CN114330647A - Model training method and device and silicon rod weight prediction method - Google Patents
Model training method and device and silicon rod weight prediction method Download PDFInfo
- Publication number
- CN114330647A CN114330647A CN202111503925.4A CN202111503925A CN114330647A CN 114330647 A CN114330647 A CN 114330647A CN 202111503925 A CN202111503925 A CN 202111503925A CN 114330647 A CN114330647 A CN 114330647A
- Authority
- CN
- China
- Prior art keywords
- data set
- data
- silicon rod
- weight
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- XUIMIQQOPSSXEZ-UHFFFAOYSA-N Silicon Chemical compound [Si] XUIMIQQOPSSXEZ-UHFFFAOYSA-N 0.000 title claims abstract description 120
- 229910052710 silicon Inorganic materials 0.000 title claims abstract description 118
- 239000010703 silicon Substances 0.000 title claims abstract description 118
- 238000000034 method Methods 0.000 title claims abstract description 66
- 238000012549 training Methods 0.000 title claims abstract description 62
- 230000009467 reduction Effects 0.000 claims abstract description 65
- 230000007787 long-term memory Effects 0.000 claims abstract description 28
- 230000006403 short-term memory Effects 0.000 claims abstract description 28
- 238000012545 processing Methods 0.000 claims abstract description 19
- 238000010606 normalization Methods 0.000 claims abstract description 16
- 238000012360 testing method Methods 0.000 claims description 41
- 230000015654 memory Effects 0.000 claims description 32
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000003860 storage Methods 0.000 claims description 10
- 238000007781 pre-processing Methods 0.000 claims description 8
- 210000002569 neuron Anatomy 0.000 claims description 5
- 238000005516 engineering process Methods 0.000 abstract description 7
- 238000004519 manufacturing process Methods 0.000 description 19
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 18
- 239000007789 gas Substances 0.000 description 15
- 239000011521 glass Substances 0.000 description 8
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 7
- 239000001257 hydrogen Substances 0.000 description 7
- 229910052739 hydrogen Inorganic materials 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- ZDHXKXAHOVTTAH-UHFFFAOYSA-N trichlorosilane Chemical compound Cl[SiH](Cl)Cl ZDHXKXAHOVTTAH-UHFFFAOYSA-N 0.000 description 7
- 239000005052 trichlorosilane Substances 0.000 description 7
- 230000006870 function Effects 0.000 description 6
- 229910021420 polycrystalline silicon Inorganic materials 0.000 description 6
- 230000004913 activation Effects 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 210000004027 cell Anatomy 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 239000002994 raw material Substances 0.000 description 2
- 239000011863 silicon-based powder Substances 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 238000005303 weighing Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000005229 chemical vapour deposition Methods 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000000151 deposition Methods 0.000 description 1
- 230000008021 deposition Effects 0.000 description 1
- RXKJFZQQPQGTFL-UHFFFAOYSA-N dihydroxyacetone Chemical compound OCC(=O)CO RXKJFZQQPQGTFL-UHFFFAOYSA-N 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000013450 outlier detection Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Landscapes
- Silicon Compounds (AREA)
Abstract
The application discloses a model training method and device and a silicon rod weight prediction method. The model training method comprises the following steps: acquiring a sample data set, wherein each sample data in the sample data set comprises: the method comprises the steps that operation data of a reduction furnace under preset operation time and generated silicon rod weight data are obtained, wherein the preset operation time corresponding to each sample data is different; carrying out normalization processing on each sample data in the sample data set to obtain a normalized data set; establishing a long-term and short-term memory network model; and training the long-term and short-term memory network model based on the normalized data set to obtain a target prediction model, wherein the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace. The method and the device solve the technical problem that the weight of the silicon rod in the reducing furnace is difficult to accurately determine in the related technology.
Description
Technical Field
The application relates to the technical field of polycrystalline silicon production, in particular to a model training method and device and a silicon rod weight prediction method.
Background
Polycrystalline silicon is a basic raw material in the solar industry and the electronic industry, and is widely applied to manufacturing of solar panels and semiconductor chips. At present, an improved Siemens method is mostly adopted in a polycrystalline silicon production technology, a polycrystalline silicon production link of the method is to electrify a silicon rod arranged in a reduction furnace, the silicon rod generates heat to generate high temperature, trichlorosilane and hydrogen are used as raw materials to generate chemical vapor deposition reaction on the surface of the high-temperature silicon rod to obtain polycrystalline silicon, and the diameter of the silicon rod gradually increases to the specified weight along with the continuous deposition of the polycrystalline silicon on the surface of the silicon rod, so that the production of one batch is finished. As the reduction furnace is a closed high-pressure container, only 1-4 parts of the wall of the furnace cylinder are provided with sight holes, and sight glasses are arranged on the sight holes for manually observing the conditions in the reduction furnace. Because the weight and the diameter of the silicon rod are in a positive correlation, production operation parameters are adjusted by regularly observing the diameter change of the silicon rod in the sight glass by production personnel, and the production is stopped when the production personnel judge that the silicon rod grows to the specified diameter. Since the weight of the silicon rod cannot be directly measured, the method for manually observing the silicon rod through a sight glass to estimate the diameter of the silicon rod and then the weight of the silicon rod is too subjective, and a quantitative result of the weight of the silicon rod cannot be obtained. This results in the fact that the operating parameters cannot be adjusted accurately in time according to the change in the silicon rod weight during the operation, and the inaccurate estimation of the silicon rod weight leads to a batch stopping production too early or too late, which leads to inconsistent silicon rod weight for each batch and poor reproducibility of production.
In order to solve the problem of measuring the weight of the silicon rod, in the prior art, an industrial camera is arranged outside a sight glass to capture a silicon rod image, so that the diameter of the silicon rod is calculated, and the weight of the silicon rod is calculated according to the diameter of the silicon rod. The method for measuring and calculating the weight of the silicon rod by measuring and calculating the diameter of the silicon rod by capturing the silicon rod image in the reducing furnace by using the sight glass has the following defects: firstly, silicon powder is sometimes generated in the operation process of the reduction furnace, the inner side of the sight glass is gradually polluted and blackened by the attachment of the silicon powder, the condition in the furnace cannot be observed, and the diameter of the silicon rod cannot be measured; secondly, the observation visual field of the sight glass is small, only a small part of the silicon rod can be measured, and images of all parts of all the silicon rods cannot be obtained; and thirdly, the gas in the furnace is disturbed, so that the edge of the silicon rod in the obtained image is distorted and deformed, and the accurate diameter of the silicon rod cannot be calculated. Therefore, the method for measuring and calculating the weight of the silicon rod by measuring and calculating the diameter of the silicon rod by acquiring the silicon rod image through the reduction furnace sight glass in the prior art has great limitation and low practicability.
In view of the above problems, no effective solution has been proposed.
Disclosure of Invention
The embodiment of the application provides a model training method and device and a silicon rod weight prediction method, and aims to at least solve the technical problem that the weight of a silicon rod in a reduction furnace is difficult to accurately determine in the related technology.
According to an aspect of an embodiment of the present application, there is provided a model training method, including: acquiring a sample data set, wherein each sample data in the sample data set comprises: running data of the reduction furnace under a preset running time and generated silicon rod weight data, wherein the preset running time corresponding to each sample data is different; carrying out normalization processing on each sample data in the sample data set to obtain a normalized data set; establishing a long-term and short-term memory network model; training the long-term and short-term memory network model based on the normalized data set to obtain a target prediction model, wherein the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace.
Optionally, determining a plurality of preset operation durations; for any one preset operation time, acquiring operation data of the reduction furnace under the preset operation time and generated silicon rod weight data; preprocessing the operating data to obtain preprocessed data; splicing the preprocessing data and the silicon rod weight data to obtain the sample data corresponding to the preset operation time; and determining the sample data corresponding to each preset operation time length to obtain the sample data set.
Optionally, determining an input layer, a hidden layer and an output layer in the long-short term memory network model, wherein the hidden layer comprises: an input gate, a forgetting gate and an output gate; determining network computation parameters of the long-short term memory network model, wherein the network computation parameters comprise: the number of hidden layer neurons, the learning rate, and the number of iterations.
Optionally, dividing the normalized data set into a training data set and a testing data set; and training the long-short term memory network model based on the training data set, and testing the trained long-short term memory network model based on the test data set to obtain the target prediction model.
Optionally, the long-short term memory network model is trained through the training data set, and the trained long-short term memory network model is tested through the test data set to obtain a test result, wherein the test result is a root mean square error between a predicted value and a true value of the long-short term memory network model; and when the test result is not less than the preset threshold value, adjusting the network calculation parameters of the long-short term memory network model, continuing to train the long-short term memory network model after the network calculation parameters are adjusted through the training data set, and testing the trained long-short term memory network model through the test data set until the test result is less than the preset threshold value.
According to another aspect of the embodiments of the present application, there is also provided a method for predicting a weight of a silicon rod, including: acquiring operation data of the reduction furnace within a target operation duration; inputting the operation data into a pre-trained target prediction model to obtain a model output value, wherein the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace; determining a weight of the silicon rod based on the model output value.
Optionally, normalizing the operation data to obtain normalized data, and inputting the normalized data into the target prediction model; and carrying out reverse normalization processing on the model output value to obtain the weight of the silicon rod.
According to another aspect of the embodiments of the present application, there is also provided a model training apparatus, including: an obtaining module, configured to obtain a sample data set, where each sample data in the sample data set includes: running data of the reduction furnace under a preset running time and generated silicon rod weight data, wherein the preset running time corresponding to each sample data is different; the processing module is used for carrying out normalization processing on each sample data in the sample data set to obtain a normalized data set; the model establishing module is used for establishing a long-term and short-term memory network model; and the training module is used for training the long-term and short-term memory network model based on the normalized data set to obtain a target prediction model, and the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace.
According to another aspect of the embodiments of the present application, there is also provided a non-volatile storage medium, which includes a stored program, wherein when the program runs, the apparatus on which the non-volatile storage medium is located is controlled to perform the model training method or the silicon rod weight prediction method.
According to another aspect of the embodiments of the present application, there is also provided a processor for executing a program, wherein the program is executed to perform the above-mentioned model training method or silicon rod weight prediction method.
In the embodiment of the present application, a sample data set is first obtained, where each sample data in the sample data set includes: the method comprises the steps that operation data of a reduction furnace under preset operation time and generated silicon rod weight data are obtained, wherein the preset operation time corresponding to each sample data is different; then, carrying out normalization processing on each sample data in the sample data set to obtain a normalized data set; establishing a long-term and short-term memory network model; and training the long-term and short-term memory network model based on the normalized data set to obtain a target prediction model, wherein the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace. By the target prediction model, the weight of the silicon rod generated in the reduction furnace can be accurately predicted in real time, reference is provided for production personnel to regulate and control the reduction furnace, and the technical problem that the weight of the silicon rod in the reduction furnace is difficult to accurately determine in the related technology is solved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the application and together with the description serve to explain the application and not to limit the application. In the drawings:
FIG. 1 is a schematic flow chart diagram of a model training method according to an embodiment of the present application;
fig. 2 is a schematic flow chart of a method for predicting the weight of a silicon rod according to an embodiment of the present application;
FIG. 3 is a schematic view illustrating a process for predicting the weight of a silicon rod according to an embodiment of the present disclosure;
FIG. 4 is a graphical illustration of the results of a silicon rod weight prediction in accordance with an embodiment of the present application;
fig. 5 is a schematic structural diagram of a silicon rod weight predicting device according to an embodiment of the present application.
Detailed Description
In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only partial embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
It should be noted that the terms "first," "second," and the like in the description and claims of this application and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Example 1
In order to solve the technical problem that the weight of the silicon rod in the reducing furnace is difficult to accurately determine in the related technology, the embodiment of the application provides a scheme for predicting the weight of the silicon rod in the reducing furnace based on a neural network model, and the basic idea is as follows: the method comprises the steps that local time sequence feature information of a reduction furnace under multiple dimensions is automatically extracted through a one-dimensional convolutional neural network to obtain feature vectors, and the feature vectors are sent into a long-term and short-term memory network to learn information of long-term and short-term local dependence of time sequence features; and then, an attention mechanism can be introduced into the prediction model to pay particular attention to the weight of the time step of the silicon rod weight time sequence prediction, the attention mechanism layer exchanges the time step with the input characteristics, the weight of each time step is calculated by using the full-connection layer, the obtained weight is multiplied by the characteristics input into each time step before, the weight of each time step can be given, and finally, a more accurate silicon rod weight prediction result can be obtained through the full-connection layer, so that reference is provided for production personnel to regulate and control the reduction furnace.
Based on the above, embodiments of the present application provide a model training method, it should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer-executable instructions, and that although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from the order shown.
Fig. 1 is a schematic flow chart of an alternative model training method according to an embodiment of the present application, as shown in fig. 1, the method at least includes steps S102-S108, where:
step S102, a sample data set is obtained, and each sample data in the sample data set comprises: the operation data of the reduction furnace under the preset operation time and the generated silicon rod weight data are obtained, wherein the preset operation time corresponding to each sample data is different.
Wherein, the operation data of the reduction furnace usually includes: the device comprises a power source, a current, a trichlorosilane flow, a hydrogen flow, a power, a mixed gas (trichlorosilane and hydrogen) pressure, a mixed gas (trichlorosilane and hydrogen) temperature, a voltage, a furnace barrel water temperature, a furnace barrel water return temperature, a furnace barrel water flow, a chassis water temperature, a chassis water return temperature, a chassis water flow, a tail gas pipeline jacket water temperature, a tail gas pipeline jacket water return temperature, a tail gas pipeline jacket water flow, a reduction furnace outlet tail gas temperature, a reduction furnace outlet tail gas pressure and the like.
In the operation process of the reduction furnace, the operation data and the generated silicon rod weight are changed in real time, the influence of time factors is considered, a plurality of operation time periods can be preset when sample data are collected, and the operation data of the reduction furnace in each operation time period and the generated silicon rod weight data are collected as sample data to obtain a sample data set. When the operation data is collected, a user can set a collection time interval (usually in a second level) according to an actual scene so as to collect multiple groups of operation data in corresponding operation time periods.
In some optional embodiments of the present application, when determining the sample data set, a plurality of preset operation durations may be determined; for any preset operation time, acquiring operation data of the reduction furnace under the preset operation time and generated silicon rod weight data; preprocessing the operation data to obtain preprocessed data; splicing the preprocessing data and the silicon rod weight data to obtain sample data corresponding to preset operation time; and determining sample data corresponding to each preset operation time length to obtain a sample data set.
Considering that data loss and errors are often caused by factors such as control, environment, measuring instrument instability, human error and the like when the operation data is collected, in order to ensure the correctness, reliability and stability of final sample data, the collected operation data generally needs to be preprocessed, and common preprocessing methods include but are not limited to: spline interpolation algorithm, local outlier detection algorithm, moving average algorithm, etc., which can be selected by the user, and are not limited herein.
Because the reduction furnace has a plurality of groups of operation data under a preset operation time, the obtained pretreatment data also has a plurality of groups, but only one silicon rod weight data is obtained, the pretreatment data and the silicon rod weight data can be spliced to obtain a group of operation-weight data, and the group of operation-weight data is used as sampling data corresponding to the preset operation time.
Specifically, a specific process for acquiring a sample data set provided in the embodiment of the present application is as follows:
1. the reduction furnace starts production according to the operation parameter table and records the operation starting time;
2. collecting operation data of the reduction furnace;
3. stopping production when the operation reaches the Nth hour (N is a preset operation time and is an integer which is generally greater than or equal to 20);
4. taking out the silicon rod in the reduction furnace, and weighing to obtain silicon rod weight data;
5. and splicing the operation data and the silicon rod weight data to obtain the operation-weight data of the first batch.
And repeating the steps when N is equal to N +1, and executing the same operation parameter table in each batch by the reduction furnace to obtain operation-weight data of a plurality of batches. Assuming that the initial value of N is 20, the operation is performed for 20 hours and 21 hours … … 100 hours respectively, and operation-weight data of 80 batches are obtained, i.e. the final sample data set is shown in table 1.
TABLE 1
It should be noted that the value of the preset operation duration N is only an example, and in actual application, a user may set the preset operation duration by himself, which is not limited herein.
And step S104, carrying out normalization processing on each sample data in the sample data set to obtain a normalized data set.
Because the sample data comprises the operating data under a plurality of characteristic dimensions, in order to facilitate the subsequent training of the model, the sample data needs to be normalized, and the normalization formula is as follows:
in the formula, x is the original data of a certain variable; x is the number of*Normalizing the variable to data; x is the number ofmaxAnd xminRespectively representing the maximum and minimum values of the variable before normalization.
Step S106, establishing a long-term and short-term memory network model.
In general, the long-short term memory network model includes an input layer, a hidden layer and an output layer, wherein the hidden layer includes: input gate, forget gate and output gate, the corresponding principle is as follows:
1) the input gate is mainly used for controlling the updating of the state of the unit at the current moment by using the input of the network at the current moment and the output of the hidden layer at the previous moment:
it=σ(wi·[ht-1,xt]+bi)
in the formula itAn input gate at time t; x is the number oftIs the input value of the variable at time t; w is aiIs the weight matrix of the input gate; biIs the bias term of the input gate; h ist-1Is the output value of the network at the time t-1; sigma is an activation function, and sigmoid is generally selected as the activation function.
2) The forgetting gate is mainly used for controlling how much unit state information at the previous moment is reserved to the current moment:
ft=σ(wf·[ht-1,xt]+bf)
in the formula (f)tA forgetting gate at the time t; x is the number oftIs the input value of the variable at time t; w is afA weight matrix for a forgetting gate; bfA bias term for a forget gate; h ist-1Is the time t-1An output value of the network; σ is the activation function.
3) The output gate is mainly used for controlling how much information is output by the unit at the current moment as the output of the network through the output gate before the unit outputs the information:
ot=σ(wo·[ht-1,xt]+bo)
in the formula otAn output gate at time t; x is the number oftIs the input value of the variable at time t; w is aoIs a weight matrix of the output gate; boIs the bias term of the output gate; h ist-1Is the output value of the network at the time t-1; σ is the activation function.
And obtaining the unit state at the time t according to the unit state at the time t-1 and the input gate and the forgetting gate:
in the formula (I), the compound is shown in the specification,the state of the input unit at the moment t; wcA weight matrix that is a cell state; bcA bias term that is a cell state; c. CtCell state at time t; c. Ct-1The cell state at time t-1.
Obtaining the network output at the time t according to the unit state and the output gate at the time t:
ht=ot·tanh(ct)
in the formula, htFor the network output at the time t, tanh is a hyperbolic tangent function, and the formula forms a long-term and short-term memory network model.
Then, determining network calculation parameters of the long-term and short-term memory network model, wherein the network calculation parameters comprise: the number of hidden layer neurons, the learning rate, and the number of iterations. For example, the hidden layer neuron number initial value is set to 32, the learning rate initial value is set to 0.05, and the iteration number initial value is set to 500.
And S108, training the long-term and short-term memory network model based on the normalized data set to obtain a target prediction model, wherein the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace.
When training the model, the normalized data set may be first divided into a training data set and a test data set, for example, for the sample data set including 80 sets of data, the data may be represented by 9: 1, dividing the first 90% of the sample data set into a training data set, and dividing the last 10% into a test data set, so as to obtain 72 groups of training data sets and 8 groups of test data sets; and then training the long-short term memory network model based on the training data set, and testing the trained long-short term memory network model based on the test data set to obtain a target prediction model.
Specifically, the long-short term memory network model can be trained through the training data set, and the trained long-short term memory network model is tested through the testing data set to obtain a testing result, wherein the testing result is the root mean square error between the predicted value and the true value of the long-short term memory network model. The root mean square error is used for evaluating the accuracy of the model prediction result, and the calculation formula is as follows:
in the formula, yRMSERepresenting the root mean square error, and n representing the number of samples of the test data set; y isact(i)And ypred(i)And respectively representing the real value and the predicted value of the silicon rod weight in the ith group of data in the test data set.
The user may set a threshold for the test result, e.g. 1 x 10-5When the test result is not less than the preset threshold value, the network calculation parameters of the long-term and short-term memory network model need to be adjusted, and then the adjustment network continues to be adjusted through the training data setTraining the long-term and short-term memory network model after the parameters are calculated, testing the trained long-term and short-term memory network model through the test data set until the test result is smaller than a preset threshold value, and obtaining a final target prediction model which can accurately predict the weight of the silicon rod generated in the reduction furnace.
Based on the target prediction model obtained by the above model training method, the embodiments of the present application further provide a silicon rod weight prediction method, it should be noted that the steps shown in the flowchart of the drawings may be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from the order shown.
Fig. 2 is a schematic flow chart of an alternative silicon rod weight prediction method according to an embodiment of the present application, as shown in fig. 2, the method at least includes steps S202-S206, wherein:
step S202, obtaining operation data of the reduction furnace in the target operation time length.
Wherein, the operation data of the reduction furnace comprises: the device comprises a power source, a current, a trichlorosilane flow, a hydrogen flow, a power, a mixed gas (trichlorosilane and hydrogen) pressure, a mixed gas (trichlorosilane and hydrogen) temperature, a voltage, a furnace barrel water temperature, a furnace barrel water return temperature, a furnace barrel water flow, a chassis water temperature, a chassis water return temperature, a chassis water flow, a tail gas pipeline jacket water temperature, a tail gas pipeline jacket water return temperature, a tail gas pipeline jacket water flow, a reduction furnace outlet tail gas temperature, a reduction furnace outlet tail gas pressure and the like.
And S204, inputting the operation data into a pre-trained target prediction model to obtain a model output value, wherein the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace.
Specifically, the acquired operation data may be normalized to obtain normalized data, and the normalized data may be input into the target prediction model to obtain a model output value.
And step S206, determining the weight of the silicon rod based on the model output value.
Specifically, the prediction result of the silicon rod weight can be obtained by performing inverse normalization processing on the model output value.
The silicon rod weight prediction method in the embodiment of the application has strong universality, and can be applied to various reduction furnace types to predict the weight of the generated silicon rod.
Fig. 3 shows a complete silicon rod weight prediction process, which includes the following steps:
1. the reduction furnace starts production according to the operation parameter table and records the operation starting time;
2. acquiring operation data of the reduction furnace;
3. when the operation is carried out for the Nth hour, the production is stopped;
4. taking out the silicon rod in the reduction furnace, and weighing to obtain silicon rod weight data;
5. repeating steps 1-4 for a number of times, except that N for each batch is the N plus 1 hour of the previous batch;
6. preprocessing the acquired data;
7. splicing the operation data and the weight data of one batch into a group of operation-weight data, and forming an operation-weight data set (sample data set) by using the data of a plurality of batches;
8. carrying out normalization processing on the operation-weight data set;
9. dividing the normalized data set into a training data set and a testing data set;
10. constructing a long-short term memory network model, which comprises an input layer, a hidden layer and an output layer, and setting network calculation parameters, including the number of neurons in the hidden layer, the learning rate and the iteration times;
11. training the constructed long-term and short-term memory network model by using a training data set;
12. testing the trained long-term and short-term memory network model by using a test data set, and taking the root mean square error as an evaluation standard of the prediction accuracy of the model;
13. when the root mean square error is not smaller than the preset threshold, adjusting the network calculation parameters, and repeating the steps 11-12 until the root mean square error is smaller than the preset threshold;
14. obtaining a finally trained target prediction model;
15. and predicting the weight of the silicon rod in the actual production of the reduction furnace through the target prediction model.
Fig. 4 shows the results of the silicon rod weight prediction, and it can be seen that the silicon rod weight in the reduction furnace can be accurately predicted by the target prediction model.
In the embodiment of the present application, a sample data set is first obtained, where each sample data in the sample data set includes: the method comprises the steps that operation data of a reduction furnace under preset operation time and generated silicon rod weight data are obtained, wherein the preset operation time corresponding to each sample data is different; then, carrying out normalization processing on each sample data in the sample data set to obtain a normalized data set; establishing a long-term and short-term memory network model; and training the long-term and short-term memory network model based on the normalized data set to obtain a target prediction model, wherein the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace. By the target prediction model, the weight of the silicon rod generated in the reduction furnace can be accurately predicted in real time, reference is provided for production personnel to regulate and control the reduction furnace, and the technical problem that the weight of the silicon rod in the reduction furnace is difficult to accurately determine in the related technology is solved.
Example 2
According to an embodiment of the present application, there is also provided a model training apparatus for implementing a model training method, as shown in fig. 5, the apparatus at least includes an obtaining module 50, a processing module 52, a model building module 54, and a training module 56, where:
an obtaining module 50, configured to obtain a sample data set, where each sample data in the sample data set includes: the operation data of the reduction furnace under the preset operation time and the generated silicon rod weight data are obtained, wherein the preset operation time corresponding to each sample data is different.
The processing module 52 is configured to perform normalization processing on each sample data in the sample data set to obtain a normalized data set.
And the model establishing module 54 is used for establishing a long-term and short-term memory network model.
And the training module 56 is used for training the long-term and short-term memory network model based on the normalized data set to obtain a target prediction model, and the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace.
It should be noted that, each module in the silicon rod weight predicting device in the embodiment of the present application corresponds to the implementation steps of the silicon rod weight predicting method in embodiment 1 one to one, and as the detailed description is already performed in embodiment 1, some details that are not shown in this embodiment may refer to embodiment 1, and are not described herein again.
Example 3
According to an embodiment of the application, a non-volatile storage medium is further provided, and the non-volatile storage medium comprises a stored program, wherein the equipment where the non-volatile storage medium is located is controlled to execute the model training method or the silicon rod weight prediction method when the program runs.
According to an embodiment of the application, a processor for running a program is also provided, wherein the above-mentioned model training method or silicon rod weight prediction method is performed when the program is running.
Specifically, the following steps are implemented when the program runs: acquiring a sample data set, wherein each sample data in the sample data set comprises: the method comprises the steps that operation data of a reduction furnace under preset operation time and generated silicon rod weight data are obtained, wherein the preset operation time corresponding to each sample data is different; carrying out normalization processing on each sample data in the sample data set to obtain a normalized data set; establishing a long-term and short-term memory network model; and training the long-term and short-term memory network model based on the normalized data set to obtain a target prediction model, wherein the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace.
Optionally, the following steps may also be executed when the program runs: acquiring operation data of the reduction furnace within a target operation duration; inputting the operation data into a pre-trained target prediction model to obtain a model output value, wherein the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace; determining the weight of the silicon rod based on the model output value.
The above-mentioned serial numbers of the embodiments of the present application are merely for description and do not represent the merits of the embodiments.
In the above embodiments of the present application, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.
In the embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, a division of a unit may be a division of a logic function, and an actual implementation may have another division, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted, or may not be executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, units or modules, and may be in an electrical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be substantially implemented or contributed to by the prior art, or all or part of the technical solution may be embodied in a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method of the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes.
The foregoing is only a preferred embodiment of the present application and it should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present application, and these improvements and modifications should also be considered as the protection scope of the present application.
Claims (10)
1. A method of model training, comprising:
acquiring a sample data set, wherein each sample data in the sample data set comprises: running data of the reduction furnace under a preset running time and generated silicon rod weight data, wherein the preset running time corresponding to each sample data is different;
carrying out normalization processing on each sample data in the sample data set to obtain a normalized data set;
establishing a long-term and short-term memory network model;
training the long-term and short-term memory network model based on the normalized data set to obtain a target prediction model, wherein the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace.
2. The method of claim 1, wherein obtaining a set of sample data comprises:
determining a plurality of preset operation durations;
for any one preset operation time, acquiring operation data of the reduction furnace under the preset operation time and generated silicon rod weight data;
preprocessing the operating data to obtain preprocessed data;
splicing the preprocessing data and the silicon rod weight data to obtain the sample data corresponding to the preset operation time;
and determining the sample data corresponding to each preset operation time length to obtain the sample data set.
3. The method of claim 1, wherein building a long-short term memory network model comprises:
determining an input layer, a hidden layer and an output layer in the long-short term memory network model, wherein the hidden layer comprises: an input gate, a forgetting gate and an output gate;
determining network computation parameters of the long-short term memory network model, wherein the network computation parameters comprise: the number of hidden layer neurons, the learning rate, and the number of iterations.
4. The method of claim 1, wherein training the long-short term memory network model based on the normalized data set to obtain a target prediction model comprises:
dividing the normalized data set into a training data set and a testing data set;
and training the long-short term memory network model based on the training data set, and testing the trained long-short term memory network model based on the test data set to obtain the target prediction model.
5. The method of claim 4, wherein training the long-short term memory network model based on the training data set and testing the trained long-short term memory network model based on the testing data set to obtain the target prediction model comprises:
training the long-short term memory network model through the training data set, and testing the trained long-short term memory network model through the testing data set to obtain a testing result, wherein the testing result is a root mean square error between a predicted value and a true value of the long-short term memory network model;
and when the test result is not less than the preset threshold value, adjusting the network calculation parameters of the long-short term memory network model, continuing to train the long-short term memory network model after the network calculation parameters are adjusted through the training data set, and testing the trained long-short term memory network model through the test data set until the test result is less than the preset threshold value.
6. A method for predicting the weight of a silicon rod, comprising:
acquiring operation data of the reduction furnace within a target operation duration;
inputting the operation data into a pre-trained target prediction model to obtain a model output value, wherein the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace;
determining a weight of the silicon rod based on the model output value.
7. The method of claim 6, wherein:
inputting the operational data into a pre-trained target prediction model, comprising: normalizing the operation data to obtain normalized data, and inputting the normalized data into the target prediction model;
determining a weight of the silicon rod based on the model output value, comprising: and carrying out reverse normalization processing on the model output value to obtain the weight of the silicon rod.
8. A model training apparatus, comprising:
an obtaining module, configured to obtain a sample data set, where each sample data in the sample data set includes: running data of the reduction furnace under a preset running time and generated silicon rod weight data, wherein the preset running time corresponding to each sample data is different;
the processing module is used for carrying out normalization processing on each sample data in the sample data set to obtain a normalized data set;
the model establishing module is used for establishing a long-term and short-term memory network model;
and the training module is used for training the long-term and short-term memory network model based on the normalized data set to obtain a target prediction model, and the target prediction model is used for predicting the weight of the generated silicon rod according to the operation data of the reduction furnace.
9. A non-volatile storage medium, comprising a stored program, wherein the program, when executed, controls an apparatus in which the non-volatile storage medium is located to perform the model training method according to any one of claims 1 to 5 or the silicon rod weight prediction method according to any one of claims 6 to 7.
10. A processor, characterized in that the processor is configured to run a program, wherein the program is configured to perform the method of model training according to any one of claims 1 to 5 or the method of silicon rod weight prediction according to any one of claims 6 to 7 when running.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111503925.4A CN114330647A (en) | 2021-12-09 | 2021-12-09 | Model training method and device and silicon rod weight prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111503925.4A CN114330647A (en) | 2021-12-09 | 2021-12-09 | Model training method and device and silicon rod weight prediction method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114330647A true CN114330647A (en) | 2022-04-12 |
Family
ID=81049967
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111503925.4A Pending CN114330647A (en) | 2021-12-09 | 2021-12-09 | Model training method and device and silicon rod weight prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114330647A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115032891A (en) * | 2022-08-11 | 2022-09-09 | 科大智能物联技术股份有限公司 | Polycrystalline silicon reduction furnace control method based on time series prediction |
CN115583654A (en) * | 2022-10-18 | 2023-01-10 | 科大智能物联技术股份有限公司 | Polysilicon reduction furnace current control method based on simulation learning |
WO2024060440A1 (en) * | 2022-09-19 | 2024-03-28 | 中控技术股份有限公司 | Operation trajectory curve determination method and apparatus, and electronic device |
-
2021
- 2021-12-09 CN CN202111503925.4A patent/CN114330647A/en active Pending
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115032891A (en) * | 2022-08-11 | 2022-09-09 | 科大智能物联技术股份有限公司 | Polycrystalline silicon reduction furnace control method based on time series prediction |
CN115032891B (en) * | 2022-08-11 | 2022-11-08 | 科大智能物联技术股份有限公司 | Polycrystalline silicon reduction furnace control method based on time series prediction |
WO2024060440A1 (en) * | 2022-09-19 | 2024-03-28 | 中控技术股份有限公司 | Operation trajectory curve determination method and apparatus, and electronic device |
CN115583654A (en) * | 2022-10-18 | 2023-01-10 | 科大智能物联技术股份有限公司 | Polysilicon reduction furnace current control method based on simulation learning |
CN115583654B (en) * | 2022-10-18 | 2023-11-21 | 科大智能物联技术股份有限公司 | Polysilicon reduction furnace current control method based on imitation learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114330647A (en) | Model training method and device and silicon rod weight prediction method | |
CN110298501B (en) | Electrical load prediction method based on long-time and short-time memory neural network | |
CN111260030B (en) | A-TCN-based power load prediction method and device, computer equipment and storage medium | |
Le Son et al. | Remaining useful lifetime estimation and noisy gamma deterioration process | |
CN111795761B (en) | Method for predicting cabinet inlet air temperature based on reinforcement learning model | |
CN111080982A (en) | Dam safety intelligent monitoring and early warning system and method based on multiple sensors | |
CN109784528A (en) | Water quality prediction method and device based on time series and support vector regression | |
CN116707331B (en) | Inverter output voltage high-precision adjusting method and system based on model prediction | |
CN110879874A (en) | Astronomical big data optical variation curve abnormity detection method | |
CN113988210A (en) | Method and device for restoring distorted data of structure monitoring sensor network and storage medium | |
CN113139605A (en) | Power load prediction method based on principal component analysis and LSTM neural network | |
CN114583767B (en) | Data-driven wind power plant frequency modulation response characteristic modeling method and system | |
CN113868957B (en) | Residual life prediction and uncertainty quantitative calibration method under Bayes deep learning | |
CN118364963A (en) | Building material price prediction method based on LSTM neural network | |
CN118211721A (en) | Building energy consumption prediction method based on artificial intelligence | |
CN103279030B (en) | Dynamic soft measuring modeling method and device based on Bayesian frame | |
CN113151842B (en) | Method and device for determining conversion efficiency of wind-solar complementary water electrolysis hydrogen production | |
CN111539573B (en) | Power prediction method and system for wind-solar hybrid off-grid system | |
CN117553840A (en) | Instrument based on intelligent management and system thereof | |
CN117538783A (en) | Lithium ion battery state of charge estimation method based on time domain fusion converter | |
CN116960962A (en) | Mid-long term area load prediction method for cross-area data fusion | |
CN115936236A (en) | Method, system, equipment and medium for predicting energy consumption of cigarette factory | |
CN115907178A (en) | Clean ecosystem CO 2 Method for predicting exchange amount | |
CN114971062A (en) | Photovoltaic power prediction method and device | |
CN113011674A (en) | Photovoltaic power generation prediction method and device, electronic equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB02 | Change of applicant information | ||
CB02 | Change of applicant information |
Country or region after: China Address after: No. 309 Liuhe Road, Binjiang District, Hangzhou City, Zhejiang Province (High tech Zone) Applicant after: Zhongkong Technology Co.,Ltd. Address before: No. six, No. 309, Binjiang District Road, Hangzhou, Zhejiang Applicant before: ZHEJIANG SUPCON TECHNOLOGY Co.,Ltd. Country or region before: China |