Seeding power determination method based on data driving and related device

Abstract

The invention discloses a seeding power determination method based on data driving and a related device, and belongs to the field of czochralski single crystal in photovoltaics, the disclosed method comprises the following steps: determining whether a shouldering broken line occurs in the last crystal pulling process according to data of the last crystal pulling process; if the shouldering broken line does not occur, obtaining the seeding process in the last crystal pulling process, the process parameters of the shouldering process and the field parameters of the current crystal pulling, and determining the seeding power of the current crystal pulling according to the obtained parameters by utilizing a first seeding power model; and if the shouldering wire breakage occurs and the record of entering the equal-diameter working procedure in the furnace is not carried out, working procedure parameters of the seeding working procedure in the last crystal pulling process and the current crystal pulling field parameters are obtained, and the seeding power of the current crystal pulling is determined according to the obtained parameters by utilizing a second seeding power model. According to the method, automatic and intelligent determination of the seeding power based on data driving is realized, the determination accuracy of the seeding power is improved, meanwhile, the production efficiency of single crystals is improved, and the line breakage rate is reduced.

Description

Technical Field

[0001] The present application belongs to the technical field of direct-pull single crystal in photovoltaics, and in particular to a method for determining seeding power based on data drive and related devices. Background Art

[0002] In the single crystal growth control system, the industry's method for determining the seeding power is still in the exploratory stage, and is mainly determined manually. The seeding power is determined by on-site process engineers based on their experience.

[0003] However, this determination method is prone to unsatisfactory and inaccurate setting of the seeding power due to subjective factors, lack of on-site experience and other issues. It also requires a large number of front-line workers to pay attention and operate at all times, which will also have a certain impact on the single crystal production cost, output and production efficiency. The seeding power determination method based on known technology will also lead to a high wire break rate during the crystal pulling process. Summary of the invention

[0004] In view of this, the present application provides a data-driven seeding power determination method and related devices, which are used to solve the problems of unsatisfactory and inaccurate seeding power setting and high disconnection rate in the known technology.

[0005] The specific technical solutions are as follows:

[0006] A method for determining seeding power based on data drive, comprising:

[0007] Get the data of the last crystal pulling process;

[0008] Determine whether shoulder breakage occurs in the last crystal pulling process according to the data of the last crystal pulling process;

[0009] If the shoulder release line breakage did not occur in the last crystal pulling process, obtain the process parameters of the seeding process, the process parameters of the shoulder release process and the field parameters of this crystal pulling process in the last crystal pulling process, and use the first seeding power model according to the process parameters of the seeding process, the process parameters of the shoulder release process and the field parameters of this crystal pulling process to determine the seeding power of this crystal pulling process;

[0010] If shoulder breakage occurred in the last crystal pulling process and there is no record of entering the equal diameter process in this furnace, obtain the process parameters of the seeding process in the last crystal pulling process and the field parameters of this crystal pulling, and use the second seeding power model to determine the seeding power of this crystal pulling according to the process parameters of the seeding process in the last crystal pulling process and the field parameters of this crystal pulling;

[0011] Among them, the second crystal seeding power model and the first crystal seeding power model are respectively aimed at different scenarios of whether shoulder breakage occurred in the last crystal pulling process in the crystal pulling preparation process. Different intelligent agents trained by data set construction are constructed by selecting features matching the targeted scenarios and solving the standard interval range of features for the process parameters in the historical crystal pulling process.

[0012] Optionally, the process parameters of the seeding process include at least some of the parameters of the seeding power, average seeding pulling speed, late seeding pulling speed, seeding time, seeding temperature, argon flow rate, crystal rotation speed, crucible rotation speed, number of re-throws and seeding residual material amount in the last crystal pulling process;

[0013] The process parameters of the shoulder release process include at least some of the parameters of the last shoulder release time, shoulder release height, shoulder release early height, shoulder release middle height, shoulder release opening time, argon gas flow rate, crystal rotation speed and crucible rotation speed;

[0014] The field parameters of this crystal pulling include at least some parameters of the remaining material amount, liquid surface temperature, number of re-throws, thermal field size, and single crystal size.

[0015] Optionally, the process of constructing the first seeding power model includes:

[0016] Obtaining source data of each process in the crystal pulling preparation process from historical data of the crystal pulling preparation process;

[0017] Performing feature selection processing that matches a first crystal pulling scenario on source data of each process, so as to obtain first features that match the first crystal pulling scenario and satisfy a first correlation condition with the seeding power based on the feature selection processing; the first crystal pulling scenario is a scenario in which no shoulder breakage occurs in the last crystal pulling process;

[0018] By screening the process data of seeding to equal diameter according to the equal diameter performance, solving the standard interval range of each of the first characteristics; the process data includes the data of each of the first characteristics;

[0019] Selecting data for each of the first features according to a standard interval range of each of the first features, and combining the selected feature data with a first seeding power corresponding to the feature data in historical data to obtain a first data set;

[0020] A machine learning model is trained based on the first data set to obtain the first seeding power model.

[0021] Optionally, performing feature selection processing that matches the first crystal pulling scenario on the source data of each process, so as to obtain first features that match the first crystal pulling scenario and satisfy a first correlation condition with the seeding power based on the feature selection processing, includes:

[0022] Determining the correlation between different process parameters in the source data of each process;

[0023] Eliminate some parameters from multiple process parameters that meet the preset correlation relationship to obtain various retained parameters;

[0024] The importance of each retention parameter relative to the seeding power is determined using a gradient boosting decision tree model, and the retention parameters whose corresponding importance meets the importance condition and matches the first crystal pulling scenario are selected as the first features.

[0025] Optionally, the process of constructing the second seeding power model includes:

[0026] Obtaining source data of each process in the crystal pulling preparation process from historical data of the crystal pulling preparation process;

[0027] Performing feature selection processing that matches a second crystal pulling scenario on the source data of each process, so as to obtain each second feature that matches the second crystal pulling scenario and satisfies a second correlation condition with the seeding power based on the feature selection processing; the second crystal pulling scenario is a scenario in which shouldering and wire breaking occurs in the last crystal pulling process and there is no record of entering the equal diameter process in the current heat;

[0028] By screening the process data of seeding to equal diameter according to the equal diameter performance, solving the standard interval range of each second feature; the process data includes the data of each second feature;

[0029] Selecting data for each of the second characteristics according to the standard interval range of each of the second characteristics, and combining the selected characteristic data with the second seeding power corresponding to the characteristic data in the historical data to obtain a second data set;

[0030] A machine learning model is trained based on the second data set to obtain the second seeding power model.

[0031] Optionally, performing feature selection processing that matches the second crystal pulling scenario on the source data of each process, so as to obtain each second feature that matches the second crystal pulling scenario and satisfies a second correlation condition with the seeding power based on the feature selection processing, includes:

[0032] Determining the correlation between different process parameters in the source data of each process;

[0033] Eliminate some parameters from multiple process parameters that meet the preset correlation relationship to obtain various retained parameters;

[0034] The importance of each retention parameter relative to the seeding power is determined using a gradient boosting decision tree model, and the retention parameters whose corresponding importance meets the importance condition and matches the second crystal pulling scenario are selected as the second features.

[0035] Optionally, by screening the process data from seeding to equal diameter according to the equal diameter performance, the standard interval range of the corresponding characteristics is solved, including:

[0036] Filter the process data of each seeding to equal diameter according to the equal diameter performance;

[0037] Based on the statistical information of the screened process data from seeding to equal diameter, the standard interval of the corresponding feature is solved;

[0038] The corresponding feature is the corresponding first feature or second feature corresponding to the process data.

[0039] Optionally, perform machine learning model training based on the corresponding dataset, including:

[0040] Using the corresponding data set to train multiple decision trees in the gradient boosting decision tree model, each decision tree improves the model performance by reducing the residual of the previous decision tree until the end condition is met, and the corresponding seeding power model is obtained;

[0041] Wherein, the corresponding data set is the first data set, and the corresponding seeding power model is the first seeding power model; or, the corresponding data set is the second data set, and the corresponding seeding power model is the second seeding power model.

[0042] A device for determining seeding power based on data drive, comprising:

[0043] An acquisition module is used to obtain the data of the last crystal pulling process;

[0044] A first determination module is used to determine whether shoulder breakage occurs in the last crystal pulling process according to the last crystal pulling process data;

[0045] The second determination module is used to obtain the process parameters of the seeding process, the process parameters of the shoulder release process and the field parameters of the current crystal pulling process in the last crystal pulling process when no shoulder release breakage occurred in the last crystal pulling process, and determine the seeding power of the current crystal pulling according to the process parameters of the seeding process, the process parameters of the shoulder release process and the field parameters of the current crystal pulling process in the last crystal pulling process using the first seeding power model;

[0046] The third determination module is used to obtain the process parameters of the seeding process in the last crystal pulling process and the field parameters of the current crystal pulling process when the shoulder release line breaks in the last crystal pulling process and there is no record of entering the equal diameter process in the current furnace, and use the second seeding power model to determine the seeding power of the current crystal pulling process according to the process parameters of the seeding process in the last crystal pulling process and the field parameters of the current crystal pulling process;

[0047] Among them, the second crystal seeding power model and the first crystal seeding power model are respectively aimed at different scenarios of whether shoulder breakage occurred in the last crystal pulling process in the crystal pulling preparation process. Different intelligent agents trained by data set construction are constructed by selecting features matching the targeted scenarios and solving the standard interval range of features for the process parameters in the historical crystal pulling process.

[0048] A computer-readable medium stores a computer program, which, when executed by a processor, can be used to implement the data-driven seeding power determination method as described in any one of the above.

[0049] According to the above scheme, the data-driven seeding power determination method and related devices provided in the present application, for different scenarios of whether shoulder breakage occurred in the last crystal pulling process during the crystal pulling preparation process, select features that match the targeted scenarios and solve the standard interval range of the features for the process parameters in the historical crystal pulling process to construct a data set and then train different intelligent agents to obtain the first seeding power model and the second seeding power model that are respectively adapted to different scenarios. On this basis, by obtaining the data of the last crystal pulling process, it is determined whether the shoulder break occurred in the last crystal pulling process according to the acquired data, and the seeding power model adapted to the current actual scenario (whether the shoulder break actually occurred or not) is called according to the judgment result. The seeding power of this crystal pulling is determined based on the called model based on the adapted model input parameters, thereby realizing data-driven automatic and intelligent determination of seeding power adapted to different breakage situations / scenarios. This can effectively overcome the problems of unsatisfactory and inaccurate setting of seeding power caused by human subjective factors, lack of on-site experience and other reasons in manual determination. At the same time, there is no need for a large number of front-line workers to pay attention and operate at all times, which will not affect the production cost, output and production efficiency, thereby improving the accuracy of determining the seeding power, while improving the single crystal production efficiency and reducing the breakage rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] For ease of reference or clarity, firstly, the relevant abbreviations or terms involved in the embodiments of the present application are explained as follows:

[0051] Crystal pulling: It is the front-end process in the photovoltaic field, specifically the process of producing silicon rods through the direct pulling method in a single crystal furnace.

[0052] Seeding: The original 16mm seed crystal is gradually reduced in size and stabilized into a uniform fine crystal with a certain coarseness.

[0053] Seed crystal: A single crystal with a certain crystal orientation, used as a replica sample so that the pulled silicon ingot has the same crystal orientation as the seed crystal.

[0054] GBDT model: GBDT stands for Gradient Boosting Decision Tree. GBDT is a gradient boosting algorithm that uses Taylor series expansion and negative gradient direction to optimize the loss function, and iteratively updates model parameters to approximate the optimal solution of the loss function.

[0055] Equal diameter performance: whether the equal diameter length reaches the expected value.

[0056] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0057] Figure 1 It is a schematic diagram of the construction process of the first seeding power model provided in the present application;

[0058] Figure 2 It is a schematic diagram of the construction process of the second seeding power model provided in the present application;

[0059] Figure 3 It is a flow chart of a method for determining seeding power based on data drive provided in the present application;

[0060] Figure 4 It is a composition structure diagram of a device for determining seeding power based on data drive provided in the present application;

[0061] Figure 5 It is a structural diagram of the electronic device provided in this application. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0063] The embodiments of the present application provide a data-driven method for determining seeding power and related devices, which are used to realize data-driven automated and intelligent prediction of seeding power by constructing a seeding power model for the seeding process of preparing single crystal silicon by the Czochralski method based on a machine learning method, thereby solving the problems of unsatisfactory and inaccurate seeding power setting and high wire breakage rate in the known technology.

[0064] Furthermore, the embodiments of the present application specifically adopt a data-driven method to select parameters in processes such as the seeding process and the shoulder release process from historical data using feature selection or even combining manual experience as model input features, and perform data screening on the selected feature parameters by solving the standard interval range of each selected feature parameter, and finally construct a data set. On this basis, a machine learning gradient boosting decision tree model is used to train the model to obtain a seeding power model, and then the seeding power in the current crystal pulling is predicted by inputting the parameters in the last seeding process, shoulder release process, etc. into the trained seeding power model.

[0065] The data-driven seeding power determination method provided in the embodiment of the present application needs to be based on a pre-constructed seeding power model. The construction process of the seeding power model is first described below.

[0066] The seeding power model constructed in the embodiment of the present application includes a first seeding power model and a second seeding power model. The second seeding power model and the first seeding power model are respectively for different scenarios of whether shoulder breakage occurred in the last crystal pulling in the crystal pulling process. Different intelligent agents trained by data set construction are achieved by selecting features matching the targeted scenarios and solving the standard interval range of the features for the process parameters in the historical crystal pulling process.

[0067] Specifically, in the actual production process on site, each process of the crystal pulling preparation process may experience wire breaks. The embodiment of the present application regards different situations of whether shoulder breakage occurs as different scenarios, and designs and constructs different crystal seeding power models for different scenarios. Among them, if shoulder breakage did not occur in the last crystal pulling process (no shoulder breakage is equivalent to entering the equal diameter process), the first crystal seeding power model is used to make an intelligent prediction of the current crystal seeding power. If shoulder breakage occurred in the last crystal pulling process and there is no record of entering the equal diameter process in this furnace, the second crystal seeding power model is used to make an intelligent prediction of the current crystal seeding power.

[0068] See also Figure 1 The schematic diagram of the construction process of the first seeding power model is shown. The construction process of the first seeding power model can be implemented as the following steps 101 to 105. These processing steps are described in detail below.

[0069] Step 101: Acquire source data of each process in the crystal pulling process from historical data of the crystal pulling process.

[0070] Optionally, the various steps in the crystal pulling process may include seeding, shoulder release, shoulder rotation, and equal diameter.

[0071] Among them, the source data of the seeding process may include but are not limited to: crystal length, crystal diameter, growth pulling speed, heater power, residual material weight, crucible lifting speed, crystal lifting speed, liquid surface temperature, argon gas flow rate, number of re-injections, and absolute liquid mouth distance.

[0072] The source data of the shoulder release process may include but are not limited to: crystal length, crystal diameter, growth pulling speed, heater power, residual material weight, crucible lifting speed, crystal lifting speed, liquid surface temperature, argon flow rate, number of re-injections, absolute liquid mouth distance, shoulder release time, shoulder release cooling amount, and shoulder release opening time.

[0073] The source data of the shoulder rotation process may include but are not limited to: crystal length, crystal diameter, growth pulling speed, heater power, residual material weight, crucible lifting speed, crystal lifting speed, liquid surface temperature, argon flow rate, number of re-injections, absolute liquid mouth distance, and shoulder rotation temperature drop amount.

[0074] The source data of the equal diameter process may include but are not limited to: crystal length, crystal diameter, growth pulling rate, heater power, residual material weight, crucible lifting speed, crystal lifting speed, liquid surface temperature, argon flow rate, number of re-injections, absolute liquid mouth distance, and process running time.

[0075] Step 102: Perform feature selection processing that matches the first crystal pulling scenario on the source data of the various steps, so as to obtain various first features that match the first crystal pulling scenario and satisfy a first correlation condition with the seeding power based on the feature selection processing.

[0076] The first crystal pulling scenario is a scenario in which the equal diameter process is entered without shoulder release and wire breakage occurring in the previous crystal pulling process.

[0077] The feature selection process in this step is intended to select representative features that can improve model performance (i.e., affect seeding power) from the parameter data (source data) of each process obtained for the first crystal pulling scenario. Specifically, the correlation between the obtained process parameters or statistical methods can be used to screen out the parameters that match the first crystal pulling scenario and satisfy the first correlation condition with the seeding power as the first features, so as to reduce the dimension of the parameter data.

[0078] Optionally, step 102 may be further implemented as the following steps 1-1 to 1-3, each of which is described in detail as follows:

[0079] 1-1. Determine the correlation between different process parameters in the source data of each process.

[0080] Optionally, a correlation coefficient matrix between process parameters may be calculated, and the correlation coefficient matrix between process parameters may be used to determine which process parameters have a higher correlation.

[0081] 1-2. Eliminate some parameters from multiple process parameters that meet the preset correlation relationship to obtain various retained parameters.

[0082] Optionally, the preset correlation relationship may be set such that the correlation coefficient between the process parameters is close to 1 or -1, that is, the absolute value of the difference between the correlation coefficient between the process parameters and 1 or -1 is less than a set threshold.

[0083] If the correlation between two or more process parameters is too high to satisfy the correlation relationship, for example, the correlation coefficient is close to 1 or -1, some of the process parameters can be eliminated to obtain the retained parameters to reduce redundancy. For example, if the correlation coefficient between two process parameters is close to 1 or -1, one of the parameters can be eliminated.

[0084] 1-3. Determine the importance of each retention parameter relative to the seeding power using a gradient boosting decision tree model, and select the retention parameters whose corresponding importance meets the importance condition and matches the first crystal pulling scenario as the first features.

[0085] Based on steps 1-2, the machine learning GBDT model can be further used to evaluate the importance of each retention parameter relative to the crystal seeding power, and output the evaluation results of the importance of each retention parameter relative to the crystal seeding power, and then select the corresponding retention parameters whose importance meets the importance conditions and matches the first crystal pulling scenario as the first features.

[0086] The importance condition may be, but is not limited to, set to a threshold value for the importance evaluation value, or a top_k in a descending sequence of importance evaluation values ​​corresponding to each parameter, so that the parameters with low importance relative to the seeding power in each retained parameter are removed by the set importance condition, and feature selection may also be performed in combination with the experience of the on-site process, wherein k is an integer greater than 0.

[0087] The first characteristics finally selected may include but are not limited to: seeding power, average seeding pulling speed, seeding pulling speed in the later stage, seeding time, seeding temperature, argon flow rate, crystal rotation speed, crucible rotation speed, number of re-throws, amount of seeding residue, shoulder release time, shoulder release height, early shoulder release height, middle shoulder release height, shoulder release opening time, argon flow rate, crystal rotation speed, crucible rotation speed, the amount of residue in this crystal pulling, liquid surface temperature, number of re-throws, thermal field size, and single crystal size.

[0088] Step 103, by screening the process data of seeding to equal diameter according to the equal diameter performance, solving the standard interval range of each of the first characteristics; the process data includes the data of each of the first characteristics.

[0089] Since historical data is generated after manual setting on site, subjective factors, lack of experience, etc. of on-site process personnel may cause data errors. To address this situation, the embodiment of the present application first performs parameter processing on the selected first feature before producing the data set to determine the standard range of the parameter value of each first feature.

[0090] Specifically, for each first feature, the process data corresponding to each first feature from each seeding to equal diameter can be first screened according to the equal-diameter performance (that is, the process data includes the data of each first feature), and then the screened process data from seeding to equal diameter can be statistically analyzed. Based on the statistical information of the data in the screened process data from seeding to equal diameter, the standard interval of the first feature can be solved. Specifically, the value interval range of the first feature that meets the preset statistical characteristics can be used as the standard interval range of the parameter value of the first feature.

[0091] It is worth noting that since the historical data and the solved standard data ranges under different bases and different equipment may not be consistent, based on this, the embodiment of the present application does not give specific standard interval ranges for each first feature (i.e., the process parameters selected as the first feature). In actual applications, as long as the above-mentioned technical concepts of the embodiment of the present application are adopted to determine the standard interval range of the characteristic parameter value, it falls within the protection scope of the embodiment of the present application.

[0092] Step 104 , selecting data for each of the first features according to the standard interval range of each of the first features, and combining the selected feature data with the first seeding power corresponding to the feature data in the historical data to obtain a first data set.

[0093] On the basis of step 103, this step can specifically filter the data of each first feature according to the standard interval range of each solved first feature. First, assuming that the seeding in front and behind each two adjacent seeding processes in the historical data are respectively called the last seeding and the current seeding in the historical data, then the record of the last seeding in the historical data that did not have shoulder breakage and entered the equal diameter process and the record of the equal diameter performance in the current seeding record that meets the expectations can be specifically filtered. Secondly, the seeding power of this time, the average pulling speed of the seeding of this time, the pulling speed of the late seeding of this time, the shoulder release time of this time, the shoulder release height and other parameter data are filtered according to the standard interval range of each first feature. Finally, the selected last feature data, the feature data of this time after filtering according to the standard interval range, and the seeding power of this seeding are combined to form a data set for model training.

[0094] The last time and this time here refer to the last seeding and this seeding respectively.

[0095] In this step, the concept of feature data screening is that no matter whether the parameters of the last seeding and shoulder release process are within the standard range, as long as the data of this seeding and shoulder release process are within the standard range, the goal of model learning is to learn and predict the seeding power according to different previous situations, so that the seeding process and subsequent process data under this seeding power are within the standard range.

[0096] Step 105: Perform machine learning model training based on the first data set to obtain the first seeding power model.

[0097] Specifically, the first data set can be used to train multiple decision trees in a gradient boosting decision tree model, and each decision tree improves model performance by reducing the residual of the previous decision tree until the termination condition is met to obtain a first seeding power model.

[0098] That is, in the embodiment of the present application, when performing model training based on the constructed first data set to obtain the first seeding power model, a machine learning gradient boosting decision tree model (GBDT) is specifically used for model training. After research and experimental verification by the applicant, the GBDT model, compared with other machine learning models, can better capture the complex relationship between a large amount of different parameter data involved in different crystal pulling processes and multiple processes in the direct-pull single crystal by gradually training multiple decision trees and making each tree reduce the residual of the previous tree. Accordingly, compared with other machine learning models, the embodiment of the present application can achieve better performance in seeding power prediction and realize accurate prediction of seeding power by using a machine learning gradient boosting decision tree model (GBDT) for model training.

[0099] Among them, the end condition can be but is not limited to being set to: the model performance of the first seeding power model reaches the set performance, or the number of iterations of the model training reaches the set number, or the model training duration reaches the set duration.

[0100] By constructing a first crystal seeding power model, this embodiment can support subsequent data-driven automated and intelligent prediction of the crystal seeding power for the first crystal pulling scenario in which no shoulder breakage occurred in the previous crystal pulling process, so as to overcome the problem of unsatisfactory and inaccurate setting of the crystal seeding power due to human subjective factors, lack of on-site experience and other reasons in the manual determination method, while improving the crystal pulling production efficiency and reducing the breakage rate.

[0101] See also Figure 2 The schematic diagram of the construction process of the second seeding power model is shown. The construction process of the second seeding power model can be implemented as the following steps 201 to 205. These processing steps are described in detail below.

[0102] Step 201: Acquire source data of each process in the crystal pulling process from historical data of the crystal pulling process.

[0103] Specifically, the source data of crystal seeding, shoulder release, shoulder rotation, and equal diameter processes in the crystal pulling preparation process can be obtained from historical data.

[0104] The source data obtained for each process can be specifically referred to the source data obtained for each process when the first seeding power model is constructed in the above text, which will not be repeated here.

[0105] Step 202: Perform feature selection processing that matches the second crystal pulling scenario on the source data of the various steps, so as to obtain various second features that match the second crystal pulling scenario and satisfy a second correlation condition with the seeding power based on the feature selection processing.

[0106] The second crystal pulling scenario is a scenario in which shoulder release and wire breakage occurred in the previous crystal pulling process and there is no record of entering the equal diameter process in this furnace.

[0107] The feature selection process of this step is intended to select representative features that can improve model performance (i.e., affect seeding power) from the parameter data (source data) of each process obtained for the second crystal pulling scenario. Specifically, the parameters that match the second crystal pulling scenario and satisfy the second correlation condition with the seeding power can be screened out as the second features through the correlation between the obtained process parameters or statistical methods to reduce the dimension of the parameter data.

[0108] Similar to the feature selection step in the process of building the first seeding power model, this step can specifically determine the correlation between different process parameters in the source data of each process, eliminate some parameters of multiple process parameters that meet the preset correlation relationship, obtain various retained parameters, and use the gradient boosting decision tree model to determine the importance of each retained parameter relative to the seeding power, and select the corresponding importance that meets the importance condition and matches the second crystal pulling scenario. Each retained parameter is selected as the second feature. At the same time, feature selection can also be combined with the experience of on-site technology.

[0109] The second characteristics finally selected may include but are not limited to: seeding power, average seeding pulling speed, seeding pulling speed in the later stage, seeding time, seeding temperature, argon flow rate, crystal rotation speed, crucible rotation speed, number of re-throws, amount of seeding residue, amount of residue in this crystal pulling, liquid surface temperature, number of re-throws, thermal field size, and single crystal size.

[0110] Compared with each of the first characteristics, each of the second characteristics does not include process parameters of the shoulder release process such as the shoulder release time length and the shoulder release height.

[0111] Step 203, by screening the process data of seeding to equal diameter according to the equal diameter performance, solving the standard interval range of each second characteristic; the process data includes the data of each second characteristic.

[0112] This step can specifically target each second feature, screen the process data from each seeding to equal diameter corresponding to each second feature according to the equal-diameter performance (that is, the process data includes the data of each second feature), perform information statistics on the screened process data from seeding to equal diameter, and solve the standard interval of the second feature based on the statistical information of the data in the screened process data from seeding to equal diameter. Specifically, the value interval range of the second feature that meets the preset statistical characteristics can be used as the standard interval range of the parameter value of the second feature.

[0113] Step 204: select data for each of the second characteristics according to the standard interval range of each of the second characteristics, and combine the selected characteristic data with the second seeding power corresponding to the characteristic data in the historical data to obtain a second data set.

[0114] On the basis of step 203, this step can specifically filter the data of each second feature according to the standard interval range of each solved second feature. First, it is assumed that the seeding in front and behind each two adjacent seeding processes in the historical data are respectively called the last seeding and the current seeding in the historical data. Specifically, the records of the last seeding in the historical data that the shoulder was broken and did not enter the equal diameter process and the records of the current seeding that the equal diameter performance met expectations can be filtered. Secondly, the parameter data of the characteristics such as the seeding power of this time, the average pulling speed of the seeding of this time, and the average pulling speed of the late stage of the seeding of this time are filtered according to the standard interval range solved for each second feature. Finally, the selected last feature data, the feature data of this time after filtering according to the standard interval range, and the seeding power of this seeding are combined to form a data set for model training.

[0115] The concept of the above feature data screening is that no matter whether the feature parameters of the last seeding and shoulder release process are within the standard range, as long as the data of this seeding and shoulder release process are within the standard range, the goal of model learning is to learn and predict the seeding power according to different previous situations, so that the seeding process and subsequent process data under this seeding power are within the standard range.

[0116] Step 205: Perform machine learning model training based on the second data set to obtain the second seeding power model.

[0117] Similarly, this step can specifically use the second data set to train multiple decision trees in the gradient boosting decision tree model, each decision tree improves the model performance by reducing the residual of the previous decision tree until the termination condition is met to obtain the second seeding power model.

[0118] That is, when performing model training based on the constructed second data set to obtain the second seeding power model, the present application also uses a machine learning gradient boosting decision tree model (GBDT) for model training, based on the GBDT model, by gradually training multiple decision trees, and making each tree better capture the complex relationship between a large amount of different parameter data involved in different crystal pulling processes and multiple processes in the direct-pull single crystal by reducing the residual of the previous tree, so as to achieve better performance in seeding power prediction and realize accurate prediction of seeding power.

[0119] The termination condition here can be, but is not limited to, set to: the model performance of the second seeding power model reaches the set performance, or the number of iterations of the model training reaches the set number, or the model training duration reaches the set duration.

[0120] By constructing a second crystal seeding power model, this embodiment can support the subsequent second crystal pulling scenario in which shoulder release and wire breakage occurred in the previous crystal pulling process and there is no record of entering the equal diameter process in the current furnace. The crystal seeding power can be determined automatically and intelligently based on data-driven, so as to overcome the problems of unsatisfactory and inaccurate setting of crystal seeding power caused by human subjective factors, lack of on-site experience and other reasons in the manual determination method, while improving the crystal pulling production efficiency and reducing the wire breakage rate.

[0121] Based on the constructed first seeding power model and second seeding power model, see Figure 3 The data-driven seeding power determination method flow chart shown in the embodiment of the present application may include the following processing steps:

[0122] Step 301, obtaining the data of the last crystal pulling process.

[0123] Specifically, the parameter data of each process of the above-mentioned crystal pulling process can be obtained.

[0124] Step 302: Determine whether shoulder breakage occurred in the last crystal pulling process based on the data of the last crystal pulling process.

[0125] Specifically, the scenario to which the seeding power prediction for this crystal pulling belongs can be determined based on the acquired data of the last crystal pulling process by determining whether shoulder breaking occurred in the last crystal pulling process. If it is determined that the shoulder breaking did not occur in the last crystal pulling process and the equal diameter process was entered, the scenario to which the seeding power prediction for this crystal pulling belongs is the first crystal pulling scenario; if it is determined that shoulder breaking occurred in the last crystal pulling process and there is no record of entering the equal diameter process in this furnace, the scenario to which the seeding power prediction for this crystal pulling belongs is the second crystal pulling scenario.

[0126] Step 303: If shoulder release and wire breakage did not occur in the last crystal pulling process, obtain the process parameters of the crystal seeding process, the process parameters of the shoulder release process and the field parameters of this crystal pulling process in the last crystal pulling process, and use the first crystal seeding power model to determine the crystal seeding power of this crystal pulling process based on the process parameters of the crystal seeding process, the process parameters of the shoulder release process and the field parameters of this crystal pulling process.

[0127] If the shoulder release and line breakage did not occur in the last crystal pulling process, that is, the scenario to which the seeding power prediction of this crystal pulling belongs is the first crystal pulling scenario, in this case, model input data matching the first crystal pulling scenario can be obtained. The obtained model input data includes process parameters of the seeding process and the shoulder release process in the last crystal pulling process and field parameters of this crystal pulling.

[0128] Among them, the process parameters of the seeding step in the last crystal pulling process include but are not limited to at least some of the parameters in the seeding power, average seeding pulling speed, late seeding pulling speed, seeding time, seeding temperature, argon flow rate, crystal rotation speed, crucible rotation speed, number of re-throws and amount of residual seeding in the last crystal pulling process. The process parameters of the shoulder release process in the last crystal pulling process include but are not limited to at least some of the parameters in the last shoulder release time, shoulder release height, early shoulder release height, middle shoulder release height, shoulder release opening time, argon flow rate, crystal rotation speed and crucible rotation speed. The on-site parameters of this crystal pulling include but are not limited to at least some of the parameters in the amount of residual material, liquid surface temperature, number of re-throws, thermal field size, and single crystal size of this crystal pulling.

[0129] Afterwards, the acquired parameter data, namely the process parameters of the crystal seeding process in the last crystal pulling process, the process parameters of the shoulder release process and the on-site parameters of this crystal pulling are input into the first crystal seeding power model, so as to obtain the crystal seeding power of this crystal pulling predicted and output by the first crystal seeding power model based on the input parameters.

[0130] Step 304: if shoulder breakage occurs in the last crystal pulling process and there is no record of entering the equal diameter process in the current furnace, obtain the process parameters of the seeding process in the last crystal pulling process and the field parameters of the current crystal pulling process, and use the second seeding power model to determine the seeding power of the current crystal pulling process according to the process parameters of the seeding process in the last crystal pulling process and the field parameters of the current crystal pulling process;

[0131] Among them, the second crystal seeding power model and the first crystal seeding power model are respectively aimed at different scenarios of whether shoulder breakage occurred in the last crystal pulling process in the crystal pulling preparation process. Different intelligent agents trained by data set construction are constructed by selecting features matching the targeted scenarios and solving the standard interval range of features for the process parameters in the historical crystal pulling process.

[0132] If the shoulder breakage occurred in the last crystal pulling process and there is no record of entering the equal diameter process in this furnace, that is, the scenario to which the seeding power prediction of this crystal pulling belongs is the second crystal pulling scenario, in this case, the model input data matching the second crystal pulling scenario can be obtained. The obtained model input data includes the process parameters of the seeding process in the last crystal pulling process and the field parameters of this crystal pulling.

[0133] Among them, the process parameters of the crystal seeding step in the last crystal pulling process and the on-site parameters of this crystal pulling process can be found in the above description and will not be repeated here.

[0134] Afterwards, the acquired parameters, namely the process parameters of the crystal seeding process in the last crystal pulling process and the field parameters of this crystal pulling, are input into the second crystal seeding power model, so as to obtain the crystal seeding power of this crystal pulling predicted and output by the second crystal seeding power model based on the input parameters.

[0135] According to the above scheme, the data-driven method for determining the crystal seeding power provided in the present application targets different scenarios of whether shoulder breakage occurred in the last crystal pulling process during the crystal pulling preparation process. By selecting features that match the targeted scenarios and solving the standard interval range of the features for the process parameters in the historical crystal pulling process, a data set is constructed and different intelligent agents are trained to obtain the first crystal seeding power model and the second crystal seeding power model that are respectively adapted to different scenarios. On this basis, by obtaining the data of the last crystal pulling process, it is determined whether the shoulder break occurred in the last crystal pulling process according to the acquired data, and the seeding power model adapted to the current actual scenario (whether the shoulder break actually occurred or not) is called according to the judgment result. The seeding power of this crystal pulling is determined based on the called model based on the adapted model input parameters, thereby realizing data-driven automatic and intelligent determination of seeding power adapted to different breakage situations / scenarios. This can effectively overcome the problems of unsatisfactory and inaccurate setting of seeding power caused by human subjective factors, lack of on-site experience and other reasons in manual determination. At the same time, there is no need for a large number of front-line workers to pay attention and operate at all times, which will not affect the production cost, output and production efficiency, thereby improving the accuracy of determining the seeding power, while improving the single crystal production efficiency and reducing the breakage rate.

[0136] Corresponding to the above method, the embodiment of the present application also provides a device for determining seeding power based on data drive, see Figure 4 The schematic diagram of the composition structure of the device comprises:

[0137] An acquisition module 401 is used to acquire the data of the last crystal pulling process;

[0138] A first determination module 402 is used to determine whether shoulder breakage occurs in the last crystal pulling process according to the last crystal pulling process data;

[0139] The second determination module 403 is used to obtain the process parameters of the seeding process and the shoulder release process in the last crystal pulling process and the field parameters of the current crystal pulling process when no shoulder release breakage occurs in the last crystal pulling process, and determine the seeding power of the current crystal pulling process according to the process parameters of the seeding process and the shoulder release process in the last crystal pulling process and the field parameters of the current crystal pulling process using the first seeding power model;

[0140] The third determination module 404 is used to obtain the process parameters of the seeding process in the previous crystal pulling process and the field parameters of the current crystal pulling process when the shoulder release line is broken in the previous crystal pulling process and there is no record of entering the equal diameter process in the current furnace, and use the second seeding power model to determine the seeding power of the current crystal pulling process according to the process parameters of the seeding process in the previous crystal pulling process and the field parameters of the current crystal pulling process;

[0141] Among them, the second crystal seeding power model and the first crystal seeding power model are respectively aimed at different scenarios of whether shoulder breakage occurred in the last crystal pulling process in the crystal pulling preparation process. Different intelligent agents trained by data set construction are constructed by selecting features matching the targeted scenarios and solving the standard interval range of features for the process parameters in the historical crystal pulling process.

[0142] In an optional embodiment, the process parameters of the seeding process include at least some of the parameters of the seeding power, average seeding pulling speed, late seeding pulling speed, seeding time, seeding temperature, argon flow rate, crystal rotation speed, crucible rotation speed, number of re-throws and seeding residual material amount in the last crystal pulling process;

[0143] The process parameters of the shoulder release process include at least some of the parameters of the last shoulder release time, shoulder release height, shoulder release early height, shoulder release middle height, shoulder release opening time, argon gas flow rate, crystal rotation speed and crucible rotation speed;

[0144] The field parameters of this crystal pulling include at least some parameters of the remaining material amount, liquid surface temperature, number of re-throws, thermal field size, and single crystal size.

[0145] In an optional embodiment, the above-mentioned device further includes a model building module for building a seeding power model; when building the first seeding power model, the model building module is specifically used to:

[0146] Obtaining source data of each process in the crystal pulling preparation process from historical data of the crystal pulling preparation process;

[0147] Performing feature selection processing that matches a first crystal pulling scenario on source data of each process, so as to obtain first features that match the first crystal pulling scenario and satisfy a first correlation condition with the seeding power based on the feature selection processing; the first crystal pulling scenario is a scenario in which no shoulder breakage occurs in the last crystal pulling process;

[0148] By screening the process data of seeding to equal diameter according to the equal diameter performance, solving the standard interval range of each of the first characteristics; the process data includes the data of each of the first characteristics;

[0149] Selecting data for each of the first features according to a standard interval range of each of the first features, and combining the selected feature data with a first seeding power corresponding to the feature data in historical data to obtain a first data set;

[0150] A machine learning model is trained based on the first data set to obtain the first seeding power model.

[0151] In an optional implementation manner, when the model building module performs feature selection processing that matches the first crystal pulling scenario on the source data of each process, so as to obtain each first feature that matches the first crystal pulling scenario and satisfies the first correlation condition with the seeding power based on the feature selection processing, the model building module is specifically used to:

[0152] Determining the correlation between different process parameters in the source data of each process;

[0153] Eliminate some parameters from multiple process parameters that meet the preset correlation relationship to obtain various retained parameters;

[0154] The importance of each retention parameter relative to the seeding power is determined using a gradient boosting decision tree model, and the retention parameters whose corresponding importance meets the importance condition and matches the first crystal pulling scenario are selected as the first features.

[0155] In an optional implementation manner, when constructing the second seeding power model, the model building module is specifically used to:

[0156] Obtaining source data of each process in the crystal pulling preparation process from historical data of the crystal pulling preparation process;

[0157] Performing feature selection processing that matches a second crystal pulling scenario on the source data of each process, so as to obtain each second feature that matches the second crystal pulling scenario and satisfies a second correlation condition with the seeding power based on the feature selection processing; the second crystal pulling scenario is a scenario in which shouldering and wire breaking occurs in the last crystal pulling process and there is no record of entering the equal diameter process in the current heat;

[0158] By screening the process data of seeding to equal diameter according to the equal diameter performance, solving the standard interval range of each second feature; the process data includes the data of each second feature;

[0159] Selecting data for each of the second characteristics according to the standard interval range of each of the second characteristics, and combining the selected characteristic data with the second seeding power corresponding to the characteristic data in the historical data to obtain a second data set;

[0160] A machine learning model is trained based on the second data set to obtain the second seeding power model.

[0161] In an optional implementation manner, when the model building module performs feature selection processing that matches the second crystal pulling scenario on the source data of each process, so as to obtain each second feature that matches the second crystal pulling scenario and satisfies the second correlation condition with the seeding power based on the feature selection processing, the model building module is specifically used to:

[0162] Determining the correlation between different process parameters in the source data of each process;

[0163] Eliminate some parameters from multiple process parameters that meet the preset correlation relationship to obtain various retained parameters;

[0164] The importance of each retention parameter relative to the seeding power is determined using a gradient boosting decision tree model, and the retention parameters whose corresponding importance meets the importance condition and matches the second crystal pulling scenario are selected as the second features.

[0165] In an optional embodiment, when the model building module selects the process data of seeding to equal diameter according to the equal diameter performance and solves the standard interval range of the corresponding feature, it is specifically used to:

[0166] Filter the process data of each seeding to equal diameter according to the equal diameter performance;

[0167] Based on the statistical information of the screened process data from seeding to equal diameter, the standard interval of the corresponding feature is solved;

[0168] The corresponding feature is the corresponding first feature or second feature corresponding to the process data.

[0169] In an optional implementation, when the model building module performs machine learning model training based on the corresponding data set, it is specifically used to:

[0170] Using the corresponding data set to train multiple decision trees in the gradient boosting decision tree model, each decision tree improves the model performance by reducing the residual of the previous decision tree until the end condition is met, and the corresponding seeding power model is obtained;

[0171] Wherein, the corresponding data set is the first data set, and the corresponding seeding power model is the first seeding power model; or, the corresponding data set is the second data set, and the corresponding seeding power model is the second seeding power model.

[0172] With regard to the data-driven seeding power determination device disclosed in the embodiments of the present application, since it corresponds to the data-driven seeding power determination method disclosed in the above method embodiments, the description is relatively simple. For relevant similarities, please refer to the description of the above method embodiments, which will not be described in detail here.

[0173] The present application also discloses an electronic device, the composition structure of the electronic device, such as Figure 5 As shown, including at least:

[0174] A memory 10, for storing a computer instruction set;

[0175] A set of computer instructions can be implemented in the form of a computer program.

[0176] The processor 20 is used to implement the data-driven seeding power determination method provided in any of the above method embodiments by executing a computer instruction set.

[0177] The processor 20 can be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a neural network processor (NPU), a deep learning processor (DPU) or other programmable logic devices, etc.

[0178] The electronic device includes a display device and / or includes a display interface and can be connected to an external display device.

[0179] In addition, the electronic device may also include components such as a communication interface and a communication bus. The memory, the processor and the communication interface communicate with each other through the communication bus.

[0180] The communication interface is used for communication between electronic devices and other devices. The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0181] In addition, the present application also provides a computer-readable medium having a computer program stored thereon, wherein the computer program includes a program code for executing the data-driven seeding power determination method disclosed in any of the method embodiments above.

[0182] In the context of the present application, a computer readable medium (machine readable medium) may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine readable medium may be a machine readable signal medium or a machine readable storage medium. A machine readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0183] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0184] For the convenience of description, the above system or device is described by dividing it into various modules or units according to its functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0185] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0186] Finally, it should be noted that, in this article, relational terms such as first, second, third and target are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0187] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for determining seeding power based on data drive, characterized in that: include: Get the data of the last crystal pulling process; Determine whether shoulder breakage occurs in the last crystal pulling process according to the data of the last crystal pulling process; If the shoulder release line breakage did not occur in the last crystal pulling process, obtain the process parameters of the seeding process, the process parameters of the shoulder release process and the field parameters of this crystal pulling process in the last crystal pulling process, and use the first seeding power model according to the process parameters of the seeding process, the process parameters of the shoulder release process and the field parameters of this crystal pulling process to determine the seeding power of this crystal pulling process; If shoulder breakage occurred in the last crystal pulling process and there is no record of entering the equal diameter process in this furnace, obtain the process parameters of the seeding process in the last crystal pulling process and the field parameters of this crystal pulling, and use the second seeding power model to determine the seeding power of this crystal pulling according to the process parameters of the seeding process in the last crystal pulling process and the field parameters of this crystal pulling; Among them, the second crystal seeding power model and the first crystal seeding power model are respectively aimed at different scenarios of whether shoulder breakage occurred in the last crystal pulling process in the crystal pulling preparation process. Different intelligent agents trained by data set construction are constructed by selecting features matching the targeted scenarios and solving the standard interval range of features for the process parameters in the historical crystal pulling process.

2. The method for determining seeding power based on data drive according to claim 1, characterized in that: The process parameters of the seeding process include at least some of the parameters of the seeding power, average seeding pulling speed, late seeding pulling speed, seeding time, seeding temperature, argon flow rate, crystal rotation speed, crucible rotation speed, number of re-throws and seeding residual material amount in the last crystal pulling process; The process parameters of the shoulder release process include at least some of the parameters of the last shoulder release time, shoulder release height, shoulder release early height, shoulder release middle height, shoulder release opening time, argon gas flow rate, crystal rotation speed and crucible rotation speed; The field parameters of this crystal pulling include at least some parameters of the remaining material amount, liquid surface temperature, number of re-throws, thermal field size, and single crystal size.

3. The method for determining seeding power based on data drive according to claim 1, characterized in that: The process of constructing the first seeding power model includes: Obtaining source data of each process in the crystal pulling preparation process from historical data of the crystal pulling preparation process; Performing feature selection processing that matches a first crystal pulling scenario on source data of each process, so as to obtain first features that match the first crystal pulling scenario and satisfy a first correlation condition with the seeding power based on the feature selection processing; the first crystal pulling scenario is a scenario in which no shoulder breakage occurs in the last crystal pulling process; By screening the process data of seeding to equal diameter according to the equal diameter performance, solving the standard interval range of each of the first characteristics; the process data includes the data of each of the first characteristics; Selecting data for each of the first features according to a standard interval range of each of the first features, and combining the selected feature data with a first seeding power corresponding to the feature data in historical data to obtain a first data set; A machine learning model is trained based on the first data set to obtain the first seeding power model.

4. The method for determining seeding power based on data drive according to claim 3, characterized in that: The performing of feature selection processing on the source data of each process that matches the first crystal pulling scenario, so as to obtain first features that match the first crystal pulling scenario and satisfy a first correlation condition with the seeding power based on the feature selection processing, comprises: Determining the correlation between different process parameters in the source data of each process; Eliminate some parameters from multiple process parameters that meet the preset correlation relationship to obtain various retained parameters; The importance of each retention parameter relative to the seeding power is determined using a gradient boosting decision tree model, and the retention parameters whose corresponding importance meets the importance condition and matches the first crystal pulling scenario are selected as the first features.

5. The method for determining seeding power based on data drive according to claim 1, characterized in that: The construction process of the second seeding power model includes: Obtaining source data of each process in the crystal pulling preparation process from historical data of the crystal pulling preparation process; Performing feature selection processing that matches a second crystal pulling scenario on the source data of each process, so as to obtain each second feature that matches the second crystal pulling scenario and satisfies a second correlation condition with the seeding power based on the feature selection processing; the second crystal pulling scenario is a scenario in which shouldering and wire breaking occurs in the last crystal pulling process and there is no record of entering the equal diameter process in the current heat; By screening the process data of seeding to equal diameter according to the equal diameter performance, solving the standard interval range of each second feature; the process data includes the data of each second feature; Selecting data for each of the second characteristics according to the standard interval range of each of the second characteristics, and combining the selected characteristic data with the second seeding power corresponding to the characteristic data in the historical data to obtain a second data set; A machine learning model is trained based on the second data set to obtain the second seeding power model.

6. The method for determining seeding power based on data drive according to claim 5, characterized in that: The performing of feature selection processing on the source data of each process that matches the second crystal pulling scenario, so as to obtain each second feature that matches the second crystal pulling scenario and satisfies a second correlation condition with the seeding power based on the feature selection processing, comprises: Determining the correlation between different process parameters in the source data of each process; Eliminate some parameters from multiple process parameters that meet the preset correlation relationship to obtain various retained parameters; The importance of each retention parameter relative to the seeding power is determined using a gradient boosting decision tree model, and the retention parameters whose corresponding importance meets the importance condition and matches the second crystal pulling scenario are selected as the second features.

7. The method for determining seeding power based on data drive according to claim 3 or 5, characterized in that: By screening the process data from seeding to equal diameter according to the equal diameter performance, the standard range of the corresponding characteristics is solved, including: Filter the process data of each seeding to equal diameter according to the equal diameter performance; Based on the statistical information of the screened process data from seeding to equal diameter, the standard interval of the corresponding feature is solved; The corresponding feature is the corresponding first feature or second feature corresponding to the process data.

8. The method for determining seeding power based on data drive according to claim 3 or 5, characterized in that: Perform machine learning model training based on the corresponding data set, including: Using the corresponding data set to train multiple decision trees in the gradient boosting decision tree model, each decision tree improves the model performance by reducing the residual of the previous decision tree until the end condition is met, and the corresponding seeding power model is obtained; Wherein, the corresponding data set is the first data set, and the corresponding seeding power model is the first seeding power model; or, the corresponding data set is the second data set, and the corresponding seeding power model is the second seeding power model.

9. A device for determining seeding power based on data drive, characterized in that: include: An acquisition module is used to obtain the data of the last crystal pulling process; A first determination module is used to determine whether shoulder breakage occurs in the last crystal pulling process according to the last crystal pulling process data; The second determination module is used to obtain the process parameters of the seeding process, the process parameters of the shoulder release process and the field parameters of the current crystal pulling process in the last crystal pulling process when no shoulder release breakage occurred in the last crystal pulling process, and determine the seeding power of the current crystal pulling according to the process parameters of the seeding process, the process parameters of the shoulder release process and the field parameters of the current crystal pulling process in the last crystal pulling process using the first seeding power model; The third determination module is used to obtain the process parameters of the seeding process in the last crystal pulling process and the field parameters of the current crystal pulling process when the shoulder release line breaks in the last crystal pulling process and there is no record of entering the equal diameter process in the current furnace, and use the second seeding power model to determine the seeding power of the current crystal pulling process according to the process parameters of the seeding process in the last crystal pulling process and the field parameters of the current crystal pulling process; Among them, the second crystal seeding power model and the first crystal seeding power model are respectively aimed at different scenarios of whether shoulder breakage occurred in the last crystal pulling process in the crystal pulling preparation process. Different intelligent agents trained by data set construction are constructed by selecting features matching the targeted scenarios and solving the standard interval range of features for the process parameters in the historical crystal pulling process.

10. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it can be used to implement the data-driven seeding power determination method as described in any one of claims 1 to 8.

Images