WO2023226226A1 - Rectification control system for preparation of electronic-grade trifluoromethane and control method therefor - Google Patents

Abstract

A rectification control system for preparation of electronic-grade trifluoromethane and a control method therefor. The method comprises: acquiring temperature data of a molecular sieve adsorber, pressure data of the molecular sieve adsorber, flow data of a crude trifluoromethane product input into the molecular sieve adsorber, and an operating power of a low-temperature heat exchanger of a rectification tower at a plurality of predetermined time points including the current time point (S110); acquiring gas chromatograms of a rectification product at the plurality of predetermined time points including the current time point (S120); arranging into a two-dimensional input matrix the temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber and the flow data of the crude trifluoromethane product input into the molecular sieve adsorber at the plurality of predetermined time points including the current time point, and then passing the two-dimensional input matrix through a first convolutional neural network, so as to generate a first feature map, wherein adjacent layers of the first convolutional neural network use convolutional kernels, which are mutually transposed (S130); passing, through a context encoder, which includes an embedding layer, the operating power of the low-temperature heat exchanger of the rectification tower at the plurality of predetermined time points including the current time point, so as to obtain a plurality of feature vectors, and arranging the plurality of feature vectors, which are in two dimensions, into a feature matrix and then passing the feature matrix through a second convolutional neural network, so as to obtain a second feature map (S140); passing, through a third convolutional neural network using a three-dimensional convolutional kernel, gas chromatograms of the rectification product at the plurality of predetermined time points including the current time point, so as to obtain a third feature map (S150); performing class-difference-based feature value correction on each of the first to third feature maps, so as to generate corrected first to third feature maps (S160); fusing the corrected first to third feature maps, so as to obtain a classification feature map (S170); and passing the classification feature map through a classifier, so as to obtain a classification result, wherein the classification result indicates whether a combination of control parameters at the current time point is reasonable (S180).

Description

Distillation control system for the preparation of electronic grade trifluoromethane and its control method Technical field

The present invention relates to the field of intelligent manufacturing of electronic grade gases, and more specifically, to a distillation control system for the preparation of electronic grade trifluoromethane and a control method thereof.

Background technique

Trifluoromethane, also known as trifluoromethane, is a colorless, slightly odorless, non-conductive gas and is an ideal substitute for alkyl halides. In the semiconductor process, CHF3 is often used in plasma etching or reactive ion etching of silicon dioxide. The characteristic of CHF3 is that it corrodes silicon dioxide quickly and corrodes silicon slowly. That is, it not only has good selectivity, but also has a large rate difference. Meet the requirements of semiconductor processes. The demand for high-purity trifluoromethane as an etchant in the manufacturing process of 8-12-inch chips continues to increase with the rapid development of the semiconductor industry.

Generally, the purity of high-purity trifluoromethane used in the semiconductor industry is 99.999%. Its purification involves deep removal technology of various impurities. Trifluoromethane is highly polar and the raw materials generally contain large amounts of CHCl3, CCl2F2, CHClF2, O2, N2, and CO2. and other impurities, the boiling points of CHF3 and CO2 are very close, and the boiling points and properties of difluorochloromethane are very similar, and the two easily form an azeotrope, making separation difficult.

The purity of my country's existing industrialized trifluoromethane is relatively low, generally 99.8%-99.9%, which does not meet the requirements of the semiconductor industry. There are some purification schemes for trifluoromethane. For example, patent 103951543A uses a purification device with multi-stage distillation to improve the purity of trifluoromethane. However, this purification scheme optimized at the structural level requires the use of complex structures. Purification equipment is expensive.

Therefore, a new purification control scheme for trifluoromethane is expected so that the purity of trifluoromethane obtained by final purification can meet the purity requirements of electronic grade gas.

Contents of the invention

In order to solve the above technical problems, this application is proposed. Embodiments of the present application provide a distillation control system and a control method for the preparation of electronic grade trifluoromethane, which use an intelligent control method based on artificial intelligence technology to conduct global, high-level operations from the perspective of optimizing control algorithms. The control parameters are adjusted accurately, dynamically and adaptively, thereby improving the purification purity of the electronic grade trifluoromethane.

According to one aspect of the present application, a distillation control system for the preparation of electronic-grade trifluoromethane is provided, which includes: a distillation parameter data acquisition unit for acquiring multiple predetermined time points including the current time point. The temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, the flow data of the trifluoromethane crude product input into the molecular sieve adsorber, and the working power of the low-temperature heat exchanger of the distillation tower; the product data acquisition unit uses In order to obtain the gas chromatogram of the distillation product at multiple predetermined time points including the current time point; the first-level distillation parameter encoding unit is used to obtain the gas chromatogram of the distillation product at multiple predetermined time points including the current time point. The temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, and the flow data of crude trifluoromethane into the molecular sieve adsorber are arranged into a two-dimensional input matrix and then passed through the first convolutional neural network to generate the first feature. Figure, wherein the adjacent layers of the first convolutional neural network use mutually transposed convolution kernels; the second-level distillation parameter encoding unit is used to convert the multiple predetermined values including the current time point into The working power of the low-temperature heat exchanger of the distillation tower at the time point is passed through the context encoder containing the embedding layer to obtain multiple feature vectors, and the multiple feature vectors are two-dimensionally arranged into a feature matrix and then passed through the second convolution The neural network obtains the second feature map; the product data encoding unit is used to pass the gas chromatogram of the distillation product at multiple predetermined time points including the current time point through a third convolution using a three-dimensional convolution kernel a neural network to obtain a third feature map; a feature map correction unit configured to perform feature value correction based on category differences on each of the first to third feature maps to generate corrected first to third feature maps; a feature map fusion unit, used to fuse the corrected first to third feature maps to obtain a classification feature map; and a control result generation unit, used to pass the classification feature map through a classifier to obtain a classification result, the classification The result is whether the combination of control parameters at the current time point is reasonable.

In the above-mentioned distillation control system for the preparation of electronic-grade trifluoromethane, the first-level distillation parameter encoding unit is further used to: use the convolution layer of the i-th layer of the first convolutional neural network. The first convolution kernel processes the input data with the following formula to generate the i-th feature map, the formula is h=f( _wi *x+b), * represents the convolution operation, and f(.) represents the activation function; And, use the convolution layer of the i+1th layer of the first convolutional neural network to use the second convolution kernel to process the i-th feature map with the following formula, the formula is h=f( _{wi +1} *x+b), * represents the convolution operation, f(.) represents the activation function, where w _i+1 = w _i ^T .

In the above-mentioned distillation control system for electronic-grade trifluoromethane preparation, the second-level distillation parameter encoding unit includes: an embedded layer subunit, used for using the encoder model containing the context of the embedded layer. The embedding layer converts the working power of the cryogenic heat exchanger of the distillation tower at multiple predetermined time points, including the current time point, into input vectors to obtain a sequence of input vectors; the context encoding subunit is used to use all The converter of the encoder model that includes the context of the embedding layer performs global contextual semantic encoding on the sequence of the input vectors obtained by the embedding layer subunit to obtain the plurality of feature vectors; the subunits are arranged two-dimensionally, Used to two-dimensionally arrange the multiple feature vectors obtained by the context encoding subunit into a feature matrix and then pass it through a second convolutional neural network to obtain a second feature map.

In the above-mentioned distillation control system for the preparation of electronic grade trifluoromethane, the product data encoding unit is further used to: each layer of the third convolutional neural network using the three-dimensional convolution kernel is in the forward direction of the layer. Convolution processing, pooling processing and activation processing are performed on the input data during the transfer to generate the third feature map from the last layer of the third convolutional neural network, wherein the third layer of the third convolutional neural network The input of one layer is the gas chromatogram of the distillation product at multiple predetermined time points including the current time point.

In the above-mentioned distillation control system for the preparation of electronic-grade trifluoromethane, the characteristic map correction unit is further used to perform category difference-based classification on each of the first to third characteristic maps using the following formula: Feature value correction to generate the corrected first to third feature maps;

Among them, the formula is:

where f _i,j,k is the feature value of the (i,j,k)th position of the feature map, and

is the global mean of the feature values at each position of the feature map.

In the above-mentioned distillation control system for the preparation of electronic-grade trifluoromethane, the control result generation unit is further configured to: the classifier processes the classification feature map according to the following formula to generate a classification result, wherein, The formula is: softmax{(W _n ,B _n ):...:(W ₁ ,B ₁ )|Project(F)}, where Project(F) represents projecting the classification feature map into a vector, W ₁ to W _n is the weight matrix of the fully connected layer of each layer, and B ₁ to B _n represent the bias matrix of the fully connected layer of each layer.

According to another aspect of the present application, a control method for a distillation control system for the preparation of electronic grade trifluoromethane, which includes:

Obtain the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point, the pressure data of the molecular sieve adsorber, the flow data of the trifluoromethane crude product input into the molecular sieve adsorber, and the low temperature of the rectification tower. The working power of the heat exchanger;

Obtain gas chromatograms of distillation products at multiple predetermined time points including the current time point;

Arrange the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point, the pressure data of the molecular sieve adsorber, and the flow data of crude trifluoromethane into the molecular sieve adsorber in a two-dimensional manner. The input matrix is then passed through the first convolutional neural network to generate the first feature map, wherein adjacent layers of the first convolutional neural network use convolution kernels that are transposed to each other;

The operating power of the cryogenic heat exchanger of the distillation tower at multiple predetermined time points including the current time point is passed through a context encoder including an embedding layer to obtain multiple feature vectors, and the multiple feature vectors are After two-dimensional arrangement into a feature matrix, a second convolutional neural network is used to obtain a second feature map; the gas chromatograms of the distillation products at multiple predetermined time points including the current time point are obtained by using three-dimensional convolution The third convolutional neural network of the core is used to obtain a third feature map; performing feature value correction based on category differences on each of the first to third feature maps to generate the corrected first to third feature maps; fusion The corrected first to third feature maps are used to obtain a classification feature map; and the classification feature map is passed through a classifier to obtain a classification result, and the classification result is whether the control parameter combination at the current time point is reasonable.

In the above control method of the distillation control system for the preparation of electronic grade trifluoromethane, the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point, the pressure of the molecular sieve adsorber Data and flow data of crude trifluoromethane input into the molecular sieve adsorber are arranged into a two-dimensional input matrix and then passed through the first convolutional neural network to generate the first feature map, including: using the first convolutional neural network The convolutional layer of the i-th layer uses the first convolution kernel to process the input data to generate the i-th feature map using the following formula. The formula is h=f( _wi *x+b), * represents the convolution operation, f(.) represents the activation function; and, use the convolution layer of the i+1th layer of the first convolutional neural network to use the second convolution kernel to process the i-th feature map with the following formula, The formula is h=f(wi ₊₁ *x+b), * represents the convolution operation, f(.) represents the activation function, where w _i+1 = w _i ^T .

In the above control method of the distillation control system for the preparation of electronic grade trifluoromethane, the working power of the low-temperature heat exchanger of the rectification tower at multiple predetermined time points including the current time point is passed through the embedded The context encoder of the layer obtains multiple feature vectors, and arranges the multiple feature vectors two-dimensionally into a feature matrix and then passes the second convolutional neural network to obtain the second feature map, including: using the embedding layer The embedding layer of the context encoder model converts the working power of the cryogenic heat exchanger of the distillation tower at multiple predetermined time points including the current time point into input vectors to obtain a sequence of input vectors; use the The converter of the encoder model that includes the context of the embedding layer performs globally-based context semantic encoding on the sequence of the input vectors obtained by the embedding layer sub-unit to obtain the plurality of feature vectors; converting the context encoder The plurality of feature vectors obtained by the unit are two-dimensionally arranged into a feature matrix and then passed through a second convolutional neural network to obtain a second feature map.

In the above control method of the distillation control system for the preparation of electronic grade trifluoromethane, the gas chromatograms of the distillation products at multiple predetermined time points including the current time point are passed through using a three-dimensional convolution kernel. The third convolutional neural network obtains the third feature map, including: each layer of the third convolutional neural network using the three-dimensional convolution kernel performs convolution processing and pooling processing on the input data in the forward pass of the layer. and activation processing to generate the third feature map from the last layer of the third convolutional neural network, wherein the input of the first layer of the third convolutional neural network is the current time point including the Gas chromatograms of distillation products at multiple predetermined time points.

In the above control method of the distillation control system for the preparation of electronic grade trifluoromethane, feature value correction based on category differences is performed on each of the first to third feature maps to generate the corrected first to third feature maps. Three feature maps, including: performing feature value correction based on category differences on each of the first to third feature maps using the following formula to generate the corrected first to third feature maps;

Among them, the formula is:

where f _i,j,k is the feature value of the (i,j,k)th position of the feature map, and

is the global mean of the feature values at each position of the feature map.

In the above control method of the distillation control system for the preparation of electronic grade trifluoromethane, the classification feature map is passed through a classifier to obtain a classification result. The classification result is whether the control parameter combination at the current time point is reasonable, including : The classifier processes the classification feature map with the following formula to generate a classification result, wherein the formula is: softmax{(W _n ,B _n ):...:(W ₁ ,B ₁ )|Project( F)}, where Project(F) represents projecting the classification feature map into a vector, W ₁ to W _n are the weight matrices of the fully connected layers of each layer, and B ₁ to B _n represent the bias matrices of the fully connected layers of each layer. .

Compared with the existing technology, the distillation control system and its control method for the preparation of electronic-grade trifluoromethane provided by this application use an intelligent control method based on artificial intelligence technology to conduct global optimization from the perspective of an optimized control algorithm. , high-precision, dynamic and adaptive adjustment of control parameters, thereby improving the purification purity of the electronic grade trifluoromethane.

Description of the drawings

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

Figure 1 is an application scenario diagram of a distillation control system for the preparation of electronic-grade trifluoromethane according to an embodiment of the present application.

Figure 2 is a block diagram of a distillation control system for electronic grade trifluoromethane preparation according to an embodiment of the present application.

Figure 3 is a block diagram of a second-stage distillation parameter encoding unit in a distillation control system for electronic-grade trifluoromethane preparation according to an embodiment of the present application.

Figure 4 is a flow chart of a control method of a distillation control system for electronic grade trifluoromethane preparation according to an embodiment of the present application.

Figure 5 is a schematic structural diagram of a control method of a distillation control system for the preparation of electronic-grade trifluoromethane according to an embodiment of the present application.

Detailed ways

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

Scenario overview

As mentioned before, trifluoromethane, also known as trifluoromethane, is a colorless, slightly odorless, non-conductive gas and is an ideal substitute for alkyl halides. In the semiconductor process, CHF3 is often used in plasma etching or reactive ion etching of silicon dioxide. The characteristic of CHF3 is that it corrodes silicon dioxide quickly and corrodes silicon slowly. That is, it not only has good selectivity, but also has a large rate difference. Meet the requirements of semiconductor processes. The demand for high-purity trifluoromethane as an etchant in the manufacturing process of 8-12-inch chips continues to increase with the rapid development of the semiconductor industry.

Generally, the purity of high-purity trifluoromethane used in the semiconductor industry is 99.999%. Its purification involves deep removal technology of various impurities. Trifluoromethane is highly polar and the raw materials generally contain large amounts of CHCl3, CCl2F2, CHClF2, O2, N2, and CO2. and other impurities, the boiling points of CHF3 and CO2 are very close, and the boiling points and properties of difluorochloromethane are very similar, and the two easily form an azeotrope, making separation difficult.

The purity of my country's existing industrialized trifluoromethane is relatively low, generally 99.8%-99.9%, which does not meet the requirements of the semiconductor industry. There are some purification schemes for trifluoromethane. For example, patent 103951543A uses a purification device with multi-stage distillation to improve the purity of trifluoromethane. However, this purification scheme optimized at the structural level requires the use of complex structures. Purification equipment is expensive.

Therefore, a new purification control scheme for trifluoromethane is expected so that the purity of trifluoromethane obtained by final purification can meet the purity requirements of electronic grade gas.

Most existing solutions improve the purity of trifluoromethane through multi-stage distillation or a combination of multiple purification methods. For example, in the technical solution disclosed in patent 201110423419.4, a two-stage distillation system is used, including a molecular sieve adsorber as the primary distillation and a distillation tower as the secondary distillation, in which the raw material trifluoromethane is heated at a certain temperature Under the conditions of pressure and pressure, it enters the molecular sieve adsorber (the adsorber is equipped with 3A molecular sieve) at a certain flow rate. After adsorbing CHCl3 and CCl2F2, it is introduced into a low-temperature rectification kettle, and intermittent distillation is performed between -155°C and -84°C to remove impurities such as CHCl3, O2, N2, etc., thereby obtaining more than 99.99% high-purity CHF3, using liquid nitrogen Collect in aluminum alloy containers by cryogenic method. The distillation tower relies on the condenser to provide reflux liquid cooling capacity, and relies on the low-temperature heat exchanger to exchange heat. The cold capacity of the low-temperature heat exchanger relies on the compressor, vacuum pump, and ethylene buffer tank to provide cooling capacity. High-purity trifluoromethane products After entering the product buffer tank, it enters the membrane compressor for filling.

However, this technical route will not only lead to an increase in the cost of a large amount of equipment, but also has a purification limit. The inventor of the present application found in experiments that in the multi-stage distillation scheme, when the number of distillation stages is increased to a predetermined number, the purity of trifluoromethane will basically not change and it is difficult to meet the purity requirements of electronic grade gas.

For this reason, the inventor of the present application tried to improve the purity of trifluoromethane through the technical route of optimizing the control algorithm. Taking the two-stage distillation system disclosed in patent 201110423419.4 as an example, it should be understood that during the working process of the actual molecular sieve adsorber, its internal temperature and pressure and the flow rate of trifluoromethane inflow have their preferred control strategies at different stages. , at the same time, temperature, pressure and flow are interrelated and synergistically affect the working effect of the molecular sieve adsorber. In addition, the first-stage purified product flowing out from the molecular sieve adsorber will flow into the distillation tower for distillation. Therefore, parameter control of the molecular sieve adsorber also requires parameter control of the subsequent distillation tower. Similarly, when setting the control strategy of the distillation tower, it is necessary to consider not only the control of the molecular sieve adsorber, but also the real-time production of distillation products. Therefore, for the control system of the trifluoromethane distillation system, a global, high-precision, dynamic and adaptive control system is expected.

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

The development of deep learning and neural networks provides new solutions and solutions for the control of trifluoromethane distillation systems.

Specifically, in the technical solution of the present application, the control parameters of multiple predetermined time points including the current time point are first obtained. The control parameters include the temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, The crude trifluoromethane product is input into the flow data of the molecular sieve adsorber and the working power of the low-temperature heat exchanger of the distillation tower.

Then, for the control parameters of the first-stage distillation, the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point, the pressure data of the molecular sieve adsorber, and the crude trifluoromethane product are input into the The flow data of the molecular sieve adsorber is arranged into a two-dimensional input matrix, and a convolutional neural network model is used to encode the input matrix to extract high-dimensional implications between various control parameters at the same predetermined time point in the input matrix. Correlation, high-dimensional implicit correlation between different control parameters at different predetermined time points, and high-dimensional implicit correlation between the same control parameter at different predetermined time points.

Next, for the control parameters of the two-stage distillation, a context encoder is used to encode the operating power of the low-temperature heat exchanger of the distillation tower at multiple predetermined time points including the current time point to extract each predetermined time point. The working power is relative to global contextual high-dimensional semantic features to obtain multiple feature vectors. And further, a convolutional neural network model is used to encode a feature matrix composed of the plurality of feature vectors to extract high-dimensional correlation implicit features between the working power at each predetermined time point relative to the global high-dimensional implicit correlation features. .

On the other hand, considering that the basic purpose of the rectification system is to obtain electronic-grade trifluoromethane that meets preset requirements, therefore, when controlling the parameters of the rectification system for the preparation of electronic-grade trifluoromethane, it is necessary to Taking into account the real-time production of trifluoromethane. Specifically, in the embodiment of the present application, a convolutional neural network model using a three-dimensional convolution kernel is used to encode the gas chromatograms of the distillation products at multiple predetermined time points including the current time point to extract The high-dimensional implicit features of the absolute amount of the distillation product at each predetermined time point and the high-dimensional implicit features of the relative change amount are used to obtain a third feature map.

Then, by fusing the first to third feature maps, the combination of control parameters of the distillation system can be classified and judged. However, when merging feature maps, considering that the convolution kernel of the convolutional network performs pixel-level correlation feature extraction on the source data, it will inevitably be affected by tiny numerical perturbations in the source data, thus affecting the features. The figure shows outlier feature values that deviate from the overall distribution. When merging feature maps, since the convolution kernels of the first to third convolutional neural networks are essentially small-scale local convolution kernels, these outlier feature values generally cannot be offset by fusion methods such as point addition. This affects the classification effect of the final fused feature map. Based on this, the feature map is modified and expressed as:

f _i,j,k are the eigenvalues of the (i,j,k)th position of the feature map, and

is the global mean of the feature values at each position of the feature map.

By taking the feature value at each position of the feature map as a single variable and calculating the negative logarithm of its class difference, the special distribution of feature values relative to the overall distribution can be generally classified, thereby making the outlier features as special examples The concealment of values within the overall distribution is enhanced to improve the classification ability of the fused feature map. In this way, the accuracy of classification judgment on the rationality of the control parameter combination at the current point in time of the distillation system is improved.

Based on this, this application proposes a distillation control system for the preparation of electronic grade trifluoromethane, which includes: a distillation parameter data acquisition unit used to obtain molecular sieves at multiple predetermined time points including the current time point. The temperature data of the adsorber, the pressure data of the molecular sieve adsorber, the flow data of the trifluoromethane crude product input into the molecular sieve adsorber, and the working power of the low-temperature heat exchanger of the distillation tower; a product data acquisition unit is used to obtain Gas chromatograms of distillation products at multiple predetermined time points including the current time point; a first-level distillation parameter encoding unit used to adsorb the molecular sieves at multiple predetermined time points including the current time point The temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, and the flow data of the trifluoromethane crude product input into the molecular sieve adsorber are arranged into a two-dimensional input matrix and then passed through the first convolutional neural network to generate the first feature map, Wherein, the adjacent layers of the first convolutional neural network use mutually transposed convolution kernels; the second-level distillation parameter encoding unit is used to convert the multiple predetermined time points including the current time point into The working power of the low-temperature heat exchanger of the distillation tower is passed through the context encoder containing the embedding layer to obtain multiple feature vectors, and the multiple feature vectors are two-dimensionally arranged into a feature matrix and then passed through the second convolutional neural network to obtain a second characteristic map; a product data encoding unit configured to pass the gas chromatograms of the distillation products at multiple predetermined time points including the current time point through a third convolutional neural network using a three-dimensional convolution kernel To obtain a third feature map; a feature map correction unit configured to perform feature value correction based on category differences on each of the first to third feature maps to generate corrected first to third feature maps; feature maps a fusion unit, used to fuse the corrected first to third feature maps to obtain a classification feature map; and a control result generation unit, used to pass the classification feature map through a classifier to obtain a classification result, the classification result Whether the combination of control parameters at the current point in time is reasonable.

Figure 1 illustrates an application scenario diagram of a distillation control system for the preparation of electronic-grade trifluoromethane according to an embodiment of the present application. As shown in Figure 1, in this application scenario, first, each sensor (for example, as shown in Figure 1) is deployed in the distillation control system (for example, H as shown in Figure 1) prepared by electronic grade trifluoromethane. The sensors T1-Tn shown in Figure 1 acquire the temperature data of the molecular sieve adsorber (for example, M shown in Figure 1) at multiple predetermined time points including the current time point, the temperature data of the molecular sieve adsorber. Pressure data, flow data of crude trifluoromethane into the molecular sieve adsorber, low temperature heat exchanger (for example, E as shown in Figure 1) of the distillation tower (for example, D as shown in Figure 1) working power, and obtain the distillation products at multiple predetermined time points including the current time point (for example, as shown in Figure 1 Gas chromatogram of P) shown in . Then, the obtained degree data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, the flow data of the crude trifluoromethane product into the molecular sieve adsorber, the working power of the low-temperature heat exchanger of the rectification tower, and The gas chromatogram of the distillation product is input into a server deployed with a distillation control algorithm for the preparation of electronic grade trifluoromethane (for example, the cloud server S illustrated in Figure 1), wherein the server can be used for The distillation control algorithm for the preparation of electronic grade trifluoromethane controls the degree data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, the flow data of the trifluoromethane crude product input into the molecular sieve adsorber, and the low temperature of the distillation tower. The working power of the heat exchanger and the gas chromatogram of the distillation product are processed to generate a classification result indicating whether the combination of control parameters at the current time point is reasonable. Furthermore, based on the classification results, unreasonable control parameters of the distillation system are dynamically adjusted to improve the purification purity of electronic grade trifluoromethane.

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

Example system

Figure 2 illustrates a block diagram of a distillation control system for electronic grade trifluoromethane preparation according to an embodiment of the present application. As shown in Figure 2, the distillation control system 200 for the preparation of electronic grade trifluoromethane according to the embodiment of the present application includes: a distillation parameter data acquisition unit 210, used to acquire multiple scheduled parameters including the current time point. The temperature data of the molecular sieve adsorber at the time point, the pressure data of the molecular sieve adsorber, the flow data of the trifluoromethane crude product input into the molecular sieve adsorber, the working power of the low-temperature heat exchanger of the distillation tower; the product data acquisition unit 220, used to obtain the gas chromatograms of the distillation products at multiple predetermined time points including the current time point; the first-level distillation parameter encoding unit 230 is used to encode the multiple predetermined time points including the current time point. The temperature data of the molecular sieve adsorber at a predetermined time point, the pressure data of the molecular sieve adsorber, and the flow data of the crude trifluoromethane product input into the molecular sieve adsorber are arranged into a two-dimensional input matrix and then passed through the first convolutional neural network to Generate a first feature map, in which adjacent layers of the first convolutional neural network use convolution kernels that are transposes of each other; the second-level distillation parameter encoding unit 240 is used to convert the information containing the current time point at The working power of the low-temperature heat exchanger of the distillation tower at multiple predetermined time points is passed through the context encoder containing the embedding layer to obtain multiple feature vectors, and the multiple feature vectors are two-dimensionally arranged into a feature matrix. The second feature map is obtained through the second convolutional neural network; the product data encoding unit 250 is used to convert the gas chromatograms of the distillation products at multiple predetermined time points including the current time point by using three-dimensional convolution The third convolutional neural network of the core is used to obtain the third feature map; the feature map correction unit 260 is used to perform feature value correction based on category differences on each of the first to third feature maps to generate the corrected third feature map. The first to third feature maps; the feature map fusion unit 270 is used to fuse the corrected first to third feature maps to obtain the classification feature map; and the control result generation unit 280 is used to fuse the classification feature map through The classifier obtains a classification result, which is whether the control parameter combination at the current time point is reasonable.

Specifically, in the embodiment of the present application, the distillation parameter data acquisition unit 210 and the product data acquisition unit 220 are used to acquire the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point. , the pressure data of the molecular sieve adsorber, the flow data of the trifluoromethane crude product input into the molecular sieve adsorber, the working power of the low-temperature heat exchanger of the distillation tower, and obtain multiple predetermined times including the current time point. Gas chromatogram of the distillation product at the point. As mentioned above, in the technical solution of this application, the technical route of optimizing the control algorithm is chosen to improve the purity of trifluoromethane. It should be understood that during the working process of the actual molecular sieve adsorber, the temperature and pressure set inside it and the flow rate of trifluoromethane inflow have their preferred control strategies at different stages. At the same time, the temperature, pressure and flow rate of the three They are interrelated and synergistically affect the working effect of the molecular sieve adsorber. In addition, the first-stage purification product flowing out from the molecular sieve adsorber will flow into the rectification tower for distillation. Therefore, parameter control of the molecular sieve adsorber also requires parameter control of the subsequent rectification tower. Similarly, when setting the control strategy of the distillation tower, it is necessary to consider not only the control of the molecular sieve adsorber, but also the real-time production of distillation products.

Therefore, in the technical solution of the present application, specifically, first, the control parameters of multiple predetermined time points including the current time point are acquired through each sensor deployed in the distillation control system for electronic grade trifluoromethane preparation, The control parameters include the temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, the flow data of the crude trifluoromethane product input into the molecular sieve adsorber, and the working power of the low-temperature heat exchanger of the distillation tower. And obtain gas chromatograms of the distillation product at multiple predetermined time points including the current time point through a gas chromatograph deployed in the distillation tower.

Specifically, in the embodiment of the present application, the first-stage distillation parameter encoding unit 230 and the second-stage distillation parameter encoding unit 240 are used to convert the multiple predetermined times including the current time point into The temperature data of the molecular sieve adsorber at the point, the pressure data of the molecular sieve adsorber, and the flow data of the crude trifluoromethane input into the molecular sieve adsorber are arranged into a two-dimensional input matrix and then passed through the first convolutional neural network to generate the third A feature map, wherein adjacent layers of the first convolutional neural network use mutually transposed convolution kernels, and combine the low-temperature values of the rectification tower at multiple predetermined time points including the current time point. The working power of the heat exchanger is passed through a context encoder containing an embedding layer to obtain multiple feature vectors, and the multiple feature vectors are two-dimensionally arranged into a feature matrix and then passed through a second convolutional neural network to obtain a second feature map. . That is to say, in the technical solution of the present application, then, for the control parameters of the first-stage distillation, the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point, the molecular sieve The pressure data of the adsorber and the flow data of the trifluoromethane crude product input into the molecular sieve adsorber are arranged into a two-dimensional input matrix, and the convolutional neural network model is used to encode the input matrix to extract the input matrix. The high-dimensional implicit correlation between the various control parameters at the same predetermined time point, the high-dimensional implicit correlation between different control parameters at different predetermined time points, and the high-dimensional implicit correlation between the same control parameter at different predetermined time points. Contains correlation to obtain the first feature map.

Next, for the control parameters of the two-stage rectification, the context encoder is used to encode the operating power of the low-temperature heat exchanger of the distillation tower at multiple predetermined time points including the current time point to extract The working power at each predetermined time point is compared with the global contextual high-dimensional semantic features to obtain the multiple feature vectors. And further, a convolutional neural network model is used to encode the feature matrix composed of the plurality of feature vectors to extract the high-dimensional correlation implicit between the working power at each predetermined time point relative to the global high-dimensional implicit correlation feature. Contains features to obtain the second feature map.

More specifically, in the embodiment of the present application, the first-level distillation parameter encoding unit is further configured to: use the i-th convolution layer of the first convolutional neural network to use the first convolution kernel to The following formula processes the input data to generate the i-th feature map. The formula is h=f(w _i *x+b), * represents the convolution operation, f(.) represents the activation function; and, using the i-th feature map The convolution layer of the i+1th layer of a convolutional neural network uses the second convolution kernel to process the i-th feature map according to the following formula, the formula is h=f(wi ₊₁ *x+b ), * represents the convolution operation, f(.) represents the activation function, where w _i+1 = w _i ^T .

More specifically, in the embodiment of the present application, the second-level distillation parameter encoding unit includes: first, using the embedding layer of the encoder model that includes the context of the embedding layer to respectively encode the parameters that include the current time point at The working power of the low-temperature heat exchanger of the distillation tower at multiple predetermined time points is converted into an input vector to obtain a sequence of input vectors; then, the converter of the encoder model containing the context of the embedded layer is used to The sequence of input vectors obtained by the embedding layer sub-unit performs global contextual semantic coding to obtain the multiple feature vectors; then, the multiple feature vectors obtained by the context encoding sub-unit are arranged two-dimensionally as The feature matrix is then passed through the second convolutional neural network to obtain the second feature map. It should be understood that since the encoder model using the converter can encode the input vector based on context, the obtained plurality of feature vectors have global working power-related feature information.

Figure 3 illustrates a block diagram of a second-stage distillation parameter encoding unit in a distillation control system for electronic-grade trifluoromethane preparation according to an embodiment of the present application. As shown in Figure 3, the second-level distillation parameter encoding unit 240 includes: an embedding layer sub-unit 241, which is used to use the embedding layer of the encoder model that includes the context of the embedding layer to respectively encode the embedding layer that includes the current time. The working power of the low-temperature heat exchanger of the distillation tower at multiple predetermined time points is converted into an input vector to obtain a sequence of input vectors; the context encoding subunit 242 is used to use the encoding of the context including the embedding layer The converter of the model performs global contextual semantic encoding on the sequence of input vectors obtained by the embedding layer sub-unit 241 to obtain the multiple feature vectors; the two-dimensional arrangement sub-unit 243 is used to convert the context The plurality of feature vectors obtained by the encoding subunit 242 are two-dimensionally arranged into a feature matrix and then passed through a second convolutional neural network to obtain a second feature map.

Specifically, in the embodiment of the present application, the product data encoding unit 250 is used to convert the gas chromatograms of the distillation products at multiple predetermined time points including the current time point through the use of a three-dimensional convolution kernel. The third convolutional neural network is used to obtain the third feature map. It should be understood that considering that the basic purpose of the rectification system is to obtain electronic grade trifluoromethane that meets preset requirements, therefore, when performing parameter control on the rectification system for the preparation of electronic grade trifluoromethane , the situation of the real-time product of trifluoromethane needs to be taken into account. Specifically, in the technical solution of the present application, a convolutional neural network model using the three-dimensional convolution kernel is used to encode the gas chromatograms of the distillation products at multiple predetermined time points including the current time point. , to extract the high-dimensional implicit features of the absolute amount of the distillation product at each predetermined time point and the high-dimensional implicit features of the relative change amount to obtain the third feature map.

Correspondingly, in a specific example, each layer of the third convolutional neural network using the three-dimensional convolution kernel performs convolution processing, pooling processing and activation processing on the input data in the forward pass of the layer to obtain the corresponding The last layer of the third convolutional neural network generates the third feature map, wherein the input of the first layer of the third convolutional neural network is the plurality of predetermined time points including the current time point. Gas chromatogram of the distillation product.

Specifically, in this embodiment of the present application, the feature map correction unit 260 is configured to perform feature value correction based on category differences on each of the first to third feature maps to generate the corrected first to third feature maps. Three feature maps. It should be understood that in the technical solution of the present application, by further fusing the first to third feature maps, classification judgment can be made on the combination of control parameters of the distillation system. However, when fusing the feature maps, considering that the convolution kernel of the convolutional network performs pixel-level correlation feature extraction on the source data, it will inevitably be affected by tiny numerical perturbations in the source data. , thus appearing in the feature map as outlier feature values that deviate from the overall distribution. When merging feature maps, since the convolution kernels of the first to third convolutional neural networks are essentially small-scale local convolution kernels, these outlier feature values generally cannot be fused through point addition methods, for example. offset, thus affecting the classification effect of the final fused feature map. Therefore, in the technical solution of the present application, it is necessary to further modify each of the first to third feature maps.

More specifically, in the embodiment of the present application, the feature map correction unit is further configured to perform feature value correction based on category differences on each of the first to third feature maps using the following formula to generate a correction The first to third feature maps described later;

Among them, the formula is:

where f _i,j,k is the feature value of the (i,j,k)th position of the feature map, and

is the global mean of the feature values at each position of the feature map. It should be understood that by taking the feature value at each position of the feature map as a single variable and calculating the negative logarithm of its category difference, the special distribution of the feature value relative to the overall distribution can be generally classified, thereby This enhances the concealment of outlier feature values as special examples within the overall distribution to improve the classification ability of the fused feature map. In this way, the accuracy of classification judgment on the rationality of the control parameter combination at the current point in time of the distillation system can be improved.

Specifically, in this embodiment of the present application, the feature map fusion unit 270 and the control result generation unit 280 are used to fuse the corrected first to third feature maps to obtain a classification feature map, and combine the The classification feature map is passed through the classifier to obtain a classification result, which is whether the combination of control parameters at the current time point is reasonable. That is to say, in the technical solution of the present application, the corrected first to third feature maps can be further fused to obtain a classification feature map for classification processing, thereby obtaining the control for representing the current time point. Classification results of whether the parameter combination is reasonable. Correspondingly, in a specific example, the classifier processes the classification feature map according to the following formula to generate a classification result, wherein the formula is: softmax{(W _n ,B _n ):...:(W ₁ ,B ₁ )|Project(F)}, where Project(F) represents projecting the classification feature map into a vector, W ₁ to W _n are the weight matrices of the fully connected layers of each layer, and B ₁ to B _n represent each The bias matrix of the fully connected layer.

In summary, the distillation control system 200 for the preparation of electronic grade trifluoromethane is clarified based on the embodiment of the present application, which uses an intelligent control method based on artificial intelligence technology to conduct global, The control parameters are adjusted with high precision, dynamically and adaptively, thereby improving the purification purity of the electronic grade trifluoromethane.

As mentioned above, the distillation control system 200 for the preparation of electronic-grade trifluoromethane according to the embodiment of the present application can be implemented in various terminal devices, such as a server for the distillation control algorithm of the preparation of electronic-grade trifluoromethane, etc. . In one example, the distillation control system 200 for the preparation of electronic grade trifluoromethane according to the embodiment of the present application can be integrated into the terminal device as a software module and/or a hardware module. For example, the distillation control system 200 for the preparation of electronic grade trifluoromethane can be a software module in the operating system of the terminal equipment, or can be an application program developed for the terminal equipment; of course, the user The distillation control system 200 for the preparation of electronic grade trifluoromethane can also be one of the many hardware modules of the terminal equipment.

Alternatively, in another example, the rectification control system 200 for the preparation of electronic grade trifluoromethane and the terminal equipment can also be separate devices, and the rectification control system for the preparation of electronic grade trifluoromethane 200 can be connected to the terminal device through a wired and/or wireless network, and transmit interactive information according to an agreed data format.

Example methods

Figure 4 illustrates a flow chart of a control method of a distillation control system for electronic grade trifluoromethane production. As shown in Figure 4, the control method of the distillation control system for the preparation of electronic grade trifluoromethane according to the embodiment of the present application includes step: S110, obtaining molecular sieve adsorption data at multiple predetermined time points including the current time point. The temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, the flow data of the trifluoromethane crude product input into the molecular sieve adsorber, and the working power of the low-temperature heat exchanger of the distillation tower; S120, obtain information including the current time point. The gas chromatograms of the distillation products at multiple predetermined time points; S130, combine the temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, and the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point. The flow data of crude fluoromethane input into the molecular sieve adsorber is arranged into a two-dimensional input matrix and then passed through the first convolutional neural network to generate the first feature map, where the adjacent layers of the first convolutional neural network use Convolution kernels that are transposed to each other; S140, pass the operating power of the low-temperature heat exchanger of the distillation tower at multiple predetermined time points including the current time point through a context encoder including an embedding layer to obtain multiple feature vectors, and arrange the multiple feature vectors into a feature matrix in two dimensions and then pass the second convolutional neural network to obtain the second feature map; S150, combine the multiple predetermined time points including the current time point The gas chromatogram of the distillation product is obtained through a third convolutional neural network using a three-dimensional convolution kernel to obtain a third feature map; S160, perform category difference-based features on each of the first to third feature maps. value correction to generate the corrected first to third feature maps; S170, fuse the corrected first to third feature maps to obtain a classification feature map; and, S180, pass the classification feature map through a classifier to obtain a classification As a result, the classification result is whether the control parameter combination at the current time point is reasonable.

FIG. 5 illustrates an architectural schematic diagram of a control method of a distillation control system for electronic-grade trifluoromethane preparation according to an embodiment of the present application. As shown in Figure 5, in the network architecture of the control method of the distillation control system for the preparation of electronic grade trifluoromethane, first, the obtained multiple predetermined time points including the current time point are obtained. The temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, and the flow data of the crude trifluoromethane product input into the molecular sieve adsorber (for example, P1 as shown in Figure 5) are arranged into a two-dimensional input matrix ( For example, M) as shown in Figure 5 is then passed through a first convolutional neural network (for example, CNN1 as shown in Figure 5) to generate a first feature map (for example, F1 as shown in Figure 5); Next, the operating power of the cryogenic heat exchanger of the distillation tower at multiple predetermined time points including the current time point (for example, P2 as shown in Figure 5) is passed through a context encoder including an embedding layer ( For example, E1 as shown in Figure 5) to obtain multiple feature vectors (for example, VF1 as shown in Figure 5), and the multiple feature vectors are two-dimensionally arranged into a feature matrix (for example, as shown in Figure MF1 as shown in Figure 5) is then passed through the second convolutional neural network (for example, CNN2 as shown in Figure 5) to obtain the second feature map (for example, F2 as shown in Figure 5); then, the The gas chromatograms of the distillation products at multiple predetermined time points including the current time point (for example, Q as shown in Figure 5) are passed through a third convolutional neural network using a three-dimensional convolution kernel (for example, as CNN3 as shown in Figure 5) to obtain a third feature map (for example, F3 as shown in Figure 5); then, perform feature value based on class difference on each of the first to third feature maps. Correction is performed to generate corrected first to third feature maps (for example, F4, F5 and F6 as illustrated in Figure 5); then, the corrected first to third feature maps are fused to obtain classification feature maps (for example, F4, F5 and F6 as illustrated in Figure 5) , F) as shown in Figure 5; and, finally, pass the classification feature map through a classifier (for example, the classifier as shown in Figure 5) to obtain a classification result, which is the current time point Whether the combination of control parameters is reasonable.

More specifically, in steps S110 and S120, the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point, the pressure data of the molecular sieve adsorber, and the input of crude trifluoromethane into the molecular sieve are obtained. The flow data of the adsorber, the working power of the low-temperature heat exchanger of the distillation tower, and the gas chromatograms of the distillation products at multiple predetermined time points including the current time point are obtained. It should be understood that in the technical solution of the present application, the technical route of optimizing the control algorithm is chosen to improve the purity of trifluoromethane. It should be understood that during the working process of the actual molecular sieve adsorber, the temperature and pressure set inside it and the flow rate of trifluoromethane inflow have their preferred control strategies at different stages. At the same time, the temperature, pressure and flow rate of the three They are interrelated and synergistically affect the working effect of the molecular sieve adsorber. In addition, the first-stage purification product flowing out from the molecular sieve adsorber will flow into the rectification tower for distillation. Therefore, parameter control of the molecular sieve adsorber also requires parameter control of the subsequent rectification tower. Similarly, when setting the control strategy of the distillation tower, it is necessary to consider not only the control of the molecular sieve adsorber, but also the real-time production of distillation products.

Therefore, in the technical solution of the present application, specifically, first, the control parameters of multiple predetermined time points including the current time point are acquired through each sensor deployed in the distillation control system for electronic grade trifluoromethane preparation, The control parameters include the temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, the flow data of the crude trifluoromethane product input into the molecular sieve adsorber, and the working power of the low-temperature heat exchanger of the distillation tower. And obtain gas chromatograms of the distillation product at multiple predetermined time points including the current time point through a gas chromatograph deployed in the distillation tower.

More specifically, in steps S130 and S140, the temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, and the crude trifluoromethane product at multiple predetermined time points including the current time point are input. The flow data of the molecular sieve adsorber is arranged into a two-dimensional input matrix and then passed through a first convolutional neural network to generate a first feature map, where adjacent layers of the first convolutional neural network use mutually transposed Convolution kernel, and pass the working power of the cryogenic heat exchanger of the distillation tower at multiple predetermined time points including the current time point through the context encoder including the embedding layer to obtain multiple feature vectors, and put the The plurality of feature vectors are two-dimensionally arranged into a feature matrix and then passed through a second convolutional neural network to obtain a second feature map. That is to say, in the technical solution of the present application, then, for the control parameters of the first-stage distillation, the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point, the molecular sieve The pressure data of the adsorber and the flow data of the trifluoromethane crude product input into the molecular sieve adsorber are arranged into a two-dimensional input matrix, and the convolutional neural network model is used to encode the input matrix to extract the input matrix. The high-dimensional implicit correlation between the various control parameters at the same predetermined time point, the high-dimensional implicit correlation between different control parameters at different predetermined time points, and the high-dimensional implicit correlation between the same control parameter at different predetermined time points. Contains correlation to obtain the first feature map.

Next, for the control parameters of the two-stage rectification, the context encoder is used to encode the operating power of the low-temperature heat exchanger of the distillation tower at multiple predetermined time points including the current time point to extract The working power at each predetermined time point is compared with the global contextual high-dimensional semantic features to obtain the multiple feature vectors. And further, a convolutional neural network model is used to encode the feature matrix composed of the plurality of feature vectors to extract the high-dimensional correlation implicit between the working power at each predetermined time point relative to the global high-dimensional implicit correlation feature. Contains features to obtain the second feature map.

More specifically, in step S150, the gas chromatograms of the distillation products at multiple predetermined time points including the current time point are passed through a third convolutional neural network using a three-dimensional convolution kernel to obtain the third feature. picture. It should be understood that considering that the basic purpose of the rectification system is to obtain electronic grade trifluoromethane that meets preset requirements, therefore, when performing parameter control on the rectification system for the preparation of electronic grade trifluoromethane , the situation of the real-time product of trifluoromethane needs to be taken into account. Specifically, in the technical solution of the present application, a convolutional neural network model using the three-dimensional convolution kernel is used to encode the gas chromatograms of the distillation products at multiple predetermined time points including the current time point. , to extract the high-dimensional implicit features of the absolute amount of the distillation product at each predetermined time point and the high-dimensional implicit features of the relative change amount to obtain the third feature map.

Correspondingly, in a specific example, each layer of the third convolutional neural network using the three-dimensional convolution kernel performs convolution processing, pooling processing and activation processing on the input data in the forward pass of the layer to obtain the corresponding The last layer of the third convolutional neural network generates the third feature map, wherein the input of the first layer of the third convolutional neural network is the plurality of predetermined time points including the current time point. Gas chromatogram of the distillation product.

More specifically, in step S160, feature value correction based on category differences is performed on each of the first to third feature maps to generate corrected first to third feature maps. It should be understood that in the technical solution of the present application, by further fusing the first to third feature maps, classification judgment can be made on the combination of control parameters of the distillation system. However, when fusing the feature maps, considering that the convolution kernel of the convolutional network performs pixel-level correlation feature extraction on the source data, it will inevitably be affected by tiny numerical perturbations in the source data. , thus appearing in the feature map as outlier feature values that deviate from the overall distribution. When merging feature maps, since the convolution kernels of the first to third convolutional neural networks are essentially small-scale local convolution kernels, these outlier feature values generally cannot be fused through point addition methods, for example. offset, thus affecting the classification effect of the final fused feature map. Therefore, in the technical solution of the present application, it is necessary to further modify each of the first to third feature maps.

More specifically, in steps S170 and S180, the corrected first to third feature maps are fused to obtain a classification feature map, and the classification feature map is passed through a classifier to obtain a classification result, and the classification result is Whether the combination of control parameters at the current time point is reasonable. That is to say, in the technical solution of the present application, the corrected first to third feature maps can be further fused to obtain a classification feature map for classification processing, thereby obtaining the control for representing the current time point. Classification results of whether the parameter combination is reasonable. Correspondingly, in a specific example, the classifier processes the classification feature map according to the following formula to generate a classification result, wherein the formula is: softmax{(W _n ,B _n ):...:(W ₁ ,B ₁ )}Project(F)}, where Project(F) represents projecting the classification feature map into a vector, W ₁ to W _n are the weight matrices of the fully connected layers of each layer, and B ₁ to B _n represent each The bias matrix of the fully connected layer.

In summary, based on the embodiments of the present application, the control method of the distillation control system for the preparation of electronic grade trifluoromethane has been clarified, which uses an intelligent control method based on artificial intelligence technology to conduct global optimization from the perspective of an optimized control algorithm. High-precision, dynamic and adaptive adjustment of control parameters, thereby improving the purification purity of the electronic grade trifluoromethane.

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

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

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

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

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

Claims (10)

1. A distillation control system for the preparation of electronic grade trifluoromethane, which is characterized by including:

   Distillation parameter data acquisition unit, used to acquire the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point, the pressure data of the molecular sieve adsorber, and the crude trifluoromethane product to be input into the molecular sieve adsorber. The flow data and the working power of the low-temperature heat exchanger of the distillation tower; the product data acquisition unit is used to obtain the gas chromatograms of the distillation products at multiple predetermined time points including the current time point; the first-stage distillation Parameter encoding unit, used to input the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point, the pressure data of the molecular sieve adsorber, and the crude trifluoromethane product into the molecular sieve adsorber. The flow data is arranged into a two-dimensional input matrix and then passed through a first convolutional neural network to generate a first feature map, wherein adjacent layers of the first convolutional neural network use convolution kernels that are transposed to each other; second A stage rectification parameter encoding unit used to pass the operating power of the low-temperature heat exchanger of the distillation tower at multiple predetermined time points including the current time point through a context encoder including an embedding layer to obtain multiple feature vectors , and arrange the multiple feature vectors into a feature matrix in two dimensions and then pass the second convolutional neural network to obtain the second feature map; the product data encoding unit is used to convert the multiple feature vectors including the current time point into The gas chromatogram of the distillation product at a predetermined time point is obtained through a third convolutional neural network using a three-dimensional convolution kernel; a feature map correction unit is used to correct each of the first to third feature maps The feature map performs feature value correction based on category differences to generate corrected first to third feature maps; a feature map fusion unit used to fuse the corrected first to third feature maps to obtain a classification feature map; and control results A generation unit is used to pass the classification feature map through a classifier to obtain a classification result. The classification result is whether the control parameter combination at the current time point is reasonable.
2. The distillation control system for the preparation of electronic grade trifluoromethane according to claim 1, wherein the first-stage distillation parameter encoding unit is further used to: use the i-th parameter of the first convolutional neural network The convolutional layer of the layer uses the first convolution kernel to process the input data to generate the i-th feature map with the following formula, the formula is h=f(w _i *x+b), * represents the convolution operation, f( .) represents the activation function; and, use the convolution layer of the i+1th layer of the first convolutional neural network to use the second convolution kernel to process the i-th feature map with the following formula, the formula is h=f(wi ₊₁ *x+b), * represents the convolution operation, f(.) represents the activation function, where w _i+1 = _wi ^T .
3. The distillation control system for the preparation of electronic grade trifluoromethane according to claim 2, wherein the second-stage distillation parameter encoding unit includes: an embedded layer subunit for using the embedded layer-containing The embedding layer of the context encoder model converts the working power of the cryogenic heat exchanger of the distillation tower at multiple predetermined time points including the current time point into input vectors to obtain a sequence of input vectors; the context encoder A unit configured to perform global context-based semantic encoding on the sequence of input vectors obtained by the embedding layer sub-unit using the converter of the encoder model containing the context of the embedding layer to obtain the plurality of feature vectors; A two-dimensional arrangement subunit, configured to two-dimensionally arrange the plurality of feature vectors obtained by the context encoding subunit into a feature matrix and then pass it through a second convolutional neural network to obtain a second feature map.
4. The distillation control system for the preparation of electronic grade trifluoromethane according to claim 3, wherein the product data encoding unit is further used to: use each of the third convolutional neural network of the three-dimensional convolution kernel. The layer performs convolution, pooling and activation on the input data in a forward pass of the layer to generate the third feature map by the last layer of the third convolutional neural network, wherein the third The input of the first layer of the convolutional neural network is the gas chromatogram of the distillation product at multiple predetermined time points including the current time point.
5. The distillation control system for the preparation of electronic grade trifluoromethane according to claim 4, wherein the characteristic map correction unit is further used to: calculate each characteristic in the first to third characteristic maps according to the following formula: The map performs feature value correction based on category differences to generate the corrected first to third feature maps; wherein, the formula is:

   

   where f _i,j,k is the feature value of the (i,j,k)th position of the feature map, and

   

   is the global mean of the feature values at each position of the feature map.
6. The distillation control system for the preparation of electronic grade trifluoromethane according to claim 5, wherein the control result generating unit is further configured to: the classifier processes the classification feature map according to the following formula to Generate classification results, where the formula is: softmax{(W _n ,B _n ):...:(W ₁ ,B ₁ )|Project(F)}, where Project(F) represents the projection of the classification feature map is a vector, W ₁ to W _n are the weight matrices of the fully connected layers of each layer, and B ₁ to B _n represent the bias matrices of the fully connected layers of each layer.
7. A control method for a distillation control system for the preparation of electronic grade trifluoromethane, which is characterized in that it includes: obtaining the temperature data of the molecular sieve adsorber at multiple predetermined time points including the current time point, the molecular sieve adsorption The pressure data of the device, the flow data of the trifluoromethane crude product input into the molecular sieve adsorber, and the working power of the low-temperature heat exchanger of the distillation tower; obtain the information of the distillation products at multiple predetermined time points including the current time point. Gas chromatogram; input the temperature data of the molecular sieve adsorber, the pressure data of the molecular sieve adsorber, and the crude trifluoromethane product at multiple predetermined time points including the current time point into the flow data of the molecular sieve adsorber. After being arranged into a two-dimensional input matrix, it is passed through a first convolutional neural network to generate a first feature map, wherein adjacent layers of the first convolutional neural network use convolution kernels that are transposes of each other; The operating power of the low-temperature heat exchanger of the distillation tower at multiple predetermined time points including the current time point is passed through a context encoder including an embedding layer to obtain multiple feature vectors, and the multiple feature vectors are arranged in a two-dimensional manner. After being the feature matrix, the second convolutional neural network is used to obtain the second feature map; the gas chromatograms of the distillation products at multiple predetermined time points including the current time point are passed through the third step using a three-dimensional convolution kernel. Convolutional neural network to obtain a third feature map; perform feature value correction based on category differences on each of the first to third feature maps to generate the corrected first to third feature maps; fuse the corrected features The first to third feature maps are used to obtain a classification feature map; and the classification feature map is passed through a classifier to obtain a classification result. The classification result is whether the control parameter combination at the current time point is reasonable.
8. The control method of a distillation control system for the preparation of electronic grade trifluoromethane according to claim 7, wherein the low-temperature heat exchanger of the rectification tower at a plurality of predetermined time points including the current time point is The working power is passed through the context encoder containing the embedding layer to obtain multiple feature vectors, and the multiple feature vectors are two-dimensionally arranged into a feature matrix and then passed through the second convolutional neural network to obtain the second feature map, including: The working power of the low-temperature heat exchanger of the distillation tower at multiple predetermined time points including the current time point is converted into an input vector using the embedding layer of the encoder model containing the context of the embedding layer to obtain the input. a sequence of vectors; using the converter of the encoder model containing the context of the embedding layer to perform global-based context semantic encoding on the sequence of input vectors obtained by the embedding layer subunit to obtain the plurality of feature vectors; The plurality of feature vectors obtained by the context encoding subunit are two-dimensionally arranged into a feature matrix and then passed through a second convolutional neural network to obtain a second feature map.
9. The control method of a distillation control system for the preparation of electronic grade trifluoromethane according to claim 8, wherein the gas chromatograms of the distillation products at multiple predetermined time points including the current time point are passed through Using a third convolutional neural network with a three-dimensional convolution kernel to obtain a third feature map includes: using each layer of the third convolutional neural network with the three-dimensional convolution kernel to convolve the input data in a forward pass of the layer. product processing, pooling processing and activation processing to generate the third feature map from the last layer of the third convolutional neural network, wherein the input of the first layer of the third convolutional neural network is the Gas chromatograms of distillation products at multiple predetermined time points including the current time point.
10. The control method of a distillation control system for the preparation of electronic grade trifluoromethane according to claim 9, wherein feature value correction based on category differences is performed on each of the first to third feature maps to generate The corrected first to third feature maps include: performing feature value correction based on category differences on each of the first to third feature maps using the following formula to generate the corrected first to third feature maps ;wherein, the formula is:

    

    where f _i,j,k is the feature value of the (i,j,k)th position of the feature map, and

    

    is the global mean of the feature values at each position of the feature map.

Images