A polysilicon production control method and device, a server and a storage medium

Abstract

This application provides a polysilicon production control method, apparatus, server, and storage medium for equipment health management. The method includes: performing dynamic principal component analysis on historical parameter samples of the polysilicon equipment and its upstream and downstream processes to obtain a dynamic principal component analysis model, including the dynamic principal component analysis equation and the Hotelling statistic and SPE test values ​​for each principal component parameter; obtaining a historical coupling model of the polysilicon equipment and its upstream and downstream processes based on the average value of each principal component parameter within the common time period with the smallest parameter fluctuations and the dynamic principal component analysis equation; obtaining a first coupling model based on multiple linear regression and a second coupling model based on principal component analysis and multiple linear regression based on the dynamic principal component analysis model; and controlling the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes of the polysilicon equipment according to the target coupling model after validity verification.

Description

Technical Field

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

[0002] Equipment fault diagnosis and early warning are crucial in the chemical industry, not only to ensure the safe operation of equipment and timely maintenance under abnormal conditions, but also to ensure the safety and stability of chemical processes.

[0003] In polysilicon production, health management of equipment such as compressors and heaters relies primarily on manufacturer-defined warning and interlocking thresholds, as well as equipment-related calculation equations (such as heat transfer coefficients), in addition to visually observable operating conditions. However, most equipment failures stem from issues with process conditions. Statistics show a high number of equipment failures caused by abnormal process conditions; therefore, establishing a relationship between process conditions and equipment parameters is necessary.

[0004] Currently, many published patents use traditional multiple linear regression to directly establish relationships between multiple parameters, covering fields such as stocks, construction, welding control, and privacy protection. However, simple multiple linear regression can easily overlook the dynamic characteristics of parameters in chemical processes such as polysilicon manufacturing. Therefore, a method is needed that takes into account the dynamic characteristics of each parameter.

[0005] Furthermore, existing multiple linear regression methods are based on existing normal data. However, this deviates from the relationship between the parameters under ideal conditions. Therefore, an ideal model is also needed as a reference to provide guidance. Summary of the Invention

[0006] The technical objective of this application is to provide a polysilicon production control method, device, server, and storage medium to solve the problem that the relationship between parameters in the current process flow, which is directly established through traditional multiple linear regression, cannot take into account the dynamic characteristics of the parameters and the relationship between the parameters, and deviates from the relationship between the parameters under ideal conditions.

[0007] To address the aforementioned technical problems, embodiments of this application provide a polysilicon production control method, comprising:

[0008] Dynamic Principal Component Analysis (DPCA) is performed on historical parameter samples to obtain a dynamic principal component analysis model. This model includes the dynamic principal component analysis equation and the Hotelling statistic test values ​​(T0) of each principal component parameter in the equation. 2 The squared prediction error (SPE) test value is used, and the parameters in the historical parameter sample include: the equipment parameters of the polysilicon equipment and the process parameters of its upstream and downstream processes;

[0009] Based on the sample time series diagram of the principal component parameters, the average value corresponding to each principal component parameter within the target time period is obtained, wherein the target time period is the common time period when the parameter fluctuations corresponding to each principal component parameter are the smallest;

[0010] Based on the average value of each principal component parameter within the target time period and the dynamic principal component analysis equation, a historical coupling model of the polysilicon equipment and its upstream and downstream processes is obtained.

[0011] Based on the dynamic principal component analysis model, a first coupling model based on multiple linear regression and a second coupling model based on principal component analysis and multiple linear regression are obtained.

[0012] The validity of the historical coupling model, the first coupling model, and the second coupling model is verified to obtain the target coupling model that passes the verification.

[0013] Based on the target coupling model, the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes of the polysilicon equipment are controlled.

[0014] Specifically, the method described above, based on the target coupling model, controls the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes of the polysilicon equipment, including:

[0015] Obtain real-time parameter samples of the polysilicon equipment and its upstream and downstream processes;

[0016] Error analysis is performed on the real-time parameter samples based on the target coupling model to obtain the analysis results;

[0017] Based on the analysis results, control the polysilicon equipment and / or the upstream and downstream process equipment corresponding to the polysilicon equipment.

[0018] Specifically, as described above, the step of performing dynamic principal component analysis on the acquired historical parameter samples to obtain a dynamic principal component analysis model includes:

[0019] Dynamic principal component analysis is performed on the historical parameter samples to determine each principal component parameter, the eigenvalues ​​corresponding to each principal component parameter, and the score matrix composed of the principal component parameters.

[0020] The loading matrix is ​​obtained based on the eigenvalues ​​corresponding to each principal component parameter;

[0021] Based on the score matrix, the loading matrix, and the historical parameter samples, the dynamic principal component equation, the Hotelling statistic test value, and the SPE test value of each principal component parameter are obtained and used as the dynamic principal component analysis model.

[0022] Specifically, as described above, the step of obtaining the historical coupling model of the polysilicon equipment and its upstream and downstream processes based on the average value of each principal component parameter within the target time period and the dynamic principal component analysis equation includes:

[0023] Substitute the average value corresponding to the principal component parameter into the dynamic principal component analysis equation to obtain the numerical value corresponding to each principal component parameter;

[0024] Based on the absolute value of the stated values, determine the first important parameter among the principal component parameters;

[0025] Based on the first important parameter, the corresponding dynamic principal component equation is transformed to obtain the historical coupling model. In the historical coupling model, the first important parameter is the first objective parameter, and the other parameters are the first response parameters.

[0026] Furthermore, according to the method described above, obtaining the first coupled model based on multiple linear regression and the second coupled model based on principal component analysis and multiple linear regression, respectively, based on the dynamic principal component analysis model, includes:

[0027] Based on the Hotelling statistic test value and the SPE test value of each principal component parameter in the dynamic principal component analysis model and their corresponding F distribution, determine the Hotelling statistic test threshold and the SPE test threshold corresponding to each dynamic principal component analysis model.

[0028] Based on the Hotling statistic test threshold and the SPE test threshold, the principal component parameters in the score matrix are screened out to obtain the target score matrix after screening.

[0029] Based on the target score matrix, the load matrix, and the historical parameter samples, the target dynamic principal component equation is obtained;

[0030] The first coupled model is obtained by performing multiple linear regression based on the target dynamic principal component equation.

[0031] Based on the target dynamic principal component equation, principal component analysis and multiple linear regression are performed to obtain the second coupled model.

[0032] Specifically, in the method described above, the step of performing multiple linear regression based on the target dynamic principal component equation to obtain the first coupled model includes:

[0033] Based on the absolute value of each principal component parameter in the target dynamic principal component equation, determine the second important parameter among each principal component parameter and the second response parameter corresponding to the second important parameter;

[0034] The second important parameter is used as the second objective parameter, and the first coupling model is obtained by performing the multiple linear regression with the corresponding second response parameter.

[0035] Specifically, in the method described above, the step of performing principal component analysis and multiple linear regression based on the target dynamic principal component equation to obtain the second coupled model includes:

[0036] Based on the absolute value of each principal component parameter in the target dynamic principal component equation, determine the third important parameter among each principal component parameter and the third response parameter corresponding to the third important parameter;

[0037] Principal component analysis is performed on each of the third response parameters to obtain the reference principal component parameters corresponding to each of the third response parameters.

[0038] The third important parameter is used as the target parameter, and a multiple linear regression is performed with the corresponding reference principal component parameter to obtain the second coupled model.

[0039] Another embodiment of this application provides a polysilicon production control device, comprising:

[0040] The first processing module is used to perform dynamic principal component analysis on historical parameter samples to obtain a dynamic principal component analysis model. The dynamic principal component analysis model includes a dynamic principal component analysis equation and the Hotelling statistic test value and SPE test value of each principal component parameter in the dynamic principal component analysis equation. The parameters in the historical parameter sample include: equipment parameters of polysilicon equipment and process parameters of its upstream and downstream processes.

[0041] The second processing module is used to obtain the average value of each principal component parameter within a target time period based on the sample time series diagram of the principal component parameters, wherein the target time period is the common time period when the parameter fluctuations of each principal component parameter are the smallest.

[0042] The third processing module is used to obtain a historical coupling model of the polysilicon equipment and its upstream and downstream processes based on the average value of each principal component parameter within the target time period and the dynamic principal component analysis equation.

[0043] The fourth processing module is used to obtain a first coupled model based on multiple linear regression and a second coupled model based on principal component analysis and multiple linear regression according to the dynamic principal component analysis model.

[0044] The fifth processing module is used to perform validity verification on the historical coupling model, the first coupling model, and the second coupling model, and obtain the target coupling model that has passed the verification.

[0045] The sixth processing module is used to control the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes of the polysilicon equipment according to the target coupling model.

[0046] 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 control method as described above.

[0047] 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 control method described above.

[0048] Compared with the prior art, the polysilicon production control method, apparatus, server, and storage medium provided in this application have at least the following beneficial effects:

[0049] This application establishes the relationships between various parameters in the process flow through dynamic principal component analysis, breaking down the information silos between parameters. It also takes into account the dynamic characteristics of the parameters, which is beneficial for monitoring and process analysis, and can be integrated with time series forecasting. In addition to constructing the first coupling model directly through multiple linear regression, this application also constructs a historical coupling model based on the "ideal state" and constructs a second coupling model by combining principal component analysis dimensionality reduction and multiple linear regression, filling the current technical gap of not having an ideal model. At the same time, this application's method can be used with different sample sizes and aggregation degrees, improving its applicability. Attached Figure Description

[0050] Figure 1 This is one of the flowcharts illustrating the polysilicon production control method of this application;

[0051] Figure 2This is the second schematic diagram of the polysilicon production control method of this application;

[0052] Figure 3 This is the third flowchart illustrating the polysilicon production control method of this application;

[0053] Figure 4 This is the fourth flowchart illustrating the polysilicon production control method of this application;

[0054] Figure 5 This is the fifth flowchart illustrating the polysilicon production control method of this application;

[0055] Figure 6 This is the sixth flowchart illustrating the polysilicon production control method of this application;

[0056] Figure 7 This is the seventh flowchart illustrating the polysilicon production control method of this application;

[0057] Figure 8 This is a schematic diagram of the polysilicon production control device of this application;

[0058] Figure 9 This is another schematic diagram of the polysilicon production control device of this application. Detailed Implementation

[0059] 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.

[0060] 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.

[0061] 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.

[0062] 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.

[0063] 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 that B is determined solely based on A; B can also be determined based on A and / or other information.

[0064] It should be noted that, for the convenience of those skilled in the art, the following description uses a cooling tower circulating pump and its upstream and downstream processes in the polysilicon production process as an example. However, this does not limit this application to cooling tower circulating pumps only in polysilicon production. Applying it to the production process of other products should also fall within the scope of protection of this application.

[0065] See Figure 1 One embodiment of this application provides a polysilicon production control method, comprising:

[0066] Step S101: Perform dynamic principal component analysis on the historical parameter samples to obtain a dynamic principal component analysis model. The dynamic principal component analysis model includes the dynamic principal component analysis equation and the Hotelling statistic test value (T0) of each principal component parameter in the dynamic principal component analysis equation. 2 The parameters in the historical parameter sample include: equipment parameters of the polysilicon equipment and process parameters of its upstream and downstream processes; and SPE test values.

[0067] Step S102: Based on the sample time series diagram of the principal component parameters, obtain the average value of each principal component parameter within the target time period, wherein the target time period is the common time period when the parameter fluctuations of each principal component parameter are the smallest.

[0068] Step S103: Based on the average value of each principal component parameter within the target time period and the dynamic principal component analysis equation, obtain the historical coupling model of the polysilicon equipment and its upstream and downstream processes.

[0069] Step S104: Based on the dynamic principal component analysis model, a first coupling model based on multiple linear regression and a second coupling model based on principal component analysis and multiple linear regression are obtained.

[0070] Step S105: Perform validity verification on the historical coupling model, the first coupling model, and the second coupling model, and obtain the target coupling model that passes the verification.

[0071] Step S106: Control the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes of the polysilicon equipment according to the target coupling model.

[0072] In this embodiment, the production control method using polysilicon as an example first acquires at least one historical parameter sample corresponding to the production process (preferably, all historical parameter samples from the start of the process to the current control point). This historical parameter sample is a collection of parameters obtained by detecting preset polysilicon equipment and various process measuring points during the process. The process measuring points are used to detect parameters such as flow rate, temperature, and liquid level collected by on-site chemical instruments. It should be noted that this detection is periodic; each historical parameter sample corresponds to a detection cycle or detection time point.

[0073] After obtaining the historical parameter samples, dynamic principal component analysis is performed on them. Dynamic principal component analysis is similar to static principal component analysis, but the difference lies in that it constructs an augmented matrix of the sample parameters and supplements the corresponding regions with parameter samples shifted forward by L steps (L being the lag factor). In a specific embodiment, the augmented matrix is ​​represented as follows:

[0074] X = [X] t X t-1 , ..., X t-L ]

[0075] Among them, X t For the parameter sample at time t or within the period;

[0076] X t-L This represents the parameter sample at time tL or within a period. Determining the value of L is complex; in one embodiment, it can be determined based on actual process experience, for example, L = 2, while X... t-1 X t-2 All are X t The parameter samples with forward time misalignment reflect the dynamic characteristics of the parameter samples.

[0077] See Figure 2 Specifically, the steps to obtain a dynamic principal component analysis model include:

[0078] Step S201: Perform dynamic principal component analysis on the historical parameter samples to determine each principal component parameter, the eigenvalues ​​corresponding to each principal component parameter, and the score matrix composed of the principal component parameters; wherein, the number of principal component parameters is k; wherein, when performing dynamic principal component analysis, the preferred cumulative contribution rate threshold is 85%;

[0079] Step S202: Obtain the loading matrix based on the eigenvalues ​​corresponding to each principal component parameter;

[0080] Step S203: Based on the score matrix, the loading matrix, and the historical parameter samples, obtain the dynamic principal component equation and the Hotelling statistic test value and the SPE test value of each principal component parameter, and use them as the dynamic principal component analysis model.

[0081] Specifically, the dynamic principal component analysis equation and the Hotelling statistic test value (T) of each principal component parameter in the dynamic principal component analysis equation are obtained. 2 The methods for obtaining the SPE test values ​​are the same as those for principal component analysis. Since principal component analysis is widely used, the specific details of each step will not be elaborated here. The resulting dynamic principal component equation can be expressed as:

[0082] Among them, a ij x is an element in the load matrix; j Let be the j-th parameter in the equipment parameters of the polysilicon equipment and the process parameters of its upstream and downstream processes; n is the number of all parameters involved in the calculation; PC i Let x be the parameter of the i-th principal component; k is the number of all principal component parameters; it should be noted that x j The equations contain parameters whose parameters are time-series-forwarded, hence they are called dynamic principal component equations.

[0083] It should be noted that the dynamic principal component analysis model obtained in this step includes: the dynamic principal component equation and the Hotelling statistic test value and SPE test value corresponding to each principal component parameter. The Hotelling statistic test value and SPE test value are used for further screening of principal component parameters, which helps to improve the accuracy of control results.

[0084] After obtaining the dynamic principal component analysis model described above, based on the sample time series diagrams of each principal component parameter, the time period with the smallest parameter fluctuation for each principal component parameter is determined. A target time period is selected, encompassing all principal component parameters, to represent the entire process flow in an "ideal state." By obtaining the average value of each principal component parameter within this target time period and coupling the parameters based on this average value and the dynamic principal component analysis equation, the "ideal state" coupling model can be obtained, which is the historical coupling model described in this embodiment.

[0085] Furthermore, a first coupling model is obtained by directly performing multiple linear regression based on the dynamic principal component analysis model, and a second coupling model is obtained by performing principal component analysis and multiple linear regression. It should be noted that both the first and second coupling models are coupling models obtained based on all historical parameter samples, and they are continuous with the current time in terms of time sequence, which is convenient for real-time detection and prediction.

[0086] Furthermore, to ensure the effectiveness of the historical coupling model, the first coupling model, and the second coupling model obtained through the above three acquisition methods, effective verification will be performed on these models. For example, significance and goodness-of-fit verification will be conducted. The significance test is performed using the t-distribution test value of coefficient analysis and the F-distribution test value of variance analysis to determine whether the corresponding P-value is greater than a preset P-value (e.g., 0.05). If it is greater than the preset P-value, it is considered insignificant, and the equation needs to be re-established. The goodness-of-fit is determined by the R-sq goodness-of-fit. If it is greater than the preset R-sq (e.g., 0.8), it is considered a good fit; otherwise, it is considered a poor fit. When any coupling model passes the effectiveness verification, it can be determined as a target coupling model. That is, the target coupling model can include at least one of the historical coupling model, the first coupling model, and the second coupling model. It should be noted that all historical coupling models are valid target coupling models.

[0087] This target coupling model enables the control of the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes by controlling at least one of the polysilicon equipment and its upstream and downstream process equipment, thereby ensuring the normal production of polysilicon.

[0088] In summary, this application establishes the relationships between various parameters in the process flow through dynamic principal component analysis, breaking down the information silos between parameters. It also considers the dynamic characteristics of the parameters, which is beneficial for monitoring and process analysis, and can be integrated with time series forecasting. Furthermore, in addition to constructing a first coupling model directly through multiple linear regression, this application also constructs a historical coupling model based on an "ideal state," and combines principal component analysis dimensionality reduction and multiple linear regression to construct a second coupling model, filling the current technological gap of lacking an ideal model. Moreover, this application's method can be used with different sample sizes and aggregation degrees, improving its applicability.

[0089] It should be noted that, when obtaining the historical parameter samples, the method further includes:

[0090] Based on the acquired historical data, determine the initial samples of historical parameters corresponding to each detection time point;

[0091] The initial samples of the historical parameters are preprocessed to obtain the historical parameter samples, wherein the preprocessing includes at least one of the following: sample sparsity processing, correlation analysis, and preliminary noise reduction processing.

[0092] Among these, sample sparsity processing is used when the sample size is large. Generally, the imported sample size is over 10,000, or even 100,000. Therefore, it is necessary to perform sampling and screening without affecting the overall trend and to form a new sample matrix. For example, when the number of parameter samples exceeds the maximum sample threshold, the parameter samples are sampled and screened according to a preset time interval.

[0093] Correlation analysis involves comparing all parameters pairwise and using the correlation coefficient R0. 2 Describe their correlation. If there is a strong correlation between the two parameters (e.g., R0), then... 2 If the value is greater than 0.9, then reduce one of the two parameters according to the actual situation.

[0094] The initial noise reduction process involves screening samples of each parameter based on process experience and warning and cascading values.

[0095] It should be noted that the purpose of this preprocessing is to obtain healthy parameter samples. Therefore, in addition to sample sparsity processing, correlation analysis and / or preliminary noise reduction, other preprocessing methods are also within the scope of protection of this application.

[0096] See Figure 3 Specifically, the method described above, based on the target coupling model, controls the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes of the polysilicon equipment, including:

[0097] Step S301: Obtain real-time parameter samples of the polysilicon equipment and its upstream and downstream processes;

[0098] Step S302: Perform error analysis on the real-time parameter samples based on the target coupling model to obtain the analysis results;

[0099] Step S303: Control the polysilicon equipment and / or the upstream and downstream process equipment corresponding to the polysilicon equipment according to the analysis results.

[0100] In this embodiment, when controlling the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes of the polysilicon equipment according to the target coupling model, the real-time parameter sample corresponding to the current detection time point is first obtained, and the error analysis of the real-time parameter sample is performed through the target coupling model obtained above to obtain the analysis result. In a specific embodiment, the relative error between the analytical value of the corresponding parameter obtained by the target coupling model based on the implementation parameter sample and its actual value is determined. If the relative error is greater than a preset error threshold (e.g., 5%), the parameter is determined to be abnormal, and the information (abnormal parameter is the corresponding relative error) is output. The process equipment of the relevant polysilicon equipment and / or the upstream and downstream processes corresponding to the polysilicon equipment is adjusted according to the abnormal parameter, thereby realizing the control or adjustment of the parameters.

[0101] It should be noted that, based on the different types (historical coupling model, first coupling model and second coupling model) and numbers of target coupling models, the analysis results include the relative errors corresponding to each target coupling model.

[0102] See Figure 4 Specifically, as described above, the step of obtaining the historical coupling model of the polysilicon equipment and its upstream and downstream processes based on the average value of each principal component parameter within the target time period and the dynamic principal component analysis equation includes:

[0103] Step S401: Substitute the average value corresponding to the principal component parameter into the dynamic principal component analysis equation to obtain the value corresponding to each principal component parameter;

[0104] Step S402: Determine the first important parameter among the principal component parameters based on the absolute value of the numerical value;

[0105] Step S403: Based on the first important parameter, the corresponding dynamic principal component equation is transformed to obtain the historical coupling model. In the historical coupling model, the first important parameter is the first target parameter, and the other parameters are the first response parameters.

[0106] In this embodiment, the process of obtaining the historical coupling model is briefly described. First, the average value of each principal component parameter within the target time period is substituted into the obtained dynamic principal component analysis equation to determine the numerical value of each principal component parameter in the dynamic principal component analysis equation. Based on the absolute value of each value, the parameter corresponding to the largest absolute value is determined as the first important parameter. It should be noted that there can be multiple first important parameters. Then, the corresponding dynamic principal component equation is transformed based on each first important parameter to obtain the target parameter expressed in terms of response parameters and principal component parameter values, i.e., the historical coupling model. For each first important parameter, in its corresponding historical coupling model, the first important parameter is the first target parameter, and all other principal component parameters are the corresponding first response parameters.

[0107] It should also be noted that, in one implementation, since the lag factor L of the augmented matrix is ​​set to 2, the historical parameter samples of a parameter will correspond to three parameters with time shifts in the augmented matrix. Therefore, determining the parameter with the largest contribution requires a comprehensive judgment based on the above three parameters, and generally, the corresponding three 'a' values ​​are selected. ij The average values ​​are compared, and a corresponding parameter is taken as the target parameter based on the reference time point. The corresponding principal component equation is transformed into the form in which the target parameter is expressed by the response parameter and the principal component value, and directly used as the historical coupling model based on the "ideal state".

[0108] It should be noted that since 82% of equipment failures originate from process conditions, the parameters that contribute the most are often not equipment parameters; and since process parameters have characteristics such as lag, the parameters that contribute the most are not the parameters with a lag factor of 0 (i.e., parameters with no time displacement based on the current time).

[0109] See Figure 5 Furthermore, according to the method described above, obtaining the first coupled model based on multiple linear regression and the second coupled model based on principal component analysis and multiple linear regression, respectively, based on the dynamic principal component analysis model, includes:

[0110] Step S501: Based on the Hotelling statistic test value and the SPE test value of each principal component parameter in the dynamic principal component analysis model and their corresponding F distribution, determine the Hotelling statistic test threshold and the SPE test threshold corresponding to each dynamic principal component analysis model.

[0111] Step S502: Based on the Hotling statistic test threshold and the SPE test threshold, the principal component parameters in the score matrix are screened out to obtain the target score matrix after screening.

[0112] Step S503: Based on the target score matrix, the load matrix, and the historical parameter samples, obtain the target dynamic principal component equation;

[0113] Step S504: Perform multiple linear regression based on the target dynamic principal component equation to obtain the first coupled model;

[0114] Step S505: Perform principal component analysis and multiple linear regression based on the target dynamic principal component equation to obtain the second coupled model.

[0115] In this embodiment, when obtaining the first coupling model and the second coupling model, the Hotling statistic test threshold and SPE test threshold corresponding to each principal component parameter are calculated in advance based on the Hotling statistic test value and SPE test value of each principal component parameter in the dynamic principal component analysis model and their corresponding F distribution. The Hotling statistic test threshold and SPE test threshold can be modified by combining them with the interlocking ledgers of various measurement points in the polysilicon equipment and its upstream and downstream processes, which are set as warning values ​​and cascading values. Furthermore, since the thresholds determined above are based on the dynamic principal component analysis model, they are also dynamic thresholds (i.e., not thresholds for single parameters).

[0116] After obtaining the thresholds for the Hotelling statistic test and the SPE test, the principal component parameters in the score matrix are screened out. Principal component parameters that do not meet the thresholds are treated as noise or outliers and removed to obtain the target score matrix. It should be noted that during the screening process, the parameter samples corresponding to the principal component parameters that are considered noise or outliers are simultaneously removed.

[0117] Furthermore, based on the target score matrix, the load matrix, and the historical parameter samples, the target dynamic principal component equation after two denoising steps can be obtained. Based on this target dynamic principal component equation, a first coupled model can be obtained through multiple linear regression, and a second coupled model can be obtained through principal component analysis and multiple linear regression. This helps improve the accuracy of the calculations for the first and second coupled models.

[0118] In another embodiment of this application, the operation of determining the threshold of the Hotelling statistic test and the SPE test, and then performing the screening, can be based on the historical coupling model obtained above. That is, the determined threshold is obtained by combining the historical coupling model, that is, with the "ideal state" as a reference, a dynamic threshold is obtained in order to provide better guidance.

[0119] See Figure 6 Specifically, in the method described above, the step of performing multiple linear regression based on the target dynamic principal component equation to obtain the first coupled model includes:

[0120] Step S601: Based on the absolute values ​​of the principal component parameters in the target dynamic principal component equation, determine the second important parameter among the principal component parameters and the second response parameter corresponding to the second important parameter;

[0121] Step S602: The second important parameter is used as the second target parameter, and the multiple linear regression is performed with the corresponding second response parameter to obtain the first coupled model.

[0122] In this embodiment, when obtaining the first coupled model, firstly, based on the absolute values ​​of each principal component parameter in the target dynamic principal component equation, the principal component parameter corresponding to the largest absolute value is determined as the second important parameter (including at least one). Then, the other principal component parameters besides the target second important parameter are determined as the second response parameters corresponding to the target second important parameter. Further, each second important parameter is used as a second target parameter, and multiple linear regression is performed to obtain the first coupled model. The method of fitting using multiple linear regression is widely used, so the specific multiple regression process will not be elaborated here.

[0123] See Figure 7 Specifically, as described above, the step of performing principal component analysis and multiple linear regression based on the target dynamic principal component equation to obtain the second coupled model includes:

[0124] Step S701: Based on the absolute values ​​of the principal component parameters in the target dynamic principal component equation, determine the third important parameter among the principal component parameters and the third response parameter corresponding to the third important parameter.

[0125] Step S702: Perform principal component analysis on each of the third response parameters to obtain the reference principal component parameters corresponding to each of the third response parameters;

[0126] Step S703: Take the third important parameter as the target parameter and perform the multiple linear regression with the corresponding reference principal component parameter to obtain the second coupled model.

[0127] In this embodiment, when obtaining the second coupled model, firstly, based on the absolute values ​​of each principal component parameter in the target dynamic principal component equation, the principal component parameter corresponding to the largest absolute value is determined as the third important parameter (including at least one). Then, the other principal component parameters besides the target third important parameter are determined as the third response parameters corresponding to the target third important parameter. Further, principal component analysis is performed on each of the third response parameters to obtain the reference principal component parameters corresponding to each third response parameter. Then, each third important parameter is used as the third target parameter, and multiple linear regression is performed to obtain the second coupled model. The method of fitting using multiple linear regression is widely used, so the specific multiple regression process will not be elaborated here.

[0128] It should be noted that the reference principal component parameters mentioned in this embodiment are the principal component parameters in the principal component equation after dimensionality reduction by principal component analysis. These are referred to as reference principal component parameters here to distinguish them from the principal component parameters in the dynamic principal component equation. It should also be noted that, since the same parameter in the dynamic principal component equation corresponds to parameters in three different time periods, this application does not merge parameters from different time periods when performing multiple linear regression. This allows for the simultaneous existence of the same parameter as the target parameter in different time periods, which is beneficial for reflecting the dynamic characteristics of the parameters.

[0129] For ease of understanding, the following example uses a parameter sample from a specific time period: April 20, 2022 to September 14, 2022. This involves 15 parameters, totaling 215,402 parameter samples. The 15 parameters are represented by Xa1, Xa2, Xa3, Xa4, Xa5, Xa6, Xa7, Xa8, Xa9, Xa10, Xa11, Xa12, Xa13, Xa14, and Xa15. The equipment parameters for the cold tower circulating pump are: current (Xa6), and the remaining parameters are: Xa1 (pressure 1), Xa2 (pressure 2), Xa3 (pressure 3), Xa4 (pressure 4), Xa5 (pressure 5), Xa7 (liquid level 4), Xa8 (liquid level 5), Xa9 (reflux temperature 1), Xa10 (reflux temperature 2), Xa11 (reflux temperature 3), Xa12 (reflux temperature 4), Xa13 (reflux temperature 5), Xa14 (flow control 1), and Xa15 (flow control 3). The subscript 'a' represents the parameter corresponding to the 'a'-th lag factor, and the total number of lag factors L is 2. After the above preprocessing and screening operations, the remaining sample size is 116179. Since time series prediction is not considered, no interpolation is performed. The augmented matrix has an L value of 2, and the augmented matrix has a total of 15*3=45 parameters. Under the cumulative contribution rate threshold of 85%, the number of principal component parameters is 5. The threshold of the Hotling statistic test value (T2) is 11.0709, and the threshold of the SPE test value is 20.3550.

[0130] Based on the load matrix obtained from principal component analysis, the parameters contributing the most to PC1 (first principal component), PC2 (second principal component), PC3 (third principal component), PC4 (fourth principal component), and PC5 (fifth principal component) are identified as Xa11, Xa1, Xa7, Xa13, and Xa8, respectively. (The suffix 'a' indicates the parameter's position across all time shifts in the augmented matrix, with the largest values ​​being X211, X11, X17, X113, and X28). It can be observed that among the five principal components, no parameter with the largest contribution is a unique device parameter (current). Meanwhile, five principal component equations were obtained (due to the length of the principal component equations, the parameters of the first principal component are considered the main ones): PC1 = 0.1579X11 - 0.09589X12 + 0.1508X13 + 0.1533X14 + 0.1499X15 - 0.05113X16 - 0.0146X17 - 0.0175X18 + 0.2035X19 + 0.2025X110 + 0.2038X111 + 0.2028X112 + 0.1734X113 + 0.10 72X114+0.1439X115+0.1580X21-0.0958X22+0.1509X23+0.1533X24+0.1500X25-0.0511X26-0.0146X27-0.0 176X28+0.2034X29+0.2025X210+0.2038X211+0.2028X212+0.1734X213+0.1071X214+0.1439X215+0.1580X31 -0.0957X32+0.1510X33+0.1533X34+0.1501X35 -0.0512X36-0.0147X37 -0.0176X38+0.2034X39+0.2024X310+0.2038X311+0.2027X312+0.1734X313+0.1070X314+0.1438X315.

[0131] Using the first principal component equation PC1, we can find that the absolute value of the coefficients before each parameter is the largest, which is 0.2038, corresponding to the parameter X211. The other principal parameters are found in the same way.

[0132] Knowing the principal parameters X211, X11, X17, X113, and X28 as the target parameters and the corresponding dynamic principal component equations, the "historical best or ideal state" values ​​are directly extracted from the samples of the five principal components. Here, this is achieved by selecting the most stable sample interval shared by the five principal components and determining the values ​​of each principal component: PC1 = 2.1301, PC2 = 1.6731, PC3 = -0.1771, PC4 = -1.0310, and PC5 = 1.4254. Substituting these values ​​into the principal component equations and transforming them into a form representing the target parameters, the historical coupling model described above can be obtained.

[0133] For this model, the corresponding Hotling statistic test value (T2) threshold is dynamic, specifically:

[0134]

[0135] The information of each of the main parameters is contained in PCi (according to the principal component equation), the value of each denominator is the square of the contribution rate corresponding to each principal component, and 11.0709 is the Hotling T2 threshold.

[0136] The above model is generally used as a reference for ideal conditions. It can show the difference from the actual conditions. In this embodiment, suggestions are given for adjusting the parameters corresponding to the quench tower circulating pump (direct adjustment of reflux temperature 3, pressure 1, liquid level 4, reflux temperature 5, and liquid level 5, and indirect adjustment of other parameters).

[0137] Using a coupled modeling method based on multiple linear regression, the equations for the five response values, which include parameters at two additional time shifts (e.g., X211, X111, and X311, representing parameters of the same physical quantity but at different time ranges), are directly obtained through the multiple linear regression model. Taking the first parameter as an example, its equation has an R-squared value of 99.98%, and the p-values ​​corresponding to the T-test and F-test values ​​for each parameter are all less than 0.05. Based on actual modeling, all equation test values ​​are less than 0.05, and R-squared values ​​are all greater than 0.8. Therefore, the output of this model not only reflects the relationship between the target parameter and other parameters but also the linear dynamic relationship of the target parameter. This model reflects the actual operation of the quench tower circulating pump under healthy conditions and does not coincide with the model under ideal conditions.

[0138] Using a coupled model based on multiple linear regression, for a single target value, all other response values ​​are reduced in dimensionality through principal component analysis (based on a cumulative contribution rate threshold of 90%, which has been experimentally verified and is the threshold for cumulative contribution rate that ensures an R-squared value above 0.8). This process is performed on each of the five target values, yielding six principal component values. These reduced principal component values ​​are then used as response parameters for multiple linear regression, resulting in five equations. The equation corresponding to the first important parameter is...

[0139] X211=150.000+0.223219tPS[1]+0.343820tPS[2]-0.180084tPS[3[

[0140] +0.125817tPS[4]±0.113285tPS[5]+0.13480tPS[6]

[0141] The R-squared value of the equation is 91.78%, and the p-values ​​corresponding to the T-test and F-test values ​​of each parameter are all less than 0.05; tPS[i] is the i-th principal component (i = 1, 2, 3, ...) obtained by principal component analysis dimensionality reduction of each response parameter.

[0142] It should be noted that the principal component values ​​in the respective equations are not the same type of principal component values, but rather principal component values ​​obtained through their respective dimensionality reduction (for example, the first principal component tPS1 obtained by dimensionality reduction of all other values ​​(response values) in the case of objective value X113 is not the same parameter as the first principal component tPS1 obtained in the case of objective value X28).

[0143] It can be observed that after dimensionality reduction of the response value at a cumulative contribution rate threshold of 90%, linear regression fitting can be performed with a smaller number of parameters, and the requirements of each threshold can be met. Furthermore, the model established by this method can also reflect the dynamic characteristics of the target parameters themselves, because the dimensionality reduction process includes information from all response parameters.

[0144] See Figure 8 Another embodiment of this application also provides a polysilicon production control device, comprising:

[0145] The first processing module 801 is used to perform dynamic principal component analysis on historical parameter samples to obtain a dynamic principal component analysis model. The dynamic principal component analysis model includes a dynamic principal component analysis equation and the Hotelling statistic test value and SPE test value of each principal component parameter in the dynamic principal component analysis equation. The parameters in the historical parameter sample include: equipment parameters of polysilicon equipment and process parameters of its upstream and downstream processes.

[0146] The second processing module 802 is used to obtain the average value of each principal component parameter within a target time period based on the sample time series diagram of the principal component parameters, wherein the target time period is the common time period when the parameter fluctuations of each principal component parameter are the smallest.

[0147] The third processing module 803 is used to obtain a historical coupling model of the polysilicon equipment and its upstream and downstream processes based on the average value of each principal component parameter within the target time period and the dynamic principal component analysis equation.

[0148] The fourth processing module 804 is used to obtain a first coupled model based on multiple linear regression and a second coupled model based on principal component analysis and multiple linear regression according to the dynamic principal component analysis model.

[0149] The fifth processing module 805 is used to perform validity verification on the historical coupling model, the first coupling model and the second coupling model, and obtain the target coupling model that has passed the verification.

[0150] The sixth processing module 806 is used to control the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes of the polysilicon equipment according to the target coupling model.

[0151] Specifically, in the apparatus described above, the sixth processing module includes:

[0152] The first processing unit is used to acquire real-time parameter samples of the polysilicon equipment and its upstream and downstream processes;

[0153] The second processing unit is used to perform error analysis on the real-time parameter samples according to the target coupling model and obtain the analysis results.

[0154] The third processing unit is used to control the polysilicon equipment and / or the upstream and downstream process equipment corresponding to the polysilicon equipment based on the analysis results.

[0155] Specifically, in the apparatus described above, the first processing module includes:

[0156] The fourth processing unit is used to perform dynamic principal component analysis on the historical parameter samples to determine each principal component parameter, the eigenvalues ​​corresponding to each principal component parameter, and the score matrix composed of the principal component parameters.

[0157] The fifth processing unit is used to obtain the loading matrix based on the eigenvalues ​​corresponding to each principal component parameter;

[0158] The sixth processing unit is used to obtain the dynamic principal component equation and the Hotelling statistic test value and the SPE test value of each principal component parameter based on the score matrix, the loading matrix and the historical parameter sample, and use them as the dynamic principal component analysis model.

[0159] Specifically, in the apparatus described above, the third processing module includes:

[0160] The seventh processing unit is used to substitute the average value corresponding to the principal component parameter into the dynamic principal component analysis equation to obtain the value corresponding to each principal component parameter;

[0161] The eighth processing unit is used to determine the first important parameter among the principal component parameters based on the absolute value of the numerical value.

[0162] The ninth processing unit is used to transform the corresponding dynamic principal component equation based on the first important parameter to obtain the historical coupling model, in which the first important parameter is the first target parameter and the other parameters are the first response parameters.

[0163] Furthermore, in the apparatus described above, the fourth processing module includes:

[0164] The tenth processing unit is used to determine the Hotling statistic test threshold and the SPE test threshold corresponding to each of the dynamic principal component analysis models based on the Hotling statistic test value and the SPE test value of each of the principal component parameters in the dynamic principal component analysis model and their corresponding F distributions.

[0165] The eleventh processing unit is used to screen out the principal component parameters in the score matrix according to the Hotling statistic test threshold and the SPE test threshold, so as to obtain the target score matrix after screening.

[0166] The twelfth processing unit is used to obtain the target dynamic principal component equation based on the target score matrix, the load matrix, and the historical parameter samples.

[0167] The thirteenth processing unit is used to perform multiple linear regression based on the target dynamic principal component equation to obtain the first coupled model.

[0168] The fourteenth processing unit is used to perform principal component analysis and multiple linear regression based on the target dynamic principal component equation to obtain the second coupled model.

[0169] Specifically, in the apparatus described above, the thirteenth processing unit includes:

[0170] The first sub-processing unit is used to determine the second important parameter and the second response parameter corresponding to the second important parameter in each of the principal component parameters according to the absolute value of each principal component parameter in the target dynamic principal component equation.

[0171] The second sub-processing unit is used to take the second important parameter as the second target parameter and perform the multiple linear regression with the corresponding second response parameter to obtain the first coupled model.

[0172] Specifically, in the apparatus described above, the fourteenth processing unit includes:

[0173] The third sub-processing unit is used to determine the third important parameter and the third response parameter corresponding to the third important parameter in each principal component parameter according to the absolute value of each principal component parameter in the target dynamic principal component equation.

[0174] The fourth sub-processing unit is used to perform principal component analysis on each of the third response parameters to obtain the reference principal component parameters corresponding to each of the third response parameters;

[0175] The fifth sub-processing unit is used to take the third important parameter as the target parameter and perform the multiple linear regression with the corresponding reference principal component parameter to obtain the second coupled model.

[0176] The apparatus embodiment of this application is an apparatus corresponding to the embodiment of the above-described polysilicon production control method. All implementation means in the above-described method embodiment are applicable to the apparatus embodiment and can achieve the same technical effect.

[0177] See Figure 9 In another embodiment of this application, a polysilicon production control device is also disclosed, comprising:

[0178] The storage module 901 includes: a detection point storage module 9011 for storing detection point information of various parameters on site; a parameter sample storage module 9012 for storing samples of various parameters on site; and a sample storage module 9013 for storing full-process information such as models and monitoring results; wherein, the sample storage module 9013 acts as a "black box".

[0179] The preprocessing module 902 includes an initial noise reduction module and a sampling module. The initial noise reduction module is used to collect threshold information such as warning values, chain values, or custom thresholds, and is used to filter the input historical data. The sampling module is used for equidistant sampling of large sample sizes to reduce the sample size.

[0180] The dynamic principal component analysis module 903 includes: an augmented matrix module 9031 and a principal component analysis module 9032. The augmented matrix module 9031 is used to construct a dynamic data matrix; the principal component analysis module 9032 is used to construct dynamic principal component equations and also has a further noise reduction function for the first coupling module and the second coupling module.

[0181] The modeling module 904 includes: a historical coupling module 9041, a first coupling module 9042, a second coupling module 9043, and a verification module 9044. The historical coupling module 9041 constructs the model using the historical coupling model based on "ideal state" as described in this application. The first coupling module 9042 constructs the model using the first coupling model based on multiple linear regression as described in this application. The second coupling module 9043 constructs the model using the second coupling model based on principal component analysis and multiple linear regression as described in this application. The verification module 9044 uses significance testing and goodness of fit to determine the effectiveness of the obtained model.

[0182] The output monitoring module 905 includes a model output module 9051 and a real-time monitoring module 9052. The model output module 9051 receives valid models as determined by the verification module 9044. The real-time monitoring module 9052 combines the valid models with the corresponding input data for each parameter for monitoring. This module can run multiple types of models simultaneously. The output models can be used by the user to guide the adjustment of process operating parameters.

[0183] Specifically, for models built using the first coupling model establishment method and the second coupling model establishment method, a threshold is set in real-time monitoring, which is usually set as the relative error between the real-time parsed value and the real-time true value.

[0184] Further explanation is needed. Figure 9 Solid arrows indicate the process of modeling and outputting a model using historical data; while dashed arrows indicate the flow of data input in real time for monitoring. The general process is as follows:

[0185] The input samples are added to the augmented matrix of the original historical data;

[0186] If the real-time monitoring module 9052 is running a historical coupling model based on "ideal state" or a first coupling model based on multiple linear regression, then the new samples in the augmented matrix are substituted into it, the corresponding analytical values ​​are obtained, and compared with the set error threshold.

[0187] If the real-time monitoring module 9052 is running a second coupled model based on principal component analysis-multiple linear regression, then in the second coupled module 9043, it is necessary to combine historical data to perform principal component dimensionality reduction, obtain new principal component samples, and input them to calculate the corresponding analytical values, and compare them with the set error threshold.

[0188] 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 control method as described above.

[0189] 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 control method described above.

[0190] 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.

[0191] 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.

[0192] 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 method for controlling polycrystalline silicon production, characterized in that, include: Dynamic principal component analysis is performed on historical parameter samples to obtain a dynamic principal component analysis model. The dynamic principal component analysis model includes a dynamic principal component equation and the Hotelling statistic test value and the squared prediction error (SPE) test value of each principal component parameter in the dynamic principal component equation. The parameters in the historical parameter sample include: equipment parameters of polysilicon equipment and process parameters of its upstream and downstream processes. Based on the sample time series diagram of the principal component parameters, the average value corresponding to each principal component parameter within the target time period is obtained, wherein the target time period is the common time period when the parameter fluctuations corresponding to each principal component parameter are the smallest; Based on the average value of each principal component parameter within the target time period and the dynamic principal component equation, a historical coupling model of the polysilicon equipment and its upstream and downstream processes is obtained. Based on the dynamic principal component analysis model, a first coupling model based on multiple linear regression and a second coupling model based on principal component analysis and multiple linear regression are obtained. The validity of the historical coupling model, the first coupling model, and the second coupling model is verified to obtain the target coupling model that passes the verification. Based on the target coupling model, the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes of the polysilicon equipment are controlled.

2. The method according to claim 1, characterized in that, Based on the target coupling model, the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes of the polysilicon equipment are controlled, including: Obtain real-time parameter samples of the polysilicon equipment and its upstream and downstream processes; Error analysis is performed on the real-time parameter samples based on the target coupling model to obtain the analysis results; Based on the analysis results, control the polysilicon equipment and / or the upstream and downstream process equipment corresponding to the polysilicon equipment.

3. The method according to claim 1, characterized in that, The process of performing dynamic principal component analysis on the acquired historical parameter samples to obtain a dynamic principal component analysis model includes: Dynamic principal component analysis is performed on the historical parameter samples to determine each principal component parameter, the eigenvalues ​​corresponding to each principal component parameter, and the score matrix composed of the principal component parameters. The loading matrix is ​​obtained based on the eigenvalues ​​corresponding to each principal component parameter; Based on the score matrix, the loading matrix, and the historical parameter samples, the dynamic principal component equation, the Hotelling statistic test value, and the SPE test value of each principal component parameter are obtained and used as the dynamic principal component analysis model.

4. The method according to claim 1, characterized in that, The step of obtaining a historical coupling model of the polysilicon equipment and its upstream and downstream processes based on the average value of each principal component parameter within the target time period and the dynamic principal component equation includes: Substitute the average value corresponding to the principal component parameter into the dynamic principal component equation to obtain the value corresponding to each principal component parameter; Based on the absolute value of the stated values, determine the first important parameter among the principal component parameters; Based on the first important parameter, the corresponding dynamic principal component equation is transformed to obtain the historical coupling model. In the historical coupling model, the first important parameter is the first objective parameter, and the other parameters are the first response parameters.

5. The method according to claim 3, characterized in that, The process of obtaining a first coupled model based on multiple linear regression and a second coupled model based on principal component analysis and multiple linear regression, according to the dynamic principal component analysis model, includes: Based on the Hotelling statistic test value and the SPE test value of each principal component parameter in the dynamic principal component analysis model and their corresponding F distribution, determine the Hotelling statistic test threshold and the SPE test threshold corresponding to each dynamic principal component analysis model. Based on the Hotling statistic test threshold and the SPE test threshold, the principal component parameters in the score matrix are screened out to obtain the target score matrix after screening. Based on the target score matrix, the load matrix, and the historical parameter samples, the target dynamic principal component equation is obtained; The first coupled model is obtained by performing multiple linear regression based on the target dynamic principal component equation. Based on the target dynamic principal component equation, principal component analysis and multiple linear regression are performed to obtain the second coupled model.

6. The method according to claim 5, characterized in that, The step of performing multiple linear regression based on the target dynamic principal component equation to obtain the first coupled model includes: Based on the absolute value of each principal component parameter in the target dynamic principal component equation, determine the second important parameter among each principal component parameter and the second response parameter corresponding to the second important parameter; The second important parameter is used as the second objective parameter, and the first coupling model is obtained by performing the multiple linear regression with the corresponding second response parameter.

7. The method according to claim 5, characterized in that, The step of performing principal component analysis and multiple linear regression based on the target dynamic principal component equation to obtain the second coupled model includes: Based on the absolute value of each principal component parameter in the target dynamic principal component equation, determine the third important parameter among each principal component parameter and the third response parameter corresponding to the third important parameter; Principal component analysis is performed on each of the third response parameters to obtain the reference principal component parameters corresponding to each of the third response parameters. The third important parameter is used as the target parameter, and a multiple linear regression is performed with the corresponding reference principal component parameter to obtain the second coupled model.

8. A polysilicon production control device, characterized in that, include: The first processing module is used to perform dynamic principal component analysis on historical parameter samples to obtain a dynamic principal component analysis model. The dynamic principal component analysis model includes a dynamic principal component equation and the Hotelling statistic test value and the squared prediction error (SPE) test value of each principal component parameter in the dynamic principal component equation. The parameters in the historical parameter sample include: equipment parameters of polysilicon equipment and process parameters of its upstream and downstream processes. The second processing module is used to obtain the average value of each principal component parameter within a target time period based on the sample time series diagram of the principal component parameters, wherein the target time period is the common time period when the parameter fluctuations of each principal component parameter are the smallest. The third processing module is used to obtain a historical coupling model of the polysilicon equipment and its upstream and downstream processes based on the average value of each principal component parameter within the target time period and the dynamic principal component equation. The fourth processing module is used to obtain a first coupled model based on multiple linear regression and a second coupled model based on principal component analysis and multiple linear regression according to the dynamic principal component analysis model. The fifth processing module is used to perform validity verification on the historical coupling model, the first coupling model, and the second coupling model, and obtain the target coupling model that has passed the verification. The sixth processing module is used to control the equipment parameters of the polysilicon equipment and / or the process parameters of the upstream and downstream processes of the polysilicon equipment according to the target coupling model.

9. A server, characterized in that, It 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 control 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 control method as described in any one of claims 1 to 7.

Images