CN-117170332-B - Multi-working-condition industrial control method and system based on integrated space-time model prediction

CN117170332BCN 117170332 BCN117170332 BCN 117170332BCN-117170332-B

Abstract

The invention discloses a multi-working-condition industrial control method and system based on integrated space-time model prediction, wherein the method comprises the steps of constructing a working condition identifier based on a typical working condition characteristic extraction method of an orthogonal test; the method comprises the steps of constructing a time dynamic model and a space distribution model under each working condition, obtaining optimal parameters of a space-time model corresponding to each working condition by adopting a data driving and integrated training method, identifying the current working condition of the multi-working-condition industrial system in real time by utilizing a working condition identifier, and obtaining the current optimal control input of the system by taking the space-time model under the current working condition as a prediction model in a rolling optimization mode. According to the method, the prediction model of the non-observable point is built by learning the space-time correlation of the observable point and the non-observable point, and the prediction model is integrated into the prediction control framework, so that the precise control of the non-observable point is realized.

Inventors

YANG CHUNHUA
ZHANG YUNFENG
HUANG KEKE
WU DEHAO
ZHOU CAN
SUN BEI
GUI WEIHUA

Assignees

中南大学

Dates

Publication Date: 20260512
Application Date: 20231016

Claims (8)

1. The utility model provides a multi-working condition industrial control method based on integration space-time model prediction, which is characterized by comprising the following steps: s1, constructing a working condition identifier based on a typical working condition characteristic extraction method of an orthogonal test; s2, constructing a time dynamic model and a space distribution model under each working condition, and obtaining optimal parameters of a space-time model corresponding to each working condition by adopting a data driving and integrated training method; aiming at non-observable points which are not arranged by the sensors in the multi-working-condition industrial system, a space distribution model is built; S3, the current working condition of the multi-working-condition industrial system is identified in real time by using the working condition identifier, a space-time model under the current working condition is used as a prediction model, and the current optimal control input of the system is obtained in a rolling optimization mode.
2. The multi-task industrial control method based on integrated space-time model prediction according to claim 1, wherein step S1 specifically comprises: firstly, determining key influence factors of the states of a multi-working-condition industrial system, and recording the key influence factors as the number of the key influence factors ; Then, for each key influence factor, determining a horizontal value according to the change range of the key influence factor, and recording the first The number of levels of the key influencing factors is ; Based on factor number And number of levels Matching and corresponding to obtain a typical orthogonal table, and acquiring operation parameters of the system under each typical working condition according to the orthogonal table, wherein each element in the typical orthogonal table corresponds to one typical working condition; And finally, constructing a training sample for identifying the working condition by utilizing the operation parameters of each typical working condition and the corresponding working condition label, and training the neural network to obtain the working condition identifier.
3. The multi-station industrial control method based on integrated space-time model prediction as set forth in claim 2, wherein the operating parameters under each typical operating condition are obtained by collecting data of a predetermined duration through sensors distributed at various positions of the system, and wherein the first step is to obtain the data of the predetermined duration by the sensors distributed at various positions of the system The operating parameters under typical conditions are expressed as : ; In the right subscript Represents the first In a typical operating mode of the device, Represent the first Time of day The sensor at the location monitors a value that, The number of the sensors is the number of the sensors, Is the training data length under each working condition.
4. The integrated space-time model prediction based multi-task industrial control method according to claim 1, wherein the first step is The input-output relationship of the time dynamic model of each sensor is as follows: ; ; In the formula, Is the first Personal sensor Predicting and outputting time; Represent the first Personal sensor A time dynamic model of the position is constructed by adopting a neural network, Model parameters representing a corresponding time dynamic model; Is the first Personal sensor Input of a temporal dynamic model at; Is a multi-working condition industrial system sequentially at sampling time points Is used for the input of the (c) to be processed, A time lag label is input; Is the first Individual sensors at sampling time points Is used for the monitoring of the value of (a), To output a time lag label.
5. The integrated space-time model prediction based multi-task industrial control method of claim 4, wherein the spatial distribution model is expressed as: ; ; In the formula, Is the spatial coordinates of any unobservable point, Is that At the position Predicting and outputting time; is a space distribution model, is constructed by adopting a neural network, Parameters representing a spatial distribution model; Time dynamic model representing all sensors The prediction of the time outputs the vector.
6. The multi-task industrial control method based on integrated space-time model prediction according to claim 5, wherein the data driving and integrated training method is adopted to obtain the optimal parameters of the space-time model corresponding to each working condition, and the used objective function is: ; In the formula, Model parameters representing the time dynamic model corresponding to the n sensors, Represent the first Each sensor corresponds to a loss function term of the time dynamic model, A loss function term representing a spatial distribution model, The relative contribution of the loss function term of each time dynamic model and the loss function term of the spatial distribution model is controlled.
7. The integrated space-time model prediction based multi-task industrial control method according to claim 6, wherein the first step is The loss function of the corresponding time dynamic model of each sensor is as follows: ; In the formula, Represent the first Personal sensor Monitoring values of time; the loss function of the spatial distribution model is: ; In the formula, Representation of At the position True value of time of day.
8. A multi-working condition industrial control system based on integrated space-time model prediction is characterized by comprising a working condition identifier and a plurality of space-time models which are in one-to-one correspondence with different working conditions; the working condition identifier is constructed based on a typical working condition characteristic extraction method of an orthogonal test and is used for identifying the current working condition of the multi-working condition industrial system in real time; The space-time model consists of a time dynamic model and a space distribution model, obtains optimal parameters by adopting a data driving and integrated training method, is used as a prediction model under the current working condition, and obtains the current optimal control input of the system by a rolling optimization mode; And establishing a space distribution model aiming at non-observable points which are not arranged by the sensors in the multi-working-condition industrial system.

Description

Multi-working-condition industrial control method and system based on integrated space-time model prediction Technical Field The invention belongs to the technical field of control, and particularly relates to a multi-working-condition industrial control method and system aiming at an unobservable state of a distributed parameter system. Background With the development and progress of modern industrial technology, the industrial process becomes more and more complex, such as 600KA aluminum electrolysis cell, 152m 2 roasting furnace. The multi-phase multi-field reaction system coupling interaction inside the system generally has the characteristics of space-time coupling, infinite and nonlinearity and the like, and is difficult to accurately model and control, so that the system cannot stably operate for a long time, and therefore, the system is very important for the accurate control of a large-scale industrial process. The existing control method is mainly divided into a lumped parameter system method and a distributed parameter system method. The lumped parameter system method considers that the state index in the system does not change along with the space, and the whole system can be controlled by observing and controlling the specific position points. Although the lumped parameter system method can control the system to a certain extent, the spatial distribution characteristics of the large-scale industrial process are ignored, so that the lumped parameter system method is difficult to accurately model and control the system. Whereas the distributed parameter system method uses partial differential equations in combination with specific initial conditions and boundary conditions to describe the object, because the partial differential equations contain time partial derivatives and space partial derivatives, the distributed parameter system method can describe the object from two angles of time dynamic and space distribution. However, conventional modeling methods of distributed parameter systems generally require PDE equations to be known, but industrial sites lack explicit knowledge of the mechanism, which is difficult to meet. With the development of information technologies such as big data and sensor technology, various sensors with sufficient quantity are deployed on an industrial site, a large amount of data is accumulated, a new path is provided for the control of a system by the data, a data-driven distributed parameter system modeling and control method is provided, the method utilizes the large amount of data accumulated on the industrial site, a space-time reduced order model of the system is obtained through learning, and a controller is designed according to the space-time reduced order model. However, conventional data-driven distributed parametric system modeling and control methods generally assume that all observed states are known and ignore the problem of model mismatch due to operating condition variations. Disclosure of Invention Aiming at the problem of accurate control of a large-scale industrial process with an undetectable state in the prior art, the invention provides a multi-working condition industrial control method and system based on integrated space-time model prediction. In order to achieve the technical purpose, the invention adopts the following technical scheme: a multi-working condition industrial control method based on integrated space-time model prediction comprises the following steps: s1, constructing a working condition identifier based on a typical working condition characteristic extraction method of an orthogonal test; s2, constructing a time dynamic model and a space distribution model under each working condition, and obtaining optimal parameters of a space-time model corresponding to each working condition by adopting a data driving and integrated training method; S3, the current working condition of the multi-working-condition industrial system is identified in real time by using the working condition identifier, a space-time model under the current working condition is used as a prediction model, and the current optimal control input of the system is obtained in a rolling optimization mode. Further, the step S1 specifically includes: firstly, determining key influence factors of the states of a multi-working-condition industrial system, and recording the key influence factors as the number of the key influence factors ; Then, for each key influence factor, determining a horizontal value according to the change range of the key influence factor, and recording the firstThe number of levels of the key influencing factors is; Based on factor numberAnd number of levelsMatching and corresponding to obtain a typical orthogonal table, and acquiring operation parameters of the system under each typical working condition according to the orthogonal table, wherein each element in the typical orthogonal table corresponds to one typical working condition; And final