Power setting method and apparatus, electronic device, and storage medium

Abstract

Embodiments of the present application provide a power setting method, device, equipment and medium. The method comprises: obtaining feature data related to the seed crystal power; the feature data comprises real-time data and setting data, and the setting data comprises a target pulling speed at the end of the seed crystal stage; inputting the feature data into a power recommendation model, the power recommendation model being trained by feature data samples related to the seed crystal power and labeled sample seed crystal power, the feature data samples comprising real-time data samples and setting data samples, the real-time data samples comprising a sample pulling speed at the end of the seed crystal stage; generating a recommended power from the power recommendation model according to the feature data; setting the seed crystal power as the recommended power to control the change of the pulling speed at the end of the seed crystal stage to the target pulling speed, so that a suitable recommended power is automatically generated, and the seed crystal power is set accordingly, so that the pulling speed at the end of the seed crystal stage can change to the target pulling speed, thereby avoiding problems such as wire breakage in the crystal pulling process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of crystal preparation, in particular to a power setting method, a power setting device, an electronic device and a storage medium. BACKGROUND

[0002] The preparation process of single crystal silicon material is mainly based on the Czochralski process (CZ), which uses the Czochralski process to refine polycrystalline silicon raw materials into single crystal silicon. The process of generating a rod-shaped single crystal silicon crystal in the Czochralski single crystal process includes the steps of loading, heating the molten material, adjusting the temperature, seeding, shoulder setting, shoulder turning, equal diameter, and tailing.

[0003] Among them, after the polycrystalline silicon raw material is melted, the seeding cannot be started immediately, because the temperature at this time is higher than the seeding temperature, and the temperature must be adjusted to the seeding temperature. Seeding is to contact the seed crystal (that is, the single crystal processed into a certain shape) previously loaded at the end of the steel wire rope with the liquid surface. At the seeding temperature, silicon molecules will grow along the lattice direction of the seed crystal, thereby forming a single crystal. Shoulder setting is to gradually grow the diameter of the crystal to the required diameter. In the shoulder setting process, a section of crystal with gradually increasing length and gradually increasing diameter to the required diameter will be pulled out to eliminate the dislocation of the crystal. When the crystal grows to the required diameter in the shoulder setting process, it enters the shoulder turning process. Shoulder turning is to control the diameter of the crystal to the required diameter. After shoulder turning is completed, it enters the equal diameter control step, in which the crystal will grow according to the set diameter through automatic control of the pulling speed and the temperature.

[0004] In the current Czochralski process, the seeding power will be reflected in the seeding pulling speed. If the power is not appropriate, it may cause abnormality such as wire breakage during crystal pulling. SUMMARY

[0005] In view of the above problems, the present application embodiment is proposed to provide a power setting method to overcome the above problems or at least partially solve the above problems, so as to solve the problem that inappropriate seeding power may cause abnormality such as wire breakage.

[0006] Correspondingly, the present application embodiment also provides a power setting device, an electronic device and a storage medium to ensure the implementation and application of the above method.

[0007] In order to solve the above problems, the present application embodiment discloses a power setting method, comprising:

[0008] In the present Czochralski single crystal process, characteristic data related to the seeding power is obtained; wherein the characteristic data includes real-time data and setting data related to the seeding power, and the setting data includes a target pulling speed at the end of the seeding stage.

[0009] inputting the feature data into a power recommendation model, wherein the power recommendation model is trained by feature data samples related to the seed power and sample seed power labeled correspondingly, the feature data samples including real-time data samples and setting data samples related to the seed power, the real-time data samples including a sample pulling speed at the end of the seed crystal growth stage;

[0010] generating a recommended power from the power recommendation model according to the feature data;

[0011] setting the seed power as the recommended power to control the change of the pulling speed at the end of the seed crystal growth stage to the target pulling speed.

[0012] Optionally, before the feature data related to the seed power is obtained in the current Czochralski process, the method further comprises:

[0013] training a plurality of candidate power recommendation models of different structures respectively by using the feature data samples and the sample power labeled correspondingly in the training set;

[0014] testing the plurality of candidate power recommendation models of different structures according to the feature data samples and the sample power labeled correspondingly in the training set to obtain training set evaluation indexes corresponding to the plurality of candidate power recommendation models of different structures;

[0015] testing the plurality of candidate power recommendation models of different structures according to the feature data samples and the sample power labeled correspondingly in the test set to obtain test set evaluation indexes corresponding to the plurality of candidate power recommendation models of different structures;

[0016] selecting a power recommendation model for recommending power from the plurality of candidate power recommendation models of different structures according to the training set evaluation indexes and the test set evaluation indexes.

[0017] Optionally, the plurality of candidate power recommendation models of different structures include at least one of the following: a random forest model, an extreme gradient boosting model, and a category gradient boosting model.

[0018] Optionally, the training set evaluation indexes include at least two of the following: mean absolute error, mean square error, and determination coefficient, the test set evaluation indexes include at least two of the following: mean absolute error, mean square error, and determination coefficient, and the selecting a power recommendation model for recommending power from the plurality of candidate power recommendation models of different structures according to the training set evaluation indexes and the test set evaluation indexes comprises:

[0019] According to the training set evaluation index and the test set evaluation index, a training set comprehensive evaluation index and a test set comprehensive evaluation index of each candidate power recommendation model are determined respectively; wherein the mean absolute error in the training set comprehensive evaluation index and the training set evaluation index has a negative correlation, the mean square error in the training set comprehensive evaluation index and the training set evaluation index has a negative correlation, the determination coefficient in the training set comprehensive evaluation index and the training set evaluation index has a positive correlation, the mean absolute error in the test set comprehensive evaluation index and the test set evaluation index has a negative correlation, the mean square error in the test set comprehensive evaluation index and the test set evaluation index has a negative correlation, and the determination coefficient in the test set comprehensive evaluation index and the test set evaluation index has a positive correlation;

[0020] From each of the candidate power recommendation models, a candidate power recommendation model whose training set comprehensive evaluation index is greater than a first threshold value, whose test set comprehensive evaluation index is greater than a second threshold value, and whose difference between the training set comprehensive evaluation index and the test set comprehensive evaluation index is less than a third threshold value is selected as a power recommendation model for recommending power.

[0021] Optionally, the feature data related to the seeding power is obtained, including:

[0022] The current processing state in the temperature adjustment stage and the seeding stage is monitored;

[0023] When each preset processing state of the current processing state is reached, the feature data is collected according to a preset feature data type corresponding to the preset processing state, and the feature data collected each time is determined as the feature data corresponding to the current processing state.

[0024] Optionally, the power recommendation model includes a plurality of sub-power recommendation models, and the feature data is input into the power recommendation model, including:

[0025] When each preset processing state of the current processing state is reached, the sub-power recommendation model corresponding to the current processing state is called;

[0026] The feature data corresponding to the current processing state is input into the sub-power recommendation model corresponding to the current processing state;

[0027] The recommended power is generated by the power recommendation model according to the feature data, including:

[0028] The recommended power in the current processing state is generated by the sub-power recommendation model according to the feature data.

[0029] Optionally, the real-time data includes liquid surface brightness, and the set data includes target brightness.

[0030] The embodiment of the present application also discloses a power setting device, comprising:

[0031] A data acquisition module is configured to acquire feature data related to the seed crystal pulling power in the current Czochralski process, wherein the feature data comprises real-time data and setting data related to the seed crystal pulling power, and the setting data comprises a target pulling speed at the end of the seed crystal pulling stage.

[0032] A data input module is configured to input the feature data into a power recommendation model, wherein the power recommendation model is trained by feature data samples related to the seed crystal pulling power and corresponding labeled sample seed crystal pulling powers, and the feature data samples comprise real-time data samples and setting data samples related to the seed crystal pulling power, and the real-time data samples comprise sample pulling speeds at the end of the seed crystal pulling stage.

[0033] A power generation module is configured to generate a recommended power from the power recommendation model according to the feature data.

[0034] A power setting module is configured to set the seed crystal pulling power as the recommended power to control the pulling speed at the end of the seed crystal pulling stage to change to the target pulling speed.

[0035] Optionally, the device further comprises:

[0036] A model training module is configured to train candidate power recommendation models of multiple structures respectively by using feature data samples and corresponding labeled sample powers in a training set before acquiring the feature data related to the seed crystal pulling power in the current Czochralski process.

[0037] A training set test module is configured to test the candidate power recommendation models of the multiple structures according to the feature data samples and corresponding labeled sample powers in the training set to obtain training set evaluation indexes corresponding to the candidate power recommendation models of the multiple structures.

[0038] A test set test module is configured to test the candidate power recommendation models of the multiple structures according to feature data samples and corresponding labeled sample powers in a test set to obtain test set evaluation indexes corresponding to the candidate power recommendation models of the multiple structures.

[0039] A model selection module is configured to select a power recommendation model for recommending power from the candidate power recommendation models of the multiple structures according to the training set evaluation indexes and the test set evaluation indexes.

[0040] Optionally, the candidate power recommendation models of the multiple structures comprise at least one of the following: a random forest model, an extreme gradient boosting model and a category gradient boosting model.

[0041] Optionally, the training set evaluation indexes include at least two of the following: mean absolute error, mean square error, and determination coefficient, the test set evaluation indexes include at least two of the following: mean absolute error, mean square error, and determination coefficient, and the model selection module includes:

[0042] a comprehensive index determination sub-module configured to determine a training set comprehensive evaluation index and a test set comprehensive evaluation index of each of the candidate power recommendation models according to the training set evaluation indexes and the test set evaluation indexes; wherein the training set comprehensive evaluation index and the mean absolute error in the training set evaluation indexes have a negative correlation, the training set comprehensive evaluation index and the mean square error in the training set evaluation indexes have a negative correlation, the training set comprehensive evaluation index and the determination coefficient in the training set evaluation indexes have a positive correlation, the test set comprehensive evaluation index and the mean absolute error in the test set evaluation indexes have a negative correlation, the test set comprehensive evaluation index and the mean square error in the test set evaluation indexes have a negative correlation, and the test set comprehensive evaluation index and the determination coefficient in the test set evaluation indexes have a positive correlation;

[0043] a selection sub-module configured to select, from the candidate power recommendation models, a candidate power recommendation model whose training set comprehensive evaluation index is greater than a first threshold, whose test set comprehensive evaluation index is greater than a second threshold, and whose difference between the training set comprehensive evaluation index and the test set comprehensive evaluation index is less than a third threshold, as the power recommendation model for recommending power.

[0044] Optionally, the data acquisition module includes:

[0045] a state monitoring sub-module configured to monitor a current processing state in a temperature adjustment stage and a crystal pulling stage;

[0046] an acquisition sub-module configured to, when each preset processing state of the current processing state is reached, acquire feature data according to a preset feature data type corresponding to the preset processing state, and determine the feature data acquired each time as feature data corresponding to the current processing state.

[0047] Optionally, the power recommendation model includes a plurality of sub-power recommendation models, and the data input module includes:

[0048] a model calling sub-module configured to, when each preset processing state of the current processing state is reached, call a sub-power recommendation model corresponding to the current processing state;

[0049] a data input sub-module configured to input the feature data corresponding to the current processing state into the sub-power recommendation model corresponding to the current processing state.

[0050] The power generation module comprises:

[0051] The power generation sub-module is configured to generate, according to the feature data, a recommended power in the current processing state by using the sub-power recommendation model.

[0052] Optionally, the real-time data comprises liquid surface brightness, and the setting data comprises target brightness.

[0053] The electronic device comprises a processor, a communication interface, a memory and a communication bus.

[0054] The memory is configured to store a computer program.

[0055] The processor is configured to execute the program stored in the memory to implement the method steps described above.

[0056] The readable storage medium comprises instructions, and when the instructions are executed by the processor of the electronic device, the electronic device can execute the power setting method described in one or more embodiments of the present application.

[0057] The embodiments of the present application have the following advantages:

[0058] According to the embodiments of the present application, the feature data related to the seed pulling power is obtained in the current single crystal pulling process, wherein the feature data comprises real-time data and setting data related to the seed pulling power, the setting data comprises a target pulling speed at the end of the seed pulling stage, the feature data is input into a power recommendation model, wherein the power recommendation model is trained by feature data samples related to the seed pulling power and corresponding labeled sample seed pulling power, the feature data samples comprise real-time data samples and setting data samples related to the seed pulling power, the real-time data samples comprise sample pulling speeds at the end of the seed pulling stage, a recommended power is generated by the power recommendation model according to the feature data, the seed pulling power is set as the recommended power, and the pulling speed at the end of the seed pulling stage is changed to the target pulling speed, so that the appropriate recommended power is automatically generated according to the feature data, the seed pulling power is set accordingly, the pulling speed at the end of the seed pulling stage can be changed to the target pulling speed, and the problems such as wire breakage in the crystal pulling process are avoided. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 is a step flow chart of a power setting method embodiment of the present application;

[0060] Figure 2is a schematic diagram of a box plot of data analysis;

[0061] Figure 3 is a schematic diagram of a density plot of data analysis;

[0062] Figure 4 is a schematic diagram of a distribution of recommendation result bias;

[0063] Figure 5 is a schematic diagram of a power recommendation process;

[0064] Figure 6 is a step flow chart of an embodiment of a power setting method of the present application;

[0065] Figure 7 is a structural block diagram of an embodiment of a power setting device of the present application;

[0066] Figure 8 is a structural block diagram of a computing device for power setting according to an exemplary embodiment. DETAILED DESCRIPTION

[0067] In order to make the above objectives, features and advantages of the present application more apparent, more comprehensible, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0068] Referring to Figure 1 , a step flow chart of an embodiment of a power setting method of the present application is shown, which can specifically include the following steps:

[0069] Step 101, in the present Czochralski process, feature data related to the seeding power is acquired; wherein the feature data includes real-time data and setting data related to the seeding power, and the setting data includes a target pulling speed at the end of the seeding stage.

[0070] In the embodiment of the present application, the Czochralski process is a process of refining raw materials into single crystals by using the Czochralski method, for example, a process of refining silicon into single crystals. The Czochralski process can be divided into a temperature adjusting stage, a seeding stage, a shoulder releasing stage, etc.

[0071] In this embodiment of the invention, during the Czochralski single crystal pulling process, the power is the power of the single crystal furnace, which has the greatest impact on the pulling speed. The crystal pulling power is a power parameter that can be set during both the temperature control and crystal pulling stages. During the Czochralski single crystal pulling process, the crystal pulling power needs to be adjusted multiple times according to the actual situation. Data related to the crystal pulling power can be divided into two types: real-time data and set data. Real-time data is data monitored during actual operation, such as liquid level brightness, etc., and can specifically include any applicable real-time data; this embodiment of the invention does not limit this. Set data is data set for the Czochralski single crystal pulling equipment, or data that is expected to be achieved, such as target liquid level brightness, target pulling speed, etc., and can specifically include any applicable set data; this embodiment of the invention does not limit this. The target pulling speed is the pulling speed set at the end of the crystal pulling stage during this Czochralski single crystal pulling process. By analyzing the relationship between the pulling speed at the end of the crystal pulling stage and the breakage rate during the crystal pulling process, the pulling speed that results in a low or minimum breakage rate can be determined in advance as the target pulling speed.

[0072] In this embodiment of the invention, optionally, the real-time data includes: liquid surface brightness, and the set data includes: target brightness. Liquid surface brightness can be obtained by reading real-time data from industrial control modules such as PLCs (Programmable Logic Controllers). The target brightness can be set according to actual needs, for example, the pulling speed corresponding to a low or minimum breakage rate; this invention does not impose any limitations on this.

[0073] In this embodiment of the invention, through data analysis, data related to the crystal-driving power is selected from the real-time data and set data of the temperature adjustment stage and the crystal-driving stage as feature data.

[0074] For example, based on historical big data statistics, real-time and set data related to chip power are first screened. Then, data analysis and processing are performed. Next, feature data is processed and screened, and the correlation between feature data and the correlation between feature data and power are analyzed to ultimately determine the feature data that can be used for data modeling. Analysis and processing methods include, but are not limited to, histograms, scatter plots, violin plots, heatmaps, box plots, density plots, and other big data statistical data processing and analysis methods. Figure 2 The diagram shows a box plot illustrating the data analysis, and as shown... Figure 3 The diagram shows a density plot of the data analysis. Through the above data analysis, real-time data and set data related to the chip extraction power were selected as the final feature data.

[0075] Step 102, input the feature data into the power recommendation model, wherein the power recommendation model is trained by a feature data sample related to the seeding power and a sample seeding power corresponding to a label, the feature data sample includes real-time data sample related to the seeding power and setting data sample, and the real-time data sample includes a sample pulling speed at the end of the seeding stage.

[0076] In the embodiment of the present application, the recommended seeding power can be obtained in a machine learning manner according to the correlation between the feature data related to the seeding power and the seeding power, and a power recommendation model that can recommend the seeding power.

[0077] In the embodiment of the present application, in order to train the power recommendation model, sample data and corresponding label data are needed, that is, a feature data sample related to the seeding power and a sample seeding power corresponding to a label. The feature data sample and the sample seeding power can be obtained through multiple experiments, or can be selected from historical data. The feature data sample includes real-time data sample related to the seeding power and setting data sample, and the real-time data sample includes a sample pulling speed at the end of the seeding stage.

[0078] Since the power recommendation model is a model that needs to output a better or best seeding power, therefore, the feature data sample should be the data of a better or best Czochralski process in history, for example, the sample pulling speed at the end of the seeding stage in the feature data sample should be in the range of a lower or lowest wire breaking rate in the crystal pulling process.

[0079] For example, considering that the label variable of the power recommendation model is the seeding power, which is a continuous variable. According to the data type of the label data, a suitable model is selected, and the power recommendation model can use a random forest, an eXtreme Gradient Boosting (XGboost) model, a Categorical Gradient Boosting (CATboost) model, or other any applicable model, and the embodiment of the present application does not limit this. The feature data sample in the training set and the sample seeding power corresponding to the label are input into the model, and then the model is trained until the model converges.

[0080] After the model training is completed, the model is evaluated, and the feature data sample of the test set is input into the power recommendation model for target value recommendation verification, for example, Figure 4The distribution of the recommended result deviation is shown in the diagram, the horizontal coordinate is the deviation between the recommended result and the sample power of the test set, and the vertical coordinate is the frequency. From the distribution of the deviation between the recommended result and the sample power of the test set, it can be seen that most of the actual production requirements are met, and the power recommendation model can be applied.

[0081] In the embodiment of the present application, in the actual application stage, the input of the power recommendation model is the feature data related to the seeding power in the temperature regulation stage and the seeding stage, and the output is the recommended value of the seeding power, denoted as recommended power.

[0082] Step 103, generating the recommended power from the power recommendation model according to the feature data.

[0083] In the embodiment of the present application, for the current Czochralski single crystal process, after obtaining the current feature data, the obtained feature data is input into the power recommendation model, the output of the power recommendation model is obtained, and then the output of the power recommendation model is used as the recommended power.

[0084] In the embodiment of the present application, before the end of the seeding stage, the feature data at the time can be input into the power recommendation model every time a preset time period elapses or the process reaches a preset processing state, so that the updated recommended power is obtained continuously.

[0085] Step 104, setting the seeding power as the recommended power to control the change of the pulling rate at the end of the seeding stage to the target pulling rate.

[0086] In the embodiment of the present application, the seeding power is set as the recommended power. Since the target pulling rate is included in the feature data, the feature data and the recommended power have a correlation. When the seeding power is set as the recommended power generated by the power recommendation model, the pulling rate changes to the target pulling rate in the seeding stage, and finally the pulling rate approaches or equals to the target pulling rate.

[0087] According to the embodiment of the present application, the feature data related to the seed power is obtained in the current Czochralski process, wherein the feature data includes real-time data related to the seed power and setting data, the setting data includes a target pulling speed at the end of the seed stage, the feature data is input into a power recommendation model, wherein the power recommendation model is trained by feature data samples related to the seed power and corresponding labeled sample seed power, the feature data samples include real-time data samples related to the seed power and setting data samples, the real-time data samples include sample pulling speed at the end of the seed stage, the recommended power is generated from the power recommendation model according to the feature data, and the seed power is set as the recommended power, so as to control the change of the pulling speed at the end of the seed stage to the target pulling speed, so that the appropriate recommended power is automatically generated according to the feature data, and the seed power is set accordingly, so that the pulling speed at the end of the seed stage can change to the target pulling speed, thereby avoiding the occurrence of problems such as wire breakage in the crystal pulling process.

[0088] In an optional embodiment of the present application, step 101 can include: monitoring the current processing state in the temperature adjustment stage and the seed stage; when each preset processing state of the current processing state is reached, collecting the feature data according to the preset feature data type corresponding to the preset processing state, and determining the feature data collected each time as the feature data corresponding to the current processing state.

[0089] In the temperature adjustment stage and the seed stage, the processing state of the process includes the state of the liquid surface brightness, the state of the temperature, the length of the crystal, and other processing states, or any other applicable processing state, and the embodiment of the present application does not limit this.

[0090] In the temperature adjustment stage and the seed stage, the processing state of the process is monitored to obtain the current processing state. For example, the liquid surface brightness is monitored to obtain the current liquid surface brightness, or whether the temperature adjustment stage is ended is monitored to obtain whether the current is the end of the temperature adjustment stage.

[0091] In order to recommend the power multiple times according to the feature data of the temperature adjustment stage and the seed stage, the feature data needs to be collected every certain period of time. Therefore, multiple preset recommendation points are set in the process of the temperature adjustment stage and the seed stage, and each preset recommendation point corresponds to a preset processing state of the process. For example, the liquid surface brightness reaches a preset brightness, the temperature adjustment stage ends, and the like.

[0092] In the process of the temperature adjustment stage and the seed stage, the feature data related to the seed power is different in different stages. Therefore, the preset processing state corresponds to a preset feature data type, and of course, the preset feature data type is the feature data related to the seed power.

[0093] Each time the current processing state reaches a preset processing state, feature data of the preset feature data type corresponding to that preset processing state is collected once. The collected feature data is then determined as the feature data corresponding to the current processing state, that is, the feature data when the process is in the preset processing state.

[0094] In an optional embodiment of the present invention, the power recommendation model includes multiple sub-power recommendation models. Step 102 may include: when the current processing state reaches a preset processing state, calling the sub-power recommendation model corresponding to the current processing state; and inputting the feature data corresponding to the current processing state into the sub-power recommendation model corresponding to the current processing state. Step 103 may include: generating a recommended power for the current processing state by the sub-power recommendation model based on the feature data.

[0095] Because the characteristic data related to crystal pulling power will differ at different stages, the types of preset characteristic data corresponding to multiple preset processing states are different. That is, in one case, the types of preset characteristic data corresponding to each preset processing state are different, and in another case, the types of preset characteristic data corresponding to two preset processing states are the same, while the types of preset characteristic data corresponding to two preset processing states are different.

[0096] Different models need to be trained for prediction depending on the type of preset feature data. The power recommendation model includes multiple sub-power recommendation models. After determining the preset processing state, a corresponding sub-power recommendation model can be obtained that can handle the feature data of the corresponding preset feature data type. When the current processing state reaches a certain preset processing state, the sub-power recommendation model corresponding to that preset processing state is called.

[0097] In the current processing state, after collecting feature data, input the sub-power recommendation model corresponding to the current processing state to obtain the recommended power in the current processing state through the sub-power recommendation model.

[0098] For example, such as Figure 5 The diagram illustrates the power recommendation process. Through data analysis, feature extraction, and feature processing, feature data samples and corresponding labeled sample crystal-leading power are obtained. Then, data modeling is performed. The temperature adjustment process is divided into 3-5 sub-power recommendation models, and the crystal-leading process is divided into 2-5 power recommendation models. Through model training, multiple trained sub-power recommendation models are obtained. These are then deployed in practical applications, with the corresponding models called at different stages. Power recommendations are performed and set as recommended power during the temperature adjustment process and during the crystal-leading process, until the shoulder-forming stage.

[0099] With reference to Figure 6 , a step flow chart of an embodiment of the power setting method of the present application is shown, which can specifically include the following steps:

[0100] In step 201, the feature data samples in the training set and the corresponding labeled sample power are used to train candidate power recommendation models of multiple structures respectively.

[0101] In the embodiment of the present application, in order to train a power recommendation model with better effect, the data used for training the model is divided into a training set and a test set, wherein the test set generally accounts for 25% to 15%, and the embodiment of the present application does not limit this.

[0102] In the embodiment of the present application, considering that the label variable of the power recommendation model is the seed power, which belongs to a continuous variable, a suitable model of multiple structures is selected according to the data type of the label data. Then, for each structure of the model, the feature data samples in the training set and the corresponding labeled sample power are input into the model, and then the model is trained until the model converges, to obtain a candidate power recommendation model of the corresponding structure.

[0103] In an optional embodiment of the present application, the candidate power recommendation model of multiple structures includes at least one of the following: a random forest model, an extreme gradient boosting model, and a category gradient boosting model.

[0104] In one embodiment, the preliminarily screened model includes one of the following three models: a random forest model, an extreme gradient boosting model, and a category gradient boosting model, for example, a random forest model with multiple structures. In one embodiment, the preliminarily screened model includes one of the following three models: a random forest model, an extreme gradient boosting model, and a category gradient boosting model, and other applicable structure models. In one embodiment, the preliminarily screened model includes two of the following three models: a random forest model, an extreme gradient boosting model, and a category gradient boosting model. In one embodiment, the preliminarily screened model includes two of the following three models: a random forest model, an extreme gradient boosting model, and a category gradient boosting model, and other applicable structure models. In one embodiment, the preliminarily screened model includes the following three models: a random forest model, an extreme gradient boosting model, and a category gradient boosting model. In one embodiment, the preliminarily screened model includes the following three models: a random forest model, an extreme gradient boosting model, and a category gradient boosting model, and other applicable structure models. The random forest model is a classifier that uses multiple trees to train and predict samples. The extreme gradient boosting model is a model that uses large-scale parallel boosting trees for prediction. The category gradient boosting model is a gradient boosting algorithm model that can well process category features.

[0105] Step 202, according to the feature data sample in the training set and the sample power corresponding to the label, test the candidate power recommendation model of the plurality of structures, and obtain the training set evaluation index corresponding to the candidate power recommendation model of the plurality of structures.

[0106] In the embodiment of the application, the evaluation index includes MAE (Mean Absolute Error), MSE (Mean Squared Error), R2 (R-Square), and the like, or other applicable evaluation indexes, and the embodiment of the application is not limited thereto.

[0107] In order to evaluate the effect of the candidate power recommendation model of the plurality of structures, the feature data sample in the training set and the sample power corresponding to the label are used to test the model. Specifically, for each structure of the candidate power recommendation model, the feature data sample in the training set is input, and then the output recommended power and the labeled sample power are compared to obtain the evaluation index corresponding to the candidate power recommendation model, which is denoted as the training set evaluation index.

[0108] For example, the training set evaluation index is shown in the following table:

[0109] cat_model MAE Score 0.93432 cat_model MSE Score 1.82532 cat_model R2 Score 0.82342 xgre_model MAE Score 0.53434 xgre_model MSE Score 0.49434 xgre_model R2 Score 0.97434 regr_model MAE Score 1.97434 regr_model MSE Score 6.76434 regr_model R2 Score 0.60434

[0110] Step 203, according to the feature data sample in the test set and the sample power corresponding to the label, test the candidate power recommendation model of the plurality of structures, and obtain the test set evaluation index corresponding to the candidate power recommendation model of the plurality of structures.

[0111] In the embodiment of the application, the feature data sample in the test set and the sample power corresponding to the label are used to test the model. Specifically, for each structure of the candidate power recommendation model, the feature data sample in the test set is input, and then the output recommended power and the labeled sample power are compared to obtain the evaluation index corresponding to the candidate power recommendation model, which is denoted as the test set evaluation index.

[0112] For example, the test set evaluation index is shown in the following table:

[0113] cat_model MAE Score 1.05432 cat_model MSE Score 2.15532 cat_model R2 Score 0.87342 xgre_model MAE Score 0.88434 xgre_model MSE Score 1.69434 xgre_model R2 Score 0.89434 regr_model MAE Score 1.97434 regr_model MSE Score 6.81434 regr_model R2 Score 0.59434

[0114] Step 204, according to the training set evaluation index and the test set evaluation index, select the power recommendation model for recommending power from the candidate power recommendation model of the plurality of structures.

[0115] In the embodiments of the present application, in the candidate power recommendation models of various structures, when selecting the power recommendation model for recommending power, the training set evaluation index and the test set evaluation index need to be screened. The model meeting the preset requirements of the training set evaluation index and the test set evaluation index is screened out as the power recommendation model for recommending power. Or if there are multiple models whose training set evaluation index and test set evaluation index meet the preset requirements, the model with the optimal training set evaluation index and test set evaluation index is selected as the power recommendation model for recommending power. For example, R2 is greater than 0.4 and the greater the better, MAE and MSE are the smallest and the smaller the better, and accordingly the best power recommendation model is selected from three models as the final model.

[0116] In an optional embodiment of the present application, the training set evaluation index includes at least two of the following: mean absolute error, mean square error, and determination coefficient, the test set evaluation index includes at least two of the following: mean absolute error, mean square error, and determination coefficient, and step 204 can include: determining the training set comprehensive evaluation index and the test set comprehensive evaluation index of each candidate power recommendation model according to the training set evaluation index and the test set evaluation index; wherein the training set comprehensive evaluation index and the mean absolute error in the training set evaluation index have a negative correlation, the training set comprehensive evaluation index and the mean square error in the training set evaluation index have a negative correlation, the training set comprehensive evaluation index and the determination coefficient in the training set evaluation index have a positive correlation, the test set comprehensive evaluation index and the mean absolute error in the test set evaluation index have a negative correlation, the test set comprehensive evaluation index and the mean square error in the test set evaluation index have a negative correlation, and the test set comprehensive evaluation index and the determination coefficient in the test set evaluation index have a positive correlation; selecting, from each candidate power recommendation model, the candidate power recommendation model whose training set comprehensive evaluation index is greater than a first threshold value, whose test set comprehensive evaluation index is greater than a second threshold value, and whose difference between the training set comprehensive evaluation index and the test set comprehensive evaluation index is less than a third threshold value, as the power recommendation model for recommending power.

[0117] Since the average absolute error and the mean square error are smaller the better, and the determination coefficient is larger the better in the average absolute error, the mean square error and the determination coefficient, in order to comprehensively evaluate the model, a comprehensive evaluation index is designed. If the evaluation index includes the average absolute error, the average absolute error in the training set comprehensive evaluation index and the training set evaluation index has a negative correlation. If the evaluation index includes the mean square error, the mean square error in the training set comprehensive evaluation index and the training set evaluation index has a negative correlation. If the evaluation index includes the determination coefficient, the determination coefficient in the training set comprehensive evaluation index and the training set evaluation index has a positive correlation. For example, the average absolute error and the mean square error are added to obtain a sum, and the determination coefficient is divided by the sum to obtain a value as the training set comprehensive evaluation index. If the evaluation index includes the average absolute error, the average absolute error in the test set comprehensive evaluation index and the test set evaluation index has a negative correlation. If the evaluation index includes the mean square error, the mean square error in the test set comprehensive evaluation index and the test set evaluation index has a negative correlation. If the evaluation index includes the determination coefficient, the determination coefficient in the test set comprehensive evaluation index and the test set evaluation index has a positive correlation. For example, the average absolute error and the mean square error are added to obtain a sum, and the determination coefficient is divided by the sum to obtain a value as the test set comprehensive evaluation index.

[0118] According to the principle that the average absolute error and the mean square error are smaller the better, and the determination coefficient is larger the better, it can be obtained that the training set comprehensive evaluation index is larger the better, and the test set comprehensive evaluation index is also larger the better. However, the training set comprehensive evaluation index and the test set comprehensive evaluation index should be as close as possible, so as to indicate that the power recommendation model does not overfit, and the accuracy on the training set and the test set is relatively good. Therefore, the candidate power recommendation model with the training set comprehensive evaluation index greater than a first threshold value, the test set comprehensive evaluation index greater than a second threshold value, and the difference between the training set comprehensive evaluation index and the test set comprehensive evaluation index less than a third threshold value is selected as the power recommendation model used for recommending power. The first threshold value, the second threshold value and the third threshold value can be set according to actual needs, and the embodiments of the present application do not limit this.

[0119] In step 205, in the current Czochralski single crystal process, feature data related to the seed power is obtained; wherein the feature data includes real-time data related to the seed power and setting data, and the setting data includes a target pulling rate at the end of the seed crystal stage.

[0120] Step 206, input the feature data into a power recommendation model, wherein the power recommendation model is trained by feature data samples related to seed crystal power and corresponding labeled sample seed crystal power, the feature data samples include real-time data samples and setting data samples related to the seed crystal power, and the real-time data samples include a sample pulling rate at the end of the seed crystal stage.

[0121] Step 207, generate a recommended power from the power recommendation model according to the feature data.

[0122] Step 208, set the seed crystal power as the recommended power to control the change of the pulling rate at the end of the seed crystal stage to the target pulling rate.

[0123] According to the embodiment of the present application, by using the feature data samples in the training set and the corresponding labeled sample power, a plurality of candidate power recommendation models of different structures are trained, the plurality of candidate power recommendation models of different structures are tested according to the feature data samples in the training set and the corresponding labeled sample power, the training set evaluation indexes corresponding to the plurality of candidate power recommendation models of different structures are obtained, the plurality of candidate power recommendation models of different structures are tested according to the feature data samples in the test set and the corresponding labeled sample power, the test set evaluation indexes corresponding to the plurality of candidate power recommendation models of different structures are obtained, the power recommendation model used for recommending power is selected from the plurality of candidate power recommendation models of different structures according to the training set evaluation indexes and the test set evaluation indexes, and the feature data related to the seed crystal power is obtained in the current Czochralski process; wherein the feature data includes real-time data and setting data related to the seed crystal power, the setting data includes a target pulling rate at the end of the seed crystal stage, the feature data is input into a power recommendation model, wherein the power recommendation model is trained by feature data samples related to seed crystal power and corresponding labeled sample seed crystal power, the feature data samples include real-time data samples and setting data samples related to the seed crystal power, and the real-time data samples include a sample pulling rate at the end of the seed crystal stage, a recommended power is generated from the power recommendation model according to the feature data, the seed crystal power is set as the recommended power to control the change of the pulling rate at the end of the seed crystal stage to the target pulling rate, so that the appropriate recommended power is automatically generated according to the feature data, the seed crystal power is set accordingly, the pulling rate at the end of the seed crystal stage can change to the target pulling rate, and the problems such as wire breakage in the crystal pulling process are avoided.

[0124] It should be noted that for the method embodiments, the series of acts complement each other to achieve the purpose of this embodiment, therefore, the sequence of the acts should not be construed as a limitation of the embodiments of this application. In addition, those skilled in the art should understand that the acts described in the specification are only preferred acts, and not all acts are necessary for the embodiments of this application.

[0125] With reference to Figure 7 , a structural block diagram of a power setting device embodiment of the present application is shown, which can specifically include the following modules:

[0126] The data acquisition module 301 is configured to acquire feature data related to the seed crystal drawing power in the current Czochralski process; wherein the feature data includes real-time data and setting data related to the seed crystal drawing power, and the setting data includes a target pulling speed at the end of the seed crystal drawing stage.

[0127] The data input module 302 is configured to input the feature data into a power recommendation model, wherein the power recommendation model is trained by feature data samples related to the seed crystal drawing power and corresponding labeled sample seed crystal drawing powers, and the feature data samples include real-time data samples and setting data samples related to the seed crystal drawing power, and the real-time data samples include sample pulling speeds at the end of the seed crystal drawing stage.

[0128] The power generation module 303 is configured to generate a recommended power from the power recommendation model according to the feature data.

[0129] The power setting module 304 is configured to set the seed crystal drawing power as the recommended power to control the pulling speed at the end of the seed crystal drawing stage to change to the target pulling speed.

[0130] Optionally, the device further includes:

[0131] The model training module is configured to train a plurality of candidate power recommendation models of different structures by using feature data samples and corresponding labeled sample powers in a training set before acquiring the feature data related to the seed crystal drawing power in the current Czochralski process.

[0132] The training set test module is configured to test the plurality of candidate power recommendation models of different structures according to the feature data samples and corresponding labeled sample powers in the training set, to obtain training set evaluation indexes corresponding to the plurality of candidate power recommendation models of different structures.

[0133] a test set test module, configured to test the candidate power recommendation models of the plurality of structures according to the feature data samples in the test set and the sample power corresponding to the labels, to obtain test set evaluation indexes corresponding to the candidate power recommendation models of the plurality of structures;

[0134] a model selection module, configured to select, from the candidate power recommendation models of the plurality of structures, a power recommendation model for recommending power according to the training set evaluation indexes and the test set evaluation indexes.

[0135] Optionally, the candidate power recommendation models of the plurality of structures include at least one of a random forest model, an extreme gradient boosting model, and a category gradient boosting model.

[0136] Optionally, the training set evaluation indexes include at least two of a mean absolute error, a mean square error, and a determination coefficient, and the test set evaluation indexes include at least two of a mean absolute error, a mean square error, and a determination coefficient, and the model selection module includes:

[0137] a comprehensive index determination sub-module, configured to determine, according to the training set evaluation indexes and the test set evaluation indexes, a training set comprehensive evaluation index and a test set comprehensive evaluation index of each of the candidate power recommendation models respectively, wherein the training set comprehensive evaluation index and the mean absolute error in the training set evaluation index have a negative correlation, the training set comprehensive evaluation index and the mean square error in the training set evaluation index have a negative correlation, the training set comprehensive evaluation index and the determination coefficient in the training set evaluation index have a positive correlation, the test set comprehensive evaluation index and the mean absolute error in the test set evaluation index have a negative correlation, the test set comprehensive evaluation index and the mean square error in the test set evaluation index have a negative correlation, and the test set comprehensive evaluation index and the determination coefficient in the test set evaluation index have a positive correlation.

[0138] a selection sub-module, configured to select, from the candidate power recommendation models, a candidate power recommendation model whose training set comprehensive evaluation index is greater than a first threshold, whose test set comprehensive evaluation index is greater than a second threshold, and whose difference between the training set comprehensive evaluation index and the test set comprehensive evaluation index is less than a third threshold, as the power recommendation model for recommending power.

[0139] Optionally, the data acquisition module includes:

[0140] a state monitoring sub-module, configured to monitor a current processing state in a temperature adjustment stage and a crystal pulling stage.

[0141] The collection submodule is configured to collect the feature data according to a preset feature data type corresponding to the preset processing state each time the current processing state reaches a preset processing state, and determine the feature data collected each time as the feature data corresponding to the current processing state.

[0142] Optionally, the power recommendation model comprises a plurality of sub-power recommendation models, and the data input module comprises:

[0143] The model calling submodule is configured to call a sub-power recommendation model corresponding to the current processing state each time the current processing state reaches a preset processing state.

[0144] The data input submodule is configured to input the feature data corresponding to the current processing state into the sub-power recommendation model corresponding to the current processing state.

[0145] The power generation module comprises:

[0146] The power generation submodule is configured to generate a recommended power in the current processing state from the sub-power recommendation model according to the feature data.

[0147] Optionally, the real-time data comprises liquid surface brightness, and the set data comprises target brightness.

[0148] According to the embodiment of the present application, the feature data related to the seed crystal pulling power is obtained in the current Czochralski process, wherein the feature data comprises real-time data and set data related to the seed crystal pulling power, the set data comprises a target pulling speed at the end of the seed crystal pulling stage, the feature data is input into a power recommendation model, wherein the power recommendation model is trained by a feature data sample related to the seed crystal pulling power and a sample seed crystal pulling power corresponding to the label, the feature data sample comprises a real-time data sample and a set data sample related to the seed crystal pulling power, the real-time data sample comprises a sample pulling speed at the end of the seed crystal pulling stage, a recommended power is generated from the power recommendation model according to the feature data, the seed crystal pulling power is set as the recommended power, and the pulling speed at the end of the seed crystal pulling stage is changed to the target pulling speed, so that the appropriate recommended power is automatically generated according to the feature data, the seed crystal pulling power is set according to the recommended power, the pulling speed at the end of the seed crystal pulling stage can be changed to the target pulling speed, and the problems such as wire breakage in the crystal pulling process are avoided.

[0149] For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the related parts are described in the part of the method embodiment.

[0150] Figure 8is a block diagram of an electronic device 400 for power setting according to an exemplary embodiment. The electronic device 400 can be, for example, a mobile phone, a computer, a digital broadcasting terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0151] Referring to Figure 8 The electronic device 400 can include one or more of the following components: a processing component 402, a memory 404, a power component 406, a multimedia component 408, an audio component 410, an input / output (I / O) interface 412, a sensor component 414, and a communication component 416.

[0152] The processing component 402 usually controls overall operations of the electronic device 400, such as operations associated with displaying, phone calling, data communication, camera operation, and recording operation. The processing component 402 can include one or more processors 420 to execute instructions to complete all or part of steps of the above-described power setting method. In addition, the processing component 402 can include one or more modules to facilitate the interaction between the processing component 402 and other components. For example, the processing component 402 can include a multimedia module to facilitate the interaction between the multimedia component 408 and the processing component 402.

[0153] The memory 404 is configured to store various types of data to support operations of the electronic device 400. Examples of these data include instructions for any application or method operating on the electronic device 400, contact data, phonebook data, messages, pictures, videos, etc. The memory 404 can be implemented by any type of volatile or non-volatile storage devices or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0154] The power component 404 provides power for various components of the electronic device 400. The power component 404 can include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power for the electronic device 400.

[0155] The multimedia component 408 includes a screen to provide an output interface between the electronic device 400 and a user. In some embodiments, the screen can include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense a touch, a slide, or a gesture on the touch panel. The touch sensor can not only sense a boundary of a touching or a sliding action, but also detect duration and intensity of the touching or sliding action. In some embodiments, the multimedia component 408 includes a front camera and / or a rear camera. The front camera and / or the rear camera can receive external multimedia data when the electronic device 400 is in an operation mode, such as a photographing mode or a video mode. Each of the front camera and the rear camera can be a fixed optical lens system or have a focal length and optical zooming capability.

[0156] The audio component 410 is configured to output and / or input an audio signal. For example, the audio component 410 includes a microphone (MIC) to receive an external audio signal when the electronic device 400 is in an operation mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal can be further stored in the memory 404 or transmitted via the communication component 416. In some embodiments, the audio component 410 further includes a speaker to output an audio signal.

[0157] The I / O interface 412 provides an interface for the processing component 402 and peripheral interface modules, such as a keypad, a click wheel, buttons, and the like. The buttons can include, but are not limited to, a home button, a volume button, a start button, and a lock button.

[0158] The sensor component 414 includes one or more sensors to provide various state assessments for the electronic device 400. For example, the sensor component 414 can detect an open / closed position of the device 400, relative positioning of components, such as a display and a keypad of the electronic device 400, a change in position of the electronic device 400 or a component of the electronic device 400, the presence or absence of user contact with the electronic device 400, the orientation or acceleration / deceleration of the electronic device 400, and a temperature change of the electronic device 400. The sensor component 414 can include a proximity sensor configured to detect presence of a nearby object without any physical touch. The sensor component 414 can further include a light sensor such as a CMOS or CCD image sensor for use in an imaging application. In some embodiments, the sensor component 414 can further include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0159] The communication component 416 is configured to facilitate wired or wireless communication between the electronic device 400 and other devices. The electronic device 400 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof. In an exemplary embodiment, the communication component 414 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 414 also includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0160] In an exemplary embodiment, the electronic device 400 can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors, or other electronic elements for performing the power setting method described above.

[0161] In an exemplary embodiment, a non-transitory computer-readable storage medium, such as the memory 404 including instructions, is also provided, which can be executed by the processor 420 of the electronic device 400 to complete the power setting method described above. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0162] A non-transitory computer-readable storage medium, when the instructions in the storage medium are executed by a processor of a terminal, enable the terminal to perform a power setting method, the method comprising:

[0163] In the present Czochralski single crystal process, characteristic data related to the seed power is obtained; wherein the characteristic data includes real-time data related to the seed power and setting data, and the setting data includes a target pulling rate at the end of the seed stage;

[0164] The characteristic data is input into a power recommendation model, wherein the power recommendation model is trained by characteristic data samples related to the seed power and corresponding labeled sample seed power, and the characteristic data samples include real-time data samples related to the seed power and setting data samples, and the real-time data samples include sample pulling rates at the end of the seed stage;

[0165] According to the characteristic data, a recommended power is generated by the power recommendation model;

[0166] The seed power is set as the recommended power to control the change of the pulling speed at the end of the seed stage to the target pulling speed.

[0167] Optionally, before the feature data related to the seed power is acquired in the current Czochralski single crystal process, the method further comprises:

[0168] The feature data samples in the training set and the corresponding labeled sample power are used to train candidate power recommendation models of multiple structures respectively;

[0169] The candidate power recommendation models of multiple structures are tested according to the feature data samples in the training set and the corresponding labeled sample power, to obtain training set evaluation indexes corresponding to the candidate power recommendation models of multiple structures;

[0170] The candidate power recommendation models of multiple structures are tested according to the feature data samples in the test set and the corresponding labeled sample power, to obtain test set evaluation indexes corresponding to the candidate power recommendation models of multiple structures;

[0171] According to the training set evaluation indexes and the test set evaluation indexes, a power recommendation model for recommending power is selected from the candidate power recommendation models of multiple structures.

[0172] Optionally, the candidate power recommendation models of multiple structures include at least one of the following: a random forest model, an extreme gradient boosting model, and a category gradient boosting model.

[0173] Optionally, the training set evaluation indexes include at least two of the following: mean absolute error, mean square error, and determination coefficient, the test set evaluation indexes include at least two of the following: mean absolute error, mean square error, and determination coefficient, and the power recommendation model for recommending power is selected from the candidate power recommendation models of multiple structures according to the training set evaluation indexes and the test set evaluation indexes.

[0174] According to the training set evaluation index and the test set evaluation index, a training set comprehensive evaluation index and a test set comprehensive evaluation index of each candidate power recommendation model are determined respectively; wherein the mean absolute error in the training set comprehensive evaluation index and the training set evaluation index has a negative correlation, the mean square error in the training set comprehensive evaluation index and the training set evaluation index has a negative correlation, the determination coefficient in the training set comprehensive evaluation index and the training set evaluation index has a positive correlation, the mean absolute error in the test set comprehensive evaluation index and the test set evaluation index has a negative correlation, the mean square error in the test set comprehensive evaluation index and the test set evaluation index has a negative correlation, and the determination coefficient in the test set comprehensive evaluation index and the test set evaluation index has a positive correlation;

[0175] From each of the candidate power recommendation models, a candidate power recommendation model whose training set comprehensive evaluation index is greater than a first threshold value, whose test set comprehensive evaluation index is greater than a second threshold value, and whose difference between the training set comprehensive evaluation index and the test set comprehensive evaluation index is less than a third threshold value is selected as a power recommendation model for recommending power.

[0176] Optionally, the feature data related to the seeding power is obtained, including:

[0177] The current processing state in the temperature adjustment stage and the seeding stage is monitored.

[0178] When each preset processing state of the current processing state is reached, the feature data is collected according to a preset feature data type corresponding to the preset processing state, and the feature data collected each time is determined as the feature data corresponding to the current processing state.

[0179] Optionally, the power recommendation model includes a plurality of sub-power recommendation models, and the feature data is input into the power recommendation model, including:

[0180] When each preset processing state of the current processing state is reached, the sub-power recommendation model corresponding to the current processing state is called;

[0181] The feature data corresponding to the current processing state is input into the sub-power recommendation model corresponding to the current processing state;

[0182] The recommended power is generated by the power recommendation model according to the feature data, including:

[0183] The recommended power in the current processing state is generated by the sub-power recommendation model according to the feature data.

[0184] Optionally, the real-time data includes liquid surface brightness, and the set data includes target brightness.

[0185] The various embodiments described in this specification are described in the context of progressive development of embodiments of the application. Each embodiment is distinguished from other embodiments by the features described in that embodiment. The same or similar features of different embodiments are identified by reference to the same or similar reference characters.

[0186] Those skilled in the art will appreciate that embodiments of the application can be devised for a method, an apparatus, or a computer program product. Accordingly, embodiments of the application can be embodied in a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of the application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0187] Embodiments of the application are described with reference to the flowchart illustrations and / or block diagrams of the methods, terminal devices (systems), and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device, or other programmable data processing terminal devices to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal devices, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 means for performing each of the one or more functions specified in the flowchart illustrations and / or block diagrams.

[0188] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing terminal device to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 means for performing each of the one or more functions specified in the flowchart illustrations and / or block diagrams.

[0189] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device to cause a series of operational steps to be performed on the computer or other programmable terminal device to produce a computer-implemented process such that the instructions which execute on the computer or other programmable terminal device provide steps for implementing the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 means for performing each of the one or more functions specified in the flowchart illustrations and / or block diagrams.

[0190] While the preferred embodiments of the application have been described above, it should be understood that many modifications and adaptations to those embodiments will be possible on the basis of the foregoing description and drawings. Therefore, the following claims are intended to cover all adaptations and modifications of the preferred embodiments as come within the scope of the application.

[0191] Finally, it should be noted that the terms "first", "second", and the like, herein do not denote any order, quantity, combination, or importance, but rather are used to distinguish one element from another, and are not intended to denote a relation or order between such elements or steps. Moreover, the term "comprising" or "comprises", or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0192] The above provides a power setting method and device, an electronic device and a readable storage medium, the principle and implementation of the application are described by applying specific examples in the text, the above embodiment is only used to help understand the method and core idea of the application; at the same time, for those skilled in the art, according to the idea of the application, the specific implementation and application range will be changed, and the above description should not be understood as the limitation of the application.

Claims

1. A power setting method, characterized in that, include: In this Czochralski single crystal pulling process, characteristic data related to crystal pulling power is acquired, including: monitoring the current processing status during the temperature adjustment stage and the crystal pulling stage; when the current processing status reaches a preset processing status, the characteristic data is collected once according to the preset characteristic data type corresponding to the preset processing status, and the characteristic data collected each time is determined as the characteristic data corresponding to the current processing status; wherein, the characteristic data includes real-time data and set data related to the crystal pulling power, and the set data includes the target pulling speed at the end of the crystal pulling stage; The feature data is input into the power recommendation model, wherein the power recommendation model is trained by feature data samples related to the crystal pulling power and the crystal pulling power of the corresponding labeled samples. The feature data samples include real-time data samples and set data samples related to the crystal pulling power. The real-time data samples include the sample pulling speed at the end of the crystal pulling stage. Based on the feature data, the power recommendation model generates a recommended power. The crystal-leading power is set to the recommended power to control the pulling speed at the end of the crystal-leading stage to change towards the target pulling speed.

2. The method according to claim 1, characterized in that, Before acquiring characteristic data related to crystal pulling power during this Czochralski single crystal growth process, the method further includes: Using the feature data samples in the training set and the corresponding labeled sample power, candidate power recommendation models with various structures are trained respectively. Based on the feature data samples and corresponding labeled sample power in the training set, the candidate power recommendation models with multiple structures are tested to obtain the training set evaluation index corresponding to the candidate power recommendation models with multiple structures. Based on the feature data samples and the corresponding labeled sample power in the test set, the candidate power recommendation models of the various structures are tested to obtain the test set evaluation index corresponding to the candidate power recommendation models of the various structures. Based on the training set evaluation metrics and the test set evaluation metrics, a power recommendation model for recommending power is selected from the candidate power recommendation models with various structures.

3. The method according to claim 2, characterized in that, The candidate power recommendation model with various structures includes at least one of the following: random forest model, extreme gradient boosting model, and categorical gradient boosting model.

4. The method according to claim 2, characterized in that, The training set evaluation metrics include at least two of the following: mean absolute error, mean squared error, and coefficient of determination. The test set evaluation metrics include at least two of the following: mean absolute error, mean squared error, and coefficient of determination. The step of selecting a power recommendation model for power recommendation from among the candidate power recommendation models with various structures based on the training set evaluation metrics and the test set evaluation metrics includes: Based on the training set evaluation metrics and the test set evaluation metrics, the comprehensive evaluation metrics for the training set and the comprehensive evaluation metrics for the test set are determined for each candidate power recommendation model. Specifically, the comprehensive evaluation metrics for the training set and the mean absolute error among the training set evaluation metrics are negatively correlated; the comprehensive evaluation metrics for the training set and the mean squared error among the training set evaluation metrics are negatively correlated; the comprehensive evaluation metrics for the training set and the coefficient of determination among the training set evaluation metrics are positively correlated; the comprehensive evaluation metrics for the test set and the mean absolute error among the test set evaluation metrics are negatively correlated; the comprehensive evaluation metrics for the test set and the mean squared error among the test set evaluation metrics are negatively correlated; and the comprehensive evaluation metrics for the test set and the coefficient of determination among the test set evaluation metrics are positively correlated. From the candidate power recommendation models, the candidate power recommendation model that has a training set comprehensive evaluation index greater than a first threshold, a test set comprehensive evaluation index greater than a second threshold, and a difference between the training set comprehensive evaluation index and the test set comprehensive evaluation index less than a third threshold is selected as the power recommendation model for recommending power.

5. The method according to claim 1, characterized in that, The power recommendation model includes multiple sub-power recommendation models, and the step of inputting the feature data into the power recommendation model includes: When the current processing state reaches a preset processing state, the sub-power recommendation model corresponding to the current processing state is invoked; Input the feature data corresponding to the current processing state into the sub-power recommendation model corresponding to the current processing state; The step of generating recommended power from the power recommendation model based on the feature data includes: Based on the feature data, the sub-power recommendation model generates a recommended power for the current processing state.

6. The method according to claim 1, characterized in that, The real-time data includes: liquid surface brightness, and the set data includes: target brightness.

7. A power setting device, characterized in that, include: The data acquisition module is used to acquire characteristic data related to crystal pulling power during this Czochralski single crystal pulling process, including: monitoring the current processing status during the temperature adjustment stage and the crystal pulling stage; when the current processing status reaches a preset processing status, acquiring the characteristic data once according to the preset characteristic data type corresponding to the preset processing status, and determining the characteristic data acquired each time as the characteristic data corresponding to the current processing status; wherein, the characteristic data includes real-time data and set data related to the crystal pulling power, and the set data includes the target pulling speed at the end of the crystal pulling stage; The data input module is used to input the feature data into the power recommendation model, wherein the power recommendation model is trained by feature data samples related to the crystal pulling power and the crystal pulling power of the corresponding labeled samples. The feature data samples include real-time data samples and set data samples related to the crystal pulling power. The real-time data samples include the sample pulling speed at the end of the crystal pulling stage. A power generation module is used to generate recommended power from the power recommendation model based on the feature data; The power setting module is used to set the crystal-leading power to the recommended power in order to control the pulling speed at the end of the crystal-leading stage to change towards the target pulling speed.

8. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the steps of the method described in any one of claims 1-6.

9. A readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform one or more of the power setting methods as described in claims 1-6.

Images