CN116164497A

CN116164497A - Rectification control system and method in liquid oxygen preparation process

Info

Publication number: CN116164497A
Application number: CN202310455075.8A
Authority: CN
Inventors: 郭剑煌; 曾育红; 江强
Original assignee: Fujian Detianchen New Material Technology Co ltd
Current assignee: Fujian Detianchen New Material Technology Co ltd
Priority date: 2023-04-25
Filing date: 2023-04-25
Publication date: 2023-05-26

Abstract

A rectification control system and a method thereof in the liquid oxygen preparation process acquire feeding speed values and liquid level values in a tower at a plurality of preset time points in a preset time period; and excavating the relevance characteristic distribution information of the time sequence change characteristic of the feeding rate value and the time sequence dynamic change characteristic of the liquid level value in the tower by adopting an artificial intelligence technology based on deep learning, and adaptively adjusting the feeding rate value based on the time sequence cooperative change condition of the feeding rate value and the liquid level value in the tower so as to optimize the rectifying efficiency and the operation stability.

Description

Rectification control system and method in liquid oxygen preparation process

Technical Field

The application relates to the technical field of intelligent control, and in particular relates to a rectification control system and a rectification control method in a liquid oxygen preparation process.

Background

Liquid oxygen has a wide range of industrial and medical uses, and is produced industrially by heating, compressing, cooling and rectifying liquid air. In the actual preparation process of liquid oxygen, the rectification control is a very important link.

In the rectification process, parameters such as the feeding rate, the liquid level in the tower and the like need to be controlled so as to ensure the stability and the safety of operation. This is due to: for the feeding rate, if the feeding rate is too high, uneven distribution of liquid in the tower is caused, and the separation effect is affected; if the feed rate is too slow, the production efficiency is lowered. For the liquid level in the tower, if the liquid level in the tower is too high, the safety problems such as pipeline blockage, overflow and the like can be caused; if the liquid level is too low, the separation effect is affected. However, in the conventional rectification parameter control scheme, only the respective parameter data are compared with the threshold value separately to complete the corresponding control, and the cooperative suitability of the respective parameter data is not concerned, so that the preparation efficiency and the operation stability of liquid oxygen are difficult to meet the requirements.

Accordingly, an optimized rectification control system in a liquid oxygen production process is desired.

Disclosure of Invention

The present application has been made in order to solve the above technical problems. The embodiment of the application provides a rectification control system and a rectification control method in a liquid oxygen preparation process, wherein the rectification control system and the rectification control method acquire feeding speed values and liquid level values in a tower at a plurality of preset time points in a preset time period; and excavating the relevance characteristic distribution information of the time sequence change characteristic of the feeding rate value and the time sequence dynamic change characteristic of the liquid level value in the tower by adopting an artificial intelligence technology based on deep learning, and adaptively adjusting the feeding rate value based on the time sequence cooperative change condition of the feeding rate value and the liquid level value in the tower so as to optimize the rectifying efficiency and the operation stability.

In a first aspect, a rectification control system in a liquid oxygen production process is provided, comprising: the data acquisition module is used for acquiring feeding speed values and liquid level values in the tower at a plurality of preset time points in a preset time period; the data time sequence distribution module is used for respectively arranging the feeding rate values and the liquid level values in the tower at a plurality of preset time points into a feeding rate time sequence input vector and a liquid level time sequence input vector in the tower according to the time dimension; the liquid level relative time sequence change module is used for calculating the difference value between the liquid level values in each two adjacent positions in the liquid level time sequence input vector in the tower to obtain the liquid level time sequence change input vector in the tower; the liquid level time sequence association module is used for carrying out association coding on the liquid level time sequence input vector in the tower and the liquid level time sequence change input vector in the tower so as to obtain an absolute and relative multidimensional association matrix of the liquid level in the tower; the liquid level time sequence change multi-scale feature extraction module is used for enabling the in-tower liquid level absolute and relative multi-dimensional correlation matrix to pass through an integrated network model comprising a first convolutional neural network model and a second convolutional neural network model to obtain an in-tower liquid level absolute and relative multi-dimensional correlation feature matrix; the feeding rate time sequence feature extraction module is used for enabling the feeding rate time sequence input vector to pass through the multi-scale neighborhood feature extraction module to obtain a feeding rate time sequence feature vector; the responsiveness transfer module is used for calculating responsiveness transfer between the absolute and relative multidimensional correlation feature matrix of the liquid level in the tower relative to the time sequence feature vector of the feeding rate so as to obtain a classification feature vector; and a feed rate control module for passing the classification feature vector through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the feed rate value at the current time point should be increased or decreased.

In the rectification control system in the liquid oxygen preparation process, the liquid level time sequence association module is used for: performing association coding on the in-tower liquid level time sequence input vector and the in-tower liquid level time sequence change input vector by using the following coding formula to obtain an in-tower liquid level absolute and relative multidimensional association matrix; wherein, the coding formula is:

wherein->

Representing the time sequence input vector of the liquid level in the tower, < >>

A transpose vector representing the timing input vector of the liquid level in the column,>

representing the time sequence change input vector of the liquid level in the tower, < >>

Representing the absolute and relative multidimensional correlation matrix of the liquid level in the tower, < >>

Representing matrix multiplication.

In the rectification control system in the liquid oxygen preparation process, the liquid level time sequence variation multi-scale feature extraction module comprises: a first scale feature extraction unit, configured to perform a convolution process, a pooling process, and a nonlinear activation process based on a two-dimensional convolution kernel on the in-tower liquid level absolute and relative multidimensional correlation matrix in forward transfer of layers by using each layer of a first convolutional neural network model of the integrated network model to output a first scale feature vector from a last layer of the first convolutional neural network model, where the first convolutional neural network model has a two-dimensional convolution kernel of a first scale; a second scale feature extraction unit for performing two-dimensional convolution kernel-based convolution processing, pooling processing and nonlinear activation processing on the in-tower liquid level absolute and relative multidimensional correlation matrix in forward transfer of layers by using layers of a second convolution neural network model of the integrated network model to output a second scale feature vector by a last layer of the second convolution neural network model, wherein the second convolution neural network model has a two-dimensional convolution kernel of a second scale, and the first scale is different from the second scale; and the cascading unit is used for cascading the first scale feature vector and the second scale feature vector to obtain the absolute and relative multidimensional correlation feature matrix of the liquid level in the tower.

In the rectification control system in the liquid oxygen preparation process, the multi-scale neighborhood feature extraction module comprises: and a fusion layer connected with the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer use two-dimensional convolution kernels with different scales.

In the rectification control system in the liquid oxygen preparation process, the feeding rate time sequence feature extraction module comprises: the first scale extraction unit is used for carrying out one-dimensional convolution coding on the feeding rate time sequence input vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following first convolution formula so as to obtain a first scale feeding rate feature vector; wherein the first convolution formula is:

wherein->

For the first one-dimensional convolution kernel>

Width in direction, ++>

For a first one-dimensional convolution kernel parameter vector, +.>

A local vector matrix for operation with a first one-dimensional convolution kernel>

For the size of the first one-dimensional convolution kernel, +.>

Representing the feed rate timing input vector, +.>

Representing one-dimensional convolutional encoding of the feed rate timing input vector; the second scale extraction unit is used for carrying out one-dimensional convolution coding on the feeding rate time sequence input vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following second convolution formula so as to obtain a second scale feeding rate feature vector; wherein the second convolution formula is: / >

Wherein->

For the second one-dimensional convolution kernel>

Width in direction, ++>

For a second one-dimensional convolution kernel parameter vector, +.>

A local vector matrix for operation with a second one-dimensional convolution kernel>

For the size of the second one-dimensional convolution kernel, +.>

Representing the feed rate timing input vector,

representing one-dimensional convolutional encoding of the feed rate timing input vector; and a multi-scale cascading unit, configured to cascade the first-scale feeding rate feature vector and the second-scale feeding rate feature vector by using a fusion layer of the multi-scale neighborhood feature extraction module to obtain the feeding rate time sequence feature vector.

In the rectification control system in the liquid oxygen preparation process, the response transfer module is used for: calculating a response transfer between the absolute and relative multidimensional correlation feature matrices of the liquid level in the tower relative to the time sequence feature vector of the feeding rate according to the following response formula to obtain a classification feature vector; wherein, the responsiveness formula is:

wherein->

Representing the absolute and relative multidimensional correlation characteristic matrix of the liquid level in the tower>

Representing the feed rate timing feature vector, +. >

Representing the classification feature vector.

In the rectification control system in the liquid oxygen preparation process, the system further comprises a training module for training the integrated network model comprising the first convolutional neural network model and the second convolutional neural network model, the multi-scale neighborhood feature extraction module and the classifier; wherein, training module includes: the training data acquisition module is used for acquiring training data, wherein the training data comprises training feeding rate values at a plurality of preset time points in a preset time period and liquid level values in a training tower, and a real value of which the feeding rate value at the current time point is required to be increased or reduced; the training data time sequence distribution module is used for respectively arranging the training feeding rate values and the liquid level values in the training tower at a plurality of preset time points into a training feeding rate time sequence input vector and a training tower liquid level time sequence input vector according to the time dimension; the training liquid level relative time sequence change module is used for calculating the difference value between the liquid level values in the training towers at every two adjacent positions in the liquid level time sequence input vector in the training towers so as to obtain the liquid level time sequence change input vector in the training towers; the training liquid level time sequence association module is used for carrying out association coding on the liquid level time sequence input vector in the training tower and the liquid level time sequence change input vector in the training tower so as to obtain an absolute and relative multidimensional association matrix of the liquid level in the training tower; the training liquid level time sequence change multi-scale feature extraction module is used for enabling the liquid level absolute and relative multi-dimensional correlation matrix in the training tower to pass through the integrated network model comprising the first convolutional neural network model and the second convolutional neural network model so as to obtain the liquid level absolute and relative multi-dimensional correlation feature matrix in the training tower; the training feeding rate time sequence feature extraction module is used for enabling the training feeding rate time sequence input vector to pass through the multi-scale neighborhood feature extraction module to obtain a training feeding rate time sequence feature vector; the training response transfer module is used for calculating the response transfer between the absolute liquid level and relative multidimensional association characteristic matrix in the training tower relative to the training feeding rate time sequence characteristic vector so as to obtain a training classification characteristic vector; the classification loss module is used for passing the training classification feature vector through the classifier to obtain a classification loss function value; the probability distribution shift information compensation loss module is used for calculating probability distribution shift information compensation loss function values of the liquid level absolute and relative multidimensional correlation feature matrix in the training tower and the training feeding rate time sequence feature vector; and a back propagation module for compensating a weighted sum of the classification loss function value and the probability distribution shift information as a loss function value, and training the integrated network model including the first convolutional neural network model and the second convolutional neural network model, the multi-scale neighborhood feature extraction module, and the classifier by back propagation of gradient descent.

In the rectification control system in the liquid oxygen preparation process, the classification loss module comprises: a training result classifying unit, configured to process the training classification feature vector by using the classifier according to the following classification formula to generate a training classification result, where the classification formula is:

wherein->

Representing the training classification feature vector, +.>

To->

Is a weight matrix>

To->

Representing a bias matrix; and a loss function value calculation unit for calculating a cross entropy value between the training classification result and a true value as the classification loss function value.

In the rectification control system in the liquid oxygen preparation process, the probability distribution shift information compensation loss module is used for: calculating the absolute and relative multidimensional correlation characteristics of the liquid level in the training tower according to the following optimization formulaCompensating a loss function value by the matrix and the probability distribution shift information of the training feed rate timing feature vector; wherein, the optimization formula is:

wherein->

And->

The liquid level absolute and relative multidimensional correlation feature matrix in the training tower and the training feeding rate time sequence feature vector are respectively +.>

Representation- >

Function (F)>

Representation->

Function (F)>

Represents a logarithmic function with base 2, +.>

And->

Compensating for shift superparameter, and +.>

For weighting superparameters, < >>

Representing the probability distribution shift information compensation loss function value.

In a second aspect, a rectification control method in a liquid oxygen preparation process is provided, which includes: acquiring feeding speed values and liquid level values in the tower at a plurality of preset time points in a preset time period; arranging the feeding rate values and the liquid level values in the tower at a plurality of preset time points into a feeding rate time sequence input vector and a liquid level time sequence input vector in the tower according to a time dimension respectively; calculating the difference value between the liquid level values in each two adjacent positions in the liquid level time sequence input vector in the tower to obtain a liquid level time sequence change input vector in the tower; performing association coding on the in-tower liquid level time sequence input vector and the in-tower liquid level time sequence change input vector to obtain an in-tower liquid level absolute and relative multidimensional association matrix; the absolute and relative multidimensional correlation matrix of the liquid level in the tower is obtained through an integrated network model comprising a first convolution neural network model and a second convolution neural network model; the feeding rate time sequence input vector is passed through a multi-scale neighborhood feature extraction module to obtain a feeding rate time sequence feature vector; calculating the response transfer between the absolute and relative multidimensional correlation feature matrices of the liquid level in the tower relative to the time sequence feature vector of the feeding rate so as to obtain a classification feature vector; and passing the classification feature vector through a classifier to obtain a classification result, wherein the classification result is used for indicating that the feeding rate value of the current time point is increased or decreased.

Compared with the prior art, the rectification control system and the method thereof in the liquid oxygen preparation process acquire the feeding speed values and the liquid level values in the tower at a plurality of preset time points in a preset time period; and excavating the relevance characteristic distribution information of the time sequence change characteristic of the feeding rate value and the time sequence dynamic change characteristic of the liquid level value in the tower by adopting an artificial intelligence technology based on deep learning, and adaptively adjusting the feeding rate value based on the time sequence cooperative change condition of the feeding rate value and the liquid level value in the tower so as to optimize the rectifying efficiency and the operation stability.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments or the description of the prior art will be briefly introduced below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is an application scenario diagram of a rectification control system in a liquid oxygen production process according to an embodiment of the present application.

Fig. 2 is a block diagram of a rectification control system in a liquid oxygen production process according to an embodiment of the present application.

FIG. 3 is a block diagram of the liquid level time varying multi-scale feature extraction module in the rectification control system in the liquid oxygen production process according to an embodiment of the present application.

Fig. 4 is a block diagram of the feed rate timing feature extraction module in the rectification control system in the liquid oxygen production process according to an embodiment of the present application.

Fig. 5 is a block diagram of the training module in the rectification control system in the liquid oxygen production process according to an embodiment of the present application.

Fig. 6 is a flow chart of a rectification control method in a liquid oxygen production process according to an embodiment of the present application.

Fig. 7 is a schematic diagram of a system architecture of a rectification control method in a liquid oxygen production process according to an embodiment of the present application.

Detailed Description

The following description of the technical solutions in the embodiments of the present application will be made with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.

Unless defined otherwise, all technical and scientific terms used in the examples of this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to limit the scope of the present application.

In the description of the embodiments of the present application, unless otherwise indicated and defined, the term "connected" should be construed broadly, and for example, may be an electrical connection, may be a communication between two elements, may be a direct connection, or may be an indirect connection via an intermediary, and it will be understood by those skilled in the art that the specific meaning of the term may be understood according to the specific circumstances.

It should be noted that, the term "first\second\third" in the embodiments of the present application is merely to distinguish similar objects, and does not represent a specific order for the objects, it is to be understood that "first\second\third" may interchange a specific order or sequence where allowed. It is to be understood that the "first\second\third" distinguishing objects may be interchanged where appropriate such that the embodiments of the present application described herein may be implemented in sequences other than those illustrated or described herein.

In the process of preparing liquid oxygen, rectification control is an important link. The process generally includes the steps of: heating: heating the liquid air to a temperature above the boiling point of the liquid air to enable the liquid air to be changed into a gaseous state; compression: compressing the gaseous air to a suitable pressure; and (3) cooling: the gaseous air is changed into liquid state again through a condenser, and liquid oxygen and liquid nitrogen are separated; and (3) rectifying: and (3) rectifying by using tower type separation equipment to separate liquid oxygen from liquid nitrogen. During the rectification process, the temperature and pressure within the column need to be controlled to ensure that the liquid oxygen and liquid nitrogen can be effectively separated. In addition, during the rectification process, the parameters such as the feeding rate, the liquid level in the tower and the like need to be controlled carefully so as to ensure the stability and the safety of operation.

Reasons for attention to control of the feed rate and column liquid level parameters during rectification include: feed rate: if the feeding rate is too high, uneven distribution of liquid in the tower is caused, and the separation effect is affected; if the feed rate is too slow, the production efficiency is lowered. It is therefore necessary to grasp the appropriate feed rate to ensure that both the separation effect and the production efficiency are satisfied. Liquid level in column: if the liquid level in the tower is too high, the safety problems such as pipeline blockage, overflow and the like can be caused; if the liquid level is too low, the separation effect is affected. It is therefore necessary to adjust the liquid level in the column by controlling the feed rate, the recovered liquid oxygen, liquid nitrogen, etc. to ensure the stability and safety of operation.

As mentioned above, for feed rates, if the feed rate is too fast, this will result in maldistribution of liquid in the column, affecting the separation effect; if the feed rate is too slow, the production efficiency is lowered. For the liquid level in the tower, if the liquid level in the tower is too high, the safety problems such as pipeline blockage, overflow and the like can be caused; if the liquid level is too low, the separation effect is affected. However, in the conventional rectification parameter control scheme, only the respective parameter data are compared with the threshold value separately to complete the corresponding control, and the cooperative suitability of the respective parameter data is not concerned, so that the preparation efficiency and the operation stability of liquid oxygen are difficult to meet the requirements. Accordingly, an optimized rectification control system in a liquid oxygen production process is desired.

Accordingly, considering that in the process of controlling the parameters of the actual rectification, due to the fact that the feeding rate value and the liquid level value in the tower have a time sequence cooperative association change rule, the corresponding self-adaptive control cannot be performed based on a single data change condition. If the control of the rectification parameters is to be accurately performed in real time, so as to optimize the rectification efficiency and the operation stability, it is necessary to sufficiently express the time-series cooperative correlation characteristics of the feed rate value and the column internal liquid level value. However, since the characteristic of the change of the liquid level value in the tower in the time dimension is characteristic information of a small scale, the capability of detecting the characteristic of the change of the small scale is weaker when characteristic capture is actually performed, that is, the expression precision of the time sequence cooperative correlation characteristic of the feeding rate value and the liquid level value in the tower is reduced. Therefore, in this process, it is difficult to mine the correlation characteristic distribution information of the time-series variation characteristic of the feed rate value and the time-series dynamic variation characteristic of the in-column liquid level value to adaptively adjust the feed rate value based on the time-series cooperative variation condition of the feed rate value and the in-column liquid level value to optimize the efficiency and the operation stability of rectification.

In recent years, deep learning and neural networks have been widely used in the fields of computer vision, natural language processing, text signal processing, and the like. The development of deep learning and neural networks provides new solutions and schemes for mining the time sequence change characteristics of the feeding rate values and the correlation characteristic distribution information of the time sequence dynamic change characteristics of the liquid level values in the towers. Those of ordinary skill in the art will appreciate that a deep learning based deep neural network model may adjust parameters of the deep neural network model by appropriate training strategies, such as by a gradient descent back-propagation algorithm, to enable modeling of complex nonlinear correlations between things, which is obviously suitable for modeling and mining correlation feature distribution information of time-varying features of the feed rate values and time-varying dynamic features of the column level values.

Specifically, in the technical scheme of the present application, first, a feed rate value and a column internal liquid level value at a plurality of predetermined time points within a predetermined period of time are obtained. Next, considering that the feeding rate value and the liquid level value in the tower have dynamic change rules in the time dimension, in order to extract change characteristic information of the feeding rate value and the liquid level value in the tower in the time dimension respectively, in the technical scheme of the application, the feeding rate value and the liquid level value in the tower at a plurality of preset time points are respectively arranged into a feeding rate time sequence input vector and a liquid level time sequence input vector according to the time dimension, so that distribution information of the feeding rate value and the liquid level value in the tower in the time sequence is respectively integrated.

Further, in order to adaptively and accurately control the feeding rate value, it is necessary to extract the dynamic change feature of the liquid level value in the tower in the time dimension, and considering that the change information of the liquid level value in the tower in the time dimension is weak, the weak change feature is small-scale change feature information relative to the liquid level value in the tower, if the time sequence dynamic change feature extraction of the liquid level value in the tower is performed by absolute change information, the calculated amount is large, the overfitting is caused, and the small-scale weak change feature of the liquid level value in the tower in the time dimension is difficult to be perceived, so that the accuracy of subsequent classification is affected.

Based on the above, in the technical scheme of the application, the time sequence dynamic characteristic extraction of the liquid level value in the tower is comprehensively carried out by adopting the time sequence relative change characteristic and the absolute change characteristic of the liquid level value in the tower. Specifically, first, a difference value between the in-tower liquid level values of every two adjacent positions in the in-tower liquid level time sequence input vector is calculated to obtain an in-tower liquid level time sequence change input vector. Then, consider that there is an association relation between the time series relative change characteristic and the time series absolute change characteristic of the liquid level value in the tower with respect to the dynamic change of the liquid level in the tower. Therefore, in order to fully explore the dynamic change rule of the liquid level value in the tower in the time dimension so as to accurately control the feeding rate, in the technical scheme of the application, the liquid level time sequence input vector in the tower and the liquid level time sequence change input vector in the tower are further subjected to associated coding so as to obtain an absolute and relative multidimensional association matrix of the liquid level in the tower.

Then, it is considered that the multi-dimensional time-series related information due to the absolute and relative amounts of the in-column liquid level values has uncertainty in the time dimension, that is, the time-series related information of the absolute and relative amounts of the in-column liquid level values has different dynamic change laws at different time period spans within the predetermined period. Therefore, in the technical scheme of the application, the absolute and relative multidimensional correlation matrix of the liquid level in the tower is obtained through an integrated network model comprising a first convolution neural network model and a second convolution neural network model. In particular, here, the first convolutional neural network model and the second convolutional neural network model have two-dimensional convolutional kernels of different sizes, so as to extract multi-scale time-series variation characteristic information of time-series related information of absolute quantity and relative quantity of the liquid level value in the tower in a time dimension.

For the feed rate values at the plurality of predetermined time points, the feed rate values also have different dynamic change characteristic information over different time period spans within the predetermined time period. In order to fully express the time sequence dynamic change characteristics of the feeding rate value, in the technical scheme of the application, the feeding rate time sequence input vector is encoded by a multi-scale neighborhood characteristic extraction module so as to extract dynamic multi-scale neighborhood associated characteristics of the feeding rate value under different time spans, and thus a feeding rate time sequence characteristic vector is obtained.

Further, the response transfer between the absolute and relative multidimensional correlation feature matrix of the liquid level in the tower relative to the time sequence feature vector of the feeding rate is further calculated, so that the correlation feature distribution information between the time sequence multiscale dynamic correlation feature of the absolute quantity and the relative quantity of the liquid level value in the tower and the dynamic multiscale time sequence change feature of the feeding rate value, namely, the suitability feature information between the time sequence change feature of the liquid level value in the tower and the time sequence change feature of the feeding rate value is represented, and the classification feature vector is obtained.

And then, classifying the classifying feature vector by a classifier to obtain a classifying result which is used for indicating whether the feeding speed value of the current time point should be increased or decreased. That is, in the technical solution of the present application, the label of the classifier includes that the feeding rate value at the current time point should be increased (first label) and that the feeding rate value at the current time point should be decreased (second label), wherein the classifier determines to which classification label the classification feature vector belongs through a soft maximum function. It is noted that the first tag p1 and the second tag p2 do not contain the concept of artificial setting, and in fact, during the training process, the computer model does not have the concept of "the feeding rate value at the current time point should be increased or should be decreased", which is only two kinds of classification tags and the probability that the output feature is under the two classification tags, i.e. the sum of p1 and p2 is one. Thus, the classification result that the feeding rate value should be increased or decreased is actually converted into a classification probability distribution conforming to the classification rule of the natural law by classifying the labels, and the physical meaning of the natural probability distribution of the labels is essentially used instead of the language text meaning that the feeding rate value at the current time point should be increased or decreased. It should be understood that, in the technical solution of the present application, the classification label of the classifier is a control policy label that the feeding rate value of the current time point should be increased or decreased, so after the classification result is obtained, the feeding rate value of the current time point may be adaptively adjusted to be increased or decreased based on the classification result, so as to optimize the efficiency and the operation stability of rectification.

In particular, in the technical solution of the present application, when the classification feature vector is obtained by calculating the transfer of the responsiveness between the absolute and relative multidimensional correlation feature matrix of the liquid level in the column and the time sequence feature vector of the feeding rate, the absolute and relative multidimensional correlation feature matrix of the liquid level in the column and the time sequence feature vector of the feeding rate are substantially subjected to distribution fusion based on the responsiveness of the feature distribution therebetween, so that when the feature distributions of the absolute and relative multidimensional correlation feature matrix of the liquid level in the column and the time sequence feature vector of the feeding rate are fused, the degradation problem of the respective feature expression information is encountered when the respective feature distribution of the absolute and relative multidimensional correlation feature matrix of the liquid level in the column and the time sequence feature vector of the feeding rate are propagated backward through the classifier in the parameter space of the model.

Based on this, the applicant of the present application introduced a correlation matrix of absolute and relative multidimensional for the liquid level in the column, for example denoted as

And said feed rate timing feature vector, e.g. denoted +.>

The probability distribution shift information compensation loss function of (2) is expressed as: />

Wherein->

And->

Compensating for shift superparameter, and +.>

Is a weighted superparameter.

Here, based on

Function is related to the characteristic matrix from absolute and relative multidimensional of the liquid level in the tower >

And said feed rate timing feature vector +.>

The respectively derived class probability values themselves follow probability distributions for the respective feature distributions, and the probability distribution shift information is used to compensate a loss function for the absolute and relative multidimensional correlation feature matrix +_ of the liquid level in the tower>

And said feed rate timing feature vector +.>

The information compensation is carried out by shifting the probability distribution of the characteristic representation, and the cross information entropy brought by compensation is maximized through a bool function, so that the characteristic distribution of the classified characteristic vector after fusion can restore the characteristic expression information of the absolute and relative multidimensional correlation characteristic matrix of the liquid level in the tower and the feeding rate time sequence characteristic vector before fusion to the greatest extent, and the accuracy of the classification result obtained by the classified characteristic vector through the classifier is improved. In this way, the feed rate value can be adaptively adjusted based on the time series cooperative variation of the feed rate and the liquid level in the column to optimize the efficiency and operational stability of the rectification.

Fig. 1 is an application scenario diagram of a rectification control system in a liquid oxygen production process according to an embodiment of the present application. As shown in fig. 1, in this application scenario, first, a feed rate value (e.g., C1 as illustrated in fig. 1) and an in-column liquid level value (e.g., C2 as illustrated in fig. 1) at a plurality of predetermined time points within a predetermined period of time are acquired; the obtained feed rate value and the in-column liquid level value are then input to a server (e.g., S as illustrated in fig. 1) deployed with a rectification control algorithm in the liquid oxygen production process, wherein the server is capable of processing the feed rate value and the in-column liquid level value based on the rectification control algorithm in the liquid oxygen production process to generate a classification result indicating that the feed rate value at the current point in time should be increased or should be decreased.

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

In one embodiment of the present application, fig. 2 is a block diagram of a rectification control system in a liquid oxygen production process according to an embodiment of the present application. As shown in fig. 2, a rectification control system 100 in a liquid oxygen production process according to an embodiment of the present application includes: a data acquisition module 110 for acquiring a feed rate value and a liquid level value in the tower at a plurality of predetermined time points within a predetermined time period; the data timing distribution module 120 is configured to arrange the feeding rate values and the in-tower liquid level values at the plurality of predetermined time points into a feeding rate timing input vector and an in-tower liquid level timing input vector according to a time dimension, respectively; the liquid level relative time sequence change module 130 is configured to calculate a difference value between the liquid level values in each two adjacent positions in the liquid level time sequence input vector in the tower to obtain a liquid level time sequence change input vector in the tower; the liquid level time sequence association module 140 is used for performing association coding on the in-tower liquid level time sequence input vector and the in-tower liquid level time sequence change input vector to obtain an in-tower liquid level absolute and relative multidimensional association matrix; the liquid level time sequence change multi-scale feature extraction module 150 is configured to pass the in-tower liquid level absolute and relative multi-dimensional correlation matrix through an integrated network model including a first convolutional neural network model and a second convolutional neural network model to obtain an in-tower liquid level absolute and relative multi-dimensional correlation feature matrix; a feeding rate timing feature extraction module 160, configured to pass the feeding rate timing input vector through a multi-scale neighborhood feature extraction module to obtain a feeding rate timing feature vector; a responsiveness transfer module 170 for calculating responsiveness transfer between the absolute and relative multi-dimensional correlation feature matrices for the liquid level in the column relative to the feed rate timing feature vector to obtain a classification feature vector; and a feed rate control module 180 for passing the classification feature vector through a classifier to obtain a classification result, wherein the classification result is used for indicating that the value of the feed rate at the current time point should be increased or decreased.

Specifically, in the embodiment of the present application, the data acquisition module 110 is configured to obtain the feeding rate value and the liquid level value in the tower at a plurality of predetermined time points within a predetermined time period. As mentioned above, for feed rates, if the feed rate is too fast, this will result in maldistribution of liquid in the column, affecting the separation effect; if the feed rate is too slow, the production efficiency is lowered. For the liquid level in the tower, if the liquid level in the tower is too high, the safety problems such as pipeline blockage, overflow and the like can be caused; if the liquid level is too low, the separation effect is affected. However, in the conventional rectification parameter control scheme, only the respective parameter data are compared with the threshold value separately to complete the corresponding control, and the cooperative suitability of the respective parameter data is not concerned, so that the preparation efficiency and the operation stability of liquid oxygen are difficult to meet the requirements. Accordingly, an optimized rectification control system in a liquid oxygen production process is desired.

Specifically, in the technical scheme of the present application, first, a feed rate value and a column internal liquid level value at a plurality of predetermined time points within a predetermined period of time are obtained.

Specifically, in the embodiment of the present application, the data timing distribution module 120 is configured to arrange the feeding rate values and the in-tower liquid level values at the plurality of predetermined time points into a feeding rate timing input vector and an in-tower liquid level timing input vector according to a time dimension, respectively. Next, considering that the feeding rate value and the liquid level value in the tower have dynamic change rules in the time dimension, in order to extract change characteristic information of the feeding rate value and the liquid level value in the tower in the time dimension respectively, in the technical scheme of the application, the feeding rate value and the liquid level value in the tower at a plurality of preset time points are respectively arranged into a feeding rate time sequence input vector and a liquid level time sequence input vector according to the time dimension, so that distribution information of the feeding rate value and the liquid level value in the tower in the time sequence is respectively integrated.

Specifically, in the embodiment of the present application, the relative liquid level timing change module 130 and the liquid level timing correlation module 140 are configured to calculate a difference value between the liquid level values in each two adjacent positions in the in-tower liquid level timing input vector to obtain an in-tower liquid level timing change input vector; and the correlation coding is used for carrying out correlation coding on the in-tower liquid level time sequence input vector and the in-tower liquid level time sequence change input vector so as to obtain an in-tower liquid level absolute and relative multidimensional correlation matrix.

Wherein, the liquid level timing correlation module 140 is configured to: performing association coding on the in-tower liquid level time sequence input vector and the in-tower liquid level time sequence change input vector by using the following coding formula to obtain an in-tower liquid level absolute and relative multidimensional association matrix; wherein, the coding formula is:

Wherein->

Representing matrix multiplication.

Specifically, in the embodiment of the present application, the liquid level time sequence variation multi-scale feature extraction module 150 is configured to pass the in-tower liquid level absolute and relative multi-dimensional correlation matrix through an integrated network model including a first convolutional neural network model and a second convolutional neural network model to obtain an in-tower liquid level absolute and relative multi-dimensional correlation feature matrix. Then, it is considered that the multi-dimensional time-series related information due to the absolute and relative amounts of the in-column liquid level values has uncertainty in the time dimension, that is, the time-series related information of the absolute and relative amounts of the in-column liquid level values has different dynamic change laws at different time period spans within the predetermined period.

Therefore, in the technical scheme of the application, the absolute and relative multidimensional correlation matrix of the liquid level in the tower is obtained through an integrated network model comprising a first convolution neural network model and a second convolution neural network model. In particular, here, the first convolutional neural network model and the second convolutional neural network model have two-dimensional convolutional kernels of different sizes, so as to extract multi-scale time-series variation characteristic information of time-series related information of absolute quantity and relative quantity of the liquid level value in the tower in a time dimension.

Fig. 3 is a block diagram of the liquid level time sequence variation multi-scale feature extraction module in the rectification control system in the liquid oxygen preparation process according to the embodiment of the present application, as shown in fig. 3, the liquid level time sequence variation multi-scale feature extraction module 150 includes: a first scale feature extraction unit 151 for performing a two-dimensional convolution kernel-based convolution process, a pooling process, and a nonlinear activation process on the in-tower liquid level absolute and relative multidimensional correlation matrix in forward transfer of layers using each layer of a first convolution neural network model of the integrated network model to output a first scale feature vector from a last layer of the first convolution neural network model, wherein the first convolution neural network model has a two-dimensional convolution kernel of a first scale; a second scale feature extraction unit 152, configured to perform a convolution process, a pooling process, and a nonlinear activation process on the in-tower liquid level absolute and relative multidimensional correlation matrix in forward transfer of layers using layers of a second convolutional neural network model of the integrated network model to output a second scale feature vector by a last layer of the second convolutional neural network model, where the second convolutional neural network model has a two-dimensional convolution kernel of a second scale, and the first scale is different from the second scale; and a cascading unit 153, configured to cascade the first scale feature vector and the second scale feature vector to obtain the in-tower liquid level absolute and relative multidimensional correlation feature matrix.

The convolutional neural network (Convolutional Neural Network, CNN) is an artificial neural network and has wide application in the fields of image recognition and the like. The convolutional neural network may include an input layer, a hidden layer, and an output layer, where the hidden layer may include a convolutional layer, a pooling layer, an activation layer, a full connection layer, etc., where the previous layer performs a corresponding operation according to input data, outputs an operation result to the next layer, and obtains a final result after the input initial data is subjected to a multi-layer operation.

The convolutional neural network model has excellent performance in the aspect of image local feature extraction by taking a convolutional kernel as a feature filtering factor, and has stronger feature extraction generalization capability and fitting capability compared with the traditional image feature extraction algorithm based on statistics or feature engineering.

Specifically, in the embodiment of the present application, the feeding rate timing feature extraction module 160 is configured to pass the feeding rate timing input vector through a multi-scale neighborhood feature extraction module to obtain a feeding rate timing feature vector. For the feed rate values at the plurality of predetermined time points, the feed rate values also have different dynamic change characteristic information over different time period spans within the predetermined time period. In order to fully express the time sequence dynamic change characteristics of the feeding rate value, in the technical scheme of the application, the feeding rate time sequence input vector is encoded by a multi-scale neighborhood characteristic extraction module so as to extract dynamic multi-scale neighborhood associated characteristics of the feeding rate value under different time spans, and thus a feeding rate time sequence characteristic vector is obtained.

The multi-scale neighborhood feature extraction module comprises: and a fusion layer connected with the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer use two-dimensional convolution kernels with different scales.

Wherein, fig. 4 is a block diagram of the feed rate timing feature extraction module in the rectification control system in the liquid oxygen preparation process according to the embodiment of the present application, as shown in fig. 4, the feed rate timing feature extraction module 160 includes: a first scale extraction unit 161 for usingThe first convolution layer of the multi-scale neighborhood feature extraction module carries out one-dimensional convolution coding on the feeding rate time sequence input vector by using a first convolution formula to obtain a first-scale feeding rate feature vector; wherein the first convolution formula is:

wherein->

For the first one-dimensional convolution kernel>

Width in direction, ++>

For a first one-dimensional convolution kernel parameter vector, +.>

For the size of the first one-dimensional convolution kernel, +.>

Representing the feed rate timing input vector, +. >

Representing one-dimensional convolutional encoding of the feed rate timing input vector; a second scale extraction unit 162, configured to perform one-dimensional convolution encoding on the feeding rate time sequence input vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following second convolution formula to obtain a second scale feeding rate feature vector; wherein the second convolution formula is: />

Wherein->

For the second dimensionConvolution kernel at +.>

Width in direction, ++>

For a second one-dimensional convolution kernel parameter vector, +.>

For the size of the second one-dimensional convolution kernel, +.>

Representing the feed rate timing input vector,

representing one-dimensional convolutional encoding of the feed rate timing input vector; and a multi-scale cascading unit 163, configured to cascade the first-scale feeding rate feature vector and the second-scale feeding rate feature vector using a fusion layer of the multi-scale neighborhood feature extraction module to obtain the feeding rate timing feature vector.

It should be noted that the multi-scale neighborhood feature extraction module is essentially a deep neural network model based on deep learning, which is capable of fitting any function by a predetermined training strategy and has a higher feature extraction generalization capability compared to the conventional feature engineering.

The multi-scale neighborhood feature extraction module comprises a plurality of parallel one-dimensional convolution layers, wherein in the process of feature extraction by the multi-scale neighborhood feature extraction module, the plurality of parallel one-dimensional convolution layers respectively carry out one-dimensional convolution coding on input data by one-dimensional convolution check with different scales so as to capture local implicit features of a sequence.

Specifically, in the embodiment of the present application, the responsiveness transfer module 170 is configured to calculate responsiveness transfer between the absolute and relative multidimensional correlation feature matrices of the liquid level in the column with respect to the time-series feature vector of the feeding rate to obtain the classification feature vector. Further, calculating the response transfer between the absolute and relative multidimensional correlation feature matrix of the liquid level in the tower relative to the time sequence feature vector of the feeding rate, so as to represent the correlation feature distribution information between the time sequence multiscale dynamic correlation feature of the absolute quantity and the relative quantity of the liquid level value in the tower and the dynamic multiscale time sequence change feature of the feeding rate value, namely, the suitability feature information between the time sequence change feature of the liquid level value in the tower and the time sequence change feature of the feeding rate value, thereby obtaining the classification feature vector.

Wherein, the responsiveness transferring module 170 is configured to: calculating a response transfer between the absolute and relative multidimensional correlation feature matrices of the liquid level in the tower relative to the time sequence feature vector of the feeding rate according to the following response formula to obtain a classification feature vector; wherein, the responsiveness formula is:

wherein->

Representing the feed rate timing feature vector, +.>

Representing the classification feature vector.

Specifically, in the embodiment of the present application, the feeding rate control module 180 is configured to pass the classification feature vector through a classifier to obtain a classification result, where the classification result is used to indicate that the feeding rate value at the current time point should be increased or should be decreased. And then, classifying the classifying feature vector by a classifier to obtain a classifying result which is used for indicating whether the feeding speed value of the current time point should be increased or decreased.

That is, in the technical solution of the present application, the label of the classifier includes that the feeding rate value at the current time point should be increased (first label) and that the feeding rate value at the current time point should be decreased (second label), wherein the classifier determines to which classification label the classification feature vector belongs through a soft maximum function. It is noted that the first tag p1 and the second tag p2 do not contain the concept of artificial setting, and in fact, during the training process, the computer model does not have the concept of "the feeding rate value at the current time point should be increased or should be decreased", which is only two kinds of classification tags and the probability that the output feature is under the two classification tags, i.e. the sum of p1 and p2 is one.

Thus, the classification result that the feeding rate value should be increased or decreased is actually converted into a classification probability distribution conforming to the classification rule of the natural law by classifying the labels, and the physical meaning of the natural probability distribution of the labels is essentially used instead of the language text meaning that the feeding rate value at the current time point should be increased or decreased. It should be understood that, in the technical solution of the present application, the classification label of the classifier is a control policy label that the feeding rate value of the current time point should be increased or decreased, so after the classification result is obtained, the feeding rate value of the current time point may be adaptively adjusted to be increased or decreased based on the classification result, so as to optimize the efficiency and the operation stability of rectification.

The feed rate control module 180 includes: the coding unit is used for carrying out full-connection coding on the classification characteristic vectors by using a plurality of full-connection layers of the classifier so as to obtain coded classification characteristic vectors; and the classification unit is used for passing the coding classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

Further, the rectification control system in the liquid oxygen preparation process further comprises a training module for training the integrated network model comprising the first convolutional neural network model and the second convolutional neural network model, the multi-scale neighborhood feature extraction module and the classifier; fig. 5 is a block diagram of the training module in the rectification control system in the liquid oxygen preparation process according to the embodiment of the present application, as shown in fig. 5, the training module 190 includes: a training data acquisition module 1901 for acquiring training data, wherein the training data comprises training feeding rate values and liquid level values in a training tower at a plurality of preset time points in a preset time period, and a true value that the feeding rate value at the current time point should be increased or decreased; the training data time sequence distribution module 1902 is configured to arrange the training feeding rate values and the liquid level values in the training tower at the plurality of predetermined time points into a training feeding rate time sequence input vector and a training liquid level time sequence input vector according to a time dimension respectively; a training liquid level relative time sequence change module 1903, configured to calculate a difference value between liquid level values in the training tower at every two adjacent positions in the training tower liquid level time sequence input vector to obtain a training tower liquid level time sequence change input vector; a training liquid level time sequence association module 1904, configured to perform association encoding on the training tower internal liquid level time sequence input vector and the training tower internal liquid level time sequence variation input vector to obtain a training tower internal liquid level absolute and relative multidimensional association matrix; a training liquid level time sequence change multi-scale feature extraction module 1905, configured to pass the training tower internal liquid level absolute and relative multi-dimensional correlation matrix through the integrated network model including the first convolutional neural network model and the second convolutional neural network model to obtain a training tower internal liquid level absolute and relative multi-dimensional correlation feature matrix; a training feed rate timing feature extraction module 1906 for passing the training feed rate timing input vector through the multi-scale neighborhood feature extraction module to obtain a training feed rate timing feature vector; a training responsiveness transfer module 1907, configured to calculate responsiveness transfer between the absolute and relative multidimensional correlation feature matrices of the liquid level in the training tower relative to the training feed rate timing feature vector to obtain a training classification feature vector; a classification loss module 1908, configured to pass the training classification feature vector through the classifier to obtain a classification loss function value; a probability distribution shift information compensation loss module 1909 for calculating a probability distribution shift information compensation loss function value of the training tower internal liquid level absolute and relative multidimensional correlation feature matrix and the training feeding rate time sequence feature vector; and a back propagation module 1910 for compensating a weighted sum of the classification loss function value and the probability distribution shift information for the loss function value as a loss function value, and training the integrated network model including the first convolutional neural network model and the second convolutional neural network model, the multi-scale neighborhood feature extraction module, and the classifier by back propagation of gradient descent.

Wherein the classification loss module 1908 comprises: a training result classifying unit, configured to process the training classification feature vector by using the classifier according to the following classification formula to generate a training classification result, where the classification formula is:

wherein->

Representing the training classification feature vector, +.>

To the point of

Is a weight matrix>

To->

And said feed rate timing feature vector, e.g. denoted +.>

The probability distribution shift information compensation loss function of (2) is expressed as: calculating the probability distribution shift information compensation loss function value of the liquid level absolute and relative multidimensional association characteristic matrix in the training tower and the training feeding rate time sequence characteristic vector according to the following optimization formula; wherein, the optimization formula is: />

Wherein->

And

Representation->

Function (F)>

Representation->

Function (F)>

Represents a logarithmic function with base 2, +.>

And->

Compensating for shift superparameter, and +.>

For weighting superparameters, < >>

Here, the characteristic matrix is correlated from absolute and relative multidimensional of the liquid level in the tower based on a Softmax function

And said feed rate timing feature vector +.>

The respectively derived class probability values themselves follow probability distributions for the respective feature distributions, and the probability distribution shift information is used to compensate a loss function for the absolute and relative multidimensional correlation feature matrix +_ of the liquid level in the tower >

And said feed rate timing feature vector +.>

Information compensation is performed by shifting probability distribution of feature representation, and cross information entropy brought by compensation is maximized through a bool function, so that feature distribution of the classification feature vector after fusion can restore the absolute and relative multidimensional correlation feature matrix of the liquid level in the tower and the characteristic of the feeding rate time sequence feature vector before fusion to the maximum extentAnd the sign expression information is used for improving the accuracy of a classification result obtained by the classification feature vector through the classifier. In this way, the feed rate value can be adaptively adjusted based on the time series cooperative variation of the feed rate and the liquid level in the column to optimize the efficiency and operational stability of the rectification.

In summary, a rectification control system 100 in a liquid oxygen production process in accordance with embodiments of the present application is illustrated that obtains feed rate values and column internal liquid level values at a plurality of predetermined time points over a predetermined period of time; and excavating the relevance characteristic distribution information of the time sequence change characteristic of the feeding rate value and the time sequence dynamic change characteristic of the liquid level value in the tower by adopting an artificial intelligence technology based on deep learning, and adaptively adjusting the feeding rate value based on the time sequence cooperative change condition of the feeding rate value and the liquid level value in the tower so as to optimize the rectifying efficiency and the operation stability.

As described above, the rectification control system 100 in the liquid oxygen production process according to the embodiment of the present application can be implemented in various terminal devices, such as a server or the like for rectification control in the liquid oxygen production process. In one example, the rectification control system 100 in a liquid oxygen production process according to embodiments of the present application may be integrated into a terminal device as a software module and/or hardware module. For example, the rectification control system 100 in the liquid oxygen production process may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the rectification control system 100 in the liquid oxygen production process can also be one of a plurality of hardware modules of the terminal device.

Alternatively, in another example, the rectification control system 100 in the liquid oxygen production process and the terminal device may be separate devices, and the rectification control system 100 in the liquid oxygen production process may be connected to the terminal device through a wired and/or wireless network, and transmit the interaction information according to an agreed data format.

In one embodiment of the present application, fig. 6 is a flow chart of a rectification control method in a liquid oxygen production process according to an embodiment of the present application. As shown in fig. 6, a rectification control method in a liquid oxygen production process according to an embodiment of the present application includes: 210, obtaining feed rate values and liquid level values in the tower at a plurality of preset time points in a preset time period; 220, arranging the feeding rate values and the liquid level values in the tower at the preset time points into a feeding rate time sequence input vector and a liquid level time sequence input vector in the tower according to the time dimension respectively; 230, calculating a difference value between the liquid level values in each two adjacent positions in the liquid level time sequence input vector in the tower to obtain a liquid level time sequence change input vector in the tower; 240, performing association coding on the in-tower liquid level time sequence input vector and the in-tower liquid level time sequence change input vector to obtain an in-tower liquid level absolute and relative multidimensional association matrix; 250, passing the in-tower liquid level absolute and relative multi-dimensional correlation matrix through an integrated network model comprising a first convolutional neural network model and a second convolutional neural network model to obtain an in-tower liquid level absolute and relative multi-dimensional correlation feature matrix; 260, passing the feed rate timing input vector through a multi-scale neighborhood feature extraction module to obtain a feed rate timing feature vector; 270, calculating a transfer of responsiveness between the absolute and relative multi-dimensional correlation feature matrices for the liquid level in the column relative to the feed rate timing feature vector to obtain a classification feature vector; and, 280, passing the classification feature vector through a classifier to obtain a classification result, wherein the classification result is used for representing that the feeding rate value of the current time point is increased or decreased.

Fig. 7 is a schematic diagram of a system architecture of a rectification control method in a liquid oxygen production process according to an embodiment of the present application. As shown in fig. 7, in the system architecture of the rectification control method in the liquid oxygen production process, first, the feed rate values and the column liquid level values at a plurality of predetermined time points within a predetermined period of time are obtained; then, arranging the feeding rate values and the liquid level values in the tower at a plurality of preset time points into a feeding rate time sequence input vector and a liquid level time sequence input vector in the tower according to a time dimension respectively; then, calculating the difference value between the liquid level values in each two adjacent positions in the liquid level time sequence input vector in the tower to obtain a liquid level time sequence change input vector in the tower; then, carrying out association coding on the in-tower liquid level time sequence input vector and the in-tower liquid level time sequence change input vector to obtain an in-tower liquid level absolute and relative multidimensional association matrix; then, the absolute and relative multidimensional correlation matrix of the liquid level in the tower passes through an integrated network model comprising a first convolution neural network model and a second convolution neural network model to obtain an absolute and relative multidimensional correlation characteristic matrix of the liquid level in the tower; then, the feeding rate time sequence input vector passes through a multi-scale neighborhood feature extraction module to obtain a feeding rate time sequence feature vector; then, calculating the response transfer between the absolute and relative multidimensional correlation feature matrices of the liquid level in the tower relative to the time sequence feature vector of the feeding rate so as to obtain a classification feature vector; and finally, the classification feature vector is passed through a classifier to obtain a classification result, wherein the classification result is used for representing that the feeding rate value of the current time point is increased or decreased.

In a specific example, in the rectification control method in the liquid oxygen preparation process, performing association coding on the in-tower liquid level time sequence input vector and the in-tower liquid level time sequence change input vector to obtain an in-tower liquid level absolute and relative multidimensional association matrix, including: performing association coding on the in-tower liquid level time sequence input vector and the in-tower liquid level time sequence change input vector by using the following coding formula to obtain an in-tower liquid level absolute and relative multidimensional association matrix; wherein, the coding formula is:

wherein->

Representing matrix multiplication.

In a specific example, in the rectification control method in the liquid oxygen preparation process, the method for obtaining the in-tower liquid level absolute and relative multidimensional correlation feature matrix by passing the in-tower liquid level absolute and relative multidimensional correlation matrix through an integrated network model comprising a first convolutional neural network model and a second convolutional neural network model comprises the following steps: performing two-dimensional convolution kernel-based convolution processing, pooling processing and nonlinear activation processing on the in-tower liquid level absolute and relative multidimensional correlation matrix in forward transfer of layers by using layers of a first convolution neural network model of the integrated network model to output a first scale feature vector by the last layer of the first convolution neural network model, wherein the first convolution neural network model has a two-dimensional convolution kernel of a first scale; performing two-dimensional convolution kernel-based convolution processing, pooling processing and nonlinear activation processing on the in-tower liquid level absolute and relative multidimensional correlation matrix in forward transfer of layers by using layers of a second convolution neural network model of the integrated network model to output a second scale feature vector by a last layer of the second convolution neural network model, wherein the second convolution neural network model has a two-dimensional convolution kernel of a second scale, and the first scale is different from the second scale; and cascading the first scale feature vector and the second scale feature vector to obtain the absolute and relative multidimensional correlation feature matrix of the liquid level in the tower.

In a specific example, in the rectification control method in the liquid oxygen preparation process, the multi-scale neighborhood feature extraction module includes: and a fusion layer connected with the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer use two-dimensional convolution kernels with different scales.

In a specific example, in the rectification control method in the liquid oxygen preparation process, the feeding rate time sequence input vector is passed through a multi-scale neighborhood feature extraction module to obtain a feeding rate time sequence feature vector, which includes: performing one-dimensional convolution encoding on the feeding rate time sequence input vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following first convolution formula to obtain a first-scale feeding rate feature vector; wherein the first convolution formula is:

wherein->

For the first one-dimensional convolution kernel>

Width in direction, ++>

For a first one-dimensional convolution kernel parameter vector, +.>

For a local vector matrix operating with a first one-dimensional convolution kernel,

for the size of the first one-dimensional convolution kernel, +. >

Representing the feed rate timing input vector, +.>

Representing one-dimensional convolutional encoding of the feed rate timing input vector; performing one-dimensional convolution encoding on the feeding rate time sequence input vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following second convolution formula to obtain a second-scale feeding rate feature vector; wherein the second convolution formula is: />

Wherein->

For the second one-dimensional convolution kernel>

Width in direction, ++>

Is a second one-dimensional convolution kernel parameter vector,

For the size of the second one-dimensional convolution kernel,

representing the feed rate timing input vector, +.>

Representing one-dimensional convolutional encoding of the feed rate timing input vector; and cascading the first scale feed rate feature vector and the second scale feed rate feature vector to obtain the feed rate timing feature vector using a fusion layer of the multi-scale neighborhood feature extraction module.

In a specific example, in the rectification control method in the liquid oxygen preparation process, calculating a response transfer between the in-column liquid level absolute and relative multidimensional correlation feature matrix with respect to the feed rate timing feature vector to obtain a classification feature vector includes: calculating a response transfer between the absolute and relative multidimensional correlation feature matrices of the liquid level in the tower relative to the time sequence feature vector of the feeding rate according to the following response formula to obtain a classification feature vector; wherein, the responsiveness formula is:

Wherein->

Representing the feed rate timing feature vector, +.>

Representing the classification feature vector.

In a specific example, in the rectification control method in the liquid oxygen preparation process, training the integrated network model including the first convolutional neural network model and the second convolutional neural network model, the multi-scale neighborhood feature extraction module and the classifier is further included; the training of the integrated network model comprising the first convolutional neural network model and the second convolutional neural network model, the multi-scale neighborhood feature extraction module and the classifier comprises the following steps: acquiring training data, wherein the training data comprises training feeding rate values at a plurality of preset time points in a preset time period and liquid level values in a training tower, and a true value that the feeding rate value at the current time point should be increased or decreased; the training feeding rate values and the liquid level values in the training tower at the preset time points are respectively arranged into training feeding rate time sequence input vectors and training tower liquid level time sequence input vectors according to the time dimension; calculating the difference value between the liquid level values in the training towers at every two adjacent positions in the liquid level time sequence input vector in the training towers to obtain a liquid level time sequence change input vector in the training towers; performing association coding on the liquid level time sequence input vector in the training tower and the liquid level time sequence change input vector in the training tower to obtain an absolute and relative multidimensional association matrix of the liquid level in the training tower; the liquid level absolute and relative multidimensional correlation matrix in the training tower passes through the integrated network model comprising the first convolutional neural network model and the second convolutional neural network model to obtain a liquid level absolute and relative multidimensional correlation characteristic matrix in the training tower; the training feeding rate time sequence input vector passes through the multi-scale neighborhood feature extraction module to obtain a training feeding rate time sequence feature vector; calculating the response transfer between the absolute liquid level and the relative multidimensional correlation feature matrix in the training tower relative to the time sequence feature vector of the training feeding rate so as to obtain a training classification feature vector; passing the training classification feature vector through the classifier to obtain a classification loss function value; calculating probability distribution shift information compensation loss function values of the absolute and relative multidimensional correlation feature matrixes of the liquid level in the training tower and the time sequence feature vector of the training feeding rate; and using the weighted sum of the classified loss function value and the probability distribution shift information to compensate the loss function value as the loss function value, and training the integrated network model comprising the first convolutional neural network model and the second convolutional neural network model, the multi-scale neighborhood feature extraction module and the classifier through back propagation of gradient descent.

In a specific example, in the rectification control method in the liquid oxygen preparation process, the step of passing the training classification feature vector through the classifier to obtain a classification loss function value includes: processing the training classification feature vector using the classifier to generate a training classification result with a classification formula:

wherein->

Representing the training classification feature vector, +.>

To->

Is a weight matrix>

To->

Representing a bias matrix; and calculating a cross entropy value between the training classification result and a true value as the classification loss function value.

In a specific example, in the rectification control method in the liquid oxygen preparation process, calculating the probability distribution shift information compensation loss function value of the training column internal liquid level absolute and relative multidimensional correlation feature matrix and the training feed rate time sequence feature vector includes: calculating the probability distribution shift information compensation loss function value of the liquid level absolute and relative multidimensional association characteristic matrix in the training tower and the training feeding rate time sequence characteristic vector according to the following optimization formula; wherein, the optimization formula is:

Wherein->

And->

Representation->

Function (F)>

Representation->

Function (F)>

Represents a logarithmic function with base 2, +.>

And->

Compensating for shift superparameter, and +.>

For weighting superparameters, < >>

It will be appreciated by those skilled in the art that the specific operation of each step in the above-described rectification control method in the liquid oxygen production process has been described in detail in the above description of the rectification control system in the liquid oxygen production process with reference to fig. 1 to 5, and thus, repetitive description thereof will be omitted.

The present application also provides a computer program product comprising instructions which, when executed, cause an apparatus to perform operations corresponding to the above-described methods.

In one embodiment of the present application, there is also provided a computer readable storage medium storing a computer program for executing the above-described method.

It should be appreciated that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the forms of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects may be utilized. Furthermore, the computer program product may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

Methods, systems, and computer program products of embodiments of the present application are described in terms of flow diagrams and/or block diagrams. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

The basic principles of the present application have been described above in connection with specific embodiments, however, it should be noted that the advantages, benefits, effects, etc. mentioned in the present application are merely examples and not limiting, and these advantages, benefits, effects, etc. are not to be considered as necessarily possessed by the various embodiments of the present application. Furthermore, the specific details disclosed herein are for purposes of illustration and understanding only, and are not intended to be limiting, as the application is not intended to be limited to the details disclosed herein as such.

The block diagrams of the devices, apparatuses, devices, systems referred to in this application are only illustrative examples and are not intended to require or imply that the connections, arrangements, configurations must be made in the manner shown in the block diagrams. As will be appreciated by one of skill in the art, the devices, apparatuses, devices, systems may be connected, arranged, configured in any manner. Words such as "including," "comprising," "having," and the like are words of openness and mean "including but not limited to," and are used interchangeably therewith. The terms "or" and "as used herein refer to and are used interchangeably with the term" and/or "unless the context clearly indicates otherwise. The term "such as" as used herein refers to, and is used interchangeably with, the phrase "such as, but not limited to.

It is also noted that in the apparatus, devices and methods of the present application, the components or steps may be disassembled and/or assembled. Such decomposition and/or recombination should be considered as equivalent to the present application.

The previous 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 generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present 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.

Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or terminal device comprising the element.

The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the application to the form disclosed herein. Although a number of example aspects and embodiments have been discussed above, a person of ordinary skill in the art will recognize certain variations, modifications, alterations, additions, and subcombinations thereof.

Claims

1. A rectification control system in a liquid oxygen preparation process, comprising: the data acquisition module is used for acquiring feeding speed values and liquid level values in the tower at a plurality of preset time points in a preset time period; the data time sequence distribution module is used for respectively arranging the feeding rate values and the liquid level values in the tower at a plurality of preset time points into a feeding rate time sequence input vector and a liquid level time sequence input vector in the tower according to the time dimension; the liquid level relative time sequence change module is used for calculating the difference value between the liquid level values in each two adjacent positions in the liquid level time sequence input vector in the tower to obtain the liquid level time sequence change input vector in the tower; the liquid level time sequence association module is used for carrying out association coding on the liquid level time sequence input vector in the tower and the liquid level time sequence change input vector in the tower so as to obtain an absolute and relative multidimensional association matrix of the liquid level in the tower; the liquid level time sequence change multi-scale feature extraction module is used for enabling the in-tower liquid level absolute and relative multi-dimensional correlation matrix to pass through an integrated network model comprising a first convolutional neural network model and a second convolutional neural network model to obtain an in-tower liquid level absolute and relative multi-dimensional correlation feature matrix; the feeding rate time sequence feature extraction module is used for enabling the feeding rate time sequence input vector to pass through the multi-scale neighborhood feature extraction module to obtain a feeding rate time sequence feature vector; the responsiveness transfer module is used for calculating responsiveness transfer between the absolute and relative multidimensional correlation feature matrix of the liquid level in the tower relative to the time sequence feature vector of the feeding rate so as to obtain a classification feature vector; and a feed rate control module for passing the classification feature vector through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the feed rate value at the current time point should be increased or decreased.

2. The rectification control system in a liquid oxygen production process according to claim 1, wherein the liquid level timing correlation module is configured to: the time sequence input vector sum of the liquid level in the tower is calculated according to the following coding formulaPerforming association coding on the input vector of the time sequence change of the liquid level in the tower to obtain an absolute and relative multidimensional association matrix of the liquid level in the tower; wherein, the coding formula is:

wherein->

Representing matrix multiplication.

3. The rectification control system in a liquid oxygen production process according to claim 2, wherein the liquid level time sequence variation multi-scale feature extraction module comprises: a first scale feature extraction unit, configured to perform a convolution process, a pooling process, and a nonlinear activation process based on a two-dimensional convolution kernel on the in-tower liquid level absolute and relative multidimensional correlation matrix in forward transfer of layers by using each layer of a first convolutional neural network model of the integrated network model to output a first scale feature vector from a last layer of the first convolutional neural network model, where the first convolutional neural network model has a two-dimensional convolution kernel of a first scale; a second scale feature extraction unit for performing two-dimensional convolution kernel-based convolution processing, pooling processing and nonlinear activation processing on the in-tower liquid level absolute and relative multidimensional correlation matrix in forward transfer of layers by using layers of a second convolution neural network model of the integrated network model to output a second scale feature vector by a last layer of the second convolution neural network model, wherein the second convolution neural network model has a two-dimensional convolution kernel of a second scale, and the first scale is different from the second scale; and the cascading unit is used for cascading the first scale feature vector and the second scale feature vector to obtain the absolute and relative multidimensional correlation feature matrix of the liquid level in the tower.

4. The rectification control system in a liquid oxygen production process of claim 3, wherein the multi-scale neighborhood feature extraction module comprises: and a fusion layer connected with the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer use two-dimensional convolution kernels with different scales.

5. The rectification control system in a liquid oxygen production process of claim 4, wherein said feed rate timing feature extraction module comprises: the first scale extraction unit is used for carrying out one-dimensional convolution coding on the feeding rate time sequence input vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following first convolution formula so as to obtain a first scale feeding rate feature vector; wherein the first convolution formula is:

wherein->

For the first one-dimensional convolution kernel>

Width in direction, ++>

For a first one-dimensional convolution kernel parameter vector, +.>

For the size of the first one-dimensional convolution kernel, +.>

Representing the feed rate timing input vector, +. >

Representing one-dimensional convolutional encoding of the feed rate timing input vector; the second scale extraction unit is used for carrying out one-dimensional convolution coding on the feeding rate time sequence input vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following second convolution formula so as to obtain a second scale feeding rate feature vector; wherein the second convolution formula is: />

Wherein->

For the second one-dimensional convolution kernel>

Width in direction, ++>

For a second one-dimensional convolution kernel parameter vector, +.>

For the size of the second one-dimensional convolution kernel, +.>

Representing the feed rate timing input vector, +.>

6. The rectification control system in a liquid oxygen production process according to claim 5, wherein said responsive transfer module is adapted to: calculating a response transfer between the absolute and relative multidimensional correlation feature matrices of the liquid level in the tower relative to the time sequence feature vector of the feeding rate according to the following response formula to obtain a classification feature vector; wherein, the responsiveness formula is:

Wherein->

Representing the absolute and relative multidimensional correlation characteristic matrix of the liquid level in the tower,

representing the feed rate timing feature vector, +.>

Representing the classification feature vector.

7. The rectification control system in a liquid oxygen production process according to claim 6, further comprising a training module for training the integrated network model comprising the first convolutional neural network model and the second convolutional neural network model, the multi-scale neighborhood feature extraction module, and the classifier; wherein, training module includes: the training data acquisition module is used for acquiring training data, wherein the training data comprises training feeding rate values at a plurality of preset time points in a preset time period and liquid level values in a training tower, and a real value of which the feeding rate value at the current time point is required to be increased or reduced; the training data time sequence distribution module is used for respectively arranging the training feeding rate values and the liquid level values in the training tower at a plurality of preset time points into a training feeding rate time sequence input vector and a training tower liquid level time sequence input vector according to the time dimension; the training liquid level relative time sequence change module is used for calculating the difference value between the liquid level values in the training towers at every two adjacent positions in the liquid level time sequence input vector in the training towers so as to obtain the liquid level time sequence change input vector in the training towers; the training liquid level time sequence association module is used for carrying out association coding on the liquid level time sequence input vector in the training tower and the liquid level time sequence change input vector in the training tower so as to obtain an absolute and relative multidimensional association matrix of the liquid level in the training tower; the training liquid level time sequence change multi-scale feature extraction module is used for enabling the liquid level absolute and relative multi-dimensional correlation matrix in the training tower to pass through the integrated network model comprising the first convolutional neural network model and the second convolutional neural network model so as to obtain the liquid level absolute and relative multi-dimensional correlation feature matrix in the training tower; the training feeding rate time sequence feature extraction module is used for enabling the training feeding rate time sequence input vector to pass through the multi-scale neighborhood feature extraction module to obtain a training feeding rate time sequence feature vector; the training response transfer module is used for calculating the response transfer between the absolute liquid level and relative multidimensional association characteristic matrix in the training tower relative to the training feeding rate time sequence characteristic vector so as to obtain a training classification characteristic vector; the classification loss module is used for passing the training classification feature vector through the classifier to obtain a classification loss function value; the probability distribution shift information compensation loss module is used for calculating probability distribution shift information compensation loss function values of the liquid level absolute and relative multidimensional correlation feature matrix in the training tower and the training feeding rate time sequence feature vector; and a back propagation module for compensating a weighted sum of the classification loss function value and the probability distribution shift information as a loss function value, and training the integrated network model including the first convolutional neural network model and the second convolutional neural network model, the multi-scale neighborhood feature extraction module, and the classifier by back propagation of gradient descent.

8. The rectification control system in a liquid oxygen production process according to claim 7, wherein said classification loss module comprises: a training result classifying unit, configured to process the training classification feature vector by using the classifier according to the following classification formula to generate a training classification result, where the classification formula is:

wherein->

Representing the training classification feature vector, +.>

To->

Is a weight matrix>

To->

9. The rectification control system in the liquid oxygen production process according to claim 8,the probability distribution shift information compensation loss module is characterized by comprising a probability distribution shift information compensation loss module, wherein the probability distribution shift information compensation loss module is used for: calculating the probability distribution shift information compensation loss function value of the liquid level absolute and relative multidimensional association characteristic matrix in the training tower and the training feeding rate time sequence characteristic vector according to the following optimization formula; wherein, the optimization formula is:

wherein->

And->

The liquid level absolute and relative multidimensional correlation feature matrix in the training tower and the training feeding rate time sequence feature vector are respectively +. >

Representation->

Function (F)>

Representation->

Function (F)>

Represents a logarithmic function with base 2, +.>

And->

Compensating for shift superparameter, and +.>

For weighting superparameters, < >>

10. A rectification control method in a liquid oxygen preparation process is characterized by comprising the following steps: acquiring feeding speed values and liquid level values in the tower at a plurality of preset time points in a preset time period; arranging the feeding rate values and the liquid level values in the tower at a plurality of preset time points into a feeding rate time sequence input vector and a liquid level time sequence input vector in the tower according to a time dimension respectively; calculating the difference value between the liquid level values in each two adjacent positions in the liquid level time sequence input vector in the tower to obtain a liquid level time sequence change input vector in the tower; performing association coding on the in-tower liquid level time sequence input vector and the in-tower liquid level time sequence change input vector to obtain an in-tower liquid level absolute and relative multidimensional association matrix; the absolute and relative multidimensional correlation matrix of the liquid level in the tower is obtained through an integrated network model comprising a first convolution neural network model and a second convolution neural network model; the feeding rate time sequence input vector is passed through a multi-scale neighborhood feature extraction module to obtain a feeding rate time sequence feature vector; calculating the response transfer between the absolute and relative multidimensional correlation feature matrices of the liquid level in the tower relative to the time sequence feature vector of the feeding rate so as to obtain a classification feature vector; and passing the classification feature vector through a classifier to obtain a classification result, wherein the classification result is used for indicating that the feeding rate value of the current time point is increased or decreased.