WO2024098604A1 - Rectification control system and method for preparing electronic-grade carbon tetrafluoride - Google Patents

Abstract

Disclosed in the present application are a rectification control system and method for preparing electronic-grade carbon tetrafluoride. Deep-learning-based artificial intelligence control technology is used to perform feature extraction on collaborative associated features of temperatures and pressures of different areas of a refining section and a multi-scale change feature of the flow velocity of a flow medium, transfer vectors of the features are further used to represent relevance feature information between the collaborative associated features of temperatures and pressures and a dynamic change feature of the flow velocity of the flow medium, and self-adaptive real-time control over the opening degree of a valve of the flow medium is performed on the basis of the relevance feature information; and spatial topological features of the different areas of the refining section are introduced during this process, so as to further enhance the extraction of the collaborative associated features of temperatures and pressures at spatial positions, thereby improving the precision for controlling the opening degree of the valve of the flow medium. In this way, the rectification efficiency can be improved, and the consumption of the cooling capacity can be reduced.

Description

Distillation control system and method for preparing electronic grade carbon tetrafluoride Technical Field

The present application relates to the field of intelligent control technology, and more specifically, to a distillation control system and method for preparing electronic grade carbon tetrafluoride.

Background technique

Carbon tetrafluoride (CF4) is the most widely used plasma etching gas in the microelectronics industry. It is widely used in etching thin film materials such as silicon, silicon dioxide, silicon nitride, phosphorus silicon glass and tungsten. It is also widely used in electronic device surface cleaning, solar cell production, laser technology, low temperature refrigeration, gas insulation, leakage detection agent, control of space rocket attitude, detergent, lubricant and brake fluid in printed circuit production. Due to its extremely strong chemical stability, CF4 can also be used in metal smelting and plastics industries.

technical problem

In recent years, due to the development of the electronics industry, the domestic market demand for high-purity carbon tetrafluoride has continued to grow, and domestic companies have also established production and purification equipment, but there are certain gaps in process stability and product purity. Therefore, it is of great significance to improve the stable operability of carbon tetrafluoride distillation and purification. In addition, given the characteristics of carbon tetrafluoride, the use of low-temperature distillation consumes a lot of cold energy. How to improve the equipment and reduce cold energy consumption has also become a key issue to be optimized.

Therefore, an optimized distillation control system for the preparation of electronic grade carbon tetrafluoride is desired.

Technical Solutions

In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a distillation control system and method for the preparation of electronic grade carbon tetrafluoride. It adopts artificial intelligence control technology based on deep learning to extract features of the associated synergistic characteristics of temperature and pressure in different regions of the refining section and the multi-scale change characteristics of the flow rate of the flow medium, and further uses the transfer vector of the two to represent the correlation feature information between the synergistic correlation characteristics of temperature and pressure and the dynamic change characteristics of the flow rate of the flow medium, and thereby performs adaptive real-time control of the valve opening of the flow medium, and in this process, the spatial topological characteristics of different regions of the refining section are introduced to further strengthen the extraction of the synergistic correlation characteristics of the temperature and pressure in the spatial position, so as to improve the control accuracy of the valve opening of the flow medium. In this way, the efficiency of distillation can be improved and the consumption of cold can be reduced.

According to one aspect of the present application, a distillation control system for preparing electronic grade carbon tetrafluoride is provided, which includes: a distillation parameter acquisition unit, which is used to obtain the temperature values and pressure values of multiple regions of the refining section at multiple predetermined time points within a predetermined time period, as well as the flow rate values of the flow medium at the multiple predetermined time points, collected by a pressure sensor and a temperature sensor; a temperature and pressure coordination unit, which is used to arrange the temperature values and pressure values of each region of the refining section at multiple predetermined time points within a predetermined time period as a temperature input vector and a pressure input vector according to the time dimension, and then calculate the product between the transpose of the temperature input vector and the pressure input vector to obtain multiple coordinated feature matrices; a temperature-pressure coordinated feature extraction unit, which is used to pass the multiple coordinated feature matrices through a first convolutional neural network model as a filter to obtain multiple coordinated feature vectors; a matrixing unit, which is used to perform two-dimensional matrixing on the multiple coordinated feature vectors to obtain a coordinated feature matrix; and a spatial topology construction unit, which is used to construct the topology of the multiple regions. A matrix, wherein the value of each position on the non-diagonal position of the topological matrix is the distance between the corresponding two areas, and the value of each position on the diagonal position of the topological matrix is zero; a topological feature extraction unit, used to pass the topological matrix through a second convolutional neural network model as a feature extractor to obtain a topological feature matrix; a graph neural network unit, used to pass the collaborative feature matrix and the topological feature matrix through a graph neural network model to obtain a topological collaborative feature matrix; a flow rate feature extraction unit, used to arrange the flow rate values of the flow medium at the multiple predetermined time points according to the time dimension as a flow medium input vector through a multi-scale neighborhood feature extraction module to obtain a flow rate feature vector; a responsiveness unit, used to calculate the transfer vector of the flow rate feature vector relative to the topological collaborative feature matrix as a classification feature vector; and a distillation control result generation unit, used to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the valve opening value for adjusting the flow medium at the current time point should be increased or decreased.

In the above-mentioned distillation control system for the preparation of electronic grade carbon tetrafluoride, the temperature-pressure collaborative feature extraction unit is further used to: use each layer of the first convolutional neural network model as a filter to perform convolution processing, feature matrix-based mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer so that the last layer of the first convolutional neural network model as a filter outputs the multiple collaborative feature vectors, wherein the input of the first layer of the first convolutional neural network model as a filter is the multiple collaborative feature matrices.

In the above-mentioned distillation control system for the preparation of electronic grade carbon tetrafluoride, the topological feature extraction unit is further used to: use each layer of the second convolutional neural network model serving as a feature extractor to perform two-dimensional convolution processing, feature matrix-based mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer so that the last layer of the second convolutional neural network model serving as a feature extractor outputs the topological feature matrix, wherein the input of the first layer of the second convolutional neural network model serving as a feature extractor is the topological matrix.

In the above-mentioned distillation control system for the preparation of electronic grade carbon tetrafluoride, the graph neural network unit is further used to use the graph neural network model to process the synergistic feature matrix and the topological feature matrix with learnable neural network parameters to obtain the topological synergistic feature matrix containing irregular spatial topological features and temperature-pressure synergistic features.

In the above-mentioned distillation control system for preparing electronic grade carbon tetrafluoride, the flow rate feature extraction unit includes: a first scale feature extraction unit, used to input the flow medium input vector into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first scale flow rate feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length; a second scale feature extraction unit, used to input the flow medium input vector into the second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second scale flow rate feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length; and a multi-scale cascade unit, used to cascade the first scale flow rate feature vector and the second scale flow rate feature vector to obtain the flow rate feature vector.

In the above-mentioned distillation control system for preparing electronic grade carbon tetrafluoride, the responsiveness unit is further used to: calculate the transfer vector of the flow rate feature vector relative to the topological synergy feature matrix as a classification feature vector using the following formula; wherein the formula is: ,in represents the flow velocity feature vector, represents the topological synergy feature matrix, represents the classification feature vector, Represents matrix multiplication.

In the above-mentioned distillation control system for preparing electronic grade carbon tetrafluoride, the distillation control result generating unit is further used to: use the classifier to process the classification feature vector according to the following formula to generate a classification result; wherein the formula is: ,in, represents the classification feature vector, is the weight matrix of the fully connected layer, Represents the bias vector of the fully connected layer.

In the above-mentioned distillation control system for the preparation of electronic grade carbon tetrafluoride, it also includes a training module for training the first convolutional neural network model as a filter, the second convolutional neural network model as a feature extractor, the graph neural network model, the multi-scale neighborhood feature extraction module and the classifier; wherein the training module includes: a training parameter acquisition unit for acquiring training data, the training data including training temperature values and training pressure values of multiple areas of the refining section at multiple predetermined time points within a predetermined time period, training flow rate values of the flow medium at the multiple predetermined time points, and the actual value of the valve opening value for adjusting the flow medium at the current time point that should be increased or decreased; a training temperature and pressure coordination unit for adjusting the refining ... After the training temperature values and training pressure values of each region of the segment at multiple predetermined time points within a predetermined time period are arranged as training temperature input vectors and training pressure input vectors according to the time dimension, the transpose of the training temperature input vector and the product of the training pressure input vector are calculated to obtain multiple training collaborative feature matrices; a training temperature-pressure collaborative feature extraction unit is used to pass the multiple training collaborative feature matrices through the first convolutional neural network model as a filter to obtain multiple training collaborative feature vectors; a training matrixing unit is used to perform two-dimensional matrixing on the multiple training collaborative feature vectors to obtain a training collaborative feature matrix; a training space topology construction unit is used to construct a training topology matrix for the multiple regions, and each of the non-diagonal positions of the training topology matrix The value of each position is the distance between the corresponding two areas, and the value of each position on the diagonal position of the training topological matrix is zero; a training topological feature extraction unit is used to pass the training topological matrix through the second convolutional neural network model as a feature extractor to obtain a training topological feature matrix; a training graph neural network unit is used to pass the training collaborative feature matrix and the training topological feature matrix through the graph neural network model to obtain a training topological collaborative feature matrix; a training flow velocity feature extraction unit is used to arrange the training flow velocity values of the flow medium at the multiple predetermined time points according to the time dimension as a training flow medium input vector through the multi-scale neighborhood feature extraction module to obtain a training flow velocity feature vector; a training responsiveness unit is used to calculate the training flow velocity feature vector A transfer vector relative to the training topological collaborative feature matrix is used as a training classification feature vector; a classification loss unit is used to pass the training classification feature vector through the classifier to obtain a classification loss function value; an internalized learning loss unit is used to calculate a sequence-to-sequence response rule internalized learning loss function value based on the distance between the training flow rate feature vector and the training classification feature vector; and a training unit is used to calculate a weighted sum of the classification loss function value and the sequence-to-sequence response rule internalized learning loss function value as a loss function value to train the first convolutional neural network model as a filter, the second convolutional neural network model as a feature extractor, the graph neural network model, the multi-scale neighborhood feature extraction module and the classifier.

In the above-mentioned distillation control system for preparing electronic grade carbon tetrafluoride, the internalized learning loss unit is further used to: calculate the sequence-to-sequence response rule internalized learning loss function value based on the distance between the training flow rate feature vector and the training classification feature vector using the following formula; wherein the formula is: ,in, is the training flow velocity feature vector, is the training classification feature vector, and and are the weight matrices of the classifier for the training flow velocity feature vector and the training classification feature vector, respectively, express Activation function, express Activation function, represents matrix multiplication, Represents the Euclidean distance between two vectors.

According to another aspect of the present application, a distillation control method for preparing electronic grade carbon tetrafluoride is provided, which includes: obtaining temperature values and pressure values of multiple regions of a refining section at multiple predetermined time points within a predetermined time period, as well as flow rate values of the flow medium at the multiple predetermined time points, collected by a pressure sensor and a temperature sensor; arranging the temperature values and pressure values of each region of the refining section at multiple predetermined time points within a predetermined time period as a temperature input vector and a pressure input vector according to the time dimension, respectively, and calculating the product between the transpose of the temperature input vector and the pressure input vector to obtain multiple collaborative feature matrices; passing the multiple collaborative feature matrices through a first convolutional neural network model as a filter to obtain multiple collaborative feature vectors; two-dimensionally matrixing the multiple collaborative feature vectors to obtain a collaborative feature matrix; constructing a topological matrix of the multiple regions , the value of each position on the non-diagonal position of the topological matrix is the distance between the corresponding two areas, and the value of each position on the diagonal position of the topological matrix is zero; the topological matrix is passed through a second convolutional neural network model as a feature extractor to obtain a topological feature matrix; the collaborative feature matrix and the topological feature matrix are passed through a graph neural network model to obtain a topological collaborative feature matrix; the flow velocity values of the flow medium at the multiple predetermined time points are arranged according to the time dimension as a flow medium input vector through a multi-scale neighborhood feature extraction module to obtain a flow velocity feature vector; the transfer vector of the flow velocity feature vector relative to the topological collaborative feature matrix is calculated as a classification feature vector; and the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the valve opening value for regulating the flow medium at the current time point should be increased or decreased.

Beneficial Effects

Compared with the prior art, the present application provides a distillation control system and method for preparing electronic grade carbon tetrafluoride. It adopts artificial intelligence control technology based on deep learning to extract features of the associated synergistic features of temperature and pressure in different regions of the refining section and the multi-scale change features of the flow rate of the flow medium, and further uses the transfer vector of the two to represent the correlation feature information between the synergistic correlation features of temperature and pressure and the dynamic change features of the flow rate of the flow medium, and thereby performs adaptive real-time control of the valve opening of the flow medium, and in this process, introduces the spatial topological features of different regions of the refining section to further strengthen the extraction of the synergistic correlation features of the temperature and pressure in the spatial position, so as to improve the control accuracy of the valve opening of the flow medium. In this way, the efficiency of distillation can be improved and the consumption of cold can be reduced.

BRIEF DESCRIPTION OF THE DRAWINGS

By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

FIG1 is an application scenario diagram of a distillation control system for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application.

FIG2 is a block diagram of a distillation control system for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application.

FIG3 is a block diagram of the flow rate feature extraction unit in the distillation control system for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application.

FIG4 is a block diagram of a training module further included in the distillation control system for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application.

FIG5 is a flow chart of a distillation control method for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application.

FIG6 is a schematic diagram of a system architecture of a distillation control method for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application.

Embodiments of the present invention

Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

Scenario Overview: As mentioned above, in recent years, due to the development of the electronics industry, the domestic market demand for high-purity carbon tetrafluoride has continued to grow. Domestic companies have also established production and purification devices, but there are certain gaps in process stability and product purity. Therefore, it is of great significance to improve the stable operability of carbon tetrafluoride distillation and purification. In addition, given the characteristics of carbon tetrafluoride, the use of low-temperature distillation consumes a lot of cold energy. How to improve the device and reduce the consumption of cold energy has also become a key issue to be optimized. Therefore, an optimized distillation control system for the preparation of electronic-grade carbon tetrafluoride is desired.

In response to the above technical problems, the present application proposes a distillation device for preparing electronic grade carbon tetrafluoride, which includes a top condenser, a reboiler, and a refining section arranged between the top condenser and the reboiler, wherein the refining section is controlled by a control system of the distillation device for preparing electronic grade carbon tetrafluoride, which controls an electronic regulating valve through PID control to control the flow rate of a flow medium, thereby improving the efficiency of distillation and reducing refrigeration consumption.

When the flow rate of the flow medium is actually controlled, considering that the refining section is controlled by the control system of the electronic grade tetrafluorocarbon distillation device, the electronic regulating valve opening of the flow medium is controlled to control the flow rate. In this process, the pressure and temperature in the refining section will affect the efficiency of distillation and the consumption of cold. Therefore, when the valve opening of the flow medium is regulated to improve the efficiency of distillation and reduce the consumption of cold, it is necessary to perform it according to the actual temperature value and pressure value of the refining section. However, since the existing control scheme has a certain hysteresis, that is, when the flow rate of the flow medium at the current time point is controlled, it is performed according to the temperature value and pressure value of the refining section at the previous moment, which will result in the effect of improving the distillation efficiency and reducing the cold is not obvious. And because temperature and pressure have a certain correlation, and different areas of the refining section have different temperature and pressure characteristics, this increases the control difficulty for the control end.

Based on this, in the technical solution of the present application, an artificial intelligence control technology based on deep learning is used to extract features of the associated synergistic features of temperature and pressure in different areas of the refining section and the multi-scale change features of the flow velocity of the flow medium, and further use the transfer vector of the two to represent the correlation feature information between the synergistic correlation features of temperature and pressure and the dynamic change features of the flow velocity of the flow medium, and use this to perform adaptive real-time control of the valve opening of the flow medium. And in this process, the spatial topological features of different areas of the refining section are introduced to further strengthen the feature extraction of the synergistic correlation of temperature and pressure in the spatial position, so as to improve the control accuracy of the valve opening of the flow medium. In this way, the efficiency of distillation can be improved and the consumption of cold can be reduced.

Specifically, in the technical solution of the present application, first, the temperature values and pressure values of multiple areas of the refining section at multiple predetermined time points within a predetermined time period are collected by a pressure sensor and a temperature sensor, and the flow rate value of the flow medium at the multiple predetermined time points is collected by a flow rate sensor. Then, the temperature values and pressure values of each area of the refining section at multiple predetermined time points within a predetermined time period are arranged as temperature input vectors and pressure input vectors according to the time dimension to integrate the information distribution of the temperature values and pressure values in the time dimension, and then the product between the transpose of the temperature input vector and the pressure input vector is calculated to obtain multiple collaborative feature matrices with temperature and pressure associated information distribution.

Then, a first convolutional neural network model as a filter having excellent performance in implicit feature extraction is used to perform feature extraction on the multiple collaborative feature matrices to respectively extract hidden feature distribution information of the collaborative correlation of temperature and pressure of each region of the refined section, thereby obtaining multiple collaborative feature vectors. The multiple collaborative feature vectors are further two-dimensionally matrixed to obtain a collaborative feature matrix having the collaborative correlation characteristics of temperature and pressure of multiple regions of the refined section as a whole.

Furthermore, considering that in multiple regions of the refined section, there is a correlation between the temperature and pressure correlation synergistic features of each region, and the characteristic distribution of this correlation is in spatial position, therefore, in the technical solution of the present application, the spatial topological features of each region are further strengthened to extract more sufficient temperature and pressure correlation synergistic features. Specifically, first, a topological matrix of the multiple regions is constructed, where the value of each position on the non-diagonal position of the topological matrix is the distance between the corresponding two regions, and the value of each position on the diagonal position of the topological matrix is zero. Then, the topological matrix is subjected to feature mining in the second convolutional neural network model as a feature extractor to extract the spatial topological feature distribution of each region of the refined section, thereby obtaining a topological feature matrix.

Then, the collaborative feature vectors of the various regions are used as feature representations of nodes, and the topological feature matrix is used as feature representations of edges between nodes. The collaborative feature matrix obtained by two-dimensionally arranging the multiple collaborative feature vectors and the topological feature matrix are passed through a graph neural network to obtain a topological collaborative feature matrix. Specifically, the graph neural network encodes the collaborative feature matrix and the topological feature matrix through graph structure data using learnable neural network parameters to obtain the topological collaborative feature matrix containing irregular logical topological features and temperature and pressure-related collaborative features of various regions.

Next, the velocity values of the flow medium at the plurality of predetermined time points are arranged as flow medium input vectors according to the time dimension to integrate the information distribution of the velocity of the flow medium in the time dimension, and then encoded through the multi-scale neighborhood feature extraction module to obtain a velocity feature vector. It should be understood that since the velocity value of the flow medium has different velocity pattern characteristics under different time period spans, the multi-scale neighborhood feature extraction module is used to perform feature encoding on it, so that the multi-scale neighborhood correlation feature information of the velocity value of the flow medium under different time spans within the predetermined time period can be extracted.

Furthermore, the transfer vector of the flow velocity feature vector relative to the topological synergistic feature matrix is calculated as a classification feature vector to represent the correlation feature information between the synergistic correlation topological characteristics of the temperature and pressure and the dynamic multi-scale change characteristics of the flow medium velocity, and to perform adaptive control of the valve opening of the flow medium to improve the efficiency of distillation and reduce cooling consumption.

In particular, in the technical solution of the present application, by calculating the transfer vector of the flow velocity feature vector relative to the topological collaborative feature matrix as the classification feature vector, the transfer response characteristics of the flow velocity feature vector in the collaborative feature topological association space of each sensor can be obtained. In addition, in order to further optimize the expression ability of the transfer response characteristics for the intrinsic feature distribution of the flow velocity feature vector, in addition to the classification loss function, a sequence-to-sequence response rule internalization learning loss function is introduced, which is expressed as: ,in, is the training flow velocity feature vector, is the classification feature vector, and and The classifiers are and The weight matrix of .

Here, the sequence-to-sequence response rule internalization learning loss function can obtain enhanced distinguishing ability between sequences through the classifier's squeeze-excitation channel attention mechanism for the weight matrix of different sequences. In this way, by training the network with this loss function, the velocity feature vector can be realized. and the classification feature vector The causality feature with better discrimination between them is restored to internalize the cause-effect response rules between vector sequences, which enhances the ability of the transfer response feature to express the intrinsic feature distribution of the flow rate feature vector, thereby improving the accuracy and precision of classification. In this way, the valve opening of the flow medium can be adaptively controlled in real time and accurately, thereby improving the efficiency of distillation and reducing the consumption of cold.

Based on this, the present application provides a distillation control system for the preparation of electronic grade carbon tetrafluoride, which includes: a distillation parameter acquisition unit, used to obtain the temperature values and pressure values of multiple regions of the refining section at multiple predetermined time points within a predetermined time period, as well as the flow rate values of the flow medium at the multiple predetermined time points, collected by a pressure sensor and a temperature sensor; a temperature and pressure coordination unit, used to arrange the temperature values and pressure values of each region of the refining section at multiple predetermined time points within a predetermined time period as a temperature input vector and a pressure input vector according to the time dimension, and then calculate the product between the transpose of the temperature input vector and the pressure input vector to obtain multiple coordinated feature matrices; a temperature-pressure coordinated feature extraction unit, used to pass the multiple coordinated feature matrices through a first convolutional neural network model as a filter to obtain multiple coordinated feature vectors; a matrixing unit, used to perform two-dimensional matrixing on the multiple coordinated feature vectors to obtain a coordinated feature matrix; a spatial topology construction unit, used to construct a topological matrix of the multiple regions , the value of each position on the non-diagonal position of the topological matrix is the distance between the corresponding two areas, and the value of each position on the diagonal position of the topological matrix is zero; a topological feature extraction unit, used to pass the topological matrix through a second convolutional neural network model as a feature extractor to obtain a topological feature matrix; a graph neural network unit, used to pass the collaborative feature matrix and the topological feature matrix through a graph neural network model to obtain a topological collaborative feature matrix; a flow rate feature extraction unit, used to arrange the flow rate values of the flow medium at the multiple predetermined time points according to the time dimension as a flow medium input vector through a multi-scale neighborhood feature extraction module to obtain a flow rate feature vector; a responsiveness unit, used to calculate the transfer vector of the flow rate feature vector relative to the topological collaborative feature matrix as a classification feature vector; and a distillation control result generation unit, used to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the valve opening value for adjusting the flow medium at the current time point should be increased or decreased.

FIG1 is an application scenario diagram of a distillation control system for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application. As shown in FIG1 , in this application scenario, a distillation device for preparing electronic grade carbon tetrafluoride includes a tower top condenser 11, a reboiler 13, and a refining section 12 disposed between the tower top condenser 11 and the reboiler 13. The temperature values (for example, D1 as shown in FIG1 ) and pressure values (for example, D2 as shown in FIG1 ) of multiple regions of the refining section at multiple predetermined time points within a predetermined time period are collected by a pressure sensor and a temperature sensor, and the flow rate values of the flow medium at the multiple predetermined time points are collected by a flow rate sensor (for example, D3 as shown in FIG1 ). Then, the obtained values of each of the refining sections are collected. The temperature values and pressure values of the region at multiple predetermined time points within a predetermined time period and the flow rate values of the flow medium at the multiple predetermined time points are input into a server (for example, S shown in FIG1 ) deployed with a distillation control algorithm for the preparation of electronic-grade carbon tetrafluoride, wherein the server is capable of using the distillation control algorithm for the preparation of electronic-grade carbon tetrafluoride to process the temperature values and pressure values of each region of the refining section at multiple predetermined time points within a predetermined time period and the flow rate values of the flow medium at the multiple predetermined time points to generate a classification result indicating whether the valve opening value for adjusting the flow medium at the current time point should be increased or decreased.

After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

Exemplary system: FIG2 is a block diagram of a distillation control system for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application. As shown in FIG2, a distillation control system 100 for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application includes: a distillation parameter acquisition unit 101, which is used to obtain the temperature values and pressure values of multiple regions of the refining section at multiple predetermined time points within a predetermined time period acquired by a pressure sensor and a temperature sensor, as well as the flow rate values of the flow medium at the multiple predetermined time points; a temperature and pressure coordination unit 102, which is used to arrange the temperature values and pressure values of each region of the refining section at multiple predetermined time points within a predetermined time period according to the time dimension as a temperature input vector and a pressure input vector, respectively, and calculate the product between the transpose of the temperature input vector and the pressure input vector to obtain multiple coordinated feature matrices; a temperature-pressure coordinated feature extraction unit 103, which is used to pass the multiple coordinated feature matrices through a first convolutional neural network model as a filter to obtain multiple coordinated feature vectors; a matrixing unit 104, which is used to perform two-dimensional matrixing on the multiple coordinated feature vectors to obtain a coordinated feature matrix; a spatial topology construction unit 105, which is used to construct the topology of the multiple regions A matrix, wherein the value of each position on the non-diagonal position of the topological matrix is the distance between the corresponding two regions, and the value of each position on the diagonal position of the topological matrix is zero; a topological feature extraction unit 106, used to pass the topological matrix through a second convolutional neural network model as a feature extractor to obtain a topological feature matrix; a graph neural network unit 107, used to pass the collaborative feature matrix and the topological feature matrix through a graph neural network model to obtain a topological collaborative feature matrix; a flow rate feature extraction unit 108, used to arrange the flow rate values of the flow medium at the multiple predetermined time points according to the time dimension as a flow medium input vector through a multi-scale neighborhood feature extraction module to obtain a flow rate feature vector; a responsiveness unit 109, used to calculate the transfer vector of the flow rate feature vector relative to the topological collaborative feature matrix as a classification feature vector; and a distillation control result generation unit 110, used to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the valve opening value for adjusting the flow medium at the current time point should be increased or decreased.

More specifically, in the embodiment of the present application, the distillation parameter acquisition unit 101 is used to obtain the temperature values and pressure values of multiple areas of the refining section at multiple predetermined time points within a predetermined time period collected by the pressure sensor and the temperature sensor, as well as the flow rate values of the flow medium at the multiple predetermined time points. Considering that the refining section is controlled by the control system of the preparation electronic grade tetrafluorocarbon distillation device, the electronic regulating valve opening of the flow medium is controlled to control the flow rate. In this process, the pressure and temperature in the refining section will affect the efficiency of distillation and the consumption of cold. Therefore, when the valve opening of the flow medium is regulated to improve the efficiency of distillation and reduce the consumption of cold, it is necessary to perform it according to the actual temperature value and pressure value of the refining section. However, since the existing control scheme has a certain hysteresis, that is, when the flow rate of the flow medium at the current time point is controlled, it is performed according to the temperature value and pressure value of the refining section at the previous moment, which will result in the effect of improving the distillation efficiency and reducing the cold is not obvious. And because temperature and pressure have a certain correlation, and different areas of the refining section have different temperature and pressure characteristics, this increases the control difficulty for the control end. Therefore, the temperature values and pressure values of multiple areas of the refining section at multiple predetermined time points within a predetermined time period, as well as the flow rate values of the flow medium at the multiple predetermined time points, are acquired by the pressure sensor and the temperature sensor, and used as the data basis for determining whether the valve opening value for adjusting the flow medium at the current time point should be increased or decreased.

More specifically, in the embodiment of the present application, the temperature and pressure coordination unit 102 is used to arrange the temperature values and pressure values of each region of the refining section at multiple predetermined time points within a predetermined time period according to the time dimension as a temperature input vector and a pressure input vector, and then calculate the product between the transpose of the temperature input vector and the pressure input vector to obtain multiple coordination feature matrices. Arranging the temperature values and pressure values of each region of the refining section at multiple predetermined time points within a predetermined time period according to the time dimension can integrate the information distribution of the temperature values and pressure values in the time dimension.

More specifically, in the embodiment of the present application, the temperature-pressure collaborative feature extraction unit 103 is used to pass the multiple collaborative feature matrices through the first convolutional neural network model as a filter to obtain multiple collaborative feature vectors. The first convolutional neural network model as a filter has excellent performance in implicit feature extraction. The first convolutional neural network model as a filter is used to extract features from the multiple collaborative feature matrices, and the hidden feature distribution information of the collaborative association of temperature and pressure in each area of the refined section can be extracted respectively.

Accordingly, in a specific example, the temperature-pressure collaborative feature extraction unit 103 is further used to: use each layer of the first convolutional neural network model as a filter to perform convolution processing, feature matrix-based mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer so that the last layer of the first convolutional neural network model as a filter outputs the multiple collaborative feature vectors, wherein the input of the first layer of the first convolutional neural network model as a filter is the multiple collaborative feature matrices.

More specifically, in the embodiment of the present application, the matrixing unit 104 is used to perform two-dimensional matrixing on the multiple collaborative feature vectors to obtain a collaborative feature matrix. Performing two-dimensional matrixing on the multiple collaborative feature vectors can obtain a collaborative feature matrix having the overall temperature and pressure collaborative correlation characteristics of the multiple regions of the refining section.

Furthermore, considering that in the multiple regions of the refining section, there is a correlation between the temperature and pressure correlation synergistic features of the respective regions, and the characteristic distribution of this correlation is in spatial position, therefore, in the technical solution of the present application, it is further strengthened based on the spatial topological features of the respective regions to extract more sufficient temperature and pressure correlation synergistic features.

More specifically, in an embodiment of the present application, the spatial topology construction unit 105 is used to construct a topology matrix of the multiple regions, the value of each position on the non-diagonal position of the topology matrix is the distance between the corresponding two regions, and the value of each position on the diagonal position of the topology matrix is zero.

More specifically, in the embodiment of the present application, the topological feature extraction unit 106 is used to pass the topological matrix through the second convolutional neural network model as a feature extractor to obtain a topological feature matrix. The topological matrix is passed through the second convolutional neural network model as a feature extractor to perform feature mining to extract the spatial topological feature distribution of each area of the refined segment.

Accordingly, in a specific example, the topological feature extraction unit 106 is further used to: use each layer of the second convolutional neural network model as a feature extractor to perform two-dimensional convolution processing, feature matrix-based mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer so that the last layer of the second convolutional neural network model as a feature extractor outputs the topological feature matrix, wherein the input of the first layer of the second convolutional neural network model as a feature extractor is the topological matrix.

More specifically, in the embodiment of the present application, the graph neural network unit 107 is used to obtain a topological collaborative feature matrix by passing the collaborative feature matrix and the topological feature matrix through a graph neural network model. Specifically, the graph neural network performs graph structure data encoding on the collaborative feature matrix and the topological feature matrix through learnable neural network parameters to obtain the topological collaborative feature matrix containing irregular logical topological features and temperature and pressure associated collaborative features of each region.

Accordingly, in a specific example, the graph neural network unit 107 is further used to use the graph neural network model to process the collaborative feature matrix and the topological feature matrix with learnable neural network parameters to obtain the topological collaborative feature matrix containing irregular spatial topological features and temperature-pressure collaborative features.

More specifically, in the embodiment of the present application, the velocity feature extraction unit 108 is used to arrange the velocity values of the flow medium at the plurality of predetermined time points according to the time dimension as a flow medium input vector and pass it through a multi-scale neighborhood feature extraction module to obtain a velocity feature vector. It should be understood that since the velocity value of the flow medium has different velocity pattern characteristics under different time period spans, the multi-scale neighborhood feature extraction module is used to perform feature encoding on it, so that the multi-scale neighborhood correlation feature information of the velocity value of the flow medium under different time spans within the predetermined time period can be extracted.

Accordingly, as shown in Figure 3, in a specific example, the flow rate feature extraction unit 108 includes: a first scale feature extraction unit 1081, used to input the flow medium input vector into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first scale flow rate feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length; a second scale feature extraction unit 1082, used to input the flow medium input vector into the second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second scale flow rate feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length; and a multi-scale cascade unit 1083, used to cascade the first scale flow rate feature vector and the second scale flow rate feature vector to obtain the flow rate feature vector.

Accordingly, in a specific example, the first scale feature extraction unit is further used to: use the first convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the flow medium input vector using the following formula to obtain the first scale flow velocity feature vector; wherein the formula is: ,in, is the first convolution kernel in Width in direction, is the first convolution kernel parameter vector, is the local vector matrix that operates with the convolution kernel function, is the size of the first convolution kernel, represents the flow medium input vector; the second scale feature extraction unit is further used to: use the second convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the flow medium input vector according to the following formula to obtain the second scale flow velocity feature vector; wherein the formula is: ,in, The second convolution kernel is Width in direction, is the second convolution kernel parameter vector, is the local vector matrix that operates with the convolution kernel function, is the size of the second convolution kernel, Represents the flow medium input vector.

More specifically, in the embodiment of the present application, the responsiveness unit 109 is used to calculate the transfer vector of the flow velocity feature vector relative to the topological synergy feature matrix as a classification feature vector.

Accordingly, in a specific example, the responsiveness unit 109 is further used to calculate the transfer vector of the flow velocity feature vector relative to the topological synergy feature matrix as a classification feature vector using the following formula; wherein the formula is: ,in represents the flow velocity feature vector, represents the topological synergy feature matrix, represents the classification feature vector, Represents matrix multiplication.

The transfer vector of the flow velocity feature vector relative to the topological synergistic feature matrix is calculated as a classification feature vector to represent the correlation feature information between the synergistic correlation topological characteristics of the temperature and pressure and the dynamic multi-scale change characteristics of the flow medium velocity, and to perform adaptive control of the valve opening of the flow medium to improve the efficiency of distillation and reduce the consumption of cold.

More specifically, in the embodiment of the present application, the distillation control result generating unit 110 is used to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the valve opening value for regulating the flow medium at the current time point should be increased or decreased.

Accordingly, in a specific example, the distillation control result generating unit 110 is further used to: use the classifier to process the classification feature vector according to the following formula to generate a classification result; wherein the formula is: ,in, represents the classification feature vector, is the weight matrix of the fully connected layer, Represents the bias vector of the fully connected layer.

Accordingly, in a specific example, the distillation control system for preparing electronic grade carbon tetrafluoride also includes a training module for training the first convolutional neural network model as a filter, the second convolutional neural network model as a feature extractor, the graph neural network model, the multi-scale neighborhood feature extraction module and the classifier; as shown in Figure 4, wherein the training module 200 includes: a training parameter acquisition unit 201, used to obtain training data, the training data including training temperature values and training pressure values of multiple areas of the refining section at multiple predetermined time points within a predetermined time period, training flow rate values of the flow medium at the multiple predetermined time points, and the actual value of the valve opening value for adjusting the flow medium at the current time point that should be increased or decreased; a training temperature and pressure coordination unit A unit 202 is used to arrange the training temperature values and training pressure values of each area of the refining section at multiple predetermined time points within a predetermined time period into a training temperature input vector and a training pressure input vector according to the time dimension, and then calculate the product between the transpose of the training temperature input vector and the training pressure input vector to obtain multiple training collaborative feature matrices; a training temperature-pressure collaborative feature extraction unit 203 is used to pass the multiple training collaborative feature matrices through the first convolutional neural network model as a filter to obtain multiple training collaborative feature vectors; a training matrixing unit 204 is used to perform two-dimensional matrixing on the multiple training collaborative feature vectors to obtain a training collaborative feature matrix; a training space topology construction unit 205 is used to construct a training topology matrix for the multiple areas, and the non-convolutional neural network model of the training topology matrix is used to obtain a training collaborative feature matrix. The value of each position on the diagonal position is the distance between the corresponding two areas, and the value of each position on the diagonal position of the training topological matrix is zero; a training topological feature extraction unit 206 is used to pass the training topological matrix through the second convolutional neural network model as a feature extractor to obtain a training topological feature matrix; a training graph neural network unit 207 is used to pass the training collaborative feature matrix and the training topological feature matrix through the graph neural network model to obtain a training topological collaborative feature matrix; a training flow velocity feature extraction unit 208 is used to arrange the training flow velocity values of the flow medium at the multiple predetermined time points according to the time dimension as a training flow medium input vector through the multi-scale neighborhood feature extraction module to obtain a training flow velocity feature vector; a training responsiveness unit 209 is used to calculate the training flow velocity The transfer vector of the feature vector relative to the training topological collaborative feature matrix is used as a training classification feature vector; a classification loss unit 210 is used to pass the training classification feature vector through the classifier to obtain a classification loss function value; an internalized learning loss unit 211 is used to calculate the sequence-to-sequence response rule internalized learning loss function value based on the distance between the training flow rate feature vector and the training classification feature vector; and a training unit 212 is used to calculate the weighted sum of the classification loss function value and the sequence-to-sequence response rule internalized learning loss function value as a loss function value to train the first convolutional neural network model as a filter, the second convolutional neural network model as a feature extractor, the graph neural network model, the multi-scale neighborhood feature extraction module and the classifier.

In particular, in the technical solution of the present application, by calculating the transfer vector of the flow velocity feature vector relative to the topological collaborative feature matrix as a classification feature vector, the transfer response feature of the flow velocity feature vector in the collaborative feature topological association space of each sensor can be obtained. In addition, in order to further optimize the expression ability of the transfer response feature for the intrinsic feature distribution of the flow velocity feature vector, in addition to the classification loss function, a sequence-to-sequence response rule internalization learning loss function is introduced.

Accordingly, in a specific example, the internalized learning loss unit 211 is further used to: calculate the sequence-to-sequence response rule internalized learning loss function value based on the distance between the training flow velocity feature vector and the training classification feature vector using the following formula; wherein the formula is: ,in, is the training flow velocity feature vector, is the training classification feature vector, and and are the weight matrices of the classifier for the training flow velocity feature vector and the training classification feature vector, respectively, express Activation function, express Activation function, represents matrix multiplication, Represents the Euclidean distance between two vectors.

Here, the sequence-to-sequence response rule internalization learning loss function can obtain enhanced distinguishing ability between sequences through the classifier's squeeze-excitation channel attention mechanism for the weight matrix of different sequences. In this way, by training the network with this loss function, the velocity feature vector can be realized. and the classification feature vector The causality feature with better discrimination between them is restored to internalize the cause-effect response rules between vector sequences, which enhances the ability of the transfer response feature to express the intrinsic feature distribution of the flow rate feature vector, thereby improving the accuracy and precision of classification. In this way, the valve opening of the flow medium can be adaptively controlled in real time and accurately, thereby improving the efficiency of distillation and reducing the consumption of cold.

In summary, the distillation control system 100 for preparing electronic grade carbon tetrafluoride based on the embodiment of the present application is explained, which adopts artificial intelligence control technology based on deep learning to extract features of the associated synergistic characteristics of temperature and pressure in different regions of the refining section and the multi-scale change characteristics of the flow rate of the flow medium, and further uses the transfer vector of the two to represent the correlation feature information between the synergistic correlation characteristics of temperature and pressure and the dynamic change characteristics of the flow rate of the flow medium, and thereby performs adaptive real-time control of the valve opening of the flow medium, and in this process, the spatial topological characteristics of different regions of the refining section are introduced to further strengthen the extraction of the synergistic correlation characteristics of the temperature and pressure in the spatial position, so as to improve the control accuracy of the valve opening of the flow medium. In this way, the efficiency of distillation can be improved and the consumption of cold can be reduced.

As described above, the distillation control system 100 for preparing electronic-grade carbon tetrafluoride according to the embodiment of the present application can be implemented in various terminal devices, such as a server having a distillation control algorithm for preparing electronic-grade carbon tetrafluoride. In one example, the distillation control system 100 for preparing electronic-grade carbon tetrafluoride can be integrated into the terminal device as a software module and/or a hardware module. For example, the distillation control system 100 for preparing electronic-grade carbon tetrafluoride 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 distillation control system 100 for preparing electronic-grade carbon tetrafluoride can also be one of the many hardware modules of the terminal device.

Alternatively, in another example, the distillation control system 100 for preparing electronic-grade carbon tetrafluoride and the terminal device may also be separate devices, and the distillation control system 100 for preparing electronic-grade carbon tetrafluoride may be connected to the terminal device via a wired and/or wireless network, and transmit interactive information in accordance with an agreed data format.

Exemplary method: FIG5 is a flow chart of a distillation control method for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application. As shown in FIG5, the distillation control method for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application includes: S101, acquiring the temperature values and pressure values of multiple regions of the refining section at multiple predetermined time points within a predetermined time period collected by a pressure sensor and a temperature sensor, as well as the flow rate values of the flow medium at the multiple predetermined time points; S102, arranging the temperature values and pressure values of each region of the refining section at multiple predetermined time points within a predetermined time period as a temperature input vector and a pressure input vector according to the time dimension, respectively, and calculating the product between the transpose of the temperature input vector and the pressure input vector to obtain multiple collaborative feature matrices; S103, passing the multiple collaborative feature matrices through a first convolutional neural network model as a filter to obtain multiple collaborative feature vectors; S104, two-dimensionally matrixing the multiple collaborative feature vectors to obtain a collaborative feature matrix; S105, constructing a topological matrix of the multiple regions, The value of each position on the non-diagonal position of the topological matrix is the distance between the corresponding two areas, and the value of each position on the diagonal position of the topological matrix is zero; S106, the topological matrix is passed through a second convolutional neural network model as a feature extractor to obtain a topological feature matrix; S107, the collaborative feature matrix and the topological feature matrix are passed through a graph neural network model to obtain a topological collaborative feature matrix; S108, the flow velocity values of the flow medium at the multiple predetermined time points are arranged according to the time dimension as a flow medium input vector and passed through a multi-scale neighborhood feature extraction module to obtain a flow velocity feature vector; S109, the transfer vector of the flow velocity feature vector relative to the topological collaborative feature matrix is calculated as a classification feature vector; and, S110, the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the valve opening value for regulating the flow medium at the current time point should be increased or decreased.

FIG6 is a schematic diagram of the system architecture of the distillation control method for preparing electronic grade carbon tetrafluoride according to an embodiment of the present application. As shown in FIG6, in the system architecture of the distillation control method for preparing electronic grade carbon tetrafluoride, first, the temperature values and pressure values of multiple regions of the refining section at multiple predetermined time points within a predetermined time period collected by the pressure sensor and the temperature sensor, as well as the flow rate values of the flow medium at the multiple predetermined time points, are obtained; then, the temperature values and pressure values of each region of the refining section at multiple predetermined time points within a predetermined time period are arranged as temperature input vectors and pressure input vectors according to the time dimension, and the product between the transpose of the temperature input vector and the pressure input vector is calculated to obtain multiple collaborative feature matrices; then, the multiple collaborative feature matrices are passed through the first convolutional neural network model as a filter to obtain multiple collaborative feature vectors; then, the multiple collaborative feature vectors are two-dimensionally matrixed to obtain a collaborative feature matrix; then, a topological matrix of the multiple regions is constructed. The value of each position on the non-diagonal position of the topological matrix is the distance between the corresponding two areas, and the value of each position on the diagonal position of the topological matrix is zero; then, the topological matrix is passed through a second convolutional neural network model as a feature extractor to obtain a topological feature matrix; then, the collaborative feature matrix and the topological feature matrix are passed through a graph neural network model to obtain a topological collaborative feature matrix; then, the flow velocity values of the flow medium at the multiple predetermined time points are arranged according to the time dimension as a flow medium input vector and passed through a multi-scale neighborhood feature extraction module to obtain a flow velocity feature vector; then, the transfer vector of the flow velocity feature vector relative to the topological collaborative feature matrix is calculated as a classification feature vector; finally, the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the valve opening value for adjusting the flow medium at the current time point should be increased or decreased.

In a specific example, in the above-mentioned distillation control method for the preparation of electronic grade carbon tetrafluoride, the multiple collaborative feature matrices are passed through the first convolutional neural network model as a filter to obtain multiple collaborative feature vectors, further including: using each layer of the first convolutional neural network model as a filter to perform convolution processing, feature matrix-based mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer to output the multiple collaborative feature vectors from the last layer of the first convolutional neural network model as a filter, wherein the input of the first layer of the first convolutional neural network model as a filter is the multiple collaborative feature matrices.

In a specific example, in the above-mentioned distillation control method for the preparation of electronic grade carbon tetrafluoride, the topological matrix is passed through the second convolutional neural network model as a feature extractor to obtain a topological feature matrix, further including: using each layer of the second convolutional neural network model as a feature extractor to perform two-dimensional convolution processing, feature matrix-based mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer to output the topological feature matrix from the last layer of the second convolutional neural network model as a feature extractor, wherein the input of the first layer of the second convolutional neural network model as a feature extractor is the topological matrix.

In a specific example, in the above-mentioned distillation control method for the preparation of electronic grade carbon tetrafluoride, the synergistic feature matrix and the topological feature matrix are passed through a graph neural network model to obtain a topological synergistic feature matrix, further comprising: using the graph neural network model to process the synergistic feature matrix and the topological feature matrix with learnable neural network parameters to obtain the topological synergistic feature matrix containing irregular spatial topological features and temperature-pressure synergistic features.

In a specific example, in the above-mentioned distillation control method for the preparation of electronic grade carbon tetrafluoride, the flow rate values of the flow medium at the multiple predetermined time points are arranged according to the time dimension as a flow medium input vector through a multi-scale neighborhood feature extraction module to obtain a flow rate feature vector, including: inputting the flow medium input vector into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale flow rate feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length; inputting the flow medium input vector into the second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale flow rate feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length; and, cascading the first-scale flow rate feature vector and the second-scale flow rate feature vector to obtain the flow rate feature vector.

Accordingly, in a specific example, the step of inputting the flow medium input vector into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale flow velocity feature vector further includes: using the first convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the flow medium input vector using the following formula to obtain the first-scale flow velocity feature vector; wherein the formula is: ,in, is the first convolution kernel in Width in direction, is the first convolution kernel parameter vector, is the local vector matrix that operates with the convolution kernel function, is the size of the first convolution kernel, represents the flow medium input vector; the second scale feature extraction unit is further used to: use the second convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the flow medium input vector according to the following formula to obtain the second scale flow velocity feature vector; wherein the formula is: ,in, The second convolution kernel is Width in direction, is the second convolution kernel parameter vector, is the local vector matrix that operates with the convolution kernel function, is the size of the second convolution kernel, Represents the flow medium input vector.

More specifically, in the embodiment of the present application, the responsiveness unit 109 is used to calculate the transfer vector of the flow velocity feature vector relative to the topological synergy feature matrix as a classification feature vector.

Accordingly, in a specific example, the responsiveness unit 109 is further used to calculate the transfer vector of the flow velocity feature vector relative to the topological synergy feature matrix as a classification feature vector using the following formula; wherein the formula is: ,in represents the flow velocity feature vector, represents the topological synergy feature matrix, represents the classification feature vector, Represents matrix multiplication.

In a specific example, in the above-mentioned distillation control method for preparing electronic grade carbon tetrafluoride, the step of passing the classification feature vector through a classifier to obtain a classification result further includes: using the classifier to process the classification feature vector using the following formula to generate a classification result; wherein the formula is: ,in, represents the classification feature vector, is the weight matrix of the fully connected layer, Represents the bias vector of the fully connected layer.

In a specific example, in the above-mentioned distillation control method for preparing electronic grade carbon tetrafluoride, it also includes: training the first convolutional neural network model as a filter, the second convolutional neural network model as a feature extractor, the graph neural network model, the multi-scale neighborhood feature extraction module and the classifier; wherein, the training of the first convolutional neural network model as a filter, the second convolutional neural network model as a feature extractor, the graph neural network model, the multi-scale neighborhood feature extraction module and the classifier further includes: acquiring training data, the training data including training temperature values and training pressure value, the training flow rate values of the flow medium at the plurality of predetermined time points, and the real value of the valve opening value for adjusting the flow medium at the current time point that should be increased or decreased; after arranging the training temperature values and the training pressure values of the various regions of the refining section at the plurality of predetermined time points within the predetermined time period as training temperature input vectors and training pressure input vectors according to the time dimension, respectively, calculating the product between the transpose of the training temperature input vector and the training pressure input vector to obtain a plurality of training collaborative feature matrices; passing the plurality of training collaborative feature matrices through the first convolutional neural network model as a filter to obtain a plurality of training collaborative feature vectors; performing two-dimensional matrixing on the plurality of training collaborative feature vectors to obtain a plurality of training collaborative feature matrices; Obtain a training collaborative feature matrix; construct a training topology matrix of the multiple regions, wherein the value of each position on the off-diagonal position of the training topology matrix is the distance between the corresponding two regions, and the value of each position on the diagonal position of the training topology matrix is zero; pass the training topology matrix through the second convolutional neural network model as a feature extractor to obtain a training topology feature matrix; pass the training collaborative feature matrix and the training topology feature matrix through the graph neural network model to obtain a training topology collaborative feature matrix; arrange the training flow velocity values of the flow medium at the multiple predetermined time points according to the time dimension as a training flow medium input vector and pass it through the multi-scale neighborhood feature extraction module to obtain a training flow velocity feature vector; calculate Calculate the transfer vector of the training flow velocity feature vector relative to the training topological collaborative feature matrix as a training classification feature vector; pass the training classification feature vector through the classifier to obtain a classification loss function value; calculate the sequence-to-sequence response rule internalized learning loss function value based on the distance between the training flow velocity feature vector and the training classification feature vector; and calculate the weighted sum of the classification loss function value and the sequence-to-sequence response rule internalized learning loss function value as the loss function value to train the first convolutional neural network model as a filter, the second convolutional neural network model as a feature extractor, the graph neural network model, the multi-scale neighborhood feature extraction module and the classifier.

In a specific example, in the above-mentioned distillation control method for preparing electronic grade carbon tetrafluoride, the calculation of the sequence-to-sequence response rule internalization learning loss function value based on the distance between the training flow rate feature vector and the training classification feature vector further includes: calculating the sequence-to-sequence response rule internalization learning loss function value based on the distance between the training flow rate feature vector and the training classification feature vector according to the following formula; wherein the formula is: ,in, is the training flow velocity feature vector, is the training classification feature vector, and and are the weight matrices of the classifier for the training flow velocity feature vector and the training classification feature vector, respectively, express Activation function, express Activation function, represents matrix multiplication, Represents the Euclidean distance between two vectors.

Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned distillation control method for preparing electronic grade carbon tetrafluoride have been described in detail in the description of the distillation control system for preparing electronic grade carbon tetrafluoride with reference to Figures 1 to 4 above, and therefore, its repeated description will be omitted.

Claims (10)

1. A distillation control system for preparing electronic grade carbon tetrafluoride, characterized in that it includes: a distillation parameter acquisition unit, used to obtain the temperature values and pressure values of multiple regions of the refining section at multiple predetermined time points within a predetermined time period, as well as the flow rate values of the flow medium at the multiple predetermined time points, collected by a pressure sensor and a temperature sensor; a temperature and pressure coordination unit, used to arrange the temperature values and pressure values of each region of the refining section at multiple predetermined time points within a predetermined time period as a temperature input vector and a pressure input vector according to the time dimension, and then calculate the product between the transpose of the temperature input vector and the pressure input vector to obtain multiple coordinated feature matrices; a temperature-pressure coordinated feature extraction unit, used to pass the multiple coordinated feature matrices through a first convolutional neural network model as a filter to obtain multiple coordinated feature vectors; a matrixing unit, used to perform two-dimensional matrixing on the multiple coordinated feature vectors to obtain a coordinated feature matrix; and a spatial topology construction unit, used to construct a topology matrix of the multiple regions. The value of each position on the non-diagonal position of the topological matrix is the distance between the corresponding two areas, and the value of each position on the diagonal position of the topological matrix is zero; a topological feature extraction unit, used to pass the topological matrix through a second convolutional neural network model as a feature extractor to obtain a topological feature matrix; a graph neural network unit, used to pass the collaborative feature matrix and the topological feature matrix through a graph neural network model to obtain a topological collaborative feature matrix; a flow rate feature extraction unit, used to arrange the flow rate values of the flow medium at the multiple predetermined time points according to the time dimension as a flow medium input vector through a multi-scale neighborhood feature extraction module to obtain a flow rate feature vector; a responsiveness unit, used to calculate the transfer vector of the flow rate feature vector relative to the topological collaborative feature matrix as a classification feature vector; and a distillation control result generation unit, used to pass the classification feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether the valve opening value for adjusting the flow medium at the current time point should be increased or decreased.
2. The distillation control system for the preparation of electronic grade carbon tetrafluoride according to claim 1 is characterized in that the temperature-pressure collaborative feature extraction unit is further used to: use each layer of the first convolutional neural network model as a filter to perform convolution processing, feature matrix-based mean pooling processing and nonlinear activation processing on the input data in the forward transfer of the layer to output the multiple collaborative feature vectors by the last layer of the first convolutional neural network model as a filter, wherein the input of the first layer of the first convolutional neural network model as a filter is the multiple collaborative feature matrices.
3. The distillation control system for the preparation of electronic grade carbon tetrafluoride according to claim 2 is characterized in that the topological feature extraction unit is further used to: use each layer of the second convolutional neural network model as a feature extractor to perform two-dimensional convolution processing, feature matrix-based mean pooling processing and nonlinear activation processing on the input data in the forward pass of the layer to output the topological feature matrix by the last layer of the second convolutional neural network model as a feature extractor, wherein the input of the first layer of the second convolutional neural network model as a feature extractor is the topological matrix.
4. The distillation control system for the preparation of electronic grade carbon tetrafluoride according to claim 3 is characterized in that the graph neural network unit is further used to use the graph neural network model to process the collaborative feature matrix and the topological feature matrix with learnable neural network parameters to obtain the topological collaborative feature matrix containing irregular spatial topological features and temperature-pressure collaborative features.
5. The distillation control system for preparing electronic grade carbon tetrafluoride according to claim 4 is characterized in that the flow rate feature extraction unit comprises: a first scale feature extraction unit, which is used to input the flow medium input vector into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first scale flow rate feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length; a second scale feature extraction unit, which is used to input the flow medium input vector into the second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second scale flow rate feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length; and a multi-scale cascade unit, which is used to cascade the first scale flow rate feature vector and the second scale flow rate feature vector to obtain the flow rate feature vector.
6. The distillation control system for preparing electronic grade carbon tetrafluoride according to claim 5, characterized in that the responsiveness unit is further used to: calculate the transfer vector of the flow velocity feature vector relative to the topological synergy feature matrix as a classification feature vector using the following formula; wherein the formula is: ,in represents the flow velocity feature vector, represents the topological synergy feature matrix, represents the classification feature vector, Represents matrix multiplication.
7. The distillation control system for preparing electronic grade carbon tetrafluoride according to claim 6 is characterized in that the distillation control result generating unit is further used to: use the classifier to process the classification feature vector according to the following formula to generate a classification result; wherein the formula is: ,in, represents the classification feature vector, is the weight matrix of the fully connected layer, Represents the bias vector of the fully connected layer.
8. The distillation control system for preparing electronic grade carbon tetrafluoride according to claim 1 is characterized in that it also includes a training module for training the first convolutional neural network model as a filter, the second convolutional neural network model as a feature extractor, the graph neural network model, the multi-scale neighborhood feature extraction module and the classifier; wherein the training module includes: a training parameter acquisition unit for acquiring training data, the training data including training temperature values and training pressure values of multiple areas of the refining section at multiple predetermined time points within a predetermined time period, training flow rate values of the flow medium at the multiple predetermined time points, and the actual value of the valve opening value for adjusting the flow medium at the current time point that should be increased or decreased; a training temperature and pressure coordination unit , used to arrange the training temperature values and training pressure values of each area of the refining section at multiple predetermined time points within a predetermined time period into a training temperature input vector and a training pressure input vector according to the time dimension, and then calculate the product between the transpose of the training temperature input vector and the training pressure input vector to obtain multiple training collaborative feature matrices; a training temperature-pressure collaborative feature extraction unit, used to pass the multiple training collaborative feature matrices through the first convolutional neural network model as a filter to obtain multiple training collaborative feature vectors; a training matrixing unit, used to perform two-dimensional matrixing on the multiple training collaborative feature vectors to obtain a training collaborative feature matrix; a training space topology construction unit, used to construct training topology matrices for the multiple areas, the non-parallel training topology matrices The value of each position on the diagonal position is the distance between the corresponding two areas, and the value of each position on the diagonal position of the training topological matrix is zero; a training topological feature extraction unit, used to pass the training topological matrix through the second convolutional neural network model as a feature extractor to obtain a training topological feature matrix; a training graph neural network unit, used to pass the training collaborative feature matrix and the training topological feature matrix through the graph neural network model to obtain a training topological collaborative feature matrix; a training flow velocity feature extraction unit, used to arrange the training flow velocity values of the flow medium at the multiple predetermined time points according to the time dimension as a training flow medium input vector through the multi-scale neighborhood feature extraction module to obtain a training flow velocity feature vector; a training responsiveness unit, used to calculate the training flow velocity feature vector. The transfer vector of the eigenvector relative to the training topological collaborative feature matrix is used as a training classification feature vector; a classification loss unit is used to pass the training classification feature vector through the classifier to obtain a classification loss function value; an internalized learning loss unit is used to calculate the sequence-to-sequence response rule internalized learning loss function value based on the distance between the training flow rate feature vector and the training classification feature vector; and a training unit is used to calculate the weighted sum of the classification loss function value and the sequence-to-sequence response rule internalized learning loss function value as a loss function value to train the first convolutional neural network model as a filter, the second convolutional neural network model as a feature extractor, the graph neural network model, the multi-scale neighborhood feature extraction module and the classifier.
9. The distillation control system for preparing electronic grade carbon tetrafluoride according to claim 8, characterized in that the internalized learning loss unit is further used to: calculate the sequence-to-sequence response rule internalized learning loss function value based on the distance between the training flow rate feature vector and the training classification feature vector according to the following formula; wherein the formula is: ,in, is the training flow velocity feature vector, is the training classification feature vector, and and are the weight matrices of the classifier for the training flow velocity feature vector and the training classification feature vector, respectively, express Activation function, express Activation function, represents matrix multiplication, Represents the Euclidean distance between two vectors.
10. A distillation control method for preparing electronic grade carbon tetrafluoride, characterized in that it includes: obtaining temperature values and pressure values of multiple regions of a refining section at multiple predetermined time points within a predetermined time period, as well as flow rate values of the flow medium at the multiple predetermined time points, collected by a pressure sensor and a temperature sensor; arranging the temperature values and pressure values of each region of the refining section at multiple predetermined time points within a predetermined time period as a temperature input vector and a pressure input vector according to the time dimension, respectively, and calculating the product between the transpose of the temperature input vector and the pressure input vector to obtain multiple collaborative feature matrices; passing the multiple collaborative feature matrices through a first convolutional neural network model as a filter to obtain multiple collaborative feature vectors; performing two-dimensional matrixing on the multiple collaborative feature vectors to obtain a collaborative feature matrix; constructing a topological matrix of the multiple regions, the topological matrix The value of each position on the non-diagonal position of the matrix is the distance between the corresponding two areas, and the value of each position on the diagonal position of the topological matrix is zero; the topological matrix is passed through a second convolutional neural network model as a feature extractor to obtain a topological feature matrix; the collaborative feature matrix and the topological feature matrix are passed through a graph neural network model to obtain a topological collaborative feature matrix; the flow velocity values of the flow medium at the multiple predetermined time points are arranged according to the time dimension as a flow medium input vector and passed through a multi-scale neighborhood feature extraction module to obtain a flow velocity feature vector; the transfer vector of the flow velocity feature vector relative to the topological collaborative feature matrix is calculated as a classification feature vector; and the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the valve opening value for regulating the flow medium at the current time point should be increased or decreased.