Method and apparatus for controlling a polycrystalline silicon reduction furnace

Abstract

The application relates to the technical field of polycrystalline silicon generation, and provides a control method and device for a polycrystalline silicon reduction furnace. The method comprises the following steps: inputting each control parameter of the polycrystalline silicon reduction furnace into a trained first neural network model, and obtaining each observation data corresponding to each control parameter; obtaining target observation data matched with at least one preset observation data from the observation data, so as to control the polycrystalline silicon reduction furnace to operate according to the control parameter corresponding to the target observation data. The control method for the polycrystalline silicon reduction furnace provided by the embodiment of the application can reduce the energy consumption of unit output and improve the production efficiency of polycrystalline silicon without reducing the quality of polycrystalline silicon.

Description

Technical Field

[0001] This application relates to the field of polysilicon generation technology, specifically to a control method and apparatus for a polysilicon reduction furnace. Background Technology

[0002] Polycrystalline silicon is produced by the chemical reduction reaction of TCS and hydrogen in the temperature range of 1000-1100°C. In this process, factors affecting the yield of polycrystalline silicon include the heating current of the reduction furnace, the raw material flow rate, and the feed temperature—all control parameters of the polycrystalline silicon reduction furnace. Inaccurate control parameters, such as excessive raw material flow rate, may result in uneven silicon rod surfaces, affecting cost, while insufficient raw material flow rate may lead to low yield. Therefore, accurately adjusting the control parameters of the polycrystalline silicon reduction furnace to reduce energy consumption per unit output without compromising polycrystalline silicon quality is a pressing issue that needs to be addressed. Summary of the Invention

[0003] This application aims to address at least one of the technical problems existing in the related art. To this end, this application proposes a control method for a polycrystalline silicon reduction furnace, which can reduce energy consumption per unit output and improve polycrystalline silicon production efficiency without reducing polycrystalline silicon quality.

[0004] This application also proposes a control device for a polysilicon reduction furnace.

[0005] This application also proposes an electronic device.

[0006] This application also proposes a computer-readable storage medium.

[0007] The control method for a polysilicon reduction furnace according to the first aspect of this application includes:

[0008] The control parameters of the polysilicon reduction furnace are input into the trained first neural network model to obtain the observation data corresponding to each control parameter.

[0009] From each of the observation data, target observation data that matches at least one preset observation data is obtained, so as to control the operation of the polysilicon reduction furnace according to the control parameters corresponding to the target observation data;

[0010] The first neural network model is trained from historical operating data, which includes historical control parameters and historical observation data corresponding to the historical control parameters.

[0011] The preset observation data refers to the historical observation data in which the corresponding polysilicon production status reaches the preset requirements. The polysilicon production status includes production energy consumption and yield.

[0012] According to one embodiment of this application, it also includes:

[0013] The historical control parameters are input into the first neural network model to obtain the predicted observation data;

[0014] The predicted observation data is matched with the historical observation data corresponding to the historical control parameters to obtain the matching result;

[0015] The network parameters of the first neural network model are adjusted according to the matching result until the predicted observation data obtained after any of the historical control parameters are input into the first neural network model matches the historical observation data corresponding to the historical control parameters.

[0016] According to one embodiment of this application, it also includes:

[0017] The historical observation data are input into the trained second neural network model for clustering to obtain the clustering results of the historical observation data.

[0018] The historical observation data whose clustering results are positive samples are the preset observation data;

[0019] The second neural network model is trained from a model training set composed of a portion of the historical observation data from each of the historical observation data.

[0020] The positive sample corresponds to polysilicon production energy consumption and yield that meet the preset requirements.

[0021] According to one embodiment of this application, each of the historical observation data in the model training set is a positive sample.

[0022] According to one embodiment of this application, obtaining target observation data that matches at least one preset observation data from the various observation data includes:

[0023] From each of the observation data, obtain each alternative observation data that matches at least one preset observation data;

[0024] From the candidate observation data, obtain the target observation data in which the polysilicon production state is greater than the average polysilicon production state of the candidate observation data.

[0025] According to one embodiment of this application, the control parameters include at least one of raw material flow rate, heating current, and feed temperature.

[0026] According to one embodiment of this application, the observation data includes at least one of the following: furnace temperature data, exhaust gas data, and cooling water data of the polysilicon reduction furnace.

[0027] A control device for a polysilicon reduction furnace according to a second aspect embodiment of this application includes:

[0028] The observation data acquisition module is used to input the control parameters of the polysilicon reduction furnace into the trained first neural network model and acquire the observation data corresponding to each control parameter.

[0029] The control parameter determination module is used to obtain target observation data that matches at least one preset observation data from each of the observation data, so as to control the operation of the polysilicon reduction furnace according to the control parameters corresponding to the target observation data;

[0030] The first neural network model is trained from historical operating data, which includes historical control parameters and historical observation data corresponding to the historical control parameters.

[0031] The preset observation data refers to the historical observation data in which the corresponding polysilicon production status reaches the preset requirements. The polysilicon production status includes production energy consumption and yield.

[0032] An electronic device according to a third aspect of this application includes a processor and a memory storing a computer program, wherein the processor executes the computer program to implement the control method for a polysilicon reduction furnace described in any of the above embodiments.

[0033] A computer-readable storage medium according to a fourth aspect of this application stores a computer program thereon, which, when executed by a processor, implements the control method for a polysilicon reduction furnace described in any of the above embodiments.

[0034] A computer program product according to a fifth aspect of this application includes: when the computer program is executed by a processor, it implements a control method for a polysilicon reduction furnace as described in any of the above embodiments.

[0035] The above-described one or more technical solutions in the embodiments of this application have at least one of the following technical effects:

[0036] By inputting the control parameters of the polysilicon reduction furnace into a trained first neural network model, and obtaining the observation data corresponding to each control parameter, the target observation data that matches at least one preset observation data that indicates the polysilicon production state has reached the preset requirements is then obtained from the observation data. Based on the control parameters corresponding to the target observation data, the operation of the polysilicon reduction furnace is controlled, thereby optimizing the control parameters of the polysilicon reduction furnace. This reduces energy consumption per unit output and improves polysilicon production efficiency without compromising polysilicon quality. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a schematic diagram of the polysilicon reduction furnace process for producing polysilicon according to an embodiment of this application;

[0039] Figure 2 This is a schematic flowchart of the control method for the polysilicon reduction furnace provided in the embodiments of this application;

[0040] Figure 3 This is a schematic diagram of the training process of the first neural network model provided in the embodiments of this application;

[0041] Figure 4 This is a schematic diagram of the process for obtaining preset observation data provided in an embodiment of this application;

[0042] Figure 5 This is a schematic diagram of the control device for the polysilicon reduction furnace provided in the embodiments of this application;

[0043] Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0045] Polycrystalline silicon is produced by the chemical reduction reaction of TCS and hydrogen in the temperature range of 1000-1100°C. During this process, factors affecting the yield of polycrystalline silicon include the heating current of the reduction furnace, the raw material flow rate, and the feed temperature—all control parameters of the polycrystalline silicon reduction furnace. Inaccurate control parameters may affect the quality of the polycrystalline silicon or production energy consumption.

[0046] The inventors noted that, as Figure 1As shown, during the polysilicon preparation process, adjusting the control parameters input to the reduction furnace, such as TCS flow rate, H flow rate, heating current, and feed temperature, affects the reaction conditions inside the furnace. While the internal reaction of the reduction furnace cannot be observed directly, data such as furnace temperature, exhaust gas volume, and cooling water volume can be monitored during polysilicon preparation. This data can be used to assess the reaction conditions within the furnace. Based on this finding, the applicant, through in-depth research, believes that the accuracy of control parameters can be determined using the observation data obtained during polysilicon preparation, thereby optimizing polysilicon quality or production energy consumption.

[0047] After in-depth research, the inventors designed a control method for a polysilicon reduction furnace. This method obtains target observation data that matches at least one preset observation data that corresponds to the polysilicon production state reaching the preset requirements from each observation data that corresponds one-to-one with each control parameter. Based on the control parameters corresponding to the target observation data, the operation of the polysilicon reduction furnace is controlled, thereby optimizing the control parameters of the polysilicon reduction furnace. This reduces energy consumption per unit output and improves polysilicon production efficiency without reducing polysilicon quality.

[0048] The control method and apparatus for the polysilicon reduction furnace provided in this application will be described in detail below through several specific embodiments.

[0049] In some embodiments, a control method for a polysilicon reduction furnace is provided. This method is applied to a server for controlling the polysilicon reduction furnace. The server can be a standalone server or a server cluster consisting of multiple servers. It can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence sampling point devices.

[0050] like Figure 2 As shown, the control method for a polycrystalline silicon reduction furnace provided in this embodiment includes:

[0051] Step 101: Input the control parameters of the polysilicon reduction furnace into the trained first neural network model to obtain the observation data corresponding to each control parameter.

[0052] Step 102: Obtain target observation data that matches at least one preset observation data from each of the observation data, so as to control the operation of the polysilicon reduction furnace according to the control parameters corresponding to the target observation data;

[0053] The first neural network model is trained from historical operating data, which includes historical control parameters and historical observation data corresponding to the historical control parameters.

[0054] The preset observation data refers to the historical observation data in which the corresponding polysilicon production status reaches the preset requirements. The polysilicon production status includes production energy consumption and yield.

[0055] By inputting the control parameters of the polysilicon reduction furnace into a trained first neural network model, and obtaining the observation data corresponding to each control parameter, the target observation data that matches at least one preset observation data that indicates the polysilicon production state has reached the preset requirements is then obtained from the observation data. Based on the control parameters corresponding to the target observation data, the operation of the polysilicon reduction furnace is controlled, thereby optimizing the control parameters of the polysilicon reduction furnace. This reduces energy consumption per unit output and improves polysilicon production efficiency without compromising polysilicon quality.

[0056] In some embodiments, the control parameters include at least one of the parameters that affect the chemical reaction within the reduction furnace, such as raw material flow rate, heating current, and feed temperature. Since polycrystalline silicon is produced by the chemical reduction reaction of TCS and hydrogen in the temperature range of 1000–1100°C, the raw material flow rate may include both the TCS flow rate and the hydrogen flow rate. The heating current is the current used to control the heating of the reduction furnace. The feed temperature is the ambient temperature at which the raw material is fed into the reduction furnace. The raw material flow rate, heating current, and feed temperature can constitute a control parameter, such as control parameter A = {raw material flow rate, heating current, feed temperature}.

[0057] In some embodiments, multiple different control parameters can be preset, such as control parameter A1 = {raw material flow rate a1, heating current a2, feed temperature a3}, control parameter A2 = {raw material flow rate a1+1, heating current a2, feed temperature a3}, control parameter A3 = {raw material flow rate a1, heating current a2+1, feed temperature a3}, control parameter A4 = {raw material flow rate a1, heating current a2, feed temperature a3+1}, etc. For each control parameter, it is input into a trained first neural network model, so that the first neural network model outputs observation data corresponding to the control parameter to predict how the observation data of the reduction furnace will change when the control parameter is adjusted. The observation data includes at least one of the following: furnace temperature data, exhaust gas data, and cooling water data of the polycrystalline silicon reduction furnace. The furnace temperature data may include the temperature of the reduction furnace during polycrystalline silicon preparation; the exhaust gas data may include the amount of exhaust gas discharged from the reduction furnace during polycrystalline silicon preparation and / or the content of each element in the exhaust gas; the cooling water data may include the consumption of cooling water in the reduction furnace during polycrystalline silicon preparation.

[0058] To improve the prediction accuracy of the first neural network model, in some embodiments, such as Figure 3 As shown, training the first neural network model includes:

[0059] Step 201: Input the historical control parameters into the first neural network model to obtain the predicted observation data;

[0060] Step 202: Match the predicted observation data with the historical observation data corresponding to the historical control parameters to obtain the matching result;

[0061] Step 203: Adjust the network parameters of the first neural network model according to the matching result until the predicted observation data obtained after inputting any of the historical control parameters into the first neural network model matches the historical observation data corresponding to the historical control parameters.

[0062] The training of the first neural network model can be performed using multiple historical operating data sets from polysilicon preparation in a reduction furnace. During a specific polysilicon preparation process in the reduction furnace, the control parameters used in that preparation can be recorded as historical control parameters, and the observed data from that preparation can be recorded as historical observation data. These historical control parameters and historical observation data are then combined into a single historical operating data set.

[0063] After acquiring multiple historical operational data sets, the historical control parameters of a specific historical operational data set can be input into the first neural network model. The observation data output by the first neural network model is then used as the predicted observation data. This predicted observation data is then compared with the historical observation data within the historical operational data set. The difference between the two, calculated using the loss function, is determined to be less than a preset difference value. If so, the two data sets match; otherwise, gradient descent is used to backpropagate the network parameters of the first neural network model using the difference value, and then training is repeated. Training of the first neural network model is complete when the difference between the predicted observation data obtained after inputting any historical control parameter into the first neural network model and the corresponding historical observation data is less than the preset difference value.

[0064] The first neural network model can be any of the following: recurrent neural network (RNN), fully connected neural network (FCN), generative adversarial network (GAN), long short-term memory network (LSTM), and Hopfield network.

[0065] After training the first neural network model, the set control parameters can be input into the trained first neural network model to predict the observation data corresponding to each control parameter.

[0066] In some embodiments, for each historical observation data point, the corresponding polysilicon production state, i.e., the polysilicon production energy consumption and yield, are known. Therefore, historical observation data points where the polysilicon production state meets preset requirements can be selected from the historical observation data points and used as preset observation data points. The preset requirements can be the average value of the polysilicon production state corresponding to each historical observation data point, i.e., the average value of the polysilicon production energy consumption and the average value of the polysilicon yield corresponding to each historical observation data point.

[0067] To more accurately determine the preset observation data and further improve the subsequent polysilicon production efficiency, such as... Figure 4 As shown, in some embodiments, it also includes:

[0068] Step 301: Input the historical observation data into the trained second neural network model for clustering to obtain the clustering results of the historical observation data.

[0069] Step 302: Determine the historical observation data whose clustering result is a positive sample as the preset observation data;

[0070] The second neural network model is trained from a model training set composed of a portion of the historical observation data from each of the historical observation data.

[0071] The positive sample corresponds to polysilicon production energy consumption and yield that meet the preset requirements.

[0072] In some embodiments, the second neural network model can be a classification model such as a BP neural network. Since the input layer and output layer of a BP neural network are the same, there can be multiple intermediate layers. The number of nodes in the intermediate layers is less than that in the output layer. In the simplest case, there is only one hidden layer in the middle, and the number of nodes in the intermediate layers is relatively small. Therefore, such intermediate layers are equivalent to compressing and abstracting the data, which can achieve unsupervised learning of the abstract features of the data.

[0073] Training the second neural network model can involve extracting historical observation data from various historical observation datasets as training samples. These training samples are then input into the second neural network model, which encodes and classifies them, determining whether each sample is a positive or negative sample. The classification result is then compared to the pre-defined category of the training sample. If the classification result is positive, and the pre-defined category is also positive, a match is successful. If a match fails, the network parameters of the second neural network model are adjusted, and training continues until the category output by the second neural network model for any training sample matches its pre-defined category. Here, a positive sample indicates that the corresponding polysilicon production energy consumption and yield meet preset requirements.

[0074] In some embodiments, training the second neural network model can focus solely on learning the patterns of positive samples. That is, each training sample can be a pre-labeled positive sample. In this case, a single-classification second neural network model can be trained using these positive training samples. The second neural network model can then determine whether the input historical observation data belongs to the positive sample by comparing the similarity between the input layer and the output layer.

[0075] For example, when classifying training samples using a second neural network model, the decision_function(X) function in the second neural network model returns the signed distance from the training sample to the hyperplane, which can be used to determine whether the training sample is a positive sample. Specifically, if the signed distance is positive, the training sample is a positive sample; if the signed distance is negative, the training sample is also a positive sample.

[0076] After training the second neural network model, the historical observation data can be input into the trained second neural network model. The second neural network model encodes the historical observation data, and then performs single-class clustering on the encoded data, thereby clustering the historical observation data into positive samples and negative samples.

[0077] After the clustering of each historical observation data is completed, each historical observation data that is clustered as a positive sample can be identified as a preset observation data. Each preset observation data is then matched with each observation data obtained through the first neural network model, so as to obtain the observation data that matches at least one preset observation data from each observation data as the target observation data.

[0078] Considering that multiple observation data points may match preset observation data, in order to extract better observation data from these matching observation data points so that the subsequently acquired control parameters can further improve the production efficiency and yield of polysilicon, in some embodiments, observation data points that match any preset observation data point can be marked as candidate observation data points. If there is only one candidate observation data point, it is directly used as the target observation data point. If there are multiple candidate observation data points, the candidate observation data point with a polysilicon production state greater than the average polysilicon production state of all candidate observation data points is selected as the target observation data point. For example, the candidate observation data point with the optimal polysilicon production state is selected as the target observation data point.

[0079] Once the target observation data is determined, the control parameters corresponding to the target observation data can be found from the various control parameters to control the operation of the polysilicon reduction furnace. This allows the polysilicon reduction furnace to operate in an environment that can improve polysilicon yield and output, thereby reducing energy consumption per unit output and improving polysilicon production efficiency without reducing polysilicon quality.

[0080] The control device for the polysilicon reduction furnace provided in this application is described below. The control device for the polysilicon reduction furnace described below can be referred to in correspondence with the control method for the polysilicon reduction furnace described above.

[0081] In some embodiments, such as Figure 5 As shown, a control device for a polysilicon reduction furnace is provided, comprising:

[0082] The observation data acquisition module 210 is used to input the control parameters of the polysilicon reduction furnace into the trained first neural network model and acquire the observation data corresponding to each of the control parameters.

[0083] The control parameter determination module 220 is used to obtain target observation data that matches at least one preset observation data from each of the observation data, so as to control the operation of the polysilicon reduction furnace according to the control parameters corresponding to the target observation data.

[0084] The first neural network model is trained from historical operating data, which includes historical control parameters and historical observation data corresponding to the historical control parameters.

[0085] The preset observation data refers to the historical observation data in which the corresponding polysilicon production status reaches the preset requirements. The polysilicon production status includes production energy consumption and yield.

[0086] By inputting the control parameters of the polysilicon reduction furnace into a trained first neural network model, and obtaining the observation data corresponding to each control parameter, the target observation data that matches at least one preset observation data that indicates the polysilicon production state has reached the preset requirements is then obtained from the observation data. Based on the control parameters corresponding to the target observation data, the operation of the polysilicon reduction furnace is controlled, thereby optimizing the control parameters of the polysilicon reduction furnace. This reduces energy consumption per unit output and improves polysilicon production efficiency without compromising polysilicon quality.

[0087] In some embodiments, the observation data acquisition module 210 is further configured to:

[0088] The historical control parameters are input into the first neural network model to obtain the predicted observation data;

[0089] The predicted observation data is matched with the historical observation data corresponding to the historical control parameters to obtain the matching result;

[0090] The network parameters of the first neural network model are adjusted according to the matching result until the predicted observation data obtained after any of the historical control parameters are input into the first neural network model matches the historical observation data corresponding to the historical control parameters.

[0091] In some embodiments, the control parameter determination module 220 is further configured to:

[0092] The historical observation data are input into the trained second neural network model for clustering to obtain the clustering results of the historical observation data.

[0093] The historical observation data whose clustering results are positive samples are the preset observation data;

[0094] The second neural network model is trained from a model training set composed of a portion of the historical observation data from each of the historical observation data.

[0095] The positive sample corresponds to polysilicon production energy consumption and yield that meet the preset requirements.

[0096] In some embodiments, each of the historical observation data in the model training set is a positive sample.

[0097] In some embodiments, the control parameter determination module 220 is specifically used for:

[0098] From each of the observation data, obtain each alternative observation data that matches at least one preset observation data;

[0099] From the candidate observation data, obtain the target observation data in which the polysilicon production state is greater than the average polysilicon production state of the candidate observation data.

[0100] In some embodiments, the control parameters include at least one of raw material flow rate, heating current, and feed temperature.

[0101] In some embodiments, the observation data includes at least one of the following: furnace temperature data, exhaust gas data, and cooling water data of the polysilicon reduction furnace.

[0102] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call a computer program stored in the memory 830 to execute a control method for the polysilicon reduction furnace, such as including:

[0103] The control parameters of the polysilicon reduction furnace are input into the trained first neural network model to obtain the observation data corresponding to each control parameter.

[0104] From each of the observation data, target observation data that matches at least one preset observation data is obtained, so as to control the operation of the polysilicon reduction furnace according to the control parameters corresponding to the target observation data;

[0105] The first neural network model is trained from historical operating data, which includes historical control parameters and historical observation data corresponding to the historical control parameters.

[0106] The preset observation data refers to the historical observation data in which the corresponding polysilicon production status reaches the preset requirements. The polysilicon production status includes production energy consumption and yield.

[0107] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] On the other hand, this application also provides a storage medium, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the control method of the polysilicon reduction furnace provided in the above embodiments.

[0109] Computer-readable storage media can be any available medium or data storage device that can be accessed by a processor, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MOs), etc.), optical storage (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).

[0110] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0111] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A control method for a polycrystalline silicon reduction furnace, characterized in that, include: The control parameters of the polysilicon reduction furnace are input into the trained first neural network model to obtain the observation data corresponding to each control parameter. From each of the observation data, target observation data that matches at least one preset observation data is obtained, so as to control the operation of the polysilicon reduction furnace according to the control parameters corresponding to the target observation data; The first neural network model is trained from historical operating data, which includes historical control parameters and historical observation data corresponding to the historical control parameters. The preset observation data are the historical observation data in which the corresponding polysilicon production status reaches the preset requirements. The polysilicon production status includes production energy consumption and yield. The method further includes: The historical observation data are input into the trained second neural network model for clustering to obtain the clustering results of the historical observation data. The historical observation data whose clustering results are positive samples are the preset observation data; The second neural network model is trained from a model training set composed of a portion of the historical observation data from each of the historical observation data. The positive sample corresponds to polysilicon production energy consumption and yield that meet the preset requirements.

2. The control method for the polycrystalline silicon reduction furnace according to claim 1, characterized in that, Also includes: The historical control parameters are input into the first neural network model to obtain the predicted observation data; The predicted observation data is matched with the historical observation data corresponding to the historical control parameters to obtain the matching result; The network parameters of the first neural network model are adjusted according to the matching result until the predicted observation data obtained after any of the historical control parameters are input into the first neural network model matches the historical observation data corresponding to the historical control parameters.

3. The control method for the polycrystalline silicon reduction furnace according to claim 1, characterized in that, The historical observation data in the model training set are the positive samples.

4. The control method for the polycrystalline silicon reduction furnace according to claim 1, characterized in that, From the observation data, target observation data matching at least one preset observation data is obtained, including: From each of the observation data, obtain each alternative observation data that matches at least one preset observation data; From the candidate observation data, obtain the target observation data in which the polysilicon production state is greater than the average polysilicon production state of the candidate observation data.

5. The control method for the polycrystalline silicon reduction furnace according to claim 1, characterized in that, The control parameters include at least one of the following: raw material flow rate, heating current, and feed temperature.

6. The control method for a polycrystalline silicon reduction furnace according to claim 1 or 5, characterized in that, The observation data includes at least one of the following: furnace temperature data, exhaust gas data, and cooling water data of the polysilicon reduction furnace.

7. A control device for a polycrystalline silicon reduction furnace, characterized in that, include: The observation data acquisition module is used to input the control parameters of the polysilicon reduction furnace into the trained first neural network model and acquire the observation data corresponding to each control parameter. The control parameter determination module is used to obtain target observation data that matches at least one preset observation data from each of the observation data, so as to control the operation of the polysilicon reduction furnace according to the control parameters corresponding to the target observation data; The first neural network model is trained from historical operating data, which includes historical control parameters and historical observation data corresponding to the historical control parameters. The preset observation data are the historical observation data in which the corresponding polysilicon production status reaches the preset requirements. The polysilicon production status includes production energy consumption and yield. The control parameter determination module is also used for: The historical observation data are input into the trained second neural network model for clustering to obtain the clustering results of the historical observation data. The historical observation data whose clustering results are positive samples are the preset observation data; The second neural network model is trained from a model training set composed of a portion of the historical observation data from each of the historical observation data. The positive sample corresponds to polysilicon production energy consumption and yield that meet the preset requirements.

8. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the control method for the polysilicon reduction furnace according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the control method for the polycrystalline silicon reduction furnace according to any one of claims 1 to 6.

Images