A method for accelerating the construction of a thermodynamic database of multi-component materials, storage medium

Abstract

The present application relates to a kind of methods for accelerating the construction of multi-component material thermodynamic database, storage medium, comprising the following steps: all end intercompound of system is divided into three subsets: marked sample pool, unmarked sample pool and test set, wherein the data of marked sample pool and test set form can be obtained by first principle calculation, unmarked sample pool is composed of end intercompound without calculation;Based on marked sample pool, a decision model is constructed, and the prediction accuracy is observed on the test set, and if the prediction accuracy reaches the predetermined standard, iteration is stopped;All unmarked samples are predicted using ensemble learning model to form energy and uncertainty distribution, and the results are used to establish the thermodynamic database of system.Compared with prior art, the present application can realize the automation process of material thermodynamic database construction by giving several initial compound formation energy samples, and the operation is simple and fast.

Description

Technical Field

[0001] This invention relates to the field of computational thermodynamics of metallic materials, and in particular to an active learning framework that couples first-principles calculations to accelerate the construction of a multi-component material thermodynamic database. Background Technology

[0002] Thermodynamic data for alloys and compounds, such as the enthalpy of formation (also known as the standard heat of formation), plays a crucial role in a variety of applications, including phase diagram calculations and materials design, and in exploring new high-melting-point materials for use in advanced coal-fired power plants, heat exchangers, filters, and turbines. In the CALculation of PHAse Diagrams (CALPHAD) method, which combines phase-based thermodynamic modeling with equilibrium calculations, the enthalpy of formation is often presented as a component of the Gibbs free energy, such as the Gibbs free energy of end-group compounds. Detailed quantification and analysis of the Gibbs free energy for each phase contributes to an intuitive understanding of the formation and phase transition processes of different phases. This understanding is a key pillar of effective and efficient materials design strategies. Density functional theory (DFT) calculations are typically used to accurately calculate the formation energy of end-group compounds, but this inevitably leads to enormous computational costs, as the number of end-group compounds can reach 10^5 for a practical alloy containing approximately 10 different elements. This presents a unique opportunity to apply machine learning (ML) in practice.

[0003] Current research using machine learning for compound energy prediction focuses primarily on non-magnetic compounds, such as the study on sigma phase formation by Jean-Claude Crivello from the Sorbonne University in Paris. These studies, using first-principles calculations on alloy compounds, do not consider the contribution of magnetism, significantly impacting the reliability of model predictions. Secondly, works like the one by Logan Ward's team at Northwestern University predicting the formation energy of inorganic compounds lack self-training and self-improvement processes in their machine learning models; they are often applied after only one training iteration, which clearly fails to meet the demands of the big data era for machine learning models. Our work considers the contribution of magnetism to the formation energy of alloys and allows for continuous iteration of the machine learning model as the database grows, improving model accuracy and generalization ability. Summary of the Invention

[0004] The purpose of this invention is to overcome the blindness of first-principles calculations in building thermodynamic databases, and to provide an active learning framework coupled with first-principles calculations to accelerate the construction of multi-component material thermodynamic databases. By providing several initial compound formation energy samples, the construction of material thermodynamic databases can be automated, and the operation is simple and fast.

[0005] The objective of this invention can be achieved through the following technical solutions: The first aspect of this invention provides a method for accelerating the construction of a multi-component material thermodynamic database, comprising the following steps: S1: Before the iteration begins, all terminal compounds in the system are divided into three subsets: labeled sample pool, unlabeled sample pool, and test set. The data formation of the labeled sample pool and test set can be obtained through first-principles calculations, while the unlabeled sample pool consists of terminal compounds that have not been calculated. S2: Build a decision model based on the labeled sample pool and observe the prediction accuracy on the test set. If the prediction accuracy reaches the predetermined standard, stop the iteration; otherwise, select unlabeled samples for computation and optimization based on the enhancement committee strategy. S3: Use an ensemble learning model to predict the formation energy and uncertainty distribution of all unlabeled samples, and use the results to build a thermodynamic database for the system.

[0006] Furthermore, S1 specifically includes: S1-1: Assuming that the formation energies of N terminal compounds need to be calculated for the multicomponent A system, the formation energies of L and T terminal compounds are first calculated using VASP software. Here, the L sample set is the initial labeled sample set, the T sample set is the test set, and the remaining N-(L+T) samples are the unlabeled sample set U. S1-2: Perform data preprocessing on N samples to construct 11 macro-scale features; S1-3: Perform data preprocessing on N samples to construct 110 microscale features.

[0007] Furthermore, in S1-1, each terminal compound is calculated using the following formula: Where n represents the number of elements in each end-member group, and k represents the different element properties. and These correspond to the composition of the i-th element and the specified element properties, respectively.

[0008] Furthermore, in S1-3, the microscale features are divided into two parts: The first part is e p features, where e represents the number of non-equivalent sites in the Wyckoff position of the compound, and p is the 11 elemental properties from Table 1, using one-hot encoding of the Wyckoff position of the compound; The second part is e p features describe the local environment of each equivalent site. The Voronoi technique is used to calculate the energy contribution of the surrounding atoms to the central atom by calculating the nearest neighbor Voronoi polyhedra of each atom in the unit cell. The calculation is obtained by the following formula: in This represents the property of the i-th central atom. This represents the property of the nth nearest neighbor atom of the central atom i. This represents the area of ​​the Voronoi polyhedron corresponding to the nth nearest neighbor atom of the central atom i.

[0009] Furthermore, S2 specifically includes: S2-1: The validated support vector machine algorithm is used as the decision model to model on the L dataset; the support vector machine regression algorithm uses the sciki-learn library based on the Python language, and uses grid search and 10-fold cross-validation to select the best model parameters and model D; S2-2: Calculate the mean absolute error (MAE) of decision model D on the test set T, as shown in the following formula: in It is the predicted value of decision model D for the i-th test set sample. Let represent the true value of the i-th test set sample, and n represent the total number of test set samples; If the MAE metric is less than the threshold, the iteration stops; otherwise, based on the enhancement committee strategy, unlabeled samples are selected for computational optimization.

[0010] Furthermore, the step of selecting unlabeled samples for computational optimization based on the enhanced committee strategy includes: A subset of terminal compounds are selected from the unlabeled sample pool as candidates. First-principles calculations are then performed on these candidates. After the calculations are completed, they are added to the labeled sample pool. S2 is repeated until the decision model achieves the expected accuracy on the test set, at which point the iteration stops.

[0011] Furthermore, the optimization of selecting unlabeled samples based on the enhanced committee strategy specifically includes: A-1: Based on the labeled sample set L, five independent training sets are obtained using bootstrap sampling, which are used for training the subsequent five committee models; A-2: Using the support vector machine algorithm as the base model, five committee models are trained on five independent training sets, and the best model parameters and model are selected using grid search and 10-fold cross-validation. A-3: Using the Enhanced-Query-By-Committee (EQBC) strategy, compute the uncertainty index on the unlabeled sample set U; A-4: Select several candidates with the largest uncertainty index from the unlabeled sample pool, then perform first-principles calculations on these candidates, and add them to the labeled sample pool L after the calculations are completed.

[0012] Furthermore, the formula for calculating the uncertainty index is as follows: Where m represents the m-th sample in the unlabeled sample set U. This represents the uncertainty of the sample, and P represents the number of committee models. This represents the predicted value of the sample on the p-th committee model. Let represent the mean predicted by the m-th sample on P committee models.

[0013] Furthermore, S3 specifically includes: S3-1: Using the p committee models trained in the previous iteration, output the distribution of the mean absolute percentage error (MAPE) estimate for the remaining unlabeled samples, as shown in the following formula: in This represents the i-th actual value. Let represent the predicted value of the i-th data point, and n represent all data points. S3-2: Combine the decision model D and the p committee models trained in the previous iteration, for a total of (1+p) models, to form an ensemble learning model. Use a stacked ensemble approach and grid search and 10-fold cross-validation to select the hyperparameters of the best ensemble learning model M. The prediction process of model M is as follows: Where j represents the number of models integrated. Let represent the predicted value of the j-th model for sample x. Representing meta-models to combine features The input is the final prediction result; S3-3: Utilize the ensemble learning model M to output the formation energy of all unlabeled samples and provide the uncertainty distribution map of the samples for the establishment of the system's thermodynamic database.

[0014] A second aspect of the present invention provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the method described above for accelerating the construction of a multi-component material thermodynamic database.

[0015] Compared with the prior art, the present invention has the following technical advantages: 1. The method of this invention overcomes the blindness of traditional first-principles calculations, saves resources and time, and couples an active learning framework with first-principles calculations. Given an initial sample set, it can automatically establish a material thermodynamics database.

[0016] 2. The method of the present invention does not involve experiments or use chemicals throughout the entire process, does not generate chemical pollution, and conforms to the concept of green environmental protection; moreover, it is simple to operate, low in cost, easy to implement, and suitable for widespread application.

[0017] 3. The method of this invention can predict the formation energy of materials in advance through the calculation results of machine learning models, select samples with low uncertainty for thermodynamic modeling, save modeling time and resources, improve modeling efficiency, play a guiding role, and avoid blindness.

[0018] 4. A multi-scale feature set that takes into account both macroscopic and microscopic scales is presented, which can more intuitively show the contribution of features at different scales to the compound formation energy.

[0019] 5. The method of this invention considers the contribution of magnetism to energy in the machine learning modeling process, and calculates the magnetic moments of some pure components through first-principles calculations, which ensures the rationality of the modeling and increases the accuracy of the model.

[0020] 6. This invention provides uncertainty estimates for all uncalculated samples, increasing the interpretability of the machine learning model. Attached Figure Description

[0021] Figure 1 The attached diagram illustrates the active learning framework flowchart of this method, which couples first-principles calculations.

[0022] Figure 2 It has a sigma phase crystal structure.

[0023] Figure 3 Voronoi polyhedron for each atom.

[0024] Figure 4 This is for the fifth iteration of MAPEE on unlabeled samples. Detailed Implementation

[0025] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Component models, material names, connection structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0026] In specific implementation, the technical solution adopted by this invention is: an active learning framework coupled with first-principles calculations, and a method for accelerating the construction of a multi-component material thermodynamic database, see [link to relevant documentation]. Figure 1 This includes the following steps: Step 1: Before the iteration begins, all terminal compounds in the system are divided into three subsets: labeled sample pool, unlabeled sample pool, and test set. The data of the labeled sample pool and test set can be calculated using first-principles calculations. The unlabeled sample pool consists of terminal compounds that have not been calculated. Step 2: Build a decision model based on the labeled sample pool and observe the prediction accuracy on the test set. If the expected accuracy is achieved, stop the iteration, skip step 3, and proceed to step 4. If the expected accuracy is not achieved, proceed to step 3. Step 3: Using an active learning framework based on an enhanced committee strategy, select a subset of candidates (i.e., terminal compounds) from the unlabeled sample pool, then perform first-principles calculations on these candidates. After the calculations are complete, add them to the labeled sample pool. Repeat Step 2 until the decision model achieves the expected accuracy on the test set, and then stop iterating. Step 4: Use the ensemble learning model to output the formation energy and uncertainty distribution of all unlabeled samples for the establishment of the system's thermodynamic database.

[0027] Step 1 includes: Step 1.1: Assuming the multicomponent system A requires calculation of the formation energies of N terminal compounds, first use VASP software to calculate the formation energies of L and T terminal compounds. The L sample set is called the initial labeled sample set, and the T sample set is called the test set. The remaining N-(L+T) samples are called the unlabeled sample set U.

[0028] Step 1.2 involves preprocessing the N samples to construct 11 macroscopic-scale features, the elemental properties of which are provided in Table 1. Each terminal compound is calculated using the following formula: Where n represents the number of elements in each end-group member, and k represents the different element properties. and These correspond to the composition of the i-th element and the specified element property, respectively; Step 1.3 involves preprocessing N samples to construct 110 microscale features, the elemental properties of which are provided in Table 1. The microscale features are divided into two parts: the first part contains e... p features, where e represents the number of inequivalent sites in the Wyckoff position of the compound, and p are 11 elemental properties from Table 1, using one-hot encoding of the Wyckoff position of the compound; the second part is also e p features describe the local environment of each equivalent site. The Voronoi technique is used to calculate the energy contribution of the surrounding atoms to the central atom by calculating the nearest-neighbor Voronoi polyhedra around each atom in the unit cell. This is calculated using the following formula: in This represents the property of the i-th central atom. This represents the property of the nth nearest neighbor atom of the central atom i. This represents the area of ​​the Voronoi polyhedron corresponding to the nth nearest neighbor atom of the central atom i.

[0029] Step 2 includes: Step 2.1: Use machine learning algorithms. Here, we recommend the validated Support Vector Machine (SVM) algorithm as the decision model, and model it on the L dataset. The SVM regression algorithm uses the sciki-learn library based on the Python language, and uses grid search and 10-fold cross-validation to select the optimal model parameters and model D. Step 2.2, calculate the mean absolute error (MAE) of decision model D on the test set T, as shown in the following formula: in It is the predicted value of decision model D for the i-th test set sample. Let represent the true value of the i-th test set sample, and n represent the total number of test set samples.

[0030] If the MAE index is less than the threshold of 500 J / mol If it is atom, then skip step 3 and proceed to step 4.

[0031] Step 3 includes: Step 3.1: Based on the labeled sample set L, use bootstrap sampling to obtain 5 independent training sets for the training of the subsequent 5 committee models; Step 3.2: Using the support vector machine algorithm as the base model, train 5 committee models on 5 independent training sets, and use grid search and 10-fold cross-validation to select the best model parameters and model; Step 3.3, using the Enhanced-Query-By-Committee (EQBC) strategy [1], calculate the uncertainty index on the unlabeled sample set U. The index calculation formula is as follows: Where m represents the m-th sample in the unlabeled sample set U. This represents the uncertainty of the sample, and P represents the number of committee models, which is 5 in this case. This represents the predicted value of the sample on the p-th committee model. Let represent the mean predicted by the m-th sample on P committee models.

[0032] Step 3.4: Select several candidates (i.e., terminal compounds) with the largest uncertainty index from the unlabeled sample pool, then perform first-principles calculations on these candidates, and add them to the labeled sample pool L after the calculations are completed.

[0033] Step 4 includes: Step 4.1: Using the p committee models trained in the previous iteration, output the distribution of the mean absolute percentage error (MAPE) estimate for the remaining unlabeled samples, as shown in the following formula: in This represents the i-th actual value. Let represent the predicted value of the i-th data point, and n represent all data points.

[0034] Step 4.2: Combine the decision model D and the p committee models trained in the previous iteration, forming a total of (1+p) models to form an ensemble learning model. Use a stacked ensemble approach and employ grid search and 10-fold cross-validation to select the hyperparameters of the optimal ensemble learning model M. The prediction process of model M is shown below: Where j represents the number of integrated models, and assuming the number of committee members p is 5, then there are a total of 6. Let represent the predicted value of the j-th model for sample x. Representing meta-models to combine features The input is the final prediction result.

[0035] Step 4.3: Utilize the ensemble learning model M to output the formation energy of all unlabeled samples and provide the uncertainty distribution diagram of the samples for the establishment of the system's thermodynamic database.

[0036] Example 1 The steps of the active learning framework coupled with first-principles calculations in this embodiment to accelerate the construction of a multi-element material thermodynamic database are as follows: Step 1: Taking the sigma phase in the Cr-Fe-Co-Ni quaternary system as an example, the formation energy of 1020 (excluding pure component energies) sigma phase end compounds needs to be calculated using first-principles calculations. Before the iteration begins, all sigma phase end compounds are divided into three subsets: labeled sample pool, unlabeled sample pool, and test set. Step 1.1: The formation energies of 1020 end compounds in the Cr-Fe-Co-Ni system need to be calculated. First, the formation energies of 180 and 168 end compounds are calculated using VASP software. The initial labeled sample set L and the test set T contain a total of 348 samples, and the remaining 672 samples are called the unlabeled sample set U.

[0037] Step 1.2 involves preprocessing the 1020 samples to construct 11 macroscopic-scale features, the elemental properties of which are provided in Table 1. Each terminal compound is calculated using the following formula: Where n represents the number of elements in each inter-group member, and k represents the different attribute types. and These correspond to the components and specified attributes of the i-th element, respectively. Table 1. Element properties used for calculating multi-scale feature sets From reference [2] †Integrating the values ​​of the density of states that are less than those of the Fermi level. ‡ Data sourced from website [http: / / phases.imet-db.ru / elements / main.aspx] Step 1.3 involves preprocessing the 1020 samples to construct 110 microscale features, the elemental properties of which are provided in Table 1. The microscale features are divided into two parts: the first part consists of 55 features, which are based on atomic microstructure and local atomic environment features. Figure 2 This section represents the microstructure of the Sigma phase. One-hot encoding of the Wyckoff positions of the Sigma phase is used; the second part contains 55 features describing the local atomic environment. Voronoi techniques are used to calculate the nearest-neighbor Voronoi polyhedra around each atom in the unit cell to weight the energy contribution of surrounding atoms to the central atom. For example... Figure 3 As shown. It is calculated using the following formula: in This represents the property of the i-th central atom. This represents the property of the nth nearest neighbor atom of the central atom i. This represents the area of ​​the Voronoi polyhedron corresponding to the nth nearest neighbor atom of the central atom i.

[0038] Step 2: Set the prediction accuracy threshold for the formation energy to 500 J / mol. Atom builds a decision model based on the labeled sample pool and observes the prediction accuracy on the test set. If the expected accuracy is achieved, the iteration stops, skipping step 3 and proceeding to step 4. If the expected accuracy is not achieved, step 3 is performed. Step 2.1: Use the support vector machine regression algorithm from the scikit-learn library to model on the labeled sample dataset of this iteration, and use grid search and 10-fold cross-validation to select the best model parameters and model D; Step 2.2, calculate the mean absolute error (MAE) of decision model D on the test set T, as shown in the following formula: in It is the predicted value of decision model D for the i-th test set sample. Let represent the true value of the i-th test set sample, and n represent the total number of test set samples.

[0039] If the MAE index is less than the threshold of 500 J / mol If it is atom, then skip step 3 and proceed to step 4.

[0040] Step 3: Select a subset of candidates (i.e., terminal compounds) from the unlabeled sample pool using an enhanced committee strategy. Then perform first-principles calculations on these candidates. After the calculations are complete, add them to the labeled sample pool. Repeat Step 2 until the decision model achieves the expected accuracy on the test set, and then stop iterating. Step 3.1: Based on the labeled sample set L, use bootstrap sampling to obtain 5 independent datasets for training the subsequent 5 committee members; Step 3.2: Using the support vector machine algorithm as the base model, train 5 committee members on 5 independent datasets, and use grid search and 10-fold cross-validation to select the best model parameters and model; Step 3.3, using the Enhanced-Query-By-Committee (EQBC) strategy [1], calculate the uncertainty index on the unlabeled sample set U. The index calculation formula is as follows: Where m represents the m-th sample in the unlabeled sample set U. This represents the uncertainty of the sample, and P represents the number of committee models, which is 5 in this case. This represents the predicted value of the sample on the p-th committee model. Let represent the mean predicted by the m-th sample on P committee models.

[0041] Step 3.4: Select 50 candidates (i.e., terminal compounds) with the largest uncertainty in the unlabeled sample pool. Then perform first-principles calculations on these candidates and add them to the labeled sample pool L after the calculations are completed.

[0042] Step 4: Use the ensemble learning model to output the formation energy and uncertainty distribution of all unlabeled samples for the establishment of the system's thermodynamic database.

[0043] Step 4.1: Using the five committee models trained in the previous iteration, output the distribution of the mean absolute percentage error estimate for the remaining unlabeled samples. The MAPE calculation formula is as follows: in This represents the i-th actual value. Let represent the predicted value of the i-th data point, and n represent all data points.

[0044] Figure 4 The MAPE-estimation index is shown after the fifth iteration, which estimates the uncertainty distribution of unlabeled samples.

[0045] Step 4.2: Combine the decision model D and the 5 committee models trained in the previous iteration, for a total of 6 models, into an ensemble learning model. Use a stacked ensemble approach and employ grid search and 10-fold cross-validation to select the parameters of the optimal ensemble learning model M. The prediction process of model M is shown below: Where j represents the number of models integrated, here p=5, for a total of 6. Let represent the predicted value of the j-th model for sample x. Representing meta-models to combine features The input is the final prediction result.

[0046] Step 4.3: Use the ensemble learning model M to output the formation energy of all unlabeled samples for the establishment of the system's thermodynamic database.

[0047] The methods described above overcome the blindness of high-throughput first-principles calculations, saving resources and time. Based on an active learning-coupled first-principles framework, only a small number of initial compound formation energies are needed to obtain the formation energies of all end compounds in the complete alloy system, making it convenient and fast. This method, through an enhanced committee query strategy, minimizes the computational resources required for first-principles calculations. Each iteration selects several of the most valuable candidate samples for first-principles calculations, saving significant computation time and resources, improving computational efficiency, providing guidance, and avoiding blindness.

[0048] In summary, the above embodiments, based on an active learning framework coupled with first-principles calculations, construct a multi-component material thermodynamic database. Before the iteration begins, all end-cell compounds in the system are divided into three subsets: a labeled sample pool, an unlabeled sample pool, and a test set. The formation energies of the labeled sample pool and test set data are calculated using first-principles calculations, while the unlabeled sample pool consists of end-cell compounds for which first-principles calculations have not been performed. Then, a decision model is constructed based on the labeled sample pool, and the prediction accuracy is observed on the test set. If the expected accuracy is achieved, the iteration stops, and the model can directly output the formation energies of all unlabeled samples for subsequent thermodynamic database construction. If the expected accuracy is not achieved, an active learning algorithm based on an enhanced committee strategy is used to select some candidates (i.e., end-cell compounds) from the unlabeled sample pool. First-principles calculations are then performed on these candidates, and they are added to the labeled sample pool after calculation. This process continues until the decision model achieves the expected accuracy on the test set, at which point the iteration stops. At this point, the decision model is combined with the committee model generated in the previous iteration, and an ensemble learning model is used to output the formation energies of all unlabeled samples for the construction of the system's thermodynamic database.

[0049] This embodiment also proposes a computer-readable storage medium storing computer instructions for instructing a computer to execute the aforementioned active learning framework coupled with first-principles calculations, thereby accelerating the construction of a multi-component material thermodynamic database. The storage medium can be an electronic medium, magnetic medium, optical medium, electromagnetic medium, infrared medium, or a semiconductor system or propagation medium. The storage medium can also include semiconductor or solid-state memory, magnetic tape, removable computer disk, random access memory (RAM), read-only memory (ROM), hard disk, and optical disc. Optical discs can include optical disc-read-only memory (CD-ROM), optical disc-read-write (CD-RW), and DVD.

[0050] References [1] D. Wu, VJ Lawhern, S. Gordon, BJ Lance, and C.-T. Lin, “Offline EEG-based driver drowsiness estimation using enhanced batch-modeactive learning (EBMAL) for regression,” in 2016 IEEE International Conference on Systems, Man, and Cybernetics (SMC) , Budapest, Hungary: IEEE, Oct. 2016, pp. 000730-000736. doi: 10.1109 / SMC.2016.7844328. [2]Y. Zhang, JRG Evans, and S. Yang, “Corrected Values ​​for Boiling Points and Enthalpies of Vaporization of Elements in Handbooks,” J. Chem. Eng. Data , vol. 56, no. 2, Art. no. 2, Feb. 2011, doi: 10.1021 / je1011086. The embodiments of the method of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made according to the purpose of the invention. Any parameter changes or calculation simplifications made based on the principle of the technical solution of the present invention, as long as they meet the purpose of the invention and do not deviate from the principle and concept of the method for constructing a multi-component material thermodynamic database using an active learning framework coupled with first-principles calculation, are all within the protection scope of the present invention.

Claims

1. A method for accelerating the construction of a multi-component material thermodynamic database, characterized in that, Includes the following steps: S1: Before the iteration begins, all terminal compounds in the system are divided into three subsets: labeled sample pool, unlabeled sample pool, and test set. The data formation of the labeled sample pool and test set can be obtained through first-principles calculations, while the unlabeled sample pool consists of terminal compounds that have not been calculated. S2: Build a decision model based on the labeled sample pool and observe the prediction accuracy on the test set. If the prediction accuracy reaches the predetermined standard, stop the iteration; otherwise, select unlabeled samples for computation and optimization based on the enhancement committee strategy. S3: Use an ensemble learning model to predict the formation energy and uncertainty distribution of all unlabeled samples, and use the results to build a thermodynamic database for the system; The optimization based on the enhanced committee strategy specifically includes: A-1: Based on the labeled sample set L, five independent training sets are obtained using bootstrap sampling, which are used for training the subsequent five committee models; A-2: Using the support vector machine algorithm as the base model, five committee models are trained on five independent training sets, and the best model parameters and model are selected using grid search and 10-fold cross-validation. A-3: Using the enhanced committee strategy EQBC, calculate the uncertainty index on the unlabeled sample set U; A-4: Select several candidates with the largest uncertainty index from the unlabeled sample pool, then perform first-principles calculations on these candidates, and add them to the labeled sample pool L after the calculations are completed; S3 specifically includes: S3-1: Using the p committee models trained in the previous iteration, output the distribution of the mean absolute percentage error (MAPE) estimate for the remaining unlabeled samples, as shown in the following formula: in This represents the i-th actual value. Let represent the predicted value of the i-th data point, and n represent all data points. S3-2: Combine the decision model D and the p committee models trained in the previous iteration, for a total of 1+p models, to form an ensemble learning model. Use a stacked ensemble approach and employ grid search and 10-fold cross-validation to select the hyperparameters of the optimal ensemble learning model M. The prediction process of model M is shown below: Where j represents the number of models integrated. Let represent the predicted value of the j-th model for sample x. Representing meta-models to combine features The input is the final prediction result; S3-3: Utilize the ensemble learning model M to output the formation energy of all unlabeled samples and provide the uncertainty distribution map of the samples for the establishment of the system's thermodynamic database.

2. The method for accelerating the construction of a multi-component material thermodynamic database according to claim 1, characterized in that, S1 specifically includes: S1-1: Assuming that the formation energies of N terminal compounds need to be calculated for the multicomponent A system, the formation energies of L and T terminal compounds are first calculated using VASP software. Here, the L sample set is the initial labeled sample set, the T sample set is the test set, and the remaining N-(L+T) samples are the unlabeled sample set U. S1-2: Perform data preprocessing on N samples to construct 11 macro-scale features; S1-3: Perform data preprocessing on N samples to construct 110 microscale features.

3. The method for accelerating the construction of a multi-component material thermodynamic database according to claim 2, characterized in that, In S1-1, each terminal compound is calculated using the following formula: Where n represents the number of elements in each end-member group, and k represents the different element properties. and These correspond to the composition of the i-th element and the specified element properties, respectively.

4. The method for accelerating the construction of a multi-component material thermodynamic database according to claim 2, characterized in that, In S1-3, the microscale features are divided into two parts: The first part is e p features, where e represents the number of non-equivalent sites in the Wyckoff position of the compound, and p is one of 11 elemental properties, using one-hot encoding of the Wyckoff position of the compound; The second part is e p features describe the local environment of each equivalent site. The Voronoi technique is used to calculate the energy contribution of the surrounding atoms to the central atom by calculating the nearest neighbor Voronoi polyhedra of each atom in the unit cell. The calculation is obtained by the following formula: in This represents the property of the i-th central atom. This represents the property of the nth nearest neighbor atom of the central atom i. This represents the area of ​​the Voronoi polyhedron corresponding to the nth nearest neighbor atom of the central atom i.

5. The method for accelerating the construction of a multi-component material thermodynamic database according to claim 1, characterized in that, S2 specifically includes: S2-1: The validated support vector machine algorithm is used as the decision model to model on the L dataset; the support vector machine regression algorithm uses the sciki-learn library based on the Python language, and uses grid search and 10-fold cross-validation to select the best model parameters and model D; S2-2: Calculate the mean absolute error (MAE) of decision model D on the test set T, as shown in the following formula: in It is the predicted value of decision model D for the i-th test set sample. Let represent the true value of the i-th test set sample, and n represent the total number of test set samples; If the MAE metric is less than the threshold, the iteration stops; otherwise, based on the enhancement committee strategy, unlabeled samples are selected for computational optimization.

6. The method for accelerating the construction of a multi-component material thermodynamic database according to claim 1, characterized in that, The formula for calculating the uncertainty index is as follows: Where m represents the m-th sample in the unlabeled sample set U. This represents the uncertainty of the sample, and P represents the number of committee models. This represents the predicted value of the sample on the p-th committee model. Let represent the mean predicted by the m-th sample on P committee models.

7. A storage medium containing computer-executable instructions, characterized in that, The storage medium of the computer-executable instructions, when executed by a computer processor, is used to perform the method for accelerating the construction of a multi-component material thermodynamic database as described in any one of claims 1 to 6.

Images