A material structure segmentation encoding method applicable to machine learning

Abstract

The present invention provides a method for segmenting and encoding material structures applicable to machine learning, including: analyzing the key atomic structures of the material structure, extracting the skeleton units of the material structure, and constructing a skeleton model; dividing the skeleton units into several fragment units, performing unique encoding based on the skeleton units and the fragment units to obtain skeleton encoding information and fragment encoding information; generating an encoding matrix of the material structure according to the skeleton encoding information and the fragment encoding information; training a machine learning model using the encoding matrix to obtain a target machine learning model, and the target machine learning model can be used to predict material properties. The present invention realizes multi-dimensional material information processing through a two-level segmentation of (one or more) skeletons-(one or more) fragments and data encoding, significantly improves the accuracy of material property prediction and the generality of material structure encoding, accelerates the material design process, and is applicable to material science research and the efficient discovery of new materials.

Description

Technical Field

[0001] The present invention relates to the technical fields of materials science and machine learning, and particularly relates to a method for segmenting and encoding the structure of materials for machine learning. Background Art

[0002] The purpose of materials design is to achieve specific functions or meet specific requirements by systematically adjusting the structure, composition, and properties of materials. Traditional materials design relies on experiments and theoretical simulations, which is a time-consuming and laborious process. With the rise of the application of machine learning in materials science, researchers establish a relationship model between material structures and properties by using a large amount of experimental data and material databases, combined with machine learning algorithms, thereby accelerating the process of material discovery and design. The first step in establishing such a model is to reasonably segment and encode the structure of materials in order to extract meaningful structure-property relationships.

[0003] In terms of material segmentation and encoding, existing methods mainly rely on manual processing or rule-based algorithms, with low efficiency and lack of generality. In addition, the encoding is usually described based on specific parameters such as lattice parameters and atomic coordinates, but the high dependence on specific parameter descriptions makes the model have problems such as poor adaptability and insufficient information expression.

[0004] Therefore, there is an urgent need for a method for segmenting and encoding the structure of materials that can achieve multi-dimensional material information processing and improve the generality and characterization accuracy of material structure encoding. Summary of the Invention

[0005] Based on this, it is necessary to provide a method for segmenting and encoding the structure of materials for machine learning in view of the above technical problems.

[0006] A method for segmenting and encoding the structure of materials for machine learning includes the following steps: analyzing the key atomic structure of the material structure, extracting the skeleton unit of the material structure, and constructing a skeleton model; dividing the skeleton unit into several fragment units, and performing unique encoding according to the skeleton unit and the fragment units to obtain skeleton encoding information and fragment encoding information; generating an encoding matrix of the material structure according to the skeleton encoding information and the fragment encoding information; training a machine learning model with the encoding matrix to obtain a target machine learning model, and the target machine learning model can predict material properties.

[0007] In one embodiment, the analyzing the key atomic structure of the material structure, extracting the skeleton unit of the material structure, and constructing a skeleton model includes: measuring and analyzing the atomic spatial arrangement in the material structure, and finding the key atomic structure forming the skeleton; forming a skeleton unit based on the key atomic structure, and constructing a skeleton model of the material according to the skeleton unit.

[0008] In one embodiment, dividing the skeleton unit into a plurality of fragment units, and performing unique encoding according to the skeleton unit and the fragment units to obtain skeleton encoding information and fragment encoding information includes: analyzing the internal structure of the skeleton unit, and dividing the key atomic structures in the skeleton unit into a plurality of fragment units; respectively performing unique encoding on the skeleton unit and the fragment units to obtain skeleton encoding information and fragment encoding information, where the skeleton encoding information and the fragment encoding information respectively correspond to different skeleton types and fragment types.

[0009] In one embodiment, generating a coding matrix of the material structure according to the skeleton encoding information and the fragment encoding information includes: using the skeleton encoding information as row elements and the fragment encoding information as column elements to generate a coding matrix representing the material structure, where the coding matrix has the same dimension as the material structure.

[0010] In one embodiment, training a machine learning model using the coding matrix to obtain a target machine learning model includes: using the coding matrix as input to train the machine learning model; optimizing the model parameters through error backpropagation until the error of the model reaches a set threshold, and validating the model through test set data. After passing the validation, a target machine learning model is obtained. The target machine learning model is used for predicting material properties, and the material properties include physical properties and chemical properties.

[0011] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows: By analyzing the key atomic structures of the material structure, extracting the skeleton unit of the material structure, constructing a skeleton model, dividing the skeleton unit into a plurality of fragment units, performing unique encoding according to the skeleton unit and the fragment units to obtain skeleton encoding information and fragment encoding information, and generating a coding matrix of the structure according to the skeleton encoding information and the fragment encoding information. Through the two-level segmentation and data encoding method of (one or more) skeletons - (one or more) fragments, it can more efficiently represent complex material structures, is applicable to various material design scenarios, is convenient for integration with machine learning models, trains a machine learning model using the coding matrix to obtain a target machine learning model, can predict material properties through the target machine learning model, can process multi-dimensional material information through multi-level skeleton and fragment encoding, simplifies the preprocessing process of model input data, significantly improves the accuracy of material structure representation and material design efficiency, and improves the generality of material structure encoding, being applicable to various material design and optimization scenarios. Description of the Drawings

[0012] Figure 1 It is a schematic flowchart of a method for segmenting and encoding a material structure for machine learning in one embodiment;

[0013] Figure 2 Schematic flow diagram of the skeleton-fragment two-level segmentation mode in one embodiment;

[0014] Figure 3 Schematic diagram of two types of atoms and the key skeleton in the graphene-like structure in one embodiment;

[0015] Figure 4 Schematic diagram of the fragment combination in the triangular unit skeleton in one embodiment; Detailed implementation manners

[0016] Before describing the detailed implementation manners of the present invention, the overall concept of the present invention is described as follows:

[0017] The present invention mainly focuses on the research and development of the material segmentation and coding process. Currently, it mainly relies on manual processing or rule-based algorithms, which are inefficient and lack generality.

[0018] Therefore, the present invention proposes a material structure segmentation and coding method for machine learning. By analyzing the key atomic structure of the material structure, extracting the skeleton unit of the material structure, constructing a skeleton model, dividing the skeleton unit into several fragment units, performing unique coding according to the skeleton unit and the fragment unit to obtain the skeleton coding information and the fragment coding information, generating the coding matrix of the structure according to the skeleton coding information and the fragment coding information. Through the two-level segmentation of skeleton-fragment and dataized coding, it can more efficiently represent complex material structures, is applicable to various material design scenarios, is convenient for integration with machine learning models, uses the coding matrix to train the machine learning model to obtain the target machine learning model, can predict material properties through the target machine learning model, can process multi-dimensional material information through multi-level skeleton and fragment coding, simplifies the preprocessing process of the model input data, significantly improves the accuracy of material structure representation and material design efficiency, improves the generality of material structure coding, and is applicable to various material design and optimization scenarios.

[0019] After introducing the overall concept of the present invention, in order to make the purpose, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below through specific implementation manners in conjunction with the drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0020] In one embodiment, as Figure 1 shown, a material structure segmentation and coding method for machine learning is provided, including the following steps:

[0021] Step S110, analyze the key atomic structure of the material structure, extract the skeleton unit of the material structure, and construct a skeleton model.

[0022] Specifically, in order to achieve efficient characterization of the material structure, it is necessary to analyze the key atomic structure of the material structure, extract the framework units that form the material skeleton, determine the key atomic structure that forms the skeleton by calculating spatial parameters such as atomic spacing and bond angle, and define it as the framework unit. A framework model is constructed through the framework unit for subsequent framework encoding.

[0023] Among them, step S110 includes: measuring and analyzing the atomic spatial arrangement in the material structure to find the key atomic structure that forms the skeleton; forming a framework unit based on the key atomic structure and constructing a framework model of the material according to the framework unit.

[0024] Specifically, by measuring and analyzing the atomic spatial arrangement in the material lattice, the key atomic structure that forms the skeleton is found. The key atomic structure is the main support structure of the material. The framework unit is extracted according to the key atomic structure, and the framework model of the entire material structure is constructed. In actual use, an initial material model can be generated using software (such as crystal analysis software VESTA or CrystalMaker), and the framework model can be extracted through parameter setting.

[0025] Step S120: Divide the framework unit into several fragment units, and perform unique encoding according to the framework unit and the fragment units to obtain framework encoding information and fragment encoding information.

[0026] Specifically, as Figure 2 shown, divide the framework unit into several fragment units. Each fragment unit represents a unique sub-structure or component distribution. Perform unique encoding on the framework unit and the fragment units respectively to obtain framework encoding information and fragment encoding information, forming a data-based encoding structure.

[0027] Among them, step S120 includes: analyzing the internal structure of the framework unit, dividing the key atomic structure within the framework unit into several fragment units; performing unique encoding on the framework unit and the fragment units respectively to obtain framework encoding information and fragment encoding information. The framework encoding information and the fragment encoding information correspond to different framework types and fragment types respectively.

[0028] Specifically, analyze the internal structure of the framework unit, divide the key atomic structure within the framework unit into several fragment units, perform unique encoding on the framework unit and the fragment units respectively. Different framework types and fragment types are represented by the framework encoding information and the fragment encoding information respectively. For example, the fragment encoding value ranges from 0 to N to represent different fragment types, and the encoded data is stored in a matrix. Through the above encoding method, data input is facilitated and the model training efficiency is improved.

[0029] As Figure 3As shown, there are two types of carbon atoms and the key skeleton in graphene obtained by analyzing the key atomic structure of the material structure; further analyzing the skeleton unit can obtain the fragment combination in the triangular unit skeleton as shown in Figure 4 shown.

[0030] Step S130: Generate a coding matrix of the material structure according to the skeleton coding information and the fragment coding information.

[0031] Specifically, generate a complete coding matrix of the material structure according to the skeleton coding information and the fragment coding information. Each element in the coding matrix represents a certain unit in the material structure. Through the two-level segmentation and data coding method of (one or more) skeletons-(one or more) fragments, the complex material structure can be characterized more efficiently, which is convenient for integration with the machine learning model.

[0032] Among them, step S130 includes: taking the skeleton coding information as the row elements and the fragment coding information as the column elements to generate a coding matrix representing the material structure, and the coding matrix has the same dimension as the material structure.

[0033] Specifically, take the skeleton coding information as the row elements and the fragment coding information as the column elements to generate a two-dimensional coding matrix representing the material structure. Here is an example. For example, set the size of the coding matrix to (2,4), and the element values are between 0 and 6, representing various coding unit types. This coding matrix can be directly input into the machine learning model for predicting the material properties.

[0034] Step S140: Train the machine learning model with the coding matrix to obtain the target machine learning model.

[0035] Specifically, use the coding matrix as the input of the machine learning model for model training to obtain the target machine learning model. The target machine learning model can predict the material properties. By inputting the encoded data into this target machine learning model, the physical or chemical properties of the material can be predicted. Through the multi-level skeleton and fragment coding method, the complex material structure data is digitalized, which is convenient for the machine learning model to process, and significantly improves the accuracy of material structure characterization and the material design efficiency.

[0036] Among them, step S140 includes: using the coding matrix as the input to train the machine learning model; optimizing the model parameters through error backpropagation until the error of the model reaches the set threshold, validating the model through the test set data, and obtaining the target machine learning model through successful validation. The target machine learning model is used for predicting the material properties, and the material properties include physical properties and chemical properties.

[0037] Specifically, select any suitable machine learning algorithm, such as a neural network model, input the encoding matrix into the machine learning model for model training. Adjust the model parameters through multiple rounds of training, and optimize the model parameters through error backpropagation until the model error reaches the set threshold, thereby obtaining a model that meets the prediction accuracy. After the model training is completed, input the test set data for model verification, evaluate the model performance through the test set error. After passing the verification, obtain the target machine learning model. Through this target machine learning model, material property prediction can be performed, including but not limited to physical properties and chemical properties, significantly improving the accuracy of material structure characterization and material design efficiency.

[0038] In this embodiment, by analyzing the key atomic structure of the material structure, the skeleton unit of the material structure is extracted, and a skeleton model is constructed. The skeleton unit is divided into several fragment units, and unique encoding is performed according to the skeleton unit and the fragment unit to obtain the skeleton encoding information and the fragment encoding information. According to the skeleton encoding information and the fragment encoding information, an encoding matrix of the structure is generated. Through the two-level segmentation of the skeleton-fragment and data encoding, complex material structures can be characterized more efficiently, applicable to various material design scenarios, facilitating integration with machine learning models. Use the encoding matrix to train the machine learning model to obtain the target machine learning model. Through the target machine learning model, material property prediction can be performed. Through multi-level skeleton and fragment encoding, multi-dimensional material information processing can be performed, simplifying the preprocessing process of model input data, significantly improving the accuracy of material structure characterization and material design efficiency, and improving the generality of material structure encoding, applicable to various material design and optimization scenarios.

[0039] Those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0040] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. Optionally, they can be implemented by program codes executable by the computing device. Thus, they can be stored in a computer storage medium (ROM / RAM, magnetic disk, optical disk) and executed by the computing device. And in some cases, the steps shown or described can be executed in a different order from here, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. Therefore, the present invention is not limited to any specific combination of hardware and software.

[0041] The above content is a further detailed description of the present invention in combination with specific implementation manners. It cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can still be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A material structure segmentation encoding method that can be used for machine learning, characterized in that: The following steps are involved: Analyze the key atomic structure of the material structure, extract the skeleton units of the material structure, and construct a skeleton model; The skeleton unit is divided into a plurality of fragment units, and uniquely encoded according to the skeleton unit and the fragment unit to obtain skeleton encoding information and fragment encoding information, including: analyzing the internal structure of the skeleton unit, dividing the key atomic structure in the skeleton unit into a plurality of fragment units; uniquely encoding the skeleton unit and the fragment unit respectively to obtain skeleton encoding information and fragment encoding information, wherein the skeleton encoding information and the fragment encoding information correspond to different skeleton types and fragment types respectively; Generate a coding matrix of the material structure according to the skeleton coding information and the fragment coding information, including: using the skeleton coding information as a row element and the fragment coding information as a column element to generate a coding matrix representing the material structure, wherein the coding matrix has the same dimension as the material structure; The coding matrix is ​​used to train the machine learning model to obtain a target machine learning model, and the target machine learning model can predict material properties.

2. A material structure segmentation coding method that can be used for machine learning according to claim 1, characterized in that: The key atomic structure of the material structure is analyzed, the skeleton unit of the material structure is extracted, and the skeleton model is constructed, including: Measure and analyze the spatial arrangement of atoms in the material structure to find the key atomic structure that forms the skeleton; A skeleton unit is formed based on the key atomic structure, and a skeleton model of the material is constructed according to the skeleton unit.

3. The material structure segmentation coding method that can be used for machine learning according to claim 1 is characterized in that: The method of training the machine learning model using the encoding matrix to obtain a target machine learning model includes: Using the encoding matrix as input, training a machine learning model; The model parameters are optimized through error back propagation until the model error reaches the set threshold. The model is verified through test set data. The target machine learning model is obtained through verification. The target machine learning model is used to predict material properties, including physical properties and chemical properties.

Images