CN116859839A

CN116859839A - Industrial control method and device based on model training

Info

Publication number: CN116859839A
Application number: CN202310799268.5A
Authority: CN
Inventors: 周爱平; 郭丽萍
Original assignee: Kyland Technology Co Ltd
Current assignee: Kyland Technology Co Ltd
Priority date: 2023-06-30
Filing date: 2023-06-30
Publication date: 2023-10-10

Abstract

The application discloses an industrial control method and device based on model training, and belongs to the field of industrial control. The industrial control method based on model training comprises the following steps: performing operation processing on real-time industrial big data in a target industrial scene by using a target algorithm prediction model to obtain a target control algorithm and a target independent variable for controlling a target controlled object in the target scene output by the target algorithm prediction model; and inputting the value of the target independent variable in the real-time industrial big data to a target controller deployed with a target control algorithm to obtain the value of the target control variable so as to control the target controlled object. According to the industrial control method based on model training, the obtained target control algorithm is high in precision and accuracy, the labor cost and the time cost can be remarkably reduced, the control efficiency and the control effect are improved, and the degree of universality is high.

Description

Industrial control method and device based on model training

Technical Field

The application belongs to the field of industrial control, and particularly relates to an industrial control method and device based on model training.

Background

With the development of artificial intelligence, industrial big data and artificial intelligence are increasingly applied to the technical field of automatic control. In the related technology, the industrial process mechanism is manually analyzed to obtain the corresponding parameter association relation, for example, an industrial expert analyzes the process mechanism in advance to obtain the association parameter related to the controlled parameter, so that a control algorithm is obtained to carry out industrial control. The control algorithm obtained by the method is low in accuracy and precision, a large amount of labor cost is consumed, and the requirement on the professional knowledge level of a user is high, so that the obtaining efficiency of the control algorithm is affected, the application range is limited, and the generalization cannot be achieved.

Disclosure of Invention

The present application aims to solve at least one of the technical problems existing in the prior art. Therefore, the application provides the industrial control method and the device based on model training, which can obviously reduce the labor and time cost, improve the control efficiency and the control effect and have higher universality.

In a first aspect, the present application provides an industrial control method based on model training, the method comprising:

performing operation processing on real-time industrial big data in a target industrial scene by using a target algorithm prediction model to obtain a target control algorithm and a target independent variable, which are output by the target algorithm prediction model and are used for controlling a target controlled object in the target scene; the target algorithm prediction model is obtained after training historical sample industrial big data, the historical sample industrial big data and the real-time industrial big data both comprise values of a plurality of independent variables and control variables, the historical sample industrial big data are used for representing historical data, the real-time industrial big data are used for representing current data, the target independent variables are independent variables which are determined in the independent variables of the real-time industrial big data and correspond to target control variables, and the target control algorithm is used for representing a control algorithm between the target independent variables and the target control variables;

And inputting the value of the target independent variable in the real-time industrial big data to a target controller deployed with the target control algorithm to obtain the value of the target control variable so as to control a target controlled object.

According to the industrial control method based on model training, the target independent variable and the target control algorithm corresponding to the target independent variable can be directly obtained by the trained target algorithm prediction model based on the input real-time industrial big data, the obtained target control algorithm is high in precision and accuracy, the independent variable related to the target control variable is not required to be manually screened by a user, the labor and time cost are obviously reduced, the application threshold is reduced, and the universality is high; and then, industrial control is performed based on the automatically acquired target independent variable value and a target control algorithm, so that the method has a higher control effect.

According to one embodiment of the present application, after the calculation processing is performed on the real-time industrial big data in the target industrial scene by using the target algorithm prediction model to obtain the target control algorithm and the target independent variable for controlling the target controlled object in the target scene output by the target algorithm prediction model, the method further includes:

Inputting the target independent variable and the target control algorithm to a simulation system under the target industrial scene, and obtaining the value of the target control variable output by the simulation system;

and respectively optimizing the target independent variable and the target control algorithm based on the value of the target control variable output by the simulation system.

According to one embodiment of the present application, the optimizing the target independent variable and the target control algorithm based on the target control variable output by the simulation system includes:

optimizing the target independent variable and the target control algorithm respectively based on the value of the target control variable output by the simulation system;

inputting the optimized target independent variable and the optimized target control algorithm into the simulation system, and obtaining the value of the target control variable output by the simulation system again;

repeating the steps of optimizing the target independent variable and the target control algorithm based on the value of the target control variable output by the simulation system until the value of the target control variable output by the simulation system meets the target control precision.

According to one embodiment of the present application, after inputting the value of the target argument in the real-time industrial big data to a target controller deployed with the target control algorithm, the method further includes:

And reusing the target algorithm prediction model to perform operation processing on the updated real-time industrial big data based on the value of the target control variable so as to optimize the target independent variable and the target control algorithm respectively.

According to one embodiment of the application, the target algorithm prediction model is trained by:

acquiring the historical sample industrial big data in the target industrial scene;

learning the historical sample industrial big data through an initial algorithm prediction model to obtain the association relation between any two elements in a plurality of independent variables and control variables in the historical sample industrial big data;

and determining a target independent variable corresponding to the target control variable from a plurality of independent variables of the historical sample industrial big data based on the association relation, and a target control algorithm corresponding to the target independent variable corresponding to the historical sample industrial big data.

According to one embodiment of the application, the acquiring the historical sample industrial big data in the target industrial scene includes:

performing data preprocessing on the obtained historical industrial big data in the target industrial scene to obtain preprocessed historical industrial big data; the historical industrial big data is at least part of historical data in the target industrial scene;

And processing the preprocessed historical industrial big data based on the processing mode corresponding to the category of each preprocessed historical industrial big data, and obtaining the historical sample industrial big data.

According to an embodiment of the present application, the processing the preprocessed historical industrial big data based on the processing manner corresponding to the respective category of the preprocessed historical industrial big data, to obtain the historical sample industrial big data, includes:

under the condition that the category is the first category, determining the preprocessed historical industrial big data as the historical sample industrial big data;

when the category is the second category, the pretreated historical industrial big data is arranged and marked based on a database time sequence arrangement mode and time information corresponding to the pretreated historical industrial big data, and the historical sample industrial big data is obtained;

under the condition that the category is a third category, the preprocessed historical industrial big data is distinguished based on the sub-category of the control system under the target industrial scene, and the classified historical industrial big data corresponding to each sub-category is respectively obtained; and based on the time sequence arrangement mode of the database and the time information corresponding to the classified historical industrial big data, arranging and marking the classified historical industrial big data, and obtaining the historical sample industrial big data.

According to one embodiment of the application, the historical sample industrial big data comprises: at least two of an operating condition variable, control system output data in the target industrial scenario, a process object variable, and a control system variable in the target industrial scenario.

According to one embodiment of the present application, the initial algorithm prediction model includes an encoder module and a decoder module connected in sequence, and the learning of the historical sample industrial big data by the initial algorithm prediction model includes:

inputting the historical sample industrial big data to the encoder module, and obtaining an intermediate representation output by the encoder module, wherein the intermediate representation is obtained by mapping the historical sample industrial big data after the encoder module learns the dependency relationship between the data at different positions in a data sequence of the historical sample industrial big data;

and inputting the intermediate representation to the decoder module, and acquiring the target independent variable and the target control algorithm which are output by the decoder module and correspond to the historical sample industrial big data.

According to one embodiment of the application, at least one of the encoder module and the decoder module comprises:

The system comprises a plurality of self-attention sub-layers and a feedforward neural network layer, wherein the output ends of the self-attention sub-layers are connected with the input ends of the feedforward neural network layer; wherein,,

the self-attention sub-layer is used for processing the data at the target position based on the data at other positions except the target position in the data sequence and acquiring the association relation between the data at the target position and the data at the other positions.

In a second aspect, the present application provides an industrial control device based on model training, the device comprising:

the first processing module is used for carrying out operation processing on real-time industrial big data in a target industrial scene by utilizing a target algorithm prediction model to obtain a target control algorithm and a target independent variable, wherein the target control algorithm and the target independent variable are used for controlling a target controlled object in the target scene and are output by the target algorithm prediction model; the target algorithm prediction model is obtained after training historical sample industrial big data, the historical sample industrial big data and the real-time industrial big data both comprise values of a plurality of independent variables and control variables, the historical sample industrial big data are used for representing historical data, the real-time industrial big data are used for representing current data, the target independent variables are independent variables which are determined in the independent variables of the real-time industrial big data and correspond to target control variables, and the target control algorithm is used for representing a control algorithm between the target independent variables and the target control variables;

And the second processing module is used for inputting the value of the target independent variable in the real-time industrial big data to a target controller deployed with the target control algorithm to obtain the value of the target control variable so as to control a target controlled object.

According to the industrial control device based on model training, the target independent variable and the target control algorithm corresponding to the target independent variable can be directly obtained by the trained target algorithm prediction model based on the input real-time industrial big data, the obtained target control algorithm is high in precision and accuracy, the independent variable related to the target control variable is not required to be manually screened by a user, the labor and time cost are obviously reduced, the application threshold is reduced, and the universality is high; and then, carrying out industrial control based on the value of the automatically acquired target independent variable and the target control algorithm, wherein the target control algorithm is obtained based on the automatically acquired target independent variable with higher control effect so as to carry out industrial control, and the acquired target control algorithm has higher precision and accuracy.

In a third aspect, the present application provides a control system based on the industrial control method based on model training according to the first aspect, including:

Training a model system by big data;

and the big data training model system is in communication connection with the edge real-time control system.

In a fourth aspect, the present application provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the model training based industrial control method as described in the first aspect above.

In a fifth aspect, the present application provides a computer program product comprising a computer program which, when executed by a processor, implements the model training based industrial control method as described in the first aspect above.

The above technical solutions in the embodiments of the present application have at least one of the following technical effects:

the target independent variable and the target control algorithm corresponding to the target independent variable can be directly obtained based on the input real-time industrial big data through the trained target algorithm prediction model, the obtained target control algorithm is high in precision and accuracy, a user does not need to manually screen the independent variable related to the target control variable, the labor and time cost are obviously reduced, the application threshold is lowered, and the universality is high; and then, industrial control is performed based on the automatically acquired target independent variable value and a target control algorithm, so that the method has a higher control effect.

Further, the historical sample industrial big data is input into the initial algorithm prediction model, so that the initial algorithm prediction model is trained with the aim of enabling the initial algorithm prediction model to output independent variables related to control variables and control algorithms, and the control algorithms are obtained by training the big data on the premise of not depending on the research of corresponding industrial process mechanisms so as to obtain corresponding parameter association relations; the user does not need to mark the sample control variable and the sample independent variable corresponding to the sample control variable in advance, so that the manpower and time cost are obviously reduced, and the learning capacity and the intelligent degree of the model are improved; and the influence of human subjective factors is effectively eliminated, so that the accuracy and the precision of the model are improved, and the method is suitable for different industrial scenes.

Furthermore, after the target independent variable and the target control algorithm corresponding to the target control variable are automatically acquired based on real-time industrial big data, simulation test is further performed based on the target independent variable and the target control algorithm, so that the target independent variable and the target control algorithm are optimized based on simulation results, the best target independent variable and the target control algorithm can be achieved, the accuracy and the precision of the acquired target independent variable and target control algorithm can be further improved, and therefore the subsequent control effect is improved.

And further, the target independent variable and the target control algorithm corresponding to the target control variable are automatically acquired based on real-time industrial big data, and the target independent variable and the target control algorithm are reversely optimized based on the actual value obtained after the target control algorithm is used for controlling the target controller, so that the target independent variable and the target control algorithm are optimal, the target independent variable and the target control algorithm can be automatically adjusted based on the actual control condition on the basis of effectively eliminating the influence of the artificial subjective factors, and the control efficiency and the control effect are improved.

Additional aspects and advantages of the application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the application.

Drawings

The foregoing and/or additional aspects and advantages of the application will become apparent and may be better understood from the following description of embodiments taken in conjunction with the accompanying drawings in which:

FIG. 1 is a schematic flow chart of an industrial control method based on model training according to an embodiment of the present application;

FIG. 2 is a second flow chart of an industrial control method based on model training according to an embodiment of the present application;

FIG. 3 is a third flow chart of an industrial control method based on model training according to an embodiment of the present application;

FIG. 4 is a schematic diagram of a control system according to an embodiment of the present application;

FIG. 5 is a schematic diagram of an industrial control device based on model training according to an embodiment of the present application;

fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The technical solutions of the embodiments of the present application will be clearly described below 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 are obtained by a person skilled in the art based on the embodiments of the present application, fall within the scope of protection of the present application.

The terms first, second and the like in the description and in the claims, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged, as appropriate, such that embodiments of the present application may be implemented in sequences other than those illustrated or described herein, and that the objects identified by "first," "second," etc. are generally of a type, and are not limited to the number of objects, such as the first object may be one or more. Furthermore, in the description and claims, "and/or" means at least one of the connected objects, and the character "/", generally means that the associated object is an "or" relationship.

The model training-based industrial control method, the model training-based industrial control device, the control system, the electronic equipment, the readable storage medium and the computer program product provided by the embodiment of the application are described in detail below through specific embodiments and application scenes thereof with reference to the accompanying drawings.

As shown in fig. 1, the model training-based industrial control method includes: step 110 and step 120.

And 110, performing operation processing on real-time industrial big data in a target industrial scene by using the target algorithm prediction model to obtain a target control algorithm and a target independent variable for controlling a target controlled object in the target scene output by the target algorithm prediction model.

In the step, the target algorithm prediction model is obtained by training the historical sample industrial big data, and the historical sample industrial big data and the real-time industrial big data both comprise values of a plurality of independent variables and control variables.

The historical sample industrial big data is used for representing historical data, the real-time industrial big data is used for representing current data, the target independent variable is an independent variable which is determined in a plurality of independent variables of the real-time industrial big data and corresponds to the target control variable, and the target control algorithm is a control algorithm used for representing between the target independent variable and the target control variable.

The real-time industrial big data is at least part of industrial big data generated in the target industrial scene, wherein the at least part of industrial big data comprises the values of a plurality of independent variables and the values of control variables.

Categories of real-time industrial big data include, but are not limited to: operating condition variables, control system output data in a target industrial scene, process object variables, control system variables in a target industrial scene, and the like.

Note that not all of the plurality of independent variables have an association relationship with the control variable.

In some embodiments, some of the plurality of independent variables may have an association with the control variable; or the association relationship between a plurality of independent variables and the control variable is not possible; or a plurality of independent variables may have an association relationship with the control variable, which is not limited herein.

The target control variable is a parameter to be controlled, and the target independent variable is a parameter associated with the target control variable in the independent variables.

The number of target control variables may be one or more.

For example, for a certain temperature control scene, parameters affecting the temperature value include a pressure value, a current value and a flow value, and if the temperature is a target control variable, the pressure value, the current value and the flow value are target independent variables, and the association relationship between the temperature and the pressure value, the current value and the flow value forms a target control algorithm.

In this step, both the target control algorithm and the target argument are automatically acquired by the system, rather than manually determined.

In the actual execution process, the target independent variable can be determined from the independent variables according to the association relation between any two elements in the independent variables and the control variable.

After the target independent variable is obtained, a target control algorithm between the target independent variable and the target control variable can be further determined.

It will be appreciated that the independent variables and control algorithms corresponding to the control variables may be different for different control types.

In some embodiments, the target algorithm prediction model may be a transducer model.

The transducer is a deep neural network model based on a self-attention mechanism, and can process sequence data in parallel with high efficiency.

It is understood that each industrial scenario may correspond to a corresponding algorithmic predictive model.

The algorithmic predictive models corresponding to different industrial scenarios may be different.

And 120, inputting the value of the target independent variable in the real-time industrial big data to a target controller deployed with a target control algorithm to obtain the value of the target control variable so as to control the target controlled object.

In this embodiment, the target controller is a controller in a target industrial scenario.

The target industrial scenario may be any industrial scenario, for example: the application is not limited by the smelting scene of fused magnesia, the thermal power generation scene, the control scene of an air conditioning system, the temperature control scene of a reaction cavity of a semiconductor etching machine and the like.

In some embodiments, the real-time industrial big data may include industrial big data in any one of an electric fused magnesia smelting scene, a thermal power generation scene, an air conditioning system control scene, and a semiconductor etcher reaction chamber temperature control scene.

It should be noted that, the industrial control method based on model training is applied to a target controller, and the target controller is deployed with a target control algorithm, where the target control algorithm is used to characterize a control relationship between a target independent variable and a target control variable.

The target control algorithm is an algorithm for controlling the target controller to adjust parameters to obtain the required target independent variable value.

The target controlled object is the object which is finally required to be controlled.

It will be appreciated that the target controlled object may be the same as, or may be different from, the target controlled variable used to control the target controlled object in different situations.

In the actual execution process, after the target independent variable in the target industrial scene is obtained, the value corresponding to the target independent variable can be extracted from the real-time industrial big data and is input to a target controller deployed with a target control algorithm to obtain the value of the target control variable so as to control the target controlled object.

In some embodiments, the target controlled object may also in turn act as an argument that affects the value of the target control variable.

For example, under PID control types, the value of the target controlled object at the last time may be conversely used as an argument for adjusting the value of the target controlled variable at the next time.

The inventor finds that in the research and development process, in the related technology, independent variables related to the control variables are mainly obtained through manual analysis or expert searching of industrial process mechanisms, after target independent variables are obtained, a control algorithm between the control variables and the independent variables is further predicted and obtained based on the manually obtained target independent variables and the target control variables, and then corresponding industrial control is carried out based on the control algorithm; the method needs to know the industrial mechanism of a specific industry in advance, and if the cross-industry is involved, the method needs to provide support based on algorithms of other industries, so that generalization cannot be achieved.

In the application, the target independent variable and the target control algorithm corresponding to the target independent variable can be directly obtained based on the input real-time industrial big data through the trained target algorithm prediction model, the target independent variable is not required to be manually obtained by a user, the labor and time cost is obviously reduced, the operation threshold is lower, and the method is suitable for any industry, thereby improving the obtaining efficiency of the control algorithm and having higher universality and universality.

In addition, by analyzing the input massive industrial big data, the calculation error caused by missed detection can be reduced, so that the precision and accuracy of the obtained target control algorithm are improved, and the control effect is improved.

According to the industrial control method based on model training provided by the embodiment of the application, the target independent variable and the target control algorithm corresponding to the target independent variable can be directly obtained by the trained target algorithm prediction model based on the input real-time industrial big data, the obtained target control algorithm has higher precision and accuracy, and the independent variable related to the target control variable is not required to be manually screened by a user, so that the labor and time cost are obviously reduced, the application threshold is reduced, and the universality is higher; and then, industrial control is performed based on the automatically acquired target independent variable value and a target control algorithm, so that the method has a higher control effect.

The following describes a training method of the target algorithm prediction model.

As shown in fig. 2, in some embodiments, the target algorithm prediction model may be trained as follows:

acquiring historical sample industrial big data in a target industrial scene;

learning the historical sample industrial big data through an initial algorithm prediction model to obtain the association relationship between any two elements of a plurality of independent variables and control variables in the historical sample industrial big data;

In this embodiment, the historical sample industrial big data is sample data for training a target algorithm predictive model.

The historical sample industrial big data is massive big data.

In some embodiments, the historical sample industrial big data may include: at least two of an operating condition variable, control system output data in a target industrial scenario, a process object variable, and a control system variable in a target industrial scenario.

It should be noted that, the sample data for training the target algorithm prediction model should belong to the same industrial scenario as the application data for applying the target algorithm prediction model later, and be different data.

For example, taking an industrial scenario of boiler temperature control of a boiler of a thermal power plant as an example, factories for performing boiler temperature control of the thermal power plant include a factory a, a factory B and a factory C, in an actual training process, a target algorithm prediction model can be trained based on industrial big data of the factory a, and the trained target algorithm prediction model can be applied to the factory a, the factory B and the factory C, respectively; in this case, the industrial big data of the plant a used in the training process is the historical sample industrial big data, and the industrial big data of the plant B and the plant C input to the trained target algorithm prediction model in the subsequent application process is the real-time industrial big data.

The historical sample industrial big data also includes values for the target control variable and the plurality of independent variables, which may be different from the values for the target control variable and the plurality of independent variables included in the real-time industrial big data.

Continuing with the above industrial scenario of furnace temperature control of a thermal power plant boiler as an example, the target control variable is temperature, and the plurality of independent variables may include any parameter related to the thermal power plant boiler.

The initial algorithmic predictive model is the model to be trained.

In some embodiments, the initial algorithmic prediction model may be a transducer model.

In the training process, the obtained historical sample industrial big data is input into an initial algorithm prediction model, such as an initial transducer model, so that the model can output a target independent variable corresponding to a target control variable and a control algorithm to train the initial algorithm prediction model, and a trained target algorithm prediction model is obtained.

For example, the historical sample industrial big data may be input to an initial algorithm prediction model, and the initial algorithm prediction model learns to obtain the association relation between the variables, so as to obtain the independent variable related to the control variable based on the screening of the association relation from the variables corresponding to the input historical sample industrial big data, and then continues training to learn further to obtain the control algorithm.

In the embodiment, the historical sample industrial big data is input into an initial algorithm prediction model, the association relation among the variables is learned by the initial algorithm prediction model, so that independent variables related to the control variables are obtained based on the screening of the association relation among the variables corresponding to the input historical sample industrial big data, and then training is continued to further learn to obtain the control algorithm.

It will be appreciated that any control variable and its corresponding independent variable and control algorithm may be obtained during the training process.

The PID control will be described below as an example.

For example, by training, the following control algorithm may be output:

y(t)＝(y(t-1),(t),d())

wherein y (t) represents the value of the controlled object at the moment t; y (t-1) represents the value of the controlled object at the moment (t-1); u (t) represents the output of the controller at the time t, namely the value of the target control variable at the time t; d (t) characterizes the disturbance suffered by the target control variable at time t.

The value u (t) of the target control variable at time t can be further expressed as:

u(t)＝(e(t),(i))

wherein p (i) is a value of a target independent variable used in a control algorithm of u (t), and the target independent variable is also obtained by training a target algorithm prediction model.

e (t) can be expressed as:

e(t)＝(t)-(t)

wherein r (t) is a correction set value under PID control, and the correction set value is further determined based on a target independent variable obtained by training a target algorithm prediction model.

In the application process, the target independent variable related to the target control variable can be accurately acquired from a plurality of independent variables only by determining the target control variable from any obtained control variable.

For the trained target algorithm prediction model, the target algorithm prediction model can be deployed to other control systems in the same industrial scene so as to output a target control algorithm control target control system based on input real-time industrial big data for industrial control.

For example, after the target algorithm prediction model in the industrial scene of the boiler temperature control of the boiler of the thermal power plant is obtained through the training of the historical sample industrial big data of the plant a, the target algorithm prediction model can be deployed to the plant B and the plant C, so that the plant B and the plant C can acquire corresponding target control algorithms based on the respective real-time industrial big data.

According to the industrial control method based on model training provided by the embodiment of the application, the historical sample industrial big data is input into the initial algorithm prediction model, so that the initial algorithm prediction model is trained with the aim of enabling the initial algorithm prediction model to output independent variables related to control variables and a control algorithm, and the control algorithm is obtained by training the big data on the premise of not depending on the research of corresponding industrial process mechanisms; the user does not need to mark the sample control variable and the sample independent variable corresponding to the sample control variable in advance, so that the manpower and time cost are obviously reduced, and the learning capacity and the intelligent degree of the model are improved; and the influence of human subjective factors is effectively eliminated, so that the accuracy and the precision of the model are improved, and the method is suitable for different industrial scenes.

In some embodiments, the initial algorithmic prediction model may include an encoder module and a decoder module connected in sequence, and learning historical sample industrial big data through the initial algorithmic prediction model may include:

inputting the historical sample industrial big data into an encoder module, and obtaining an intermediate representation output by the encoder module, wherein the intermediate representation is obtained by mapping the historical sample industrial big data after the encoder module learns the dependency relationship between the data at different positions in a data sequence of the historical sample industrial big data;

and inputting the intermediate representation into a decoder module, and acquiring a target independent variable and a target control algorithm which are output by the decoder module and correspond to the historical sample industrial big data.

In this embodiment, the initial algorithmic prediction model may include an encoder module and a decoder module.

Wherein the encoder is configured to map an input data sequence to a set of intermediate representations, and the decoder is configured to convert the intermediate representations to a target sequence for output.

According to the application, the encoder module learns the dependency relationship between the data at different positions in the data sequence of the historical sample industrial big data, so that the intermediate representation is obtained; and converting the intermediate representation by the decoder module to output the target independent variable and the control algorithm, thereby training to obtain a target algorithm prediction model.

In some embodiments, at least one of the encoder module and the decoder module may include: the system comprises a plurality of self-attention sub-layers and a feedforward neural network layer, wherein the output ends of the self-attention sub-layers are connected with the input ends of the feedforward neural network layer; the self-attention sub-layer is used for processing the data at the target position based on the data at other positions except the target position in the data sequence and acquiring the association relation between the data at the target position and the data at other positions.

In this embodiment, the number of layers of the multi-layer self-care sublayer may be customized based on a user, and the present application is not limited.

The output end of the multi-layer self-attention sub-layer is connected with the input end of the feedforward neural network layer.

The target location may be any location in the data sequence.

In some embodiments, both the encoder module and the decoder module may include multiple self-attention sub-layers and feedforward neural network layers.

Wherein the self-attention sub-layer can learn the dependency between different positions in the data sequence, i.e. the model takes into account the information at all other positions in the sequence when processing the information at each position.

Based on the multi-layer training, the association relation between the variables is obtained, and a control algorithm aiming at a certain target control variable is further obtained.

According to the industrial control method based on model training, provided by the embodiment of the application, the dependency relationship among different positions in the sequence is learned by utilizing the transducer architecture, so that the association relationship among variables can be effectively obtained based on the input industrial big data prediction, the control algorithm aiming at a certain variable is obtained based on the association relationship, the learning capacity is strong, and the model accuracy is high.

The manner in which large industrial data of historical samples are obtained will be described below.

With continued reference to FIG. 2, in some embodiments, acquiring historical sample industrial big data in a target industrial scenario may include:

performing data preprocessing on the obtained historical industrial big data in the target industrial scene to obtain preprocessed historical industrial big data; the historical industrial big data is at least part of historical data in a target industrial scene;

and processing the preprocessed historical industrial big data based on the processing modes corresponding to the categories of the preprocessed historical industrial big data, and acquiring historical sample industrial big data.

In this embodiment, preprocessing includes data cleansing, integration, conversion, discretization, and reduction.

The historical industrial big data is at least part of historical data in the target industrial scene, and the data volume of the historical industrial big data is large.

Historical industrial big data may include: operating condition variables, control system output data in a target industrial scene, process object variables, control system variables in a target industrial scene, and the like.

The operating condition variables are used for representing factory operating condition data information, such as field sensors, instrument data, equipment operating state data and the like.

The output data of the control system in the target industrial scene is used for representing information generated by the control system, such as operator operation instructions, alarm information, intermediate calculated variable results and the like.

The process object variables are used for representing parameters of the process object of the factory, such as a V-F curve of a frequency converter, the pipe diameter of a pipeline, the material of the pipeline, the power of an electric heater and the like.

The control system variables in the target industrial scene are used for representing parameters of the control system, such as controller manufacturers, controller CPU types, core numbers, main frequencies, controller storage spaces, controller timer jitter, control task periods, I/O bus types, I/O channel numbers, I/O channel types and the like.

It will be appreciated that for operating condition variables and control system output data in a target industrial scenario, this data is typically stored in the SCADA system database of the plant or the information system database of the plant, or may be acquired from a control system (PLC/DCS), which itself is time stamped.

For process object variables, the process object variables need to be obtained from factory design files, and in the actual implementation process, the design institute submits the corresponding design files to the factory for storage after the factory design stage is completed.

For control system variables in a target industrial scenario, the variables are typically found in configuration engineering files and control system technical specification files of the control system.

Based on the differences among the operating condition variables, the control system output data in the target industrial scene, the process object variables, and the control system variables in the target industrial scene, the preprocessed historical industrial big data can be further divided into a plurality of categories.

And then processing the preprocessed historical industrial big data based on the corresponding processing modes of the respective categories to obtain the historical sample industrial big data.

In some embodiments, processing the preprocessed historical industrial big data based on the processing manner corresponding to the respective category of the preprocessed historical industrial big data, obtaining the historical sample industrial big data may include:

under the condition that the category is the first category, the preprocessed historical industrial big data is determined to be historical sample industrial big data;

under the condition that the category is the second category, the pretreated historical industrial big data is arranged and marked based on the time sequence arrangement mode of the database and the time information corresponding to the pretreated historical industrial big data, and the historical sample industrial big data is obtained;

Under the condition that the category is the third category, splitting the preprocessed historical industrial big data based on the sub-category of the control system under the target industrial scene, and respectively acquiring the classified historical industrial big data corresponding to each sub-category; and based on the time sequence arrangement mode of the database and the time information corresponding to the classified historical industrial big data, arranging and labeling the classified historical industrial big data, and obtaining the historical sample industrial big data.

In this embodiment, the time information is a time stamp corresponding to each data.

The first category includes operating condition variables and control system output data in a target industrial scenario.

The second category includes process object variables.

The third class includes control system variables in the target industrial scenario.

For the first type of data, the data is provided with a time stamp, so that secondary labeling is not needed.

For the second type of data, the data is input and marked according to the arrangement mode of the database and by adding a time stamp.

For the third type of data, the control system configuration engineering file data is classified and split, and a time stamp is added according to the arrangement mode of a database for data input and marking.

Based on the processing mode, the industrial big data of the historical sample can be obtained.

As shown in fig. 3, in some embodiments, after performing an operation process on real-time industrial big data in a target industrial scene by using a target algorithm prediction model to obtain a target control algorithm and a target independent variable for controlling a target controlled object in the target scene output by the target algorithm prediction model, the method may further include:

inputting the target independent variable and the target control algorithm into a simulation system under a target industrial scene, and obtaining the value of the target control variable output by the simulation system;

In this embodiment, the target control variable takes the value of the simulation value output by the simulation system.

After the target independent variable and the target control algorithm corresponding to the target independent variable are obtained through the target algorithm prediction model prediction, the target control algorithm can be input into a simulation system to carry out simulation test, so that the target independent variable and the target control algorithm are optimized in turn based on a simulation result.

The value of the target control variable is a simulation value output by a simulation system.

It will be appreciated that optimizing the variables based on the simulation values may be repeated a number of times.

In some embodiments, optimizing the target independent variable and the target control algorithm based on the values of the target control variable output by the simulation system, respectively, may include:

optimizing a target independent variable and a target control algorithm respectively based on the value of a target control variable output by the simulation system;

inputting the optimized target independent variable and the optimized target control algorithm into a simulation system, and obtaining the value of the target control variable output by the simulation system again;

In this embodiment, the simulation system is a simulation system corresponding to the control system in the target industrial scenario.

The target control accuracy may be user-defined based.

Optimizing the target independent variable and the target control algorithm based on the value of the target control variable to obtain the optimized target independent variable and target control algorithm; and then inputting the optimized target independent variable and the optimized target control algorithm into the simulation system again for simulation to obtain a new simulation value, and optimizing the target independent variable and the target control algorithm after the last optimization based on the new simulation value.

Repeating the operation until the value of the target control variable output by the simulation system for the last time meets the target control precision, thereby achieving the simulation effect required by the user.

According to the industrial control method based on model training, after the target independent variable and the target control algorithm corresponding to the target control variable are automatically acquired based on real-time industrial big data, simulation test is further carried out based on the target independent variable and the target control algorithm, so that the target independent variable and the target control algorithm are optimized based on simulation results, the best is achieved, the accuracy and the accuracy of the acquired target independent variable and target control algorithm can be further improved, and therefore the follow-up control effect is improved.

With continued reference to fig. 2, in some embodiments, after step 120, the method may further include:

and (3) carrying out operation processing on the updated real-time industrial big data by utilizing the target algorithm prediction model again based on the value of the target control variable so as to optimize the target independent variable and the target control algorithm respectively.

In this embodiment, the target control variable is an actual value obtained by performing a control operation by a deployed target control algorithm in the application process.

The actual value is used as new historical data to update the historical sample industrial big data, and the target algorithm prediction model is updated based on the updated historical sample industrial big data to update the target independent variable and the target control algorithm output by the target algorithm prediction model, so that further optimization of the target independent variable and the target control algorithm is realized.

For example, in the running process, the data of the actual algorithm running is fed back to the target algorithm prediction model for further training to obtain the optimal solution of the control algorithm and the target independent variable, and then the control algorithm package in the control system is synchronously updated.

According to the industrial control method based on model training, which is provided by the embodiment of the application, the target independent variable and the target control algorithm corresponding to the target control variable are automatically acquired based on real-time industrial big data, and the target independent variable and the target control algorithm are reversely optimized based on the actual value obtained after the target control algorithm is controlled by the target controller, so that the target independent variable and the target control algorithm are optimal, the target independent variable and the target control algorithm can be automatically regulated based on the actual control condition on the basis of effectively eliminating the influence of human subjective factors, and the control efficiency and the control effect are improved.

The control of the fused magnesia smelting process will be described below as an example.

The traditional fused magnesia smelting process adopts an HMI+PLC control system.

The main function of the HMI is to input a current set value, the PLC mainly realizes current tracking control through a PID process, the output three-phase electrode current tracks the set value input by the HMI, and the output value u (k) is regulated according to the deviation e (t) between the actual current value y (k) and the set value r.

The specific control algorithm is as follows:

wherein u (k) is an adjusting value for adjusting the controlled object y (k), namely the value of the target control variable; t is the target duration, and T is the time.

The initial algorithm prediction model provided by the embodiment of the application is used for processing the collected historical industrial big data to obtain the historical sample industrial big data and performing training calculation based on the historical sample industrial big data.

Wherein, the historical industrial big data may be the collection of operation history data and operation data from fused magnesia factories, including but not limited to: target value r of energy consumption per ton ^* Upper limit value r _min Lower limit value r _max Each time, a current set value r (k), a smelting voltage B1, an electrode diameter B2, and a current lower limit value y _min Upper limit value y of current _max Actual current value y (k), control period T of the control system, and instantaneous power P (k) of the electric furnace.

As shown in fig. 3, after the above-mentioned historical industrial big data is processed or marked to obtain the historical sample industrial big data, the historical sample industrial big data is input into the initial algorithm prediction model to train, so that the trained target algorithm prediction model can output other variables related to the current value and control algorithms between the current value and other variables.

For example, a plurality of target independent variables associated with a target control variable may be obtained: target value r of energy consumption per ton ^* Upper limit value r _min Lower limit value r _max Smelting voltage B1, electrode diameter B2 and current lower limit y _min Upper limit value y of current _max And the actual value y (k) of the controlled object, etc., and the control algorithm therebetween.

After the target independent variable and the control algorithm are obtained, the simulation system calculates an optimized set value y (nk) based on the target independent variable and the control algorithm, and then based on the current feedback value y _i (+1), current value y (k) and deviation e _i () Automatically calculating a set compensation value required by PID control, and adding the set compensation value to an optimized set value y (nk) to obtainTo the final set value y _sp () Finally, the final set value y _sp () And deploying the optimized control algorithm to a target controller to obtain the value of a target control variable so as to control a target controlled object.

The target control algorithm is obtained through training the target algorithm prediction model provided by the embodiment of the application, the target independent variable is optimized based on the target control algorithm, the optimal set value can be automatically obtained according to the working condition of the field smelting environment, the original set value input by means of manual experience is replaced, the influence of subjective factors of people is eliminated, and the problem that an operator cannot accurately identify abnormal working conditions in time and adjust the current set value is avoided.

In practical deployment application, the method can improve the stability of electric current control of the electric smelting magnesia furnace, reduce abnormal working conditions and reduce the power consumption of the electric smelting magnesia smelting process.

Of course, in other embodiments, the model training-based industrial control method of the present application may be applied to other industrial scenarios, such as prediction analysis for economic value of plant operation, obtaining a mathematical model (i.e., a control algorithm) between yield, energy consumption and raw material consumption based on comprehensive analysis of data such as yield, energy consumption and raw material consumption, so as to provide an optimal yield plan, and provide a control strategy, for example, performing production strategy adjustment and control strategy adjustment in combination with the price of peak electricity and valley electricity and the price of raw material on the market.

According to the industrial control method based on model training provided by the embodiment of the application, the execution subject can be an industrial control device based on model training. In the embodiment of the application, an industrial control device based on model training is taken as an example to execute an industrial control method based on model training.

The embodiment of the application also provides an industrial control device based on model training.

As shown in fig. 5, the model training-based industrial control device may include: a first processing module 510 and a second processing module 520.

The first processing module 510 is configured to perform operation processing on real-time industrial big data in a target industrial scene by using a target algorithm prediction model, so as to obtain a target control algorithm and a target independent variable for controlling a target controlled object in the target scene output by the target algorithm prediction model; the target algorithm prediction model is obtained after training the historical sample industrial big data, wherein the historical sample industrial big data and the real-time industrial big data comprise a plurality of independent variables and values of control variables, the historical sample industrial big data are used for representing the historical data, the real-time industrial big data are used for representing the current data, the target independent variable is an independent variable which is determined from the plurality of independent variables of the real-time industrial big data and corresponds to the target control variable, and the target control algorithm is a control algorithm used for representing between the target independent variable and the target control variable;

The second processing module 520 is configured to input the value of the target independent variable in the real-time industrial big data to a target controller deployed with a target control algorithm, so as to obtain the value of the target control variable, so as to control the target controlled object.

According to the industrial control device based on model training provided by the embodiment of the application, the target independent variable associated with the target control variable is automatically determined from a plurality of independent variables according to the incidence relation between any two elements in the plurality of independent variables and the control variable in the input industrial big data by the system, and the target control algorithm is further obtained based on the automatically obtained target independent variable, and is deployed to the control system to perform corresponding industrial control, so that a user does not need to manually obtain the target independent variable, the labor and time cost is remarkably reduced, the operation threshold is lower, the industrial control device is suitable for any industry, the obtaining efficiency of the control algorithm is improved, and the universality are higher.

In some embodiments, the model-based trained industrial control device may further comprise: a third processing module for:

the method comprises the steps of performing operation processing on real-time industrial big data in a target industrial scene by utilizing a target algorithm prediction model to obtain a target control algorithm and a target independent variable which are used for controlling a target controlled object in the target scene output by the target algorithm prediction model, inputting the target independent variable and the target control algorithm into a simulation system in the target industrial scene, and obtaining the value of the target control variable output by the simulation system;

In some embodiments, the third processing module may be further configured to:

In some embodiments, the apparatus may further include a fourth processing module configured to, after inputting the value of the target independent variable in the real-time industrial big data to the target controller deployed with the target control algorithm, obtain the value of the target control variable, re-utilize the target algorithm prediction model to perform operation processing on the updated real-time industrial big data based on the value of the target control variable, so as to optimize the target independent variable and the target control algorithm, respectively.

In some embodiments, the apparatus may further include a fifth processing module for:

acquiring historical sample industrial big data in a target industrial scene;

In some embodiments, the fifth processing module may be further configured to:

In some embodiments, the initial algorithmic prediction model includes an encoder module and a decoder module connected in sequence, and the fifth processing module may be further configured to:

The embodiment of the application also provides a control system.

As shown in fig. 4, the control system includes: and the big data trains the model system and the edge real-time control system.

The big data training model system is in communication connection with the edge real-time control system.

The big data training model system is used for training the historical sample industrial big data to obtain a target algorithm prediction model capable of outputting a target control algorithm and a target independent variable.

The big data training model system is combined with the edge real-time control system for performing the model training based industrial control method as described in any of the embodiments above.

The edge real-time control system may include a control system development environment and a control system operating environment.

For example, the big data training model system may send the generated target algorithm prediction model to the control system development environment, the control system development environment completes the overall control program coding based on the target control algorithm, configures the input and output of the control task, associates with the physical I/O mapping, configures the control task operation period, etc., then compiles and downloads to the control system operation environment, and the edge control system operation environment is responsible for the signal docking of the field sensor and the actuator, thereby completing the real-time control of the production field.

In the running process, the data of the actual algorithm running can be fed back to the target algorithm prediction model for further training to obtain the optimal solution of the control algorithm and the target independent variable, and then the control algorithm package for managing the development environment of the edge control system is synchronously updated.

Through multiple experiments by the inventor, the edge real-time control system provided by the application has a jitter time level as low as 5us, and can be suitable for high-precision control scenes such as industrial robot control, numerical control machine tool control and the like.

In some embodiments, the edge real-time control system may further include a cloud-edge collaborative algorithm management interface and a data communication interface, where the algorithm management interface is configured to obtain a target control algorithm generated by the big data training model system, and the data communication interface is configured to synchronize real-time data of the edge real-time control system to the big data training model system for training.

As shown in fig. 2, in some embodiments, the big data training model system may include, connected in series: the system comprises a data acquisition module, a data processing module, a model training module, an encoding module and a control simulation module.

In this embodiment, the control simulation module is electrically connected to the model training module and the edge real-time control system, respectively.

The data acquisition module is used for acquiring historical industrial big data.

As shown in fig. 4, the data processing module is used for preprocessing and calibrating the historical sample industrial big data, so as to obtain the historical sample industrial big data.

The model training module is used for training an initial algorithm prediction model based on the historical sample industrial big data to obtain a target algorithm prediction model.

The control simulation module is used for performing simulation test based on a target control algorithm output by the target algorithm prediction model and optimizing the target control algorithm and the target independent variable based on a simulation result.

The specific implementation steps of each module are described in the above embodiments, and are not described herein.

With continued reference to fig. 2, in some embodiments, the edge real-time control system may include: a first module and an execution module.

In this embodiment, the first module is electrically connected to the control emulation module;

the first module may include a download module and a deployment module.

For example, the first module may develop an environment for the control system described above.

The downloading module is used for downloading a target control algorithm and a target independent variable which are obtained by optimizing the control simulation module; the deployment module is used for deploying the target control algorithm and the target independent variable to the execution module.

The execution module is electrically connected with the first module and the data acquisition module respectively.

The execution module is used for carrying out control operation based on the value of the target independent variable in the target industrial scene through a target control algorithm to obtain the value of the target control variable so as to control the target controlled object.

The execution module may be the edge real-time control system described above.

In addition, the execution module can send the output actual value to the data acquisition module to update the historical industrial big data, so as to update the target algorithm prediction model.

Taking the control system as a PID control system as an example, the execution module may include: the system comprises a current setting control module, a self-optimizing correction module and a loop control module with output compensation.

The current setting control module is used for calculating an optimized set value y (nk) based on a target independent variable output by the algorithm prediction model and a target control algorithm.

The self-optimizing correction module is used for feeding back the value y according to the current _i (+1), current value y (k) and deviation e _i () And calculating a set compensation value.

The loop control module includes: PID controller u _i1 () Higher order nonlinear term rate of change compensation u _i2 () And a previous moment higher order nonlinear term compensator u _i3 () To jointly achieve current control.

According to the control system provided by the embodiment of the application, the target algorithm prediction model is obtained through training of the big data training model system, and the edge real-time control system is deployed by the target control algorithm which is further output based on the target algorithm prediction model, so that industrial control is performed, a user does not need to manually acquire a target independent variable, the influence of human subjective factors is effectively eliminated, and the accuracy and the precision of the acquired target control algorithm are effectively improved; in addition, the target independent variable and the target control algorithm can be automatically adjusted based on the actual control condition, so that the control efficiency and the control effect are improved.

In some embodiments, as shown in fig. 6, an electronic device 600 is further provided in the embodiments of the present application, which includes a processor 601, a memory 602, and a computer program stored in the memory 602 and capable of running on the processor 601, where the program when executed by the processor 601 implements the above-mentioned embodiments of the industrial control method based on model training, or the industrial control algorithm determining method, or each process of the algorithm prediction model training method, and the same technical effects can be achieved, and are not repeated herein.

The electronic device in the embodiment of the application includes the mobile electronic device and the non-mobile electronic device.

The embodiment of the application also provides a non-transitory computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements each process of the above embodiment of the industrial control method based on model training, and can achieve the same technical effects, and in order to avoid repetition, the description is omitted here.

Wherein the processor is a processor in the electronic device described in the above embodiment. The readable storage medium includes computer readable storage medium such as computer readable memory ROM, random access memory RAM, magnetic or optical disk, etc.

The embodiment of the application also provides a computer program product, which comprises a computer program, wherein the computer program is executed by a processor to realize the industrial control method embodiment based on model training.

The embodiment of the application further provides a chip, the chip comprises a processor and a communication interface, the communication interface is coupled with the processor, the processor is used for running programs or instructions, the processes of the industrial control method embodiment based on model training can be realized, the same technical effects can be achieved, and the repetition is avoided, and the description is omitted here.

It should be understood that the chips referred to in the embodiments of the present application may also be referred to as system-on-chip chips, chip systems, or system-on-chip chips, etc.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus 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 apparatus. 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 apparatus that comprises the element. Furthermore, it should be noted that the scope of the methods and apparatus in the embodiments of the present application is not limited to performing the functions in the order shown or discussed, but may also include performing the functions in a substantially simultaneous manner or in an opposite order depending on the functions involved, e.g., the described methods may be performed in an order different from that described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a computer software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal (which may be a mobile phone, a computer, a server, or a network device, etc.) to perform the method according to the embodiments of the present application.

The embodiments of the present application have been described above with reference to the accompanying drawings, but the present application is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those having ordinary skill in the art without departing from the spirit of the present application and the scope of the claims, which are to be protected by the present application.

In the description of the present specification, reference to the terms "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the application. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

While embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that: many changes, modifications, substitutions and variations may be made to the embodiments without departing from the spirit and principles of the application, the scope of which is defined by the claims and their equivalents.

Claims

1. An industrial control method based on model training, comprising:

2. The model training-based industrial control method according to claim 1, wherein after the real-time industrial big data in a target industrial scene is processed by using a target algorithm prediction model to obtain a target control algorithm and a target independent variable for controlling a target controlled object in the target scene output by the target algorithm prediction model, the method further comprises:

3. The model training-based industrial control method according to claim 2, wherein the optimizing the target independent variable and the target control algorithm based on the values of the target control variable output by the simulation system, respectively, comprises:

4. A model training-based industrial control method according to any one of claims 1-3, characterized in that after inputting the value of the target independent variable in the real-time industrial big data to a target controller where the target control algorithm is deployed, the method further comprises:

5. A model training based industrial control method according to any of claims 1-3, characterized in that the target algorithm predictive model is trained by:

6. The model training-based industrial control method of claim 5, wherein the obtaining the historical sample industrial big data in the target industrial scene comprises:

7. The model training-based industrial control method according to claim 6, wherein the processing the preprocessed historical industrial big data based on the processing mode corresponding to the respective category of the preprocessed historical industrial big data, and obtaining the historical sample industrial big data, comprises:

8. The model training-based industrial control method of claim 5, wherein the historical sample industrial big data comprises: at least two of an operating condition variable, control system output data in the target industrial scenario, a process object variable, and a control system variable in the target industrial scenario.

9. The model training-based industrial control method of claim 5, wherein the initial algorithmic prediction model comprises an encoder module and a decoder module connected in sequence, the learning of the historical sample industrial big data by the initial algorithmic prediction model comprising:

10. The model training-based industrial control method of claim 9, wherein at least one of the encoder module and the decoder module comprises:

11. An industrial control device based on model training, comprising:

12. A control system based on the model training-based industrial control method according to any one of claims 1-10, comprising:

training a model system by big data;