Superconductor performance prediction method and system, and network training method and system

Abstract

The embodiment of the invention provides a superconductor performance prediction method and system and a network training method and system. In the prediction method, a prediction system generates a plurality of core descriptors and a plurality of non-core descriptors for describing a target superconductor based on proportion information corresponding to each element in the target superconductor and attribute values of a plurality of attributes, and inputting at least part of the descriptors into the target network to obtain the performance index of the target performance of the target superconductor. The target network is a network obtained by training a plurality of sample superconductors by a training system, and comprises a plurality of sub-networks which have the same structure and are arranged according to levels, each sub-network comprises N input nodes and one output node, the output node of the i-th level sub-network is used as the input node of the (i + 1)-th level sub-network, and the output node of the (i + 1)-th level sub-network is used as the output node of the (i + 1)-th level sub-network; n core descriptors input into each sub-network in the first hierarchy are the same, N-N1 non-core descriptors are not completely the same, both N and N1 are greater than 1, and i is greater than or equal to 1.

Description

[0001] This application claims priority to Chinese Patent Application No. 2024111718620, filed on August 23, 2024, and Chinese Patent Application No. 202411194667X, filed on August 28, 2024, the contents of which are incorporated herein by reference in their entirety. TECHNICAL FIELD

[0002] The present specification relates to the field of chemical materials, and in particular, to a method and system for predicting the performance of a superconductor, and a method and system for training a network. BACKGROUND

[0003] Superconductors can exhibit zero resistance and perfect diamagnetic properties below a critical temperature (Tc), which is a material with unique physical properties, and has shown great application potential in many fields. These fields include but are not limited to power transmission, railway transportation, nuclear fusion, strong magnetic fields in medical imaging, and quantum computing, etc. The critical temperature of a superconductor is a key factor affecting the practicality of a superconductor, so accurate prediction of the critical temperature is crucial to promote the application of superconductors.

[0004] The contents of the background section only represent the inventor's own knowledge, and do not mean that the above information has entered the public domain before the filing date of the present disclosure, nor does it mean that it can be prior art of the present disclosure. SUMMARY

[0005] The present specification provides a method and system for predicting the performance of a superconductor, and a method and system for training a network, which improves the accuracy of predicting the performance of a superconductor, and can solve the problem that the network training does not meet the expected problem caused by insufficient training samples but more descriptors of superconductors.

[0006] In a first aspect, the present specification provides a method for predicting a performance of a superconductor, configured to predict a target performance of a target superconductor, the target superconductor comprising at least one element, the method comprising: generating K descriptors for describing the target superconductor based on information about proportions of the at least one element in the target superconductor and attribute values of X attributes corresponding to the at least one element respectively; determining N1 core descriptors and K-N1 non-core descriptors from the K descriptors, the core descriptors having a higher influence on the target performance than the non-core descriptors; and inputting at least part of the K descriptors into a target network for prediction to obtain a performance index corresponding to the target performance, wherein the target network is a network trained using Y sample superconductors, the target network comprising a plurality of sub-networks with the same structure and arranged in a hierarchy, each sub-network comprising N input nodes and 1 output node, the output node of a sub-network in an i-th level being used as the input node of a sub-network in an (i+1)-th level, each sub-network in the 1-st level being inputted with the N1 core descriptors and N-N1 non-core descriptors, the N-N1 non-core descriptors being part of the K-N1 non-core descriptors, and the non-core descriptors inputted to different sub-networks being not completely identical; wherein X, K, N1, Y, and N are integers greater than 1, and i is an integer greater than or equal to 1.

[0007] In some embodiments, the X attributes comprise at least two dimensions from among the following: a periodic table dimension; a thermodynamic dimension; a physical dimension; or a crystallographic dimension.

[0008] In some embodiments, the K descriptors comprise at least one of the following groups: a first group of descriptors for describing statistical characteristics of the target superconductor in the X attributes; a second group of descriptors for describing valence electron occupation states of the target superconductor; a third group of descriptors for describing ionization of the target superconductor; and a fourth group of descriptors for describing conformational entropy of the target superconductor.

[0009] In some embodiments, the first group of descriptors is generated by:

[0010] For each of the X attributes, P statistical values are obtained by performing statistics on characteristics of the target superconductor in the attribute from a plurality of statistical dimensions based on information about proportions of the at least one element in the target superconductor and attribute values of the at least one element in the attribute respectively; and a first group of descriptors is generated based on P statistical values corresponding to the X attributes of the target superconductor respectively, the number of the first group of descriptors being equal to X*P.

[0011] In some embodiments, the second group of descriptors comprises occupancy information of Q valence electrons in the target superconductor, wherein the occupancy information of the z-th valence electron is obtained by:

[0012] For each element in the at least one element, based on the attribute values of at least part of the X attributes corresponding to the element, the number of each valence electron in the element and the total number of valence electrons are determined; and for the z-th valence electron, based on the proportion information of the at least one element in the target superconductor, the number of the z-th valence electron corresponding to each of the at least one element, and the total number of valence electrons corresponding to each of the at least one element, the occupancy information of the z-th valence electron is determined.

[0013] In some embodiments, the third group of descriptors is obtained based on the proportion information of the at least one element in the target superconductor, the electronegativity corresponding to each of the at least one element, and any one of the following: the maximum value of the electronegativity corresponding to each of the at least one element; or the average value of the electronegativity corresponding to each of the at least one element.

[0014] In some embodiments, the fourth group of descriptors is obtained based on the proportion information of the at least one element in the target superconductor, the Boltzmann constant and the room temperature.

[0015] In some embodiments, the target performance is the critical temperature, and the N1 core descriptors comprise multiple descriptors selected from the following: a descriptor representing the average value of the deviation of different elements in the at least one element at the BCC Fermi level; a descriptor representing the average value of the attribute value of the first ionization energy of the at least one element; a descriptor representing the minimum value of the deviation of different elements in the at least one element at the ground state BCC lattice constant; a descriptor representing the average value of the attribute value of the total number of valence electrons filled by the at least one element; and a descriptor representing the minimum value of the deviation of different elements in the at least one element at the thermal conductivity.

[0016] In some embodiments, the target network comprises F levels, and the inputting of at least part of the K descriptors into the target network to obtain the performance index corresponding to the target performance comprises:

[0017] generating m1 groups of descriptors based on the K descriptors, each group of descriptors comprising the N1 core descriptors and N-N1 non-core descriptors; inputting the m1 groups of descriptors into m1 sub-networks of the first level of the target network respectively to obtain m1 intermediate performance indexes of the target performance; in order, according to the value of i from 2 to F, inputting the m 1-1 i-1 sub-networks of the i-1 level of the target network to obtain m i-1Each intermediate performance metric is divided into m i The m-th layer of the target network is then input after the group. i A sub-network is used to obtain m of the target performance. i An intermediate performance metric; and the intermediate performance metric of the target performance output by the sub-network of layer F as the final performance metric of the target performance; wherein, m1, F, and m i-1 All are integers greater than 1, and m i It is an integer greater than or equal to 1.

[0018] Secondly, this specification provides a training method for a target network used to predict the target performance of a superconductor. The method includes: obtaining experimental values ​​of Y sample superconductors for the target performance, each sample superconductor including at least one element; for each sample superconductor, generating K descriptors to describe the sample superconductor based on the proportion information of the at least one element in the sample superconductor and the attribute values ​​of X attributes corresponding to the at least one element, and determining N1 core descriptors and K-N1 non-core descriptors from the K descriptors, wherein the core descriptors have a greater influence on the target performance than the non-core descriptors; and inputting a portion of the descriptors from the K descriptors corresponding to each sample superconductor into the target network for pre-training. The target network is tested to obtain the predicted value corresponding to the target performance, wherein: the target network includes multiple sub-networks with the same structure and arranged in layers, each sub-network includes N input nodes and 1 output node, the output node of the sub-network in the i-th layer serves as the input node of the sub-network in the (i+1)-th layer, each sub-network in the first layer is input with the N1 core descriptors and N-N1 non-core descriptors, the N-N1 non-core descriptors are some of the descriptors in the K-N1 non-core descriptors, and the non-core descriptors input to different sub-networks are not completely the same; the parameters of the target network are updated with the training objective of minimizing the difference between the predicted value and the experimental value; wherein X, K, N1, Y, and N are all integers greater than 1, and i is an integer greater than or equal to 1.

[0019] In some embodiments, the X properties include at least two of the following dimensions: periodic table dimension; thermodynamic dimension; physical dimension; or crystallographic dimension.

[0020] In some embodiments, the K descriptors include at least one of: a first group of descriptors for describing statistical features of the sample superconductor on the X properties; a second group of descriptors for describing valence electron occupancy states of the sample superconductor; a third group of descriptors for describing ionization of the sample superconductor; and a fourth group of descriptors for describing conformational entropy of the sample superconductor.

[0021] In some embodiments, the first group of descriptors is generated by:

[0022] For each of the X properties, P statistical values are obtained by performing statistics on characteristics of the sample superconductor on the property from multiple statistical dimensions based on the proportion information of the at least one element in the sample superconductor and the property values of the at least one element on the property, respectively; and a first group of descriptors is generated based on the P statistical values of the sample superconductor on the X properties, respectively, and the number of the first group of descriptors is equal to X*P.

[0023] In some embodiments, the second group of descriptors includes occupancy information of Q valence electrons in the sample superconductor, wherein the occupancy information of the z-th valence electron is obtained by:

[0024] For each of the at least one element, the number of each valence electron and the total number of valence electrons in the element are determined based on the property values of at least part of the X properties corresponding to the element; and for the z-th valence electron, the occupancy information of the z-th valence electron is determined based on the proportion information of the at least one element in the sample superconductor, the number of the z-th valence electron corresponding to the at least one element, respectively, and the total number of valence electrons corresponding to the at least one element, respectively.

[0025] In some embodiments, the third group of descriptors is obtained based on the proportion information of the at least one element in the sample superconductor, the electronegativity corresponding to the at least one element, and any one of: a maximum value of the electronegativity corresponding to the at least one element; or an average value of the electronegativity corresponding to the at least one element.

[0026] In some embodiments, the fourth group of descriptors is obtained based on the proportion information of the at least one element in the sample superconductor, the Boltzmann constant and the room temperature.

[0027] In some embodiments, the target property is a critical temperature, and the N1 core descriptors include one or more of: a descriptor representing an average of deviations of different elements in the at least one element from a BCC Fermi level; a descriptor representing an average of attribute values of the at least one element at a first ionization energy; a descriptor representing a minimum of deviations of different elements in the at least one element from a ground state BCC lattice constant; a descriptor representing an average of attribute values of the at least one element at a total valence electron filling number; and a descriptor representing a minimum of deviations of different elements in the at least one element from a thermal conductivity.

[0028] In some embodiments, the target network includes F layers, and the inputting of the part of the K descriptors corresponding to each sample superconductor into the target network to obtain a predicted value of the target property includes:

[0029] generating m1 groups of descriptors based on the K descriptors, each group of descriptors including the N1 core descriptors and N-N1 non-core descriptors; inputting the m1 groups of descriptors into m1 sub-networks of a first layer of the target network to obtain m1 intermediate predicted values of the target property; in order of i taking values from 2 to F, dividing the m 1-1 intermediate predicted values of the target property obtained by the m i-1 sub-networks of the i-1th layer of the target network into m i groups, and inputting the m i groups into m i sub-networks of the ith layer of the target network to obtain m i-1 intermediate predicted values of the target property; and taking the intermediate predicted values of the target property output by the sub-networks of the Fth layer as a final predicted value of the target property; wherein the m1, the F, and the m i are integers greater than 1, and the m i is an integer greater than or equal to 1.

[0030] In a third aspect, the present specification also provides a prediction system configured to predict a target property of a target superconductor, including: at least one storage medium storing at least one instruction set; and at least one processor in communication connection with the at least one storage medium, wherein when the prediction system is running, the at least one processor reads the at least one instruction set and executes the prediction method of the first aspect according to the instructions of the at least one instruction set.

[0031] In a fourth aspect, the present specification also provides a training system configured to train a target network for predicting a target performance of a superconductor, comprising: at least one storage medium storing at least one instruction set; and at least one processor in communication connection with the at least one storage medium, wherein when the training system is running, the at least one processor reads the at least one instruction set and executes the training method of the second aspect according to the indication of the at least one instruction set.

[0032] Other functions of the prediction method and system of the superconductor performance, the training method and system of the network provided by the present specification will be partially listed in the following description. The creative aspects of the prediction method and system of the superconductor performance, the training method and system of the network provided by the present specification can be fully explained by practicing or using the methods, systems and combinations described in the following detailed examples. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present specification, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present specification, and other drawings can also be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0034] Figure 1 An application scenario diagram for predicting the performance of a superconductor is shown according to an embodiment of the present specification;

[0035] Figure 2 A hierarchical structure diagram of a target network is shown according to an embodiment of the present specification;

[0036] Figure 3 A structure diagram of a sub-network is shown according to an embodiment of the present specification;

[0037] Figure 4 A hardware structure diagram of a system is shown according to an embodiment of the present specification;

[0038] Figure 5 A flowchart of a training method of a target network is shown according to an embodiment of the present specification;

[0039] Figure 6 A diagram showing the change of the superconductor Tc with the element proportion is shown according to an embodiment of the present specification;

[0040] Figure 7 A diagram showing the GA iteration process is shown according to an embodiment of the present specification;

[0041] Figure 8A schematic diagram showing the relationship between the core descriptor and Tc provided according to embodiments of the present specification is shown;

[0042] Figure 9 A schematic diagram showing the input and output between the target network hierarchies provided according to embodiments of the present specification is shown;

[0043] Figure 10 A schematic diagram showing the learning results of different network models provided according to embodiments of the present specification is shown;

[0044] Figure 11 A specific schematic diagram showing the learning results of different network models provided according to embodiments of the present specification is shown;

[0045] Figure 12 A schematic diagram showing the positions of the elements included in different superconductors in the periodic table of elements provided according to embodiments of the present specification is shown;

[0046] Figure 13 A schematic diagram showing the prediction results of HEAs by different network models provided according to embodiments of the present specification is shown; and

[0047] Figure 14 A flowchart of a method for predicting the performance of a superconductor provided according to embodiments of the present specification is shown. DETAILED DESCRIPTION

[0048] The following description provides specific applications and requirements of the present specification, which is intended to enable a person skilled in the art to manufacture and use the contents of the present specification. Various modifications to the disclosed embodiments are apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of the present specification. Therefore, the present specification is not limited to the embodiments shown, but is consistent with the widest scope of the claims.

[0049] The terms used herein are only for the purpose of describing specific example embodiments, and are not limiting. For example, unless the context clearly indicates otherwise, as used herein, the singular forms "a", "an", and "the" can also include the plural forms. When used in the present specification, the terms "comprise", "include" and / or "contain" mean that the associated integers, steps, operations, elements and / or components exist, but do not exclude the presence of one or more other features, integers, steps, operations, elements, components and / or groups.

[0050] These and other features, and characteristics of the present specification, as well as the methods of operation and functions of the related elements of structure and the combination of parts and economies of manufacture, will become more apparent upon consideration of the following description and the appended claims with reference to the accompanying drawings, all of which form a part of this specification. It is to be expressly understood, however, that the drawings are for the purpose of illustration and description only and are not intended as a definition of the limits of the present specification. As such, the

[0051] The flow diagrams used in this specification illustrate operations according to some embodiments of the present specification. It will be understood that each operation can be implemented by various means, such as hardware, software, firmware, or a combination thereof. In one embodiment, for example, one or more procedures are implemented in hardware, such as by one or more ASICs. In another embodiment, for example, one or more procedures are implemented in software and / or firmware, such as by one or more programs running on a processor(s), which would typically be provided in a general or application-specific computer system. As will be realized, the present specification includes any combination of the preceding methods and / or operations.

[0052] Bardeen-Cooper-Schrieffer (BCS) theory explains the properties of many superconductors, such as low-temperature superconductors, however, the BCS theory has limited capability in predicting Tc and studying new superconductors.

[0053] Artificial Intelligence (AI) neural networks have been applied in the process of studying Tc of superconductors. In the application process, the descriptors input by the AI neural network can be divided into three categories, such as descriptors describing the properties of the chemical composition of the superconductor, descriptors describing the properties of the structure of the superconductor, and descriptors describing the properties of the elements constituting the superconductor. Among them, due to the lack of material property parameters, the predictive ability of the neural network trained by the descriptors describing the properties of the chemical composition of the superconductor is limited, and the neural network trained by the descriptors describing the properties of the structure of the superconductor cannot predict the Tc of the superconductor with unknown structure. In contrast, it is feasible to train a neural network by descriptors describing the properties of the elements constituting the superconductor.

[0054] Descriptors refer to information describing the superconductor or the performance of the superconductor, such as information about the components (such as elements) of the superconductor that have an impact on the critical temperature, information about the properties of the superconductor itself, and the like.

[0055] In some embodiments, the AI neural network can be trained by 145 element property descriptors, but due to the small number of sample superconductors, the AI neural network trained by 145 element property descriptors cannot accurately predict Tc.

[0056] Generally, the performance of an AI neural network is highly dependent on the number and quality of training samples. For sample superconductors, the more descriptors describing the sample superconductors, the more likely it is to train a good performance prediction network. For ordinary neural networks, the more descriptors, the more sample superconductors are needed, but the acquisition of sample superconductors is difficult and requires a large amount of time cost. Therefore, how to overcome the difficulty of training the prediction network to the expected difficulty caused by the lack of samples and the more descriptors of superconductors while ensuring the accuracy of the prediction result is a difficult problem in the process of superconductor research.

[0057] The following will be described in conjunction with Figure 1 The application scenario of predicting the performance of superconductors provided in the specification is introduced.

[0058] Figure 1 The application scenario of predicting the performance of superconductors provided in the specification is introduced. Figure 1 As shown in the figure, scenario 001 can include training system 100 and prediction system 200.

[0059] Scenario 001 can be divided into two stages, a training stage and a prediction stage.

[0060] Training stage

[0061] The training system 100 can train a target network capable of predicting the target performance of the superconductor based on a plurality of sample superconductors, such as critical temperature. The training system 100 can deploy the target network in the prediction system 200 after training.

[0062] Prediction stage

[0063] The prediction system 200 can call the target network to predict the performance index corresponding to the target performance of the target superconductor.

[0064] In some embodiments, the training system 100 can store data and instructions for implementing the training method of the target network, and can execute or be used to execute the data and instructions. In some embodiments, the training system 100 can include a hardware device with data information processing function and the necessary programs required to drive the hardware device to work.

[0065] In some embodiments, the prediction system 200 can store data and instructions for implementing the prediction method of the superconductor performance, and can execute or be used to execute the data and instructions. In some embodiments, the prediction system 200 can include a hardware device with data information processing function and the necessary programs required to drive the hardware device to work.

[0066] It can be understood that the training system 100 and the prediction system 200 can correspond to the same system, or can correspond to different systems, and the present specification does not limit this.

[0067] It should be noted that the training system 100 can correspond to one device, or can correspond to a device cluster, and the present specification does not limit this. When the training system 100 corresponds to one device, the training method of the target network can be completely executed on the device. When the training system 100 corresponds to a device cluster, the training method of the target network can be cooperatively executed on multiple devices corresponding to the device cluster, or the training method of the target network can have other execution manners, and the present specification does not limit this.

[0068] The prediction system 200 can correspond to one device, or can correspond to a device cluster, and the present specification does not limit this. When the prediction system 200 corresponds to one device, the prediction method of the superconductor performance can be completely executed on the device. When the prediction system 200 corresponds to a device cluster, the prediction method of the superconductor performance can be cooperatively executed on multiple devices corresponding to the device cluster, or the prediction method of the superconductor performance can have other execution manners, and the present specification does not limit this.

[0069] The network structure of the target network, for example Figure 2 As shown, includes F levels, which are the first level network to the F level network. Each level network can include one or more sub-networks, and the number of sub-networks included in each level network gradually decreases from the first level to the F level. For example, the first level network includes m1 sub-networks, the second level network includes m2 sub-networks, m2 represents the first sub-network in the first level, and the sub-network represents the first sub-network in the second level, and so on. The F level network includes one sub-network. The output of each sub-network in the first level is used as the input of each sub-network in the second level, and so on. The network structure of each sub-network can be the same, and the present specification does not limit this.

[0070] The present specification does not limit the decreasing manner of the number of sub-networks in each level. For example, the number of sub-networks in each level decreases regularly, such as decreasing by a fixed number in turn, that is, the difference between the number of sub-networks in the first level and the number of sub-networks in the second level is the same as the difference between the number of sub-networks in the second level and the number of sub-networks in the third level. For another example, the number of sub-networks in each level decreases irregularly, such as decreasing by a non-fixed number.

[0071] It should be noted that the present specification does not limit the number of network levels of the target network, and the number of sub-networks included in each level.

[0072] For example, the sub-network is an Artificial Neural Network (ANN). The network structure of an ANN is as follows: Figure 3 As shown, it includes an input layer, hidden layers, and an output layout. Each neuron in an ANN layer is connected to the neuron in the previous layer, for example... Figure 3 shown Until The input to each neuron in the process includes Until Therefore, ANN can also be called a fully connected neural network. Here, j, j1, j2, and j3 are all integers greater than 0.

[0073] For example, the subnetwork can also be a convolutional neural network (CNN), and this specification does not limit the type of subnetwork.

[0074] Figure 4 A hardware structure diagram of a system 400 provided according to an embodiment of this specification is shown.

[0075] like Figure 4 As shown, the system 400 can be Figure 1 The training system 100 or the prediction system 200 are used.

[0076] The system 400 includes at least one storage medium 430 and at least one processor 420. In some embodiments, the system 400 may further include a communication port 450 and an internal communication bus 410. Furthermore, the system 400 may also include I / O components 460.

[0077] The internal communication bus 410 can connect to different system components. For example, the internal communication bus 410 can connect to storage medium 430, processor 420, communication port 450, and I / O component 460.

[0078] I / O component 460 supports input / output between the system 400 and other components.

[0079] Communication port 450 is used for data communication between the system 400 and the outside world. For example, communication port 450 can be used for data communication between the system 400 and a network. Communication port 450 can be a wired communication port or a wireless communication port.

[0080] In some embodiments, the network can be any type of wired or wireless network, or a combination thereof. For example, the network can include a cable network, a wired network, a fiber optic network, a telecommunication network, an intranet, the Internet, a Local Area Network (LAN), a Wide Area Network (WAN), a Wireless Local Area Network (WLAN), a Metropolitan Area Network (MAN), a Public Switched Telephone Network (PSTN), a Bluetooth network™, a short-range wireless network (ZigBee™), a Near Field Communication (NFC) network, or the like.

[0081] In some embodiments, the network can include one or more network access points. For example, the network can include wired or wireless network access points, such as a base station or an Internet exchange point. Through the access point, one or more components of each device corresponding to the system 400 can be connected to the network to exchange data or information.

[0082] The storage medium 430 can include a data storage device. The data storage device can be a non-transitory storage medium or a transitory storage medium. For example, the data storage device can include one or more of a disk 432, a read-only memory (ROM) 434, or a random access memory (RAM) 436. The storage medium 430 further includes at least one instruction set stored in the data storage device. The instruction set can include computer program code, which can include programs, routines, objects, components, data structures, procedures, modules, and the like, that perform the prediction method of superconductor performance or the training method of a target network provided in the present specification.

[0083] The processor 420 can be communicatively connected to the storage medium 430. The processor 420 is configured to execute the at least one instruction set. When the system 400 is running, the processor 420 reads the at least one instruction set and performs the prediction method of superconductor performance or the training method of a target network provided in the present specification according to the instructions of the at least one instruction set.

[0084] The processor 420 can be in the form of one or more processors. In some embodiments, the processor 420 can include one or more hardware processors, such as a microcontroller, a microprocessor, a reduced instruction set computer (RISC), an application specific integrated circuit (ASIC), an application specific instruction set processor (ASIP), a central processing unit (CPU), a graphics processing unit (GPU), a physics processing unit (PPU), a microcontroller unit, a digital signal processor (DSP), a field programmable gate array (FPGA), an advanced RISC machine (ARM), a programmable logic device (PLD), any circuit or processor capable of executing one or more functions, or the like, or any combination thereof.

[0085] For the sake of illustration only, only one processor 420 is shown in the system 400 in the drawings. However, it should be noted that the system 400 described in the present specification can also include multiple processors. Therefore, the operations and / or method steps disclosed in the present specification can be performed by one processor as described in the present specification, or jointly performed by multiple processors. For example, if the processor 420 of the system 400 described in the present specification performs step A and step B, it should be understood that step A and step B can also be performed jointly or separately by two different processors 420 (for example, a first processor performs step A, and a second processor performs step B, or the first and second processors jointly perform steps A and B).

[0086] Figure 5 A flowchart of a training method P500 of a target network is shown, which is provided according to an embodiment of the present specification. The training system 100 can perform the training method P500 of the target network, which is used to predict a target performance of a superconductor. As shown, the training method P500 of the target network includes the following steps. Figure 5

[0087] S510: Obtain experimental values of Y sample superconductors on a target performance respectively, each sample superconductor comprising at least one element.

[0088] Wherein Y is an integer greater than 1. The target performance can be understood as a parameter representing the properties of the superconductor, such as the critical temperature, etc. The number of elements included in different sample superconductors can be different, and the element proportion information corresponding to the elements can also be different.

[0089] A part of the Y sample superconductors include the same elements but different element proportion information in the sample superconductors, such as B 33.3333 C 16.6667 Ni 30.8333 Cu 2.5000 Y 16.6667 and B 33.3333 C 16.6667 Ni​23.3333 Cu 10.000 0Y 16.6667 ; some sample superconductors include different elements, and the proportion of elements in the sample superconductors is also different, such as B 33.3333 C 16.6667 Sc 8.3333 Ni 33.3333 Lu 8.3333 and B 33.3333 C 16.6667 Ni 33.3333 La 8.3333 Th 8.3333 ; some sample superconductors also include different numbers of elements, such as Al 13.0435 Ti 9.5652 Zr 9.5652 Nb 29.1304 Hf 9.5652 Ta 29.1304 and B 20.6186 N 27.8351 Ni 20.6186 La 29.8969 Ce 1.0309 .

[0090] The present specification takes the SuperCon database as the data source of the sample superconductors, which contains the Tc obtained by researchers through experimental research on known superconductors, and the superconductors involved are extensive, thereby ensuring the reliability of the target network training. The database contains related records of more than 14,794 superconductors, and some records have the following defects:

[0091] 1. Some records of superconductors include composition information, but lack Tc. There are 3052 such records, and the training system 100 excludes the records of these superconductors when determining the sample superconductors.

[0092] 2. The composition of some superconductors in the record is not clear. For example, materials such as Ca1Nb2O z and the like, it is difficult to determine the proportion information 'z' of oxygen atoms. The training system 100 also excludes the records of these superconductors when determining the sample superconductors, and clears 24 such records.

[0093] 3. Data redundancy. For example, superconductors with the same composition record different T c values, such as the T c value of Nb3Pt1 is recorded as 8.1K and 11K. For another example, different superconductors have the same composition elements, but the proportion information is proportional, or the order of elements in the chemical formula is different. For example, Ga7Pt3 and Ga 0.7 Pt 0.3 , C 1.35 Ti 0.1 Y0.9 and Y 1.8 Ti 0.2 C 2.7 represent the same superconductor. If the superconductors with the above-mentioned record problems are used as sample superconductors to train the target network, the following problems may exist: too much noise is introduced in the process of training the target network, which may affect the prediction accuracy of the target network; the same superconductor may exist in the training set and the test set at the same time, which affects the optimization of the target network, and further affects the prediction accuracy of the target network. Therefore, the training system 100 also excludes the records of these superconductors when determining the sample superconductors, and removes 6,505 such records.

[0094] After the training system 100 removes 9,581 records with the above-mentioned defects in the SuperCon database, 5,213 sample superconductors and experimental values of Tc of each sample superconductor are obtained. The SuperCon database mainly involves records about low-temperature superconductors.

[0095] Figure 6 The variation diagram of Tc of superconductors with element proportion information is given. a) gives the variation of Tc of La 2-x Sr x CuO4 system superconductor with element proportion information; b) gives the variation of Tc of Bi2Sr2CaCu2O y system superconductor with element proportion information.

[0096] Referring to a), in cuprate superconductors, Tc changes with the carrier density introduced by the doping element Sr, La 2- x Sr x CuO4 system superconductor shows a dome-shaped feature, indicating that superconductivity appears within a certain doping range. However, it is well known that the oxygen content is difficult to accurately measure, that is, the doping level caused by the change of oxygen content cannot be determined. Referring to b), Bi2Sr2CaCu2O y system superconductor does not show a dome-shaped feature, because the oxygen content cannot be accurately determined. This shows that the data of high-temperature superconductors contains many unreliable records before being strictly cleaned. Before using the experimental values of Tc of high-temperature superconductors for training of the target network, reliable data need to be determined, and whether the low-temperature superconductor (traditional superconductor) and the high-temperature superconductor have the same mechanism in predicting Tc has not been determined. Therefore, the related records of high-temperature superconductors are not used in the training process in the present specification, that is, the Y sample superconductors use the experimental values of Tc of low-temperature superconductors.

[0097] The experimental values of Tc of some sample superconductors (containing 5 elements or 6 elements) in the Y sample superconductors can be seen in Table 1.

[0098] Table 1

[0099]

[0100]

[0101] Table 1 continued

[0102]

[0103]

[0104] Table 1 continued

[0105]

[0106]

[0107] Table 1 continued

[0108]

[0109]

[0110] Table 1 continued

[0111]

[0112]

[0113] Table 1 continued

[0114] Chemical Formula T c experimental value (K) B 44.4444 Sc 2.7778 Y 8.3333 Rh 4.4444 Ir 40.0000 ]]> 2.92 B 33.3333 C 16.6667 Ni 33.3333 Dy 10.0000 Tm 6.6667 ]]> 5.77 B 33.3333 C 16.6667 Ni 33.3333 La 1.3333 Lu 15.3333 ]]> 13.6 B 33.3333 C 16.6667 Ni 21.6667 Y 16.6667 Pd 11.6667 ]]> 12.1 B 32.2581 C 16.1290 Co 6.4516 Ni 29.0323 Lu 16.1290 ]]> 2.6 B 25.0000 C 25.0000 Ni 20.0000 Cu 5.0000 Lu 25.0000 ]]> 3.38 B 33.3056 C 16.6528 Co 0.1665 Ni 33.2223 Dy 16.6528 ]]> 4.1 B 32.4149 C 16.2075 Co 5.5105 Ni 29.6596 Lu 16.2075 ]]> 3.8

[0115] The sample superconductors shown in Table 1 cover a large number of element types and element combinations, which is conducive to improving the prediction accuracy of the target network and expanding the application range of the prediction.

[0116] S520: For each sample superconductor, based on the proportion information of at least one element in the sample superconductor and the attribute values of the X attributes corresponding to the at least one element respectively, K descriptors for describing the sample superconductor are generated, and N1 core descriptors and K1 non-core descriptors are determined in the K descriptors, the influence degree of the core descriptor on the target performance is higher than that of the non-core descriptor, K1 = K-N1.

[0117] In some embodiments, the X properties include at least two dimensions from the following: a periodic table dimension; a thermodynamics dimension; a physics dimension; and a crystallography dimension. In this embodiment, the element properties of multiple dimensions are considered, and the K descriptors thus obtained can more comprehensively depict the sample superconductor, widely cover the superconductor properties, avoid missing important descriptors, and provide a good foundation for improving the prediction accuracy and generalization ability of the target network.

[0118] For example, the X properties refer to at least part of Table 2.

[0119] Table 2

[0120]

[0121]

[0122] In Table 2, BCC refers to a body-centered cubic lattice (Body Center Cubic), FCC refers to a face-centered cubic lattice (Face Center Cubic), ICSD refers to an inorganic crystal structure database (Inorganic Crystal Structure Database), and GS refers to the ground state of an object.

[0123] For example, the sample superconductor is B 44.4444 Sc 2.7778 Y 8.3333 Rh 4.4444 Ir 40.0000 The K descriptors describing B 44.4444 Sc 2.7778 Y 8.3333 Rh 4.4444 Ir 40.0000 may be generated based on the following: the proportion of B elements the proportion of Sc elements the proportion of Y elements the proportion of Rh elements the proportion of Ir elements and the attribute values of at least part of the properties listed in Table 2 corresponding to B elements, the attribute values of the at least part of the properties corresponding to Sc elements, the attribute values of the at least part of the properties corresponding to Y elements, the attribute values of the at least part of the properties corresponding to Rh elements, and the attribute values of the at least part of the properties corresponding to Ir elements.

[0124] The following describes the generation process of the K descriptors describing the sample superconductor #1 by taking the sample superconductor #1 as an example. The generation process of the K descriptors describing other sample superconductors can be similarly performed.

[0125] In some embodiments, the K descriptors describing the sample superconductor #1 include at least one of the following groups:

[0126] a first group of descriptors for describing statistical characteristics of the sample superconductor #1 on the X attributes;

[0127] a second group of descriptors for describing valence electron occupation states of the sample superconductor #1;

[0128] a third group of descriptors for describing ionicity of the sample superconductor #1; and

[0129] a fourth group of descriptors for describing configurational entropy (CE) of the sample superconductor #1.

[0130] In this embodiment, the K descriptors can describe different aspects of the characteristics of the sample superconductor, and the description of the sample superconductor is more comprehensive, which can support the prediction of complex multi-element superconductors, improve the prediction accuracy of the target network, and enhance the generalization ability of the target network.

[0131] For example, the first group of descriptors is generated by training the system 100 to obtain P statistical values of the characteristics of the sample superconductor #1 on each of the X attributes from multiple statistical dimensions based on the proportion information of each element included in the sample superconductor #1 in the sample superconductor #1 and the attribute value of each element on the attribute, and then generating the first group of descriptors based on the P statistical values of the sample superconductor #1 on the X attributes. The number of the first group of descriptors is equal to X*P. In this example, the first group of descriptors involves multiple statistical dimensions, the number of descriptors and the statistical dimensions involved are more, the basic statistical characteristics of the sample superconductor are mined, the sample superconductor can be accurately and comprehensively described, and the training of the target network is facilitated.

[0132] Taking attribute #1 as an example, the system 100 can obtain P statistical values of the characteristics of the sample superconductor #1 on attribute #1 from multiple statistical dimensions, and the P statistical values are all the descriptors corresponding to attribute #1 in the first group of descriptors. The generation process of all the descriptors corresponding to attribute #1 in the first group of descriptors is mainly introduced below, and the descriptors corresponding to other attributes in the first group of descriptors are similar and are not described in detail.

[0133] For example, the plurality of statistical dimensions includes statistical dimension #1, statistical dimension #2, statistical dimension #3, statistical dimension #4 and statistical dimension #5. Statistical dimension #1 is used to statistically calculate a weighted average of attribute values of each element included in the sample superconductor #1 on attribute #1. The formula corresponding to statistical dimension #1 is as follows:

[0134]

[0135] wherein x i represents the proportion of the i-th element in the chemical formula of the sample superconductor #1, f i represents the attribute value of the i-th element on attribute #1, represents the descriptor corresponding to attribute #1 obtained from statistical dimension #1. Assuming that the sample superconductor #1 is B 44.4444 Sc 2.7778 Y 8.3333 Rh 4.4444 Ir 40.0000 , the training system 100 can determine

[0136] Statistical dimension #2 is used to statistically calculate the weighted deviation of attribute values of each element included in the sample superconductor #1 on attribute #1. The formula corresponding to statistical dimension #2 is as follows:

[0137]

[0138] wherein x represents the descriptor corresponding to attribute #1 obtained from statistical dimension #2.

[0139] Statistical dimension #3 is used to statistically calculate the standard deviation of attribute values of each element included in the sample superconductor #1 on attribute #1. The formula corresponding to statistical dimension #3 is as follows:

[0140]

[0141] wherein f represents the descriptor corresponding to attribute #1 obtained from statistical dimension #3.

[0142] Statistical dimension #4 is used to statistically calculate the maximum value, the minimum value, the value range of attribute values or weighted attribute values of each element included in the sample superconductor #1 on attribute #1. The formula corresponding to statistical dimension #4 is as follows:

[0143] (f i )_min,max,range (4)

[0144] (x i *f i )_min,max,range (5)

[0145] The training system 100 can obtain 6 descriptors corresponding to attribute #1 from the statistical dimension #4, as follows:

[0146] (f i )_min represents the minimum value of the attribute values of each element included in the sample superconductor #1 on attribute #1;(f i )_max represents the maximum value of the attribute values of each element included in the sample superconductor #1 on attribute #1;(f i )_range represents the value range of the attribute values of each element included in the sample superconductor #1 on attribute #1, which can be the difference between (f i )_max and (f i )_min.(x i *f i )_min represents the minimum value of the weighted attribute values of each element included in the sample superconductor #1 on attribute #1;(x i *f i )_max represents the maximum value of the weighted attribute values of each element included in the sample superconductor #1 on attribute #1;(x i *f i )_range represents the value range of the weighted attribute values of each element included in the sample superconductor #1 on attribute #1, which can be the difference between (x i *f i )_max and (x i *f i )_min.

[0147] The statistical dimension #5 is used to statistically analyze the deviation degree of different elements in the sample superconductor #1 on attribute #1. The formula corresponding to the statistical dimension #5 is as follows:

[0148]

[0149] wherein, x i1 represents the proportion of the i1th element in the chemical formula of the sample superconductor #1, x i2 represents the proportion of the i2th element in the chemical formula of the sample superconductor #1, f i1 represents the attribute value of the i1th element on attribute #1, f i2 represents the attribute value of the i2th element on attribute #1, AP represents the deviation degree of different elements on attribute #1, and AP 加权 represents the deviation degree of the weighted attribute values of different elements on attribute #1.

[0150] The training system 100 can obtain 8 descriptors corresponding to attribute #1 from the statistical dimension #5, as follows:

[0151] AP_ave represents the average of the deviations of different elements in sample superconductor #1 at attribute #1; AP_min represents the minimum of the deviations of different elements in sample superconductor #1 at attribute #1; AP_max represents the maximum of the deviations of different elements in sample superconductor #1 at attribute #1; AP_range represents the range of the deviations of different elements in sample superconductor #1 at attribute #1, which can be the difference between AP_max and AP_min; AP 加权 _ave represents the average of the weighted attribute value deviations of different elements in sample superconductor #1 at attribute #1; AP 加权 _min represents the minimum of the weighted attribute value deviations of different elements in sample superconductor #1 at attribute #1; AP 加权 _max represents the maximum of the weighted attribute value deviations of different elements in sample superconductor #1 at attribute #1; AP 加权 _range represents the range of the weighted attribute value deviations of different elements in sample superconductor #1 at attribute #1, which can be the difference between AP 加权 _max and AP 加权 _min. AP, AP 加权 The APs and APs are related to the atomic size and the change of electronegativity between different elements, which have important influences on the structure and performance of superconductor materials. The training system 100 considers the APs and APs when generating the first group of descriptors. 加权 The APs and APs are very important to describe the comprehensiveness of the superconductor. In addition, if the sample superconductor #1 is composed of a single element, the above eight descriptors are all zero.

[0152] As can be seen, the training system 100 can obtain 17 descriptors corresponding to attribute #1 through the above multiple statistical dimensions. If the X attributes are the 53 kinds shown in Table 2, the first group of descriptors obtained by the training system 100 for the sample superconductor #1 can include 17*53=901 descriptors, which comprehensively cover the basic statistical attributes of the sample superconductor #1.

[0153] In some embodiments, the second group of descriptors includes the occupation information of Q kinds of valence electrons in the sample superconductor #1, wherein the occupation information of the zth kind of valence electron is obtained by the following way:

[0154] For each element in the sample superconductor #1, the training system 100 determines, based on attribute values of at least some of the X attributes corresponding to the element, a number of each valence electron in the element and a total number of valence electrons in the element; and for the z-th valence electron, determines, based on the percentage information corresponding to each element in the sample superconductor #1, the number of the z-th valence electron, and the total number of valence electrons corresponding to each element, the occupation information of the z-th valence electron.

[0155] That is, each descriptor in the second group of descriptors is used to describe a valence electron occupation state, and the second group of descriptors obtained by the training system 100 includes Q descriptors, and each valence electron can correspond to a descriptor. Valence electrons have an important influence on the superconductor reaction process, and therefore considering the occupation states of multiple valence electrons in the descriptors can characterize key attributes of the sample superconductor, and can improve the prediction accuracy of the target network on the catalytic performance.

[0156] For example, the training system 100 obtains the descriptor corresponding to the z-th valence electron based on the following formula:

[0157]

[0158] wherein F z represents the descriptor corresponding to the z-th valence electron, E z represents the number of the z-th valence electron in the i-th element in the chemical formula of the sample superconductor #1, E n represents the total number of Q valence electrons in the i-th element, such as valence electrons p, s, d, and f. Specifically, the descriptor corresponding to the valence electron p is The descriptors corresponding to other valence electrons are similar, and thus the training system 100 can obtain four descriptors.

[0159] In some embodiments, the third group of descriptors is obtained based on the percentage information corresponding to each element in the sample superconductor #1, and the electronegativity corresponding to each element and any one of the following: the maximum value of the electronegativity corresponding to each element included in the sample superconductor #1; and the average value of the electronegativity corresponding to each element included in the sample superconductor #1. The electronegativity is a measure of the ability of an atom of an element to attract electrons in a compound, and therefore considering the electronegativity in the descriptors can characterize key attributes of the sample superconductor, and can improve the prediction accuracy of the target network on the catalytic performance.

[0160] For example, the training system 100 obtains the third group of descriptors based on the following formula:

[0161]

[0162] wherein I represents the maximum value of electronegativity corresponding to each element included in the sample superconductor #1, or the average value of electronegativity corresponding to each element included in the sample superconductor #1; when when the maximum value of electronegativity is represented, the training system 100 can obtain the first descriptor in the third group of descriptors based on formula (9); when when the average value of electronegativity is represented, the training system 100 can obtain the second descriptor in the third group of descriptors based on formula (9); when when the average value of electronegativity is represented, the training system 100 can further obtain the third descriptor in the third group of descriptors, specifically, when the I calculated by the training system 100 based on formula (9) is greater than 1.7, the third descriptor is 1, and if the calculated I is not greater than 1.7, the third descriptor is 0 (which is also the standard for determining the formation of ionic bonds).

[0163] In some embodiments, the fourth group of descriptors is obtained based on the proportion information corresponding to each element in the sample superconductor #1, the Boltzmann constant and the room temperature. The configurational entropy can indicate the degree of disorder of atomic distribution, and considering the configurational entropy in the descriptors can depict the key attributes of the sample superconductor and improve the prediction accuracy of the target network on the catalytic performance.

[0164] For example, the training system 100 obtains the fourth group of descriptors based on the following formula:

[0165] ΔS con = -k B T∑x i *lnx i (10)

[0166] wherein ΔS con represents the configurational entropy; k B represents the Boltzmann constant; and T represents the room temperature, such as 298K. The training system 100 can obtain one descriptor based on formula (10).

[0167] In summary, in the case of 53 element attributes, for each sample superconductor, the training system 100 generates 901 first group of descriptors, 4 second group of descriptors, 3 third group of descriptors and 1 fourth group of descriptors based on formulas (1) to (10), i.e., the total number of K descriptors is 909.

[0168] ​In the field of materials, it takes a lot of time cost to obtain the Tc value of a sample superconductor through experiments, and therefore, in the case of a small number of sample superconductors, it is easy to have the problem of overfitting when a neural network is trained by using 909 descriptors as the characteristics of a sample superconductor, especially when the sample superconductor contains a large number of elements, it is more difficult to train a target network to predict more accurately. For example, in an ANN with two hidden layers, inputting 909 descriptors will require training to determine more than one billion hyperparameters, and in the case of a small number of sample superconductors, it is difficult to train to achieve the expected effect. Therefore, the present specification uses a genetic algorithm (Genetic Algorithm, GA) to reduce the number of input descriptors in a single subnetwork of the target network.

[0169] For example, the training system 100 groups the K descriptors based on a competitive swarm optimizer (CSO) algorithm, and the descriptors in each group compete in pairs. After each competition, the winning descriptor enters the next iteration, and the losing descriptor can learn from the winning descriptor to update its position and speed. The basic principle of the CSO algorithm can be seen from the following formula:

[0170]

[0171] where t represents the number of iterations; and are three vectors randomly generated in the range [0, 1] n For each descriptor, '1' indicates that the descriptor is selected after competition, and '0' indicates that the descriptor is not selected after competition; 'n' represents the number of descriptors, for example, 'n' is 909; the variable and represent the winning descriptor and the losing descriptor, respectively, and the variable represents the average position of the descriptor group in the tth iteration, and φ determines the influence of , represents the speed of the i-th descriptor in the tth iteration.

[0172] In this example, the training system 100 uses 1-R 2As the fitness function, the fitness function is minimized by strategically selecting descriptors. Specifically, at the beginning of the iteration, the training system 100 randomly selects a plurality of descriptor combinations and their velocities, and then randomly selects two of the descriptor combinations for fitness function evaluation. The descriptor combination with the lower fitness function value is the winner, and the descriptor combination with the higher fitness function value is the loser. During the iteration process, the update and iteration are performed according to formulas (11) and (12). When the fitness function value no longer decreases, the iteration process is terminated.

[0173] Figure 7 The GA iteration process is exemplified, the x-axis represents the number of GA iterations, the left y-axis represents the number of descriptors (d), and the right y-axis represents the test R 2 During the GA iteration process, a plurality of preferred descriptor combinations appear. The difference between the predicted value of the critical temperature Tc of the sample superconductor predicted by the preferred descriptor combination and the experimental value is less than or equal to a preset threshold 1.

[0174] For example, continuing to refer to Figure 7 After the iteration selection, the training system 100 determines that when N is 49, the prediction accuracy reaches a turning point, the test R 2 is 0.92, and tends to be stable. That is, when the number of descriptors input for each subnetwork is 49, Tc can be relatively accurately predicted.

[0175] It should be noted that the determination of N can be related to the number of sample superconductors participating in the GA, and therefore the specific value of N is not limited in the specification.

[0176] In addition, during the iteration process, the descriptors with a frequency greater than or equal to a preset threshold 2 in the plurality of preferred descriptor combinations are used as core descriptors required for predicting Tc, and the descriptors with a frequency less than the preset threshold 2 are used as non-core descriptors required for predicting Tc.

[0177] Table 3 gives the core descriptors required for predicting Tc. The frequencies of the five core descriptors MD1 to MD5 are 1055, 960, 944, 914, and 902, respectively, and the frequencies of other non-core descriptors are all less than 500, which indicates the importance of the core descriptors.

[0178] Table 3

[0179]

[0180]

[0181] For example, the AP_ave_BCCfermi represents the average value of the deviation of different elements in the sample superconductor at the BCC Fermi level, which can be obtained based on formula (6). The ave_FirstIonizationEnergy represents the average value of the attribute value of each element in the sample superconductor at the first ionization energy, which can be obtained based on formula (4). The AP_min_GSestBCClatcnt represents the minimum value of the deviation of each element in the sample superconductor at the ground state BCC lattice constant, which can be obtained based on formula (6). The ave_NValance represents the average value of the attribute value of each element in the sample superconductor at the filling number of total valence electrons, which can be obtained based on formula (4). The AP_min_thermal_conductivity represents the minimum value of the deviation of different elements in the sample superconductor at the thermal conductivity, which can be obtained based on formula (6).

[0182] The example gives a plurality of descriptors with high correlation with Tc, which are taken as core descriptors in the training target network process. The target network configures appropriate weights for the core descriptors in the training learning process to reflect the importance of the core descriptors in predicting Tc.

[0183] Figure 8 The relationship between the above core descriptors and Tc is illustrated. In which, a) shows the relationship between MD1 and Tc; b) shows the relationship between MD2 (eV) and Tc; c) shows the relationship between MD4 and Tc; d) shows the relationship between MD3 and Tc; f) shows the relationship between MD5 (W / m*K) and Tc.

[0184] S530: inputting part of the K descriptors corresponding to each sample superconductor into the target network for prediction to obtain a predicted value of the target performance of the sample superconductor. Wherein: the target network includes a plurality of sub-networks with the same structure and arranged in levels, each sub-network includes N input nodes and 1 output node, the output node of the sub-network in the i-th level is used as the input node of the sub-network in the i+1-th level, each sub-network in the first level is inputted with the N1 core descriptors and N2 non-core descriptors, the N2 non-core descriptors are part of the K1 non-core descriptors, and the non-core descriptors inputted into different sub-networks are not completely the same, N2=N-N1.

[0185] Wherein, X, K, K1, N1, N2, Y, N are integers greater than 1, and i is an integer greater than or equal to 1. The target network can avoid the overfitting problem of the target network training when there are too many descriptors but the number of sample superconductors is insufficient. It should be noted that each sub-network can have different weighting factors during training.

[0186] In some embodiments, see Figure 9 The training system 100 generates m1 sets of descriptors based on the K descriptors, each set including the N1 core descriptors and N-N1 non-core descriptors; the m1 sets of descriptors are input into the m1 subnetworks of the first layer of the target network to obtain m1 intermediate prediction values ​​of the target performance; according to the order of i from 2 to F, the m1 subnetworks of the (i-1)th layer of the target network are... 1-1 The target performance m obtained from the subnetwork i-1 The intermediate predicted values ​​are divided into m i Group input of m of the i-th layer of the target network i A subnetwork is used to obtain the target performance m. i The intermediate predicted values ​​are then used as the final predicted values ​​of the target performance output by the sub-network of layer F. Where m1, F, and m are intermediate predicted values. i-1 All are integers greater than 1, m i It is an integer greater than or equal to 1.

[0187] In this embodiment, dividing the K descriptors into multiple groups and inputting them into multiple sub-networks avoids the problem of overfitting caused by too many descriptors inputting into a single sub-network, which would result in an excessive number of hyperparameters required for training due to insufficient sample superconductors. Figure 9 The hierarchical structure of the target network shown allows descriptors to be distributed across multiple subnetworks, avoiding overfitting and expanding the number of descriptors to 909. This enriches the characterization of superconductors. Each subnetwork receives a relatively small number of input descriptors, resolving the machine learning contradiction arising from a large number of descriptors but a limited number of superconductor samples. Furthermore, testing (see below for the testing process) demonstrates that the target network's prediction accuracy surpasses that of other neural networks, which is beneficial for superconductor research.

[0188] For example, when K is 909, the training system 100 can obtain the following from 909-5=904 non-core descriptors through random combination. There are 5 non-core descriptor groups, and each non-core descriptor group plus 5 core descriptors can be obtained. There are 10 descriptor groups. In other words, the number of subnetworks in the first level can reach 100,000. However, this specification does not limit the number of subnetworks in the first level.

[0189] Different descriptor groups play different degrees in predicting Tc, and the degrees can be determined in the process of training the target network. The input descriptor groups of each subnetwork in the first layer are different, which can improve the generalization ability of the target network and reduce the possibility of overfitting. Compared with the traditional training of a network through a descriptor group, the training method of the present specification considers various combinations of different descriptors, which can capture or mine the potential influence of different descriptor combinations on predicting Tc, so as to train a target network with high prediction accuracy. Traditional algorithms such as bagging, boosting and stacking usually rely on simple geometric mean, without considering the contribution difference of different descriptor groups to predicting Tc, nor the difference of different descriptor groups to the performance of the target network.

[0190] For the first layer of the target network, the input of each subnetwork is a descriptor group, and each descriptor group includes N (for example, 49) descriptors; that is, each input node of each subnetwork in the first layer corresponds to a descriptor. For example, the input of subnetwork is descriptor group 1, and so on. Each descriptor group includes N1 (for example, 5) core descriptors and N2 (for example, 44) non-core descriptors. Different descriptor groups include at least one different N2 non-core descriptor. For the second layer, the input of each subnetwork is N intermediate prediction values output by the subnetworks in the first layer, for example, the input of subnetwork is N intermediate prediction values output by subnetworks to subnetwork , which are to For the third layer, the input of each subnetwork is N intermediate prediction values output by the subnetworks in the second layer, and so on. For the Fth layer, the input of subnetwork F is N intermediate prediction values output by the subnetworks in the F-1th layer.

[0191] The target network can learn the iterative knowledge of different subnetworks in the training process, and different subnetworks can include overlapping knowledge and unique knowledge, so as to improve the prediction accuracy. In the case of limited number of sample superconductors, the number of hyperparameters is too high when training a neural network through 909 descriptors, which affects the training accuracy. The target network gradually converges the prediction value of the target performance with the increase of the number of layers.

[0192] For example, in the process of training the target network, the target performance corresponding to the sample superconductor is Tc, the core descriptors input by each subnetwork in the first layer are MD1 to MD5, and the output of each subnetwork after the first layer is the prediction value of Tc.

[0193] S540: updating the parameters of the target network with the training objective of minimizing the difference between the predicted value and the experimental value of the target performance of the sample superconductor.

[0194] Illustratively, the present specification takes the coefficient of determination (R 2 ) and the mean absolute percentage error (MAPE) as the evaluation indicators of the target network. The calculation formulas of MAPE and R 2 are as follows:

[0195]

[0196] wherein n represents the number of sample superconductors in the current training round, y i represents the experimental value of the i-th sample superconductor with respect to the target performance, represents the average experimental value of the sample superconductor with respect to the target performance, F(x i ) represents the predicted value of the i-th sample superconductor with respect to the target performance predicted by the target network. The higher the R 2 value and the lower the MAPE value, the better the performance of the target network, i.e., the more accurate the prediction result.

[0197] Illustratively, the training system 100 divides the Y sample superconductors into a training set and a test set, and trains the following network models including the target network: 49-ANN, 49-XGBoost, 49-CNN, 145-ANN, 145-XGBoost, 145-HNN(ANN), 145-CNN, 145-HNN(CNN), 909-ANN, 909-HNN(ANN), 909-CNN, 909-XGBoost and 909-HNN(CNN). Among them, the number represents the number of descriptors input into the network, and HNN represents the target network.

[0198] For example, "49-ANN" indicates that the network to be trained is an ANN, and the input of the ANN is 49 descriptors obtained through GA (including the 5 core descriptors given in Table 3); "49-XGBoost", "49-CNN" and "49-ANN" have the same input descriptors, but the networks to be trained are different; "145-ANN" indicates that the network to be trained is an ANN, and the input of the ANN is 145 traditional descriptors (different from the 909 descriptors provided in this specification); "145-XGBoost", "145-CNN" and "145-ANN" have the same input descriptors, but the networks to be trained are different; "145-HNN(ANN)" indicates that the network to be trained is the target network, and each subnetwork in the target network is an ANN, and the input of the target network is the aforementioned 145 descriptors; "14 "5-HNN(CNN)" indicates that the network to be trained is the target network, and each subnetwork in the target network is a CNN, and the input of the target network is the 145 descriptors; "909-ANN" indicates that the network to be trained is an ANN, and the input of the ANN is the 909 descriptors provided in this specification; "909-XGBoost", "909-CNN" and "909-ANN" have the same input descriptors, but different networks to be trained; "909-HNN(ANN)" indicates that the network to be trained is the target network, and each subnetwork in the target network is an ANN, and the input of the target network is the 909 descriptors; "909-HNN(CNN)" indicates that the network to be trained is the target network, and each subnetwork in the target network is a CNN, and the input of the target network is the 909 descriptors.

[0199] Figure 10 The training results of the aforementioned network models are shown. The horizontal axis represents the different network models, and the left vertical axis represents the test R. 2 The left ordinate passes through ΔR 2 Indicate training R 2 and test R 2 The differences between them. Training R 2 This represents the relationship between the network model's predicted and experimental values ​​of Tc for superconductors in the training set; testing R... 2 This represents the relationship between the network model's predicted and experimental values ​​of Tc for superconductors in the test set.

[0200] See Figure 10 The above network model's test R on the test set 2The values from low to high in order are: 49-ANN, 145-ANN, 49-XGBoost, 145-CNN, 145-XGBoost, 145-HNN(ANN), 145-HNN(CNN), 49-CNN, 909-ANN, 909-XGBoost, 909-HNN(ANN), 909-CNN, and 909-HNN(CNN).

[0201] Figure 11 The detailed learning results of the above partial network models are given. Among them, a) shows the learning result of 145-ANN; b) shows the learning result of 145-HNN(ANN); c) shows the relationship between the network layer number and the test R 2 of 145-HNN(ANN); d) shows the learning result of 145-CNN; e) shows the learning result of 145-HNN(CNN); f) shows the relationship between the network layer number and the test R 2 of 145-HNN(CNN); g) shows the learning result of 909-ANN; h) shows the learning result of 909-HNN(ANN); i) shows the relationship between the network layer number and the test R 2 of 909-HNN(ANN); j) shows the learning result of 909-CNN; k) shows the learning result of 909-HNN(CNN); and l) shows the relationship between the network layer number and the test R 2 of 909-HNN(CNN).

[0202] From Figure 10 and Figure 11 the following conclusions can be drawn:

[0203] 1. Referring to a) of Figure 11 , when T c is greater than 20K, “145-ANN” shows a large error on the training set, indicating that “145-ANN” is under-fitted, further indicating that the 145 descriptor training network model is incomplete.

[0204] 2. Referring to k) of Figure 11 , “909-HNN(ANN)” has almost no error on the training set and the test set, and has strong prediction ability, which improves the test R 2 of “145-ANN” from 0.830 to 0.956.

[0205] 3. The performance of the CNN network model is better than that of the ANN.

[0206] 4、The performance of the network models such as “145-HNN(ANN)”, “145-HNN(CNN)”, “909-HNN(ANN)”, “909-HNN(CNN)” increases with the increase of the number of network layers, and increases with the increase of the number of sub-networks in the network layer, see Figure 11 c), f), i) and l) of FIG. 14, the test R 2 is increased at the 4th network level.

[0207] 5、When the number of descriptors describing a sample superconductor is reduced from 909 to 49, although the single ANN / CNN / XGBoost 2 , the test R 2 is also reduced, which indicates that the reduction of the number of descriptors alleviates the overfitting effect, but may miss some information related to Tc. Compared with the single ANN / CNN / XGBoost network model, the HNN network model such as “909-HNN(ANN)”, “909-HNN(CNN)” not only realizes a higher test R 2 , but also reduces the 2 , i.e. reduces the overfitting.

[0208] In order to further demonstrate the prediction ability of “909-HNN(CNN)”, the present specification uses high-entropy alloys (HEAs) for testing, because HEAs have the following common characteristics with the Y sample superconductors: first, HEAs are metals, similar to the Y sample superconductors in electrical conductivity, and candidate materials for superconductors usually need to be conductors. The Y sample superconductors almost cover all known conductive materials, so possible new conductors mainly come from two categories: amorphous alloys and HEAs. Given that the long-range disordered atomic structure of amorphous alloys may have different superconducting mechanisms, the present specification selects HEAs as new test superconductors. Second, the Tc of HEAs is usually lower than 10K, much lower than the McMillan limit (about 40K), which means that the superconducting pairing mechanism in HEAs is consistent with that in the Y sample superconductors. Third, the elements constituting HEAs are all contained in the Y sample superconductors, Figure 12 The elements contained in the Y sample superconductors and HEAs are visually displayed, wherein all the elements contained in 45 HEAs are contained in the Y sample superconductors, and the number of elements of the 45 HEAs is also similar to the number of elements of the sample superconductors listed in Table 1, with more than 4 elements. Therefore, the target network trained by the method P500 should have the ability to predict the Tc of the 45 HEAs

[0209] Exemplarily, the 13 trained network models above were used to predict the Tc of the 45 HEAs, respectively. Figure 13 The prediction results of the 45 HEAs by the network models above are given. Wherein a) gives the relationship between the predicted values and experimental values of the Tc of the 45 HEAs by “909-HNN(CNN)”; b) gives the test R 2 and ΔR’ 2 ; c) gives the relationship between the experimental values of the Tc of the Y sample superconductors, the element serial number count and the element number.

[0210] Referring to Figure 13 , the R 2 of “49-ANN”, “145-ANN”, “49-XGBoost”, “145-CNN”, “145-XGBoost”, “145-HNN(ANN)”, “145-HNN(CNN)”, “49-CNN”, “909-ANN”, “909-XGBoost”, “909-HNN(ANN)”, “909-CNN” and “909-HNN(CNN)” are 0.72, 0.162, 0.67, 0.502, 0.512, 0.563, 0.694, 0.764, 0.53, 0.585, 0.81, 0.651 and 0.92, respectively. The prediction results above are consistent with the prediction results of the network models given in Table 2. Figure 10 The R 2 of “909-HNN(CNN)” decreases from 0.956 to 0.92, see Fig. a), and the overfitting of “909-HNN(CNN)” can be ignored. In contrast, the R 2 of other network models decreases more significantly. For example, the R 2 of “145-ANN” decreases from 0.891 to 0.162, and the overfitting is very serious.

[0211] In addition, Fig. c) shows the distribution of the T c of the Y sample superconductors according to the element number, which contains the 215 sample superconductors given in Table 1. From Fig. c), it is found that the T c of the sample superconductors with the element number between 2 and 4 is higher.

[0212] Table 4 gives the experimental values of Tc of the 45 HEAs and the predicted values of Tc predicted by the "909-HNN(CNN)", in which 23 HEAs contain 5 elements and 22 HEAs contain 6 elements. In the prediction process of the HEAs containing 5 elements and the HEAs containing 6 elements by the "909-HNN(CNN)", the MAPEs are 5.3% and 6.1% respectively, and the overall MAPE of the 45 HEAs is 5.7%, less than 6%, which is relatively accurate. The above results show the prediction stability of the target network. The method P500 can train a large number (e.g., more than 10 3 ) of sub-networks in parallel in the process of training the target network, learn the iterative knowledge of different sub-networks, and include overlapping knowledge and unique knowledge between different sub-networks, so as to improve the prediction accuracy.

[0213] Table 4

[0214]

[0215]

[0216] In summary, the training method and the training system of the target network provided in the specification expand the number of descriptors describing the sample superconductor, enrich the description dimension of the sample superconductor, and can group K descriptors for parallel training of sub-networks through multiple levels of target networks, avoiding the contradiction between excessive number of descriptors and insufficient training samples. First, the significant increase in the number of descriptors combined with the target network can improve the prediction accuracy of the performance of the superconductor, and even a small number of training samples can significantly enhance the prediction ability of the target network, which can reduce the experimental cost in the research process of the superconductor. Moreover, when grouping the descriptors, each group includes the same core descriptor and different non-core descriptors, which considers the importance of the core descriptor and excavates the potential influence degree of different non-core descriptor combinations on the performance of the superconductor. Combined with the prediction performance data given in the foregoing, the prediction accuracy of the target network is higher than that of other networks, which is beneficial to accelerate the process of superconductor material innovation research.

[0217] Figure 14 A flowchart of a prediction method P1400 of a performance of a superconductor is shown, which is provided by an embodiment of the specification. The prediction system 200 can perform the prediction method P1400 of the performance of the superconductor to predict a target performance of a target superconductor, and the target superconductor includes at least one element. As shown in the flowchart, the prediction method P1400 of the performance of the superconductor includes the following steps. Figure 14

[0218] ​S1410: generating K descriptors for describing the target superconductor based on the proportion information of the at least one element in the target superconductor and the attribute values of the X attributes corresponding to the at least one element respectively.

[0219] In some embodiments, the X attributes include at least two dimensions in the following: element periodic table dimension; thermodynamic dimension; physics dimension; and crystallographic dimension.

[0220] Specifically, the X attributes can refer to Table 2, which will not be described here.

[0221] In some embodiments, the K descriptors include at least one of the following groups:

[0222] The first group of descriptors is used to describe the statistical characteristics of the target superconductor in the X attributes.

[0223] The second group of descriptors is used to describe the valence electron occupation state of the target superconductor.

[0224] The third group of descriptors is used to describe the ionization of the target superconductor; and

[0225] The fourth group of descriptors is used to describe the conformational entropy of the target superconductor.

[0226] In some embodiments, the first group of descriptors is generated by the following method:

[0227] For each of the X attributes, the prediction system 200 obtains P statistical values of the target superconductor in the attribute from multiple statistical dimensions based on the proportion information of the at least one element in the target superconductor and the attribute values of the at least one element in the attribute respectively; and generates the first group of descriptors based on the P statistical values of the target superconductor in the X attributes respectively, and the number of the first group of descriptors is equal to X*P.

[0228] Specifically, the generation method of the first group of descriptors corresponding to the target superconductor is similar to the generation method of the first group of descriptors corresponding to the sample superconductor #1, and the specific generation method can refer to the above, which will not be described here.

[0229] In some embodiments, the second group of descriptors includes the occupation information of Q valence electrons in the target superconductor, wherein the occupation information of the zth valence electron is obtained by the following method:

[0230] For each of the at least one element, the prediction system 200 determines the number of each valence electron and the total number of valence electrons in the element based on the attribute values of at least part of the X attributes corresponding to the element; and for the zth valence electron, the prediction system 200 determines the occupation information of the zth valence electron based on the proportion information of the at least one element in the target superconductor, the number of the zth valence electron corresponding to each of the at least one element, and the total number of valence electrons corresponding to each of the at least one element.

[0231] Specifically, the third group of descriptors corresponding to the target superconductor is generated in a manner similar to the third group of descriptors corresponding to the sample superconductor #1. For details, reference can be made to the foregoing, and details are not repeated herein.

[0232] In some embodiments, the third group of descriptors is obtained based on the proportion information of the at least one element in the target superconductor, the electronegativity corresponding to each of the at least one element, and any one of the following:

[0233] the maximum value of the electronegativity corresponding to each of the at least one element;

[0234] the average value of the electronegativity corresponding to each of the at least one element.

[0235] Specifically, the third group of descriptors corresponding to the target superconductor is generated in a manner similar to the third group of descriptors corresponding to the sample superconductor #1. For details, reference can be made to the foregoing, and details are not repeated herein.

[0236] In some embodiments, the fourth group of descriptors is obtained based on the proportion information of the at least one element in the target superconductor, the Boltzmann constant, and the room temperature.

[0237] Specifically, the fourth group of descriptors corresponding to the target superconductor is generated in a manner similar to the fourth group of descriptors corresponding to the sample superconductor #1. For details, reference can be made to the foregoing, and details are not repeated herein.

[0238] S1420: Determine N1 core descriptors and K1 non-core descriptors from the K descriptors, the influence degree of the core descriptors on the target performance being higher than the influence degree of the non-core descriptors on the target performance.

[0239] In some embodiments, the target performance is the critical temperature Tc, and the N1 core descriptors include a plurality of the following:

[0240] MD1, representing the average value of the deviation of different elements in the at least one element at the BCC Fermi level;

[0241] MD2, representing the average value of the attribute values of the first ionization energy of the at least one element;

[0242] MD3, characterizing the minimum deviation of the different elements in the ground-state BCC lattice constant among the at least one element;

[0243] MD4, representing the average of the attribute values ​​of the at least one element in terms of the total number of valence electrons; and

[0244] MD5 is the minimum deviation in thermal conductivity among the at least one element.

[0245] For example, after the prediction system 200 determines the N1 core descriptors from MD1 to MD5, it uses the remaining descriptors among the K descriptors as non-core descriptors.

[0246] S1430: Input at least a portion of the K descriptors into the target network for prediction to obtain the performance index corresponding to the target performance. The target network is a network trained using Y sample superconductors. The target network includes multiple sub-networks with the same structure and arranged in layers. Each sub-network includes N input nodes and 1 output node. The output node of the sub-network in the i-th layer serves as the input node of the sub-network in the (i+1)-th layer. Each sub-network in the first layer is input with the N1 core descriptors and N2 non-core descriptors. The N2 non-core descriptors are a portion of the K1 non-core descriptors, and the non-core descriptors input to different sub-networks are not completely identical.

[0247] In some embodiments, the prediction system 200 generates m1 sets of descriptors based on the K descriptors, each set including the N1 core descriptors and N-N1 non-core descriptors; the m1 sets of descriptors are input into the m1 sub-networks of the first layer of the target network to obtain m1 intermediate performance indicators of the target performance; the m1 sub-networks of the (i-1)th layer of the target network are then processed according to the order of i values ​​from 2 to F. 1-1 The target performance m obtained from the subnetwork i-1 Each intermediate performance metric is divided into m i Grouping inputs the m of the i-th layer of the target network i A subnetwork is used to obtain the target performance m. i There are several intermediate performance metrics; and the intermediate performance metrics of the target performance output by the subnetwork of layer F are used as the final performance metrics of the target performance.

[0248] For example, when the target performance is Tc, the prediction system 200 can invoke the target network. Specifically, the prediction system 200 can input the m1 sets of descriptors respectively as follows: Figure 9 The target network has m1 subnetworks in its first layer, resulting in m1 intermediate performance metrics for Tc. Following the order of i from 2 to F, the m1 subnetworks in the (i-1)th layer of the target network T are...1-1 m intermediate performance indicators of Tc obtained by the m sub-networks of the i-th layer of the target network i-1 i m intermediate performance indicators of Tc obtained by the m sub-networks of the i-th layer of the target network i m intermediate performance indicators of Tc obtained by the m sub-networks of the i-th layer of the target network i m intermediate performance indicators of Tc obtained by the m sub-networks of the i-th layer of the target network

[0249] It can be understood that part of the technical details involved in the superconductor performance prediction method P1400 can refer to the description in the target network training method P500, which will not be repeated. The beneficial effects of the above-mentioned embodiments of the method P1400 can also refer to the beneficial effect description of the embodiments of the method P500, which will not be repeated.

[0250] In summary, the superconductor performance prediction method and the training system provided by the present specification expand the number of descriptors describing the target superconductor, enrich the description dimension of the target superconductor, and can group K descriptors and input them into the sub-network for parallel prediction through the target network of multiple levels. First, the significant increase in the number of descriptors combined with the target network can improve the prediction accuracy of the catalytic performance of the target superconductor, and can reduce the experimental cost in the research process of the superconductor. Furthermore, when grouping the descriptors, each group includes the same core descriptor and the non-core descriptor which is not completely the same. It not only considers the importance of the core descriptor, but also excavates the potential influence degree of different non-core descriptor combinations on the superconductor performance of the target superconductor. Combined with the prediction data given in the foregoing, the prediction accuracy of the target network is higher than that of other networks, which is beneficial to accelerate the process of superconductor material innovation research.

[0251] ​Another aspect of the present specification provides a computer-readable non-transitory storage medium storing at least one set of executable instructions for performing the prediction of superconductor performance or the training of a target network. When the executable instructions are executed by a processor, the executable instructions direct the processor to implement the steps of the prediction of superconductor performance method P1400 or the training of a target network method P500 described in the present specification. In some possible implementation, each aspect of the present specification can also be implemented as a program product in the form of a computer program code, which includes a program code. When the program product is run on the system 400, the program code is used to make the system 400 perform the steps of the prediction of superconductor performance method P1400 or the training of a target network method P500 described in the present specification. The program product for implementing the above method can include the program code in a portable compact disc read-only memory (CD-ROM) and can be run on the system 400. However, the program product of the present specification is not limited to this, and in the present specification, the readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system. The program product can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium may, for example, be but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples of the readable storage medium include an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. The computer-readable storage medium can include a data signal propagating in a baseband or as a carrier wave in a propagated signal, in which the readable program code is borne. Such a propagated signal can take on multiple forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination thereof. The readable storage medium can also be any readable medium that is not a storage medium and that can send, propagate or transfer a program for use by or in connection with an instruction execution system, apparatus or device. The program code contained in the readable storage medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, and the like, or any suitable combination thereof. The program code for performing the operations of the present specification can be written in any combination of one or more programming languages, including an object-oriented programming language, such as Java, C++, and the like, and a conventional procedural programming language, such as the "C" programming language or similar programming languages.The program code can execute entirely on the system 400, partly on the system 400, as a stand-alone software package, partly on the system 400 and partly on a remote computing device, or entirely on the remote computing device.

[0252] The term "and / or" in the embodiments of the present specification describes an associated relationship between associated objects, which means that there can be three relationships. For example, A and / or B can mean that there are three cases of A alone, A and B together, and B alone. The character " / " generally represents that the associated objects before and after are in an "or" relationship.

[0253] The terms "first", "second", and the like in the present specification are used to distinguish similar or similar objects or entities, and do not necessarily mean to limit a specific order or sequence.

[0254] The term "a plurality of" in the present specification shall be understood as two or more, unless otherwise specifically stated.

[0255] The above describes specific embodiments of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different than the order in the embodiments and still achieve the desired result. In addition, the processes depicted in the accompanying drawings do not necessarily require the particular order or sequential order to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0256] In summary, after reading the detailed disclosure, those skilled in the art can understand that the foregoing detailed disclosure can be presented only in an exemplary manner, and can not be limiting. Although it is not explicitly stated here, those skilled in the art can understand that the present specification requires various reasonable changes, improvements and modifications to the embodiments. These changes, improvements and modifications are intended to be presented by the present specification, and are within the spirit and scope of the exemplary embodiments of the present specification.

[0257] In addition, certain terms in the present specification have been used to describe the embodiments of the present specification. For example, "one embodiment", "embodiment" and / or "some embodiments" means that the specific features, structures or characteristics described in connection with the embodiment can be included in at least one embodiment of the present specification. Therefore, it can be emphasized and should be understood that two or more references to "embodiments" or "one embodiment" or "alternative embodiments" in various parts of the present specification do not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics can be appropriately combined in one or more embodiments of the present specification.

[0258] It is to be understood that the foregoing description of the embodiments of the present specification is intended to be illustrative only and that other embodiments of the present specification are possible being encompassed by the present specification. For the purpose of simplicity of the present specification, various features of the present specification are grouped together in one or more embodiments, drawings or descriptions thereof. However, the features described as grouped together are not to be understood as being indispensable to a single embodiment of the present specification. In fact, some of the features as described above can be implemented independently of all other features and thus may be implemented in one or more embodiments, in any combination. Similarly, the description of the embodiments of the present specification is to be understood as meaning that the various embodiments of the present specification can also be understood as a combination of the various sub-embodiments. In other words, the embodiments of the present specification can be understood as a combination of the various sub-embodiments, each of which is characterized by less than all features of the single embodiment as described above.

[0259] Each patent, patent application, publication of a patent application, and other material, such as articles, books, specifications, publications, documents, literature, and the like (including any and all related prosecution history documents) cited in this disclosure are hereby incorporated by reference in their entirety for all purposes to the same extent as if each were specifically and individually indicated to be incorporated by reference herein. In the event of any inconsistency between the description, definition, and / or terminology in the aforementioned materials and that used in the present disclosure, the description, definition, and / or terminology in the present disclosure control.

[0260] Finally, it is to be understood that the embodiments of the application disclosed herein are illustrative of the principles of the present specification. Other modifications that fall within the scope of the present specification can also be made. Thus, the present specification discloses the best mode contemplated for carrying out the application. Of course, those skilled in the art will recognize improvements in the preferred embodiments. Accordingly, the patent application discloses the principles of the application and their practical application and uses as desired. The present specification is therefore to be construed in such a light as not to foreshadow any disclosure in the prior art which is not intended to be incorporated by reference in the present specification.

Claims

1. A method for predicting a performance of a target superconductor, the target superconductor comprising at least one element, the method comprising: generating K descriptors for describing the target superconductor based on information about a proportion of the at least one element in the target superconductor and attribute values of X attributes corresponding to the at least one element respectively; determining N1 core descriptors and K-N1 non-core descriptors from the K descriptors, the core descriptors having a higher influence on the target performance than the non-core descriptors; and inputting at least part of the K descriptors into a target network to obtain a performance index corresponding to the target performance, wherein the target network is a network trained by Y sample superconductors, the target network comprising a plurality of sub-networks with the same structure and arranged in a hierarchical manner, each sub-network comprising N input nodes and one output node, and an output node of a sub-network in an i-th level being an input node of a sub-network in an (i+1)-th level, each sub-network in a first level being inputted with the N1 core descriptors and N-N1 non-core descriptors, the N-N1 non-core descriptors being part of the K-N1 non-core descriptors, and different sub-networks being inputted with non-core descriptors that are not completely the same, wherein X, K, N1, Y and N are integers greater than 1, and i is an integer greater than or equal to 1. the X attributes comprise at least two dimensions of the following: a periodic table dimension; a thermodynamic dimension; a physical dimension; or a crystallographic dimension. the K descriptors comprise at least one of the following groups: a first group of descriptors for describing statistical characteristics of the target superconductor in the X attributes; a second group of descriptors for describing valence electron occupation states of the target superconductor; a third group of descriptors for describing ionization of the target superconductor; and a fourth group of descriptors for describing conformational entropy of the target superconductor. the first group of descriptors is generated by: for each of the X attributes, obtaining P statistical values of characteristics of the target superconductor in the attribute from a plurality of statistical dimensions based on information about a proportion of the at least one element in the target superconductor and attribute values of the at least one element in the attribute respectively; and generating the first group of descriptors based on the P statistical values of the target superconductor in the X attributes respectively, the number of the first group of descriptors being equal to X*P. the second group of descriptors comprises occupation information of Q valence electrons in the target superconductor, wherein occupation information of a z-th valence electron is obtained by: for each of the at least one element, determining a number of each valence electron and a total number of valence electrons in the element based on attribute values of at least part of the X attributes corresponding to the element; and ​ ​ ​ ​ ​ ​ ​ ​ 2. The method of claim 1, wherein, ​ ​ ​ ​ ​ 3. The method of claim 1, wherein, ​ ​ ​ ​ ​ ​ 4. The method of claim 3, wherein, ​ ​ ​ ​ 5. The method of claim 3, wherein, ​ ​ For the zth valence electron, based on the proportion information of the at least one element in the target superconductor, the number of the zth valence electron corresponding to each of the at least one element, and the total number of valence electrons corresponding to each of the at least one element, the occupation information of the zth valence electron is determined.

6. The method of claim 3, wherein, The third group of descriptors is obtained based on the proportion information of the at least one element in the target superconductor, the electronegativity corresponding to each of the at least one element, and any one of the following: The maximum value of the electronegativity corresponding to each of the at least one element; or The average value of the electronegativity corresponding to each of the at least one element.

7. The method of claim 3, wherein, The fourth group of descriptors is obtained based on the proportion information of the at least one element in the target superconductor, the Boltzmann constant and the room temperature.

8. The method of claim 1, wherein, The target performance is the critical temperature, and the N1 core descriptors include the following: A descriptor representing the average value of the deviation of different elements in the at least one element on the BCC Fermi level; A descriptor representing the average value of the attribute value of the first ionization energy of the at least one element; A descriptor representing the minimum value of the deviation of different elements in the at least one element on the ground state BCC lattice constant; A descriptor representing the average value of the attribute value of the total number of filled valence electrons of the at least one element; and A descriptor representing the minimum value of the deviation of different elements in the at least one element on the thermal conductivity. The target network includes F levels, and the input of at least part of the K descriptors into the target network for prediction to obtain the performance index corresponding to the target performance includes:

9. The method of claim 1, wherein, Based on the K descriptors, m1 groups of descriptors are generated, each group of descriptors including the N1 core descriptors and N-N1 non-core descriptors; The m1 groups of descriptors are respectively input into the m1 sub-networks of the first level of the target network to obtain m1 intermediate performance indexes of the target performance; The intermediate performance index of the target performance output by the sub-network of the Fth level is taken as the final performance index of the target performance; Following the order of i values ​​from 2 to F, the m values ​​of the (i-1)th layer of the target network are... 1-1 The target performance m obtained from the subnetwork i-1 Each intermediate performance metric is divided into m i The m-th layer of the target network is then input after the group. i A sub-network is used to obtain m of the target performance. i One intermediate performance metric; and 10. A training method of a target network, the target network being used for predicting a target performance of a superconductor, the method comprising: wherein said m1, said F, said m i-1 are integers greater than 1, said m i is an integer greater than or equal to 1. Obtaining experimental values of Y sample superconductors on the target performance, each sample superconductor including at least one element; For each sample superconductor, based on the proportion information of the at least one element in the sample superconductor and the attribute values of X attributes corresponding to each of the at least one element, K descriptors describing the sample superconductor are generated, and N1 core descriptors and K-N1 non-core descriptors are determined in the K descriptors, the influence degree of the core descriptors on the target performance being higher than the influence degree of the non-core descriptors on the target performance; Part of the K descriptors corresponding to each sample superconductor is input into the target network for prediction to obtain a predicted value corresponding to the target performance, wherein: ​ The target network comprises a plurality of sub-networks with the same structure and arranged in a hierarchy, each sub-network comprising N input nodes and 1 output node, the output node of a sub-network in an i-th level serving as an input node of a sub-network in an (i+1)-th level, each sub-network in a first level is inputted with N1 core descriptors and N-N1 non-core descriptors, the N-N1 non-core descriptors being part of the K-N1 non-core descriptors, and the non-core descriptors inputted to different sub-networks are not completely identical; and parameters of the target network are updated with a training objective of minimizing the difference between the predicted value and the experimental value; wherein the X, the K, the N1, the Y, the N are all integers greater than 1, and the i is an integer greater than or equal to 1.

11. The method of claim 10, wherein, The X attributes comprise at least two of the following dimensions: a periodic table dimension; a thermodynamic dimension; a physical dimension; or a crystallographic dimension.

12. The method of claim 10, wherein, The K descriptors comprise at least one of the following groups: a first group of descriptors for describing statistical characteristics of the sample superconductor on the X attributes; a second group of descriptors for describing valence electron occupation states of the sample superconductor; a third group of descriptors for describing ionization of the sample superconductor; and a fourth group of descriptors for describing configurational entropy of the sample superconductor. The first group of descriptors is generated by:

13. The method of claim 12, wherein, for each of the X attributes, based on the proportion information of the at least one element in the sample superconductor and the attribute values of the at least one element on the attribute, respectively, a plurality of statistical dimensions are used to statistically obtain P statistical values of the sample superconductor on the attribute; and based on the P statistical values of the sample superconductor on the X attributes, respectively, a first group of descriptors is generated, the number of the first group of descriptors being equal to X*P. The second group of descriptors comprises occupation information of Q valence electrons in the sample superconductor, wherein the occupation information of the z-th valence electron is obtained by: for each of the at least one element, based on the attribute values of at least part of the X attributes corresponding to the element, the number of each valence electron in the element and the total number of valence electrons in the element are determined; and 14. The method of claim 12, wherein, for the z-th valence electron, based on the proportion information of the at least one element in the sample superconductor, the number of the z-th valence electron corresponding to the at least one element, respectively, and the total number of valence electrons corresponding to the at least one element, respectively, the occupation information of the z-th valence electron is determined. The third group of descriptors is obtained based on the proportion information of the at least one element in the sample superconductor, the electronegativity corresponding to the at least one element, and any one of the following: a maximum value of the electronegativity corresponding to the at least one element; or 15. The method of claim 12, wherein, an average value of the electronegativity corresponding to the at least one element. The fourth group of descriptors is obtained based on the proportion information of the at least one element in the sample superconductor, the Boltzmann constant and the room temperature. ​ 16. The method of claim 12, wherein, ​ 17. The method of claim 10, wherein, The target property is a critical temperature, and the N1 core descriptors include the following: a descriptor representing an average of deviations of different elements in the at least one element on a BCC Fermi level; a descriptor representing an average of attribute values of the at least one element on a first ionization energy; a descriptor representing a minimum of deviations of different elements in the at least one element on a ground state BCC lattice constant; a descriptor representing an average of attribute values of the at least one element on a total valence electron filling number; and a descriptor representing a minimum of deviations of different elements in the at least one element on a thermal conductivity. The target network includes F levels, and the input of part of the K descriptors corresponding to each sample superconductor into the target network for prediction obtains a predicted value corresponding to the target property, including:

18. The method of claim 10, wherein, generating m1 groups of descriptors based on the K descriptors, each group of descriptors including the N1 core descriptors and N-N1 non-core descriptors; inputting the m1 groups of descriptors into m1 sub-networks of the first level of the target network to obtain m1 intermediate predicted values of the target property; taking the intermediate predicted value of the target property output by the sub-network of the Fth level as the final predicted value of the target property; Following the order of i from 2 to F, the m values ​​of the (i-1)th layer of the target network are... 1-1 The target performance m obtained from the subnetwork i-1 The intermediate predicted values ​​are divided into m i After grouping, input the m of the i-th layer of the target network. i A sub-network is used to obtain m of the target performance. i One intermediate predicted value; and a prediction system configured to predict a target property of a target superconductor, including: wherein said m1, said F, said m i-1 are integers greater than 1, said m i is an integer greater than or equal to 1.

19. A prediction system characterized by, at least one storage medium storing at least one instruction set; and at least one processor in communication connection with the at least one storage medium, wherein, when the prediction system is running, the at least one processor reads the at least one instruction set, and executes the method of any one of claims 1-9 according to the indication of the at least one instruction set. a training system configured to train a target network for predicting a target property of a superconductor, including:

20. A training system, characterized by at least one storage medium storing at least one instruction set; and at least one processor in communication connection with the at least one storage medium, wherein, when the training system is running, the at least one processor reads the at least one instruction set, and executes the method of any one of claims 10-18 according to the indication of the at least one instruction set. ​

Images