Inorganic material melting point prediction method and system, and network training method and system

Abstract

The invention discloses an inorganic material melting point prediction method and system and a network training method and system. The method comprises the following steps: acquiring element information of a to-be-tested inorganic material; generating a plurality of descriptors according to the element information of the inorganic material to be detected; analyzing the plurality of descriptors through a genetic algorithm, determining a model input dimension and M main descriptors, and determining the descriptors except the main descriptors as other descriptors; and inputting the main descriptor and other descriptors into a preset inorganic material melting point prediction model based on the model input dimension, and outputting inorganic material melting point prediction data. By combining a convolutional neural network (CNN) and a generative adversarial network (GAN) and utilizing a hierarchical learning strategy, the melting point of the multi-component material is efficiently and accurately predicted.

Description

Technical Field

[0001] The present application relates to the fields of materials science and thermodynamics, and more specifically, to a method and system for predicting the melting point of inorganic materials, and a method and system for training a network. Background Art

[0002] Melting point is a fundamental thermodynamic property of materials, determining their processing and operating conditions. Phase diagrams commonly used in industry and scientific research are composed of numerous melting points. The study of high-temperature resistant materials is inseparable from their melting point. However, methods for determining melting points are very limited. The traditional method involves producing the material and then measuring its melting point. While accurate, this method is inefficient. For some difficult-to-process materials, this method is not only time-consuming but also costly.

[0003] The development of materials science has gone through four major phases: the early empirical paradigm, which relied on experimentation and experience; the theoretical modeling paradigm, which formulated laws using mathematical equations; the computational science paradigm, which used complex simulations based on theoretical models; and the more recent data-driven science paradigm, which uses big data and machine learning to study materials. Each phase has driven the development of materials science, culminating in the modern era of using machine learning to predict new materials.

[0004] In addition, some melting point prediction models based on machine learning have been proposed in recent years, but these models still have shortcomings in accuracy, versatility and reliability. For example: 1. High computational cost: Although the molecular dynamics simulation method based on DFT is accurate, it is computationally expensive and time-consuming, and is not suitable for large-scale data sets.

[0005] 2. Limited prediction accuracy: The prediction accuracy of some deep learning methods still needs to be improved when dealing with complex multi-component systems. For example, the kNN method introduces systematic errors in melting point prediction, and the molecular descriptors fail to capture liquid-state interaction information well.

[0006] 3. Insufficient model generalization ability: The QSPR model has certain risks when predicting the melting points of compounds outside the database, and the generalization ability of the model needs to be improved.

[0007] 4. Data type limitations: Due to data type limitations, the prediction results of some machine learning models are not universal.

[0008] 5. Insufficient descriptors: Some models use a small number of descriptors and cannot fully study the complex relationship between melting point and material, resulting in low model accuracy.

[0009] Therefore, the prior art has defects and is in urgent need of improvement. Summary of the Invention

[0010] In view of the above problems, the purpose of the present invention is to provide a method and system for predicting the melting point of inorganic materials, and a network training method and system, by combining convolutional neural networks (CNN) and generative adversarial networks (GAN) and utilizing a hierarchical learning strategy to efficiently and accurately predict the melting point of multi-component materials.

[0011] Compared with the existing technology, it has the following advantages: 1. Reduce computing costs: By combining convolutional neural networks and generative adversarial networks and utilizing a hierarchical learning strategy, computing costs can be significantly reduced and prediction efficiency can be improved.

[0012] 2. Improve prediction accuracy: By integrating a large number of common descriptors and optimizing the model structure, the accuracy of melting point prediction is improved and systematic errors are reduced.

[0013] 3. Enhance model generalization ability: By studying the impact of different types of network structures and descriptors on prediction results, the generalization ability of the model is enhanced, making its prediction performance more stable in different compounds and multi-component systems.

[0014] 4. Expand data types: Ensure the universality of model prediction results by using multiple types of data.

[0015] 5. Increase the number of descriptors: By using more descriptors, the complex relationship between melting point and material can be fully studied and the accuracy of the model can be improved.

[0016] A first aspect of the present invention provides a method for predicting the melting point of an inorganic material, comprising: Obtain element information of inorganic materials to be tested; Generate multiple descriptors based on the element information of the inorganic material to be tested; the descriptors include configuration entropy, electron orbital occupation state, ionicity index and element property statistical parameters; Analyzing the multiple descriptors by a genetic algorithm to determine a model input dimension and M main descriptors, and determining descriptors other than the main descriptors as other descriptors; The main descriptor and the other descriptors are input into a preset inorganic material melting point prediction model based on the model input dimension, and inorganic material melting point prediction data is output.

[0017] In this solution, the generating of multiple descriptors according to the element information of the inorganic material to be tested includes: The configuration entropy of the inorganic material to be tested is calculated according to the proportion of each element in the inorganic material to be tested: ; Where, ΔS con is the configuration entropy, k B is the Boltzmann constant, T is the Kelvin temperature, c iis the proportion of element i in all elements; Calculate the ratio of s, p, d and f valence electrons to the total valence of s, p, d and f valence electrons in the elements present in the material, and determine the occupancy state of s, p, d and f valence electron orbitals; Determining a plurality of ionicity indices according to the electronegativity of element i in the inorganic material to be tested; The multiple element properties of the inorganic material to be tested are statistically analyzed using multiple preset statistical methods to determine multiple element property statistical parameters.

[0018] In this solution, the genetic algorithm is used to analyze the multiple descriptors to determine the model input dimension and M main descriptors, including: Step 1: Randomly select multiple descriptors to create individuals and build the initial population; Step 2: Evaluate the fitness of each individual in each group using a preset evaluation criterion; the preset evaluation criterion includes the coefficient of determination R 2 , mean square error MSE and root mean square error RMSE; Step 3, select individuals from various populations to create a new population based on their fitness; Step 4: Select individuals from the new population and perform genetic operations to generate new individuals; the genetic operations include selection, mutation, and crossover; Step 5, replace the old population with the new population; Repeat steps 2-5 until the evaluation criteria converge, determine the converged descriptor dimension as the model input dimension, and determine the M descriptors that appear most frequently during the iteration as the main descriptors.

[0019] This plan also includes: The preset inorganic material melting point prediction model is a hierarchical neural network structure composed of multiple sub-networks with the same structure and arranged in a hierarchical manner, each sub-network includes q input nodes and 1 output node, and the output data of the n-th sub-network is the input data of the n+1-th sub-network; The number of input nodes q of each sub-network is determined according to the model input dimension, and M main descriptors and qM other descriptors are input to each sub-network in the first layer respectively. The other descriptors input to each sub-network are determined by random selection.

[0020] A second aspect of the present invention provides a method for training a network for predicting the melting point of an inorganic material, the method comprising: Obtain melting point data of sample materials; performing data cleaning on the melting point data of the sample material to construct a melting point data set of the sample material; constructing a plurality of descriptors based on element information of the sample material in the sample material melting point dataset; Analyzing the multiple descriptors by a genetic algorithm to determine a model input dimension and M main descriptors, and determining descriptors other than the main descriptors as other descriptors; The number of input nodes of each sub-network in the hierarchical neural network is determined based on the model input dimension, the main descriptor and the other descriptors are trained through the hierarchical neural network, and a preset inorganic material melting point prediction model is established.

[0021] In this solution, the step of constructing multiple descriptors based on the elemental information of the sample material in the sample material melting point dataset includes: The configuration entropy of the sample material is calculated according to the proportion of each element in the sample material: ; Where, ΔS con is the configuration entropy, k B is the Boltzmann constant, T is the Kelvin temperature, c i is the proportion of element i in all elements; Calculate the ratio of s, p, d and f valence electrons to the total valence of s, p, d and f valence electrons in the elements present in the material, and determine the occupancy state of s, p, d and f valence electron orbitals; determining a plurality of ionicity indices according to the electronegativity of element i in the sample material; Statistics are performed on various elemental properties of the sample material using a plurality of preset statistical methods to determine a plurality of elemental property statistical parameters.

[0022] In this solution, the genetic algorithm is used to analyze the multiple descriptors to determine the model input dimension and M main descriptors, including: Step 1: Randomly select multiple descriptors to create individuals and build the initial population; Step 2: Evaluate the fitness of each individual in each group using a preset evaluation criterion; the preset evaluation criterion includes the coefficient of determination R 2 , mean square error MSE and root mean square error RMSE; Step 3, select individuals from various populations to create a new population based on their fitness; Step 4: Select individuals from the new population and perform genetic operations to generate new individuals; the genetic operations include selection, mutation, and crossover; Step 5, replace the old population with the new population; Repeat steps 2-5 until the evaluation criteria converge, determine the converged descriptor dimension as the model input dimension, and determine the M descriptors that appear most frequently during the iteration as the main descriptors.

[0023] This plan also includes: The preset inorganic material melting point prediction model is a hierarchical neural network structure composed of multiple sub-networks with the same structure and arranged in a hierarchical manner, each sub-network includes q input nodes and 1 output node, and the output data of the n-th sub-network is the input data of the n+1-th sub-network; The number of input nodes q of each sub-network is determined according to the model input dimension, and M main descriptors and qM other descriptors are input to each sub-network in the first layer respectively. The other descriptors input to each sub-network are determined by random selection.

[0024] This plan also includes: Get the user's voice information; Recognize the user's first physical state information based on the sound information; The first body status information is sent to a terminal.

[0025] A third aspect of the present invention provides an inorganic material melting point prediction system, characterized in that it is configured to predict the melting point of an inorganic material, comprising: at least one storage medium storing at least one instruction set; and at least one processor, in communication 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 any one of the methods for predicting the melting point of an inorganic material according to the instructions of the at least one instruction set.

[0026] A fourth aspect of the present invention provides a training system, characterized in that it is configured to train a prediction network for predicting the melting point of an inorganic material, comprising: at least one storage medium storing at least one instruction set; and at least one processor, in communication 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 any one of the network training methods according to the instructions of the at least one instruction set.

[0027] The present invention discloses a method and system for predicting the melting point of inorganic materials, as well as a network training method and system. The method comprises: obtaining elemental information of the inorganic material to be tested; generating multiple descriptors based on the elemental information of the inorganic material to be tested; analyzing the multiple descriptors using a genetic algorithm to determine the model input dimension and M primary descriptors, and identifying descriptors other than the primary descriptors as other descriptors; inputting the primary descriptors and other descriptors into a preset inorganic material melting point prediction model based on the model input dimension, and outputting inorganic material melting point prediction data. By combining a convolutional neural network (CNN) and a generative adversarial network (GAN), and utilizing a hierarchical learning strategy, the melting point of multi-component materials can be efficiently and accurately predicted. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flow chart showing a method for predicting the melting point of an inorganic material provided by the present invention is shown; Figure 2 A flow chart of a network training method provided by the present invention is shown; Figure 3 A block diagram of a preset inorganic material melting point prediction model provided by the present invention is shown; Figure 4 A schematic diagram showing the elemental information of inorganic materials provided by the present invention; Figure 5 A schematic diagram of the genetic algorithm screening process provided by the present invention is shown; Figure 6 A schematic diagram showing a comparison of prediction effects of different models provided by the present invention is shown; Figure 7 A schematic diagram showing the comparison of ternary phase diagrams of the prediction effects of different models provided by the present invention. DETAILED DESCRIPTION

[0029] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0030] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0031] Figure 1 The flowchart of the method for predicting the melting point of an inorganic material provided by the present invention is shown.

[0032] like Figure 1 As shown, the present invention discloses a method for predicting the melting point of an inorganic material, comprising: S102, obtaining element information of the inorganic material to be tested; S104, generating a plurality of descriptors based on the element information of the inorganic material to be tested; the descriptors include configurational entropy, electron orbital occupation state, ionicity index and element property statistical parameters; S106, analyzing the multiple descriptors by a genetic algorithm to determine the model input dimension and M main descriptors, and determining the descriptors other than the main descriptors as other descriptors; S108 , inputting the main descriptor and other descriptors into a preset inorganic material melting point prediction model based on the model input dimension, and outputting inorganic material melting point prediction data.

[0033] According to an embodiment of the present invention, a preset inorganic material melting point prediction model (hierarchical neural network statistical ensemble integrated model HNN, the model structure is as follows Figure 3 As shown in the figure, the melting point of the inorganic material to be tested is predicted. First, the 53 element information of the inorganic material to be tested is obtained through experimental data and other means. The types of element information are as follows: Figure 3 As shown. 909 descriptors are determined using the system's preset descriptor calculation method, including 1 configurational entropy, 4 electron orbital occupancy states, 3 ionicity indices, and 901 elemental property statistical parameters. The obtained descriptors are analyzed using a genetic algorithm, and the converged descriptor dimension is determined as the optimal model input dimension. The M descriptors with the highest frequency of occurrence during the genetic iteration process are determined as the main descriptors, and descriptors other than the main descriptors are determined as other descriptors. The number of input nodes of the subnetwork in the first layer of the preset inorganic material melting point prediction model is determined based on the model input dimension. All the main descriptors are input into each subnetwork, and some other descriptors are randomly selected based on the number of input nodes of the subnetwork and input into each subnetwork. The other descriptors input into each subnetwork are not exactly the same. The main descriptors and other descriptors are analyzed using the preset inorganic material melting point prediction model, and combined with the melting point data of sample materials in the database (obtained through various channels such as the Internet and laboratories), the inorganic material melting point prediction data for the inorganic material to be tested is output.

[0034] According to an embodiment of the present invention, multiple descriptors are generated based on the element information of the inorganic material to be tested, including: The configuration entropy of the inorganic material to be tested is calculated according to the proportion of each element in the inorganic material to be tested: ; Where, ΔS con is the configuration entropy, k B is the Boltzmann constant, T is the Kelvin temperature, c i is the proportion of element i in all elements; Calculate the ratio of s, p, d and f valence electrons to the total valence of s, p, d and f valence electrons in the elements present in the material, and determine the occupancy state of s, p, d and f valence electron orbitals; Determine multiple ionicity indices based on the electronegativity of element i in the inorganic material to be tested; The various element properties of the inorganic material to be tested are statistically analyzed using a variety of preset statistical methods to determine a plurality of element property statistical parameters.

[0035] It should be noted that the element types contained in the inorganic material to be tested are obtained, represented by element i, and the proportion c of element i in all elements is calculated. i The proportion of element i among all elements is input into the system's preset configurational entropy calculation formula to determine the configurational entropy of the inorganic material being tested. Configurational entropy represents the spatial arrangement of atoms in a material. Including configurational entropy in descriptors can characterize key properties of the inorganic material being tested.

[0036] Count the number of s, p, d and f valence electrons in each element respectively, and calculate the s, p, d and f valence electron orbital occupancy states using the system's preset electron orbital occupancy state calculation formula. The calculation formula for the electron orbital occupancy state is: ; Among them, F x represents the electronic structure property corresponding to the xth valence electron, c i is the proportion of the i-th element in all elements, E x It represents the number of x-th valence electrons (s, p, d or f valence electrons) in the i-th element in the sample to be tested, E n represents the sum of the number of s, p, d, and f valence electrons of the i-th element.

[0037] The properties of ionic compounds are calculated using the formula: ; Where I is the property of ionic compounds, e is the natural constant, c i is the proportion of the i-th element in all elements, f i is the electronegativity corresponding to the i-th element, is the maximum value or average value of the electronegativity of each element. When is the maximum value among the electronegativity corresponding to each element, determine the first ionic compound property; when When I is the average value of the electronegativity corresponding to each element, the second ionic compound property is determined. At the same time, the third ionic compound property is determined by the second ionic compound property. When I>1.7, the third ionic compound property is 1; when I≤1.7, the third ionic compound property is 0.

[0038] like Figure 4 The figure shows 53 elemental properties of inorganic materials. These 53 elemental properties are statistically analyzed using various preset statistical methods, resulting in 901 elemental property statistical parameters. These 17 statistical methods include calculating the minimum, maximum, and range of elemental properties; the (weighted) minimum, maximum, and range of weighted sorting; and the (weighted) minimum, maximum, range, and mean of absolute percentages of elemental properties.

[0039] According to an embodiment of the present invention, a genetic algorithm is used to analyze multiple descriptors to determine the model input dimension and M main descriptors, including: Step 1: Randomly select multiple descriptors to create individuals and build the initial population; Step 2: Evaluate the fitness of each individual in each group using the preset evaluation criteria; the preset evaluation criteria include the coefficient of determination R 2 , mean square error MSE and root mean square error RMSE; Step 3, select individuals from various populations to create a new population based on their fitness; Step 4: Select individuals from the new population and perform genetic operations to generate new individuals; genetic operations include selection, mutation, and crossover; Step 5, replace the old population with the new population; Repeat steps 2-5 until the evaluation criteria converge, determine the converged descriptor dimension as the model input dimension, and determine the M descriptors that appear most frequently during the iteration as the main descriptors.

[0040] It's important to note that due to the scarcity of data in materials science, training all 909 descriptors using a single neural network model is impractical. In a neural network with only two hidden layers and 909 input descriptors, over a billion hyperparameters would need to be tuned. This has been a long-standing challenge for neural network-based AI models to solve materials science problems. Given the limited size of datasets, reducing the dimensionality of the input descriptors for a single neural network model is crucial. Genetic algorithms (GAs) are commonly used to reduce the dimensionality of neural network descriptors (feature extraction).

[0041] The results of genetic algorithm screening are as follows Figure 5As shown. The genetic algorithm (GA) first randomly selects a system-preset number of descriptors to form an initial group. These groups then undergo a crossover and mutation process with the goal of obtaining higher training and test scores and reducing the number of descriptors. GA encodes the descriptors into binary form "1" or "0", indicating whether the descriptor is selected or not. Subsequently, the fitness function of each model is calculated and the population is developed based on the new set of descriptors. Subsequent iterations of the descriptor selection process produce a new set of descriptors (group) with a reduced number of descriptors. After iterative GA selection, the descriptors reach a converged descriptor dimension (optimal dimension) in a single neural network input in a given dataset. In addition, the frequency of occurrence of descriptors during the GA iteration may vary, such as Figure 5 Figure b shows the five most frequently occurring descriptors. The top five most frequently occurring descriptors (the initial value of M is 5, which can be adjusted by those skilled in the art based on practical needs) are defined as "primary descriptors." This classification simplifies the subsequent HNN algorithm architecture design. Many sub-models based on different descriptors reached the upper limit of their test scores. Compared to traditional descriptor selection methods, this method classifies descriptors into primary and other descriptors based on their frequency of occurrence in genetic statistics.

[0042] Figure 5 The progress of the genetic algorithm iteration process is shown in a, where the x-axis represents the number of iterations, the left y-axis represents the descriptor dimension (d), and the right y-axis represents the test set. R 2 When d=50, we observe a turning point in the accuracy of the base learner model. The importance of the descriptor is reflected in Figure 5 In b, a histogram of descriptor frequencies is shown.

[0043] According to an embodiment of the present invention, the further embodiment includes: The preset inorganic material melting point prediction model is a hierarchical neural network structure composed of multiple sub-networks with the same structure and arranged in a hierarchical manner. Each sub-network includes q input nodes and 1 output node. The output data of the n-th sub-network is the input data of the n+1-th sub-network. The number of input nodes q of each sub-network is determined according to the model input dimension, and M main descriptors and qM other descriptors are input to each sub-network in the first layer respectively. The other descriptors input to each sub-network are determined by random selection.

[0044] It should be noted that other descriptors for each sub-network can be input using , where 909-M represents other descriptors, i.e., descriptors other than the primary descriptor; and qM represents the number of other descriptors that can be input into the subnetwork in addition to the primary descriptor. qM other descriptors are randomly (and unorderedly) selected from the 909-M other descriptors and input into the subnetwork along with the M primary descriptors. The other descriptors input into each subnetwork are randomly selected, so they are not exactly the same in each subnetwork.

[0045] In a hierarchical neural network, the number of subnetworks included in each layer of the network gradually decreases. That is, the number of subnetworks included in the current layer is less than that of the previous layer, and the highest layer contains only one subnetwork. The output data of each subnetwork in the current layer serves as the input data for each subnetwork in the next layer. The network structure of each subnetwork is the same, and those skilled in the art can modify the network structure of any subnetwork in any layer according to actual needs.

[0046] Figure 2 The flowchart of the network training method provided by the present invention is shown.

[0047] like Figure 2 As shown, the second aspect of the present invention provides a method for training a network for predicting the melting point of an inorganic material, the method comprising: S202, obtaining melting point data of sample material; S204, performing data cleaning on the melting point data of the sample material to construct a melting point data set of the sample material; S206, constructing a plurality of descriptors according to the element information of the sample material in the sample material melting point data set; S208, analyzing the multiple descriptors by a genetic algorithm to determine the model input dimension and M main descriptors, and determining the descriptors other than the main descriptors as other descriptors; S210, determining the number of input nodes of each sub-network in the hierarchical neural network based on the model input dimension, training the main descriptors and other descriptors through the hierarchical neural network, and establishing a preset inorganic material melting point prediction model.

[0048] It should be noted that, first, about 15,000 sample material melting point data were obtained through various channels such as the Internet and laboratories, and these sample material melting point data were cleaned by methods such as deleting duplicate values, supplementing missing values, and processing outliers. Finally, more than 8,000 high-quality sample material melting point data were obtained, and a huge, comprehensive and clean data database was established.

[0049] Secondly, the success of neural networks in materials science depends largely on the quality of experimental data, the choice of material descriptors, and the artificial intelligence algorithms used. However, existing neural network algorithms perform poorly when processing small material data sets and high-dimensional material descriptors. The main reason is that it is difficult to effectively train a large number of neural network hyperparameters with a small amount of data. To solve this problem, this paper proposes a novel hierarchical neural network statistical ensemble integration model (HNN), which is a preset inorganic material melting point prediction model. The model structure is as follows: Figure 3 shown.

[0050] The model first constructs multiple descriptors based on the elemental information of the sample material and analyzes the obtained descriptors using a genetic algorithm to determine the optimal model input dimension and corresponding primary descriptors. It then integrates millions of neural network sub-models (sub-networks, each with the same input nodes, determined by the optimal model input dimension) into a "neural network statistical ensemble" with neural network learning elements. This method preserves the unique correlation between the descriptor and the target attribute in each sub-model at the optimal dimension of a given dataset and can be trained in parallel using a limited dataset, resolving the conflict between small datasets and high-dimensional descriptors.

[0051] Because the "hierarchical neural network statistical ensemble" algorithm successfully resolves the significant mismatch between the dimensionality of "descriptors" and the resulting number of hyperparameters and the volume of specialized data, it can significantly increase the number of "descriptors," thereby improving predictive accuracy and capabilities. Materials are composed of elements. To develop universal predictive capabilities, it is necessary to establish a correlation between material properties and information (properties) about the constituent elements. The physical and chemical properties of individual elements are related not only to the quantum properties of the element but also to the crystal structure formed by the element. Given a specific crystal structure, it is necessary to extract intrinsic physical / chemical properties that are independent of impurities. Furthermore, the kinetic and thermodynamic parameters of the transformation from the amorphous phase to the crystalline phase of the element are also relevant. These parameters form the basic descriptors of the element.

[0052] Finally, an artificial neural network (ANN) and a convolutional neural network (CNN) were trained separately. The training results were compared, and then a similar adversarial algorithm was used to combine the outputs of the ANN and CNN, dynamically adjusting the loss function to ultimately obtain a fused HNN prediction model.

[0053] During the implementation process, a neural network training model program was written in Python. The neural network model architecture was constructed using the nn (neural network) module in the open-source artificial intelligence library Pytorch, and data computation and processing were performed using the numpy and pandas libraries. The neural network structure in the program was custom-written, using the nn.Module and Sequential functions inherited from nn to define the neural network structure. All parameters (number of layers, number of nodes, and activation function) can be modified. For data input, the train_test_split tool in the sklearn library in Python was used to split the three datasets into three groups. A random seed was set, and 90% of the data was randomly selected as the training set, which served as reference data for model training. The remaining 10% was used as the validation set, which was used to test the predictive power of the entire model. The pandas library was used to read the partitioned descriptor and material property files, and the numpy library was used to normalize the data.

[0054] In order to verify the effects of different descriptors and algorithms in material melting point prediction, several sets of experiments were conducted, and the results are shown as follows: Figure 6 shown. Figure 6 Figure a shows the prediction performance of an artificial neural network (ANN) model using a traditional 208 descriptors. Although this model can predict the melting point of the material to a certain extent, its accuracy and stability are low, with a low R² value and a large error. Figure 6 Figure b shows the effect of using the expanded ANN model with 909 descriptors. Compared to the traditional ANN model with 208 descriptors, the 909-descriptor ANN model significantly improves prediction accuracy and stability, with an increased R² value and a reduced error. This indicates that increasing the number of descriptors helps improve the model's predictive performance. Figure 6 Figure c shows the results of a convolutional neural network (CNN) model using 909 descriptors. Compared to the ANN model, the CNN model excels in processing high-dimensional data and capturing complex features, further improving the accuracy and stability of predictions, with a higher R² value and a lower error. Figure 6 Figure d shows the performance of a generative adversarial network (GAN) model that fuses CNN and ANN. This model combines the advantages of CNN and ANN and further optimizes its performance through adversarial training. Experimental results show that the GAN model performs best in predicting material melting points, achieving the highest R² value and the lowest error, significantly outperforming all the previous models.

[0055] In summary, experimental results show that the accuracy and stability of material melting point predictions gradually improve with the increase in the number of descriptors and the complexity of the model. In particular, the generative adversarial network model that integrates CNN and ANN achieves the best prediction performance, demonstrating its great potential and application prospects in the field of materials science.

[0056] In addition, if Figure 7 As shown in the figure, the ternary phase diagrams of four different prediction models were also predicted, and the results were highly consistent with the scatter plots obtained experimentally. Figure 7 The ternary phase diagram of the artificial neural network (ANN) model using the traditional 208 descriptors is shown in a. Figure 7 b shows the ternary phase diagram of the ANN model using the expanded 909 descriptors. Figure 7 The ternary phase diagram of the convolutional neural network (CNN) model using 909 descriptors is shown in c and Figure 7 Figure d shows a ternary phase diagram from a generative adversarial network (GAN) model that combines CNN and ANN. This demonstrates that training the new algorithm on a subset of material melting point data allows for comprehensive melting point contour maps across the entire compositional range. This generative content (GC) approach reduces the workload of manual repetitive experimentation while providing valuable and timely feedback for further exploration of multi-element materials. This highlights the effectiveness of systematic, realistic, and validated datasets in generating reasonably accurate predictions, even with relatively small datasets.

[0057] The inorganic material melting point prediction model presented in this invention has the following key technical features: 1. Large and clean dataset: A database containing more than 8,000 high-quality material melting point data has been established, providing a solid data foundation for model training.

[0058] 2. Hierarchical Neural Network Statistical Ensemble Integration Model (HNN): Genetic algorithms reduce element descriptors, determine the optimal input dimensions and main descriptors through genetic algorithms, and perform "neural network statistical ensemble integration" on millions of neural network sub-models, resolving the contradiction between small data sets and high-dimensional descriptors.

[0059] 3. Rich descriptor library: A library of 53 basic descriptors based on element performance was established, and calculated through 17 statistical methods, a total of 909 descriptor dimensions were generated, significantly exceeding the number of descriptors in existing technologies.

[0060] 4. Training method for integrating ANN and CNN: Use artificial neural network (ANN) and convolutional neural network (CNN) for training respectively, compare their training results, and then use a similar adversarial algorithm to combine the outputs of ANN and CNN, and dynamically adjust them through the loss function to finally obtain a fused HNN prediction model.

[0061] 5. Efficient implementation method: Use Python to write neural network training model programs, and use open source libraries such as Pytorch, numpy, and pandas for data calculation and processing to ensure the efficiency and accuracy of the model.

[0062] A third aspect of the present invention provides an inorganic material melting point prediction system, characterized in that it is configured to predict the melting point of an inorganic material, comprising: at least one storage medium storing at least one instruction set; and at least one processor, in communication with at least one storage medium, When the prediction system is running, at least one processor reads at least one instruction set and executes any one of the methods for predicting the melting point of an inorganic material according to the instructions of the at least one instruction set.

[0063] A fourth aspect of the present invention provides a training system, characterized in that it is configured to train a prediction network for predicting the melting point of an inorganic material, comprising: at least one storage medium storing at least one instruction set; and at least one processor, in communication with at least one storage medium, When the training system is running, at least one processor reads at least one instruction set and executes any one of the network training methods according to the instructions of the at least one instruction set.

[0064] The information involved in this application (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals (including but not limited to signals transmitted between user terminals and other devices, etc.) are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions. For example, the "element information of the inorganic material to be tested" and "melting point data of the sample material" involved in this disclosure are all obtained with full authorization.

[0065] The present invention discloses a method and system for predicting the melting point of inorganic materials, as well as a network training method and system. The method comprises: obtaining elemental information of the inorganic material to be tested; generating multiple descriptors based on the elemental information of the inorganic material to be tested; analyzing the multiple descriptors using a genetic algorithm to determine the model input dimension and M primary descriptors, and identifying descriptors other than the primary descriptors as other descriptors; inputting the primary descriptors and other descriptors into a preset inorganic material melting point prediction model based on the model input dimension, and outputting inorganic material melting point prediction data. By combining a convolutional neural network (CNN) and a generative adversarial network (GAN), and utilizing a hierarchical learning strategy, the melting point of multi-component materials can be efficiently and accurately predicted.

[0066] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0067] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0068] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0069] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0070] Alternatively, if the integrated units described above are implemented as software modules and sold or used as standalone products, they can also be stored on a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product, stored on a storage medium, includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute all or part of the methods described in various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as removable storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A method for predicting the melting point of an inorganic material, characterized in that: include: Obtain element information of inorganic materials to be tested; Generate multiple descriptors based on the element information of the inorganic material to be tested; the descriptors include configuration entropy, electron orbital occupation state, ionicity index and element property statistical parameters; Analyzing the multiple descriptors by a genetic algorithm to determine a model input dimension and M main descriptors, and determining descriptors other than the main descriptors as other descriptors; The main descriptor and the other descriptors are input into a preset inorganic material melting point prediction model based on the model input dimension, and inorganic material melting point prediction data is output.

2. The method for predicting the melting point of an inorganic material according to claim 1, wherein The generating of a plurality of descriptors according to the element information of the inorganic material to be tested comprises: The configuration entropy of the inorganic material to be tested is calculated according to the proportion of each element in the inorganic material to be tested: ； Where, ΔS con is the configuration entropy, k B is the Boltzmann constant, T is the Kelvin temperature, c i is the proportion of element i in all elements; Calculate the ratio of s, p, d and f valence electrons to the total valence of s, p, d and f valence electrons in the elements present in the material, and determine the occupancy state of s, p, d and f valence electron orbitals; Determining multiple ionicity indices according to the electronegativity of element i in the inorganic material to be tested; The multiple element properties of the inorganic material to be tested are statistically analyzed using multiple preset statistical methods to determine multiple element property statistical parameters.

3. The method for predicting the melting point of an inorganic material according to claim 1, wherein The analyzing the multiple descriptors by a genetic algorithm to determine the model input dimension and M main descriptors includes: Step 1: Randomly select multiple descriptors to create individuals and build the initial population; Step 2: Evaluate the fitness of each individual in each group using a preset evaluation criterion; the preset evaluation criterion includes the coefficient of determination R 2 , mean square error MSE and root mean square error RMSE; Step 3, select individuals from various populations to create a new population based on their fitness; Step 4: Select individuals from the new population and perform genetic operations to generate new individuals; the genetic operations include selection, mutation, and crossover; Step 5, replace the old population with the new population; Repeat steps 2-5 until the evaluation criteria converge, determine the converged descriptor dimension as the model input dimension, and determine the M descriptors that appear most frequently during the iteration as the main descriptors.

4. The method for predicting the melting point of an inorganic material according to claim 1, wherein Also includes: The preset inorganic material melting point prediction model is a hierarchical neural network structure composed of multiple sub-networks with the same structure and arranged in a hierarchical manner, each sub-network includes q input nodes and 1 output node, and the output data of the n-th sub-network is the input data of the n+1-th sub-network; The number of input nodes q of each sub-network is determined according to the model input dimension, and M main descriptors and qM other descriptors are input to each sub-network in the first layer respectively. The other descriptors input to each sub-network are determined by random selection.

5. A method for training a network for predicting the melting point of an inorganic material, the method comprising: Obtain melting point data of sample materials; performing data cleaning on the melting point data of the sample material to construct a melting point data set of the sample material; constructing a plurality of descriptors based on element information of the sample material in the sample material melting point dataset; Analyzing the multiple descriptors by a genetic algorithm to determine a model input dimension and M main descriptors, and determining descriptors other than the main descriptors as other descriptors; The number of input nodes of each sub-network in the hierarchical neural network is determined based on the model input dimension, the main descriptor and the other descriptors are trained through the hierarchical neural network, and a preset inorganic material melting point prediction model is established.

6. The training method according to claim 5, characterized in that The constructing of a plurality of descriptors according to the element information of the sample material in the sample material melting point dataset comprises: The configuration entropy of the sample material is calculated according to the proportion of each element in the sample material: ； Where, ΔS con is the configuration entropy, k B is the Boltzmann constant, T is the Kelvin temperature, c i is the proportion of element i in all elements; Calculate the ratio of s, p, d and f valence electrons to the total valence of s, p, d and f valence electrons in the elements present in the material, and determine the occupancy state of s, p, d and f valence electron orbitals; determining a plurality of ionicity indices according to the electronegativity of element i in the sample material; Statistics are performed on various elemental properties of the sample material using a plurality of preset statistical methods to determine a plurality of elemental property statistical parameters.

7. The training method according to claim 5, characterized in that The analyzing the multiple descriptors by a genetic algorithm to determine the model input dimension and M main descriptors includes: Step 1: Randomly select multiple descriptors to create individuals and build the initial population; Step 2: Evaluate the fitness of each individual in each group using a preset evaluation criterion; the preset evaluation criterion includes the coefficient of determination R 2 , mean square error MSE and root mean square error RMSE; Step 3, select individuals from various populations to create a new population based on their fitness; Step 4: Select individuals from the new population and perform genetic operations to generate new individuals; the genetic operations include selection, mutation, and crossover; Step 5, replace the old population with the new population; Repeat steps 2-5 until the evaluation criteria converge, determine the converged descriptor dimension as the model input dimension, and determine the M descriptors that appear most frequently during the iteration as the main descriptors.

8. The training method according to claim 5, characterized in that Also includes: The preset inorganic material melting point prediction model is a hierarchical neural network structure composed of multiple sub-networks with the same structure and arranged in a hierarchical manner, each sub-network includes q input nodes and 1 output node, and the output data of the n-th sub-network is the input data of the n+1-th sub-network; The number of input nodes q of each sub-network is determined according to the model input dimension, and M main descriptors and qM other descriptors are input to each sub-network in the first layer respectively. The other descriptors input to each sub-network are determined by random selection.

9. A system for predicting the melting point of inorganic materials, characterized in that: Configured to predict melting points of inorganic materials, including: at least one storage medium storing at least one instruction set; and at least one processor, in communication 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 according to any one of claims 1 to 4 according to the instructions of the at least one instruction set.

10. A training system, characterized in that: Configured to train a prediction network for predicting the melting points of inorganic materials, including: at least one storage medium storing at least one instruction set; and at least one processor, in communication 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 according to any one of claims 5-8 according to the instructions of the at least one instruction set.

Images