Polysilicon production equipment data prediction method and device, server and storage medium

Abstract

The application provides a polysilicon production equipment data prediction method and device, a server and a storage medium, wherein the method comprises: obtaining a plurality of first parameter samples, the first sample parameters in the first parameter samples comprising equipment parameters of polysilicon production equipment and process parameters of upstream and downstream processes thereof; obtaining filtered second sample parameters according to EMD and principal component analysis, and determining at least one main parameter and a corresponding response parameter through principal component analysis; obtaining a correlation model between parameters according to multivariate linear regression; and obtaining corresponding predicted parameter samples and a predicted robust distance according to the second parameter samples, the correlation model and an EMD component set corresponding to each second sample parameter. The application can effectively reduce noise, construct a correlation model based on the correlation between parameters to make the prediction result more accurate, and improve the accuracy and stability of the prediction result through the predicted robust distance.

Description

Technical Field

[0001] This application relates to the field of equipment health management technology, and in particular to a data prediction method, device, server and storage medium for polysilicon production equipment. Background Technology

[0002] Equipment data prediction and early warning are crucial in the large-scale chemical industry, not only ensuring the safe operation of equipment and timely maintenance under abnormal conditions, but also guaranteeing the safe and stable operation of chemical processes. Currently, for equipment parameters, various chemical industries generally build correlation models based on known historical data, and these models can be directly used for real-time monitoring of parameter updates. However, accurate prediction of future conditions is still not possible. In polysilicon processes, current predictions rely solely on coarse lag and the changing trends of end-sample data. While methods such as multiple linear regression and principal component analysis for noise reduction can correct biases to some extent, they ultimately cannot achieve accurate predictions and are prone to parameter silos, failing to explain the relationships between various parameter sampling points. For example, the method and device for predicting combined power loads based on EEMD provided in patent CN111784068A. Therefore, to be applicable to large-scale chemical processes such as polysilicon, a comprehensive approach is needed that establishes relationships between various parameters based on detailed predictions of each parameter, thereby achieving accurate prediction of polysilicon production equipment data. Summary of the Invention

[0003] The technical objective of this application is to provide a method, apparatus, server, and storage medium for predicting data from polysilicon production equipment, in order to solve the current problem of the inability to accurately predict data from polysilicon production equipment.

[0004] To address the aforementioned technical problems, embodiments of this application provide a method for predicting data from polysilicon production equipment, comprising:

[0005] Multiple first parameter samples are obtained. The first sample parameters in the first parameter samples include the equipment parameters of the polysilicon production equipment and the process parameters of its upstream and downstream processes. Each first parameter sample corresponds to one of multiple time points within a preset time period.

[0006] Based on Empirical Mode Decomposition (EMD) and Principal Component Analysis (PCA), each first sample parameter in the first parameter sample is individually subjected to interference filtering to obtain the filtered second sample parameters and the second parameter sample including the second sample parameters.

[0007] The principal component analysis is performed on the second parameter sample to determine at least one major parameter affecting the polysilicon production equipment and its upstream and downstream processes, as well as the response parameters corresponding to each major parameter, wherein the response parameters corresponding to the major parameters are the second sample parameters in the second parameter sample other than the major parameters;

[0008] Based on the main parameters and the corresponding response parameters, a multiple linear regression is performed to obtain the correlation model of the polysilicon production equipment and its upstream and downstream processes.

[0009] Based on the second parameter sample, the correlation model, and the EMD component set corresponding to each second sample parameter, the prediction results of the polysilicon production equipment and its upstream and downstream processes are obtained. The prediction results include: prediction parameter sample and prediction robust distance.

[0010] Preferably, the method described above further includes:

[0011] The polysilicon production equipment is controlled based on the prediction results of the polysilicon production equipment and its upstream and downstream processes.

[0012] Specifically, in the method described above, the step of individually performing interference filtering on each first sample parameter in the first parameter sample based on EMD and principal component analysis to obtain filtered second sample parameters and a second parameter sample including the second sample parameters includes:

[0013] EMD is performed on each of the first sample parameters to obtain multiple intrinsic mode function (IMF) components and residual components;

[0014] Principal component analysis is performed on each of the IMF components and the residual components according to the first cumulative contribution rate threshold to obtain the set of EMD components of the second sample parameters corresponding to each of the first sample parameters;

[0015] The components in each of the EMD component sets are merged to obtain the corresponding second sample parameters and the second parameter samples.

[0016] Specifically, in the method described above, performing principal component analysis on the second parameter sample to determine at least one major parameter affecting the polysilicon production equipment and its upstream and downstream processes, and the response parameters corresponding to each major parameter, includes:

[0017] Based on the second cumulative contribution rate threshold, the principal component analysis is performed on the second parameter sample to obtain the load matrix, score matrix and eigenvalues ​​corresponding to the polysilicon production equipment and its upstream and downstream processes.

[0018] The principal component equation is obtained based on the loading matrix and the score matrix, and the second sample parameter existing in the principal component equation is determined as the principal parameter.

[0019] Based on the main parameters and the second sample parameters, the response parameters corresponding to each of the main parameters are determined.

[0020] Specifically, in the method described above, the step of performing multiple linear regression based on the main parameters and the corresponding response parameters to obtain the correlation model of the polysilicon production equipment and its upstream and downstream processes includes:

[0021] By linear fitting, a multiple linear regression model is obtained between the main parameters and their corresponding response parameters.

[0022] Obtain the T-distribution test value and F-distribution test value for each of the aforementioned multiple linear regression models;

[0023] The multiple linear regression model whose T-distribution test value is greater than the corresponding first threshold and whose F-distribution test value is greater than the corresponding first threshold is identified as the correlation model.

[0024] Specifically, as described above, obtaining the prediction results of the polysilicon production equipment and its upstream and downstream processes based on the second parameter sample, the correlation model, and the EMD component set corresponding to each of the second sample parameters includes:

[0025] Based on the association model, the main parameters of the target and their corresponding target response parameters are determined, and the set of EMD components corresponding to the target response parameters is obtained;

[0026] The components in the EMD component set are subjected to time-series prediction using a Long Short-Term Memory (LSTM) artificial neural network to obtain a predicted component set.

[0027] The components in the predicted component set corresponding to each target response parameter are merged to obtain the target predicted response parameter corresponding to each target response parameter;

[0028] Based on the association model and the target prediction response parameters, the target prediction main parameters corresponding to the target main parameters are obtained;

[0029] The prediction parameter sample is determined based on the target prediction response parameters and the target prediction main parameters;

[0030] The prediction robust distance is determined based on the prediction parameter sample and the second parameter sample.

[0031] Furthermore, in the method described above, determining the robust prediction distance based on the predicted parameter samples and the second parameter samples includes:

[0032] Based on the second parameter sample, obtain the robust distance, covariance matrix, and mean matrix of the polysilicon production equipment and its upstream and downstream processes;

[0033] The prediction robust distance is obtained based on the covariance matrix, the mean matrix, and the prediction parameter samples.

[0034] Another embodiment of this application provides an apparatus for predicting data from polysilicon production equipment, comprising:

[0035] The first processing module is used to acquire multiple first parameter samples. The first sample parameters in the first parameter samples include the equipment parameters of the polysilicon production equipment and the process parameters of its upstream and downstream processes. Each first parameter sample corresponds to one of multiple time points within a preset time period.

[0036] The second processing module is used to perform interference filtering on each first sample parameter in the first parameter sample according to EMD and principal component analysis to obtain the filtered second sample parameters and the second parameter sample including the second sample parameters.

[0037] The third processing module is used to perform principal component analysis on the second parameter sample to determine at least one main parameter affecting the polysilicon production equipment and its upstream and downstream processes, as well as the response parameters corresponding to each main parameter, wherein the response parameters corresponding to the main parameters are the second sample parameters in the second parameter sample other than the main parameters;

[0038] The fourth processing module is used to perform multiple linear regression based on the main parameters and the response parameters corresponding to each main parameter to obtain the correlation model of the polysilicon production equipment and its upstream and downstream processes.

[0039] The fifth processing module is used to obtain the prediction results of the polysilicon production equipment and its upstream and downstream processes based on the second parameter sample, the correlation model and the EMD component set corresponding to each second sample parameter. The prediction results include: prediction parameter sample and prediction robust distance.

[0040] Another embodiment of this application provides a server including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the polysilicon production equipment data prediction method as described above.

[0041] Another embodiment of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the polysilicon production equipment data prediction method as described above.

[0042] Compared with the prior art, the polysilicon production equipment data prediction method, apparatus, server, and storage medium provided in this application have at least the following beneficial effects:

[0043] This application combines EMD and principal component analysis for dimensionality reduction, which helps to screen out unstable components and effectively reduce noise without reducing the number of samples. It also considers the correlation between various parameters and constructs a correlation model through multiple linear regression for prediction, avoiding information silos and making the prediction results more accurate. Furthermore, it increases the prediction robustness distance, which helps to improve fault detection efficiency and further improves the accuracy and stability of the prediction results. Attached Figure Description

[0044] Figure 1 This is one of the flowcharts illustrating the data prediction method for polysilicon production equipment in this application;

[0045] Figure 2 This is the second flowchart illustrating the data prediction method for polysilicon production equipment in this application;

[0046] Figure 3 This is the third flowchart illustrating the data prediction method for polysilicon production equipment in this application;

[0047] Figure 4 This is the fourth flowchart illustrating the data prediction method for polysilicon production equipment in this application;

[0048] Figure 5 This is the fifth flowchart illustrating the data prediction method for polysilicon production equipment in this application;

[0049] Figure 6 This is the sixth flowchart illustrating the data prediction method for polysilicon production equipment in this application;

[0050] Figure 7 This is a schematic diagram of the linear regression results with parameter 4 as the main parameter in a specific embodiment;

[0051] Figure 8 This is a schematic diagram of robust distance results with parameter 4 as the main parameter in a specific embodiment;

[0052] Figure 9 This is one of the structural schematic diagrams of the device for data prediction in polysilicon production equipment according to this application;

[0053] Figure 10 This is the second schematic diagram of the device for predicting data in polysilicon production equipment according to this application. Detailed Implementation

[0054] To make the technical problems, technical solutions, and advantages of this application clearer, a detailed description will be provided below in conjunction with the accompanying drawings and specific embodiments. In the following description, specific details such as particular configurations and components are provided merely to aid in a comprehensive understanding of the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Furthermore, for clarity and brevity, descriptions of known functions and structures have been omitted.

[0055] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0056] In the various embodiments of this application, it should be understood that the sequence number of each process described below does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0057] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0058] In the embodiments provided in this application, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0059] See Figure 1 One embodiment of this application provides a data prediction method for polysilicon production equipment, including:

[0060] Step S101: Obtain multiple first parameter samples. The first sample parameters in the first parameter samples include the equipment parameters of the polysilicon production equipment and the process parameters of its upstream and downstream processes. Each first parameter sample corresponds to one of multiple time points within a preset time period.

[0061] Step S102: Based on EMD and principal component analysis, each first sample parameter in the first parameter sample is individually subjected to interference filtering to obtain the filtered second sample parameters and the second parameter sample including the second sample parameters.

[0062] Step S103: Perform principal component analysis on the second parameter sample to determine at least one main parameter affecting the polysilicon production equipment and its upstream and downstream processes, as well as the response parameters corresponding to each main parameter, wherein the response parameters corresponding to the main parameters are the second sample parameters in the second parameter sample other than the main parameters;

[0063] Step S104: Perform multiple linear regression based on the main parameters and the response parameters corresponding to each main parameter to obtain the correlation model of the polysilicon production equipment and its upstream and downstream processes;

[0064] Step S105: Based on the second parameter sample, the correlation model, and the EMD component set corresponding to each second sample parameter, the prediction results of the polysilicon production equipment and its upstream and downstream processes are obtained. The prediction results include: prediction parameter sample and prediction robust distance.

[0065] In this embodiment, when performing data prediction on polysilicon production equipment, multiple first parameter samples related to the polysilicon production equipment are first acquired. Each parameter sample corresponds to one of multiple time points within a preset time period. That is, each parameter sample is a set of first sample parameter values ​​collected at preset sampling points in the polysilicon production equipment and its upstream and downstream processes at the corresponding time point. The first sample parameters include, but are not limited to, the equipment parameters of the polysilicon production equipment and the process parameters of the upstream and downstream processes of the polysilicon production equipment. It should be noted that the preset time period can be a time period with a fixed duration or the total time period from the start of the process to the present.

[0066] It should be noted that the equipment parameters are the operating parameters of the reflux pump, such as current and amplitude, while the upstream and downstream process parameters can be material-related parameters, such as flow rate, temperature, pressure, and liquid level. In a specific embodiment, taking the washing and dust removal process in the cold hydrogenation workshop as an example, the polysilicon production equipment is a reflux pump, and the corresponding equipment parameter is the operating current. The process parameters include, but are not limited to, at least one of the following: pressure in the feed buffer tank of the coarse fractionation tower, hydraulic pressure of the condensed chlorosilane (upstream), pressure of the liquid in the bottom of the quench tower, hydraulic pressure of the condensed chlorosilane (upstream), pressure at the top of the quench tower, liquid level in the bottom of the quench tower, liquid level in the feed buffer tank of the coarse fractionation tower, temperature of the mixed gas after dust removal, temperature in the feed buffer tank of the coarse fractionation tower, flow rate of chlorosilane flowing downstream of the reflux pump into the feed buffer tank of the coarse fractionation tower, and flow rate of chlorosilane flowing downstream of the reflux pump back from the top of the quench tower to the quench tower.

[0067] After obtaining each first parameter sample, interference filtering is performed on each first sample parameter in the first parameter sample based on EMD and principal component analysis. During the filtering process, the IMF component and residual component corresponding to each first sample parameter are obtained by EMD decomposition, and principal component analysis is performed on each component to reduce the dimensionality. The IMF components that do not meet the preset conditions are removed as interference components, and the remaining components are merged to obtain new sample parameters, which are denoted as second sample parameters (stationary parameters) to ensure the stability of the filtered second sample parameters. At the same time, after converting all first sample parameters in the same first parameter sample into second sample parameters, the first parameter sample at this time is denoted as the second parameter sample. That is, in this application, the parameter types of the first sample parameters and the corresponding second sample parameters are the same, but the second sample parameters are the filtered parameters. The numerical values ​​of the two may be different. For easy distinction, the filtered parameters are denoted as second sample parameters, and the set of second sample parameters at the same time point is denoted as the second parameter sample.

[0068] It should also be noted that in this step, principal component analysis is performed on each parameter individually, not by combining all components of all parameters into a common set for dimensionality reduction, nor by directly using the dimensionality-reduced principal components as the reduced components. For example, after performing EMD and principal component analysis on the sample of pressure 1, when performing EMD and principal component analysis on pressure 2, only the components of pressure 2 can be reduced in dimensionality, and the components of pressure 1 cannot be added.

[0069] After obtaining the second parameter samples, principal component analysis will be performed again on multiple second parameter samples to determine at least one main parameter affecting polysilicon production equipment and its upstream and downstream processes. After determining at least one main parameter, the corresponding response parameter will be determined based on each main parameter, that is, the second sample parameter in the second parameter samples other than the main parameter. Then, by performing multiple linear regression on each main parameter and its corresponding response parameter, the correlation model of polysilicon production equipment and its upstream and downstream processes can be obtained. This correlation model is used to indicate the relationship between the main parameter and the response parameter, including but not limited to at least one of numerical relationship and mapping relationship.

[0070] Furthermore, based on the correlation model and the EMD component set corresponding to each second sample parameter, prediction parameters can be obtained from the prediction results of polysilicon production equipment and its upstream and downstream processes. Then, based on the prediction parameter samples and the second parameter samples, a robust distance can be obtained to obtain the robust distance in the prediction results of polysilicon production equipment and its upstream and downstream processes, thereby realizing the prediction of polysilicon production equipment and its upstream and downstream processes.

[0071] In summary, this embodiment combines EMD and principal component analysis for dimensionality reduction, which helps to screen out unstable components, thereby effectively reducing noise without reducing the number of samples. It also considers the correlation between various parameters and constructs a correlation model through multiple linear regression for prediction, avoiding information silos and making the prediction results more accurate. Furthermore, it increases the prediction robustness distance, which helps to improve fault detection efficiency and further improves the accuracy and stability of the prediction results.

[0072] In one specific embodiment, during the process of acquiring multiple first parameter samples, since the collected sample parameters may have missing samples or strong correlations between parameters, in order to ensure the temporal integrity of the sample parameters and reduce the amount of computation, each first parameter sample will be preprocessed. The preprocessing operation includes, but is not limited to, at least one of parameter validity analysis, correlation analysis, and interpolation. The following describes the preprocessing through correlation analysis and interpolation in a specific embodiment.

[0073] In this embodiment, correlation analysis involves analyzing the degree of correlation between any two parameters in the original sample data. If the correlation degree is within a preset range (R... 2 If the two parameters are within the range of 0.8, then the data corresponding to one of the parameters is reduced.

[0074] Interpolation is applied to samples with missing parameters. Specifically, in the exported data, the values ​​of missing parameters are displayed as "*". In most cases, samples with missing parameters are relatively scattered, while in a few cases they are relatively concentrated. Therefore, appropriate interpolation processes need to be performed based on their distribution.

[0075] When the number of missing parameters in multiple consecutive samples, i.e., the number of missing consecutive data, is less than or equal to the first missing threshold (e.g., 10 samples), data is supplemented by linear interpolation.

[0076] When the continuous data missing amount is greater than the first missing amount threshold but less than or equal to the second missing amount threshold (e.g., 50 samples), a preset number (e.g., 5 to 10 samples) of reference sample data on both sides of the data missing point are obtained, and data interpolation is performed to supplement the data based on the mean of the reference sample data, a bidirectional sliding data window, and linear proportional interpolation. For example, the mean of the reference sample data plus the bidirectional sliding data window is used to calculate the predicted value of each data point, and the final filling value is calculated using a linear proportional method.

[0077] When the amount of missing continuous data is greater than the second missing data threshold, the device or apparatus corresponding to the sampling point is determined to be abnormal due to the lack of large continuous sample data, and therefore the corresponding parameters will be filtered out.

[0078] It should be noted that the reason for maintaining the integrity of the parameter timing is mainly to facilitate subsequent timing prediction of the parameters.

[0079] Preferably, the method described above further includes:

[0080] The polysilicon production equipment is controlled based on the prediction results of the polysilicon production equipment and its upstream and downstream processes.

[0081] In this embodiment, after obtaining the above-mentioned prediction results, the polysilicon production equipment will also be controlled according to the prediction results. Specifically, the polysilicon production equipment will be controlled according to the trend of the prediction results. At the same time, the equipment corresponding to the upstream and downstream processes can also be controlled to achieve at least the adjustment of the corresponding processes and ensure the normal operation of the equipment and processes.

[0082] See Figure 2 Specifically, in the method described above, the step of individually performing interference filtering on each first sample parameter in the first parameter sample based on EMD and principal component analysis to obtain filtered second sample parameters and a second parameter sample including the second sample parameters includes:

[0083] Step S201: Perform EMD on each of the first sample parameters to obtain multiple IMF components and residual components;

[0084] Step S202: Perform principal component analysis on each of the IMF components and the residual components according to the first cumulative contribution rate threshold to obtain the EMD component set of the second sample parameters corresponding to each of the first sample parameters;

[0085] Step S203: Merge the components in each of the EMD component sets to obtain the corresponding second sample parameters and the second parameter samples.

[0086] In this embodiment, when performing interference filtering on each first sample parameter individually, EMD is specifically performed on each first sample parameter to obtain multiple IMF components and residual components. For example, the components of each parameter are denoted as IMFi1, IMFj2, IMFk3, ..., IMFqn (where the suffixes i, j, k, q, etc., represent a certain intrinsic mode function component of a parameter, and the suffixes 1, 2, 3, ..., n belong to the parameters of each mode). In addition to the IMF components, each parameter also has a residual component, denoted as Res1, Res2, Res3, ..., Resn. Further, principal component analysis is performed on the IMF components and residual components to remove interference components. The method for removing interference components is not to directly merge them into several fewer principal component components through principal component analysis, but rather to determine the number of principal components through principal component analysis and reduce the number of IMF components so that the number of residual components for each parameter is equal to the number of principal components. Specifically, the method involves determining the number of principal components (IMFs) as 'a' by using the first cumulative contribution rate of principal component analysis (PCA) to ensure it exceeds a threshold (e.g., 90%). Then, taking the IMF components of parameter X1 (totaling b, but actually b > a) IMFi1 as an example, components are removed directly starting from IMF11 until IMF(ab)1 is removed. The remaining components (including residual components) are the retained components. In simpler terms, the interference components are the first 1-3 components of each parameter.

[0087] The set of remaining components after filtering out the interference component is denoted as the EMD component set of the second sample parameter corresponding to the first sample parameter. By merging the components (including IMF components and residual components) in this EMD component set, the second sample parameter can be obtained. All the second sample parameters corresponding to the first sample parameters in the same first parameter sample constitute the second parameter sample.

[0088] See Figure 3 Specifically, as described above, the step of performing principal component analysis on the second parameter sample to determine at least one major parameter affecting the polysilicon production equipment and its upstream and downstream processes, and the response parameters corresponding to each major parameter, includes:

[0089] Step S301: Perform principal component analysis on the second parameter sample according to the second cumulative contribution rate threshold to obtain the load matrix, score matrix and eigenvalues ​​corresponding to the polysilicon production equipment and its upstream and downstream processes;

[0090] Step S302: Obtain the principal component equation based on the load matrix and the score matrix, and determine the second sample parameter existing in the principal component equation as the principal parameter;

[0091] Step S303: Determine the response parameters corresponding to each of the main parameters based on the main parameters and the second sample parameters.

[0092] In this embodiment, when performing principal component analysis on the second parameter sample, dimensionality reduction is performed based on a second cumulative contribution rate threshold (e.g., 85%) to obtain the load matrix, score matrix, and eigenvalues ​​corresponding to the polysilicon production equipment and its upstream and downstream processes. The load matrix, score matrix, and eigenvalues ​​are important results obtained from the principal component analysis. The load matrix reflects the importance of the corresponding parameters, the score matrix reflects the results obtained after dimensionality reduction, and the eigenvalues ​​reflect the importance of each principal component to the principal component analysis model. The specific acquisition method is not detailed here. Specifically, the method for determining the main parameters affecting the polysilicon production equipment and its upstream and downstream processes involves obtaining the principal component equation based on the load matrix and score matrix. This principal component equation is as follows:

[0093]

[0094] Among them, a ij x is an element in the load matrix. j Let be the second sample parameter with the j-th subscript in the second parameter sample, and n be the number of all second sample parameters in the second parameter sample. PC i Let be the parameter of the i-th principal component, and k be the total number of principal component parameters. Finally, determine the largest |a| in the expression for each principal component PCi. ij | Corresponding parameter x j These are the main parameters of the polysilicon production equipment and its upstream and downstream processes.

[0095] That is, the sample parameters existing in the principal component equation are identified as the main parameters affecting the polysilicon production equipment and its upstream and downstream processes, and backups are made to be used to construct the correlation model of the polysilicon production equipment and its upstream and downstream processes.

[0096] At this point, the cumulative contribution rate threshold for principal component analysis dimensionality reduction is 85%, which can also be customized by the user. The principal parameters are determined by the maximum absolute value of each element in the loading vector corresponding to each principal component. Correspondingly, all other parameters are response parameters. It is important to emphasize that there is not necessarily one target parameter. When there are multiple target parameters, each target parameter is used as a reference, and all other parameters are response parameters (for example, A and B are both principal parameters; when A is the principal parameter, B is also considered a response parameter of A; when B is the principal parameter, A is also considered a target parameter of B. Parameter A is the principal parameter of principal component equation a, and parameter B is the principal parameter of principal component equation b).

[0097] See Figure 4Specifically, as described above, the step of performing multiple linear regression based on the main parameters and the corresponding response parameters to obtain the correlation model of the polysilicon production equipment and its upstream and downstream processes includes:

[0098] Step S401: Through linear fitting, a multiple linear regression model is obtained between the main parameters and their corresponding response parameters.

[0099] Step S402: Obtain the T-distribution test value and F-distribution test value of each of the multiple linear regression models;

[0100] Step S403: Determine the multiple linear regression model whose T-distribution test value is greater than the corresponding first threshold and whose F-distribution test value is greater than the corresponding first threshold as the correlation model.

[0101] In this embodiment, when obtaining the correlation model of polysilicon production equipment and its upstream and downstream processes, a linear fit is performed between the main parameters and the corresponding response parameters to obtain the corresponding multiple linear regression model. The main parameters are the target parameters in the multiple linear regression model. The method for obtaining the multiple linear regression model is relatively advanced and will not be elaborated here. After obtaining the multiple linear regression model, it is screened based on the model's significance and goodness of fit. Significance is determined by the t-distribution test value of coefficient analysis and the F-distribution test value of variance analysis (e.g., a p-value less than 0.05 is considered significant), and the goodness of fit is determined by R-sq (e.g., R-sq greater than 0.8 is considered a good fit). If any one of these criteria is not met, the corresponding model is invalid. If all models are invalid, the parameters need to be reselected, and the process must start again from the data acquisition point. It should be noted that, due to the relatively "hidden" correlation between upstream and downstream processes of moving equipment such as quench tower circulating pumps in polysilicon production, the situation where all models are invalid is generally not present, thus ensuring the effectiveness of the method used in this embodiment.

[0102] See Figure 5 Specifically, as described above, obtaining the prediction results of the polysilicon production equipment and its upstream and downstream processes based on the second parameter sample, the correlation model, and the EMD component set corresponding to each of the second sample parameters includes:

[0103] Step S501: Determine the main parameters of the target and their corresponding target response parameters according to the association model, and obtain the EMD component set corresponding to the target response parameters;

[0104] Step S502: Perform LSTM time series prediction on the components in the EMD component set to obtain the predicted component set;

[0105] Step S503: Merge the components in the predicted component set corresponding to each target response parameter to obtain the target predicted response parameter corresponding to each target response parameter;

[0106] Step S504: Based on the association model and the target prediction response parameters, obtain the target prediction main parameters corresponding to the target main parameters;

[0107] Step S505: Determine the prediction parameter sample based on the target prediction response parameters and the target prediction main parameters;

[0108] Step S506: Determine the prediction robust distance based on the prediction parameter sample and the second parameter sample.

[0109] In this embodiment, when obtaining the prediction results of polysilicon production equipment and its upstream and downstream processes, the target principal parameters and target response parameters are determined based on the multiple linear regression model corresponding to the determined correlation model. Then, the EMD component set is obtained again through EMD or based on the aforementioned EMD and principal component analysis of each second sample parameter. The EMD component set corresponding to each target response parameter is then determined. Furthermore, LSTM time-series prediction is performed on the components in the EMD component set corresponding to each target response parameter to obtain the corresponding prediction components, thus obtaining the prediction component set. Specifically, since the LSTM model is a special neural network model, it can solve the gradient vanishing and gradient exploding problems in the long sequence training process. In the modeling process, historical samples are divided into training and test sets, while future time-series prediction continues to predict backwards based on the previous model (specifically, a sliding data window is used). Therefore, when performing LSTM time series prediction, it is necessary to determine the training set and test set for each EMD component in each target response parameter, obtain the LSTM model, and then obtain the fitted values ​​of the training set and test set for each EMD component in each target response parameter of the LSTM model. The predicted value of each EMD component in each target response parameter is obtained by using a sliding data window.

[0110] It should be noted that, due to the significant randomness of the LSTM model fitting results, this implementation employs a method of repeatedly selecting the optimal objective index. Three metrics are used to evaluate the LSTM model: Mean Square Error (MSE), Mean Absolute Error (MAE), and Goodness of Fit Rfit. 2 Because under the same training samples, the MSE, MAE, and R-values ​​of different models built repeatedly vary. 2 All three are different, and their trends are correlated; therefore, the goodness of fit R is used here.2 As the target indicator, select the largest R 2 The values ​​correspond to the LSTM model. Results from multiple experiments show that with 3-5 sets of experiments, each repeated more than 50 times, the final fitting results are consistent.

[0111] After obtaining the set of predicted components corresponding to each target response parameter, the target predicted response parameters corresponding to each target response parameter can be obtained by merging the predicted components in the same set of predicted components. Furthermore, based on the aforementioned association model obtained from the response parameters and principal parameters, the corresponding target predicted principal parameters can be obtained from each target predicted response parameter. The target predicted principal parameters and each target predicted response parameter can constitute the prediction parameter sample.

[0112] After obtaining the prediction parameter samples, the robust prediction distance corresponding to the prediction parameter samples can be calculated by combining them with the second parameter samples mentioned above.

[0113] See Figure 6 Furthermore, in the method described above, determining the robust prediction distance based on the predicted parameter sample and the second parameter sample includes:

[0114] Step S601: Based on the second parameter sample, obtain the robust distance, covariance matrix, and mean matrix of the polysilicon production equipment and its upstream and downstream processes.

[0115] Step S602: Obtain the prediction robust distance based on the covariance matrix, the mean matrix, and the prediction parameter samples.

[0116] In this embodiment, when obtaining the predicted robust distance, the robust distances (i.e., historical robust distances) of the polysilicon production equipment and its upstream and downstream processes are first obtained based on the aforementioned multiple second parameter samples. The covariance matrix and mean matrix involved in calculating the robust distance are also obtained. The robust distance is an optimized model of Mahalanobis distance, based on the fast minimum covariance determinant algorithm. Since it is already publicly available, it will not be elaborated upon here. The formula for calculating the predicted robust distance, based on the covariance matrix and mean matrix obtained when acquiring the historical robust distance, is as follows:

[0117]

[0118] Where, x m is a vector of all parameter values ​​for the m-th predicted parameter sample; n is the number of second parameter samples; T is the mean matrix; S is the covariance matrix.

[0119] This predictive robust distance can improve fault detection efficiency, build a robust space for more accurate prediction, and serve as an important basis for judging overall stability.

[0120] To facilitate understanding by those skilled in the art, the following example illustrates 12 parameters (including equipment parameters and process parameters of upstream and downstream processes) of a circulating pump A (polysilicon production equipment) in the cold hydrogenation process: X1 Pressure 1, X2 Pressure 2, X3 Pressure 3, X4 Pressure 4, X5 Pump A Current, X6 Liquid Level 4, X7 Liquid Level 5, X8 Temperature 1, X9 Temperature 5, X10 Flow Control 1, X11 Flow Control 3, X12 Flow Control 4.

[0121] The parameter attributes and specific meanings of these 12 parameters are shown in Table 1.

[0122] Table 1

[0123]

[0124]

[0125] A total of 1032 samples were selected from April 20, 2022 to June 1, 2022, at hourly intervals. Due to faults between April 25 and 28, fluctuations occurred, necessitating noise reduction. This involved EMD decomposition and PCA dimensionality reduction to remove components (cumulative contribution rate of 90%, a threshold obtained experimentally; if the cumulative contribution rate is below 90%, historical data is not fully optimized into healthy sample data), resulting in the denoised parameters. The parameters before and after processing are shown in Table 2.

[0126] Table 2

[0127]

[0128]

[0129] The parameters X1', X2', X3', X4', X5', X6', X7', X8', X9', X10', X11', and X12' are used to form a new historical data sample matrix. PCA dimensionality reduction is then performed (with a cumulative contribution rate threshold of 85%), resulting in four principal component equations. Based on the parameters corresponding to the maximum absolute values ​​of the elements in each principal component loading vector, the principal parameters are selected as X2', X4', X5', and X7'. Four multiple linear regression equations are obtained using these four principal parameters. The R-sq of the multiple linear regression equation corresponding to the principal parameter X4' is 94.72%, while the R-sq of the multiple linear regression equations corresponding to the other principal parameters are all less than 80%. Furthermore, the p-values ​​corresponding to the T-test and F-test values ​​for each parameter are all less than 0.05. Therefore, parameter X4' is selected as the target parameter for the linear regression model.

[0130] For the remaining components, LSTM modeling was performed. Of the 1032 samples, 80% were used as the training set and 20% as the test set. Future time series prediction was also performed, forecasting 24 future samples (i.e., samples from the next day). In this embodiment, the LSTM modeling process used the LSTM(a) function from the keras.layers database in Python, where 'a' was set to 128, and the number of training samples was represented by the batch_size. The LSTM modeling was repeated 100 times, and the goodness of fit R was selected. 2 The highest-performing model. In practice, the highest fit R of LSTM modeling for each component is achieved. 2 All are above 95%.

[0131] The component values ​​of all 24 predicted samples were combined according to their corresponding parameters, and the results were substituted into a linear regression model with the principal parameter X4' as the target parameter. The results are as follows. Figure 7 As shown, the fitting results are good, and the last 24 future prediction samples show a decreasing and stabilizing trend. In fact, on June 2nd, the corresponding parameter X4 (i.e., pressure 4) did indeed decrease, with the data before noise reduction dropping from 3.41 MPa to 3.28 MPa. Figure 7 The results were similar. Other parameters also matched the expected results for June 2nd.

[0132] Furthermore, a robust distance model was established using the denoised X1', X2', X3', X4', X5', X6', X7', X8', X9', X10', X11', and X12', and the corresponding 24 sets of future prediction samples were input. The corresponding robust distances are as follows: Figure 8 As shown. The threshold value is... p=12. It is clear that the robust distances of the denoised data did not exceed the threshold, indicating that the denoised samples were all in a healthy state. It can also be observed that the robust distances of the predicted 24 samples did not exceed the threshold either. In fact, no malfunctions occurred on June 2nd, and the equipment operated smoothly.

[0133] See Figure 9 Another embodiment of this application also provides an apparatus for predicting data from polysilicon production equipment, comprising:

[0134] The first processing module 901 is used to acquire multiple first parameter samples. The first sample parameters in the first parameter samples include the equipment parameters of the polysilicon production equipment and the process parameters of its upstream and downstream processes. Each first parameter sample corresponds to one of multiple time points within a preset time period.

[0135] The second processing module 902 is used to perform interference filtering on each first sample parameter in the first parameter sample according to EMD and principal component analysis to obtain the filtered second sample parameters and the second parameter sample including the second sample parameters.

[0136] The third processing module 903 is used to perform principal component analysis on the second parameter sample to determine at least one main parameter affecting the polysilicon production equipment and its upstream and downstream processes, as well as the response parameters corresponding to each main parameter, wherein the response parameters corresponding to the main parameters are the second sample parameters in the second parameter sample other than the main parameters;

[0137] The fourth processing module 904 is used to perform multiple linear regression based on the main parameters and the response parameters corresponding to each main parameter to obtain the correlation model of the polysilicon production equipment and its upstream and downstream processes.

[0138] The fifth processing module 905 is used to obtain the prediction results of the polysilicon production equipment and its upstream and downstream processes based on the second parameter sample, the correlation model and the EMD component set corresponding to each second sample parameter. The prediction results include: prediction parameter sample and prediction robust distance.

[0139] Preferably, in the apparatus described above, the apparatus further includes:

[0140] The sixth processing module is used to control the polysilicon production equipment based on the prediction results of the polysilicon production equipment and its upstream and downstream processes.

[0141] Specifically, in the method described above, the second processing module includes:

[0142] The first processing unit is used to perform EMD on each of the first sample parameters to obtain multiple IMF components and residual components.

[0143] The second processing unit is used to perform principal component analysis on each of the IMF components and the residual components according to the first cumulative contribution rate threshold, so as to obtain the EMD component set of the second sample parameters corresponding to each of the first sample parameters;

[0144] The third processing unit is used to merge the components in each of the EMD component sets to obtain the corresponding second sample parameters and the second parameter samples.

[0145] Specifically, in the apparatus described above, the third processing module includes:

[0146] The fourth processing unit is used to perform principal component analysis on the second parameter sample according to the second cumulative contribution rate threshold to obtain the load matrix, score matrix and eigenvalues ​​corresponding to the polysilicon production equipment and its upstream and downstream processes.

[0147] The fifth processing unit is used to obtain the principal component equation based on the load matrix and the score matrix, and to determine the second sample parameter existing in the principal component equation as the principal parameter;

[0148] The sixth processing unit is used to determine the response parameter corresponding to each of the main parameters based on the main parameters and the second sample parameters.

[0149] Specifically, in the apparatus described above, the fourth processing module includes:

[0150] The seventh processing unit is used to obtain a multiple linear regression model between the main parameters and their corresponding response parameters through linear fitting.

[0151] The eighth processing unit is used to obtain the T-distribution test value and F-distribution test value of each of the multiple linear regression models.

[0152] The ninth processing unit is used to determine that the multiple linear regression model whose T-distribution test value is greater than the corresponding first threshold and whose F-distribution test value is greater than the corresponding first threshold is the correlation model.

[0153] Specifically, in the aforementioned device, the fifth processing module includes:

[0154] The tenth processing unit is used to determine the main parameters of the target and their corresponding target response parameters according to the association model, and to obtain the set of EMD components corresponding to the target response parameters;

[0155] The eleventh processing unit is used to perform long short-term memory (LSTM) time-series prediction on the components in the EMD component set to obtain a predicted component set.

[0156] The twelfth processing unit is used to merge the components in the predicted component set corresponding to each of the target response parameters to obtain the target predicted response parameters corresponding to each of the target response parameters.

[0157] The thirteenth processing unit is used to obtain the target prediction main parameters corresponding to the target main parameters based on the association model and the target prediction response parameters;

[0158] The fourteenth processing unit is used to determine the prediction parameter sample based on the target prediction response parameters and the target prediction main parameters;

[0159] The fifteenth processing unit is used to determine the prediction robust distance based on the prediction parameter sample and the second parameter sample.

[0160] Furthermore, in the apparatus described above, the fifteenth processing unit is configured to include:

[0161] The first sub-processing unit is used to obtain the robust distance, covariance matrix, and mean matrix of the polysilicon production equipment and its upstream and downstream processes based on the second parameter sample.

[0162] The second sub-processing unit is used to obtain the prediction robust distance based on the covariance matrix, the mean matrix, and the prediction parameter samples.

[0163] The device embodiment of this application is a device corresponding to the embodiment of the above-described polysilicon production equipment data prediction method. All implementation means in the above-described method embodiment are applicable to the device embodiment and can achieve the same technical effect.

[0164] See Figure 10 In another embodiment of this application, the apparatus for predicting data from polysilicon production equipment includes:

[0165] The collection module 1001 (first processing module) mainly collects parameter samples from various sampling points on-site within the corresponding process range of the polysilicon production equipment.

[0166] The preprocessing module 1002, while retaining the interlock values ​​and warning values ​​of each parameter, incorporates algorithms for EMD decomposition, PCA dimensionality reduction, component screening, and merging. It is used to perform EMD decomposition on each parameter and PCA dimensionality reduction to screen out interference components, thereby performing noise reduction processing.

[0167] The master data determination module 1003 mainly uses the processed full-parameter principal component analysis to find the main parameters, and the corresponding remaining parameters are used as response parameters.

[0168] Modeling module 1004 mainly includes multiple linear regression and LSTM algorithms, as well as corresponding methods for determining the effectiveness of the model. The multiple linear regression algorithm only requires input of target parameter information and response parameter samples, while the LSTM algorithm requires manual adjustment of the input repetition count and the number of training sets.

[0169] The prediction module 1005 mainly receives the model established by the modeling module and uses the model and the number of samples input for future time series prediction to perform full-parameter future time series prediction.

[0170] The stability module 1006 mainly determines whether the data is stable by calculating the robust distance of all parameters and all samples and setting a threshold.

[0171] The coordination and invocation module 1007 is mainly used to retrieve and allocate resources from the storage module according to the needs of the corresponding instructions of the above modules.

[0172] Storage module 1008 is mainly used to store samples and corresponding calculation examples, steps, and fault point markers.

[0173] Display module 1009 is mainly used to display the time series graphs of various parameters and indicators, as well as the health status of the parameters.

[0174] Output module 10010 is used to output and export the processing results of each step.

[0175] Another embodiment of this application provides a server including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the polysilicon production equipment data prediction method as described above.

[0176] Another embodiment of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the polysilicon production equipment data prediction method as described above.

[0177] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0178] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.

[0179] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A data prediction method for polysilicon production equipment, characterized in that, include: Multiple first parameter samples are obtained. The first sample parameters in the first parameter samples include the equipment parameters of the polysilicon production equipment and the process parameters of its upstream and downstream processes. Each first parameter sample corresponds to one of multiple time points within a preset time period. Based on empirical mode decomposition (EMD) and principal component analysis, each first sample parameter in the first parameter sample is individually subjected to interference filtering to obtain the filtered second sample parameters and the second parameter sample including the second sample parameters. The principal component analysis is performed on the second parameter sample to determine at least one major parameter affecting the polysilicon production equipment and its upstream and downstream processes, as well as the response parameters corresponding to each major parameter, wherein the response parameters corresponding to the major parameters are the second sample parameters in the second parameter sample other than the major parameters; Based on the main parameters and the corresponding response parameters, a multiple linear regression is performed to obtain the correlation model of the polysilicon production equipment and its upstream and downstream processes. Based on the second parameter sample, the correlation model, and the EMD component set corresponding to each second sample parameter, the prediction results of the polysilicon production equipment and its upstream and downstream processes are obtained. The prediction results include: prediction parameter sample and prediction robust distance.

2. The method according to claim 1, characterized in that, The method further includes: The polysilicon production equipment is controlled based on the prediction results of the polysilicon production equipment and its upstream and downstream processes.

3. The method according to claim 1, characterized in that, The step of performing individual interference filtering on each first sample parameter in the first parameter sample based on Empirical Mode Decomposition (EMD) and Principal Component Analysis (PCA) to obtain filtered second sample parameters and a second parameter sample including the second sample parameters includes: EMD is performed on each of the first sample parameters to obtain multiple intrinsic mode function (IMF) components and residual components; Principal component analysis is performed on each of the IMF components and the residual components according to the first cumulative contribution rate threshold to obtain the set of EMD components of the second sample parameters corresponding to each of the first sample parameters; The components in each of the EMD component sets are merged to obtain the corresponding second sample parameters and the second parameter samples.

4. The method according to claim 1, characterized in that, The principal component analysis performed on the second parameter sample to determine at least one major parameter affecting the polysilicon production equipment and its upstream and downstream processes, and the response parameters corresponding to each major parameter, includes: Based on the second cumulative contribution rate threshold, the principal component analysis is performed on the second parameter sample to obtain the load matrix, score matrix and eigenvalues ​​corresponding to the polysilicon production equipment and its upstream and downstream processes. The principal component equation is obtained based on the loading matrix and the score matrix, and the second sample parameter existing in the principal component equation is determined as the principal parameter. Based on the main parameters and the second sample parameters, the response parameters corresponding to each of the main parameters are determined.

5. The method according to claim 1, characterized in that, The step of performing multiple linear regression based on the main parameters and the corresponding response parameters to obtain the correlation model of the polysilicon production equipment and its upstream and downstream processes includes: By linear fitting, a multiple linear regression model is obtained between the main parameters and their corresponding response parameters. Obtain the T-distribution test value and F-distribution test value for each of the aforementioned multiple linear regression models; The multiple linear regression model whose T-distribution test value is greater than the corresponding first threshold and whose F-distribution test value is greater than the corresponding first threshold is identified as the correlation model.

6. The method according to claim 1, characterized in that, The step of obtaining the prediction results of the polysilicon production equipment and its upstream and downstream processes based on the second parameter sample, the correlation model, and the EMD component set corresponding to each second sample parameter includes: Based on the association model, the main parameters of the target and their corresponding target response parameters are determined, and the set of EMD components corresponding to the target response parameters is obtained; Perform time-series prediction on the components in the EMD component set using a Long Short-Term Memory Artificial Neural Network (LSTM) to obtain a predicted component set. The components in the predicted component set corresponding to each target response parameter are merged to obtain the target predicted response parameter corresponding to each target response parameter; Based on the association model and the target prediction response parameters, the target prediction main parameters corresponding to the target main parameters are obtained; The prediction parameter sample is determined based on the target prediction response parameters and the target prediction main parameters; The prediction robust distance is determined based on the prediction parameter sample and the second parameter sample.

7. The method according to claim 6, characterized in that, Determining the robust prediction distance based on the predicted parameter sample and the second parameter sample includes: Based on the second parameter sample, obtain the robust distance, covariance matrix, and mean matrix of the polysilicon production equipment and its upstream and downstream processes; The prediction robust distance is obtained based on the covariance matrix, the mean matrix, and the prediction parameter samples.

8. A device for predicting data in polysilicon production equipment, characterized in that, include: The first processing module is used to acquire multiple first parameter samples. The first sample parameters in the first parameter samples include the equipment parameters of the polysilicon production equipment and the process parameters of its upstream and downstream processes. Each first parameter sample corresponds to one of multiple time points within a preset time period. The second processing module is used to perform interference filtering on each first sample parameter in the first parameter sample according to EMD and principal component analysis to obtain the filtered second sample parameters and the second parameter sample including the second sample parameters. The third processing module is used to perform principal component analysis on the second parameter sample to determine at least one main parameter affecting the polysilicon production equipment and its upstream and downstream processes, as well as the response parameters corresponding to each main parameter, wherein the response parameters corresponding to the main parameters are the second sample parameters in the second parameter sample other than the main parameters; The fourth processing module is used to perform multiple linear regression based on the main parameters and the response parameters corresponding to each main parameter to obtain the correlation model of the polysilicon production equipment and its upstream and downstream processes. The fifth processing module is used to obtain the prediction results of the polysilicon production equipment and its upstream and downstream processes based on the second parameter sample, the correlation model and the EMD component set corresponding to each second sample parameter. The prediction results include: prediction parameter sample and prediction robust distance.

9. A server, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the polysilicon production equipment data prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the steps of the polysilicon production equipment data prediction method as described in any one of claims 1 to 7.

Images