WO2024026993A1

WO2024026993A1 - Automatic batching control system for ammonium fluoride preparation and control method thereof

Info

Publication number: WO2024026993A1
Application number: PCT/CN2022/119586
Authority: WO
Inventors: 丘添明; 邱汉林; 廖鸿辉; 陈三凤; 蓝丽萍; 罗丽华; 陈蜂
Original assignee: 福建省龙氟新材料有限公司
Priority date: 2022-08-05
Filing date: 2022-09-19
Publication date: 2024-02-08
Also published as: CN115309215A; CN115309215B

Abstract

An automatic batching control system for ammonium fluoride preparation and a control method thereof. The present application uses artificial intelligence control technology. Starting from flow velocity values of ammonia, flow velocity values of anhydrous hydrogen fluoride, reaction temperature values, and pH values of a reaction solution, association information of implicit features of each piece of data in a time series dimension is mined by using a deep neural network model, and associated feature information of said piece of data is fused by using a Bayesian model so as to perform real-time dynamic control on the flow velocities at which the ammonia and the anhydrous hydrogen fluoride are added into a reaction tank, thereby improving the reaction efficiency and the product quality.

Description

Automatic batching control system and control method for ammonium fluoride preparation

Technical field

The present invention relates to the field of intelligent manufacturing, and more specifically, to an automatic batching control system for ammonium fluoride preparation and a control method thereof.

Background technique

Ammonium fluoride, the molecular formula is NH4F, the relative molecular mass is 37.04, the relative density is 1.015 (25℃), colorless leaf-like or needle-like crystals, after sublimation, it becomes hexagonal columnar crystals; easy to deliquesce and agglomerate, soluble in cold water, slightly Soluble in alcohol, insoluble in acetone and liquid ammonia. When heated or exposed to hot water, it decomposes and loses ammonia and converts it into more stable ammonium fluoride. Ammonium fluoride is widely used, such as as a glass etching agent, a chemical polishing agent for metal surfaces, a wood and wine preservative, a disinfectant, a mordant for fibers, and a solvent for extracting rare elements. It can also be used as a component for ion detection in chemical analysis. Masking agent, disinfectant for wine making, preservative, fiber mordant, etc.

The traditional production method of ammonium fluoride is the liquid phase method: put a certain amount of hydrofluoric acid into a lead or plastic container. Cool with water outside the container, and slowly introduce ammonia gas while stirring until the pH value of the reaction solution reaches about 4. The reaction solution is cooled and crystallized, centrifuged, and air-dried to obtain ammonium fluoride product. Ammonium fluoride produced by the traditional liquid phase method has shortcomings such as high water content, easy caking, and inability to be stored for a long time.

Therefore, an optimized preparation scheme for ammonium fluoride is expected.

Contents of the invention

In order to solve the above technical problems, this application is proposed. The embodiment of the present application provides an automatic batching control system and a control method for ammonium fluoride preparation, which uses artificial intelligence control technology to determine the flow rate value of liquid ammonia, the flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and Starting from the pH value of the reaction solution, a deep neural network model is used to dig out the implicit feature correlation information of each data in the time series dimension, and Bayesian is used to fuse the correlation feature information of each data to compare the relationship between the liquid ammonia and all the data. The flow rate of the anhydrous hydrogen fluoride added into the reaction tank is dynamically controlled in real time, thereby improving the efficiency of the reaction and the quality of the product.

According to one aspect of the present application, an automatic batching control system for ammonium fluoride preparation is provided, which includes:

The data acquisition module is used to obtain the first flow rate value of liquid ammonia, the second flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and the pH value of the reaction liquid at multiple predetermined time points within a predetermined time period; the batching speed structured correlation module, For arranging the first flow rate values of liquid ammonia and the second flow rate values of anhydrous hydrogen fluoride at multiple predetermined time points within the predetermined time period into first flow rate vectors and second flow rate vectors, calculating the first The product between the transpose vector of the flow velocity vector and the second flow velocity vector is used to obtain the flow velocity control matrix; the batching velocity characteristic filtering module is used to pass the flow velocity control matrix through the first convolutional neural network as a filter to obtain Flow rate control feature vector; time series encoding module, used to pass the reaction temperature values and the pH value of the reaction liquid at multiple predetermined time points within the predetermined time period through a time series encoder including a one-dimensional convolution layer to obtain the reaction temperature feature vector and PH timing feature vector; a feature correction module, configured to correct the eigenvalues of each position in the flow rate control feature vector based on the reaction temperature feature vector and the PH timing feature vector to obtain a corrected flow rate control feature vector; A Bayesian fusion module for using a Bayesian probability model to fuse the corrected flow rate control feature vector, the reaction temperature feature vector and the PH timing feature vector to obtain a posterior feature vector; and

A batching control result generation module is used to pass the posterior feature vector through a classifier to obtain a classification result. The classification result is used to indicate that the first flow rate value of liquid ammonia at the current time point should be increased or should be reduced and there is no The second flow rate value of water hydrogen fluoride should be increased or should be decreased.

In the above-mentioned automatic batching control system for the preparation of ammonium fluoride, the batching speed characteristic filtering module is further used: each layer of the first convolutional neural network performs on the input data respectively in the forward transmission of the layer: Perform convolution processing on the input data to obtain a convolution feature map; perform mean pooling based on the local feature matrix on the convolution feature map to obtain a pooled feature map; and perform nonlinear activation on the pooled feature map. To obtain the activation feature map; wherein, the output of the last layer of the first convolutional neural network is the flow rate control feature vector, and the input of the first layer of the first convolutional neural network is the flow rate control matrix. .

In the above-mentioned automatic batching control system for ammonium fluoride preparation, the time sequence encoding module includes: a temperature time sequence encoding unit, used to arrange the reaction temperature values at multiple predetermined time points within the predetermined time period according to the time dimension as Temperature input vector; use the fully connected layer of the temporal encoder to fully connect the temperature input vector with the following formula to extract the high-dimensional hidden features of the eigenvalues of each position in the temperature input vector, where, The formula is:

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

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

Among them, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix that operates with the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector; PH timing encoding unit, used to arrange the PH values of the reaction solution at multiple predetermined time points within the predetermined time period into a PH value input vector according to the time dimension; use the fully connected layer of the timing encoder to calculate the The PH value input vector is fully connected to extract the high-dimensional hidden features of the eigenvalues at each position in the PH value input vector, where the formula is:

represents matrix multiplication; use the one-dimensional convolution layer of the temporal encoder to perform one-dimensional convolution encoding on the PH value input vector with the following formula to extract the high-dimensional hidden between the eigenvalues of each position in the PH value input vector. Contains associated features, where the formula is:

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

In the above-mentioned automatic batching control system for the preparation of ammonium fluoride, the feature correction module is further used to control the feature values of each position in the feature vector based on predetermined hyperparameters and the flow rate and each position in the reaction temperature feature vector. The greater of the difference between the first distance between the characteristic values and the first distance, and the characteristic value of each position in the predetermined hyperparameter and the flow rate control characteristic vector and the The sum value between the difference between the second distances between the characteristic values of each position in the PH time series feature vector and the larger of the second distances, for each position in the flow rate control feature vector The characteristic values are corrected to obtain the corrected flow rate control characteristic vector.

In the above-mentioned automatic batching control system for the preparation of ammonium fluoride, the characteristic correction module is further used to: based on the reaction temperature characteristic vector and the pH timing characteristic vector, calculate each position in the flow rate control characteristic vector according to the following formula The eigenvalues are corrected to obtain the corrected flow rate control eigenvector; wherein, the formula is:

Where f ₁ , f ₂ and f ₃ are respectively the eigenvalues of the corresponding positions of the reaction temperature eigenvector, the PH timing eigenvector and the flow rate control eigenvector normalized to the [0,1] interval, d(f ₃ , f ₁ ) represents the first distance between the eigenvalues of each position in the flow rate control eigenvector and the eigenvalues of each position in the reaction temperature eigenvector, d(f ₃ , f ₂ ) represents the second distance between the eigenvalues of each position in the flow rate control eigenvector and the eigenvalues of each position in the PH timing feature vector, and ρ is the predetermined hyperparameter.

In the above-mentioned automatic batching control system for the preparation of ammonium fluoride, the Bayesian fusion module is further used to: use the Bayesian probability model to fuse the corrected flow rate control feature vector and the reaction temperature using the following formula: Feature vector and the PH time series feature vector to obtain the posterior feature vector; wherein, the formula is:

qi＝pi*ai/bi

Wherein, pi is the characteristic value of each position in the corrected flow rate control characteristic vector, ai and bi are the characteristic values of each position in the reaction temperature characteristic vector and the pH timing characteristic vector respectively, and qi is the characteristic value of each position in the corrected flow rate control characteristic vector. The eigenvalues of each position in the posterior eigenvector.

In the above-mentioned automatic batching control system for ammonium fluoride preparation, the batching control result generation module is further used to use the classifier to process the posterior feature vector with the following formula to obtain the classification result, where , the formula is: softmax{(W _n ,B _n ):...:(W ₁ ,B ₁ )|X}, where W ₁ to W _n are weight matrices, B ₁ to B _n are bias vectors, X is the posterior feature vector.

According to another aspect of the present application, a control method for an automatic batching control system for ammonium fluoride preparation, which includes:

Obtain the first flow rate value of liquid ammonia, the second flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and the pH value of the reaction liquid at multiple predetermined time points within the predetermined time period; After the first flow rate value of liquid ammonia and the second flow rate value of anhydrous hydrogen fluoride are arranged into the first flow rate vector and the second flow rate vector respectively, calculate the relationship between the transpose vector of the first flow rate vector and the second flow rate vector. The product of The liquid PH value passes through a time series encoder containing a one-dimensional convolution layer to obtain a reaction temperature feature vector and a PH time series feature vector; based on the reaction temperature feature vector and PH time series feature vector, each position in the flow rate control feature vector is The eigenvalues are corrected to obtain the corrected flow rate control eigenvector; a Bayesian probability model is used to fuse the corrected flow rate control eigenvector, the reaction temperature eigenvector and the PH timing eigenvector to obtain a posteriori feature vector ;as well as

The posterior feature vector is passed through a classifier to obtain a classification result, which is used to indicate that the first flow rate value of liquid ammonia at the current time point should be increased or decreased and the second flow rate value of anhydrous hydrogen fluoride should be increase or should decrease.

In the above control method of the automatic batching control system for ammonium fluoride preparation, the flow rate control matrix is passed through the first convolutional neural network as a filter to obtain the flow rate control feature vector, including: the first convolutional neural network Each layer of the network processes the input data separately in the forward pass of the layer: performs convolution processing on the input data to obtain the convolution feature map; performs mean pooling based on the local feature matrix on the convolution feature map to obtain the pool feature map; and perform nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the first convolutional neural network is the flow rate control feature vector, and the third The input to the first layer of a convolutional neural network is the flow rate control matrix.

In the control method of the above-mentioned automatic batching control system for ammonium fluoride preparation, the reaction temperature values and the pH values of the reaction liquid at multiple predetermined time points within the predetermined time period are passed through a time series encoder including a one-dimensional convolution layer. To obtain the reaction temperature feature vector and the PH timing feature vector, the method includes: arranging the reaction temperature values at multiple predetermined time points within the predetermined time period into a temperature input vector according to the time dimension; using the fully connected layer of the timing encoder to The following formula performs fully connected encoding on the temperature input vector to extract high-dimensional implicit features of the eigenvalues of each position in the temperature input vector, where the formula is:

Among them, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix that operates with the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector; Arrange the PH values of the reaction solution at multiple predetermined time points within the predetermined time period into a PH value input vector according to the time dimension; use the fully connected layer of the timing encoder to fully connect the PH value input vector with the following formula Encoding to extract high-dimensional hidden features of the eigenvalues at each position in the PH value input vector, where the formula is:

In the above control method of the automatic batching control system for ammonium fluoride preparation, based on the reaction temperature characteristic vector and the pH timing characteristic vector, the characteristic values of each position in the flow rate control characteristic vector are corrected to obtain the corrected flow rate Controlling a feature vector, including: a difference between a predetermined hyperparameter and a first distance between a feature value of each position in the flow rate control feature vector and a first distance between a feature value of each position in the reaction temperature feature vector and the The larger of the first distance, and the second distance between the predetermined hyperparameter and the characteristic value of each position in the flow rate control characteristic vector and the characteristic value of each position in the PH timing characteristic vector. The sum value between the larger difference between the two distances and the second distance is corrected to obtain the corrected flow speed control feature vector.

In the above control method of the automatic batching control system for ammonium fluoride preparation, based on the reaction temperature characteristic vector and the pH timing characteristic vector, the characteristic values of each position in the flow rate control characteristic vector are corrected to obtain the corrected flow rate Controlling the feature vector includes: based on the reaction temperature feature vector and the pH timing feature vector, correcting the feature values of each position in the flow rate control feature vector with the following formula to obtain the corrected flow rate control feature vector; wherein, The formula is:

In the above control method of the automatic batching control system for ammonium fluoride preparation, a Bayesian probability model is used to fuse the corrected flow rate control feature vector, the reaction temperature feature vector and the pH timing feature vector to obtain the following A posteriori feature vector, including: using a Bayesian probability model to fuse the corrected flow rate control feature vector, the reaction temperature feature vector and the PH timing feature vector with the following formula to obtain the posterior feature vector;

Among them, the formula is:

qi＝pi*ai/bi

In the above-mentioned control method of the automatic batching control system for ammonium fluoride preparation, passing the posterior feature vector through a classifier to obtain a classification result includes: using the classifier to perform the following calculation on the posterior feature vector: Process to obtain the classification result, where the formula is: softmax{(W _n ,B _n ):...:(W ₁ ,B ₁ )|X}, where W ₁ to W _n are weight matrices, B ₁ to B _n are bias vectors, and X is the posterior feature vector.

According to yet another aspect of the present application, a computer-readable medium is provided with computer program instructions stored thereon. The computer program instructions, when executed by a processor, cause the processor to perform the preparation of ammonium fluoride as described above. The control method of the automatic batching control system used.

Compared with the existing technology, the automatic batching control system and its control method for the preparation of ammonium fluoride provided by this application use artificial intelligence control technology to determine the flow rate value of liquid ammonia, the flow rate value of anhydrous hydrogen fluoride, and the reaction temperature. Starting from the pH value of the reaction solution and the pH value of the reaction solution, a deep neural network model is used to dig out the implicit feature correlation information of each data in the time series dimension, and Bayesian is used to fuse the correlation feature information of each data to obtain the liquid ammonia. The flow rate of the anhydrous hydrogen fluoride added into the reaction tank is dynamically controlled in real time, thereby improving the efficiency of the reaction and the quality of the product.

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 1A is a flow chart of the preparation process of an automatic batching control system for ammonium fluoride preparation according to an embodiment of the present application.

Figure 1B is an application scenario diagram of an automatic batching control system for ammonium fluoride preparation according to an embodiment of the present application.

Figure 2 is a block diagram of an automatic batching control system for ammonium fluoride preparation according to an embodiment of the present application.

Figure 3 is a block diagram of the timing coding module in the automatic batching control system for ammonium fluoride preparation according to an embodiment of the present application.

Figure 4 is a flow chart of a control method of an automatic batching control system for ammonium fluoride preparation according to an embodiment of the present application.

Figure 5 is a schematic structural diagram of a control method of an automatic batching control system for ammonium fluoride preparation 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, ammonium fluoride has a molecular formula of NH4F, a relative molecular mass of 37.04, and a relative density of 1.015 (25°C). It is colorless leaf-like or needle-like crystals and turns into hexagonal columnar crystals after sublimation. It is easy to deliquesce and agglomerate, and can Soluble in cold water, slightly soluble in alcohol, insoluble in acetone and liquid ammonia. When heated or exposed to hot water, it decomposes and loses ammonia and converts it into more stable ammonium fluoride. Ammonium fluoride is widely used, such as as a glass etching agent, a chemical polishing agent for metal surfaces, a wood and wine preservative, a disinfectant, a mordant for fibers, and a solvent for extracting rare elements. It can also be used as a component for ion detection in chemical analysis. Masking agent, disinfectant for wine making, preservative, fiber mordant, etc.

Therefore, an optimized preparation scheme for ammonium fluoride is expected.

As shown in Figure 1A, in one preparation scheme, the preparation process is:

Step 1: Add the mother liquor to the reaction tank, then add liquid ammonia and anhydrous hydrogen fluoride under stirring to react; Step 2: Prepare the reaction solution by cooling, crystallizing, centrifuging and drying to prepare ammonium fluoride.

The mother liquid is the liquid obtained by centrifugal separation of the reaction liquid in step 2. The prepared ammonium fluoride has the advantages of low product moisture content, not easy to agglomerate, durable in storage, and of high quality. The mother liquid is a liquid obtained after centrifugal separation of the reaction liquid, and its main components are ammonium fluoride and ammonia water. In the early stage of preparation, a certain amount of mother liquor can be prepared in advance to start the preparation process. Then during the preparation process, the liquid obtained by centrifugation of the reaction liquid can be recycled as the mother liquid. After the preparation is completed, the liquid obtained by centrifugation of the reaction liquid can be retained. Make the mother liquor needed for the next preparation, no need to prepare another mother liquor.

In this way, by adding mother liquor to the reaction tank, equipment damage and pollution are avoided, because directly adding anhydrous hydrogen fluoride when the tank is empty will cause pollution and damage the equipment. And because the mother liquor only needs to be prepared once in the initial stage of preparation, it can be recycled during the preparation process and subsequent preparations without the need for separate preparation, thus greatly reducing production costs and simplifying the production process.

After adding the mother liquor, add liquid ammonia and anhydrous hydrogen fluoride to the reaction tank under stirring to react. Since the ammonium fluoride generated by the reaction is prone to stratification, resulting in uneven acidity, the reaction tank is constantly cleaned during the reaction. The reaction solution is stirred to prevent the generation of substandard products and inaccurate sampling and analysis. Stirring can be completed by the electric stirring device provided in the reaction tank, or an additional stirring device can be added to enhance the stirring effect.

In particular, the best effect is to add liquid ammonia and anhydrous hydrogen fluoride in the following order: first add 50 to 60 kg of liquid ammonia, then add 100 to 110 kg of anhydrous hydrogen fluoride, and finally add the remaining liquid ammonia and anhydrous hydrogen fluoride at the same time. This is because anhydrous hydrogen fluoride has a high density. If added first, it will sink to the bottom, resulting in uneven reactions. Adding a certain amount of liquid ammonia first and then adding a certain amount of anhydrous hydrogen fluoride can effectively avoid uneven reactions while ensuring When the reaction is uniform, adding the remaining liquid ammonia and anhydrous hydrogen fluoride at the same time will help improve production efficiency and avoid excessively long production cycles and increased production costs. During the reaction process, liquid ammonia and anhydrous hydrogen fluoride should be added slowly to control the reaction temperature between 90 and 110°C. A cooling water pipe can also be set up on the reaction tank, supplemented by cooling water for cooling. If the reaction temperature If the reaction rises too fast, it can be adjusted by reducing the feed amount or increasing the cooling water. The reaction pressure should be controlled at normal pressure so that the reaction proceeds continuously, uniformly, slowly and stably. The pH value of the reaction end point is preferably controlled at 5 to 6. The specific control method can be carried out in the following way: when there is still 5% difference from the end point of the feed (calculated as liquid ammonia), use pH test paper or other pH detection device to detect the pH value of the reaction solution. pH value, and then adjust the remaining amount of liquid ammonia and anhydrous hydrogen fluoride accordingly according to the test results, so that the pH value at the end of the reaction is controlled at 5 to 6.

Accordingly, the inventor of the present application found that in the above preparation scheme, the synergy of the flow rate control and reaction temperature of adding liquid ammonia and anhydrous hydrogen fluoride to the reaction tank is of great significance for improving reaction efficiency and product quality. Therefore, it is expected that in the process of preparing ammonium fluoride, the liquid ammonia should be prepared based on the flow rate values of the liquid ammonia and the anhydrous hydrogen fluoride added at multiple predetermined time points, as well as the reaction temperature value and the pH value of the reaction solution. The flow rate of the anhydrous hydrogen fluoride added into the reaction tank is dynamically controlled in real time, thereby improving the efficiency of the reaction and the quality of the product.

Specifically, in the technical solution of the present application, first, multiple sensors are used to obtain the first flow rate value of liquid ammonia, the second flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and the reaction temperature at multiple predetermined time points within a predetermined time period. Liquid pH. Then, in order to extract the hidden correlation feature information between the first flow rate value and the second flow rate value, the first flow rate value of liquid ammonia at multiple predetermined time points within the predetermined time period and the anhydrous After the second flow rate values of hydrogen fluoride are respectively arranged into the first flow rate vector and the second flow rate vector, the product between the transposed vector of the first flow rate vector and the second flow rate vector is calculated to obtain a flow rate control matrix to integrate The flow rate information facilitates subsequent feature mining. Further, a convolutional neural network model with excellent performance in implicit correlation feature extraction is used to extract and filter features of the flow rate control matrix, thereby obtaining a flow rate control feature vector.

It should be understood that for the reaction temperature values and reaction liquid PH values at multiple predetermined time points within the predetermined time period, it is considered that the reaction temperature values and the reaction liquid PH values have special implications in time. Contains feature information. Therefore, in order to more fully extract such implicit features, in the technical solution of this application, a temporal encoder including a one-dimensional convolution layer is used to separately encode multiple predetermined times within the predetermined time period. The reaction temperature value of the point and the pH value of the reaction solution are encoded to obtain the reaction temperature feature vector and pH time series feature vector. In one example, the temporal encoder consists of alternately arranged fully connected layers and one-dimensional convolutional layers, which respectively extract the reaction temperature value and the pH value of the reaction solution in the temporal dimension through one-dimensional convolutional encoding. The high-dimensional hidden features of the reaction temperature value and the pH value of the reaction solution are respectively extracted through the correlation and fully connected coding.

It should be understood that when the flow rate control matrix is passed through the first convolutional neural network as a filter to obtain the flow rate control feature vector, the feature extraction of the first convolutional neural network as a filter cannot be completely preserved. The timing characteristics of the first input vector and the second input vector result in a feature distribution between the flow rate control feature vector and the reaction temperature feature vector and the PH timing feature vector that follow the timing distribution. deviation.

Therefore, the flow rate control feature vector is corrected based on the time series distribution of the reaction temperature feature vector and the pH time series feature vector, as follows:

That is, considering that there is anisotropy between the flow rate control feature vector, the reaction temperature feature vector, and the pH timing feature vector due to the difference in feature distribution, it is reflected that its vector representation resides in high-dimensional features. a narrow subset of space. Therefore, the comparison search space of the flow rate control eigenvector with respect to the reaction temperature eigenvector and the PH time series eigenvector is isotropic through the above correction, so that the flow rate control eigenvector is converted to the same direction as the reaction temperature eigenvector. The isotropic and differentiated representation space of the temperature feature vector and the PH time series feature vector enhances the distribution consistency of the feature representation, thereby improving the accuracy of classification.

Furthermore, considering the use of the corrected flow rate control feature vector as a priori probability, the purpose of the technical solution in this application is to update the prior probability based on new evidence, that is, when the reaction temperature value and the pH value of the reaction solution change. The posterior probability is obtained by the posterior probability. Then according to the Bayesian formula, the posterior probability is the prior probability multiplied by the event probability divided by the evidence probability. Therefore, in the technical solution of this application, the Bayesian probability model is used to fuse the corrected flow rate control feature vector, The reaction temperature feature vector and the PH timing feature vector are used to obtain a posteriori feature vector, where the corrected flow rate control feature vector serves as a priori, the reaction temperature feature vector serves as an event, and the PH timing feature vector serves as evidence. In this way, the posterior feature vector can be passed through the classifier to obtain a first flow rate value indicating that the liquid ammonia at the current time point should increase or decrease and the second flow rate value of anhydrous hydrogen fluoride should increase or decrease. Classification results should be reduced.

Based on this, this application proposes an automatic batching control system for the preparation of ammonium fluoride, which includes: a data acquisition module, used to obtain the first flow rate value of liquid ammonia at multiple predetermined time points within a predetermined time period, anhydrous The second flow rate value of hydrogen fluoride, the reaction temperature value and the pH value of the reaction liquid; the batching speed structured correlation module is used to combine the first flow rate value of liquid ammonia and the first flow rate value of anhydrous hydrogen fluoride at multiple predetermined time points within the predetermined time period. After the second flow velocity values are arranged into the first flow velocity vector and the second flow velocity vector respectively, calculate the product between the transpose vector of the first flow velocity vector and the second flow velocity vector to obtain the flow velocity control matrix; batching velocity characteristic filtering A module for passing the flow rate control matrix through the first convolutional neural network as a filter to obtain a flow rate control feature vector; a time series encoding module for converting the reaction temperature values of multiple predetermined time points within the predetermined time period and the pH value of the reaction solution are respectively passed through a time series encoder containing a one-dimensional convolution layer to obtain the reaction temperature feature vector and the pH time series feature vector; the feature correction module is used to calculate the reaction temperature feature vector and the pH time series feature vector based on the reaction temperature feature vector and the pH time series feature vector. The eigenvalues of each position in the flow rate control eigenvector are corrected to obtain the corrected flow rate control eigenvector; a Bayesian fusion module is used to use a Bayesian probability model to fuse the corrected flow rate control eigenvector and the reaction The temperature feature vector and the PH time series feature vector are used to obtain a posterior feature vector; and, a batching control result generation module is used to pass the posterior feature vector through a classifier to obtain a classification result, and the classification result is used to represent the current The first flow rate value of liquid ammonia at the time point should be increased or should be decreased and the second flow rate value of anhydrous hydrogen fluoride should be increased or should be decreased.

Figure 1B illustrates an application scenario diagram of an automatic batching control system for ammonium fluoride preparation according to an embodiment of the present application. As shown in Figure 1B, in this application scenario, first, multiple predetermined time points within a predetermined time period are acquired through various sensors (for example, the flow rate sensor T1, the temperature sensor T2 and the pH value sensor T3 as shown in Figure 1B) The first flow rate value of liquid ammonia, the second flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and the pH value of the reaction liquid. Then, the obtained first flow rate value of liquid ammonia, the second flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and the pH value of the reaction liquid at multiple predetermined time points within the predetermined time period are input into the prepared system equipped with ammonium fluoride. in a server using an automatic batching control algorithm (for example, the cloud server S as shown in FIG. 1B), wherein the server can use the automatic batching control algorithm for ammonium fluoride preparation to process multiple batches within the predetermined time period. The first flow rate value of liquid ammonia at the predetermined time point, the second flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and the pH value of the reaction liquid are processed to generate a first flow rate value of liquid ammonia that represents the current time point. The classification result is that the second flow rate value of anhydrous hydrogen fluoride should be increased or should be reduced.

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

Example system

Figure 2 illustrates a block diagram of an automatic batching control system for ammonium fluoride preparation according to an embodiment of the present application. As shown in Figure 2, the automatic batching control system 200 for ammonium fluoride preparation according to the embodiment of the present application includes: a data acquisition module 210, used to obtain the first flow rate of liquid ammonia at multiple predetermined time points within a predetermined time period. value, the second flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and the pH value of the reaction liquid; the batching speed structured correlation module 220 is used to combine the first flow rate value of liquid ammonia at multiple predetermined time points within the predetermined time period and the second flow rate values of anhydrous hydrogen fluoride are arranged into the first flow rate vector and the second flow rate vector respectively, then calculate the product between the transposed vector of the first flow rate vector and the second flow rate vector to obtain the flow rate control matrix ; Ingredient speed feature filtering module 230, used to pass the flow speed control matrix through the first convolutional neural network as a filter to obtain the flow speed control feature vector; Time series encoding module 240, used to combine multiple data within the predetermined time period The reaction temperature value and reaction liquid PH value at the predetermined time point are respectively passed through a time series encoder including a one-dimensional convolution layer to obtain a reaction temperature feature vector and a PH time series feature vector; the feature correction module 250 is used to based on the reaction temperature feature vector and PH time series feature vector, correcting the feature values of each position in the flow speed control feature vector to obtain the corrected flow speed control feature vector; the Bayesian fusion module 260 is used to use the Bayesian probability model to fuse the correction The posterior flow rate control feature vector, the reaction temperature feature vector and the PH timing feature vector are used to obtain a posterior feature vector; and a batching control result generation module 270 is used to pass the posterior feature vector through a classifier to obtain classification As a result, the classification result is used to indicate that the first flow rate value of liquid ammonia at the current time point should be increased or decreased and the second flow rate value of anhydrous hydrogen fluoride should be increased or decreased.

Specifically, in the embodiment of the present application, the data collection module 210, the batching speed structured correlation module 220 and the batching speed feature filtering module 230 are used to obtain liquid data at multiple predetermined time points within a predetermined time period. The first flow rate value of ammonia, the second flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and the pH value of the reaction liquid, and the first flow rate value of liquid ammonia and anhydrous hydrogen fluoride at multiple predetermined time points within the predetermined time period After the second flow velocity values are arranged into the first flow velocity vector and the second flow velocity vector respectively, calculate the product between the transpose vector of the first flow velocity vector and the second flow velocity vector to obtain the flow velocity control matrix, and then The flow rate control matrix is passed through the first convolutional neural network as a filter to obtain the flow rate control feature vector. As mentioned above, it should be understood that in the preparation scheme, the synergy of flow rate control and reaction temperature of liquid ammonia and anhydrous hydrogen fluoride added to the reaction tank is of great significance for improving reaction efficiency and product quality. Therefore, in the technical solution of the present application, it is expected that in the process of preparing ammonium fluoride, the flow rate values of the liquid ammonia and the anhydrous hydrogen fluoride added at multiple predetermined time points, as well as the reaction temperature value and the reaction liquid The pH value is used to dynamically control the flow rate of the liquid ammonia and anhydrous hydrogen fluoride added to the reaction tank in real time, thereby improving the efficiency of the reaction and the quality of the product.

That is, specifically, in the technical solution of the present application, first, multiple sensors are used to obtain the first flow rate value of liquid ammonia, the second flow rate value of anhydrous hydrogen fluoride, and the reaction temperature at multiple predetermined time points within a predetermined time period. value and the pH value of the reaction solution. Then, in order to extract the hidden correlation feature information between the first flow rate value and the second flow rate value, the first flow rate value of liquid ammonia at multiple predetermined time points within the predetermined time period and the anhydrous After the second flow rate values of hydrogen fluoride are respectively arranged into the first flow rate vector and the second flow rate vector, the product between the transposed vector of the first flow rate vector and the second flow rate vector is calculated to obtain a flow rate control matrix to integrate The flow rate information facilitates subsequent feature mining. Further, a convolutional neural network model with excellent performance in implicit correlation feature extraction is used to extract and filter features of the flow rate control matrix, thereby obtaining a flow rate control feature vector.

More specifically, in the embodiment of the present application, the batching speed feature filtering module is further used to perform: each layer of the first convolutional neural network on the input data in the forward transfer of the layer: The data is convolved to obtain a convolution feature map; the convolution feature map is subjected to mean pooling based on a local feature matrix to obtain a pooled feature map; and, the pooled feature map is subjected to nonlinear activation to obtain Activating the feature map; wherein, the output of the last layer of the first convolutional neural network is the flow rate control feature vector, and the input of the first layer of the first convolutional neural network is the flow rate control matrix.

Specifically, in the embodiment of the present application, the time series encoding module 240 is used to pass the reaction temperature values and the pH values of the reaction liquid at multiple predetermined time points within the predetermined time period through a time series including a one-dimensional convolution layer. encoder to obtain the reaction temperature feature vector and PH timing feature vector. It should be understood that for the reaction temperature values and reaction liquid PH values at multiple predetermined time points within the predetermined time period, it is considered that the reaction temperature values and the reaction liquid PH values have special implications in time. Contains feature information. Therefore, in order to more fully extract such implicit features, in the technical solution of this application, a temporal encoder including a one-dimensional convolution layer is used to separately encode multiple predetermined times within the predetermined time period. The reaction temperature value of the point and the pH value of the reaction solution are encoded to obtain the reaction temperature feature vector and pH time series feature vector. Correspondingly, in a specific example, the temporal encoder consists of alternately arranged fully connected layers and one-dimensional convolutional layers, which respectively extract the reaction temperature value and the reaction liquid PH through one-dimensional convolutional coding. The correlation of values in the time series dimension and the high-dimensional hidden features of the reaction temperature value and the pH value of the reaction solution are respectively extracted through fully connected coding.

More specifically, in the embodiment of the present application, the time series encoding module includes: first, arranging the reaction temperature values at multiple predetermined time points within the predetermined time period into a temperature input vector according to the time dimension; then, using the The fully connected layer of the temporal encoder performs fully connected encoding on the temperature input vector using the following formula to extract high-dimensional implicit features of the eigenvalues of each position in the temperature input vector, where the formula is:

represents the matrix multiplication; then, use the one-dimensional convolution layer of the temporal encoder to perform one-dimensional convolution encoding on the temperature input vector with the following formula to extract the high-dimensional hidden between the eigenvalues of each position in the temperature input vector Contains associated features, where the formula is:

Among them, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix that operates with the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector. Further, the pH values of the reaction solution at multiple predetermined time points within the predetermined time period are arranged into a pH value input vector according to the time dimension; then, the fully connected layer of the temporal encoder is used to calculate the pH value according to the following formula The input vector is fully connected to extract the high-dimensional hidden features of the eigenvalues at each position in the PH value input vector, where the formula is:

represents matrix multiplication; then, use the one-dimensional convolution layer of the temporal encoder to perform one-dimensional convolution encoding on the PH value input vector with the following formula to extract the feature value between each position in the PH value input vector The high-dimensional implicit correlation characteristics of , where the formula is:

Figure 3 illustrates a block diagram of the timing coding module in the automatic batching control system for ammonium fluoride preparation according to an embodiment of the present application. As shown in Figure 3, the time series encoding module 240 includes: a temperature time series encoding unit 241, used to arrange the reaction temperature values at multiple predetermined time points within the predetermined time period into a temperature input vector according to the time dimension; using the The fully connected layer of the temporal encoder performs fully connected encoding on the temperature input vector using the following formula to extract high-dimensional implicit features of the eigenvalues of each position in the temperature input vector, where the formula is:

Among them, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix that operates with the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector. The PH timing encoding unit 242 is used to arrange the PH values of the reaction solution at multiple predetermined time points within the predetermined time period into a PH value input vector according to the time dimension; use the fully connected layer of the timing encoder to calculate the PH values using the following formula: The PH value input vector is fully connected to extract the high-dimensional hidden features of the eigenvalues at each position in the PH value input vector, where the formula is:

Specifically, in the embodiment of the present application, the feature correction module 250 is used to correct the feature values of each position in the flow rate control feature vector based on the reaction temperature feature vector and the pH timing feature vector to obtain a correction. Post flow velocity control eigenvector. It should be understood that when the flow rate control matrix is passed through the first convolutional neural network as a filter to obtain the flow rate control feature vector, the feature extraction of the first convolutional neural network as a filter cannot be completely preserved. The timing characteristics of the first input vector and the second input vector result in a feature distribution between the flow rate control feature vector and the reaction temperature feature vector and the PH timing feature vector that follow the timing distribution. deviation. Therefore, in the technical solution of the present application, the flow rate control feature vector is further corrected based on the time series distribution of the reaction temperature feature vector and the pH time series feature vector. That is, considering that there is anisotropy between the flow rate control feature vector, the reaction temperature feature vector, and the pH timing feature vector due to the difference in feature distribution, it is reflected that its vector representation resides in high-dimensional features. a narrow subset of space. Therefore, the comparison search space of the flow rate control eigenvector with respect to the reaction temperature eigenvector and the PH timing eigenvector is isotropic through the above correction, so that the flow rate control eigenvector is converted to the same direction as the reaction temperature eigenvector. The isotropic and differentiated representation space of the temperature feature vector and the PH time series feature vector enhances the distribution consistency of the feature representation, thereby improving the accuracy of classification.

More specifically, in the embodiment of the present application, the feature correction module is further configured to: based on predetermined hyperparameters and the feature values of each position in the flow rate control feature vector and the features of each position in the reaction temperature feature vector. The greater of the difference between the first distance between values and the first distance, and the characteristic value of each position in the predetermined hyperparameter and the flow rate control characteristic vector and the PH timing sequence The sum value between the difference between the second distances between the characteristic values of each position in the feature vector and the larger of the second distances, for the characteristics of each position in the flow velocity control feature vector The value is corrected to obtain the corrected flow rate control characteristic vector. Correspondingly, in a specific example, based on the reaction temperature feature vector and the pH timing feature vector, the feature values of each position in the flow rate control feature vector are corrected with the following formula to obtain the corrected flow rate control feature vector ;wherein, the formula is:

Specifically, in the embodiment of the present application, the Bayesian fusion module 260 and the batching control result generation module 270 are used to use a Bayesian probability model to fuse the corrected flow rate control feature vector, the reaction The temperature feature vector and the PH time series feature vector are used to obtain a posteriori feature vector, and the posteriori feature vector is passed through a classifier to obtain a classification result. The classification result is used to represent the first flow rate of liquid ammonia at the current time point. The value should be increased or should be decreased and the second flow rate value of anhydrous hydrogen fluoride should be increased or should be decreased. It should be understood that, taking into account the use of the corrected flow rate control feature vector as a priori probability, the purpose of the technical solution in this application is to update based on new evidence, that is, when there is a change in the reaction temperature value and the pH value of the reaction solution. The prior probability gives the posterior probability. Then according to the Bayesian formula, the posterior probability is the prior probability multiplied by the event probability divided by the evidence probability. Therefore, in the technical solution of this application, the Bayesian probability model is used to fuse the corrected flow rate control feature vector, The reaction temperature feature vector and the PH timing feature vector are used to obtain a posteriori feature vector, where the corrected flow rate control feature vector serves as a priori, the reaction temperature feature vector serves as an event, and the PH timing feature vector serves as evidence. In this way, the posterior feature vector can be passed through the classifier to obtain a first flow rate value indicating that the liquid ammonia at the current time point should increase or decrease and the second flow rate value of anhydrous hydrogen fluoride should increase or decrease. Classification results should be reduced.

Correspondingly, in a specific example, the classifier is used to process the posterior feature vector with the following formula to obtain the classification result, where the formula is: softmax{(W _n ,B _n ): …:(W ₁ ,B ₁ )|X}, where W ₁ to W _n are weight matrices, B ₁ to B _n are bias vectors, and X is the posterior feature vector.

More specifically, in the embodiment of the present application, the Bayesian fusion module is further used to: use the Bayesian probability model to fuse the corrected flow rate control feature vector, the reaction temperature feature vector and the following formula: The PH time series feature vector is used to obtain the posterior feature vector; wherein, the formula is:

qi＝pi*ai/bi

In summary, the automatic batching control system 200 for the preparation of ammonium fluoride based on the embodiment of the present application has been clarified, which uses artificial intelligence control technology to determine the flow rate value of liquid ammonia, the flow rate value of anhydrous hydrogen fluoride, and the reaction temperature value. Starting from the pH value of the reaction liquid, a deep neural network model is used to mine the implicit feature correlation information of each data in the time series dimension, and Bayes is used to fuse the correlation feature information of each data to compare the liquid ammonia and The flow rate of the anhydrous hydrogen fluoride added into the reaction tank is dynamically controlled in real time, thereby improving the efficiency of the reaction and the quality of the product.

As mentioned above, the automatic batching control system 200 for the preparation of ammonium fluoride according to the embodiment of the present application can be implemented in various terminal devices, such as a server of the automatic batching control algorithm for the preparation of ammonium fluoride, etc. In one example, the automatic batching control system 200 for ammonium fluoride preparation according to the embodiment of the present application can be integrated into the terminal device as a software module and/or hardware module. For example, the automatic batching control system 200 for ammonium fluoride preparation 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 ammonium fluoride preparation The automatic batching control system 200 used can also be one of the many hardware modules of the terminal equipment.

Alternatively, in another example, the automatic batching control system 200 for ammonium fluoride preparation and the terminal device can also be separate devices, and the automatic batching control system 200 for ammonium fluoride preparation can be connected via wired and/or Or a wireless network is connected to the terminal device, and interactive information is transmitted according to the agreed data format.

Example methods

Figure 4 illustrates a flow chart of a control method of an automatic batching control system for ammonium fluoride preparation. As shown in Figure 4, the control method of the automatic batching control system for ammonium fluoride preparation according to the embodiment of the present application includes the steps: S110, obtaining the first flow rate value of liquid ammonia at multiple predetermined time points within a predetermined time period, The second flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and the pH value of the reaction liquid; S120, separate the first flow rate value of liquid ammonia and the second flow rate value of anhydrous hydrogen fluoride at multiple predetermined time points within the predetermined time period, respectively. After being arranged into the first flow velocity vector and the second flow velocity vector, calculate the product between the transpose vector of the first flow velocity vector and the second flow velocity vector to obtain the flow velocity control matrix; S130, pass the flow velocity control matrix through Use the first convolutional neural network as a filter to obtain the flow rate control feature vector; S140, pass the reaction temperature values and reaction liquid PH values at multiple predetermined time points within the predetermined time period through a time series including a one-dimensional convolution layer. encoder to obtain the reaction temperature feature vector and the PH timing feature vector; S150, based on the reaction temperature feature vector and the PH timing feature vector, correct the eigenvalues of each position in the flow rate control feature vector to obtain the corrected flow rate control Feature vector; S160, use a Bayesian probability model to fuse the corrected flow rate control feature vector, the reaction temperature feature vector and the PH timing feature vector to obtain a posterior feature vector; and, S170, convert the posterior feature vector The experimental feature vector is passed through the classifier to obtain a classification result, which is used to indicate that the first flow rate value of liquid ammonia at the current time point should be increased or decreased and the second flow rate value of anhydrous hydrogen fluoride should be increased or should be increased. decrease.

Figure 5 illustrates a schematic structural diagram of a control method of an automatic batching control system for ammonium fluoride preparation according to an embodiment of the present application. As shown in Figure 5, in the network architecture of the control method of the automatic batching control system for ammonium fluoride preparation, first, the first flow rates of liquid ammonia at multiple predetermined time points within the predetermined time period are obtained. The value (e.g., P1 as illustrated in Figure 5) and the second flow rate value of anhydrous hydrogen fluoride (e.g., P2 as illustrated in Figure 5) are respectively arranged as a first flow rate vector (e.g., as illustrated in Figure 5 After V1) and the second flow velocity vector (for example, V2 as illustrated in Figure 5), the product between the transpose vector of the first flow velocity vector and the second flow velocity vector is calculated to obtain the flow velocity control matrix ( For example, M) as shown in Figure 5; then, the flow rate control matrix is passed through the first convolutional neural network as a filter (for example, CNN1 as shown in Figure 5) to obtain the flow rate control feature vector ( For example, VF1 as shown in Figure 5); then, the obtained reaction temperature values (for example, Q1 as shown in Figure 5) and the reaction liquid PH value (for example, Q1 as shown in Figure 5) at multiple predetermined time points within the predetermined time period ( For example, Q2 as shown in Figure 5) are respectively passed through a temporal encoder (for example, E as shown in Figure 5) including a one-dimensional convolution layer to obtain the reaction temperature feature vector (for example, as shown in Figure 5 VF2) and PH timing feature vector (for example, VF3 as shown in Figure 5); then, based on the reaction temperature feature vector and PH timing feature vector, the eigenvalues of each position in the flow rate control feature vector are Calibrate to obtain the corrected flow rate control feature vector (for example, VF4 as illustrated in Figure 5); then, use a Bayesian probability model to fuse the corrected flow rate control feature vector, the reaction temperature feature vector and the PH the temporal feature vector to obtain the posterior feature vector (for example, VF as illustrated in Figure 5); and, finally, pass the posterior feature vector through the classifier (for example, the circle S as illustrated in Figure 5) To obtain a classification result, the classification result is used to indicate that the first flow rate value of liquid ammonia at the current time point should be increased or decreased and the second flow rate value of anhydrous hydrogen fluoride should be increased or decreased.

Use a semantic understanding model (for example, SUM as shown in Figure 5) to separately apply the declared registration information of the multiple wireless APs (for example, IN1 as shown in Figure 5) when the obtained multiple wireless APs simultaneously declare registration. -INn) into semantic feature vectors (for example, VS1-VSn as shown in Figure 5); then, the flow rate control matrix is passed through the first convolutional neural network as a filter to obtain the flow rate control feature vector (for example, VS1-VSn as shown in Figure 5) , MR as shown in Figure 5); then, use a convolutional neural network (for example, CNN as shown in Figure 5) to extract the registration information feature map (for example, as shown in Figure 5) from the registration information matrix FR as shown), the scale of the registration information feature map is L*S*C, L represents the length of the semantic feature vector, S represents the number of wireless APs, and C represents the number of channels; then, the registration information features The eigenvalue matrix of each L*C of the graph in the S dimension is eigenvalue decomposed to obtain the diagonal eigenvalue matrix corresponding to the eigenmatrix of each said L*C (for example, as illustrated in Figure 5 MD1-MDn) and the eigenvector matrix (for example, ME1-MEn as illustrated in Figure 5); then, a Bayesian probability model is used to fuse the corrected flow rate control eigenvector, the reaction temperature eigenvector and The PH time series feature vector is used to obtain a posterior feature vector (for example, VE as shown in Figure 5); and, finally, the eigenvalue vector is input into a classifier as a classification vector (for example, as shown in Figure 5 The schematic circle S) is used to obtain the classification result, which is used to indicate whether the declaration and registration of the wireless AP is correct.

More specifically, in steps S110, S120 and S130, the first flow rate value of liquid ammonia, the second flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and the reaction liquid PH at multiple predetermined time points within the predetermined time period are obtained value, and after arranging the first flow rate values of liquid ammonia and the second flow rate values of anhydrous hydrogen fluoride at multiple predetermined time points within the predetermined time period into the first flow rate vector and the second flow rate vector respectively, calculate the The product of the transposed vector of the first flow vector and the second flow vector is used to obtain the flow control matrix, and then the flow control matrix is passed through the first convolutional neural network as a filter to obtain the flow control feature vector. It should be understood that in the preparation scheme, the synergy of the flow rate control and reaction temperature of liquid ammonia and anhydrous hydrogen fluoride added to the reaction tank is of great significance for improving reaction efficiency and product quality. Therefore, in the technical solution of the present application, it is expected that in the process of preparing ammonium fluoride, the flow rate values of the liquid ammonia and the anhydrous hydrogen fluoride added at multiple predetermined time points, as well as the reaction temperature value and the reaction liquid The pH value is used to dynamically control the flow rate of the liquid ammonia and anhydrous hydrogen fluoride added to the reaction tank in real time, thereby improving the efficiency of the reaction and the quality of the product.

More specifically, in step S140, the reaction temperature values and reaction liquid PH values at multiple predetermined time points within the predetermined time period are passed through a temporal encoder including a one-dimensional convolution layer to obtain the reaction temperature feature vector and PH. Time series feature vector. It should be understood that for the reaction temperature values and reaction liquid PH values at multiple predetermined time points within the predetermined time period, it is considered that the reaction temperature values and the reaction liquid PH values have special implications in time. Contains feature information. Therefore, in order to more fully extract such implicit features, in the technical solution of this application, a temporal encoder including a one-dimensional convolution layer is used to separately encode multiple predetermined times within the predetermined time period. The reaction temperature value of the point and the pH value of the reaction solution are encoded to obtain the reaction temperature feature vector and pH time series feature vector. Correspondingly, in a specific example, the temporal encoder consists of alternately arranged fully connected layers and one-dimensional convolutional layers, which respectively extract the reaction temperature value and the reaction liquid PH through one-dimensional convolutional coding. The correlation of values in the time series dimension and the high-dimensional hidden features of the reaction temperature value and the pH value of the reaction solution are respectively extracted through fully connected coding.

More specifically, in step S150, based on the reaction temperature feature vector and the pH timing feature vector, the feature values of each position in the flow rate control feature vector are corrected to obtain a corrected flow rate control feature vector. It should be understood that when the flow rate control matrix is passed through the first convolutional neural network as a filter to obtain the flow rate control feature vector, the feature extraction of the first convolutional neural network as a filter cannot be completely preserved. The timing characteristics of the first input vector and the second input vector result in a feature distribution between the flow rate control feature vector and the reaction temperature feature vector and the PH timing feature vector that follow the timing distribution. deviation. Therefore, in the technical solution of the present application, the flow rate control feature vector is further corrected based on the time series distribution of the reaction temperature feature vector and the pH time series feature vector. That is, considering that there is anisotropy between the flow rate control feature vector, the reaction temperature feature vector, and the pH timing feature vector due to the difference in feature distribution, it is reflected that its vector representation resides in high-dimensional features. a narrow subset of space. Therefore, the comparison search space of the flow rate control eigenvector with respect to the reaction temperature eigenvector and the PH time series eigenvector is isotropic through the above correction, so that the flow rate control eigenvector is converted to the same direction as the reaction temperature eigenvector. The isotropic and differentiated representation space of the temperature feature vector and the PH time series feature vector enhances the distribution consistency of the feature representation, thereby improving the accuracy of classification.

More specifically, in steps S160 and S170, a Bayesian probability model is used to fuse the corrected flow rate control feature vector, the reaction temperature feature vector and the PH timing feature vector to obtain a posterior feature vector, and The posterior feature vector is passed through a classifier to obtain a classification result, which is used to indicate that the first flow rate value of liquid ammonia at the current time point should be increased or decreased and the second flow rate value of anhydrous hydrogen fluoride should be increase or should decrease. It should be understood that, taking into account the use of the corrected flow rate control feature vector as a priori probability, the purpose of the technical solution in this application is to update based on new evidence, that is, when there is a change in the reaction temperature value and the pH value of the reaction solution. The prior probability gives the posterior probability. Then according to the Bayesian formula, the posterior probability is the prior probability multiplied by the event probability divided by the evidence probability. Therefore, in the technical solution of this application, the Bayesian probability model is used to fuse the corrected flow rate control feature vector, The reaction temperature feature vector and the PH timing feature vector are used to obtain a posteriori feature vector, where the corrected flow rate control feature vector serves as a priori, the reaction temperature feature vector serves as an event, and the PH timing feature vector serves as evidence. In this way, the posterior feature vector can be passed through the classifier to obtain a first flow rate value indicating that the liquid ammonia at the current time point should increase or decrease and the second flow rate value of anhydrous hydrogen fluoride should increase or decrease. Classification results should be reduced.

In summary, the control method of the automatic batching control system for the preparation of ammonium fluoride based on the embodiments of the present application has been clarified. By using artificial intelligence control technology, the flow rate value of liquid ammonia, the flow rate value of anhydrous hydrogen fluoride, and the reaction Starting from the temperature value and pH value of the reaction liquid, a deep neural network model is used to mine the implicit feature correlation information of each data in the time series dimension, and Bayesian is used to fuse the correlation feature information of each data to predict the liquid. The flow rate of ammonia and the anhydrous hydrogen fluoride added to the reaction tank is dynamically controlled in real time, thereby improving the efficiency of the reaction and the quality of the product.

Example computer program products and computer-readable storage media

In addition to the above-mentioned methods and devices, embodiments of the present application may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to execute the “exemplary method” described above in this specification. The steps in the functions of the control method of the automatic batching control system for ammonium fluoride preparation according to various embodiments of the present application are described in the section.

The computer program product can be used to write program codes for performing the operations of the embodiments of the present application in any combination of one or more programming languages, including object-oriented programming languages, such as Java, C++, etc. , also includes conventional procedural programming languages, such as the "C" language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server execute on.

In addition, embodiments of the present application may also be a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are run by a processor, the computer program instructions cause the processor to execute the above-mentioned "example method" part of this specification. The steps in the control method of the automatic batching control system for ammonium fluoride preparation described in.

The computer-readable storage medium may be any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. Readable storage media may include, for example, but are not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: electrical connection with one or more conductors, portable disk, hard disk, random access memory (RAM), read only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.

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

An automatic batching control system for the preparation of ammonium fluoride, characterized in that it includes: a data acquisition module, used to obtain the first flow rate value of liquid ammonia and the second flow rate value of anhydrous hydrogen fluoride at multiple predetermined time points within a predetermined time period. Flow rate value, reaction temperature value and reaction liquid PH value; batching speed structured correlation module, used to combine the first flow rate value of liquid ammonia and the second flow rate value of anhydrous hydrogen fluoride at multiple predetermined time points within the predetermined time period After being arranged into the first flow velocity vector and the second flow velocity vector respectively, calculate the product between the transpose vector of the first flow velocity vector and the second flow velocity vector to obtain the flow velocity control matrix; the batching velocity characteristic filter module is used to The flow rate control matrix is passed through the first convolutional neural network as a filter to obtain the flow rate control feature vector; a time series encoding module is used to combine the reaction temperature values and reaction liquid PH at multiple predetermined time points within the predetermined time period. The values are respectively passed through a time series encoder including a one-dimensional convolution layer to obtain a reaction temperature feature vector and a PH time series feature vector; a feature correction module is used to calculate the flow rate control feature based on the reaction temperature feature vector and PH time series feature vector. The eigenvalues of each position in the vector are corrected to obtain the corrected flow rate control eigenvector; a Bayesian fusion module is used to use a Bayesian probability model to fuse the corrected flow rate control eigenvector, the reaction temperature eigenvector and The PH time series feature vector is used to obtain a posteriori feature vector; and a batching control result generation module is used to pass the posteriori feature vector through a classifier to obtain a classification result, and the classification result is used to represent the liquid ammonia at the current time point. The first flow rate value of should be increased or should be decreased and the second flow rate value of anhydrous hydrogen fluoride should be increased or should be decreased.
The automatic batching control system for ammonium fluoride preparation according to claim 1, characterized in that the batching speed characteristic filter module is further used to: each layer of the first convolutional neural network is in the forward direction of the layer. During the transfer, the input data are separately processed: convolution processing is performed on the input data to obtain a convolution feature map; mean pooling based on the local feature matrix is performed on the convolution feature map to obtain a pooled feature map; and

Nonlinear activation is performed on the pooled feature map to obtain an activation feature map; wherein, the output of the last layer of the first convolutional neural network is the flow rate control feature vector, and the output of the last layer of the first convolutional neural network is the flow rate control feature vector. The input to the first layer is the flow rate control matrix.
The automatic batching control system for ammonium fluoride preparation according to claim 2, characterized in that the time sequence encoding module includes: a temperature time sequence encoding unit for converting the temperature at multiple predetermined time points within the predetermined time period. The reaction temperature values are arranged into temperature input vectors according to the time dimension; use the fully connected layer of the temporal encoder to fully connect the temperature input vector with the following formula to extract the characteristic values of each position in the temperature input vector. High-dimensional latent features, where the formula is:
where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector,
represents a matrix multiplication; and, using the one-dimensional convolution layer of the temporal encoder to perform one-dimensional convolution encoding on the temperature input vector with the following formula to extract the high-dimensional hidden between the eigenvalues of each position in the temperature input vector Contains associated features, where the formula is:

Among them, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix that operates with the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector; PH timing encoding unit, used to arrange the PH values of the reaction solution at multiple predetermined time points within the predetermined time period into a PH value input vector according to the time dimension; use the fully connected layer of the timing encoder to calculate the The PH value input vector is fully connected and encoded to extract the high-dimensional hidden features of the eigenvalues at each position in the PH value input vector, where the formula is:
where X is the input vector, Y is the output vector, W is the weight matrix, and B is the bias vector,
represents matrix multiplication; and uses the one-dimensional convolution layer of the temporal encoder to perform one-dimensional convolution encoding on the PH value input vector with the following formula to extract the relationship between the feature values of each position in the PH value input vector High-dimensional implicit correlation features, where the formula is:

Among them, a is the width of the convolution kernel in the x direction, F is the convolution kernel parameter vector, G is the local vector matrix that operates with the convolution kernel function, w is the size of the convolution kernel, and X represents the input vector.
The automatic batching control system for the preparation of ammonium fluoride according to claim 3, characterized in that the feature correction module is further used to control the eigenvalues of each position in the eigenvector based on predetermined hyperparameters and the flow rate. The greater of the difference between the first distance between the characteristic values of each position in the reaction temperature feature vector and the first distance, and the predetermined hyperparameter and the flow rate control feature vector, whichever is greater The sum value between the larger of the difference between the feature value of each position and the second distance between the feature value of each position in the PH time series feature vector and the second distance, for all The characteristic values of each position in the flow velocity control characteristic vector are corrected to obtain the corrected flow velocity control characteristic vector.
The automatic batching control system for ammonium fluoride preparation according to claim 4, characterized in that the characteristic correction module is further used to: based on the reaction temperature characteristic vector and the pH timing characteristic vector, use the following formula to calculate the The characteristic values of each position in the flow velocity control characteristic vector are corrected to obtain the corrected flow velocity control characteristic vector; wherein, the formula is:

Where f 1 , f 2 and f 3 are respectively the eigenvalues of the corresponding positions of the reaction temperature eigenvector, the PH timing eigenvector and the flow rate control eigenvector normalized to the [0,1] interval, d(f 3 , f 1 ) represents the first distance between the eigenvalues of each position in the flow rate control eigenvector and the eigenvalues of each position in the reaction temperature eigenvector, d(f 3 , f 2 ) represents the second distance between the eigenvalues of each position in the flow rate control eigenvector and the eigenvalues of each position in the PH timing feature vector, and ρ is the predetermined hyperparameter.
The automatic batching control system for ammonium fluoride preparation according to claim 5, characterized in that the Bayesian fusion module is further used to: use a Bayesian probability model to fuse the corrected flow rate with the following formula Control the eigenvector, the reaction temperature eigenvector and the PH timing eigenvector to obtain the posterior eigenvector; wherein, the formula is:

qi＝pi*ai/bi

Wherein, pi is the characteristic value of each position in the corrected flow rate control characteristic vector, ai and bi are the characteristic values of each position in the reaction temperature characteristic vector and the pH timing characteristic vector respectively, and qi is the characteristic value of each position in the corrected flow rate control characteristic vector. The eigenvalues of each position in the posterior eigenvector.
The automatic batching control system for ammonium fluoride preparation according to claim 6, characterized in that the batching control result generation module is further used to use the classifier to process the posterior feature vector with the following formula To obtain the classification result, the formula is: softmax{W n ,B n ):...:(W 1 ,B 1 )|X}, where W 1 to W n are weight matrices, and B 1 to B n is the bias vector, and X is the posterior feature vector.
A control method for an automatic batching control system for ammonium fluoride preparation, which is characterized by including:

Obtain the first flow rate value of liquid ammonia, the second flow rate value of anhydrous hydrogen fluoride, the reaction temperature value and the pH value of the reaction liquid at multiple predetermined time points within the predetermined time period; After the first flow rate value of liquid ammonia and the second flow rate value of anhydrous hydrogen fluoride are arranged into the first flow rate vector and the second flow rate vector respectively, calculate the relationship between the transpose vector of the first flow rate vector and the second flow rate vector. The product of The liquid PH value passes through a time series encoder containing a one-dimensional convolution layer to obtain a reaction temperature feature vector and a PH time series feature vector; based on the reaction temperature feature vector and PH time series feature vector, each position in the flow rate control feature vector is The eigenvalues are corrected to obtain the corrected flow rate control eigenvector; a Bayesian probability model is used to fuse the corrected flow rate control eigenvector, the reaction temperature eigenvector and the PH timing eigenvector to obtain a posteriori feature vector ;as well as

The posterior feature vector is passed through a classifier to obtain a classification result, which is used to indicate that the first flow rate value of liquid ammonia at the current time point should be increased or decreased and the second flow rate value of anhydrous hydrogen fluoride should be increase or should decrease.
The control method of the automatic batching control system for ammonium fluoride preparation according to claim 8, characterized in that the flow rate control matrix is passed through the first convolutional neural network as a filter to obtain the flow rate control feature vector. , including: each layer of the first convolutional neural network separately performs convolution processing on the input data in the forward transfer of the layer to obtain a convolution feature map; performs a convolution-based feature map on the convolution feature map. Mean pooling of the local feature matrix to obtain a pooled feature map; and nonlinear activation of the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the first convolutional neural network is The flow rate control feature vector is the flow rate control matrix, and the input of the first layer of the first convolutional neural network is the flow rate control matrix.
The control method of the automatic batching control system for ammonium fluoride preparation according to claim 9, characterized in that, based on the reaction temperature feature vector and the pH timing feature vector, each position in the flow rate control feature vector is Correcting the eigenvalues to obtain the corrected flow rate control eigenvector, including: based on the first difference between the predetermined hyperparameter and the eigenvalues of each position in the flow rate control eigenvector and the eigenvalues of each position in the reaction temperature eigenvector. The larger of the difference between a distance and the first distance, and the characteristic value of each position in the predetermined hyperparameter and the flow rate control feature vector and each position in the PH timing feature vector The sum value between the difference between the second distance between the characteristic values and the larger of the second distances, the characteristic values of each position in the flow control characteristic vector are corrected to obtain The corrected flow rate control feature vector.