CN115629579B - Control method and device of CSTR system - Google Patents
Control method and device of CSTR system Download PDFInfo
- Publication number
- CN115629579B CN115629579B CN202211252946.8A CN202211252946A CN115629579B CN 115629579 B CN115629579 B CN 115629579B CN 202211252946 A CN202211252946 A CN 202211252946A CN 115629579 B CN115629579 B CN 115629579B
- Authority
- CN
- China
- Prior art keywords
- data
- material concentration
- moment
- time
- data set
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 30
- 239000000463 material Substances 0.000 claims abstract description 84
- 238000005457 optimization Methods 0.000 claims abstract description 47
- 238000012417 linear regression Methods 0.000 claims description 21
- 230000002068 genetic effect Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 2
- 230000001276 controlling effect Effects 0.000 abstract description 7
- 238000013178 mathematical model Methods 0.000 abstract description 3
- 230000001105 regulatory effect Effects 0.000 abstract description 3
- 239000002826 coolant Substances 0.000 description 4
- 239000000110 cooling liquid Substances 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000035772 mutation Effects 0.000 description 3
- 239000002994 raw material Substances 0.000 description 3
- 239000000376 reactant Substances 0.000 description 3
- 238000013500 data storage Methods 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000012824 chemical production Methods 0.000 description 1
- 239000003153 chemical reaction reagent Substances 0.000 description 1
- 239000012809 cooling fluid Substances 0.000 description 1
- 239000000975 dye Substances 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006116 polymerization reaction Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000003756 stirring Methods 0.000 description 1
- 229920002994 synthetic fiber Polymers 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
- G05B19/41885—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by modeling, simulation of the manufacturing system
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/32—Operator till task planning
- G05B2219/32339—Object oriented modeling, design, analysis, implementation, simulation language
Landscapes
- Engineering & Computer Science (AREA)
- Manufacturing & Machinery (AREA)
- General Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
Abstract
The embodiment of the application discloses a control method and a device of a CSTR system, which are applied to regulating and controlling control parameters of the CSTR system. Firstly, historical data of a CSTR system at a plurality of continuous moments is obtained, a data set is established, the data set is split based on a tree model and state parameters in the historical data, and a plurality of data subsets are obtained. Then, a linear model of each data subset is built separately, constituting a hierarchical linear model. And finally, collecting real-time data of the CSTR system at the current moment and a plurality of continuous moments before the current moment, solving a preset optimization problem for the control parameters and the material concentration according to the real-time data and the layered linear model, and predicting to obtain the numerical value of the control parameters of the CSTR system at the next continuous moment. The control method and the control device provided by the embodiment of the application can reduce the modeling time of the system physical mathematical model, and simultaneously can design different preset optimization problems according to different requirements of users so as to meet the regulation and control of the control parameters of the CSTR system with different working conditions and structures.
Description
Technical Field
The application belongs to the technical field of CSTR systems, and particularly relates to a control method and device of a CSTR system.
Background
The continuous stirred tank reactor (CSTR, continuous Stirred Tank Reactor) system is the most commonly used reactor for polymerization chemistry in the chemical industry, plays a very important role in core equipment for chemical production, and is widely used in the industries of dyes, foods, pharmaceutical reagents and synthetic materials.
In a CSTR system, the reactant feed flows into the reactor at a constant flow rate to form the reactant, while the reactant flows out of the reactor at the same steady flow rate. After the reaction raw materials flow into the CSTR system, the reaction raw materials and the materials remained in the reactor are completely mixed instantly under the strong stirring action in the reactor, so that the concentration and the temperature in the reactor are equal everywhere. Therefore, the CSTR system can effectively treat the raw materials with high suspended solid content, and avoids layering of the materials.
The CSTR system comprises control parameters for controlling the CSTR system and status parameters for reflecting the status of the CSTR system. In early control of a CSTR system, a position control device or PID (Proportional-Integral-Differential) control consisting of unit meters is often used, and the control mode is to regulate and control parameters of the CSTR system. However, since chemical reaction processes generally have strong nonlinearity and hysteresis, and the position control device and the PID control are more suitable for a linear system with an accurate mathematical model, the position control device and the PID control do not perform well in CSTR system control. Thus, there is a need for a method and apparatus that enables rapid, stable control of CSTR systems.
Disclosure of Invention
The embodiment of the application provides a control method and a device for a CSTR system, which are used for solving the problem that the control parameters of the CSTR system cannot be regulated and controlled rapidly and stably in the prior art.
In order to solve the technical problems, the embodiment of the application discloses the following technical scheme: .
In a first aspect, some embodiments of the present application provide a method for controlling a CSTR system, which is applied to regulate control parameters of the CSTR system; comprising the following steps:
acquiring historical data of a CSTR system at a plurality of continuous moments and establishing a data set, wherein a preset time interval is reserved between every two adjacent continuous moments, and the historical data comprise data of control parameters, data of state parameters and data of material concentration;
splitting the data set based on the tree model and the state parameters to obtain a plurality of data subsets;
respectively establishing a linear model of each data subset to form a layered linear model, wherein input variables of the layered linear model are control parameters and material concentration, and output variables of the layered linear model are material concentration;
collecting real-time data of a CSTR system at the current moment and a plurality of continuous moments before the current moment, wherein the real-time data comprise data of control parameters, data of state parameters and data of material concentration;
and solving a preset optimization problem for the control parameters and the material concentration according to the real-time data and the layered linear model, and predicting to obtain the numerical value of the control parameters of the CSTR system at the next continuous moment.
In a second aspect, some embodiments of the present application provide a control apparatus for a CSTR system, which is used for adjusting and controlling control parameters of the CSTR system; comprising the following steps:
the system comprises a data set establishing unit, a data processing unit and a data processing unit, wherein the data set establishing unit is used for acquiring historical data of a CSTR system at a plurality of continuous moments and establishing a data set, a preset time interval is reserved between every two adjacent continuous moments, and the historical data comprises data of control parameters, data of state parameters and data of material concentration;
a data set splitting unit for splitting the data set based on the tree model and the state parameter to obtain a plurality of data subsets;
the system comprises a layered linear model building unit, a data storage unit and a data storage unit, wherein the layered linear model building unit is used for respectively building a linear model of each data subset to form a layered linear model, input variables of the layered linear model are control parameters and material concentration, and output variables of the layered linear model are material concentration;
the real-time data acquisition unit is used for acquiring real-time data of the CSTR system at the current moment and a plurality of continuous moments before the current moment, wherein the real-time data comprise data of control parameters, data of state parameters and data of material concentration;
and the control parameter prediction unit is used for solving the preset optimization problem of the control parameters and the material concentration according to the real-time data and the layered linear model, and predicting to obtain the numerical value of the control parameters of the CSTR system at the next continuous moment.
The control method and the device for the CSTR system are applied to regulating and controlling the control parameters of the CSTR system. Firstly, historical data of a CSTR system at a plurality of continuous moments is obtained, a data set is established, the data set is split based on a tree model and state parameters in the historical data, and a plurality of data subsets are obtained. Then, a linear model of each data subset is built separately, constituting a hierarchical linear model. And finally, collecting real-time data of the CSTR system at the current moment and a plurality of continuous moments before the current moment, solving a preset optimization problem for the control parameters and the material concentration according to the real-time data and the layered linear model, and predicting to obtain the numerical value of the control parameters of the CSTR system at the next continuous moment.
The control method and the control device provided by the embodiment of the application are suitable for a nonlinear and high-hysteresis CSTR system, can reduce the modeling time of a system physical mathematical model, and can realize accurate and effective estimation of the nonlinear system. Meanwhile, different preset optimization problems can be designed according to different requirements of users, so that the regulation and control of the CSTR system control parameters of different working conditions and different structures can be met.
Drawings
FIG. 1 is a schematic flow chart of a method for controlling a CSTR system according to an embodiment of the present application;
fig. 2 is a schematic flow chart of step S102 in fig. 1 according to an embodiment of the application;
FIG. 3 is a diagram illustrating splitting of a data set according to an embodiment of the present application;
fig. 4 is a schematic flow chart of step S105 in fig. 1 according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of a control device of a CSTR system according to an embodiment of the present application.
Detailed Description
The present application will be described in further detail below with reference to the drawings and detailed description for the purpose of better understanding of the technical solution of the present application to those skilled in the art.
In the disclosed embodiment of this application, the differential relationship of the material concentration of the CSTR system is as follows:
wherein Ca is the material concentration; t is the temperature in the tank, tc is the temperature of the cooling liquid, and T and Tc are the state parameters of the CSTR system; tr is the reference temperature of the cooling fluid and is a control parameter of the CSTR system. According to the relation of the CSTR system, the CSTR system is a system with a strong nonlinear relation due to the existence of the exponential function, and the reference temperature Tr of the cooling liquid can realize the control of the material concentration Ca by controlling the temperature Tc of the cooling liquid and the temperature T in the tank, so that the hysteresis inertia of the CSTR system is also strong. Therefore, accurate regulation of the value of the CSTR system control parameter is important to obtain the desired material concentration.
Fig. 1 is a schematic flow chart of a control method of a CSTR system according to an embodiment of the present application, which can be used to regulate control parameters of the CSTR system. As shown in fig. 1, the control method includes the steps of:
step S101: historical data of the CSTR system at a plurality of successive times is acquired and a data set is established.
The historical data comprise data of control parameters, data of state parameters and data of material concentration.
In CSTR systems, the state parameters and the control parameters are of a large variety, and for ease of understanding, the embodiments disclosed herein take only the in-tank temperature, the coolant temperature, and the coolant reference temperature as examples. Wherein the state parameters of the CSTR system, including the in-tank temperature and the coolant temperature, can be collected by sensors provided on the CSTR system. The control parameter of the CSTR system is the reference temperature of the coolant.
The method comprises the steps of acquiring state parameters, control parameters and material concentration data of a CSTR system at a plurality of continuous moments in the past as historical data, wherein a preset time interval is reserved between every two adjacent continuous moments. In one embodiment of the present disclosure, historical data is obtained for a CSTR system at a plurality of consecutive times over the past five days, with a 1 minute time interval between adjacent consecutive times. For example, the history data at a certain time represents the state of the CSTR system within one minute from the time, and the history data at the next continuous time adjacent to the time represents the state of the CSTR system within the next minute.
And establishing a data set from the acquired historical data, wherein samples in the data set are state parameter values, control parameter values and material concentration values of the CSTR system at one continuous moment.
Step S102: the data set is split based on the tree model and the state parameters to obtain a plurality of data subsets.
In one embodiment of the present disclosure, as shown in FIG. 2, the following sub-steps may be employed to split the data set.
Step S1021: a plurality of branch nodes of the data set are determined based on a binary tree model.
In the disclosed embodiment of the application, the input variable of the binary tree model is a state parameter, and the output variable is a material concentration. This step may be accomplished by:
firstly, a binary tree model is established according to the following method, the data set to be split is pre-split, the purpose of pre-split is only to determine branch nodes of the data set, and the data set is not actually split.
D 1 ＝(Y i ,Cv i )∈D|Cv i <a,i＝1,2,3...N
D 2 ＝(Y i ,Cv i )∈D|Cv i ≥a,i＝1,2,3...N
And selecting a sample corresponding to the SSE minimum value as a pending branch node a, namely selecting a plurality of samples of the data set to be split, trying one by one, and finally taking the sample adopted when the SSE minimum value is obtained as the pending branch node a.
Wherein D is the data set to be split, N is the total number of samples in the data set to be split, a is the undetermined branch node of the data set to be split, Y i Outputting the data value of the variable, cv, for the ith sample in the data set D to be split i Inputting the data value of the variable for the ith sample in the data set to be split D 1 And D 2 Respectively generating two data subsets based on pre-splitting data set D of undetermined branch node a, c 1 And c 2 Respectively D 1 And D 2 Mean of the output variables.
Whether the number of samples of each pre-split data subset is larger than the total number of preset samples is judged, for example, the total number of the preset samples is 10, and whether the number of samples of each pre-split data subset is larger than 10 is judged.
If the number of samples of each pre-split data subset is greater than the total number of preset samples, the scale of each pre-split data subset is enough, and the current undetermined branch node can be used as a branch node a of the data set for final splitting of the data set. Then, continuing to pre-split each pre-split data subset by adopting the method until the size of the pre-split data subset does not meet the requirement, i.e. the number of samples of the pre-split data subset is smaller than the total number of preset samples.
If the number of samples of each pre-split data subset is not greater than the total number of preset samples, or the number of samples of one of the pre-split data subsets is not greater than the total number of preset samples, the pre-split data subsets are not continuously split, and the to-be-determined branch node adopted in the pre-split process can not be used as the branch node of the data set.
Step S1022: the data set is split into a plurality of data subsets with each branch node.
Splitting the data set into a plurality of data subsets D using all the branching nodes determined in step S1021 i For specific splitting, reference is made to the description of pre-splitting in the above embodiments. Taking a simple example of fig. 3, where rounded rectangles represent the data sets to be split and circles represent the data subsets that are ultimately split.
The values of the samples in each data subset are close together, whereby the data set can be divided into subsets according to the degree of similarity of the samples.
The data subset after splitting the data set by the branch node can be expressed as:
S＝{D 1 ,D 2 ,D 3 ......D k }
wherein S is a data set established by historical data, D i For the ith data subset, k is the number of data subsets, and the number of samples of each data subset is greater than the total number of preset samples.
Step S103: and respectively establishing a linear model of each data subset to form a layered linear model.
Because the samples in each data subset have similarities, in the disclosed embodiment of the application, a linear model is built for each data subset, so that each linear model can be adapted to the samples in the corresponding data subset to the greatest extent.
After each linear model of the data subset is built, each linear model is combined to form a layered linear model, wherein input variables of the layered linear model are control parameters and material concentration, output variables of the layered linear model are material concentration, and each layer in the layered linear model corresponds to the linear model of one data subset one by one.
In one embodiment of the present disclosure, step S103 may be accomplished in the following manner.
And (I) respectively establishing a linear regression model of each data subset, wherein the input variable of the linear regression model is the value of the control parameter and the material concentration at m+1 continuous moments, and the output variable is the value of the material concentration at the corresponding m continuous moments and the next continuous moment.
In the embodiment of the application, the material concentration at a certain moment is considered to be related to the control parameters of the CSTR system and the material concentration at a certain time before the moment, so that when the linear regression model of each data subset is built, the relevant parameters of the CSTR system at a previous time are included.
For each subset of data, a linear regression model is built according to the following equation:
F t ＝A×F t-1 +B×U t-1
wherein t is the current time or a corresponding historical time in the historical data; a and B are coefficient matrixes in the linear regression model respectively, and the linear regression models of different data subsets have different coefficient matrixes; a, a i And b i Coefficients of coefficient matrices a and B, respectively; m is a preset number, and is the observation length, namely m continuous moments; y is Y t-i The value representing the output variable at time t-i, i=1, 2,3. Once again, m is chosen, the t-i moment is the ith continuous moment before the current moment or the historical moment; y is Y t Outputting the numerical value of the variable for the current time or the historical time; xc t-i The values representing the input variables at time t-i, i=1, 2,3.
And (II) fitting the linear regression models of each data subset to form a layered linear model, wherein each layer in the layered linear model is respectively in one-to-one correspondence with the linear regression model of one data subset.
The hierarchical linear model is shown as follows:
F t ＝A 1 ×F t-1 +B 1 ×U t-1 F∈D 1 ,U∈D 1
F t ＝A 2 ×F t-1 +B 2 ×U t-1 F∈D 2 ,U∈D 2
……
F t ＝A K ×F t-1 +B k ×U t-1 F∈D k ,U∈D k
wherein A is i 、B i Coefficient matrixes of the ith layer are respectively obtained, and the coefficient matrixes in different layers are different; d (D) i Is the i-th subset of data.
Step S104: real-time data of the CSTR system at the current moment and a plurality of continuous moments before the current moment are collected.
The current moment and m continuous moments before the current moment are obtained as real-time data by means of a sensor arranged on the CSTR system, wherein m is a preset number consistent with the previous embodiments. The real-time data includes data of control parameters, data of status parameters, and data of material concentrations.
Step S105: according to the real-time data and the layered linear model, solving the preset optimization problem of the control parameters and the material concentration, and predicting to obtain the numerical value of the control parameters of the CSTR system at the next continuous moment.
When the real-time data is substituted into the hierarchical linear model, it is first determined to which layer in the hierarchical linear model the real-time data belongs, i.e., to which subset of data the real-time data has the highest similarity. And predicting the material concentration at a continuous moment after the current moment by using a corresponding linear regression model in the layered linear model. In the embodiment of the application, the material concentration predicted value and the control parameter predicted value of a plurality of continuous moments in the future are solved simultaneously together with the layered linear model by utilizing the preset optimization problem of the control parameters and the material concentration, so that the predicted control parameters can meet the optimization target of the preset optimization problem.
In one embodiment of the present disclosure, as shown in FIG. 4, step S105 may be implemented using the following substeps.
Step S1051: substituting the real-time data into a layered linear model, and simultaneously solving the material concentration predicted values of a plurality of continuous moments after the current moment and the control parameter predicted values of a plurality of continuous moments after the current moment by adopting a genetic algorithm according to a preset optimization problem.
In one embodiment of the disclosure, the genetic algorithm employs a meta-heuristic optimization algorithm, wherein three operators, selection, crossover and mutation, determine the performance of the genetic algorithm: the selection operator determines how to select the individuals producing the offspring, and generally speaking, the individuals with higher fitness have higher selection probability; the crossover operator defines the rule of information transfer between individuals, namely, the mode of generating next generation individuals between individuals; mutation operators are random manifestations in genetic algorithms, and can enable individuals to break through the limit of current search and generate individuals with brand new characteristics.
The meta heuristic optimization algorithm can be simply understood as being implemented according to the following steps:
1. binary encoding is carried out on the data characteristics;
2. initializing a population;
3. evaluating fitness of individuals in the population;
4. selecting a plurality of individuals from the population with a certain probability;
5. performing cross operation on the selected individuals to generate next generation individuals;
6. reducing the mutation of a segment in the individual with small probability, and then re-executing the steps 3-6 until the preset iteration times are reached.
In one embodiment of the present disclosure, this step S1051 may be implemented by:
the predicted value of the material concentration at h continuous moments after the current moment t is expressed by the following formula:
wherein t is the current time,for the current momentthe predicted value of the material concentration at the i-th continuous time after t, i=1, 2,3 … … h, where h is a preset number, and represents h continuous times after the current time, for example, h=10.
According to the real-time data, the layered linear model and the predicted value of the control parameter, the material concentration at the future continuous moment after the current moment can be obtained.
The expression of the material concentration predicted values of h continuous moments after the current moment t is defined so as to be substituted into a preset optimization problem, and the material concentration predicted values and the control parameter predicted values at the future moment are obtained by simultaneously solving the preset optimization problem.
Acquiring material concentration reference values of h continuous moments after the current moment t, wherein the material concentration reference values are expressed by the following formula:
ref＝[r tp1 ,r t+2 ,...,r t+h ]
wherein r is t+i I=1, 2,3 … … h, which is the material concentration reference value at the i-th continuous time after the current time t.
In a specific embodiment of the present disclosure, a material concentration reference value of the CSTR system at a future continuous time is preset, and the material concentration reference value can represent a value that can be reached by the expected material concentration at a future time.
The preset optimization problem is as follows:
Opt＝min loss(result,ref)
wherein, opt is an optimization target of a preset optimization problem, loss is a loss function of the optimization target Opt;g and h are respectively the optimization conditions of a preset optimization problem, lower and upper are respectively the lower and upper bounds of a control parameter predicted value, g and h respectively represent inequality and equality constraint functions, incons and equens are respectively the inequality and equality constraint conditions of the optimization problem, Y is the material concentration predicted value to be solved in the optimization problem, and%>The predicted value of the control parameter to be solved in the optimization problem is expressed as:
wherein,i=1, 2,3 and … … h for the predicted value of the control parameter to be solved at the ith continuous moment after the current moment t.
In practical application, the optimization target and the optimization condition of the preset optimization problem in the embodiment of the present application may be set according to the needs of the user, and are not limited to the modes listed in the above embodiment.
In one embodiment of the present disclosure, the loss function of the optimization objective is calculated using the mean square error MSE, with the following calculation formula:
wherein result [ i ] is a predicted value of the material concentration at the ith continuous moment after the current moment, ref [ i ] is a reference value of the material concentration at the ith continuous moment after the current moment, and i=1, 2,3 … … h. The preset optimization problem is to take the minimum realization loss function as an optimization target, and then solve and obtain a control parameter predicted value and a material concentration predicted value.
In one embodiment of the present disclosure,
the optimization objective is set as follows:
Opt＝min loss(result,ref)
the optimization conditions were set as follows:
wherein Y represents the concentration of the material to be solved in the preset optimization problem,the control parameters to be solved in the preset optimization problem are expressed, the above expression indicates that the material concentration must be between 0.35 and 0.65mol/L, and the control parameter cooling liquid reference temperature must be between 335 and 372K.
Step S1052: and determining the predicted value of the control parameter of the next continuous moment at the current moment as the value of the control parameter of the CSTR system at the next continuous moment.
According to solving the preset optimization problem, control parameter predicted values of h continuous moments can be obtained, and the control parameter predicted value of the first moment, namely the next continuous moment t+1 of the current moment t, is determined to be the numerical value of the control parameter of the CSTR system at the next continuous moment, so that the control of the control parameter of the CSTR system is realized, and the true material concentration value of the CSTR system at the next continuous moment is enabled to be as close as possible to the expected reference value.
Fig. 5 is a schematic structural diagram of a control device of a CSTR system according to an embodiment of the present application, which is applied to regulate control parameters of the CSTR system. As shown in fig. 5, the control device includes the following units:
a data set establishing unit 11 configured to acquire historical data of the CSTR system at a plurality of continuous moments and establish a data set, wherein a preset time interval is arranged between every two adjacent continuous moments, and the historical data comprises data of control parameters, data of state parameters and data of material concentration;
a data set splitting unit 12 configured to split the data set based on the tree model and the state parameter, obtaining a plurality of data subsets;
a layered linear model building unit 13 configured to build a linear model of each data subset, respectively, to constitute a layered linear model, wherein input variables of the layered linear model are control parameters and material concentrations, and output variables are material concentrations;
a real-time data acquisition unit 14 configured to acquire real-time data of the CSTR system at a current time and a plurality of consecutive times before the current time, the real-time data including data of control parameters, data of status parameters, and data of material concentration;
the control parameter prediction unit 15 is configured to solve a preset optimization problem on the control parameters and the material concentration according to the real-time data and the layered linear model, and predict and obtain the numerical value of the control parameters of the CSTR system at the next continuous moment.
It is to be understood that the above embodiments are merely illustrative of the application of the principles of the present application, but not in limitation thereof. Various modifications and improvements may be made by those skilled in the art without departing from the spirit and substance of the application, and are also considered to be within the scope of the application.
Claims (5)
1. A control method of CSTR system is applied to control parameters of CSTR system; the control method is characterized by comprising the following steps:
acquiring historical data of a CSTR system at a plurality of continuous moments and establishing a data set, wherein a preset time interval is reserved between every two adjacent continuous moments, and the historical data comprise data of control parameters, data of state parameters and data of material concentration;
splitting the data set based on the tree model and the state parameters to obtain a plurality of data subsets, including:
determining a plurality of branch nodes of a data set based on a binary tree model, wherein the input variable of the binary tree model is a state parameter, and the output variable is material concentration;
a binary tree model is established according to the following method, and the data set to be split is pre-split:
D 1 ＝(Y i ，Cv i )∈D|Cv i ＜a，i＝1，2，3...N
D 2 ＝(Y i ，Cv i )∈D|Cv i ≥a,i＝1，2，3...N
selecting a sample corresponding to the SSE minimum value as a pending branch node a;
wherein D is the data set to be split, N is the total number of samples in the data set to be split, a is the undetermined branch node of the data set to be split, Y i Cv is the value of the output variable in the ith sample in the data set D to be split i For the value of the input variable in the ith sample in the data set D to be split, D 1 And D 2 Respectively generating two data subsets based on pre-splitting data set D of undetermined branch node a, c 1 And c 2 Respectively D 1 And D 2 Average value of the medium output variable;
determining whether the number of samples of each pre-split data subset is greater than a preset total number of samples,
if yes, taking the undetermined branch node as a branch node of the data set, and adopting the method to continuously pre-split each pre-split data subset;
if not, stopping pre-splitting the data subset;
establishing a linear model of each data subset respectively to form a layered linear model, wherein the method comprises the following steps:
the method comprises the following steps of respectively establishing a linear regression model of each data subset, wherein input variables of the linear regression model are values of control parameters and material concentration at m+1 continuous moments, and output variables are values of the material concentration at corresponding m continuous moments and the next continuous moment:
F t ＝A×F t-1 +B×U t-1
wherein t is the current time or a corresponding historical time in the historical data; a and B are coefficient matrixes in the linear regression model respectively, and the linear regression models of different data subsets have different coefficient matrixes; a, a i And b i Coefficients of coefficient matrices a and B, respectively; m is a preset number, namely m continuous moments; y is Y t-i The value representing the output variable at time t-i, i=1, 2,3. Once again, m is chosen, the t-i moment is the i th continuous moment before the current moment or the historical moment; y is Y t Outputting the numerical value of the variable for the current time or the historical time; xc t-i The value representing the input variable at time t-i, i=1, 2,3. M;
fitting the linear regression models of each data subset to form a layered linear model, wherein each layer in the layered linear model is respectively in one-to-one correspondence with the linear regression model of one data subset; the input variables of the layered linear model are control parameters and material concentration, and the output variables are material concentration;
collecting real-time data of a CSTR system at the current moment and a plurality of continuous moments before the current moment, wherein the real-time data comprise data of control parameters, data of state parameters and data of material concentration;
according to the real-time data and the layered linear model, solving a preset optimization problem of the control parameters and the material concentration, and predicting to obtain the numerical value of the control parameters of the CSTR system at the next continuous moment, wherein the method comprises the following steps:
substituting the real-time data into a layered linear model, and simultaneously solving material concentration predicted values at a plurality of continuous moments after the current moment and control parameter predicted values at a plurality of continuous moments after the current moment by adopting a genetic algorithm according to a preset optimization problem;
and determining the predicted value of the control parameter of the next continuous moment at the current moment as the value of the control parameter of the CSTR system at the next continuous moment.
2. The control method according to claim 1, wherein splitting the data set based on the tree model and the state parameter to obtain a plurality of data subsets comprises:
the data set is split into a plurality of data subsets with each of the branching nodes.
3. The control method according to claim 1, wherein the substituting of the real-time data into the hierarchical linear model and the simultaneous solving of the predicted values of the material concentration at a plurality of consecutive times after the current time and the predicted values of the control parameter at a plurality of consecutive times after the current time by using the genetic algorithm according to the preset optimization problem includes:
the predicted values of the material concentration at h continuous moments after the current moment are expressed by the following formula:
wherein t is the current time,i=1, 2,3, which is the predicted value of the material concentration at the i-th successive moment after the current moment t... h, h is a predetermined number;
acquiring material concentration reference values of h continuous moments after the current moment, wherein the material concentration reference values are expressed by the following formula:
ref＝[r t+1 ,r t+2 ,...,r t+h ]
wherein r is t+i As a material concentration reference value at the i-th successive time after the current time t, i=1, 2,3.
The preset optimization problem is as follows:
Opt＝minloss(result,ref)
wherein, opt is an optimization target of a preset optimization problem, loss is a loss function of the Opt, lower and upper are respectively a lower bound and an upper bound of a predicted value of a control parameter, g and h respectively represent an inequality constraint function and an equality constraint function, incons and equs respectively represent an inequality constraint condition and an equality constraint condition, Y is a predicted value of the concentration of a material to be solved in the preset optimization problem,the predicted value of the control parameter to be solved in the preset optimization problem is expressed as:
wherein,to-be-solved control for ith continuous moment after current moment tParameter predictions, i=1, 2,3.
4. A control method according to claim 3, characterized by further comprising:
and calculating a loss function of the optimization target by adopting an average square error MSE, wherein the calculation formula is as follows:
wherein result [ i ] is a predicted value of the material concentration at the i-th continuous time after the current time, ref [ i ] is a reference value of the material concentration at the i-th continuous time after the current time, i=1, 2,3.
5. A control device of a CSTR system is applied to regulate and control parameters of the CSTR system; characterized in that the control device comprises:
the system comprises a data set establishing unit, a data processing unit and a data processing unit, wherein the data set establishing unit is used for acquiring historical data of a CSTR system at a plurality of continuous moments and establishing a data set, a preset time interval is reserved between every two adjacent continuous moments, and the historical data comprises data of control parameters, data of state parameters and data of material concentration;
a data set splitting unit, configured to split a data set based on a tree model and a state parameter, to obtain a plurality of data subsets, including:
determining a plurality of branch nodes of a data set based on a binary tree model, wherein the input variable of the binary tree model is a state parameter, and the output variable is material concentration;
a binary tree model is established according to the following method, and the data set to be split is pre-split:
D 1 ＝(Y i ，Cv i )∈D|Cv i ＜a，i＝1，2，3...N
D 2 ＝(Y i ，Cv i )∈D|Cv i ≥a，i＝1，2，3...N
selecting a sample corresponding to the SSE minimum value as a pending branch node a;
wherein D is the data set to be split, N is the total number of samples in the data set to be split, a is the undetermined branch node of the data set to be split, Y i Cv is the value of the output variable in the ith sample in the data set D to be split i For the value of the input variable in the ith sample in the data set D to be split, D 1 And D Z Respectively generating two data subsets based on pre-splitting data set D of undetermined branch node a, c 1 And c 2 Respectively D 1 And D 2 Average value of the medium output variable;
determining whether the number of samples of each pre-split data subset is greater than a preset total number of samples,
if yes, taking the undetermined branch node as a branch node of the data set, and adopting the method to continuously pre-split each pre-split data subset;
if not, stopping pre-splitting the data subset;
the hierarchical linear model building unit is configured to build a linear model of each data subset, and form a hierarchical linear model, and includes:
the method comprises the following steps of respectively establishing a linear regression model of each data subset, wherein input variables of the linear regression model are values of control parameters and material concentration at m+1 continuous moments, and output variables are values of the material concentration at corresponding m continuous moments and the next continuous moment:
F t ＝A×F t-1 +B×U t-1
wherein t is the current time or a corresponding historical time in the historical data; a and B are coefficient matrixes in the linear regression model respectively, and the linear regression models of different data subsets have different coefficient matrixes; a, a i And b i Coefficients of coefficient matrices a and B, respectively; m is a preset number, namely m continuous moments; y is Y ti The value representing the output variable at time t-i, i=1, 2,3. Once again, m is chosen, the t-i moment is the i th continuous moment before the current moment or the historical moment; y is Y t Outputting the numerical value of the variable for the current time or the historical time; x is X ti The value representing the input variable at time t-i, i=1, 2,3. M;
fitting the linear regression models of each data subset to form a layered linear model, wherein each layer in the layered linear model is respectively in one-to-one correspondence with the linear regression model of one data subset; the input variables of the layered linear model are control parameters and material concentration, and the output variables are material concentration;
the real-time data acquisition unit is used for acquiring real-time data of the CSTR system at the current moment and a plurality of continuous moments before the current moment, wherein the real-time data comprise data of control parameters, data of state parameters and data of material concentration;
the control parameter prediction unit is configured to solve a preset optimization problem for the control parameter and the material concentration according to the real-time data and the layered linear model, and predict to obtain a numerical value of the control parameter of the CSTR system at the next continuous moment, where the numerical value comprises:
substituting the real-time data into a layered linear model, and simultaneously solving material concentration predicted values at a plurality of continuous moments after the current moment and control parameter predicted values at a plurality of continuous moments after the current moment by adopting a genetic algorithm according to a preset optimization problem;
and determining the predicted value of the control parameter of the next continuous moment at the current moment as the value of the control parameter of the CSTR system at the next continuous moment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211252946.8A CN115629579B (en) | 2022-10-13 | 2022-10-13 | Control method and device of CSTR system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211252946.8A CN115629579B (en) | 2022-10-13 | 2022-10-13 | Control method and device of CSTR system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115629579A CN115629579A (en) | 2023-01-20 |
| CN115629579B true CN115629579B (en) | 2023-11-07 |
Family
ID=84905678
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211252946.8A Active CN115629579B (en) | 2022-10-13 | 2022-10-13 | Control method and device of CSTR system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115629579B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101420796A (en) * | 2008-12-02 | 2009-04-29 | 郭震 | Power-on heating device |
| CN103927506A (en) * | 2014-04-25 | 2014-07-16 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | RIFD anti-collision method based on dynamic timeslot conflict tracking tree |
| CN107533995A (en) * | 2015-05-08 | 2018-01-02 | 科磊股份有限公司 | Hot spot monitoring based on model |
| CN109214117A (en) * | 2018-10-15 | 2019-01-15 | 南京天洑软件有限公司 | A kind of intelligent industrial algorithm for design based on value network |
| CN109997154A (en) * | 2017-10-30 | 2019-07-09 | 上海寒武纪信息科技有限公司 | Information processing method and terminal device |
| CN112308273A (en) * | 2019-07-31 | 2021-02-02 | 中国石油化工股份有限公司 | Memory, petrochemical enterprise pollution discharge management method, device and equipment |
| WO2021080824A1 (en) * | 2019-10-25 | 2021-04-29 | Dow Global Technologies Llc | Nonlinear model predictive control of a process |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US12081651B2 (en) * | 2018-12-06 | 2024-09-03 | Schneider Electric Systems Usa, Inc. | One-time pad encryption for industrial wireless instruments |
| US20220058317A1 (en) * | 2020-08-23 | 2022-02-24 | Ricardo Reis De Carvalho | Smart process control system for continuous treatment of felts |
-
2022
- 2022-10-13 CN CN202211252946.8A patent/CN115629579B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101420796A (en) * | 2008-12-02 | 2009-04-29 | 郭震 | Power-on heating device |
| CN103927506A (en) * | 2014-04-25 | 2014-07-16 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | RIFD anti-collision method based on dynamic timeslot conflict tracking tree |
| CN107533995A (en) * | 2015-05-08 | 2018-01-02 | 科磊股份有限公司 | Hot spot monitoring based on model |
| CN109997154A (en) * | 2017-10-30 | 2019-07-09 | 上海寒武纪信息科技有限公司 | Information processing method and terminal device |
| CN109214117A (en) * | 2018-10-15 | 2019-01-15 | 南京天洑软件有限公司 | A kind of intelligent industrial algorithm for design based on value network |
| CN112308273A (en) * | 2019-07-31 | 2021-02-02 | 中国石油化工股份有限公司 | Memory, petrochemical enterprise pollution discharge management method, device and equipment |
| WO2021080824A1 (en) * | 2019-10-25 | 2021-04-29 | Dow Global Technologies Llc | Nonlinear model predictive control of a process |
Non-Patent Citations (3)
| Title |
|---|
| Lo S N.Optimization of output fluctuation model of non-ideal CSTR's in series with a sinusoidal input of concentration.《Chemical Engineering Science》.1984,全文. * |
| 基于小波域隐马尔可夫树模型的过程趋势分析;李成, 宋执环, 李平;信息与控制(03);全文 * |
| 面向Fluent的PABR反应器三维模型分割方法;孙浩鹏;;科技创新与应用(27);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115629579A (en) | 2023-01-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhu et al. | Scalable reinforcement learning for plant-wide control of vinyl acetate monomer process | |
| Angelov | Evolving rule-based models: a tool for design of flexible adaptive systems | |
| Angelov et al. | Automatic generation of fuzzy rule-based models from data by genetic algorithms | |
| Ellis et al. | An encoder–decoder LSTM-based EMPC framework applied to a building HVAC system | |
| Tan et al. | A three-step machine learning framework for energy profiling, activity state prediction and production estimation in smart process manufacturing | |
| JP3790129B2 (en) | Process simulator application control apparatus and method | |
| Ravi et al. | Design of Intelligent Self‐Tuning GA ANFIS Temperature Controller for Plastic Extrusion System | |
| Sarkar et al. | ANNSA: a hybrid artificial neural network/simulated annealing algorithm for optimal control problems | |
| CN108287474A (en) | Based on the probabilistic catalytic reforming reactor robust operation optimization method of raw material | |
| CN117557016A (en) | A stamping resource scheduling method for vehicle manufacturing based on deep reinforcement learning | |
| CN115629579B (en) | Control method and device of CSTR system | |
| CN118318274A (en) | Method for generating artificial intelligence model for process control, process control system based on artificial intelligence model and reactor including the same | |
| Rohkohl et al. | Multi-criteria and real-time control of continuous battery cell production steps using deep learning | |
| Tan et al. | Soft computing applications in dynamic model identification of polymer extrusion process | |
| CN107885086A (en) | Autonomous navigation device control parameter on-line control method based on MCMC optimization Q study | |
| Chan et al. | Modelling and optimization of fluid dispensing for electronic packaging using neural fuzzy networks and genetic algorithms | |
| CN112016241A (en) | Soft measurement modeling method based on mLASSO-MLP model | |
| Safari et al. | A hybrid attention‐based long short‐term memory fast model for thermal regulation of smart residential buildings | |
| Soukkou et al. | Optimal control of a CSTR process | |
| Zhang et al. | Optimization control of a fed-batch process using an improved reinforcement learning algorithm | |
| Sagdatullin | Application of fuzzy logic and neural networks methods for industry automation of technological processes in oil and gas engineering | |
| Zhang et al. | Active learning strategy for online prediction of particle size distribution in cobalt oxalate synthesis process | |
| CN113657038B (en) | Data-driven reaction temperature optimization method and device for hydrocracking process | |
| Baron et al. | Reliable on-line re-optimization control of a fed-batch fermentation process using bootstrap aggregated extreme learning machine | |
| CN106127326B (en) | Chemical material processing melt index forecasting method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |