WO2024021254A1

WO2024021254A1 - Intelligent separation and purification system for electronic-grade chlorine trifluoride

Info

Publication number: WO2024021254A1
Application number: PCT/CN2022/119303
Authority: WO
Inventors: 李嘉磊; 陈施华; 华辉; 肖珏英; 陈碧灵
Original assignee: 福建德尔科技股份有限公司
Priority date: 2022-07-29
Filing date: 2022-09-16
Publication date: 2024-02-01
Also published as: CN115231525B; CN115231525A

Abstract

The present application relates to the field of intelligent production lines and particularly discloses an intelligent separation and purification system for electronic-grade chlorine trifluoride. The intelligent separation and purification system for electronic-grade chlorine trifluoride adopts artificial intelligence-based control technology, takes a purity value of a chlorine trifluoride product obtained after primary purification, a first temperature of a first alkali metal adsorbent bed, a second temperature of a second alkali metal adsorbent bed, and a third temperature of a third alkali metal adsorbent bed as input data, and uses a deep neural network model as a feature extractor, to comprehensively carry out intelligent control and judgment of a purification apparatus for the electronic-grade chlorine trifluoride. In this way, the purification and separation effect can be accurately regulated and controlled in real time to optimize the purity, thereby improving the purification effect of the electronic-grade chlorine trifluoride.

Description

Intelligent separation and purification system for electronic grade chlorine trifluoride

Technical field

The present invention relates to the field of smart production lines, and more specifically, to an intelligent separation and purification system for electronic grade chlorine trifluoride.

Background technique

Chlorine trifluoride is a highly oxidizing chip etching cleaning agent. Currently, there are very few manufacturers in the world that can prepare electronic-grade chlorine trifluoride, and only domestic applicants have the ability to prepare electronic-grade chlorine trifluoride. This is because chlorine trifluoride easily associates with hydrogen fluoride to form multi-polymers with special intermolecular forces. Traditional separation methods cannot completely solve the problem of multi-polymer separation. This is the key to preparing electronic-grade chlorine trifluoride. technical challenge.

Therefore, it is crucial to develop an intelligent separation and purification solution for electronic-grade chlorine trifluoride.

Contents of the invention

In order to solve the above technical problems, this application is proposed. Embodiments of the present application provide an intelligent separation and purification system for electronic-grade chlorine trifluoride, which uses artificial intelligence-based control technology to pass the purity value of the chlorine trifluoride product after primary purification, the first alkali metal adsorbent The first temperature of the layer bed, the second temperature of the second alkali metal adsorbent layer bed, and the third temperature of the third alkali metal adsorbent layer bed are used as input data, and a deep neural network model is used as a feature extractor to comprehensively analyze the electron The purification device of grade chlorine trifluoride is intelligently controlled. In this way, the effect of purification and separation can be accurately controlled in real time to optimize the purity, thereby improving the purification effect of the electronic grade chlorine trifluoride.

According to one aspect of the present application, an intelligent separation and purification system for electronic grade chlorine trifluoride is provided, which includes:

The data acquisition module is used to obtain the purity value of the chlorine trifluoride product after primary purification at multiple predetermined time points within a predetermined time period, the first temperature of the first alkali metal adsorbent layer bed, the second alkali metal adsorbent The second temperature of the layer bed and the third temperature of the third alkali metal adsorbent layer bed; a temperature data structuring module used to combine the values of the first alkali metal adsorbent layer bed at multiple predetermined time points within the predetermined time period. The first temperature, the second temperature of the second alkali metal adsorbent layer bed, and the third temperature of the third alkali metal adsorbent layer bed are arranged as a temperature control matrix according to the time dimension and sample dimension; the temperature data local correlation encoding module is used to Pass the temperature control matrix through the first convolutional neural network as a feature extractor to obtain a temperature control local correlation feature map;

The temperature data global correlation encoding module is used to pass the temperature control local correlation feature map through a non-local neural network to obtain the temperature control global correlation feature map; the fusion module is used to fuse the temperature control local correlation feature map and the temperature Control the global correlation feature map to obtain the temperature control feature map; the dimensionality reduction module is used to perform global mean pooling on each feature matrix of the temperature control feature map to obtain the temperature control feature vector; the correction module is used to perform the temperature control feature vector The control feature vector is corrected to obtain the corrected temperature control feature vector; the product purity data encoding module is used to encode the purity value of the first-level purified chlorine trifluoride product at multiple predetermined time points within the predetermined time period by including A temporal encoder of a one-dimensional convolutional layer to obtain a product purity feature vector; a responsiveness estimation module for calculating a control transfer matrix of the corrected temperature control feature vector relative to the product purity feature vector; and

A control result generation module is used to pass the control transfer matrix through a classifier to obtain a classification result. The classification result is used to indicate whether the temperature control combination of the three-stage metal adsorbent layer bed within a predetermined time period meets the predetermined requirements.

In the above-mentioned intelligent separation and purification system of electronic grade chlorine trifluoride, the temperature data structuring module includes: a row vector construction unit for adsorbing the first alkali metal at multiple predetermined time points within the predetermined time period. The first temperature of the agent bed, the second temperature of the second alkali metal adsorbent bed, and the third temperature of the third alkali metal adsorbent bed are respectively arranged as row vectors according to the time dimension to obtain multiple row vectors; A matrix construction unit configured to arrange the plurality of row vectors into the temperature control matrix according to the sample dimensions.

In the above-mentioned intelligent separation and purification system of electronic grade chlorine trifluoride, the temperature data local correlation encoding module is further used to: use each layer of the first convolutional neural network as a feature extractor in the forward direction of the layer. During the transfer, the input data are separately processed: convolution processing is performed on the input data to obtain a convolution feature map; mean pooling processing is performed on the convolution feature map to obtain a pooled feature map; and, the pooled feature map is obtained. Perform nonlinear activation to obtain an activation feature map; wherein, the output of the last layer of the first convolutional neural network as a feature extractor is the temperature control local correlation feature map, and the first layer as a feature extractor The input of the first layer of the convolutional neural network is the temperature control matrix.

In the above-mentioned intelligent separation and purification system of electronic grade chlorine trifluoride, the temperature data global correlation encoding module includes: a point convolution unit for inputting the temperature control local correlation feature map into the non-local neural network respectively. The first point convolution layer, the second point convolution layer and the third point convolution layer to obtain the first feature map, the second feature map and the third feature map; the first fusion unit is used to calculate the first The position-weighted sum of the feature map and the second feature map is used to obtain an intermediate fusion feature map; a normalization unit is used to input the intermediate fusion feature map into a Softmax function to calculate the position of each position in the intermediate fusion feature map. The feature values are normalized to obtain the normalized intermediate fusion feature map; the second fusion unit is used to calculate the position-weighted sum of the normalized intermediate fusion feature map and the third feature map to obtain the re-fusion feature Figure; a global perception unit, used to embed the re-fused feature map into a Gaussian similarity function to calculate the similarity between the feature values of each position in the re-fused feature map to obtain a global perception feature map; the channel number adjustment unit , used to pass the global perceptual feature map through the fourth point convolution layer of the non-local neural network to obtain a channel-adjusted global perceptual feature map; and, a third fusion unit, used to calculate the channel-adjusted global perceptual feature The position-weighted sum of the temperature control local correlation feature map and the temperature control local correlation feature map is used to obtain the temperature control global correlation feature map.

In the above intelligent separation and purification system of electronic grade chlorine trifluoride, the fusion module is further used to: fuse the temperature control local correlation feature map and the temperature control global correlation feature map using the following formula to obtain the temperature Control feature map; where, the formula is:

F _s =αF ₁ +βF ₂

Among them, F _s is the temperature control feature map, F ₁ is the temperature control local correlation feature map, F ₂ is the temperature control global correlation feature map, "+" means the temperature control local correlation feature map and all the temperature control local correlation feature maps. The elements at corresponding positions in the temperature control global correlation feature map are added together, α and β are used to control the relationship between the temperature control local correlation feature map and the temperature control global correlation feature map in the temperature control feature map. Balanced weighting parameters.

In the above-mentioned intelligent separation and purification system of electronic grade chlorine trifluoride, the correction module is further used to: correct the temperature control feature vector with the following formula to obtain the corrected temperature control feature vector; wherein, The formula is:

Where V represents the temperature control eigenvector, ∑ is the autocovariance matrix of the temperature control eigenvector, μ and σ are the global mean and variance of the temperature control eigenvector respectively, and exp(·) represents the exponential operation of the vector. , the exponential operation raised to the power of a vector represents the value of the natural exponential function raised to the power of the value of each position of the vector,

and

Represents position-wise subtraction and addition of feature vectors respectively,

represents matrix multiplication, ||·|| ₂ represents the second norm of the eigenvector.

In the above-mentioned intelligent separation and purification system of electronic grade chlorine trifluoride, the product purity data encoding module is further used to: convert the first-level purified chlorine trifluoride product at multiple predetermined time points within the predetermined time period. The purity values are arranged into a one-dimensional input vector according to the time dimension; use the fully connected layer of the temporal encoder to perform fully connected encoding on the input vector with the following formula to extract the feature values of each position in the input vector High-dimensional latent features, where the formula is:

where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector,

represents matrix multiplication; use the one-dimensional convolution layer of the temporal encoder to perform one-dimensional convolution encoding on the input vector with the following formula to extract high-dimensional implicit correlation features between the eigenvalues of each position in the input vector, Among them, the formula is:

Among them, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix that operates with the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector.

In the above-mentioned intelligent separation and purification system of electronic grade chlorine trifluoride, the responsiveness estimation module is further used to: calculate the control of the corrected temperature control characteristic vector relative to the product purity characteristic vector according to the following formula Transfer matrix; where the formula is:

S＝T*F

Where F represents the corrected temperature control eigenvector, T represents the control transfer matrix, and S represents the product purity eigenvector.

In the above-mentioned intelligent separation and purification system of electronic grade chlorine trifluoride, the control result generation module is further used: the classifier processes the control transfer matrix according to the following formula to generate a classification result, wherein: The formula is: softmax{(W _n ,B _n ):...:(W ₁ ,B ₁ )|Project(F)}, where Project(F) represents projecting the control transfer matrix into a vector, W ₁ to W _n is the weight matrix of the fully connected layer of each layer, and B ₁ to B _n represent the bias matrix of the fully connected layer of each layer.

Compared with the existing technology, the intelligent separation and purification system of electronic grade chlorine trifluoride provided by this application adopts artificial intelligence control technology to determine the purity value of the chlorine trifluoride product after first-level purification, the first alkali metal The first temperature of the adsorbent bed, the second temperature of the second alkali metal adsorbent bed, and the third temperature of the third alkali metal adsorbent bed are used as input data, and a deep neural network model is used as a feature extractor to synthesize Intelligent control of electronic grade chlorine trifluoride purification device. In this way, the effect of purification and separation can be accurately controlled in real time to optimize the purity, thereby improving the purification effect of the electronic grade chlorine trifluoride.

Description of drawings

The above and other objects, features and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. They are used to explain the present application together with the embodiments of the present application and do not constitute a limitation of the present application. In the drawings, like reference numbers generally represent like components or steps.

Figure 1 is an application scenario diagram of an intelligent separation and purification system for electronic grade chlorine trifluoride according to an embodiment of the present application.

Figure 2 is a block diagram of an intelligent separation and purification system for electronic grade chlorine trifluoride according to an embodiment of the present application.

Figure 3 is a block diagram of the temperature data global correlation encoding module in the intelligent separation and purification system of electronic grade chlorine trifluoride according to the embodiment of the present application.

Figure 4 is a flow chart of the intelligent separation and purification method of electronic grade chlorine trifluoride according to the embodiment of the present application.

Figure 5 is a schematic structural diagram of an intelligent separation and purification method for electronic grade chlorine trifluoride according to an embodiment of the present application.

Detailed ways

Hereinafter, example embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the example embodiments described here.

Scenario overview

As mentioned before, chlorine trifluoride is an extremely oxidizing chip etching cleaning agent. Currently, there are very few manufacturers in the world that can prepare electronic-grade chlorine trifluoride, and only domestic applicants have the ability to prepare electronic-grade chlorine trifluoride. This is because chlorine trifluoride easily associates with hydrogen fluoride to form multi-polymers with special intermolecular forces. Traditional separation methods cannot completely solve the problem of multi-polymer separation. This is the key to preparing electronic-grade chlorine trifluoride. technical challenge. Therefore, it is crucial to develop an intelligent separation and purification solution for electronic-grade chlorine trifluoride.

Currently, the technical principles and steps of the solution provided by patent CN114538381A are as follows: S1: By heating the alkali metal adsorbent in the 3-level metal adsorbent layer bed, the alkali metal adsorbent and hydrogen fluoride molecules are associated with each other to form a stronger Separate through hydrogen bonding to achieve primary purification; S2: further disperse hydrogen fluoride and chlorine trifluoride associated molecules through a 2-stage cryogenic distillation device to achieve secondary purification.

Wherein, the three-level metal adsorbent layer bed includes a first alkali metal adsorbent layer bed, a second alkali metal adsorbent layer bed, and a third alkali metal adsorbent layer bed that are connected in sequence, and the three-level metal adsorbent layer bed The layered bed is used to adsorb free hydrogen fluoride. Each alkali metal adsorbent bed includes a mixture of Al2O3+LiF. The reaction temperature of the 3-stage metal adsorbent layer bed is 150°C to 200°C. The height of each alkali metal adsorbent bed is 1.8 to 2.5 meters.

Wherein, the 2-stage low-temperature rectification device includes a low-boiling tower and a high-boiling tower that are connected in sequence, and the third alkali metal adsorbent layer bed is connected with the low-boiling tower. The 2-stage low-temperature rectification device includes There are extractants used to further disperse the hydrogen fluoride and chlorine trifluoride associated molecules. The low boiling tower includes a first reboiler, a first low boiling tower packing section, a second low boiling tower packing section and a first condenser from bottom to top. The high boiling tower includes, from bottom to top, a second reboiler, a first high boiling tower packing section, a second high boiling tower packing section, a third high boiling tower packing section and a second condenser. An extraction agent is provided in each packing section to further disperse the associated molecules of hydrogen fluoride and chlorine trifluoride. The temperature of the second tray at the upper end of the first reboiler is 10°C-12°C, and the temperature of the second tray at the lower end of the first condenser is -22.5°C-24°C; control the second reboiler. The temperature at the upper end of the boiler is 11°C-12°C, and the temperature at the lower end of the second condenser is -6°C--4°C.

Accordingly, the inventor of the present application found that in the existing electronic-grade chlorine trifluoride purification and separation device, the condition control of each reaction equipment is random or controlled according to predetermined conditions. This aspect will make the purification and separation effect inaccurate. Adjust for purity optimization. That is to say, there is a certain degree of randomness in the purification accuracy control of existing electronic-grade chlorine trifluoride purification and separation devices. On the other hand, since the electronic-grade chlorine trifluoride purification and separation device requires many parameters to be controlled, and there are complex linear and/or non-linear relationships between each parameter, it is difficult to purify electronic-grade chlorine trifluoride. There are high technical difficulties in controlling the device.

In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, speech signal processing and other fields. In addition, deep learning and neural networks have also shown that they are close to or even beyond human performance in areas such as image classification, object detection, semantic segmentation, and text translation.

The development of deep learning and neural networks provides new solutions and solutions for the control of electronic-grade chlorine trifluoride purification devices.

Specifically, in the technical solution of the present application, first, each sensor is used to obtain the purity value of the chlorine trifluoride product after primary purification at multiple predetermined time points within a predetermined time period, and the purity value of the first alkali metal adsorbent layer bed. a first temperature, a second temperature of the second alkali metal adsorbent bed, and a third temperature of the third alkali metal adsorbent bed. Then, consider the relationship between the first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed, and the third temperature of the third alkali metal adsorbent layer bed. has special correlation characteristics in time series. Therefore, in order to extract the implicit correlation between the three in time, the third layer of the first alkali metal adsorbent bed at multiple predetermined time points in the predetermined time period is further A temperature, a second temperature of the second alkali metal adsorbent layer bed, and a third temperature of the third alkali metal adsorbent layer bed are arranged as a temperature control matrix according to the time dimension and the sample dimension. And the temperature control matrix is passed through the first convolutional neural network as a feature extractor for feature extraction to extract the local high-dimensional implicit correlation feature information of each position in the temperature control matrix to obtain the temperature control local correlation. Feature map.

It should be understood that considering that convolution is a typical local operation, for the first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed and the third Regarding the third temperature of the three alkali metal adsorbent beds, the temperatures of the respective alkali metal adsorbent beds do not exist in isolation, and the correlation between the temperatures of the respective alkali metal adsorbent beds creates prospects. Target. Therefore, in the technical solution of the present application, in order to extract the first temperature of the first alkali metal adsorbent bed, the second temperature of the second alkali metal adsorbent bed and the third alkali metal adsorption Based on the correlation of the third temperature of the agent bed, a non-local neural network is used to further extract the features of the feature map. That is, the temperature control local correlation feature map is passed through a non-local neural network to obtain a temperature control global correlation feature map. In particular, here, the non-local neural network calculates the first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed and the third alkali metal adsorbent layer. The third temperature similarity of the agent bed captures hidden dependency information, and then models contextual features, allowing the network to focus on the overall content of the electrical power data, thereby improving the feature extraction capabilities of the backbone network in classification and detection tasks.

In this way, the feature information in the temperature control local correlation feature map and the temperature control global correlation feature map is fused to obtain a temperature control feature map. Furthermore, in order to reduce parameter data and thus reduce the amount of calculation, global mean pooling is performed on each feature matrix of the temperature control feature map to obtain the temperature control feature vector. This can prevent overfitting and improve subsequent classification. accuracy.

However, in the technical solution of the present application, the temperature control feature vector combines the temperature control local correlation features and the temperature control global correlation features in the spatial dimensions of each feature matrix of the temperature control feature map, and through the temperature control feature vector The control feature map is obtained by global mean pooling along the channel dimension, which makes the feature value of each position of the temperature control feature vector likely to produce correlation deviations in information fusion, so that forward propagation correlation guidance correction is preferably performed. ,Right now:

Where V represents the temperature control eigenvector, Σ is the autocovariance matrix of the temperature control eigenvector, that is, the value of each position of the matrix is the variance between the eigenvalues of each two positions of the vector V, μ and σ is the global mean and variance of the temperature control feature vector respectively, exp(·) represents the exponential operation of the vector, and the exponential operation with the vector as the power represents the natural exponential function value with the value of each position of the vector as the power,

and

Here, the forward propagation correlation guided correction is based on the characteristics of downsampling forward propagation of features based on global mean pooling along the channel dimension, and is effectively guided by learnable normal sampling offset engineering. Model the long-range dependence in the spatial dimension within the feature matrix and the channel dimension between feature matrices, and consider the local and non-local neighborhoods of the feature matrix to repair the correlation between each eigenvalue of the feature vector, thereby improving This improves the prediction ability of the temperature control feature vector for class probability, thereby improving the accuracy of classification.

It should be understood that for the purity value of the chlorine trifluoride product after primary purification at multiple predetermined time points within the predetermined time period, since the purity value of the chlorine trifluoride product after primary purification varies over time. has special implicit correlation features. Therefore, in order to more fully extract this correlation feature information, in the technical solution of this application, the first-level purification of multiple predetermined time points within the predetermined time period is further carried out. The purity value of the final chlorine trifluoride product is passed through a temporal encoder containing a one-dimensional convolutional layer to obtain the product purity feature vector. In one example, the temporal encoder consists of alternately arranged fully connected layers and one-dimensional convolutional layers, which extract the purity value of the first-level purified chlorine trifluoride product through one-dimensional convolutional coding at Correlation in the time series dimension and extraction of high-dimensional hidden features of the purity value of the chlorine trifluoride product after primary purification through fully connected coding.

Then, considering that the characteristic scales of the temperature data of the alkali metal adsorbent layer bed and the purity value data of the chlorine trifluoride product after primary purification are different, and the product purity characteristics can be used in high-dimensional space is regarded as a responsive feature for the temperature control feature. Therefore, in order to better integrate the feature information of the two for classification, the control of the corrected temperature control feature vector relative to the product purity feature vector is further calculated. transfer matrix. Furthermore, a classifier is used to perform classification processing on the control transfer matrix to obtain a classification result indicating whether the temperature control combination of the three-stage metal adsorbent bed within a predetermined time period meets the predetermined requirements.

Based on this, this application proposes an intelligent separation and purification system for electronic-grade chlorine trifluoride, which includes: a data acquisition module used to obtain first-level purified chlorine trifluoride at multiple predetermined time points within a predetermined time period. The purity value of the product, the first temperature of the first alkali metal adsorbent bed, the second temperature of the second alkali metal adsorbent bed, and the third temperature of the third alkali metal adsorbent bed; the temperature data structuring module, used to change the first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed and the third alkali metal adsorbent layer bed at multiple predetermined time points within the predetermined time period. The third temperature is arranged into a temperature control matrix according to the time dimension and the sample dimension; a temperature data local correlation encoding module is used to pass the temperature control matrix through the first convolutional neural network as a feature extractor to obtain a temperature control local correlation feature map ; Temperature data global correlation encoding module, used to pass the temperature control local correlation feature map through a non-local neural network to obtain the temperature control global correlation feature map; a fusion module, used to fuse the temperature control local correlation feature map and the temperature control local correlation feature map The temperature control globally correlates the feature map to obtain the temperature control feature map; the dimensionality reduction module is used to perform global mean pooling on each feature matrix of the temperature control feature map to obtain the temperature control feature vector; the correction module is used to perform the temperature control feature vector. The temperature control feature vector is corrected to obtain the corrected temperature control feature vector; the product purity data encoding module is used to pass the purity value of the first-level purified chlorine trifluoride product at multiple predetermined time points within the predetermined time period. A temporal encoder including a one-dimensional convolutional layer to obtain a product purity eigenvector; a responsiveness estimation module for calculating a control transfer matrix of the corrected temperature control eigenvector relative to the product purity eigenvector; and, control results A generation module is used to pass the control transfer matrix through a classifier to obtain a classification result. The classification result is used to indicate whether the temperature control combination of the three-stage metal adsorbent layer bed within a predetermined time period meets the predetermined requirements.

Figure 1 illustrates the application scenario diagram of the intelligent separation and purification system of electronic grade chlorine trifluoride according to the embodiment of the present application. As shown in Figure 1, in this application scenario, first, through each sensor (for example, the purity detector T1 and the temperature sensor T2 illustrated in Figure 1), the first-level data of multiple predetermined time points within a predetermined time period are obtained. The purity value of the purified chlorine trifluoride product, the first temperature of the first alkali metal adsorbent layer bed (for example, M1 as shown in Figure 1), the second alkali metal adsorbent layer bed (for example, as shown in Figure 1 The second temperature of the third alkali metal adsorbent layer bed (for example, M3 as shown in Figure 1). Then, the obtained purity values of the chlorine trifluoride product after primary purification at multiple predetermined time points within the predetermined time period, the first to third values of the first to third alkali metal adsorbent beds The temperature is input into a server deployed with an intelligent separation and purification algorithm of electronic grade chlorine trifluoride (for example, the cloud server S as shown in Figure 1), wherein the server is capable of intelligent separation and purification of electronic grade chlorine trifluoride. The purification algorithm is performed on the purity values of the chlorine trifluoride product after primary purification at multiple predetermined time points within the predetermined time period and the first to third temperatures of the first to third alkali metal adsorbent beds. Processing to generate a classification result indicating whether the temperature control combination of the 3-stage metal adsorbent layer bed meets the predetermined requirements within a predetermined time period.

After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be specifically introduced below with reference to the accompanying drawings.

Example system

Figure 2 illustrates a block diagram of an intelligent separation and purification system for electronic grade chlorine trifluoride according to an embodiment of the present application. As shown in Figure 2, the intelligent separation and purification system 200 of electronic grade chlorine trifluoride according to the embodiment of the present application includes: a data acquisition module 210, used to obtain the first-level purified data at multiple predetermined time points within a predetermined time period. The purity value of the chlorine trifluoride product, the first temperature of the first alkali metal adsorbent bed, the second temperature of the second alkali metal adsorbent bed, and the third temperature of the third alkali metal adsorbent bed; temperature data The structured module 220 is used to combine the first temperature of the first alkali metal adsorbent bed, the second temperature of the second alkali metal adsorbent bed and the third alkali metal at multiple predetermined time points within the predetermined time period. The third temperature of the adsorbent bed is arranged into a temperature control matrix according to the time dimension and the sample dimension; the temperature data local correlation encoding module 230 is used to pass the temperature control matrix through the first convolutional neural network as a feature extractor to obtain Temperature control local correlation feature map; temperature data global correlation encoding module 240, used to pass the temperature control local correlation feature map through a non-local neural network to obtain a temperature control global correlation feature map; fusion module 250, used to fuse the temperature Control the local correlation feature map and the temperature control global correlation feature map to obtain the temperature control feature map; the dimensionality reduction module 260 is used to perform global mean pooling on each feature matrix of the temperature control feature map to obtain the temperature control feature vector. ; Correction module 270, used to correct the temperature control feature vector to obtain the corrected temperature control feature vector; Product purity data encoding module 280, used to convert the first-level data at multiple predetermined time points within the predetermined time period; The purity value of the purified chlorine trifluoride product is passed through a temporal encoder including a one-dimensional convolution layer to obtain a product purity feature vector; the responsiveness estimation module 290 is used to calculate the corrected temperature control feature vector relative to the product The control transfer matrix of the purity feature vector; and, the control result generation module 300 is used to pass the control transfer matrix through a classifier to obtain a classification result, the classification result is used to represent the 3-level metal adsorbent layer bed within a predetermined time period Whether the temperature control combination meets the predetermined requirements.

Specifically, in the embodiment of the present application, the data acquisition module 210, the temperature data structuring module 220 and the temperature data local correlation encoding module 230 are used to obtain the experience of multiple predetermined time points within a predetermined time period. The purity value of the chlorine trifluoride product after primary purification, the first temperature of the first alkali metal adsorbent bed, the second temperature of the second alkali metal adsorbent bed, and the third temperature of the third alkali metal adsorbent bed. temperature, and combine the first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed and the third alkali metal adsorbent layer bed at multiple predetermined time points within the predetermined time period. The third temperature is arranged into a temperature control matrix according to the time dimension and the sample dimension, and then the temperature control matrix is passed through the first convolutional neural network as a feature extractor to obtain the temperature control local correlation feature map. As mentioned before, in the existing electronic-grade chlorine trifluoride purification and separation device, the condition control of each reaction equipment is random or controlled according to predetermined conditions. This aspect will make it impossible to accurately control the purification and separation effect. Perform purity optimization. That is to say, there is a certain degree of randomness in the purification accuracy control of existing electronic-grade chlorine trifluoride purification and separation devices. On the other hand, since the purification and separation device of electronic grade chlorine trifluoride requires many parameters to be controlled, and there are complex linear and/or nonlinear relationships between each parameter, it is expected that the purification and separation device of electronic grade chlorine trifluoride will be The purification device is intelligently controlled.

That is, specifically, in the technical solution of the present application, first, each sensor is used to obtain the purity values of the chlorine trifluoride product after primary purification at multiple predetermined time points within a predetermined time period, the first alkali metal adsorbent The first temperature of the layer bed, the second temperature of the second alkali metal adsorbent layer bed, and the third temperature of the third alkali metal adsorbent layer bed. Then, it should be understood that considering the first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed and the third temperature of the third alkali metal adsorbent layer bed, The three temperatures have special correlation characteristics in time series. Therefore, in order to extract the implicit correlation between the three temperatures in time, the first alkali metal adsorbent at multiple predetermined time points within the predetermined time period is further The first temperature of the layer bed, the second temperature of the second alkali metal adsorbent layer bed, and the third temperature of the third alkali metal adsorbent layer bed are arranged as a temperature control matrix according to the time dimension and the sample dimension. And the temperature control matrix is passed through the first convolutional neural network as a feature extractor for feature extraction to extract the local high-dimensional implicit correlation feature information of each position in the temperature control matrix to obtain the temperature control local correlation. Feature map.

More specifically, in the embodiment of the present application, the temperature data structuring module includes: a row vector construction unit for converting the first alkali metal adsorbent layer bed at multiple predetermined time points within the predetermined time period. The first temperature, the second temperature of the second alkali metal adsorbent layer bed, and the third temperature of the third alkali metal adsorbent layer bed are respectively arranged as row vectors according to the time dimension to obtain a plurality of row vectors; the matrix construction unit, For arranging the plurality of row vectors into the temperature control matrix according to the sample dimension.

More specifically, in the embodiment of the present application, the temperature data local correlation encoding module is further configured to: use each layer of the first convolutional neural network as a feature extractor to encode the input in the forward pass of the layer. The data are processed separately: performing convolution processing on the input data to obtain a convolution feature map; performing mean pooling processing on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map. To obtain the activation feature map; wherein, the output of the last layer of the first convolutional neural network as a feature extractor is the temperature control local correlation feature map, and the first convolutional neural network as a feature extractor The input of the first layer is the temperature control matrix.

Specifically, in this embodiment of the present application, the temperature data global correlation encoding module 240 is used to pass the temperature control local correlation feature map through a non-local neural network to obtain a temperature control global correlation feature map. It should be understood that considering that convolution is a typical local operation, for the first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed and the third Regarding the third temperature of the three alkali metal adsorbent beds, the temperatures of the respective alkali metal adsorbent beds do not exist in isolation, and the correlation between the temperatures of the respective alkali metal adsorbent beds creates prospects. Target. Therefore, in the technical solution of the present application, in order to extract the first temperature of the first alkali metal adsorbent bed, the second temperature of the second alkali metal adsorbent bed and the third alkali metal adsorption Based on the correlation of the third temperature of the agent bed, a non-local neural network is used to further extract the features of the feature map. That is, the temperature control local correlation feature map is passed through a non-local neural network to obtain a temperature control global correlation feature map. In particular, here, the non-local neural network calculates the first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed and the third alkali metal adsorbent layer. The third temperature similarity of the agent bed captures hidden dependency information, and then models contextual features, allowing the network to focus on the overall content of the electrical power data, thereby improving the feature extraction capabilities of the backbone network in classification and detection tasks.

More specifically, in the embodiment of the present application, the temperature data global correlation encoding module includes: first, inputting the temperature control local correlation feature map into the first point convolution layer and the third point convolution layer of the non-local neural network respectively. A two-point convolution layer and a third point convolution layer are used to obtain the first feature map, the second feature map and the third feature map. Then, a position-weighted sum of the first feature map and the second feature map is calculated to obtain an intermediate fused feature map. Next, the intermediate fusion feature map is input into the Softmax function to normalize the feature values of each position in the intermediate fusion feature map to obtain a normalized intermediate fusion feature map. Then, a position-weighted sum of the normalized intermediate fused feature map and the third feature map is calculated to obtain a re-fused feature map. Then, the re-fused feature map is embedded in a Gaussian similarity function to calculate the similarity between the feature values of each position in the re-fused feature map to obtain a global perceptual feature map. Then, the global perceptual feature map is passed through the fourth point convolution layer of the non-local neural network to obtain a channel-adjusted global perceptual feature map. Finally, the position-weighted sum of the channel adjustment global perception feature map and the temperature control local correlation feature map is calculated to obtain the temperature control global correlation feature map.

Figure 3 illustrates a block diagram of the temperature data global correlation encoding module in the intelligent separation and purification system of electronic grade chlorine trifluoride according to the embodiment of the present application. As shown in Figure 3, the temperature data global correlation encoding module 240 includes: a point convolution unit 241, which is used to input the temperature control local correlation feature map into the first point convolution layer of the non-local neural network. , the second point convolution layer and the third point convolution layer to obtain the first feature map, the second feature map and the third feature map; the first fusion unit 242 is used to calculate the first feature map and the third feature map. The position-weighted sum of the two feature maps is used to obtain the intermediate fusion feature map; the normalization unit 243 is used to input the intermediate fusion feature map into the Softmax function to normalize the feature values of each position in the intermediate fusion feature map. to obtain a normalized intermediate fusion feature map; the second fusion unit 244 is used to calculate the position-weighted sum of the normalized intermediate fusion feature map and the third feature map to obtain a re-fusion feature map; global perception Unit 245 is used to embed the re-fused feature map into a Gaussian similarity function to calculate the similarity between the feature values of each position in the re-fused feature map to obtain a global perceptual feature map; the channel number adjustment unit 246 is used Passing the global perceptual feature map through the fourth point convolution layer of the non-local neural network to obtain a channel-adjusted global perceptual feature map; and a third fusion unit 247 for calculating the channel-adjusted global perceptual feature map and the position-weighted sum of the temperature control local correlation feature map to obtain the temperature control global correlation feature map.

Specifically, in this embodiment of the present application, the fusion module 250 and the dimensionality reduction module 260 are used to fuse the temperature control local correlation feature map and the temperature control global correlation feature map to obtain a temperature control feature map, And perform global mean pooling on each feature matrix of the temperature control feature map to obtain a temperature control feature vector. That is, in the technical solution of the present application, the feature information in the temperature control local correlation feature map and the temperature control global correlation feature map are further fused to obtain a temperature control feature map. Then, in order to reduce the parameter data and thus the amount of calculation, global mean pooling is performed on each feature matrix of the temperature control feature map to obtain the temperature control feature vector. This can prevent overfitting and improve the accuracy of subsequent classification. accuracy.

More specifically, in the embodiment of the present application, the fusion module is further configured to: fuse the temperature control local correlation feature map and the temperature control global correlation feature map using the following formula to obtain the temperature control feature map; Among them, the formula is:

F _s =αF ₁ +βF ₂

Specifically, in this embodiment of the present application, the correction module 270 is used to correct the temperature control feature vector to obtain a corrected temperature control feature vector. It should be understood that in the technical solution of the present application, the temperature control feature vector combines the temperature control local correlation features and the temperature control global correlation features in the spatial dimensions of each feature matrix of the temperature control feature map, and through the The temperature control feature map is obtained by global mean pooling along the channel dimension, which makes the feature value of each position of the temperature control feature vector likely to produce correlation deviations in information fusion, so that forward propagation is preferably performed. Relevance guided corrections. In this way, the forward propagation correlation guided correction is based on the characteristics of downsampling forward propagation of features based on global mean pooling along the channel dimension, and is effectively guided by learnable normal sampling offset engineering. Model the long-range dependence in the spatial dimension within the feature matrix and the channel dimension between feature matrices, and consider the local and non-local neighborhoods of the feature matrix to repair the correlation between each eigenvalue of the feature vector, thereby improving This improves the prediction ability of the temperature control feature vector for class probability, thereby improving the accuracy of classification.

More specifically, in the embodiment of the present application, the correction module is further configured to: correct the temperature control feature vector with the following formula to obtain the corrected temperature control feature vector; wherein the formula is:

and

Specifically, in the embodiment of the present application, the product purity data encoding module 280 is used to encode the purity values of the chlorine trifluoride product after primary purification at multiple predetermined time points within the predetermined time period by including a Dimensional convolutional layer temporal encoder to obtain the product purity feature vector. It should be understood that for the purity value of the chlorine trifluoride product after primary purification at multiple predetermined time points within the predetermined time period, since the purity value of the chlorine trifluoride product after primary purification varies over time. has special implicit correlation features. Therefore, in order to more fully extract this correlation feature information, in the technical solution of this application, the first-level purification of multiple predetermined time points within the predetermined time period is further carried out. The purity value of the final chlorine trifluoride product is passed through a temporal encoder containing a one-dimensional convolutional layer to obtain the product purity feature vector. Correspondingly, in a specific example, the temporal encoder consists of alternately arranged fully connected layers and one-dimensional convolutional layers, which extract the first-level purified chlorine trifluoride product through one-dimensional convolutional encoding. Correlation of the purity value in the time series dimension and extraction of high-dimensional hidden features of the purity value of the chlorine trifluoride product after primary purification through fully connected coding.

More specifically, in the embodiment of the present application, the product purity data encoding module is further used to: calculate the purity value of the chlorine trifluoride product after primary purification at multiple predetermined time points within the predetermined time period according to The time dimension is arranged as a one-dimensional input vector; the fully connected layer of the temporal encoder is used to fully connect the input vector with the following formula to extract the high-dimensional implicit features of the eigenvalues of each position in the input vector, Among them, the formula is:

Specifically, in this embodiment of the present application, the responsiveness estimation module 290 and the control result generation module 300 are used to calculate the control transfer matrix of the corrected temperature control feature vector relative to the product purity feature vector, And the control transfer matrix is passed through a classifier to obtain a classification result, which is used to indicate whether the temperature control combination of the three-stage metal adsorbent layer bed within a predetermined time period meets the predetermined requirements. It should be understood that considering that the characteristic scales of the temperature data of the alkali metal adsorbent layer bed and the purity value data of the chlorine trifluoride product after primary purification are different, and the product purity characteristics are in a high-dimensional space can be regarded as a responsive feature for the temperature control feature. Therefore, in order to better integrate the feature information of the two for classification, in the technical solution of this application, the corrected temperature control feature vector is further calculated. Control transfer matrix relative to the product purity eigenvector. Furthermore, a classifier is used to perform classification processing on the control transfer matrix to obtain a classification result indicating whether the temperature control combination of the three-stage metal adsorbent bed within a predetermined time period meets the predetermined requirements. Correspondingly, in a specific example, the classifier processes the control transfer matrix with the following formula to generate a classification result, where the formula is: softmax{(W _n ,B _n ):...:(W ₁ ,B ₁ )|Project(F)}, where Project(F) represents projecting the control transfer matrix into a vector, W ₁ to W _n are the weight matrices of the fully connected layers of each layer, and B ₁ to B _n represent each The bias matrix of the fully connected layer.

More specifically, in the embodiment of the present application, the responsiveness estimation module is further configured to: calculate the control transfer matrix of the corrected temperature control feature vector relative to the product purity feature vector using the following formula; Among them, the formula is:

S＝T*F

In summary, the intelligent separation and purification system 200 of electronic grade chlorine trifluoride based on the embodiment of the present application is clarified, which adopts artificial intelligence control technology to determine the purity value of the chlorine trifluoride product after the first-level purification, the third The first temperature of the first alkali metal adsorbent bed, the second temperature of the second alkali metal adsorbent bed, and the third temperature of the third alkali metal adsorbent bed are used as input data, and a deep neural network model is used as the feature extractor. , to comprehensively implement intelligent control of the electronic grade chlorine trifluoride purification device. In this way, the effect of purification and separation can be accurately controlled in real time to optimize the purity, thereby improving the purification effect of the electronic grade chlorine trifluoride.

As mentioned above, the intelligent separation and purification system 200 for electronic-grade chlorine trifluoride according to the embodiment of the present application can be implemented in various terminal devices, such as servers for intelligent separation and purification algorithms for electronic-grade chlorine trifluoride. In one example, the intelligent separation and purification system 200 for electronic grade chlorine trifluoride according to an embodiment of the present application can be integrated into a terminal device as a software module and/or a hardware module. For example, the electronic-grade chlorine trifluoride intelligent separation and purification system 200 can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the electronic-grade chlorine trifluoride can be a software module in the operating system of the terminal device. The intelligent separation and purification system 200 for chlorine fluoride can also be one of the many hardware modules of the terminal equipment.

Alternatively, in another example, the intelligent separation and purification system 200 for electronic grade chlorine trifluoride and the terminal device can also be separate devices, and the intelligent separation and purification system 200 for electronic grade chlorine trifluoride can be connected via a wired And/or a wireless network is connected to the terminal device, and the interactive information is transmitted according to the agreed data format.

Example methods

Figure 4 illustrates a flow chart of an intelligent separation and purification method of electronic grade chlorine trifluoride. As shown in Figure 4, the intelligent separation and purification method of electronic grade chlorine trifluoride according to the embodiment of the present application includes the step: S110, obtaining the first-level purified chlorine trifluoride product at multiple predetermined time points within a predetermined time period. purity value, the first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed, and the third temperature of the third alkali metal adsorbent layer bed; S120, change the predetermined time The first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed and the third temperature of the third alkali metal adsorbent layer bed at multiple predetermined time points in the segment are summed according to the time dimension. The sample dimensions are arranged into a temperature control matrix; S130, pass the temperature control matrix through the first convolutional neural network as a feature extractor to obtain a temperature control local correlation feature map; S140, pass the temperature control local correlation feature map through a non- Local neural network to obtain the temperature control global correlation feature map; S150, fuse the temperature control local correlation feature map and the temperature control global correlation feature map to obtain the temperature control feature map; S160, perform each of the temperature control feature maps The feature matrix performs global mean pooling to obtain the temperature control feature vector; S170, correct the temperature control feature vector to obtain the corrected temperature control feature vector; S180, combine the temperature control feature vectors at multiple predetermined time points within the predetermined time period. The purity value of the chlorine trifluoride product after primary purification is passed through a temporal encoder including a one-dimensional convolution layer to obtain a product purity feature vector; S190, calculate the corrected temperature control feature vector relative to the product purity feature vector. Control transfer matrix; and, S200, pass the control transfer matrix through a classifier to obtain a classification result, which is used to indicate whether the temperature control combination of the 3-stage metal adsorbent layer bed within a predetermined time period meets the predetermined requirements.

Figure 5 illustrates a schematic structural diagram of an intelligent separation and purification method for electronic grade chlorine trifluoride according to an embodiment of the present application. As shown in Figure 5, in the network architecture of the intelligent separation and purification method of electronic grade chlorine trifluoride, first, the first alkali metal adsorbent layer bed obtained at multiple predetermined time points within the predetermined time period is The first temperature of , the second temperature of the second alkali metal adsorbent layer bed, and the third temperature of the third alkali metal adsorbent layer bed (for example, P1 as shown in Figure 5) are arranged according to the time dimension and the sample dimension as Temperature control matrix (for example, M as shown in Figure 5); then, the temperature control matrix is passed through the first convolutional neural network as a feature extractor (for example, CNN1 as shown in Figure 5) to obtain Temperature control local correlation feature map (for example, F1 as shown in Figure 5); then, the temperature control local correlation feature map is passed through a non-local neural network (for example, CNN2 as shown in Figure 5) to obtain the temperature Control the global correlation feature map (for example, F2 as shown in Figure 5); then, fuse the temperature control local correlation feature map and the temperature control global correlation feature map to obtain a temperature control feature map (for example, as shown in Figure 5 F as shown in Figure 5); then, global mean pooling is performed on each feature matrix of the temperature control feature map to obtain a temperature control feature vector (for example, VF1 as shown in Figure 5); then, the temperature The control feature vector is corrected to obtain a corrected temperature control feature vector (for example, VF2 as shown in Figure 5); then, the first-level purified chlorine trifluoride at multiple predetermined time points within the predetermined time period is The purity value of the product (e.g., P2 as illustrated in Figure 5) is passed through a temporal encoder (e.g., E as illustrated in Figure 5) including a one-dimensional convolutional layer to obtain a product purity feature vector (e.g., as shown in Figure VF shown in Figure 5); then, calculate the control transfer matrix (for example, MF shown in Figure 5) of the corrected temperature control feature vector relative to the product purity feature vector; and, finally, convert the The control transfer matrix is passed through a classifier (for example, the classifier as shown in Figure 5) to obtain a classification result, which is used to indicate whether the temperature control combination of the 3-stage metal adsorbent bed within a predetermined time period meets the predetermined Require.

In summary, the intelligent separation and purification method of electronic grade chlorine trifluoride based on the embodiments of the present application has been clarified, which uses artificial intelligence-based control technology to determine the purity value of the chlorine trifluoride product after primary purification, the first The first temperature of the alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed, and the third temperature of the third alkali metal adsorbent layer bed are used as input data, and a deep neural network model is used as a feature extractor, To comprehensively control the electronic grade chlorine trifluoride purification device intelligently. In this way, the effect of purification and separation can be accurately controlled in real time to optimize the purity, thereby improving the purification effect of the electronic grade chlorine trifluoride.

The basic principles of the present application have been described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, advantages, effects, etc. mentioned in this application are only examples and not limitations. These advantages, advantages, effects, etc. cannot be considered as Each embodiment of this application must have. In addition, the specific details disclosed above are only for the purpose of illustration and to facilitate understanding, and are not limiting. The above details do not limit the application to be implemented using the above specific details.

The block diagrams of the devices, devices, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, devices, equipment, and systems may be connected, arranged, and configured in any manner. Words such as "includes," "includes," "having," etc. are open-ended terms that mean "including, but not limited to," and may be used interchangeably therewith. As used herein, the words "or" and "and" refer to the words "and/or" and are used interchangeably therewith unless the context clearly dictates otherwise. As used herein, the word "such as" refers to the phrase "such as, but not limited to," and may be used interchangeably therewith.

It should also be pointed out that in the device, equipment and method of the present application, each component or each step can be decomposed and/or recombined. These decompositions and/or recombinations shall be considered equivalent versions of this application.

The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, this application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for the purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the form disclosed herein. Although various example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

An intelligent separation and purification system for electronic-grade chlorine trifluoride, characterized by including: a data acquisition module, used to obtain the purity value of the chlorine trifluoride product after primary purification at multiple predetermined time points within a predetermined time period , the first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed, and the third temperature of the third alkali metal adsorbent layer bed; the temperature data structuring module is used to convert all the The first temperature of the first alkali metal adsorbent layer bed, the second temperature of the second alkali metal adsorbent layer bed and the third temperature of the third alkali metal adsorbent layer bed at multiple predetermined time points within the predetermined time period are as follows: The time dimension and the sample dimension are arranged into a temperature control matrix; a temperature data local correlation encoding module is used to pass the temperature control matrix through the first convolutional neural network as a feature extractor to obtain a temperature control local correlation feature map; the temperature data global A correlation coding module, used to pass the temperature control local correlation feature map through a non-local neural network to obtain a temperature control global correlation feature map; a fusion module, used to fuse the temperature control local correlation feature map and the temperature control global correlation feature map to obtain a temperature control feature map; a dimensionality reduction module for performing global mean pooling on each feature matrix of the temperature control feature map to obtain a temperature control feature vector; a correction module for performing a global mean pooling on each feature matrix of the temperature control feature map to obtain a temperature control feature vector; Calibration is performed to obtain the corrected temperature control feature vector; a product purity data encoding module is used to pass the purity value of the first-level purified chlorine trifluoride product at multiple predetermined time points within the predetermined time period through a one-dimensional volume. Stacked time series encoders to obtain product purity eigenvectors; a responsiveness estimation module for calculating a control transfer matrix of the corrected temperature control eigenvector relative to the product purity eigenvector; and a control result generation module for The control transfer matrix is passed through a classifier to obtain a classification result, which is used to indicate whether the temperature control combination of the three-stage metal adsorbent layer bed within a predetermined time period meets the predetermined requirements.
The intelligent separation and purification system of electronic grade chlorine trifluoride according to claim 1, characterized in that the temperature data structuring module includes: a row vector construction unit for converting multiple predetermined times within the predetermined time period. The first temperature of the first alkali metal adsorbent bed, the second temperature of the second alkali metal adsorbent bed, and the third temperature of the third alkali metal adsorbent bed at the time point are respectively arranged in rows according to the time dimension. vector to obtain multiple row vectors; a matrix construction unit configured to arrange the multiple row vectors into the temperature control matrix according to the sample dimensions.
The intelligent separation and purification system of electronic grade chlorine trifluoride according to claim 2, characterized in that the temperature data local correlation encoding module is further used to: use the first convolutional neural network as a feature extractor Each layer of each performs a convolution process on the input data in the forward pass of the layer: perform convolution processing on the input data to obtain a convolution feature map; perform mean pooling processing on the convolution feature map to obtain a pooled feature map; and Nonlinear activation is performed on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the first convolutional neural network as a feature extractor is the temperature control local correlation feature map, and the The input of the first layer of the first convolutional neural network as a feature extractor is the temperature control matrix.
The intelligent separation and purification system of electronic grade chlorine trifluoride according to claim 3, characterized in that the temperature data global correlation encoding module includes: a point convolution unit for converting the temperature control local correlation feature map The first point convolution layer, the second point convolution layer and the third point convolution layer of the non-local neural network are respectively input to obtain the first feature map, the second feature map and the third feature map; the first fusion unit , used to calculate the position-weighted sum of the first feature map and the second feature map to obtain an intermediate fusion feature map; a normalization unit, used to input the intermediate fusion feature map into the Softmax function to calculate the The feature values of each position in the intermediate fusion feature map are normalized to obtain the normalized intermediate fusion feature map; the second fusion unit is used to calculate the pressure of the normalized intermediate fusion feature map and the third feature map. The position weighted sum is used to obtain the re-fused feature map; the global perception unit is used to embed the re-fused feature map into a Gaussian similarity function to calculate the similarity between the feature values of each position in the re-fused feature map to obtain the global a perceptual feature map; a channel number adjustment unit for passing the global perceptual feature map through the fourth point convolution layer of the non-local neural network to obtain a channel-adjusted global perceptual feature map; and a third fusion unit for calculating The channel adjusts the position-weighted sum of the global perceptual feature map and the temperature control local correlation feature map to obtain the temperature control global correlation feature map.
The intelligent separation and purification system of electronic grade chlorine trifluoride according to claim 4, characterized in that the fusion module is further used to: fuse the temperature control local correlation feature map and the temperature control global according to the following formula Correlate the characteristic map to obtain the temperature control characteristic map; wherein, the formula is:

F s =αF 1 +βF 2

Among them, F s is the temperature control feature map, F 1 is the temperature control local correlation feature map, F 2 is the temperature control global correlation feature map, "+" means the temperature control local correlation feature map and all the temperature control local correlation feature maps. The elements at corresponding positions in the temperature control global correlation feature map are added together, α and β are used to control the relationship between the temperature control local correlation feature map and the temperature control global correlation feature map in the temperature control feature map. Balanced weighting parameters.
The intelligent separation and purification system of electronic grade chlorine trifluoride according to claim 5, characterized in that the correction module is further used to: correct the temperature control feature vector with the following formula to obtain the corrected Temperature control feature vector; where, the formula is:

Where V represents the temperature control eigenvector, Σ is the autocovariance matrix of the temperature control eigenvector, μ and σ are the global mean and variance of the temperature control eigenvector respectively, and exp(·) represents the exponential operation of the vector. , the exponential operation raised to the power of a vector represents the value of the natural exponential function raised to the power of the value of each position of the vector,
and
Represents position-wise subtraction and addition of feature vectors respectively,
represents matrix multiplication, ||·|| 2 represents the second norm of the eigenvector.
The intelligent separation and purification system of electronic grade chlorine trifluoride according to claim 6, characterized in that the product purity data encoding module is further used to: combine the data at multiple predetermined time points within the predetermined time period. The purity value of the chlorine trifluoride product after stage purification is arranged into a one-dimensional input vector according to the time dimension; the fully connected layer of the temporal encoder is used to perform fully connected encoding on the input vector with the following formula to extract the input The high-dimensional implicit features of the eigenvalues at each position in the vector, where the formula is:

where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector,
represents matrix multiplication; and uses the one-dimensional convolution layer of the temporal encoder to perform one-dimensional convolution encoding on the input vector with the following formula to extract high-dimensional implicit correlation features between the feature values of each position in the input vector , where the formula is:

Among them, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix that operates with the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector.
The intelligent separation and purification system of electronic grade chlorine trifluoride according to claim 7, characterized in that the responsiveness estimation module is further used to: calculate the corrected temperature control characteristic vector relative to the The control transfer matrix of the product purity characteristic vector; wherein, the formula is:

S＝T*F

Where F represents the corrected temperature control eigenvector, T represents the control transfer matrix, and S represents the product purity eigenvector.
The intelligent separation and purification system of electronic grade chlorine trifluoride according to claim 8, characterized in that the control result generation module is further used: the classifier processes the control transfer matrix according to the following formula to Generate classification results, where the formula is: softmax{(W n ,B n ):...:(W 1 ,B 1 )|Project(F)}, where Project(F) represents the projection of the control transfer matrix is a vector, W 1 to W n are the weight matrices of the fully connected layers of each layer, and B 1 to B n represent the bias matrices of the fully connected layers of each layer.