Shoulder release control method, device, computer equipment and storage medium

Abstract

The present application relates to a shoulder release control method, device, computer equipment and storage medium. The method comprises: obtaining a predicted diameter value based on time series monitoring data within a preset time range; inputting a target shoulder release angle and the predicted diameter value into a long short-term memory mechanism model to obtain a target pulling speed value; and controlling a target device to release the shoulder based on the target pulling speed value. The method can improve the shoulder release survival rate.

Description

Technical Field

[0001] The present application relates to the technical field of crystal preparation, and in particular to a shoulder release control method, device, computer equipment and storage medium. Background Art

[0002] The shoulder release survival rate can be defined as the percentage of crystals that have successfully released their shoulders and entered the equal diameter stage among all the crystals that entered the shoulder release stage within a cycle. In the existing crystal preparation field, the pulling speed control in the shoulder release stage is usually achieved by manual shoulder release and PID controlled shoulder release. However, manual shoulder release is time-consuming and labor-intensive, and the process standards vary among operators. In the long-term operation, the difference in operation is magnified, resulting in disconnection. PID speed shoulder release is prone to control overshoot or undershoot due to the time lag between the pulling speed and the control target during the shoulder release process, which leads to disconnection in the shoulder release process.

[0003] Therefore, the current shoulder release control technology still has the problem of low shoulder release survival rate. Summary of the invention

[0004] Based on this, it is necessary to provide a shoulder release control method, device, computer equipment and storage medium that can improve the shoulder release survival rate in response to the above technical problems.

[0005] In a first aspect, the present application provides a shoulder release control method, the shoulder release control method comprising:

[0006] Based on the time series monitoring data within a preset time range, a predicted diameter value is obtained;

[0007] Inputting the target shoulder angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target pulling speed value; the construction process of the long short-term memory mechanism model includes: constructing a short-term loss index based on a first storage step; constructing a long-term loss index based on a second storage step; the first storage step is smaller than the second storage step; constructing the long short-term memory mechanism model based on the short-term loss index and the long-term loss index;

[0008] Based on the target pulling speed value, the target device is controlled to release the shoulder.

[0009] In one embodiment, obtaining the predicted diameter value based on the time series monitoring data within a preset time range includes:

[0010] Based on the time series monitoring data within a preset time range and a pre-trained diameter prediction model, a predicted diameter value is obtained; the diameter prediction model is trained by multiple historically collected complete monitoring data of the same type of equipment as the current equipment during the shoulder release stage.

[0011] In one embodiment, obtaining the predicted diameter value based on the time series monitoring data within a preset time range and a pre-trained diameter prediction model includes:

[0012] Acquire historically collected complete monitoring data of the same type of equipment as the current equipment during the shoulder release stage; the complete monitoring data includes measured diameter data and shoulder release stage data; the shoulder release stage data includes at least one of crystal state parameters, environmental state parameters and operating condition parameters;

[0013] Decomposing the shoulder release phase data into trend component data and residual component data;

[0014] Based on the trend component data, the residual component data and the measured diameter data, the linear layer is trained to obtain the diameter prediction model.

[0015] In one embodiment, before inputting the target shouldering angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target pulling speed value, the method further includes:

[0016] Obtaining process condition parameters of the current equipment; the process condition parameters include at least one of equipment parameters, process parameters and material supply parameters of the current equipment;

[0017] Perform thermal stress simulation based on the process condition parameters to obtain a first shoulder release angle; the first shoulder release angle is a shoulder angle with minimum thermal stress under the process condition parameters;

[0018] Based on the first shoulder release angle, a target shoulder release angle is determined.

[0019] In one embodiment, performing thermal stress simulation based on the process condition parameters to obtain the first shoulder angle includes:

[0020] Based on the equipment parameters, process parameters and material supply parameters of the current equipment, construct an equipment model corresponding to the equipment parameters;

[0021] Calculating thermal stresses for multiple shoulder angles based on the device model, elastic modulus and thermal expansion coefficient of the target material;

[0022] The shoulder angle with the smallest thermal stress is taken as the first shoulder angle.

[0023] In one embodiment, before inputting the target shouldering angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target pulling speed value, the method further includes:

[0024] Acquire complete monitoring data of the same type of equipment as the current equipment collected historically during the shoulder release phase;

[0025] Based on the complete monitoring data, a target shoulder angle is determined; the target shoulder angle is the shoulder angle with the highest survival rate in the complete monitoring data.

[0026] In one embodiment, the long short-term memory mechanism model includes a long short-term memory loss function, and the target shoulder angle and the predicted diameter value are input into the long short-term memory mechanism model to obtain the target pulling speed value, including:

[0027] Inputting the target shoulder angle and the predicted diameter value into the long short-term memory loss function to obtain a predicted loss value;

[0028] If the predicted loss value is less than the preset loss threshold, the target pulling speed value is determined based on the historical pulling speed control value at the previous moment and the target shoulder angle.

[0029] In a second aspect, the present application provides a shoulder release control device, the shoulder release control device comprising:

[0030] A diameter prediction module, used to obtain a predicted diameter value based on time series monitoring data within a preset time range;

[0031] A pulling speed determination module is used to input the target shoulder angle and the predicted diameter value into the long short-term memory mechanism model to obtain a target pulling speed value; the construction process of the long short-term memory mechanism model includes: constructing a short-term loss index based on a first storage step; constructing a long-term loss index based on a second storage step; the first storage step is smaller than the second storage step; constructing the long short-term memory mechanism model based on the short-term loss index and the long-term loss index;

[0032] The shoulder release control module is used to control the target device to release the shoulder based on the target pulling speed value.

[0033] In a third aspect, the present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned method when executing the computer program.

[0034] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the method described above when executed by a processor.

[0035] The shoulder release control method, device, computer equipment and storage medium obtain a predicted diameter value based on time series monitoring data within a preset time range; simulate based on the process condition parameters of the current equipment to obtain a target shoulder release angle; input the target shoulder release angle and the predicted diameter value into the long short-term memory mechanism model to obtain a target pulling speed value; the construction process of the long short-term memory mechanism model includes: constructing a short-term loss index based on a first storage step; constructing a long-term loss index based on a second storage step; the first storage step is smaller than the second storage step; constructing the long short-term memory mechanism model based on the short-term loss index and the long-term loss index; based on the target pulling speed value, controlling the target equipment to perform shoulder release, and being able to determine the optimal shoulder release angle of the current equipment based on historical time series monitoring data, and determine the optimal pulling speed value in real time in combination with the long short-term memory mechanism model, thereby solving the time lag, uncertainty and nonlinearity of pulling speed control during the shoulder release process, and achieving the effect of improving the survival rate of shoulder release. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A diagram showing an application environment of a shoulder release control method in one embodiment;

[0037] Figure 2 A schematic flow chart of a shoulder release control method in one embodiment;

[0038] Figure 3 is a flow chart of a shoulder release control method according to another embodiment;

[0039] Figure 4 A comparison diagram of the predicted diameter and the actual diameter of the shoulder in one embodiment;

[0040] Figure 5 A comparative shoulder diagram of a comparative example and an embodiment in one embodiment;

[0041] Figure 6 A schematic diagram of a furnace for a shoulder placement process in one embodiment;

[0042] Figure 7 is a structural block diagram of a shoulder release control device in one embodiment;

[0043] Figure 8 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] The shoulder release control method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The terminal 102 can obtain the time series monitoring data and the process condition parameters of the current equipment within a preset time range by communicating with the server 104, and obtain the predicted diameter value based on the time series monitoring data within the preset time range; simulate based on the process condition parameters of the current equipment to obtain the target shoulder release angle; input the target shoulder release angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target pulling speed value; based on the target pulling speed value, control the target equipment to release the shoulder. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers and Internet of Things devices. The server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0046] In one embodiment, Figure 2 As shown, a shoulder release control method is provided, which is applied to Figure 1 The terminal 102 in the example is used as an example to illustrate, and the following steps are included:

[0047] Step S100, obtaining a predicted diameter value based on time series monitoring data within a preset time range.

[0048] Among them, the preset time range can be a time range selected by a fixed time sliding window method with the current moment as the end time, or it can be a time range selected with the current device entering the shoulder release stage as the starting time and the current moment as the end time, or it can be a time range obtained based on other selection rules, which is not limited in this embodiment.

[0049] The timing monitoring data may be monitoring data obtained by monitoring the current device during the shoulder release process and in a time sequence. Among them, the timing monitoring data may include one or more of the crystal state parameters, environmental state parameters and operating condition parameters of the current device during the shoulder release process. The crystal state parameters may be measured parameters in the shoulder release stage, such as diameter, temperature, growth rate, crystal length, etc.; environmental state parameters may be measured parameters of the environment in which the crystal is located, such as liquid level, ambient temperature, etc.; operating condition parameters may be execution parameters or measured parameters of the current device itself during operation, and the execution parameters may include control instructions for each component, such as target speed, target rising speed, etc., and the measured parameters may include state parameters of each component, such as speed, torque, load, temperature, etc. of the component.

[0050] By analyzing past time-series monitoring data, future changes can be predicted more accurately, thereby obtaining more accurate predicted diameter values.

[0051] Step S200, inputting the target shoulder angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target pulling speed value; the construction process of the long short-term memory mechanism model includes: constructing a short-term loss index based on the first storage step; constructing a long-term loss index based on the second storage step; constructing the long short-term memory mechanism model based on the short-term loss index and the long-term loss index.

[0052] Among them, the long short-term memory mechanism model can be obtained by training with a long short-term memory network based on the short term, or other models that generate pulling speed values ​​based on short-term data and long-term data training or processing. The storage step size is used to specify the duration of the stored historical data, so that when the model calculates the loss, it calculates the loss based on historical data of different lengths. The first storage step size and the second storage step size can be obtained based on user settings, and the first storage step size is smaller than the second storage step size. By constructing long-term loss indicators and short-term loss indicators respectively by different step sizes, the loss calculation of long-term and short-term memory can be realized, and the long short-term memory mechanism model can be obtained.

[0053] In some of the embodiments, the long-short-term memory mechanism model can be a long-short-term memory loss function, which can calculate the short-term loss value and the long-term loss value according to the short-term loss index and the long-term loss index, respectively, and then perform a comprehensive calculation based on the short-term loss value and the long-term loss value to obtain the predicted loss value.

[0054] Constructing the short-term loss indicator through the first storage step may be to measure the difference between the prediction and the actual value within the shorter time window corresponding to the first storage step.

[0055] Constructing the long-term loss indicator through the second storage step can be based on a longer time window corresponding to the second storage step, and measure the difference between the prediction and the actual value within the time window.

[0056] Based on the short-term loss index and the long-term loss index, a long-short-term memory mechanism model is constructed. The applicable conditions of the short-term loss index and the long-term loss index can be set, or the weight ratio of the output loss values ​​of the short-term loss index and the long-term loss index can be set to obtain the long-short-term memory mechanism model.

[0057] Compared with the traditional PID control method, the long short-term memory mechanism model can better understand and adapt to complex process conditions, thereby providing a more accurate and smooth speed adjustment strategy, which helps to avoid problems caused by response lag.

[0058] Step S300, based on the target pulling speed value, control the target device to release the shoulder.

[0059] Among them, the actual shoulder release speed of the current device can be adjusted based on the determined target pulling speed value.

[0060] A shoulder release control method provided in the present embodiment obtains a predicted diameter value based on time series monitoring data within a preset time range; performs simulation based on process condition parameters of the current equipment to obtain a target shoulder release angle; inputs the target shoulder release angle and the predicted diameter value into a long short-term memory mechanism model to obtain a target pulling speed value; and controls the target equipment to perform shoulder release based on the target pulling speed value. The method can determine the optimal shoulder release angle of the current equipment based on historical time series monitoring data, and determine the optimal pulling speed value in real time in combination with the long short-term memory mechanism model, thereby solving the problems of time lag, uncertainty and nonlinearity of pulling speed control during the shoulder release process, and achieving the effect of improving the survival rate of shoulder release.

[0061] In one embodiment, obtaining the predicted diameter value based on the time series monitoring data within a preset time range includes:

[0062] Based on the time series monitoring data within a preset time range and the pre-trained diameter prediction model, the predicted diameter value is obtained.

[0063] Among them, the pre-trained diameter prediction model is trained by multiple historically collected complete monitoring data of the same type of equipment as the current equipment in the shoulder release stage.

[0064] The equipment of the same type as the current equipment may be equipment of the same model, or equipment of the same type in terms of application field, process stage or specific use.

[0065] The complete monitoring data can be the complete data collected from the first rod to the last rod during the shoulder release process. Multiple complete monitoring data can be the complete monitoring data of multiple shoulder releases of a single device, the complete monitoring data of shoulder releases of multiple devices, or the complete monitoring data of multiple shoulder releases of multiple devices.

[0066] Training based on complete monitoring data may be performed by extracting the measured diameter value and other monitoring data from the complete monitoring data, and using the other monitoring data and the measured diameter value as training data. For example, linear regression, decision tree, support vector machine, neural network, deep learning model, etc. may be trained, or other existing artificial intelligence algorithms may be used for training, which is not limited in this embodiment.

[0067] The present embodiment provides a shoulder release control method, which predicts the diameter value through a pre-trained diameter prediction model and time series monitoring data, and can obtain the predicted diameter value at the current moment. Compared with simple trend analysis or rule-based methods, the trained model can predict the results more accurately and reduce the errors caused by subjective judgment differences, thereby achieving the effect of improving the survival rate of shoulder release.

[0068] In one embodiment, obtaining the predicted diameter value based on the time series monitoring data within a preset time range and the pre-trained diameter prediction model includes:

[0069] Obtain complete monitoring data of the same type of equipment as the current equipment during the shoulder release phase collected historically;

[0070] Decompose the shoulder release phase data into trend component data and residual component data;

[0071] Based on the trend component data, residual component data and measured diameter data, the linear layer is trained to obtain the diameter prediction model.

[0072] The complete monitoring data may include measured diameter data and shoulder release phase data, the measured diameter data may be the measured diameter value at each diameter measurement sampling point, and the shoulder release phase data may include at least one of crystal state parameters, environmental state parameters, and operating condition parameters. The crystal state parameters, environmental state parameters, and operating condition parameters are not described in detail herein.

[0073] Decomposing the shoulder-releasing phase data into trend component data and participating component data can be achieved by using a technology suitable for time series decomposition, such as the STL decomposition method. The trend component data represents the long-term change trend in the data. For example, the smooth part can be extracted from the shoulder-releasing phase data as the trend component data, which is used to characterize the change trend of each data type in the shoulder-releasing phase data; the residual component data can be the residual data after removing the trend component data from the shoulder-releasing phase data. In a specific embodiment, the value of the residual component data can be the relative value of the trend component data.

[0074] In this embodiment, the diameter prediction model can be obtained based on a linear layer, and the linear layer can be a linear regression model for prediction. The linear layer is trained based on the decomposed trend component data and residual component data as input parameters and the measured diameter data as output parameters.

[0075] The present embodiment provides a shoulder release control method, which decomposes the shoulder release stage data into trend component data and residual component data, and trains the linear layer based on the trend component data, the residual component data and the measured diameter data to obtain a diameter prediction model. This can achieve a more detailed decomposition of the data and model them separately, so as to better capture the different characteristics in the data and obtain more accurate prediction results, thereby achieving the effect of improving the survival rate of shoulder release.

[0076] It is understandable that in the shouldering process, the control of the shouldering angle plays an important role in the shouldering survival rate of the crystal. If the shouldering angle is too large, it may cause the appearance of grain boundaries or grains during the growth of the crystal. If the shouldering angle is too small, it will make it difficult to control the gradient during the crystal growth process, resulting in difficulty in crystal growth. Therefore, the control of the shouldering angle of the crystal is very important.

[0077] In one embodiment, a target shoulder angle can be obtained according to the process condition parameters of the current equipment, and the target shoulder angle and the predicted diameter value are input into the long short-term memory mechanism model. Before obtaining the target pulling speed value, the following steps are also included:

[0078] Obtain the process condition parameters of the current equipment;

[0079] Based on the process condition parameters, thermal stress simulation is performed to obtain the first shoulder angle; the first shoulder angle is the shoulder angle with the minimum thermal stress under the process condition parameters;

[0080] Based on the first shoulder drop angle, determine the target shoulder drop angle.

[0081] The thermal stress simulation may be performed by simulating the thermal stress distribution of the crystal when it grows under given process conditions through numerical simulation software, such as finite element analysis tools.

[0082] Through the process condition parameters, each shoulder release angle can be simulated, the thermal stress value under each shoulder release angle can be calculated, and the shoulder angle with the smallest thermal stress value can be selected as the target shoulder release angle.

[0083] The process condition parameters include at least one of the equipment parameters, process parameters and material supply parameters of the current equipment, wherein the equipment parameters may be the hardware parameters of the equipment itself, such as equipment model, rated power, etc.; the process parameters may be the process flow, setting parameters, etc. used by the equipment in the shouldering process; the material supply parameters may be the material supply data in the shouldering process, which may be characterized by the control parameters of the control elements corresponding to each material. For example, the weight of the supplied silicon material may be obtained by mapping the weight change data fed back by the sensor, or by mapping the operating data of the valve motor at the feeding port.

[0084] Based on the first shoulder angle, the target shoulder angle is determined. The first shoulder angle may be used as the target shoulder angle, or the first shoulder angle may be further processed, such as adjusted or corrected in combination with other shoulder angles to obtain the target shoulder angle.

[0085] In the shoulder release control method provided in this embodiment, thermal stress simulation is performed through process condition parameters, and the shoulder angle with the minimum thermal stress value can be obtained as the target shoulder release angle. The shoulder angle with the minimum thermal stress value can be obtained as the target shoulder release angle, so that the internal thermal stress of the crystal in the shoulder release stage can be maintained in an ideal state, thereby achieving the effect of improving the shoulder release survival rate.

[0086] In one embodiment, thermal stress simulation is performed based on process condition parameters to obtain the first shoulder angle, which includes:

[0087] Based on the equipment parameters, process parameters and material supply parameters of the current equipment, build an equipment model corresponding to the equipment parameters;

[0088] Calculate thermal stresses for multiple shoulder angles based on the device model, elastic modulus and thermal expansion coefficient of the target material;

[0089] The shoulder angle with the smallest thermal stress is taken as the first shoulder angle.

[0090] For example, when the current equipment is a single crystal furnace, the equipment parameter may be the furnace type, the process parameter may be the drawing specification, the material supply parameter may be the oxygen content, and the like.

[0091] Based on the equipment parameters, process parameters and material supply parameters of the current equipment, a equipment model corresponding to the equipment parameters is constructed. The above parameters can be input into a third-party software and the equipment model is constructed through the software.

[0092] Based on the device model, the elastic modulus and thermal expansion coefficient of the target material, the thermal stresses of multiple shoulder angles are calculated. The elastic modulus and thermal expansion coefficient of the target material are filled into the solution formula under the current device model to obtain the thermal stresses of multiple shoulder angles.

[0093] In order to maximize the survival rate in the shouldering stage, the shoulder angle with the smallest thermal stress can be taken as the first shouldering angle.

[0094] The present embodiment provides a shoulder release control method, which constructs an equipment model and calculates the thermal stress of multiple shoulder angles in combination with the elastic modulus and thermal expansion coefficient of the target material. The shoulder angle with the minimum thermal stress can be obtained as the first shoulder angle, thereby maintaining the internal thermal stress of the crystal in the shoulder release stage in an ideal state, thereby achieving the effect of improving the shoulder release survival rate.

[0095] In one embodiment, the target shoulder angle can be obtained based on the complete monitoring data collected historically, and the target shoulder angle and the predicted diameter value can be input into the long short-term memory mechanism model. Before obtaining the target pulling speed value, the following steps are also included:

[0096] Obtain complete monitoring data of the same type of equipment as the current equipment during the shoulder release phase collected historically;

[0097] Based on the complete monitoring data, the target shoulder angle is determined; the target shoulder angle is the shoulder angle with the highest survival rate in the complete monitoring data.

[0098] It is understandable that the complete monitoring data of the same type of equipment as the current equipment in the shoulder release stage collected in history records the process data of each successful shoulder release. Therefore, through the complete monitoring data, the shoulder angle corresponding to each successful shoulder release can be counted.

[0099] Based on the complete monitoring data, the target shoulder angle is determined. This can be based on the shoulder angle corresponding to each successful shoulder release. The shoulder angle with the most successful shoulder releases or the highest shoulder release success rate is counted as the target shoulder angle.

[0100] A shoulder release control method provided in this embodiment determines a target shoulder release angle based on complete monitoring data. The target shoulder release angle is a shoulder angle obtained based on statistics of historical complete monitoring data. By selecting the shoulder angle with the highest survival probability as the target shoulder angle, the survival rate of shoulder release can be improved.

[0101] Further, in one embodiment, based on the first shoulder release angle, determining a target shoulder release angle further includes:

[0102] Based on the process condition parameters of the current equipment, thermal stress simulation is performed to obtain the first shoulder release angle;

[0103] Determine a historical shoulder release angle set based on historically collected complete monitoring data of the same type of equipment as the current equipment during the shoulder release phase;

[0104] Based on the first shoulder release angle and the historical shoulder release angle set, a target shoulder release angle is obtained.

[0105] In this embodiment, the judgment can be made in combination with the process condition parameters and the complete monitoring data. In particular, the first shoulder release angle can be obtained based on the process condition parameters.

[0106] The first shoulder angle may be a shoulder angle with the smallest thermal stress under the process condition parameters of the current equipment.

[0107] The historical shoulder angle set may be a shoulder angle set in time series monitoring data whose survival rate is higher than a preset survival rate threshold.

[0108] Based on the first shoulder angle and the set of historical shoulder angles, the target shoulder angle is obtained. The target shoulder angle can be obtained by calculating the first shoulder angle and the set of historical shoulder angles through statistical methods, such as weighted average and other optimization algorithms. The best shoulder angle obtained under statistics is used as the target shoulder angle. Alternatively, the target shoulder angle can be an angle in the set of historical shoulder angles that is close to the first shoulder angle.

[0109] The present embodiment provides a shoulder release control method, which calculates the target shoulder release angle by combining process condition parameters and complete monitoring data. It can also combine the special conditions of the current equipment with the experience under historical complete monitoring data to obtain the optimal target shoulder release angle for the current equipment, thereby achieving the effect of improving the survival rate of shoulder release.

[0110] In one embodiment, obtaining a target shoulder release angle based on the first shoulder release angle and the historical shoulder release angle set includes:

[0111] Based on the first shoulder release angle, the historical shoulder release angle set is screened to obtain at least one second shoulder release angle having a similarity with the first shoulder release angle higher than a preset similarity threshold;

[0112] Based on the first shoulder release angle and the second shoulder release angle, a target shoulder release angle is determined.

[0113] Wherein, based on the first shoulder angle, the historical shoulder angle set is screened, and the shoulder angle with the highest similarity to the first shoulder angle or higher than a preset threshold is found in the historical shoulder angle set as the second shoulder angle. The similarity can be obtained by using an existing similarity calculation method, and can be calculated by using, for example, an absolute difference, a relative error, or a more complex statistical distance (such as Euclidean distance). The preset similarity threshold needs to be set according to the actual situation to ensure that the screened angle is close to the theoretical optimal solution and has a priori knowledge of practical operation.

[0114] Furthermore, processing is performed based on the first shoulder angle and the second shoulder angle, such as by using an average method, a weighted average or other optimization algorithms, such as a genetic algorithm, a multi-objective optimization technique, etc., to find a balanced shoulder angle between the first shoulder angle and the second shoulder angle as the target shoulder angle.

[0115] The present embodiment provides a shoulder release control method, which obtains a second shoulder release angle by screening a historical shoulder release angle set based on a first shoulder release angle, and obtains a target shoulder release angle based on the first shoulder release angle and the second shoulder release angle. The optimal target shoulder release angle under the current equipment can be obtained, thereby achieving the effect of improving the survival rate of shoulder release.

[0116] In one embodiment, the target shoulder angle and the predicted diameter value are input into the long short-term memory mechanism model to obtain the target pulling speed value including:

[0117] Input the target shoulder angle and the predicted diameter value into the long short-term memory loss function to obtain the predicted loss value;

[0118] If the predicted loss value is less than the preset loss threshold, the target pulling speed value is determined based on the historical pulling speed control value and the target shoulder angle at the previous moment.

[0119] The long short-term memory mechanism model includes a long short-term memory loss function, which can be used to measure the difference between the predicted value and the actual value of the model. In this embodiment, the loss function can be used to evaluate the deviation between the target pulling speed value obtained by the currently input target shoulder angle and predicted diameter value and the ideal state.

[0120] It can be understood that the loss value calculation of the long-term and short-term memory loss function is the loss value obtained based on the comprehensive judgment of long-term data and short-term data, thereby realizing the loss calculation of long-term and short-term memory.

[0121] The preset loss threshold may be a threshold set by a user or operator based on actual needs or prior knowledge. Under the preset loss threshold, it can be ensured that the pulling speed is controlled within a safe range.

[0122] If the predicted loss value is less than the preset loss threshold, the target pulling speed value is determined based on the historical pulling speed control value and the target shoulder angle at the previous moment. The required speed increase or deceleration compared to the historical pulling speed control value at the previous moment can be determined based on the historical pulling speed control value at the previous moment and the required target shoulder angle, so as to obtain the target pulling speed value.

[0123] Furthermore, if the predicted loss value is greater than or equal to the preset loss threshold, it is necessary to redetermine the target shoulder angle and predicted diameter value, so as to further calculate the target pulling speed value.

[0124] A shoulder release control method provided in this embodiment introduces a loss function to evaluate the loss value under the target shoulder release angle and the predicted diameter value, and then determines the target pulling speed value based on the historical pulling speed control value and the target shoulder release angle at the previous moment. It can filter out unreliable prediction results and unstable factors caused by frequent adjustments, thereby maintaining a high control accuracy under complex and changeable production conditions, and can achieve the effect of improving the shoulder release survival rate.

[0125] In order to more clearly illustrate the technical solution of the present application, the present application also provides a detailed embodiment.

[0126] In one embodiment, Figure 3As shown, a shoulder release control method is provided, comprising:

[0127] Obtain the time series big data of monitoring parameters of the shouldering process of CZ single crystal silicon, that is, complete monitoring data. The time series big data can be the operating condition data of 200 CZ single crystal furnace equipment in at least one furnace process, that is, shouldering from the first to the last rod. The complete monitoring data may include the state distribution of the maximum and minimum values ​​of the shouldering pulling speed reaching the design limit, and the initial temperature and power state of the shouldering covering the historical statistics. The parameters collected for the data are shouldering process variables including but not limited to temperature, liquid level, diameter, crystal growth rate, and crystal length.

[0128] Long prediction of diameter in shoulder release process based on DLinear method. Traditional long short-term memory neural network (LSTM) cannot adapt to data drift caused by process changes, which leads to degradation of model prediction performance. In order to solve the problem of data offset caused by the time-varying characteristics of large-scale single crystal silicon shoulder release process, this embodiment uses Dlinear method to achieve stable prediction of shoulder release process diameter.

[0129] Furthermore, the Dlinear method includes: first decomposing the original data into a trend component and a residual component by moving average, then applying a linear layer to these two components, and finally adding the two components to obtain the final prediction. Decomposition into trend components and residual components ; The decomposition formula of the trend moving average method is as follows:

[0130]

[0131] in, For trend components , residual components ;

[0132] Trend component and residual components The linear layer prediction is established according to the following formula:

[0133]

[0134] Among them, W is the linear layer of the time axis with a dimension of T×L. and are the input and predicted values ​​of the ith variable respectively. For the predicted value of the trend component And the residual prediction model prediction value Accumulate and finally realize the prediction of diameter.

[0135] The steps for establishing the above diameter prediction model include: dividing the collected time series big data, that is, the complete monitoring data, into a training set and a test set according to the number of data, with the ratio of the training set to the test set being 7:3, and then evaluating on the training set and the test set, and using the Dlinear method to predict the mean absolute error (MAE) of the K (K>=3) step diameter. The calculation formula is as follows:

[0136]

[0137] in is the target diameter value, is the predicted value of diameter, and N is the number of samples.

[0138] Coupling time series big data and control objectives of silicon single crystal growth mechanism.

[0139] It is understandable that the optimal shoulder target that only relies on historical data requires a large amount of historical data under the same process conditions. The generated shoulder target does not necessarily meet the crystallization requirements of the process, and cannot meet the rapid adjustment of the silicon single crystal growth process and the crystallization mechanism. Therefore, the setting of the control target should comprehensively consider the time series big data and the silicon single crystal growth mechanism process.

[0140] Firstly, according to the process condition parameters, including setting the furnace type, drawing specifications, resistivity and oxygen content, the isosceles shoulder with relatively uniform thermal stress distribution is simulated in the CGSim software, and the angle A of the shoulder is determined.

[0141] The simulation solution process of angle A is as follows:

[0142]

[0143] Where E is the elastic modulus of the single crystal silicon material, α is the thermal expansion coefficient of the single crystal silicon material, ΔT i is the temperature difference between adjacent intervals, l is the shoulder length, and for each shoulder angle A belonging to the shoulder angle set {A}, the optimal shoulder angle with the minimum thermal stress is found.

[0144] Then, based on the optimal setting angle of the shoulder type, the actual growth shoulder type with the highest shoulder survival rate is selected from the historical time series distribution data, and the shoulder type setting angle is matched. , and its objective function of multi-objective programming is:

[0145]

[0146] Among them, A i is the actual angle of the shoulder for the living shoulder type, {Shoulder} is the historical shoulder type set, S(A i) function calculates the survival rate of the shoulder type subset with a similarity greater than 0.8 with the surviving shoulder type in the historical shoulder type set based on a dynamic regularization algorithm, thereby completing the setting of the shoulder type target in the shoulder release process.

[0147] Long short-term memory adaptive controller. Due to the differences in silicon single crystal growth between furnaces and between different equipment during the continuous CZ-type silicon single crystal growth process, the long short-term memory adaptive controller is designed as follows:

[0148] 1. Define the control loss function:

[0149]

[0150] Where e(k) is the error loss function, y r (k) is the set diameter value at time k, y(k) is the actual diameter value at time k, To set the diameter change rate, is the actual diameter change rate.

[0151] 2. Long and short-term memory mechanism:

[0152] Short-term indicators:

[0153] Long-term indicators:

[0154] Where s and l are both historical step storage steps, and the step length s in the short-term indicator is less than the step length l in the long-term indicator.

[0155]

[0156] in is the error loss function of the diameter change rate at moment k.

[0157] 3. Optimize and find the best pulling speed value:

[0158]

[0159] Among them, u(k) is the response value at the current time k, u(k-1) is the speed control value of k at time k-1; ΔU(k) ​​is the loss function solved based on the control loss function and the shoulder diameter prediction model in step 2 , the increase in pulling speed.

[0160] Control performance evaluation. This paper uses control following deviation and shoulder release survival rate to describe the performance of the control method. Control following deviation is defined as the average value of the difference between the control target and the actual index. The shoulder release survival rate calculation formula is defined as follows: shoulder release survival rate = number of shoulder releases entering equal diameter / number of shoulder releases × 100%;

[0161] The embodiment of this article is a two-week actual shoulder release process of 211 times in 20 vertical single crystal furnaces. The whole shoulder release process is taken over by the control program without human intervention. The shoulder release process control of the comparative example is a PID control algorithm. The cooling power adjustment and crucible lifting control of the shoulder release process in the embodiment and the comparative example are the same, and the upper and lower limits of the pulling speed are also the same. The quality of the polysilicon raw materials fed, the batch of quartz crucibles, the experimental environment, the feeding and disassembly personnel, and the execution process standards of the embodiment and the comparative example are the same. The furnace type is the Jingsheng 160 furnace type produced in the same batch, and the production specification is 12 inches. According to the shoulder release diameter prediction model obtained in steps one and two, as shown in Figure 4 As shown in the figure, the comparison between the predicted shoulder diameter and the actual diameter shows that the model can accurately predict the shoulder diameter value, and the MAE on the test set is 0.2 mm.

[0162] Table 1 Comparative analysis framework diagram

[0163]

[0164] The shoulder release control method of the experimental example of the present invention was used to produce a 12-inch single crystal silicon rod. The shoulder release survival rate and yield were higher than the shoulder release PID control of the comparative example. The survival rate was increased by 4.6%. The target following deviation of the control method was 50% lower than that of the PID control. In addition, it can be found that the shoulder shape consistency of the experimental example in the embodiment is significantly higher than that of the comparative example. Specifically, the shoulder picture is as follows Figure 5 As shown in FIG. 1 , it can be found that the shoulder has no obvious step under the control method, and the shoulder is relatively smooth, which meets the requirements of the silicon crystal growth process. Figure 6 The image of the shoulder release process is shown. In summary, the shoulder release process method of the present invention has a significant effect in preparing 12-inch single crystal silicon rods.

[0165] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0166] Based on the same inventive concept, the embodiment of the present application also provides a shoulder release control device for implementing the shoulder release control method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the shoulder release control device provided below can refer to the limitations of the shoulder release control method above, and will not be repeated here.

[0167] In one embodiment, Figure 7 As shown, a shoulder release control device is provided, which includes a diameter prediction module 100, a pulling speed determination module 200 and a shoulder release control module 300, wherein:

[0168] The diameter prediction module 100 is used to obtain a predicted diameter value based on the time series monitoring data within a preset time range;

[0169] The pulling speed determination module 200 is used to input the target shoulder angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target pulling speed value; the construction process of the long short-term memory mechanism model includes: constructing a short-term loss index based on a first storage step; constructing a long-term loss index based on a second storage step; the first storage step is smaller than the second storage step; constructing the long short-term memory mechanism model based on the short-term loss index and the long-term loss index;

[0170] The shoulder release control module 300 is used to control the target device to release the shoulder based on the target pulling speed value.

[0171] In one embodiment, the diameter prediction module 100 is further configured to:

[0172] The predicted diameter value is obtained based on the time series monitoring data within a preset time range and a pre-trained diameter prediction model; the diameter prediction model is trained by multiple historically collected complete monitoring data of the same type of equipment as the current equipment during the shoulder release stage.

[0173] In one embodiment, the shoulder release control device further includes a model training module for:

[0174] Acquire historically collected complete monitoring data of the same type of equipment as the current equipment during the shoulder release stage; the complete monitoring data includes measured diameter data and shoulder release stage data; the shoulder release stage data includes at least one of crystal state parameters, environmental state parameters and operating condition parameters;

[0175] Decompose the shoulder release phase data into trend component data and residual component data;

[0176] Based on the trend component data, residual component data and measured diameter data, the linear layer is trained to obtain the diameter prediction model.

[0177] In one embodiment, the shoulder release control device further includes a first angle determination module, which is used to:

[0178] Obtaining process condition parameters of the current equipment; the process condition parameters include at least one of equipment parameters, process parameters and material supply parameters of the current equipment;

[0179] Based on the process condition parameters, thermal stress simulation is performed to obtain the first shoulder angle; the first shoulder angle is the shoulder angle with the minimum thermal stress under the process condition parameters;

[0180] Based on the first shoulder release angle, a target shoulder release angle is determined.

[0181] In one embodiment, the first angle determination module is further configured to:

[0182] Based on the equipment parameters, process parameters and material supply parameters of the current equipment, construct an equipment model corresponding to the equipment parameters;

[0183] Calculating thermal stresses for multiple shoulder angles based on the device model, elastic modulus and thermal expansion coefficient of the target material;

[0184] The shoulder angle with the smallest thermal stress is taken as the first shoulder angle.

[0185] In one embodiment, the shoulder release control device further includes a second angle determination module, which is used to:

[0186] Obtain complete monitoring data of the same type of equipment as the current equipment during the shoulder release phase collected historically;

[0187] Based on the complete monitoring data, the target shoulder angle is determined; the target shoulder angle is the shoulder angle with the highest survival rate in the complete monitoring data.

[0188] In one embodiment, the long short-term memory mechanism model includes a long short-term memory loss function, and the pulling speed determination module 200 is further used to:

[0189] Input the target shoulder angle and the predicted diameter value into the long short-term memory loss function to obtain the predicted loss value;

[0190] If the predicted loss value is less than the preset loss threshold, the target pulling speed value is determined based on the historical pulling speed control value and the target shoulder angle at the previous moment.

[0191] Each module in the shoulder release control device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0192] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a shoulder control method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0193] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0194] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the shoulder release control method of any of the above embodiments is implemented:

[0195] Based on the time series monitoring data within a preset time range, a predicted diameter value is obtained;

[0196] Inputting the target shoulder angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target pulling speed value; the construction process of the long short-term memory mechanism model includes: constructing a short-term loss index based on a first storage step; constructing a long-term loss index based on a second storage step; the first storage step is smaller than the second storage step; constructing the long short-term memory mechanism model based on the short-term loss index and the long-term loss index;

[0197] Based on the target pulling speed value, the target device is controlled to release the shoulder.

[0198] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the shoulder release control method of any of the above embodiments is implemented:

[0199] Based on the time series monitoring data within a preset time range, a predicted diameter value is obtained;

[0200] Inputting the target shoulder angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target pulling speed value; the construction process of the long short-term memory mechanism model includes: constructing a short-term loss index based on a first storage step; constructing a long-term loss index based on a second storage step; the first storage step is smaller than the second storage step; constructing the long short-term memory mechanism model based on the short-term loss index and the long-term loss index;

[0201] Based on the target pulling speed value, the target device is controlled to release the shoulder.

[0202] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0203] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0204] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0205] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A shoulder release control method, characterized in that: The shoulder release control method comprises: Based on the time series monitoring data within a preset time range, a predicted diameter value is obtained; Input the target shoulder release angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target pulling speed value; the construction process of the long short-term memory mechanism model includes: constructing a short-term loss index based on the first storage step; constructing a long-term loss index based on the second storage step; the first storage step is smaller than the second storage step; constructing the long short-term memory mechanism model based on the short-term loss index and the long-term loss index; inputting the target shoulder release angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target pulling speed value also includes: obtaining the process condition parameters of the current equipment; the process condition parameters include at least one of the equipment parameters, process parameters and material supply parameters of the current equipment; performing thermal stress simulation based on the process condition parameters to obtain a first shoulder release angle; the first shoulder release angle is the shoulder angle with the minimum thermal stress under the process condition parameters; determining the target shoulder release angle based on the first shoulder release angle; Based on the target pulling speed value, the target device is controlled to release the shoulder.

2. The shoulder release control method according to claim 1, characterized in that: The step of obtaining the predicted diameter value based on the time series monitoring data within a preset time range includes: Based on the time series monitoring data within a preset time range and a pre-trained diameter prediction model, a predicted diameter value is obtained; the diameter prediction model is trained by multiple historically collected complete monitoring data of the same type of equipment as the current equipment during the shoulder release stage.

3. The shoulder release control method according to claim 2, characterized in that: The method of obtaining the predicted diameter value based on the time series monitoring data within the preset time range and the pre-trained diameter prediction model includes: Acquire historically collected complete monitoring data of the same type of equipment as the current equipment during the shoulder release stage; the complete monitoring data includes measured diameter data and shoulder release stage data; the shoulder release stage data includes at least one of crystal state parameters, environmental state parameters and operating condition parameters; Decomposing the shoulder release phase data into trend component data and residual component data; Based on the trend component data, the residual component data and the measured diameter data, the linear layer is trained to obtain the diameter prediction model.

4. The shoulder release control method according to claim 1, characterized in that: The thermal stress simulation is performed based on the process condition parameters to obtain the first shoulder release angle, which includes: Based on the equipment parameters, process parameters and material supply parameters of the current equipment, construct an equipment model corresponding to the equipment parameters; Calculating thermal stresses for multiple shoulder angles based on the device model, elastic modulus and thermal expansion coefficient of the target material; The shoulder angle with the smallest thermal stress is taken as the first shoulder angle.

5. The shoulder release control method according to claim 1, characterized in that: Before inputting the target shouldering angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target pulling speed value, the following steps are also included: Obtain complete monitoring data of the same type of equipment as the current equipment during the shoulder release phase collected historically; Based on the complete monitoring data, a target shoulder angle is determined; the target shoulder angle is the shoulder angle with the highest survival rate in the complete monitoring data.

6. The shoulder release control method according to claim 1, characterized in that: The long short-term memory mechanism model includes a long short-term memory loss function, and the target shoulder angle and the predicted diameter value are input into the long short-term memory mechanism model to obtain the target pulling speed value, which includes: Inputting the target shoulder angle and the predicted diameter value into the long short-term memory loss function to obtain a predicted loss value; If the predicted loss value is less than the preset loss threshold, the target pulling speed value is determined based on the historical pulling speed control value at the previous moment and the target shoulder angle.

7. A shoulder release control device, characterized in that: The shoulder release control device comprises: A diameter prediction module, used to obtain a predicted diameter value based on time series monitoring data within a preset time range; A drawing speed determination module is used to input the target shoulder angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target drawing speed value; the construction process of the long short-term memory mechanism model includes: constructing a short-term loss index based on a first storage step; constructing a long-term loss index based on a second storage step; the first storage step is smaller than the second storage step; constructing the long short-term memory mechanism model based on the short-term loss index and the long-term loss index; before inputting the target shoulder angle and the predicted diameter value into the long short-term memory mechanism model to obtain the target drawing speed value, it also includes: obtaining the process condition parameters of the current equipment; the process condition parameters include at least one of the equipment parameters, process parameters and material supply parameters of the current equipment; performing thermal stress simulation based on the process condition parameters to obtain a first shoulder angle; the first shoulder angle is the shoulder angle with the minimum thermal stress under the process condition parameters; determining the target shoulder angle based on the first shoulder angle; The shoulder release control module is used to control the target device to release the shoulder based on the target pulling speed value.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images