained critical temperature prediction model to obtain a reconstructed temperature;

[0061] Performing a critical temperature prediction operation on the updated latent vector using a critical temperature prediction model to obtain an updated temperature, and calculating the absolute temperature difference based on the reconstructed temperature and the updated temperature;

[0062] A target confirmation module is used to compare the absolute temperature difference with a preset absolute difference threshold;

[0063] If it is confirmed that the absolute temperature difference is greater than the absolute difference threshold, the low-dimensional latent vector is updated using the updated latent vector, and the updated low-dimensional latent vector is used to return to the above step of adding multiple random perturbations to the low-dimensional latent vector until it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold;

[0064] If it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold, return to the above step of adding multiple random perturbations to the low-dimensional latent vector to obtain an updated temperature sequence, and the updated temperature sequence includes 6 updated temperatures, and the absolute temperature differences corresponding to the updated temperatures are all less than or equal to the absolute difference threshold;

[0065] Extract initial update temperatures from the update temperature sequence in sequence, and perform the following operations on the extracted initial update temperatures:

[0066] According to the initial update temperature, a target update temperature is determined in the update temperature sequence, wherein the target update temperature is adjacent to the initial update temperature in the update temperature sequence and lags behind the initial update temperature;

[0067] After confirming that the initial update temperature is greater than or equal to the target update temperature, the target update temperature is used as the initial update temperature, and the process returns to the step of confirming the target update temperature in the update temperature sequence based on the initial update temperature. If the initial update temperatures are all greater than or equal to the target update temperature, the target temperature is confirmed in the update temperature sequence, wherein the target temperature is the last update temperature in the update temperature sequence, and the target temperature is greater than 140K.

[0068] In order to solve the above problem, the present invention further provides an electronic device, comprising:

[0069] a memory storing at least one instruction;

[0070] The processor executes the instructions stored in the memory to implement the above-mentioned gradient-guided Tc>140K superconductor inverse design method.

[0071] In order to solve the above problems, the present invention also provides a computer-readable storage medium, which stores at least one instruction. The at least one instruction is executed by a processor in an electronic device to implement the above-mentioned gradient-guided Tc>140K superconductor inverse design method.

[0072] The present invention solves the problems described in the background technology. First, a superconducting data set is constructed, and unrecorded critical data and repeated critical data in the superconducting data set are detected. The unrecorded critical data and repeated critical data are removed from the superconducting data set to obtain a preliminary data set. By removing the unrecorded critical data and repeated critical data, the quality and accuracy of the preliminary data set can be ensured, and deviations caused by data errors or repeated records can be avoided, ensuring that the subsequent superconducting generation model and superconducting diffusion model learn valid data features; secondly, a superconducting representation matrix is ​​constructed to represent the complex data of the superconductor through a three-dimensional matrix, which can better capture the characteristics of the superconductor in a higher dimension, providing rich resources for the subsequent training of the superconducting generation model and the superconducting diffusion model. Using a three-dimensional matrix as input, the superconductor generative model can better learn the relationships between superconductor components and more accurately reflect the material's properties. Subsequently, multiple random perturbations are added to the low-dimensional latent vector to obtain a perturbation-added vector. This allows exploration of more regions in the latent space, helping the trained generative model find the optimal solution. A perturbation gradient estimate is calculated based on the perturbation-added vector. The perturbation gradient estimate is a direction vector representing the direction in the latent space where the critical temperature of the superconductor corresponding to the low-dimensional latent vector increases most rapidly, thereby increasing the critical temperature of the generated superconductor. Finally, an updated temperature sequence is obtained, and a target temperature is identified in the updated temperature sequence, which is greater than 140K. Therefore, the present invention can accelerate the design and discovery of high-critical-temperature superconducting materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 A schematic flow chart of a gradient-guided reverse design method for a Tc>140K superconductor according to an embodiment of the present invention;

[0074] Figure 2 A functional module diagram of a gradient-guided reverse design system for a Tc>140K superconductor according to an embodiment of the present invention;

[0075] Figure 3 A schematic structural diagram of an electronic device for implementing the gradient-guided Tc>140K superconductor reverse design method provided in one embodiment of the present invention.

[0076] Description of reference numerals:

[0077] 1. Electronic device; 10. Processor; 11. Memory; 12. Bus.

[0078] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0079] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0080] The embodiment of the present application provides a Tc>140K superconductor reverse design method based on gradient guidance. The execution subject of the said Tc>140K superconductor reverse design method based on gradient guidance includes but is not limited to at least one of the electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the said Tc>140K superconductor reverse design method based on gradient guidance can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0081] Reference Figure 1 FIG. 1 is a flow chart of a gradient-guided reverse design method for a Tc>140K superconductor according to an embodiment of the present invention. In this embodiment, the gradient-guided reverse design method for a Tc>140K superconductor includes:

[0082] S1. Acquire superconducting data, construct a superconducting data set based on the superconducting data, detect unrecorded critical data and repeated critical data according to the superconducting data set, remove the unrecorded critical data and repeated critical data from the superconducting data set to obtain a preliminary data set, and perform data cleaning on the preliminary data set to obtain a cleaned data set, wherein the cleaned data includes: superconductor components.

[0083] Superconducting data refers to data related to superconductors. Superconductor-related data includes: superconductor composition, superconductor chemical formula, and superconductor critical temperature. Superconductor composition refers to the elemental composition of the superconductor. The superconductor chemical formula refers to the chemical formula of the superconductor. For example, the chemical formula of magnesium diboride superconductor is MgB2. The superconductor critical temperature refers to the corresponding superconducting critical temperature of the superconductor. For example, the superconducting critical temperature of magnesium diboride superconductor is 39K. Acquiring superconducting data refers to collecting superconducting data from the SuperCon database, which yielded approximately 5,000 superconducting data sets. A superconducting dataset refers to a collection of superconducting data. Unrecorded critical data refers to superconducting data without a clearly recorded superconducting critical temperature. For example, hydride superconductors are considered potential high-temperature superconductors. However, hydride superconductors require extremely high pressures to maintain their superconducting state, making it difficult to accurately measure their superconducting critical temperature. Repeated critical data refers to duplicate superconducting data, for example, obtaining superconducting data for two identical magnesium diboride superconductors. The preliminary data set refers to the set obtained after removing unrecorded critical data and repeated critical data from the superconducting data set. The cleaned data set refers to the set obtained after performing data cleaning operations on the preliminary data set. Data cleaning of the preliminary data set refers to first obtaining the superconducting critical temperature with multiple different records in the preliminary data set, and then using the truncated average method to correct the superconducting critical temperature with multiple different records to obtain a corrected data set, and finally using pre-built cleaning rules to clean the corrected data set to obtain a cleaned data set. The truncated average method is an existing technology and will not be described in detail here. The cleaning rules include: atomic number rule, chemical formula rule, radioactivity rule and matrix atomic number rule. The atomic number rule refers to the rule for adjusting the corrected data with uncertain atomic number to the optimal content in a stable state. For example, La 1.85 Sr 0.15 Cu 1-x Ni x The number of O4 atoms is uncertain, so it is adjusted to the optimal content in the stable state: La 1.85 Sr 0.15 CuO4. Chemical formula rules refer to the rules for deleting incorrect chemical formulas, for example, Hg1Ba2Ca2Cu3O 8+δ For incorrect chemical formulas, delete them directly. Radioactive rules refer to the rules for deleting radioactive substances, for example, UBe 13 For radioactive substances, delete them directly. The matrix atomic number rule refers to the rule for adjusting the correction data of atomic number errors, for example, Y 0.5 Y of Ba2Cu3O7 0.5 There is an error in the number of atoms, so Y 0.5 Ba2Cu3O7 is adjusted to YBa2Cu3O7.

[0084] S2. Construct a superconducting representation matrix, and train a pre-constructed superconducting generation model and a pre-constructed superconducting diffusion model based on the superconducting representation matrix and the cleaned data set to obtain a training generation model and a training diffusion model, wherein the superconducting representation matrix is ​​a three-dimensional matrix, and the training generation model includes: a training encoding unit and a training decoding unit.

[0085] The superconducting generative model refers to a VAE model, or a variational autoencoder. The superconducting generative model includes an encoder and a decoder. The superconducting diffusion model refers to a diffusion model, or a diffusion model. These are prior art and are not described in detail here. The trained generative model refers to the model obtained after training the superconducting generative model. The trained diffusion model refers to the model obtained after training the superconducting diffusion model. The trained encoding unit refers to the encoder included in the trained generative model, which is used to map the superconductor components input to the training encoding unit to a low-dimensional latent space. The trained decoding unit refers to the decoder included in the trained generative model, which is used to reconstruct the superconductor components mapped to the low-dimensional latent space. These are prior art and are not described in detail here. Latent space refers to a compressed, low-dimensional space that can encode raw data to capture the core features of the raw data. For example, a picture of a dog, the pixel data of the picture itself is very complex, including the RGB value of each pixel. This information is high-dimensional, and the latent space will convert the complex information of the picture into a more concise representation, retaining the core features of the dog. The core features of a dog refer to the unique features of the dog, such as the shape of the dog, the color of the dog, etc., but there is no RGB value for each pixel. The representation of the latent space is a low-dimensional vector that represents the core features of the dog in the picture.

[0086] In detail, the constructing of the superconducting representation matrix includes:

[0087] Obtaining a conventional element set, wherein the conventional element set consists of 86 common elements found in superconductors;

[0088] Set the matrix ordinal number, where the matrix ordinal number consists of ten numbers from 0 to 9;

[0089] Set a ten-dimensional matrix and a matrix channel set, wherein the ten-dimensional matrix is ​​a one-hot encoding representation of the stoichiometric coefficients, and the matrix channel set consists of a first channel, a second channel, and a third channel;

[0090] A superconducting representation matrix is ​​constructed based on the matrix ordinal number, the ten dimensions of the matrix, the first channel, the second channel, and the third channel, wherein the shape of the superconducting representation matrix is ​​3×86×10.

[0091] Explainably, the superconducting representation matrix refers to a three-dimensional matrix based on the stoichiometric coefficients, constructed based on the matrix ordinal number, the matrix tenth dimension, the first channel, the second channel, and the third channel. The conventional element set refers to the set of conventional elements. Conventional elements refer to the 86 elements that have appeared in existing superconductors, sorted according to the order of elements in the periodic table. The matrix ordinal number is used to represent the rows of the superconducting representation matrix. The matrix tenth dimension refers to the one-hot encoding representation of the stoichiometric coefficients. When the matrix tenth dimension is 0, it means that the element does not exist. When the matrix tenth dimension is 1, it means that the element exists. For example, YbBa2Cu3O 6.86 In the first channel, the tenth dimension of the matrix of Yb, Cu, and O is 1, and the tenth dimension of the matrix of other elements is 0. The first channel corresponds to the first dimension of the superconducting representation matrix. The first dimension is used to represent the integer part of the stoichiometric coefficient of the superconductor. The second channel corresponds to the second dimension of the superconducting representation matrix. The second dimension is used to represent the first decimal place in the decimal part of the stoichiometric coefficient of the superconductor. The third dimension is used to represent the second decimal place in the decimal part of the stoichiometric coefficient of the superconductor. For example, the stoichiometric coefficient of the superconductor is: YbBa2Cu3O 6.86 , O 6.86 The decimal part is 6.86, with 8 as the first decimal and 6 as the second decimal. The columns of each dimension of the superconducting representation matrix are composed of a set of conventional elements, each column represents a different conventional element, and the rows of the three-dimensional matrix are composed of matrix ordinals. The values ​​of the numbers in the matrix ordinals represent the values ​​of the stoichiometric coefficients. The values ​​of the numbers in the matrix ordinals in the first channel represent the values ​​of the integer part of the stoichiometric coefficients, the values ​​of the numbers in the matrix ordinals in the second channel represent the values ​​of the first decimal of the stoichiometric coefficients, and the values ​​of the numbers in the matrix ordinals in the third channel represent the values ​​of the second decimal of the stoichiometric coefficients. For example, YbBa2Cu3O 6.86 In the first channel, the matrix ordinal number of Yb is 1, the matrix ordinal number of Ba is 2, the matrix ordinal number of Cu is 3, and the matrix ordinal number of O is 6.

[0092] In detail, the pre-built superconducting generation model and the pre-built superconducting diffusion model are trained based on the superconducting representation matrix and the cleaned data set to obtain the trained generation model and the trained diffusion model, including:

[0093] Using a superconducting representation matrix to represent the cleaned data in the cleaned data set to obtain superconducting representation data, inputting the superconducting representation data into a superconducting generation model, and using the superconducting generation model to generate training space data;

[0094] The training space data is transformed using a superconducting generative model to obtain reconstructed representation data;

[0095] Calculate the reconstruction loss and KL divergence based on the training space data and the reconstructed representation data respectively;

[0096] Set the batch size, learning rate, and number of training rounds. The batch size is 128, the learning rate is 1e-4, and the number of training rounds is 2000.

[0097] The superconducting generative model is adjusted based on a preset optimizer, a preset generation loss threshold, a preset divergence threshold, a batch size, a learning rate, a number of training rounds, a reconstruction loss, and a KL divergence to obtain a trained generative model, wherein the optimizer is an Adam optimizer;

[0098] The superconducting representation data is input into a training generation model to obtain a training sample, and the superconducting diffusion model is trained using the training sample to obtain a training diffusion model.

[0099] It can be explained that superconducting representation data refers to the data obtained by representing the superconductor components of the cleaned data using a superconducting representation matrix, training space data refers to the data obtained by encoding the superconducting representation data using the encoder of the superconducting generative model, and reconstruction representation data refers to the data obtained by decoding the training space data using the decoder of the superconducting generative model. The data obtained by decoding the training space data using the decoder is a new superconductor. Reconstruction loss refers to the mean squared error between the superconductor components of the training space data and the superconductor components of the reconstructed representation data. The batch size, learning rate, and number of training rounds are all hyperparameters in the process of adjusting the superconducting generative model. They are prior art and will not be further described here. Adjusting the superconducting generative model based on a preset optimizer, a preset generation loss threshold, a preset divergence threshold, a batch size, a learning rate, a number of training rounds, a reconstruction loss, and a KL divergence refers to using a gradient descent method, an optimizer, a batch size, a learning rate, and a number of training rounds to make the reconstruction loss and KL divergence less than the generation loss threshold and the divergence threshold, respectively. This is prior art and will not be further described here. The generation loss threshold refers to a manually set value used to determine whether the reconstruction loss is stable. When the reconstruction loss is less than the generation loss threshold, the reconstruction loss is confirmed to be stable. The divergence threshold refers to a manually set value used to determine whether the KL divergence is stable. When the KL divergence is less than the divergence threshold, the KL divergence is confirmed to be stable. When both the reconstruction loss and the KL divergence are stable, the adjustment of the superconducting generation model is completed, and a training generation model is obtained. The training sample refers to the superconductor component obtained by inputting superconducting representation data into the training generation model, encoding the superconducting representation data using the training encoding unit of the training generation unit, and then decoding the encoded superconducting representation data using the training decoding unit of the training generation unit. Training the superconducting diffusion model using the training samples refers to using the training samples to cause the superconducting diffusion model to learn a reverse diffusion process. When the diffusion loss of the superconducting diffusion model is less than the diffusion loss threshold, the training of the superconducting diffusion model is completed, and the training diffusion model is obtained. This is a prior art and will not be further described here.

[0100] In detail, the superconducting diffusion model is trained using training samples to obtain a training diffusion model, including:

[0101] Calculating diffusion loss based on the training samples;

[0102] A probability restoration function is constructed, and the superconducting diffusion model is trained according to the diffusion loss, optimizer, batch size, learning rate, number of training rounds, probability restoration function and preset diffusion loss threshold to obtain a trained diffusion model.

[0103] The diffusion loss is the mean square error between the superconductor composition of the training sample and the superconductor composition of the sample generated by the superconductor diffusion model. This is a prior art technique and will not be described in detail here. The diffusion loss threshold is a manually set value used to judge the diffusion loss.

[0104] In detail, the construction of the probability restoration function includes:

[0105] Obtain the noise level and noise superconductor components, and construct a probability recovery function based on the noise level and noise superconductor components:

[0106] p(X0|X s )=∫p(X s-1 |X s )p(X0|X s-1 )dX s-1

[0107] Among them, p(X0|X s ) refers to the probability of reducing the noise superconductor component with noise level s to the noise superconductor component with noise level 0, X0 refers to the noise superconductor component with noise level 0, and X s Refers to the noise superconductor component with noise level s, p(X s-1 |X s ) refers to the probability of reducing a noise superconductor component with a noise level of s to a noise superconductor component with a noise level of s-1, p(X0|X s-1 ) refers to the probability of reducing a noise superconductor component with a noise level of s-1 to a noise superconductor component with a noise level of 0, X s-1 Refers to a noisy superconductor component with a noise level of s-1.

[0108] The probability reduction function is interpretable. The reduction probability is the probability of reducing a noisy superconductor component to a noise-free superconductor component. The noise level is manually set to categorize noisy superconductor components. Noisy superconductor components refer to the noisy superconductor components generated by the VAE model. The process of reducing noisy superconductor components to noise-free superconductor components is the reverse diffusion process. The complete superconducting diffusion model training process is as follows: the noisy superconductor components generated by the VAE model are input into the superconducting diffusion model. The superconducting diffusion model learns the reverse diffusion process based on the probability reduction function, optimizer, batch size, learning rate, and number of training rounds. The noisy superconductor components generated by the VAE model are then reduced based on the superconducting diffusion model after learning the reverse diffusion process. The diffusion loss is calculated based on the reduced superconductor components. When the diffusion loss is less than the diffusion loss threshold, the superconducting diffusion model training is completed.

[0109] S3. Utilize the training coding unit to perform a low-dimensional latent space mapping operation on the superconductor component to obtain a low-dimensional latent vector.

[0110] An interpretable, low-dimensional latent vector refers to the vector obtained after the machine learning model: the VAE model maps the input data to a low-dimensional latent space. The low-dimensional latent vector contains the key information of the input data and can help the VAE model generate or predict data. The optimized latent vector refers to the vector obtained after gradient optimization of the low-dimensional latent vector using the training diffusion model.

[0111] S4. Perform multiple random perturbations on the low-dimensional latent vector to obtain a perturbation-added vector, calculate a perturbation gradient estimate based on the perturbation-added vector, and use the perturbation gradient estimate to update the low-dimensional latent vector to obtain an updated latent vector.

[0112] Interpretable, perturbation addition vector refers to the new vector obtained after adding a random perturbation vector to the low-dimensional latent vector, random perturbation vector refers to a random vector used to simulate exploration in different directions in the latent space, perturbation gradient estimate refers to a direction vector, indicating the direction in which the critical temperature of the superconductor corresponding to the low-dimensional latent vector increases fastest in the latent space, and update latent vector refers to the vector obtained after updating the low-dimensional latent vector according to the perturbation gradient estimate.

[0113] Specifically, the step of performing multiple random perturbations on the low-dimensional latent vector to obtain a perturbation-added vector includes:

[0114] Get the random perturbation vector and calculate the perturbation addition vector based on the preset amplitude parameter, random perturbation vector and low-dimensional potential vector:

[0115] Z=γ×δ+H

[0116] Here, Z refers to the perturbation addition vector, γ refers to the amplitude parameter, δ refers to the random perturbation vector, and H refers to the low-dimensional latent vector.

[0117] It can be explained that the amplitude parameter refers to a parameter set artificially for adjusting the amplitude of the random perturbation vector. The larger the amplitude parameter is, the greater the influence of the random perturbation vector on the perturbation addition vector is.

[0118] In detail, the calculation of the disturbance gradient estimate based on the disturbance addition vector includes:

[0119] Decoding the disturbance adding vector by using the training decoding unit to obtain a disturbance adding sample;

[0120] Optimizing the disturbance addition samples using the training diffusion model to obtain optimized addition samples;

[0121] Obtain the number of disturbances, and calculate the disturbance gradient estimate based on the optimized added samples, reconstructed optimized samples, the number of disturbances, the critical temperature prediction model, and the random disturbance vector:

[0122]

[0123] Among them, Grad(H) refers to the perturbation gradient estimation, Q refers to the number of perturbations, q refers to the preset ergodic parameter, f(*) refers to the critical temperature prediction model, Dec(Z) refers to the optimized added samples, Dec(H) refers to the reconstructed optimized samples, δ q Refers to the random perturbation vector when the ergodic parameter is q.

[0124] Interpretably, the perturbation-added sample refers to the superconductor sample obtained after decoding the perturbation-added vector using the training decoding unit, the optimized-added sample refers to the superconductor sample obtained after optimizing the perturbation-added sample using the training diffusion model, the number of perturbations refers to the number of random perturbations added to the low-dimensional latent vector, and the traversal parameter refers to the parameter used to traverse the number of perturbations.

[0125] In detail, the method of updating the low-dimensional potential vector using the perturbation gradient estimation to obtain the updated potential vector includes:

[0126] Set learning parameters and update the potential vector based on the low-dimensional potential vector, perturbation gradient estimation calculation and learning parameter calculation:

[0127] Z1=H+∈Grad(H)

[0128] Here, Z1 refers to the updated latent vector and ∈ refers to the learning parameters.

[0129] Interpretably, learning parameters refer to the hyperparameters used to control the step size of the low-dimensional latent vector in each update, which determines the speed at which the low-dimensional latent vector is updated.

[0130] S5. Use the trained decoding unit to perform a high-dimensional output space mapping operation on the low-dimensional latent vector to obtain a reconstructed sample, use the trained diffusion model to optimize the reconstructed sample to obtain a reconstructed optimized sample, and use the pre-trained critical temperature prediction model to predict the critical temperature of the reconstructed optimized sample to obtain a reconstructed temperature.

[0131] Interpretable, reconstructed samples refer to superconductor samples obtained by decoding low-dimensional latent vectors using the trained decoding unit in the trained generative model. Reconstructed optimized samples refer to superconductor samples obtained by denoising the reconstructed samples using the trained diffusion model. The critical temperature prediction model refers to a manually set model for predicting the critical temperature of superconductors. Optionally, the critical temperature prediction model is a hybrid expert model that can effectively capture the differences between different types of superconductors and make specialized predictions for different systems, thereby improving overall prediction accuracy. The reconstructed temperature refers to the critical temperature of the reconstructed optimized sample obtained by using the critical temperature prediction model to predict the critical temperature of the reconstructed optimized sample.

[0132] S6. Use the critical temperature prediction model to perform a critical temperature prediction operation on the updated latent vector to obtain an updated temperature, and calculate the absolute temperature difference based on the reconstructed temperature and the updated temperature.

[0133] To be explained, the absolute temperature difference refers to the absolute value of the difference between the reconstructed temperature and the updated temperature.

[0134] Specifically, performing a critical temperature prediction operation on the updated potential vector using the critical temperature prediction model to obtain an updated temperature includes:

[0135] Performing a decoding operation on the updated latent vector using the training decoding unit to obtain an updated decoded sample;

[0136] Optimizing the updated decoding samples according to the training diffusion model to obtain updated optimized samples;

[0137] The critical temperature prediction model is used to perform a critical temperature prediction operation on the updated optimized sample to obtain an updated temperature.

[0138] It can be explained that the updated decoding sample refers to the superconductor sample obtained after decoding the updated latent vector using the trained decoding unit, the updated optimization sample refers to the superconductor sample obtained after denoising the updated decoding sample using the trained diffusion model, and the updated temperature refers to the critical temperature of the updated optimization sample obtained after performing the critical temperature prediction operation on the updated optimization sample using the critical temperature prediction model.

[0139] S7. Compare the absolute temperature difference with a preset absolute difference threshold.

[0140] It can be explained that the absolute difference threshold refers to a threshold set manually for determining the absolute temperature difference. Optionally, the absolute difference threshold is 0.5K.

[0141] S8. If it is confirmed that the absolute temperature difference is greater than the absolute difference threshold, the low-dimensional latent vector is updated using the updated latent vector, and the updated low-dimensional latent vector is used to return to the above step of adding multiple random perturbations to the low-dimensional latent vector until it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold.

[0142] S9. If it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold, return to the above step of adding multiple random perturbations to the low-dimensional latent vector to obtain an updated temperature sequence, and the updated temperature sequence includes 6 updated temperatures, and the absolute temperature differences corresponding to the updated temperatures are all less than or equal to the absolute difference threshold.

[0143] It can be explained that the updated temperature sequence refers to a sequence composed of updated temperatures. Returning to the above step of adding multiple random perturbations to the low-dimensional latent vector, the updated temperature sequence refers to obtaining six updated temperatures in succession, and the absolute differences of the temperatures corresponding to the six updated temperatures are all less than the absolute difference threshold.

[0144] S10. Extract the initial update temperatures from the update temperature sequence in sequence, and perform the following operations on the extracted initial update temperatures: confirm the target update temperature in the update temperature sequence based on the initial update temperature, wherein the target update temperature is adjacent to the initial update temperature in the update temperature sequence and lags behind the initial update temperature; after confirming that the initial update temperature is greater than or equal to the target update temperature, use the target update temperature as the initial update temperature, and return to the step of confirming the target update temperature in the update temperature sequence based on the initial update temperature; if the initial update temperatures are all greater than or equal to the target update temperature, confirm the target temperature in the update temperature sequence, wherein the target temperature is the last update temperature in the update temperature sequence, and the target temperature is greater than 140K.

[0145] It can be explained that the initial updated temperature refers to the updated temperature in the updated temperature sequence.

[0146] The present invention solves the problems described in the background technology. First, a superconducting data set is constructed, and unrecorded critical data and repeated critical data in the superconducting data set are detected. The unrecorded critical data and repeated critical data are removed from the superconducting data set to obtain a preliminary data set. By removing the unrecorded critical data and repeated critical data, the quality and accuracy of the preliminary data set can be ensured, and deviations caused by data errors or repeated records can be avoided, ensuring that the subsequent superconducting generation model and superconducting diffusion model learn valid data features; secondly, a superconducting representation matrix is ​​constructed to represent the complex data of the superconductor through a three-dimensional matrix, which can better capture the characteristics of the superconductor in a higher dimension, providing rich resources for the subsequent training of the superconducting generation model and the superconducting diffusion model. Using a three-dimensional matrix as input, the superconductor generative model can better learn the relationships between superconductor components and more accurately reflect the material's properties. Subsequently, multiple random perturbations are added to the low-dimensional latent vector to obtain a perturbation-added vector. This allows exploration of more regions in the latent space, helping the trained generative model find the optimal solution. A perturbation gradient estimate is calculated based on the perturbation-added vector. The perturbation gradient estimate is a direction vector representing the direction in the latent space where the critical temperature of the superconductor corresponding to the low-dimensional latent vector increases most rapidly, thereby increasing the critical temperature of the generated superconductor. Finally, an updated temperature sequence is obtained, and a target temperature is identified in the updated temperature sequence, which is greater than 140K. Therefore, the present invention can accelerate the design and discovery of high-critical-temperature superconducting materials.

[0147] like Figure 2 , which is a functional module diagram of a gradient-guided Tc>140K superconductor reverse design system provided by an embodiment of the present invention.

[0148] The gradient-guided reverse design system 100 for superconductors with a Tc greater than 140K described in the present invention can be installed in an electronic device. Depending on the functionality implemented, the gradient-guided reverse design system 100 for superconductors with a Tc greater than 140K can include a model training module 101, a vector update module 102, a difference calculation module 103, and a target confirmation module 104. A module, also referred to as a unit, is a series of computer program segments that can be executed by an electronic device processor and perform a fixed function, and is stored in the memory of the electronic device.

[0149] The model training module 101 is used to obtain superconducting data, construct a superconducting data set based on the superconducting data, detect unrecorded critical data and repeated critical data according to the superconducting data set, remove the unrecorded critical data and repeated critical data from the superconducting data set to obtain a preliminary data set, and perform data cleaning on the preliminary data set to obtain a cleaned data set, wherein the cleaned data includes: superconductor components;

[0150] Constructing a superconducting representation matrix, and training a pre-constructed superconducting generation model and a pre-constructed superconducting diffusion model based on the superconducting representation matrix and the cleaned data set to obtain a training generation model and a training diffusion model, wherein the superconducting representation matrix is ​​a three-dimensional matrix, and the training generation model includes: a training encoding unit and a training decoding unit;

[0151] The vector updating module 102 is configured to perform a low-dimensional latent space mapping operation on the superconductor component using the training coding unit to obtain a low-dimensional latent vector;

[0152] Perform multiple random perturbations on the low-dimensional latent vector to obtain a perturbation-added vector, calculate a perturbation gradient estimate based on the perturbation-added vector, and use the perturbation gradient estimate to update the low-dimensional latent vector to obtain an updated latent vector;

[0153] The difference calculation module 103 is used to use the trained decoding unit to perform a high-dimensional output space mapping operation on the low-dimensional potential vector to obtain a reconstructed sample, optimize the reconstructed sample using the trained diffusion model to obtain a reconstructed optimized sample, and use the pre-trained critical temperature prediction model to predict the critical temperature of the reconstructed optimized sample to obtain a reconstructed temperature;

[0154] Performing a critical temperature prediction operation on the updated latent vector using a critical temperature prediction model to obtain an updated temperature, and calculating the absolute temperature difference based on the reconstructed temperature and the updated temperature;

[0155] The target confirmation module 104 is used to compare the absolute temperature difference with a preset absolute difference threshold;

[0156] If it is confirmed that the absolute temperature difference is greater than the absolute difference threshold, the low-dimensional latent vector is updated using the updated latent vector, and the updated low-dimensional latent vector is used to return to the above step of adding multiple random perturbations to the low-dimensional latent vector until it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold;

[0157] If it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold, return to the above step of adding multiple random perturbations to the low-dimensional latent vector to obtain an updated temperature sequence, and the updated temperature sequence includes 6 updated temperatures, and the absolute temperature differences corresponding to the updated temperatures are all less than or equal to the absolute difference threshold;

[0158] Extract initial update temperatures from the update temperature sequence in sequence, and perform the following operations on the extracted initial update temperatures:

[0159] According to the initial update temperature, a target update temperature is determined in the update temperature sequence, wherein the target update temperature is adjacent to the initial update temperature in the update temperature sequence and lags behind the initial update temperature;

[0160] After confirming that the initial update temperature is greater than or equal to the target update temperature, the target update temperature is used as the initial update temperature, and the process returns to the step of confirming the target update temperature in the update temperature sequence based on the initial update temperature. If the initial update temperatures are all greater than or equal to the target update temperature, the target temperature is confirmed in the update temperature sequence, wherein the target temperature is the last update temperature in the update temperature sequence, and the target temperature is greater than 140K.

[0161] In detail, the modules in the gradient-guided Tc>140K superconductor reverse design system 100 according to the embodiment of the present invention are used in the same manner as above. Figure 1 The same technical means are used as the gradient-guided Tc>140K superconductor reverse design method described in and can produce the same technical effects, which will not be repeated here.

[0162] like Figure 3 1 is a schematic structural diagram of an electronic device for implementing a gradient-guided reverse design method for a Tc>140K superconductor provided by an embodiment of the present invention.

[0163] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a gradient-guided inverse design method program for a Tc>140K superconductor.

[0164] The memory 11 includes at least one type of readable storage medium, including a flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 11 may 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 may also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 1. Furthermore, the memory 11 includes both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can be used not only to store application software installed in the electronic device 1 and various types of data, such as the code of the gradient-guided Tc>140K superconductor reverse design method program, but can also be used to temporarily store data that has been output or is about to be output.

[0165] In some embodiments, the processor 10 may be comprised of an integrated circuit, such as a single packaged integrated circuit or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and a combination of various control chips. The processor 10 is the control core (Control Unit) of the electronic device, connecting the various components of the entire electronic device using various interfaces and circuits. It executes or runs programs or modules stored in the memory 11 (e.g., a gradient-guided Tc>140K superconductor reverse design method program) and calls data stored in the memory 11 to perform various functions of the electronic device 1 and process data.

[0166] The bus 12 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus 12 may be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to enable communication between the memory 11 and at least one processor 10.

[0167] Figure 3 Only the electronic device with components is shown, and it can be understood by those skilled in the art that Figure 3 The structure shown does not constitute a limitation on the electronic device 1 , and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0168] For example, although not shown, the electronic device 1 may further include a power source (such as a battery) for powering the various components. Preferably, the power source may be logically connected to the at least one processor 10 via a power management device, thereby implementing functions such as charging management, discharging management, and power consumption management through the power management device. The power source may further include any components such as one or more DC or AC power sources, a recharging device, a power failure detection circuit, a power converter or inverter, a power status indicator, etc. The electronic device 1 may further include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0169] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0170] Optionally, the electronic device 1 may further include a user interface, which may be a display or an input unit (such as a keyboard). Optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0171] The gradient-guided Tc>140K superconductor reverse design method program stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve the following:

[0172] Acquire superconducting data, construct a superconducting data set based on the superconducting data, detect unrecorded critical data and repeated critical data according to the superconducting data set, remove the unrecorded critical data and repeated critical data from the superconducting data set to obtain a preliminary data set, and perform data cleaning on the preliminary data set to obtain a cleaned data set, wherein the cleaned data includes: superconductor components;

[0173] Constructing a superconducting representation matrix, and training a pre-constructed superconducting generation model and a pre-constructed superconducting diffusion model based on the superconducting representation matrix and the cleaned data set to obtain a training generation model and a training diffusion model, wherein the superconducting representation matrix is ​​a three-dimensional matrix, and the training generation model includes: a training encoding unit and a training decoding unit;

[0174] performing a low-dimensional latent space mapping operation on the superconductor component using the training coding unit to obtain a low-dimensional latent vector;

[0175] Perform multiple random perturbations on the low-dimensional latent vector to obtain a perturbation-added vector, calculate a perturbation gradient estimate based on the perturbation-added vector, and use the perturbation gradient estimate to update the low-dimensional latent vector to obtain an updated latent vector;

[0176] Using the trained decoding unit to perform a high-dimensional output space mapping operation on the low-dimensional latent vector to obtain a reconstructed sample, using the trained diffusion model to optimize the reconstructed sample to obtain a reconstructed optimized sample, and using the pre-trained critical temperature prediction model to predict the critical temperature of the reconstructed optimized sample to obtain a reconstructed temperature;

[0177] Performing a critical temperature prediction operation on the updated latent vector using a critical temperature prediction model to obtain an updated temperature, and calculating the absolute temperature difference based on the reconstructed temperature and the updated temperature;

[0178] Comparing the absolute temperature difference with a preset absolute difference threshold;

[0179] If it is confirmed that the absolute temperature difference is greater than the absolute difference threshold, the low-dimensional latent vector is updated using the updated latent vector, and the updated low-dimensional latent vector is used to return to the above step of adding multiple random perturbations to the low-dimensional latent vector until it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold;

[0180] If it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold, return to the above step of adding multiple random perturbations to the low-dimensional latent vector to obtain an updated temperature sequence, and the updated temperature sequence includes 6 updated temperatures, and the absolute temperature differences corresponding to the updated temperatures are all less than or equal to the absolute difference threshold;

[0181] Extract initial update temperatures from the update temperature sequence in sequence, and perform the following operations on the extracted initial update temperatures:

[0182] According to the initial update temperature, a target update temperature is determined in the update temperature sequence, wherein the target update temperature is adjacent to the initial update temperature in the update temperature sequence and lags behind the initial update temperature;

[0183] After confirming that the initial update temperature is greater than or equal to the target update temperature, the target update temperature is used as the initial update temperature, and the process returns to the step of confirming the target update temperature in the update temperature sequence based on the initial update temperature. If the initial update temperatures are all greater than or equal to the target update temperature, the target temperature is confirmed in the update temperature sequence, wherein the target temperature is the last update temperature in the update temperature sequence, and the target temperature is greater than 140K.

[0184] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 3 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0185] Furthermore, if the modules / units integrated into 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 can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0186] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. When the computer program is executed by a processor of an electronic device, the computer program can implement:

[0187] Acquire superconducting data, construct a superconducting data set based on the superconducting data, detect unrecorded critical data and repeated critical data according to the superconducting data set, remove the unrecorded critical data and repeated critical data from the superconducting data set to obtain a preliminary data set, and perform data cleaning on the preliminary data set to obtain a cleaned data set, wherein the cleaned data includes: superconductor components;

[0188] Constructing a superconducting representation matrix, and training a pre-constructed superconducting generation model and a pre-constructed superconducting diffusion model based on the superconducting representation matrix and the cleaned data set to obtain a training generation model and a training diffusion model, wherein the superconducting representation matrix is ​​a three-dimensional matrix, and the training generation model includes: a training encoding unit and a training decoding unit;

[0189] performing a low-dimensional latent space mapping operation on the superconductor component using the training coding unit to obtain a low-dimensional latent vector;

[0190] Perform multiple random perturbations on the low-dimensional latent vector to obtain a perturbation-added vector, calculate a perturbation gradient estimate based on the perturbation-added vector, and use the perturbation gradient estimate to update the low-dimensional latent vector to obtain an updated latent vector;

[0191] Using the trained decoding unit to perform a high-dimensional output space mapping operation on the low-dimensional latent vector to obtain a reconstructed sample, using the trained diffusion model to optimize the reconstructed sample to obtain a reconstructed optimized sample, and using the pre-trained critical temperature prediction model to predict the critical temperature of the reconstructed optimized sample to obtain a reconstructed temperature;

[0192] Performing a critical temperature prediction operation on the updated latent vector using a critical temperature prediction model to obtain an updated temperature, and calculating the absolute temperature difference based on the reconstructed temperature and the updated temperature;

[0193] Comparing the absolute temperature difference with a preset absolute difference threshold;

[0194] If it is confirmed that the absolute temperature difference is greater than the absolute difference threshold, the low-dimensional latent vector is updated using the updated latent vector, and the updated low-dimensional latent vector is used to return to the above step of adding multiple random perturbations to the low-dimensional latent vector until it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold;

[0195] If it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold, return to the above step of adding multiple random perturbations to the low-dimensional latent vector to obtain an updated temperature sequence, and the updated temperature sequence includes 6 updated temperatures, and the absolute temperature differences corresponding to the updated temperatures are all less than or equal to the absolute difference threshold;

[0196] Extract initial update temperatures from the update temperature sequence in sequence, and perform the following operations on the extracted initial update temperatures:

[0197] According to the initial update temperature, a target update temperature is determined in the update temperature sequence, wherein the target update temperature is adjacent to the initial update temperature in the update temperature sequence and lags behind the initial update temperature;

[0198] After confirming that the initial update temperature is greater than or equal to the target update temperature, the target update temperature is used as the initial update temperature, and the process returns to the step of confirming the target update temperature in the update temperature sequence based on the initial update temperature. If the initial update temperatures are all greater than or equal to the target update temperature, the target temperature is confirmed in the update temperature sequence, wherein the target temperature is the last update temperature in the update temperature sequence, and the target temperature is greater than 140K.

[0199] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are only exemplary, and actual implementations may have other division methods.

[0200] 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, that is, they may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of the solution of this embodiment according to actual needs.

[0201] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0202] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0203] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A gradient-guided reverse design method for superconductors with Tc>140K, characterized in that: The method comprises: Acquire superconducting data, construct a superconducting data set based on the superconducting data, detect unrecorded critical data and repeated critical data according to the superconducting data set, remove the unrecorded critical data and repeated critical data from the superconducting data set to obtain a preliminary data set, and perform data cleaning on the preliminary data set to obtain a cleaned data set, wherein the cleaned data includes: superconductor components; Constructing a superconducting representation matrix, and training a pre-constructed superconducting generation model and a pre-constructed superconducting diffusion model based on the superconducting representation matrix and the cleaned data set to obtain a training generation model and a training diffusion model, wherein the superconducting representation matrix is ​​a three-dimensional matrix, and the training generation model includes: a training encoding unit and a training decoding unit; performing a low-dimensional latent space mapping operation on the superconductor component using the training coding unit to obtain a low-dimensional latent vector; Perform multiple random perturbations on the low-dimensional latent vector to obtain a perturbation-added vector, calculate a perturbation gradient estimate based on the perturbation-added vector, and use the perturbation gradient estimate to update the low-dimensional latent vector to obtain an updated latent vector; Using the trained decoding unit to perform a high-dimensional output space mapping operation on the low-dimensional latent vector to obtain a reconstructed sample, using the trained diffusion model to optimize the reconstructed sample to obtain a reconstructed optimized sample, and using the pre-trained critical temperature prediction model to predict the critical temperature of the reconstructed optimized sample to obtain a reconstructed temperature; Performing a critical temperature prediction operation on the updated latent vector using a critical temperature prediction model to obtain an updated temperature, and calculating the absolute temperature difference based on the reconstructed temperature and the updated temperature; Comparing the absolute temperature difference with a preset absolute difference threshold; If it is confirmed that the absolute temperature difference is greater than the absolute difference threshold, the low-dimensional latent vector is updated using the updated latent vector, and the updated low-dimensional latent vector is used to return to the above step of adding multiple random perturbations to the low-dimensional latent vector until it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold; If it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold, return to the above step of adding multiple random perturbations to the low-dimensional latent vector to obtain an updated temperature sequence, and the updated temperature sequence includes 6 updated temperatures, and the absolute temperature differences corresponding to the updated temperatures are all less than or equal to the absolute difference threshold; Extract initial update temperatures from the update temperature sequence in sequence, and perform the following operations on the extracted initial update temperatures: According to the initial update temperature, a target update temperature is determined in the update temperature sequence, wherein the target update temperature is adjacent to the initial update temperature in the update temperature sequence and lags behind the initial update temperature; After confirming that the initial update temperature is greater than or equal to the target update temperature, the target update temperature is used as the initial update temperature, and the process returns to the step of confirming the target update temperature in the update temperature sequence based on the initial update temperature. If the initial update temperatures are all greater than or equal to the target update temperature, the target temperature is confirmed in the update temperature sequence, wherein the target temperature is the last update temperature in the update temperature sequence, and the target temperature is greater than 140K.

2. The gradient-guided reverse design method for a Tc>140K superconductor according to claim 1, characterized in that: The constructing of the superconducting representation matrix includes: Obtaining a conventional element set, wherein the conventional element set consists of 86 common elements found in superconductors; Set the matrix ordinal number, where the matrix ordinal number consists of ten numbers from 0 to 9; Set a ten-dimensional matrix and a matrix channel set, wherein the ten-dimensional matrix is ​​a one-hot encoding representation of the stoichiometric coefficients, and the matrix channel set consists of a first channel, a second channel, and a third channel; A superconducting representation matrix is ​​constructed based on the matrix ordinal number, the ten dimensions of the matrix, the first channel, the second channel, and the third channel, wherein the shape of the superconducting representation matrix is ​​3×86×10.

3. The gradient-guided reverse design method for a Tc>140K superconductor according to claim 2, characterized in that: The method of training a pre-built superconducting generation model and a pre-built superconducting diffusion model based on the superconducting representation matrix and the cleaned data set to obtain a training generation model and a training diffusion model includes: Using a superconducting representation matrix to represent the cleaned data in the cleaned data set to obtain superconducting representation data, inputting the superconducting representation data into a superconducting generation model, and using the superconducting generation model to generate training space data; The training space data is transformed using a superconducting generative model to obtain reconstructed representation data; Calculate the reconstruction loss and KL divergence based on the training space data and the reconstructed representation data respectively; Set the batch size, learning rate, and number of training rounds. The batch size is 128, the learning rate is 1e-4, and the number of training rounds is 2000. The superconducting generative model is adjusted based on a preset optimizer, a preset generation loss threshold, a preset divergence threshold, a batch size, a learning rate, a number of training rounds, a reconstruction loss, and a KL divergence to obtain a trained generative model, wherein the optimizer is an Adam optimizer; The superconducting representation data is input into a training generation model to obtain a training sample, and the superconducting diffusion model is trained using the training sample to obtain a training diffusion model.

4. The gradient-guided reverse design method for a Tc>140K superconductor according to claim 3, characterized in that: The method of training the superconducting diffusion model using the training samples to obtain the training diffusion model includes: Calculating diffusion loss based on the training samples; A probability restoration function is constructed, and the superconducting diffusion model is trained according to the diffusion loss, optimizer, batch size, learning rate, number of training rounds, probability restoration function and preset diffusion loss threshold to obtain a trained diffusion model.

5. The gradient-guided reverse design method for a Tc>140K superconductor according to claim 4, characterized in that: The constructing of the probability restoration function includes: Obtain the noise level and noise superconductor components, and construct a probability recovery function based on the noise level and noise superconductor components: p(X0|X s )=∫p(X s-1 |X s )p(X0|X s-1 )dX s-1 Among them, p(X0|X s ) refers to the probability of reducing the noise superconductor component with noise level s to the noise superconductor component with noise level 0, X0 refers to the noise superconductor component with noise level 0, and X s Refers to the noise superconductor component with noise level s, p(X s-1 |X s ) refers to the probability of reducing a noise superconductor component with a noise level of s to a noise superconductor component with a noise level of s-1, p(X0|X s-1 ) refers to the probability of reducing a noise superconductor component with a noise level of s-1 to a noise superconductor component with a noise level of 0, X s-1 Refers to a noisy superconductor component with a noise level of s-1.

6. The gradient-guided reverse design method for a Tc>140K superconductor according to claim 5, characterized in that: The step of performing multiple random perturbation additions on the low-dimensional potential vector to obtain a perturbation addition vector includes: Get the random perturbation vector and calculate the perturbation addition vector based on the preset amplitude parameter, random perturbation vector and low-dimensional potential vector: Z=γ×δ+H Here, Z refers to the perturbation addition vector, γ refers to the amplitude parameter, δ refers to the random perturbation vector, and H refers to the low-dimensional latent vector.

7. The gradient-guided reverse design method for a Tc>140K superconductor according to claim 6, characterized in that: The calculation of the disturbance gradient estimation based on the disturbance addition vector includes: Decoding the disturbance adding vector by using the training decoding unit to obtain a disturbance adding sample; Optimizing the disturbance addition samples using the training diffusion model to obtain optimized addition samples; Obtain the number of disturbances, and calculate the disturbance gradient estimate based on the optimized added samples, reconstructed optimized samples, the number of disturbances, the critical temperature prediction model, and the random disturbance vector: Among them, Grad(H) refers to the perturbation gradient estimation, Q refers to the number of perturbations, q refers to the preset ergodic parameter, f(*) refers to the critical temperature prediction model, Dec(Z) refers to the optimized added samples, Dec(H) refers to the reconstructed optimized samples, δ q Refers to the random perturbation vector when the ergodic parameter is q.

8. The gradient-guided reverse design method for a Tc>140K superconductor according to claim 7, characterized in that: The method of updating the low-dimensional potential vector by using the perturbation gradient estimation to obtain the updated potential vector includes: Set learning parameters and update the potential vector based on the low-dimensional potential vector, perturbation gradient estimation calculation and learning parameter calculation: Z1=H+∈Grad(H) Here, Z1 refers to the updated latent vector and ∈ refers to the learning parameters.

9. The gradient-guided reverse design method for a Tc>140K superconductor according to claim 8, characterized in that: The method of performing a critical temperature prediction operation on the updated potential vector using the critical temperature prediction model to obtain an updated temperature includes: Performing a decoding operation on the updated latent vector using the training decoding unit to obtain an updated decoded sample; Optimizing the updated decoding samples according to the training diffusion model to obtain updated optimized samples; The critical temperature prediction model is used to perform a critical temperature prediction operation on the updated optimized sample to obtain an updated temperature.

10. A gradient-guided reverse design system for superconductors with a Tc>140K, characterized by: The system comprises: a model training module, configured to acquire superconducting data, construct a superconducting data set based on the superconducting data, detect unrecorded critical data and repeated critical data according to the superconducting data set, remove the unrecorded critical data and repeated critical data from the superconducting data set to obtain a preliminary data set, and perform data cleaning on the preliminary data set to obtain a cleaned data set, wherein the cleaned data includes: superconductor components; Constructing a superconducting representation matrix, and training a pre-constructed superconducting generation model and a pre-constructed superconducting diffusion model based on the superconducting representation matrix and the cleaned data set to obtain a training generation model and a training diffusion model, wherein the superconducting representation matrix is ​​a three-dimensional matrix, and the training generation model includes: a training encoding unit and a training decoding unit; a vector updating module, configured to perform a low-dimensional latent space mapping operation on the superconductor component using the training coding unit to obtain a low-dimensional latent vector; Perform multiple random perturbations on the low-dimensional latent vector to obtain a perturbation-added vector, calculate a perturbation gradient estimate based on the perturbation-added vector, and use the perturbation gradient estimate to update the low-dimensional latent vector to obtain an updated latent vector; a difference calculation module for performing a high-dimensional output space mapping operation on the low-dimensional latent vector using the trained decoding unit to obtain a reconstructed sample, optimizing the reconstructed sample using the trained diffusion model to obtain a reconstructed optimized sample, and predicting the critical temperature of the reconstructed optimized sample using the pre-trained critical temperature prediction model to obtain a reconstructed temperature; Performing a critical temperature prediction operation on the updated latent vector using a critical temperature prediction model to obtain an updated temperature, and calculating the absolute temperature difference based on the reconstructed temperature and the updated temperature; A target confirmation module is used to compare the absolute temperature difference with a preset absolute difference threshold; If it is confirmed that the absolute temperature difference is greater than the absolute difference threshold, the low-dimensional latent vector is updated using the updated latent vector, and the updated low-dimensional latent vector is used to return to the above step of adding multiple random perturbations to the low-dimensional latent vector until it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold; If it is confirmed that the absolute temperature difference is not greater than the absolute difference threshold, return to the above step of adding multiple random perturbations to the low-dimensional latent vector to obtain an updated temperature sequence, and the updated temperature sequence includes 6 updated temperatures, and the absolute temperature differences corresponding to the updated temperatures are all less than or equal to the absolute difference threshold; Extract initial update temperatures from the update temperature sequence in sequence, and perform the following operations on the extracted initial update temperatures: According to the initial update temperature, a target update temperature is determined in the update temperature sequence, wherein the target update temperature is adjacent to the initial update temperature in the update temperature sequence and lags behind the initial update temperature; After confirming that the initial update temperature is greater than or equal to the target update temperature, the target update temperature is used as the initial update temperature, and the process returns to the step of confirming the target update temperature in the update temperature sequence based on the initial update temperature. If the initial update temperatures are all greater than or equal to the target update temperature, the target temperature is confirmed in the update temperature sequence, wherein the target temperature is the last update temperature in the update temperature sequence, and the target temperature is greater than 140K.

Images