A method, apparatus and device for predicting a topological center point

Abstract

The application provides a topological center point prediction method, device and equipment, and belongs to the technical field of crystal growth. The application aims to provide a rapid analysis method for crystal growth simulation. The method comprises the following steps: obtaining first flow field data corresponding to different working condition parameter combinations by simulating crystal growth environments under different working condition parameter combinations; for each working condition parameter combination, converting the first flow field data corresponding to the working condition parameter combination into second flow field data under a preset structured grid, and determining the topological center point coordinates corresponding to the working condition according to the second flow field data; constructing a preset model, training the preset model based on multiple working condition parameter combinations and the topological center point coordinates corresponding to each working condition parameter combination, and obtaining a topological center point prediction model; and inputting a target working condition parameter combination into the topological center point prediction model to obtain the topological center point coordinates corresponding to the target working condition parameter combination.

Description

Technical Field

[0001] This invention relates to the field of data prediction technology, and in particular to a method, apparatus and equipment for predicting topological center points. Background Technology

[0002] Currently, when sampling data under different operating conditions, there are differences in data distribution. For example, when collecting data for crystal growth, the center of gravity of data collection is different under different operating conditions. This makes it difficult to analyze data under different operating conditions in a unified manner, making data analysis quite challenging. Summary of the Invention

[0003] Based on the background technology, this invention proposes a method, apparatus, and device for predicting topological center points.

[0004] In a first aspect, the present invention provides a method for predicting topological center points, comprising: The crystal growth environment under different combinations of operating parameters was simulated to obtain the first flow field data corresponding to different combinations of operating parameters. The first flow field data includes the position information of multiple discrete data points and the flow field parameters of each discrete data point, including temperature parameters and flow velocity parameters. For each combination of operating parameters, the first flow field data corresponding to the combination of operating parameters is converted into second flow field data under a preset structured grid, and the topology center point coordinates corresponding to the combination of operating parameters are determined based on the second flow field data. The topology center point coordinates represent the coordinates of the center point of the gas vortex during the crystal growth process. A preset model is constructed, and the preset model is trained based on multiple combinations of operating condition parameters and the coordinates of the topology center point corresponding to each combination of operating condition parameters to obtain a topology center point prediction model. The target operating condition parameter combination is input into the topology center point prediction model to obtain the topology center point coordinates corresponding to the target operating condition parameter combination.

[0005] Optionally, converting the first flow field data corresponding to the combination of operating parameters into second flow field data under a preset structured grid includes: Linear interpolation is performed on multiple discrete data points included in the first flow field data using an interpolation function to generate a grid dataset. The grid dataset includes the coordinates of each discrete data point in the preset structured grid and the flow field parameters corresponding to each coordinate. Based on the grid dataset, the preset structured grid is extended at the boundary to obtain the second flow field data.

[0006] Optionally, the step of extending the boundary of the preset structured grid based on the grid dataset includes: Obtain the boundary data of the grid dataset; The boundaries of the preset structured grid are expanded, and the flow field parameters corresponding to the expanded points are determined based on the boundary data. Based on the flow field parameters corresponding to multiple expansion points and the grid dataset, second flow field data is generated.

[0007] Optionally, determining the topology center point coordinates corresponding to the combination of operating parameters based on the second flow field data includes: Based on the second flow field data, the vector field index of each coordinate point in the preset structured grid is determined, and the vector field index represents the gas flow state of multiple coordinate points adjacent to the coordinate point. The coordinates of the topological center point are determined from a plurality of coordinate points based on the vector field index.

[0008] Optionally, determining the vector field index of each coordinate point in the preset structured grid based on the second flow field data includes: For each coordinate point, the velocity vectors corresponding to multiple adjacent coordinate points are obtained from the second flow field data. The angle difference between two adjacent velocity vectors is determined, and the sum of multiple angle differences is obtained to obtain the vector field index.

[0009] Optionally, determining the coordinates of the topological center point from a plurality of coordinate points based on the vector field index includes: Multiple target coordinate points that satisfy preset conditions are determined from the multiple coordinate points, the preset conditions including the vector field index being located within a first target range and the flow velocity parameter being located within a second target range; Based on the distance between multiple target coordinate points, the multiple target coordinate points are clustered to obtain multiple clusters; Determine the centroid coordinates of each cluster, and use the centroid coordinates as the coordinates of the topological center point.

[0010] Optionally, the step of training the preset model based on multiple combinations of operating condition parameters and the coordinates of the topology center point corresponding to each combination of operating condition parameters to obtain a topology center point prediction model includes: According to a preset ratio, the various combinations of operating condition parameters and the coordinates of the topology center point corresponding to each combination of operating condition parameters are divided into a training set and a test set. Under different model parameter configurations, the preset model is trained using the training set to obtain multiple candidate models; The test set is used to evaluate multiple candidate models in order to determine the topology center point prediction model from among the multiple candidate models.

[0011] Optionally, the method further includes: When a new combination of operating conditions is received, the coordinates of the topology center point corresponding to the new combination of operating conditions are obtained, and the training set is updated based on the new combination of operating conditions and the coordinates of the topology center point corresponding to the new operating condition. The updated training set is used to retrain the preset model in order to re-obtain the topology center point prediction model.

[0012] A second aspect of the present invention provides a topological center point prediction apparatus, comprising: The acquisition module is used to simulate the crystal growth environment under different combinations of working parameters to obtain the first flow field data corresponding to the different combinations of working parameters. The first flow field data includes the position information of multiple discrete data points and the flow field parameters of each discrete data point. The flow field parameters include temperature parameters and flow velocity parameters. The determination module is used to convert the first flow field data corresponding to each combination of operating parameters into second flow field data under a preset structured grid, and determine the topological center point coordinates corresponding to the combination of operating parameters based on the second flow field data. The topological center point coordinates represent the coordinates of the center point of the gas vortex during the crystal growth process. The model generation module is used to construct a preset model and train the preset model based on multiple combinations of operating condition parameters and the topology center point coordinates corresponding to each combination of operating condition parameters to obtain a topology center point prediction model. The prediction module is used to input the target operating condition parameter combination into the topology center point prediction model to obtain the topology center point coordinates corresponding to the target operating condition parameter combination.

[0013] A third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executed, implements the steps of the topological center point prediction method as described in the first aspect above.

[0014] The method for predicting the topology center point provided by this invention includes: simulating the crystal growth environment under different combinations of operating parameters to obtain first flow field data corresponding to different combinations of operating parameters, wherein the first flow field data includes the position information of multiple discrete data points and flow field parameters for each discrete data point, the flow field parameters including temperature parameters and flow velocity parameters; for each combination of operating parameters, converting the first flow field data corresponding to the combination of operating parameters into second flow field data under a preset structured grid, and determining the topology center point coordinates corresponding to the combination of operating parameters based on the second flow field data, wherein the topology center point coordinates represent the coordinates of the center point of the gas vortex during the crystal growth process; constructing a preset model, and training the preset model based on multiple combinations of operating parameters and the topology center point coordinates corresponding to each combination of operating parameters to obtain a topology center point prediction model; inputting a target combination of operating parameters into the topology center point prediction model to obtain the topology center point coordinates corresponding to the target combination of operating parameters. Therefore, this invention converts flow field data under different combinations of operating parameters into second flow field data under the same preset structured grid, which facilitates the study of the variation law of flow field data under different combinations of operating parameters. This makes it easier to predict the topology center point when at least one operating parameter in the combination of operating parameters changes. Thus, the topology center point under the required combination of operating parameters can be directly determined through the topology center point prediction model, which makes it easier to determine the topology center point under different combinations of operating parameters without performing multiple simulation processes, thereby improving the optimization process efficiency of crystal growth.

[0015] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the scale in the drawings is for illustration only and does not represent the actual scale.

[0017] Figure 1 A flowchart illustrating the steps of the topology center point prediction method provided in an embodiment of the present invention is shown. Figure 2 A schematic diagram of the prediction process for topology center points in an embodiment of the present invention is shown; Figure 3 A schematic diagram of the process for acquiring the second flow field data is shown; Figure 4 A schematic diagram illustrating the process of determining the coordinates of the topology center point is shown. Figure 5 A schematic diagram of the model training process is shown; Figure 6 A schematic diagram of the topological center point prediction device provided in an embodiment of the present invention is shown. Detailed Implementation

[0018] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Silicon carbide (SiC), as a core representative of third-generation wide-bandgap semiconductor materials, boasts excellent physical and electrical properties such as wide bandgap, high breakdown electric field, high saturated electron drift velocity, and high thermal conductivity. It has demonstrated irreplaceable application value in high-end fields such as power electronic devices, radio frequency communication devices, new energy power generation and storage, and aerospace, and has become a key basic material supporting the upgrading of the next-generation semiconductor industry. The preparation process of SiC crystals requires extremely high temperatures and complex physicochemical coupling environments. Currently, mainstream crystallization processes include physical vapor transport (PVT), liquid phase epitaxy (LPE), and high-temperature chemical vapor deposition (HTCVD). Among these, PVT has become the widely adopted SiC crystal growth technology in industry and academia due to its advantages such as relatively simple equipment structure, mature and controllable process, high crystal growth quality, and ease of achieving large-size single crystal mass production. In the preparation of SiC ingots by the PVT method, high-purity SiC powder is first placed at the bottom of a graphite crucible, and a seed crystal of the same crystal type is fixed at the top. The crucible is then induction heated to 2100–2400°C to establish a stable temperature gradient. The temperature is maintained for tens to hundreds of hours to ensure the continuous sublimation and transport of the raw materials and to maintain stable pressure, temperature gradient and atmosphere until SiC crystallization is completed.

[0020] However, the PVT method for SiC crystal growth involves complex physicochemical behaviors such as high-temperature sublimation, gas-phase transport, interface crystallization, and multi-field coupling. A strong nonlinear coupling relationship exists between process parameters and the final crystal quality, and even small fluctuations in process parameters can lead to a significant increase in crystal defect density, severely impacting the performance of the SiC substrate and the yield of subsequent device fabrication. Currently, research on PVT-based SiC crystal growth, both domestically and internationally, still largely relies on numerous orthogonal experiments, single-factor variable experiments, and engineering experience summaries. This research approach not only incurs enormous experimental costs but also struggles to accurately capture the synergistic mechanisms between various process parameters, failing to achieve globally optimal control of process parameters. This results in insufficient SiC crystal quality stability, making it difficult to meet the application requirements of high-end power devices.

[0021] In recent years, machine learning has begun to be applied to crystallization process optimization, such as using models to accelerate computational fluid dynamics simulations and employing algorithms to achieve global optimality searches for multiple process parameters. However, the application of machine learning to crystal growth processes still has limitations. For example, it requires fitting and optimizing a large number of experimental and simulated process parameters, and there is the problem of uneven data sampling under different growth conditions, which makes data analysis difficult. Furthermore, the lack of deep integration with the evolution mechanisms of core physical fields such as thermal and flow fields greatly reduces the efficiency of crystallization process optimization.

[0022] In view of this, embodiments of the present invention provide a method, apparatus, and device for predicting topological center points. By converting discrete data collected under different operating conditions into second flow field data under the same preset structured grid, it is easier to analyze the variation law of topological center points under different operating conditions. In this case, the topological center point prediction model can be directly applied to the prediction of topological center points under different operating conditions without the need for complex analysis and conversion of data, thereby improving the efficiency of crystal growth process optimization.

[0023] Reference Figure 1 , Figure 1 A flowchart illustrating the steps of the topological center point prediction method provided in an embodiment of the present invention is shown, as follows: Figure 1 As shown, the prediction method specifically includes: S101 uses simulation to model the crystal growth environment under different combinations of operating parameters to obtain the first flow field data corresponding to different combinations of operating parameters.

[0024] The first flow field data includes the location information of multiple discrete data points and the flow field parameters of each discrete data point, including temperature parameters and flow velocity parameters.

[0025] In this embodiment, the crystal growth environment of silicon carbide under different combinations of operating parameters can be simulated to obtain the temperature and fluid velocity at each position in the crucible during the crystal growth process corresponding to different combinations of operating parameters. It should be noted that the fluid velocity at each position is a vector to represent the gas flow direction at each position, which will also affect the crystal quality during the crystal growth process. That is, the flow rate parameter actually includes the gas flow direction and gas flow velocity at each position.

[0026] The combination of operating parameters includes various operating parameters during the crystal growth process. Specifically, these parameters may include crystal growth power, coil moving distance, and top hole size. Taking the growth of silicon carbide crystals by PVT method as an example, crystal growth power represents the electrical power input to the induction coil, coil moving distance is the moving distance of the induction coil, and top hole size is the top hole size of the graphite crucible. In other words, the combination of operating parameters includes various operating parameters that affect the crystal growth environment.

[0027] S102, for each combination of operating parameters, the first flow field data corresponding to the combination of operating parameters is converted into second flow field data under a preset structured grid, and the coordinates of the topology center point corresponding to the combination of operating parameters are determined based on the second flow field data.

[0028] Among them, the topological center point coordinates represent the coordinates of the center point of the gas vortex during crystal growth.

[0029] In this embodiment, after obtaining the first flow field data corresponding to different combinations of operating parameters through simulation, the first flow field data is processed to obtain the coordinates of the topology center point under different combinations of operating parameters. It is understood that the sampling distribution of the first flow field data obtained under different combinations of operating parameters may differ. Therefore, the coordinates of the topology center point obtained directly from the first flow field data corresponding to different combinations of operating parameters are not uniform and are difficult to apply to training a topology center point prediction model. Therefore, the first flow field data corresponding to each combination of operating parameters can be converted into coordinate points under the same structured grid, so that the flow field data under multiple combinations of operating parameters are located in a unified grid model, facilitating the subsequent establishment of a prediction model for the topology center point.

[0030] The preset structured mesh can be determined based on the space in which the crystal grows. For example, the container used for actual crystal growth can be modeled and then meshed to obtain the preset structured mesh. Considering that the container generally uses an axisymmetric structure during crystal growth, the cross-section of the container's center point can be meshed as the preset structured mesh to reduce data complexity. For example, taking crystal growth in a cube-shaped container as an example, the square of the cross-section can be selected to draw the mesh, resulting in the preset structured mesh. The resolution of the preset structured mesh can be selected according to requirements.

[0031] S103. Construct a preset model and train the preset model based on multiple combinations of operating condition parameters and the coordinates of the topology center point corresponding to each combination of operating condition parameters to obtain a topology center point prediction model.

[0032] S104. Input the target working condition parameter combination into the topology center point prediction model to obtain the coordinates of the topology center point corresponding to the target working condition parameter combination.

[0033] In this context, the topological center point represents the vortex center of the gas vortex during crystal growth. It defines the location of the gas vortex, i.e., the gas flow path, which significantly impacts the quality of the crystal. Therefore, the location of the topological center point can be used as an indicator for process optimization during crystal growth. This allows for the construction of a topological center point prediction model. By inputting different combinations of operating parameters, the location of the topological center point can be determined quickly without extensive experiments or simulations. The preset model is a regression model, which establishes a relationship between input and output. This model can be trained to establish the relationship between the operating parameter combinations and the topological center point coordinates, resulting in the topological center point prediction model. Thus, a regression algorithm can be used to construct the mapping relationship between operating parameter combinations and topological center point coordinates. Multiple operating parameter combinations are used as training samples, and the corresponding topological center point coordinates are used as labels to train the regression model, resulting in the topological center point prediction model.

[0034] In this process, various combinations of operating parameters can be standardized to enable the model to accurately identify different types of operating parameters. This standardization process can specifically include data format standardization, arrangement order standardization of multiple operating parameters, etc. Therefore, when using the topology center point prediction model to determine the topology center point coordinates corresponding to the target operating parameter combination, the input target operating parameter combination can be standardized first, and then the topology center point coordinates corresponding to the target operating parameter combination can be determined.

[0035] In some embodiments, multiple different model parameters can be configured for the preset model. During the training process, preset models with different model parameters can be trained to obtain multiple candidate models. Then, the candidate model with accurate results can be selected from the multiple candidate models as the topology center point prediction model to improve the accuracy of model prediction.

[0036] The topology center point prediction method provided in this invention converts unstructured discrete point data into data under a structured grid during the simulated crystal growth process. Then, the coordinates of the topology center points corresponding to different operating parameters are determined using this structured grid data. The topology center point prediction model is then trained using these coordinates. This allows for direct determination of the corresponding topology center point coordinates based on the operating parameters during subsequent crystal growth simulations, improving the efficiency of topology center point determination. Furthermore, the topology center points under different operating conditions can be obtained by directly adjusting the operating parameters input to the topology center point prediction model. Since the topology center points under different operating conditions are obtained using the same structured grid, analysis can be performed under the same structured grid for crystal growth under different operating conditions. Applying this to the crystal growth simulation process helps to efficiently and accurately select the optimal operating parameters.

[0037] In one embodiment, the process of converting the first flow field data corresponding to different combinations of operating condition parameters into the second flow field data under a preset structured grid can be as follows: First, an interpolation function is used to linearly interpolate the multiple discrete data points included in the first flow field data to generate a grid dataset. The grid dataset includes the coordinates of each discrete data point in the preset structured grid and the flow field parameters corresponding to each coordinate. Then, based on the grid dataset, the preset structured grid is extended to obtain the second flow field data.

[0038] Specifically, the griddata function and linear interpolation can be used to map multiple discrete data points in the first flow field data to a preset structured grid to generate a grid dataset. The grid dataset can include the coordinate information of each discrete data point in the preset structured grid, as well as its corresponding flow field parameters, which can include axial velocity, radial velocity, and temperature.

[0039] In addition, after converting the first flow field data into the second flow field data under the preset structured grid, the boundary of the preset structured grid can be extended to avoid the interference of boundary effects on the detection topology center point.

[0040] In one embodiment, the specific method for extending the boundary of a preset structured grid may be to first obtain the boundary data of the grid dataset; then, extend the boundary of the preset structured grid and determine the flow field parameters corresponding to the extension points based on the boundary data; and then, generate second flow field data based on the flow field parameters corresponding to multiple extension points and the grid dataset.

[0041] Specifically, the boundary of a pre-defined structured grid can be mirrored and expanded. The mirror width can be selected according to requirements; for example, a mirror width of one grid point means that the boundary of the pre-defined structured grid is expanded outward by one grid point. Then, data is filled into the expanded grid points to fill in the flow field parameters. For temperature parameters, the temperature field value can be preprocessed to 0. For velocity parameters, the velocity values ​​of adjacent internal nodes can be copied to complete the mirror filling, obtaining the flow field parameters corresponding to multiple expanded points. The flow field parameters of multiple expanded points and the grid dataset are then used as the second flow field data. The coordinate positions of multiple expanded points can be marked to indicate the boundary position of the expanded pre-defined structured grid. After determining the flow field data corresponding to multiple expanded points, the flow field data at the boundary positions and the acquired grid dataset are combined to generate the second flow field data. That is, the second flow field data includes the flow field data of each coordinate point in the pre-defined structured grid and the flow field data of the expanded boundary points. In this way, when analyzing the flow field data of each coordinate point in the pre-defined structured grid to obtain the coordinates of the topology center point, the interference of the boundary of the pre-defined structured grid can be avoided.

[0042] It is understandable that the topological center point represents the center point of the gas vortex during crystal growth. Its characteristics are that the gas flow velocity is zero and the surrounding gas rotates in the same direction. Therefore, the coordinate point that meets the above characteristics can be determined based on the flow field data, and this coordinate point can be used as the topological center point. Identifying the position of the topological center point helps to analyze the flow field situation during crystal growth under different working conditions.

[0043] Specifically, it can be done by first determining the vector field index of each coordinate point in the preset structured grid based on the second flow field data. This vector field index represents the gas flow state of multiple adjacent coordinate points. Then, based on the vector field index and the second flow field data, the coordinates of the topology center point are determined from the multiple coordinate points.

[0044] In this embodiment, the topological center point represents the center of the gas vortex. Therefore, the vector field index of each coordinate point can be defined as the sum of the angles of the velocity vectors of multiple adjacent coordinate points located at the intersection of the tic-tac-toe shape within the area centered on that coordinate point. Its value can be the sum of the angles of the velocity vectors of multiple adjacent coordinate points divided by 2π. When the vector field index is 1, it indicates that the coordinate point is the topological center point; when the vector field index is 0, it indicates that the coordinate point is the boundary point of multiple gas vortices, which can be named a saddle point. Thus, based on the second flow field data, the vector field index of each coordinate point in the preset structured grid can be determined. Then, the points with a vector field index of 0 and the points with a vector field index of 1 are determined. It can be understood that the points with a vector field index of 0, i.e., saddle points, are the boundary points of multiple gas vortices. The number of saddle points helps determine the number of gas vortices. Therefore, based on the points with a vector field index of 0 and the points with a vector field index of 1, the position coordinates of the topological center point can be determined from the multiple coordinate points in the second flow field data.

[0045] It is understood that in this embodiment, the vector field index is defined as the sum of the gas flow directions at adjacent positions of a coordinate point. Therefore, the gas flow direction can be determined based on the gas velocity vector in the second flow field data. Thus, the method for determining the vector field index of each coordinate point in the preset structured grid based on the second flow field data can be as follows: for each coordinate point, first obtain the velocity vectors corresponding to multiple adjacent coordinate points from the second flow field data; then, determine the angle difference between two adjacent velocity vectors; and finally, sum the multiple angle differences to obtain the vector field index.

[0046] In this embodiment, based on the definition of the vector field index, for each coordinate point, the velocity vector of each intersection point in the tic-tac-toe structure formed with that coordinate point as the center can be obtained; then the difference between the velocity vectors of two adjacent coordinate points is used to determine the vector angle; then, the sum of multiple vector angles is divided by 2π to obtain the vector field index.

[0047] Understandably, when the unit grid size of the pre-defined structured grid is small, there may be cases where the same gas vortex includes multiple topological center points. In such cases, after determining multiple coordinate points that meet the conditions, adjacent coordinate points can be clustered into one class, and then the topological center point corresponding to each cluster can be determined. Specifically, multiple target coordinate points that meet the pre-defined conditions can be determined from the multiple coordinate points. The pre-defined conditions include that the vector field index is within the first target range and the flow velocity parameter is within the second target range. Then, based on the distance between the multiple target coordinate points, the multiple target coordinate points are clustered to obtain multiple clusters. After that, the centroid coordinates of each cluster are determined, and the centroid coordinates are determined as the topological center point coordinates.

[0048] In this context, the vortex center is the point with a velocity of 0, and the gas flow in the vicinity of the vortex center all revolves around it. Therefore, the coordinate point with a velocity close to 0 and a vector field exponent close to 1 can be considered the vortex center, i.e., the topological center point. Thus, the first target range can be 0.7~1.3, preferably 0.9~1.1, and the second target range can be 1e. -6 ~1e -2 When the vector field index of a coordinate point satisfies the first target range and the velocity value satisfies the second target range, that coordinate point is the target coordinate point.

[0049] After obtaining multiple target coordinate points, these points can be clustered. Specifically, a radius-based clustering algorithm compatible with the DBSCAN algorithm can be used to cluster multiple target coordinate points belonging to the same vortex into the same cluster. The clustering radius threshold can be 0.5 to 5 times the unit size of the structured grid.

[0050] It is understandable that if multiple target coordinate points in the same cluster are close to the vortex center point, then the centroid coordinates of the same cluster can be calculated and used as the topological center point coordinates to avoid the situation where the same gas vortex has multiple topological center point coordinates.

[0051] In one embodiment, when training the topology center point prediction model, multiple combinations of operating parameters and their corresponding topology center point coordinates can be first divided into training samples and test samples to train a preset model with different parameter configurations to obtain multiple candidate models. Then, the test samples are used to evaluate the multiple candidate models and select the topology center point prediction model to improve the accuracy of the prediction model. Specifically, multiple combinations of operating parameters and the corresponding topology center point coordinates can be divided into training sets and test sets according to a preset ratio. Then, under different model parameter configurations, the preset model is trained using the training set to obtain multiple candidate models. After that, the multiple candidate models are evaluated using the test set to determine the topology center point prediction model from the multiple candidate models.

[0052] In this embodiment, the number of operating condition parameter combinations in the training set and the number of operating condition parameter combinations in the test set can be determined based on the total number of operating condition parameter combinations. This allows for the division of various operating condition parameter combinations into training and test sets, with each combination labeled using the coordinates of its corresponding topology center point. The preset ratio used for this division can be selected based on actual needs.

[0053] The preset model is a regression model, which can be a Gaussian process regression, support vector regression, random forest regression, ridge regression, or K-nearest neighbor regression, etc. It can achieve the mapping prediction of the combination of operating parameters to the coordinates of the topology center point.

[0054] The model parameter configuration of the preset model includes one or more of the following: kernel function form, kernel parameter boundary, noise term, regularization strength, and number of optimization restarts. When training the preset model with the training set, preset models with different model parameter configurations can be trained to obtain candidate models corresponding to different model parameter configurations. For example, if multiple kernel function forms can be configured, the preset models corresponding to different kernel function forms can be trained with the training set to obtain multiple candidate models corresponding to different kernel function forms.

[0055] After obtaining multiple candidate models, each candidate model can be evaluated using a test set. The accuracy of the candidate models' predictions is then assessed based on the test results. When evaluating candidate models, RMSE, MAE, and R-squared can be used. 2 One or more of these can be used as evaluation metrics to quantify the performance of candidate models based on the evaluation metrics.

[0056] The system can display the evaluation results of different candidate models to the user, allowing the user to select the topology center point prediction model based on the evaluation results. Alternatively, it can directly select the candidate model with the smallest error or the best fit based on the evaluation results of multiple candidate models.

[0057] The obtained prediction model of the topology center point can be used to determine the coordinates of the corresponding topology center point based on the input combination of operating parameters. This eliminates the need to re-perform the crystal growth simulation every time the parameters are adjusted. Furthermore, the coordinates of the topology center point corresponding to different combinations of operating parameters can be directly used to analyze the variation law of the topology center point under different combinations of operating parameters. Applying this method to the flow field analysis process of crystal growth simulation can improve the analysis efficiency.

[0058] In one embodiment, if a new operating condition is added, it is necessary to obtain the topology center point of the flow field data corresponding to the new operating condition and add the corresponding data to the training set to retrain the preset model. Specifically, when a new operating condition parameter combination is received, the coordinates of the topology center point corresponding to the new operating condition parameter are first obtained, and the training set is updated based on the new operating condition parameter combination and the coordinates of the topology center point corresponding to the new operating condition parameter combination. Then, the preset model is retrained using the updated training set to obtain the topology center point prediction model again.

[0059] In this embodiment, when adding a new combination of operating parameters, the new combination of operating parameters and the corresponding topological center coordinates can be added to the training set. After adding the new combination of operating parameters to the training set, the preset model is retrained to obtain a new prediction model for the topological center. This allows for the direct calling of the solidified file for consistent inference when adding new research data on crystal growth or new operating conditions, thus achieving reproducible and iterative training.

[0060] In the case where the prediction model for the topology center point is selected from multiple candidate models, one approach is to retest the multiple candidate models trained after updating the training set, and obtain the candidate model with the highest accuracy as the prediction model for the topology center point. Alternatively, one approach is to directly use the model parameters corresponding to the candidate model selected in the previous training as the preset model parameters to improve training efficiency.

[0061] The topology center point prediction method provided in this invention converts unstructured discrete point data into data under a structured grid during the simulated crystal growth process. Then, the coordinates of the topology center points corresponding to different operating parameters are determined using this structured grid data. The topology center point prediction model is then trained using these coordinates. This allows for direct determination of the corresponding topology center point coordinates based on the operating parameters during subsequent crystal growth simulations, improving the efficiency of topology center point determination. Furthermore, the topology center points under different operating conditions can be obtained by directly adjusting the operating parameters input to the topology center point prediction model. Since the topology center points under different operating conditions are obtained using the same structured grid, analysis can be performed under the same structured grid for crystal growth under different operating conditions. Applying this to the crystal growth simulation process helps to efficiently and accurately select the optimal operating parameters.

[0062] The following describes the topological center point prediction method provided by the embodiments of the present invention with specific examples: Reference Figure 2 , Figure 2 A flowchart illustrating the topological center point prediction method provided in this disclosure embodiment is shown, as follows: Figure 2As shown, firstly, the crystal growth environment under different combinations of operating parameters is simulated to obtain the first flow field data corresponding to different combinations of operating parameters. In this process, two types of core input files can be prepared and stored according to a specified directory structure: one type is a flow field data file, used to store the flow field data of the crystal growth simulation. The format supports CSV / TXT / DAT and includes required columns such as x-coordinate, y-coordinate, axial-velocity, radial-velocity, and temperature. Optional columns include the cellnumber column. All files are stored in the . / data folder. The other type is a condition mapping table named arinf.csv, containing a data column and, importantly, operating parameter columns such as P (crystal growth power), dxc (coil movement distance), and dd (top hole size), used to establish the correlation between the operating parameter combinations and the flow field data. After simulating crystal growth using CFD, the flow field data of the gas phase region within the geometric region of crystal growth is obtained, and then the operating parameter combinations and flow field data files are stored separately.

[0063] Next, the first flow field data is processed to determine the topology center point corresponding to each combination of operating parameters. At this point, refer to... Figure 3 , Figure 3 A schematic diagram of the process for acquiring the second flow field data is shown, as follows: Figure 3 As shown, after reading multiple first flow field data from the stored flow field data file, each first flow field data is converted into a grid dataset under a preset structured grid. The resolution of the preset structured grid can be set as needed. The grid dataset includes gridded temperature data and gridded gas velocity data, that is, the grid dataset includes the temperature parameters and gas velocity parameters of each node of the preset structured grid. Then, the grid dataset is extended to avoid boundary effects, resulting in an extended dataset, which includes the temperature parameters and gas velocity parameters of the extended boundary points. The grid dataset and the extended dataset are combined to obtain the second flow field data, realizing the conversion of discrete data points into data nodes under the same preset structured grid.

[0064] Specifically, the first flow field data in the . / data folder can be read first. The griddata function is called and linear interpolation is used to map the unstructured grid flow field data to a structured grid of a specified resolution, generating a grid dataset including x-coordinate grid, y-coordinate grid, axial velocity grid, radial velocity grid, and temperature grid. Then, the boundary of the preset structured grid can be extended, with a radial width of 1 grid point. For boundary points with a temperature field value of 0, if the temperature field is non-zero in the effective flow field region, the temperature field of the boundary region or blank region is assigned a value of 0, or it can be preprocessed to 0 by the user. By copying the velocity values ​​of adjacent internal nodes and filling them radially, an extended grid dataset and a boundary marker matrix for identifying the boundary position are generated. Combining this extended dataset and the grid dataset yields multiple second flow field data.

[0065] Next, based on the second flow field data, the coordinates of the topology center point corresponding to the second flow field data were determined, with reference to... Figure 4 , Figure 4 A schematic diagram illustrating the process of determining the coordinates of the topology center point is shown, such as... Figure 4 As shown, the vector field index of each coordinate point is first determined based on the second flow field data. Then, based on the vector field indices corresponding to multiple coordinate points, topological feature points are determined. Furthermore, based on the distance between multiple topological feature points, they are clustered to obtain multiple clusters. Afterward, the centroid coordinates of each cluster are calculated and used as the coordinates of the topological center point. Specifically, the vector field index can be defined as the cumulative sum of vector angles within an 8-neighborhood of the velocity vector, with the direction sign included, divided by 2π. Based on this definition, the vector field index corresponding to each coordinate point is calculated according to the gas velocity vector at each coordinate point in the second flow field data. Then, based on the vector field index corresponding to each coordinate point, topological feature points are detected, excluding mirror boundaries. These topological feature points include saddle points and topological center points. The topological center point is a coordinate point where the vector field index is close to 1.0 and the velocity value is close to 0. The saddle point is a point where the vector field index is close to -1.0 and the velocity value is close to 0.

[0066] After identifying multiple saddle points and topological feature points, a radius-based brute-force algorithm compatible with the DBSCAN algorithm can be used to cluster the detected topological feature points, filtering out noise points, and calculating the centroid of each cluster as the final actual topological center point coordinates. The output can also integrate multi-dimensional flow field data to form a visualization for easy viewing. Before obtaining the vector field index, key parameters can be configured for vector field index acquisition and model training parameter configuration. Specifically, key parameters can include topological feature extraction parameters and model training configuration parameters. Topological feature extraction parameters include grid_resolution (structured grid resolution), tolerance (vector field index judgment tolerance), tolerance_v (judgment threshold for velocity approaching 0), eps (cluster radius), and min_samples (minimum cluster size). Model training configuration parameters include version (model save directory name), random_state (random seed), test_size (test set proportion), n_restarts_optimizer (number of optimizer restarts), and alpha (noise level), etc. Pre-configuring topological feature extraction parameters and model training configuration parameters can be used in the process of obtaining vector field indices, determining the coordinates of topological center points, and subsequent model training.

[0067] Then, refer to Figure 5 , Figure 5 A schematic diagram of the model training process is shown, such as... Figure 5As shown, after determining the coordinates of the topology center point corresponding to each combination of operating parameters, multiple combinations of operating parameters and the coordinates of the topology center point corresponding to each combination of operating parameters can be divided into a training set and a test set. A pre-set model is established using the Gaussian process regression algorithm. The pre-set model is trained using the training set, and the trained model is evaluated using the test set. Multiple parameter configurations for the pre-set models can be stored. When training the pre-set model using the training set, pre-set models with different parameter configurations are trained to obtain multiple candidate models. Then, the test set is used to evaluate each candidate model to determine the prediction model for the topology center point. Specifically, when constructing the preset model, the combination of operating parameters is used as the input features, and the coordinates of the topology center point are used as the prediction target to construct supervised learning samples. Before training, multiple combinations of operating parameters can be standardized to determine the stability of model training, and the standardized parameters are saved for subsequent reproduction. A fixed random seed ensures repeatability. During the training of the preset model, multiple combinations of operating parameters and the corresponding topology center point coordinates can be divided into training and test sets in a 7:3 ratio. The training set is used to train the preset model, and the test set is used to test the candidate models obtained from the training. Simultaneously, a parameter configuration set can be constructed, which includes at least the kernel function form, kernel function boundary, noise term, regularization strength, and number of optimization restarts. Therefore, when training the preset model using the training set, all parameter configurations can be traversed to obtain multiple candidate models.

[0068] Next, the candidate models with different parameter configurations are tested using a test set, and the performance of the candidate models is evaluated using error metrics and fit metrics, such as RMSE, MAE, and R-squared. 2 One or more of these are used to determine the predictive model as the topological center coordinate based on the evaluation results.

[0069] Finally, when it is necessary to determine the coordinates of the topology center point under different combinations of operating parameters, the combination of operating parameters can be directly input into the prediction model to determine the coordinates of the topology center point corresponding to that combination of operating parameters.

[0070] In addition, when adding new research data or predicting new working conditions, the solidified files can be called for consistent inference, enabling reproducible and iterative training.

[0071] Based on the same inventive concept, the present invention also provides a topological center point prediction device, referring to... Figure 6 , Figure 6 A schematic diagram of the topology center point prediction device provided in an embodiment of the present invention is shown, as follows: Figure 6 As shown, the prediction device specifically includes: The acquisition module 201 is used to simulate the crystal growth environment under different combinations of working parameters to obtain the first flow field data corresponding to different combinations of working parameters. The first flow field data includes the position information of multiple discrete data points and the flow field parameters of each discrete data point, including temperature parameters and flow velocity parameters. The determination module 202 is used to convert the first flow field data corresponding to each combination of operating parameters into second flow field data under a preset structured grid, and determine the topological center point coordinates corresponding to the combination of operating parameters based on the second flow field data. The topological center point coordinates represent the coordinates of the center point of the gas vortex during crystal growth. The model generation module 203 is used to construct a preset model and train the preset model based on multiple combinations of working condition parameters and the topological center point coordinates corresponding to each combination of working condition parameters to obtain a pallet center point prediction model. The prediction module 204 is used to input the target working condition parameter combination into the topology center point prediction model to obtain the topology center point coordinates corresponding to the target working condition parameter combination.

[0072] In some embodiments, the determining module 202 includes: The interpolation module is used to perform linear interpolation on multiple discrete data points included in the first flow field data using an interpolation function to generate a grid dataset. The grid dataset includes the coordinates of each discrete data point in a preset structured grid, as well as the flow field parameters corresponding to each coordinate. The extension module is used to extend the boundaries of a preset structured grid based on a grid dataset to obtain a second flow field data.

[0073] In some embodiments, the extension module includes: The first determining unit is used to determine the vector field index of each coordinate point in the preset structured grid based on the second flow field data. The vector field index represents the gas flow state of multiple coordinate points adjacent to the coordinate point. The second determining unit is used to determine the coordinates of the topology center point from multiple coordinate points based on the vector field index and the second flow field data.

[0074] In some embodiments, the model generation module 203 includes: The partitioning module is used to divide multiple combinations of operating condition parameters and the coordinates of the topology center point corresponding to each combination of operating condition parameters into training set and test set according to a preset ratio. The training module is used to train a preset model using a training set under different model parameter configurations to obtain multiple candidate models; The evaluation module is used to evaluate multiple candidate models using a test set to determine the topology center point prediction model from among the multiple candidate models.

[0075] In some embodiments, the prediction device further includes: The update module is used to obtain the coordinates of the topology center point corresponding to the new operating condition parameter combination when a new operating condition parameter combination is received, and to update the training set based on the new operating condition parameter combination and the coordinates of the topology center point corresponding to the new operating condition parameter combination. The retraining module is used to retrain the preset model using the updated training set in order to re-obtain the topology center point prediction model.

[0076] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the steps of the topological center point prediction method as described in any of the above embodiments.

[0077] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0078] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0079] The above provides a detailed description of the method, apparatus, and device for predicting topological center points provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

[0080] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the following claims.

[0081] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

[0082] The terms "an embodiment," "embodiment," or "one or more embodiments" as used herein mean that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment of the invention. Furthermore, please note that the examples of the phrase "in one embodiment" do not necessarily all refer to the same embodiment.

[0083] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0084] In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising a plurality of different elements and by means of a suitably programmed computer. In a unit claim enumerating a plurality of means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words may be interpreted as names.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting topological center points, characterized in that, The method includes: The crystal growth environment under different combinations of operating parameters was simulated to obtain the first flow field data corresponding to different combinations of operating parameters. The first flow field data includes the position information of multiple discrete data points and the flow field parameters of each discrete data point, including temperature parameters and flow velocity parameters. For each combination of operating parameters, the first flow field data corresponding to the combination of operating parameters is converted into second flow field data under a preset structured grid, and the topology center point coordinates corresponding to the operating condition are determined based on the second flow field data. The topology center point coordinates represent the coordinates of the center point of the gas vortex during crystal growth. A preset model is constructed, and the preset model is trained based on multiple combinations of operating condition parameters and the coordinates of the topology center point corresponding to each combination of operating condition parameters to obtain a topology center point prediction model. The target operating condition parameter combination is input into the topology center point prediction model to obtain the topology center point coordinates corresponding to the target operating condition parameter combination.

2. The method for predicting topological center points according to claim 1, characterized in that, The step of converting the first flow field data corresponding to the combination of operating parameters into second flow field data under a preset structured grid includes: Linear interpolation is performed on multiple discrete data points included in the first flow field data using an interpolation function to generate a grid dataset. The grid dataset includes the coordinates of each discrete data point in the preset structured grid and the flow field parameters corresponding to each coordinate. Based on the grid dataset, the preset structured grid is extended at the boundary to obtain the second flow field data.

3. The method for predicting topological center points according to claim 2, characterized in that, The step of extending the boundaries of the preset structured grid based on the grid dataset includes: Obtain the boundary data of the grid dataset; The boundaries of the preset structured grid are expanded, and the flow field parameters corresponding to the expanded points are determined based on the boundary data. Based on the flow field parameters corresponding to multiple expansion points and the grid dataset, second flow field data is generated.

4. The method for predicting topological center points according to claim 1, characterized in that, The step of determining the topology center point coordinates corresponding to the combination of operating parameters based on the second flow field data includes: Based on the second flow field data, the vector field index of each coordinate point in the preset structured grid is determined, and the vector field index represents the gas flow state of multiple coordinate points adjacent to the coordinate point. Based on the vector field index and the second flow field data, the coordinates of the topology center point are determined from a plurality of coordinate points.

5. The method for predicting topological center points according to claim 4, characterized in that, The step of determining the vector field index of each coordinate point in the preset structured grid based on the second flow field data includes: For each coordinate point, the velocity vectors corresponding to multiple adjacent coordinate points are obtained from the second flow field data. Determine the angle difference between two adjacent velocity vectors; The vector field index is obtained by summing the multiple angle differences.

6. The method for predicting topological center points according to claim 4, characterized in that, The step of determining the coordinates of the topology center point from multiple coordinate points based on the vector field index and the second flow field data includes: Multiple target coordinate points that satisfy preset conditions are determined from the multiple coordinate points, the preset conditions including the vector field index being located within a first target range and the flow velocity parameter being located within a second target range; Based on the distance between multiple target coordinate points, the multiple target coordinate points are clustered to obtain multiple clusters; Determine the centroid coordinates of each cluster, and use the centroid coordinates as the coordinates of the topological center point.

7. The method for predicting topological center points according to claim 1, characterized in that, The process of training the preset model based on multiple combinations of operating condition parameters and the coordinates of the topology center point corresponding to each combination of operating condition parameters to obtain a topology center point prediction model includes: According to a preset ratio, the various combinations of operating condition parameters and the coordinates of the topology center point corresponding to each combination of operating condition parameters are divided into a training set and a test set. Under different model parameter configurations, the preset model is trained using the training set to obtain multiple candidate models; The test set is used to evaluate multiple candidate models in order to determine the topology center point prediction model from among the multiple candidate models.

8. The method for predicting topological center points according to claim 7, characterized in that, The method further includes: When a new combination of operating conditions is received, the coordinates of the topology center point corresponding to the new combination of operating conditions are obtained, and the training set is updated based on the new combination of operating conditions and the coordinates of the topology center point corresponding to the new combination of operating conditions. The updated training set is used to retrain the preset model in order to re-obtain the topology center point prediction model.

9. A device for predicting topological center points, characterized in that, include: The acquisition module is used to simulate the crystal growth environment under different combinations of operating parameters to obtain the first flow field data corresponding to different combinations of operating parameters. The first flow field data includes the location information of multiple discrete data points and the flow field parameters of each discrete data point. The flow field parameters include temperature parameters and flow velocity parameters. The determination module is used to convert the first flow field data corresponding to each combination of operating parameters into second flow field data under a preset structured grid, and determine the topological center point coordinates corresponding to the combination of operating parameters based on the second flow field data. The topological center point coordinates represent the coordinates of the center point of the gas vortex during the crystal growth process. The model generation module is used to construct a preset model and train the preset model based on multiple combinations of operating condition parameters and the topology center point coordinates corresponding to each combination of operating condition parameters to obtain a topology center point prediction model. The prediction module is used to input the target operating condition parameter combination into the topology center point prediction model to obtain the topology center point coordinates corresponding to the target operating condition parameter combination.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes, it implements the steps of the topology center point prediction method as described in any one of claims 1-8.

Images