CN116880191A - Intelligent control method of process industrial production system based on time sequence prediction - Google Patents
Intelligent control method of process industrial production system based on time sequence prediction Download PDFInfo
- Publication number
- CN116880191A CN116880191A CN202310879391.8A CN202310879391A CN116880191A CN 116880191 A CN116880191 A CN 116880191A CN 202310879391 A CN202310879391 A CN 202310879391A CN 116880191 A CN116880191 A CN 116880191A
- Authority
- CN
- China
- Prior art keywords
- manipulated
- variable
- time sequence
- controlled variable
- variables
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 162
- 230000008569 process Effects 0.000 title claims abstract description 119
- 238000009776 industrial production Methods 0.000 title claims abstract description 56
- 230000004044 response Effects 0.000 claims description 81
- 238000012360 testing method Methods 0.000 claims description 45
- 238000012549 training Methods 0.000 claims description 27
- 238000005457 optimization Methods 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 16
- 230000004913 activation Effects 0.000 claims description 13
- 238000012795 verification Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 12
- 230000015654 memory Effects 0.000 claims description 12
- 239000002245 particle Substances 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 7
- 230000001413 cellular effect Effects 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 5
- 230000007787 long-term memory Effects 0.000 claims description 3
- 230000006403 short-term memory Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 9
- 230000007246 mechanism Effects 0.000 abstract description 3
- 238000004519 manufacturing process Methods 0.000 description 8
- 239000013643 reference control Substances 0.000 description 4
- 239000000126 substance Substances 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 239000004566 building material Substances 0.000 description 1
- -1 color Substances 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 239000002699 waste material Substances 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
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
Abstract
The application discloses an intelligent control method of a process industrial production system based on time sequence prediction, and relates to the technical field of process industry. Therefore, the integral control process can automatically learn and fit the characteristics of the nonlinear process, simultaneously process a plurality of controlled variables and fit the characteristics of the nonstationary process, the integral control process is stable and controllable, the problems of external disturbance, multiple constraints, complex mechanism modeling and the like in the industrial production process of the process can be solved, and a better control effect is achieved.
Description
Technical Field
The application relates to the technical field of process industry, in particular to an intelligent control method of a process industrial production system based on time sequence prediction.
Background
The process industry refers to a production process through physical change and chemical change, mainly comprises basic raw material industries such as petroleum, chemical industry, steel, color, building materials and the like, and most of the industries are the support and basic industries of national economy and are important supporting forces for continuous growth of the national economy.
The production efficiency can be effectively improved by reasonably controlling the manipulated variables of the industrial production system, and two main control methods for the manipulated variables exist at present: one is to rely on manual experience control to reach the preset production index, but the manual operation experience is uneven, which is easy to cause resource waste, the fluctuation of the production index is large, and the working condition is difficult to operate in the optimal state. Another common control method is PID control, which has a good effect on the control of most conventional industrial production systems.
However, the process industrial production system has the following characteristics compared with the traditional industrial production system: (1) Industrial production systems are more susceptible to external disturbances and to fluctuations in the process parameters of the previous stage. (2) There are more constraints in the control process of industrial production systems due to physical limitations and safety considerations of the equipment in the industrial production systems. (3) The industrial production system involves physical and chemical changes, the nonlinearity is complex and serious, the coupling is strong, and an accurate mechanism model is difficult to build. These characteristics result in undesirable control effects of the PID control on the industrial production system because the gain parameters of the conventional PID control are affected by the environment, and because the industrial production system has high complexity, strong nonlinearity, strong correlation and uncertainty compared with the conventional industrial production system, the controlled object of the PID is in a constantly changing environment, and the PID gain needs to be adjusted according to the change of the environment, so that the control difficulty is high. The PID control is still a linear combination of weighting in three aspects, so the problem of strong nonlinearity in an industrial production system cannot be well explained, and the PID control reduces and eliminates errors by error feedback, so that overshoot of the system behavior is often caused by too great initial control force.
Disclosure of Invention
Aiming at the problems and the technical requirements, the inventor provides an intelligent control method of a process industrial production system based on time sequence prediction, and the technical scheme of the application is as follows:
an intelligent control method of a process industrial production system based on time sequence prediction, the intelligent control method comprising:
for any one controlled variable in the process industrial production system, training to obtain a time sequence prediction model by using the controlled variable, the controlled variable corresponding to the controlled variable and the historical time sequence data of the disturbance variable corresponding to the controlled variable, wherein each controlled variable is influenced by the corresponding controlled variable and the disturbance variable, the time sequence prediction model is established based on a long-short-period memory network, and is used for outputting the predicted value of the controlled variable at the future time t+n according to the historical time sequence data of the controlled variable and the historical time sequence data of the controlled variable from the time t-n of the input history to the current time t, and t and n are integer parameters respectively;
for each controlled variable corresponding to any controlled variable, taking a predicted value of the controlled variable, which is output by a time sequence prediction model of the controlled variable, at a future time t+n as an optimization target, optimizing the value of each controlled variable at the current time t by using a quantum particle swarm optimization algorithm until the value of each controlled variable when the optimization target is achieved is obtained, and taking the value of each controlled variable as a reference controlled value of each controlled variable based on the controlled variable;
performing overall optimization on each manipulated variable in the process industrial production system based on the reference manipulated value of each controlled variable to obtain a target manipulated value of each manipulated variable in the process industrial production system;
each manipulated variable in the process industrial production system is controlled in accordance with the target manipulated value of each manipulated variable.
The further technical scheme is that the whole optimization of each manipulated variable in the process industrial production system based on the reference manipulated value of each controlled variable comprises the following steps of:
when the manipulated variable corresponds to and affects only one controlled variable, the reference manipulated value of the controlled variable affected by the manipulated variable is directly used as the target manipulated value of the manipulated variable;
when the manipulated variables correspond to and influence a plurality of controlled variables, weighting calculation is carried out on the manipulated variables based on the reference manipulated values of different controlled variables respectively, so as to obtain target manipulated values of the manipulated variables.
When the control variable is respectively weighted and calculated based on the reference control values of different controlled variables, the weighted weight of the control variable based on the reference control value of each controlled variable is related to the priority of the corresponding controlled variable, and the higher the priority of the controlled variable is, the larger the corresponding weighted weight is.
The further technical scheme is that the weighting calculation of the control variables based on the reference control values of different controlled variables respectively further comprises:
under the condition that the manipulated variables are based on the reference manipulated values of each controlled variable, determining the values of all process variables affecting the manipulated variables in the process industrial production system, and correcting the reference manipulated values of the manipulated variables based on the controlled variables according to a data correction strategy corresponding to the values of the process variables;
the manipulated variables are weighted based on corrected reference manipulated values of different controlled variables.
The further technical scheme is that training by using historical time sequence data to obtain a time sequence prediction model corresponding to any one controlled variable comprises the following steps:
acquiring a historical data set corresponding to a controlled variable, wherein the historical data set comprises a plurality of groups of historical time sequence data, and each group of historical time sequence data comprises the historical time sequence data of the controlled variable, a manipulated variable corresponding to the controlled variable and a disturbance variable corresponding to the controlled variable;
dividing the historical data set into a training set, a verification set and a step response test set;
in any round of model iteration, training the time sequence prediction model by using a training set, and verifying the prediction precision of the time sequence prediction model by using a verification set to calculate and obtain a precision index P val Verifying the process logic rationality of the time sequence prediction model by using the step response test set to obtain a step response index F sep Precision index P val The higher the prediction accuracy, the higher the step response index F sep The higher the process logic rationality is;
according to the accuracy index P of the previous model iteration val And a step response index F sep And determining model parameters of the time sequence prediction model in the next round of model iteration, and entering the next round of model iteration until the total iteration times are reached.
According to the further technical scheme, the method comprises the following steps of according to the accuracy index P of the current round of model iteration val And a step response index F sep Determining model parameters of the time sequence prediction model in the next round of model iteration comprises the following steps:
precision index P of previous model iteration val Reaching the precision index P of the previous model iteration val And the step response index F of the previous model iteration sep Reach the step response index F of the previous model iteration sep When the time sequence prediction model is used, the model parameters of the time sequence prediction model are directly saved for the next round of model iteration;
step response index F at the previous iteration of the model sep Reach the step response index F of the previous model iteration sep Precision index P of current round of model iteration val The precision index P of the previous model iteration is not reached val But the accuracy index P of two-round model iteration val When the difference value of the time sequence prediction model does not exceed the difference value threshold value, directly storing model parameters of the time sequence prediction model for the next round of model iteration;
otherwise, the model parameters of the time sequence prediction model are adjusted according to a preset strategy to serve as the model parameters of the time sequence prediction model in the next round of model iteration.
The further technical proposal is that a step response test set is utilized to verify the process logic rationality of a time sequence prediction model so as to obtain a step response index F sep Comprising:
initializing the number f of times the j-th operating variable corresponding to the controlled variable accords with the process logic on the step response test set j =0, j is an integer parameter with a start value of 1;
traversing any g-th group of historical time sequence data in the step response test set, and fixing the g-th group of historical time sequence dataThe values of the rest manipulated variables and disturbance variables except the jth manipulated variable are unchanged, the value of the jth manipulated variable is subjected to step adjustment in the value range of the jth manipulated variable, and when the change trend of the controlled variable output by the time sequence prediction model in the step adjustment process of the jth manipulated variable is detected to be consistent with the process logic, f is caused to be j ＝f j +1, otherwise hold f j The value is unchanged, g is an integer parameter with a starting value of 1;
if all the G groups of history time sequence data in the step response test set are not traversed, g=g+1 and the step of traversing any G groups of history time sequence data in the step response test set is executed again;
if all G groups of historical time sequence data in the step response test set are traversed, the frequency f of the jth manipulated variable conforming to the process logic on the step response test set is obtained j The method comprises the steps of carrying out a first treatment on the surface of the The j=j+1 is made and the step of initializing the number of times that any J-th manipulated variable corresponding to the controlled variable accords with the process logic on the step response test set is executed again until the number of times that all J manipulated variables corresponding to the controlled variable accord with the process logic on the step response test set is obtained, wherein J is an integer parameter;
obtaining a step response index F according to the times that all the manipulated variables corresponding to the controlled variables accord with the process logic on the step response test set sep 。
The further technical proposal is that a step response index F is obtained sep Comprising the following steps:
q initial variables are respectively applied to the time sequence prediction model, and the sum F of the times of the process logic conforming to the step response test set of all J manipulated variables corresponding to the controlled variables is determined on the basis of applying any Q initial variables to the time sequence prediction model q Determining a step response indicatorq is an integer parameter with a start value of 1.
The method comprises the further technical scheme that a time sequence prediction model corresponding to each controlled variable comprises two long-short-period memory networks and two full-connection layers which are sequentially arranged from input to output; in each long and short term memory network, the activation functions in the input gate, the forget gate and the output gate use PReLU, the activation functions of the hidden layer and the cell state use Tanh, and the cell state is added to the memory for calculating the input gate, the forget gate and the output gate.
The further technical proposal is that in each long-short-period memory network, the input gate i of the time step s s Forgetting door f s And an output gate o s The calculation formula of (2) is as follows:
wherein x is s Is the input layer of time step s, C s Is the cellular state of time step s, C s-1 Is the cellular state of time step s-1, h s Is the hidden layer of time step s, h s-1 Is a hidden layer of time step s-1 and has C s ＝f s ×C s-1 +i s ×Tanh(W C h s-1 +W C x s )，h s ＝o s ×Tanh(C s )；W i 、W f 、W o 、W C 、b i 、b f And b o Respectively, parameters of model learning.
The beneficial technical effects of the application are as follows:
the application discloses an intelligent control method of a process industrial production system based on time sequence prediction, which effectively predicts the value of a controlled variable in the future by establishing a time sequence prediction model to determine the value of the controlled variable through an optimizing algorithm, then the target control value of each controlled variable is obtained through overall consideration, the characteristics of a nonlinear process can be automatically learned and fitted, a plurality of controlled variables can be simultaneously processed, the characteristics of the nonstationary process can be fitted, the overall control process is stable and controllable, and the related improvement and promotion can be carried out later according to different practical application scenes, and the established time sequence model can automatically learn and eliminate noise, so that the sensitivity to the noise is lower, and a more stable prediction result can be provided. The intelligent control method can solve the problems of external disturbance, multiple constraint, complex mechanism modeling and the like in the process of industrial production, and can also make up for the defects of control in the process industrial field caused by the characteristics of the traditional PID.
Drawings
Fig. 1 is a method flow diagram of an intelligent control method according to an embodiment of the present application.
FIG. 2 is a logic diagram of a single long-short term memory network in any one of the timing prediction models in accordance with one embodiment of the present application.
FIG. 3 is a flow chart of a method of training a timing prediction model in one embodiment of the application.
FIG. 4 is a flow chart of a method of obtaining a step response indicator in one embodiment of the application.
Detailed Description
The following describes the embodiments of the present application further with reference to the drawings.
The application discloses an intelligent control method of a process industrial production system based on time sequence prediction, referring to a flow chart shown in fig. 1, the intelligent control method comprises the following steps:
and step 1, determining each controlled variable in the process industrial production system, and corresponding manipulated variables and disturbance variables of each controlled variable.
The controlled variables in a process industrial production system are some of the production operating parameters that need to be controlled in a process industrial production system, with different actual meanings in different process industrial production systems. The controlled variables are all or some of the critical production operating parameters in the process industry production system.
Each controlled variable is influenced by a corresponding manipulated variable and a disturbance variable, wherein the manipulated variable is a variable which can be manually adjusted to be valued in a process industrial production system and has a definite adjusting mode, and the disturbance variable is a variable which cannot be manually controlled in the process industrial production system and is mainly caused by external disturbance. Each controlled variable corresponds to one or more manipulated variables and to one or more disturbance variables. Likewise, in different process industrial production systems, the manipulated variables and the disturbance variables have different actual meanings.
The process industrial production system is provided with a plurality of controlled variables, each controlled variable is controlled by a corresponding operating variable and a corresponding disturbance variable, the operating variables corresponding to any two controlled variables are the same and different, and the disturbance variables corresponding to any two controlled variables are the same and different.
A process industrial production system contains a large number of manipulated variables, which for any one manipulated variable have three cases: (1) Only one controlled variable in the process industrial production system is affected, then the manipulated variable corresponds to only one controlled variable. (2) Influencing a plurality of controlled variables in the process industrial production system, the manipulated variable corresponds to the plurality of controlled variables. (3) Without affecting any controlled variable in the process industrial production system, the manipulated variable does not correspond to any controlled variable.
For example, in one embodiment, there is a controlled variable ctrl in the process industrial production system 1 、ctrl 2 And ctrl 3 Controlled variable ctrl 1 Corresponding manipulated variable M 1 Manipulated variable M 2 Disturbance variable D 1 Controlled variable ctrl 2 Corresponding manipulated variable M 1 Disturbance variable D 2 Controlled variable ctrl 3 Corresponding manipulated variable M 3 Disturbance variable D 3 . Then the controlled variable ctrl 1 Is influenced by two manipulated variables, while the controlled variable ctrl 2 And a controlled variable ctrl 3 Are influenced by only one manipulated variable, and the controlled variable ctrl 1 And a controlled variable ctrl 2 All subject to the same manipulated variable M 1 Is a function of (a) and (b).
And 2, training any one controlled variable in the process industrial production system by using historical time sequence data of the controlled variable, the controlled variable corresponding to the controlled variable and the disturbance variable corresponding to the controlled variable to obtain a time sequence prediction model.
The time sequence prediction model corresponding to any controlled variable is used for outputting the predicted value of the controlled variable at the future time t+n according to the historical time sequence data of the controlled variable and the historical time sequence data of the controlled variable from the time t-n of the input history to the current time t, and t and n are integer parameters respectively.
1. First, a network structure of a time sequence prediction model is constructed.
The time sequence prediction model corresponding to any controlled variable is established based on the long-short-period memory network, and in one embodiment, the time sequence prediction model corresponding to each controlled variable comprises two long-short-period memory networks and two full-connection layers which are sequentially arranged from input to output. In addition, improvements have been made to long-term memory networks, in each of which:
(1) The activation functions in the input gate, the forget gate, and the output gate use the PReLU to replace the conventional Sigmoid, with the following benefits:
the use of the PReLU as an activation function better solves for derivatives, while having no effect such as an exponential function in other complex activation functions, can make network computation faster.
The use of the PReLU as an activation function also prevents gradient explosions and vanishes when the value is too large or too small, the derivative of activation functions such as Sigmoid and Tanh being close to 0, while the PReLU is an unsaturated activation function, which does not exist.
The calculation formula of the activation function PReLU isThe update formula of the parameter a along with the training process of the model is +.>Mu is momentum, ζ is learning rate, ++>Ladder for representing parameter aDegree. Therefore, compared with the ReLU, the activation function PReLU introduces the parameter a to solve the problem that the input cannot be negative, and the value of the parameter a is not a fixed value, so that the parameter a can be updated along with the training process of the model, and the integral fitting effect of the model can be optimized.
(2) The activation function of the hidden layer and the cell state uses Tanh.
(3) The cell status is added to calculate the input gate, the forget gate and the output gate.
Based on the improved design of the three parts, as shown in fig. 2, in each long-short-period memory network, the input gate i of the time step s s Forgetting door f s And an output gate o s The calculation formula of (2) is as follows:
wherein x is s Is the input layer of time step s, C s Is the cellular state of time step s, C s-1 Is the cellular state of time step s-1, h s Is the hidden layer of time step s, h s-1 Is a hidden layer of time step s-1 and has C s ＝f s ×C s-1 +i s ×Tanh(W C h s-1 +W C x s )，h s ＝o s ×Tanh(C s )。W i 、W f 、W o 、W C 、b i 、b f And b o Respectively, parameters of model learning.
2. After the network structure of the time sequence prediction model is constructed, the model training is performed by using the controlled variable, the controlled variable corresponding to the manipulated variable and the historical time sequence data of the disturbance variable corresponding to the controlled variable, which comprises the following steps, please refer to the flow chart shown in fig. 3:
(1) And acquiring a historical data set corresponding to the controlled variable.
The historical data set comprises a plurality of groups of historical time sequence data, and each group of historical time sequence data comprises a controlled variable, a manipulated variable corresponding to the controlled variable and a disturbance variable corresponding to the controlled variable.
(2) The historical data set is divided into a training set, a verification set and a step response test set, and when the data set is divided, the data set can be divided according to the required proportion.
(3) In any round of model iteration, training the time sequence prediction model by using a training set, and verifying the prediction precision of the time sequence prediction model by using a verification set to calculate and obtain a precision index P val Verifying the process logic rationality of the time sequence prediction model by using the step response test set to obtain a step response index F sep . Precision index P val The higher the prediction accuracy, the higher the step response index F sep The higher the process logic rationality.
In a traditional model training scenario, a historical data set is generally divided into a training set and a verification set, then in each round of model iteration, model training is carried out on a time sequence prediction model by using the training set, and the prediction precision of the time sequence prediction model is verified by using the verification set until model parameters enabling the numerical value of a loss function on the verification set to be minimum are obtained. In the scene with the time sequence prediction result as the target, the method is generally free from problems, but in the scene with the control optimization as the target, in addition to the prediction precision of the time sequence model, whether the control logic relation between the manipulated variable and the controlled variable meets the process is more critical, if the method is only based on the mode of model iteration, a model parameter with very high prediction precision but completely not meeting the control logic is obtained with a certain probability, and even if the obtained model has high prediction precision, the optimal manipulated variable calculated by an optimization algorithm is definitely not meeting the actual control requirement, so that in any round of model iteration, the method not only utilizes a verification set to verify the prediction precision of the time sequence prediction model, but also utilizes a step response test set to verify the process logic rationality of the time sequence prediction model. By adding the verification test of the step response, the model parameters of the time sequence prediction model obtained after training can be ensured to accord with the process logic in the actual scene, and the actual application requirements are more met.
In theory, the historical time sequence data in the historical data set under the normal working condition can be used for verifying the step response to verify the process logic rationality of the time sequence prediction model, but the situation that the model pre-learning exists when the training set is directly used or the historical time sequence data in the verification set is considered, so that the application directly divides a single step response test set for verifying the process logic rationality in the step (2).
In any round of model iteration, the prediction precision of the time sequence prediction model is verified by utilizing a verification set to calculate and obtain a precision index P val The method is similar to the conventional method, and the application is not repeated, and the step response index F is obtained by verifying the process logic rationality of the time sequence prediction model by using a step response test set sep The method comprises the following steps, please refer to fig. 4:
(a) Initializing the number f of times the any jth manipulated variable corresponding to the controlled variable meets the process logic on the step response test set j =0, j is an integer parameter with a start value of 1.
(b) Traversing any g-th set of historical time sequence data in the step response test set, wherein the values of the rest manipulated variables and disturbance variables except the j-th manipulated variable in the g-th set of historical time sequence data are unchanged, and performing step adjustment on the value of the j-th manipulated variable in the value range of the j-th manipulated variable, wherein the step adjustment comprises step type increase or decrease of the value of the j-th manipulated variable.
And detecting the change trend of the controlled variable output by the time sequence prediction model in the process of carrying out step adjustment on the value of the j-th manipulated variable. When detecting that the change trend of the controlled variable output by the time sequence prediction model in the step adjustment process of the j-th manipulated variable accords with the process logic, enabling f j ＝f j +1, otherwise hold f j The value is unchanged, and g is an integer parameter with a starting value of 1.
(c) If all G sets of historical time series data in the step response test set are not traversed, let g=g+1 and execute the step of traversing any G th set of historical time series data in the step response test set again, i.e. execute (b) again.
(d) If all G group histories in the step response test set are traversedThe time sequence data is obtained, and the frequency f of the jth manipulated variable conforming to the process logic on the step response test set is obtained j 。
If the number of times that all J manipulated variables corresponding to the controlled variable meet the process logic on the step response test set is not obtained, j=j+1 is made and the step of initializing the number of times that any jth manipulated variable corresponding to the controlled variable meets the process logic on the step response test set is executed again, that is, the step of executing (a) again is executed, and the number of times that the next manipulated variable meets the process logic on the step response test set is determined.
If all J manipulated variables corresponding to the controlled variables are obtained to accord with the times of the process logic on the step response test set, ending the cycle and executing (e), wherein J is an integer parameter.
(e) Obtaining a step response index F according to the times that all the manipulated variables corresponding to the controlled variables accord with the process logic on the step response test set sep 。
In practical application, when the step response test is performed according to the method, an initial variable is required to be applied to the time sequence prediction model, in one embodiment, the step response test is performed according to the method on the basis of applying any q-th initial variable to the time sequence prediction model, and then the sum of the times that all J manipulated variables corresponding to the controlled variables conform to the process logic on the step response test set is determinedq is an integer parameter with a start value of 1. Then Q initial variables are respectively applied to the time sequence prediction model, and the process is repeated, so that F when each initial variable is applied can be obtained q Then determine the step response index +.>Q is an integer parameter.
(4) According to the accuracy index P of the previous model iteration val And a step response index F sep Determining model parameters of the time sequence prediction model in the next round of model iteration, and entering the next round of modelIterating until the total iteration times are reached.
In one embodiment, the precision index P at the current round of model iteration val Reaching the precision index P of the previous model iteration val And the step response index F of the previous model iteration sep Reach the step response index F of the previous model iteration sep And when the model parameters of the time sequence prediction model are directly saved for the next round of model iteration. Step response index F at the previous iteration of the model sep Reach the step response index F of the previous model iteration sep Precision index P of current round of model iteration val The precision index P of the previous model iteration is not reached val But the accuracy index P of two-round model iteration val When the difference value of the time sequence prediction model does not exceed the difference value threshold value, directly storing the model parameters of the time sequence prediction model for the next round of model iteration. Otherwise, the model parameters of the time sequence prediction model are adjusted according to a preset strategy to serve as the model parameters of the time sequence prediction model in the next round of model iteration.
By the method, a time sequence prediction model corresponding to the controlled variable can be finally obtained through training, and each controlled variable is trained according to the method, so that the time sequence prediction model corresponding to each controlled variable can be obtained through training, and as shown in fig. 1, taking the process industrial production system comprising H controlled variables as an example, the time sequence prediction models 1-H corresponding to the controlled variable 1-the controlled variable H are respectively obtained through training.
And 3, regarding each controlled variable corresponding to any controlled variable, taking a predicted value of the controlled variable, which is output by a time sequence prediction model of the controlled variable, at a future time t+n as an optimization target, optimizing the value of each controlled variable at the current time t by utilizing a quantum particle swarm optimization algorithm until the value of each controlled variable when the optimization target is achieved is obtained, and taking the value of each controlled variable as a reference controlled value of each controlled variable. In this step, the predicted value is considered to reach the corresponding target value when the error between the predicted value of the controlled variable at the future time t+n and the corresponding target value is smaller than the error threshold.
Because the disturbance variable is uncontrollable, the disturbance variable is kept unchanged, only the operation variable is optimized, when each operation variable is optimized, the process constraints such as a value range, an optimizing direction, an optimizing step length and the like are set for each operation variable according to the process condition to limit the optimizing boundary of the optimizing algorithm, and the optimizing efficiency is improved. The updating of particles in the particle swarm optimization algorithm is to obtain a new individual by observation, namely, a probability is given to observe the particles, then a position is obtained, a plurality of probabilities are randomly generated for each particle, the Monte Carlo idea is utilized for observation, a plurality of individuals are obtained, then the individual is selected to be optimal, the rest individuals are evaluated in sequence, and finally, the next generation of individuals are obtained, so that searching is performed. Compared with the traditional particle swarm optimization algorithm, the method improves the global searching capability by introducing the probability idea, and improves the defect that the traditional particle swarm optimization algorithm is easy to fall into an optimal solution.
And 4, integrally optimizing each control variable in the process industrial production system based on the reference control value of each controlled variable to obtain the target control value of each control variable in the process industrial production system.
As described in step 1 above, in a process industrial production system, some manipulated variables have an effect on a plurality of controlled variables at the same time, so that step 3, after optimizing the manipulated variables based on the time-series prediction model and the target values of the respective controlled variables, may result in the existence of one reference manipulated value of the manipulated variables based on the different controlled variables. For example, in the example of step 1, the controlled variable ctrl 1 Corresponding manipulated variable M 1 And manipulated variable M 2 Controlled variable ctrl 2 Corresponding manipulated variable M 1 Thus manipulating variable M 1 In step 3, a control variable ctrl is obtained 1 And a control variable ctrl 2 Resulting in multiple reference manipulated values for the same manipulated variable, it is difficult to give accurate control commands.
This step is therefore primarily intended to solve this problem, and is optimized as a whole such that each manipulated variable has only a unique target manipulated value. Including for any one of the manipulated variables in the process industry production system:
when the manipulated variable corresponds to and affects only one controlled variable, the reference manipulated value of the controlled variable based on the manipulated variable is directly used as the target manipulated value of the manipulated variable.
And when the manipulated variable corresponds to and affects a plurality of controlled variables, respectively carrying out weighted calculation on the manipulated variable based on the reference manipulated values of different controlled variables to obtain the target manipulated value of the manipulated variable. When weighting calculation is performed on the manipulated variables based on the reference manipulated values of the different controlled variables, respectively, the respective weighting weights may be custom-set in advance. Or in another embodiment, the weighted weight of the manipulated variable based on the reference manipulated value of each controlled variable is related to the priority of the corresponding controlled variable, and the higher the priority of the controlled variable is, the larger the corresponding weighted weight is, so that the obtained target manipulated value better meets the priority requirement of the controlled variable, and better control effect is achieved. The priority of each controlled variable is preset, and the specific value of the weighting weight corresponding to each controlled variable with different priorities is preset.
For example, in the above example, when the controlled variable ctrl 1 Priority of (c) is greater than the controlled variable ctrl 2 The manipulated variable M can be configured when the priority of (a) 1 Based on the controlled variable ctrl 1 The weighting of the reference manipulated value of (2) is 0.7, and the manipulated variable M 1 Based on the controlled variable ctrl 2 The weighting of the reference manipulation value of (2) is 0.3. Or more extreme examples, the manipulated variable M may be configured directly 1 Based on the controlled variable ctrl 1 The weighting of the reference manipulated value of (2) is 1, and the manipulated variable M 1 Based on the controlled variable ctrl 2 The weighting of the reference manipulated value of (2) is 0, so that the reference manipulated value of the manipulated variable based on the controlled variable with the highest priority is directly taken as the target manipulated value.
In another embodiment, certain manipulated variables can also be affected by some process variables in the process plant, which are intermediate quantities that the process plant produces during operation, with different actual meanings in different process plant systems. The method further comprises the step of correcting the data of the reference manipulated values of the manipulated variables based on the different controlled variables before the step of weighting the reference manipulated values of the manipulated variables based on the different controlled variables:
in the case where the manipulated variables are based on the reference manipulated values of each controlled variable, the values of the respective process variables affecting the manipulated variables in the process industrial production system at that time are determined, and then the reference manipulated values of the manipulated variables based on the controlled variables are corrected according to the data correction strategy corresponding to the values of the process variables. The correspondence between different values of the process variable and the data correction strategy is preset, for example, in one example, when the process variable reaches a certain set value, 2 is subtracted from the reference manipulated value of the manipulated variable based on the controlled variable, and the configuration of the data correction strategy is performed according to the actual service requirement.
After the reference manipulated values of the manipulated variables based on different controlled variables are corrected according to the method, the corrected reference manipulated values of the manipulated variables based on different controlled variables are weighted and calculated.
And step 5, controlling each manipulated variable in the process industrial production system according to the target manipulated value of each manipulated variable, so that each controlled variable tends to be stable and tends to the respective target value at the future time t+n, and the effect of overall control is achieved.
The above is only a preferred embodiment of the present application, and the present application is not limited to the above examples. It is to be understood that other modifications and variations which may be directly derived or contemplated by those skilled in the art without departing from the spirit and concepts of the present application are deemed to be included within the scope of the present application.
Claims (10)
1. An intelligent control method of a process industrial production system based on time sequence prediction is characterized by comprising the following steps:
for any one controlled variable in the process industrial production system, training to obtain a time sequence prediction model by utilizing the controlled variable, the controlled variable corresponding to the controlled variable and the historical time sequence data of the disturbance variable corresponding to the controlled variable, wherein each controlled variable is influenced by the corresponding controlled variable and the disturbance variable, the time sequence prediction model is built based on a long-short-period memory network, and the time sequence prediction model is used for outputting predicted values of the controlled variable at future t+n moments according to the input historical time sequence data of the controlled variable and the historical time sequence data of the controlled variable from t-n moment to current t moment, and t and n are integer parameters respectively;
for each manipulated variable corresponding to any one controlled variable, taking a predicted value of the controlled variable, which is output by a time sequence prediction model of the controlled variable, at a future time t+n as an optimization target, optimizing the value of each manipulated variable at the current time t by using a quantum particle swarm optimization algorithm until the value of each manipulated variable when the optimization target is achieved is obtained, and taking the value of each manipulated variable as a reference manipulated value of each manipulated variable based on the controlled variable;
performing overall optimization on each manipulated variable in the process industrial production system based on the reference manipulated value of each controlled variable to obtain a target manipulated value of each manipulated variable in the process industrial production system;
individual manipulated variables in the process industrial production system are controlled in accordance with target manipulated values of the individual manipulated variables.
2. The intelligent control method according to claim 1, wherein the overall optimization of each manipulated variable in the process industrial production system based on the reference manipulated value of each controlled variable comprises, for any one manipulated variable in the process industrial production system:
when the manipulated variable corresponds to and affects only one controlled variable, the reference manipulated value of the controlled variable affected by the manipulated variable is directly used as the target manipulated value of the manipulated variable;
and when the manipulated variable corresponds to and affects a plurality of controlled variables, respectively carrying out weighted calculation on the manipulated variable based on the reference manipulated values of different controlled variables to obtain the target manipulated value of the manipulated variable.
3. The intelligent control method according to claim 2, wherein when weighting calculation is performed on the manipulated variables based on the reference manipulated values of the different controlled variables, respectively, the weighted weight of the manipulated variables based on the reference manipulated value of each controlled variable is related to the priority of the corresponding controlled variable, and the higher the priority of the controlled variable is, the larger the corresponding weighted weight is.
4. The intelligent control method according to claim 2, wherein weighting calculation of the manipulated variables based on reference manipulated values of different controlled variables, respectively, further comprises:
under the condition that the manipulated variables are based on the reference manipulated values of each controlled variable, determining the values of the process variables affecting the manipulated variables in the process industrial production system, and correcting the reference manipulated values of the manipulated variables based on the controlled variables according to a data correction strategy corresponding to the values of the process variables;
and carrying out weighted calculation on the corrected reference manipulation values of the manipulated variables based on different controlled variables.
5. The intelligent control method according to claim 1, wherein training with historical time series data to obtain a time series prediction model corresponding to any one controlled variable comprises:
acquiring a historical data set corresponding to the controlled variable, wherein the historical data set comprises a plurality of groups of historical time sequence data, and each group of historical time sequence data comprises the historical time sequence data of the controlled variable, the controlled variable corresponding to the manipulated variable and the disturbance variable corresponding to the controlled variable;
dividing the historical data set into a training set, a verification set and a step response test set;
in any round of model iteration, a training set is utilized to predict a time sequence modelModel training is carried out, and the prediction precision of the time sequence prediction model is verified by utilizing a verification set so as to calculate and obtain a precision index P val Verifying the process logic rationality of the time sequence prediction model by using a step response test set to obtain a step response index F sep Precision index P val The higher the prediction accuracy, the higher the step response index F sep The higher the process logic rationality is;
according to the accuracy index P of the previous model iteration val And a step response index F sep And determining model parameters of the time sequence prediction model in the next round of model iteration, and entering the next round of model iteration until the total iteration times are reached.
6. The intelligent control method according to claim 5, wherein the precision index P according to the current model iteration is the same as the precision index P according to the current model iteration val And a step response index F sep Determining model parameters of the time sequence prediction model in the next round of model iteration comprises the following steps:
precision index P of previous model iteration val Reaching the precision index P of the previous model iteration val And the step response index F of the previous model iteration sep Reach the step response index F of the previous model iteration sep When the time sequence prediction model is used, directly storing model parameters of the time sequence prediction model for the next round of model iteration;
step response index F at the previous iteration of the model sep Reach the step response index F of the previous model iteration sep Precision index P of current round of model iteration val The precision index P of the previous model iteration is not reached val But the accuracy index P of two-round model iteration val When the difference value of the time sequence prediction model does not exceed the difference value threshold value, directly storing model parameters of the time sequence prediction model for the next round of model iteration;
otherwise, the model parameters of the time sequence prediction model are adjusted according to a preset strategy to serve as the model parameters of the time sequence prediction model in the next round of model iteration.
7. The intelligent control method according to claim 5, wherein the step response test set is used to verify the process logic of the time series prediction model to obtain a step response index F sep Comprising:
initializing the number of times f that any jth manipulated variable corresponding to the controlled variable accords with process logic on the step response test set j =0, j is an integer parameter with a start value of 1;
traversing any g-th set of historical time sequence data in the step response test set, fixing the values of the rest manipulated variables and disturbance variables except the j-th manipulated variable in the g-th set of historical time sequence data, carrying out step adjustment on the value of the j-th manipulated variable in the value range of the j-th manipulated variable, and when detecting that the change trend of a controlled variable output by the time sequence prediction model in the step adjustment process of the j-th manipulated variable accords with process logic, enabling f to be j ＝f j +1, otherwise hold f j The value is unchanged, g is an integer parameter with a starting value of 1;
if not, making g=g+1 and executing the step of traversing any G-th group history time sequence data in the step response test set again;
if all G groups of historical time sequence data in the step response test set are traversed, the frequency f of the j-th operating variable conforming to the process logic on the step response test set is obtained j The method comprises the steps of carrying out a first treatment on the surface of the The step of enabling j=j+1 and executing the number of times of initializing any J-th manipulated variable corresponding to the controlled variable to accord with the process logic on the step response test set again until the number of times of obtaining all J manipulated variables corresponding to the controlled variable to accord with the process logic on the step response test set is obtained, wherein J is an integer parameter;
obtaining the step response index F according to the times that all the manipulated variables corresponding to the controlled variables conform to the process logic on the step response test set sep 。
8. The intelligent control method according to claim 7, wherein the step response index F is obtained sep Comprising the following steps:
q initial variables are respectively applied to the time sequence prediction model, and the sum F of the times of the process logic conforming to the step response test set of all J manipulated variables corresponding to the controlled variables is determined on the basis of applying any Q initial variables to the time sequence prediction model q Determining the step response indexq is an integer parameter with a start value of 1.
9. The intelligent control method according to claim 1, wherein the time sequence prediction model corresponding to each controlled variable comprises two long-short-period memory networks and two full-connection layers which are sequentially arranged from input to output; in each long and short term memory network, the activation functions in the input gate, the forget gate and the output gate use PReLU, the activation functions of the hidden layer and the cell state use Tanh, and the cell state is added to the memory for calculating the input gate, the forget gate and the output gate.
10. The intelligent control method according to claim 9, wherein in each long-short-term memory network, the input gate i of the time step s s Forgetting door f s And an output gate o s The calculation formula of (2) is as follows:
wherein x is s Is the input layer of time step s, C s Is the cellular state of time step s, C s-1 Is the cellular state of time step s-1, h s Is the hidden layer of time step s, h s-1 Is a hidden layer of time step s-1 and has C s ＝f s ×C s-1 +i s ×Tanh(W C h s-1 +W C x s )，h s ＝o s ×Tanh(C s )；W i 、W f 、W o 、W C 、b i 、b f And b o Respectively, parameters of model learning.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310879391.8A CN116880191A (en) | 2023-07-17 | 2023-07-17 | Intelligent control method of process industrial production system based on time sequence prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310879391.8A CN116880191A (en) | 2023-07-17 | 2023-07-17 | Intelligent control method of process industrial production system based on time sequence prediction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116880191A true CN116880191A (en) | 2023-10-13 |
Family
ID=88264051
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310879391.8A Pending CN116880191A (en) | 2023-07-17 | 2023-07-17 | Intelligent control method of process industrial production system based on time sequence prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116880191A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117196544A (en) * | 2023-11-07 | 2023-12-08 | 恒实建设管理股份有限公司 | Intelligent management method and system for engineering information |
-
2023
- 2023-07-17 CN CN202310879391.8A patent/CN116880191A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117196544A (en) * | 2023-11-07 | 2023-12-08 | 恒实建设管理股份有限公司 | Intelligent management method and system for engineering information |
| CN117196544B (en) * | 2023-11-07 | 2024-01-30 | 恒实建设管理股份有限公司 | Intelligent management method and system for engineering information |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Event-triggered adaptive dynamic programming for non-zero-sum games of unknown nonlinear systems via generalized fuzzy hyperbolic models | |
| CN107193212B (en) | Aero-engine nonlinear predictive control method based on novel wolf optimization algorithm | |
| US8260441B2 (en) | Method for computer-supported control and/or regulation of a technical system | |
| CN108008627B (en) | Parallel optimization reinforcement learning self-adaptive PID control method | |
| Yang et al. | Control of nonaffine nonlinear discrete-time systems using reinforcement-learning-based linearly parameterized neural networks | |
| Lu et al. | A hybrid-adaptive dynamic programming approach for the model-free control of nonlinear switched systems | |
| JP2010514986A (en) | Method for computer-aided closed-loop control and / or open-loop control of technical systems, in particular gas turbines | |
| CN110471276B (en) | Apparatus for creating model functions for physical systems | |
| CN116880191A (en) | Intelligent control method of process industrial production system based on time sequence prediction | |
| Dominic et al. | An adaptive, advanced control strategy for KPI-based optimization of industrial processes | |
| Schwedersky et al. | Nonlinear model predictive control algorithm with iterative nonlinear prediction and linearization for long short-term memory network models | |
| Kosmatopoulos | Control of unknown nonlinear systems with efficient transient performance using concurrent exploitation and exploration | |
| CN112819224B (en) | Unit output prediction and confidence evaluation method based on deep learning fusion model | |
| Koumboulis et al. | A metaheuristic approach for controller design of multivariable processes | |
| Sethuramalingam et al. | PID controller tuning using soft computing methodologies for industrial process-A comparative approach | |
| CN111240201B (en) | Disturbance suppression control method | |
| CN115167102A (en) | Reinforced learning self-adaptive PID control method based on parallel dominant motion evaluation | |
| CN114995106A (en) | PID self-tuning method, device and equipment based on improved wavelet neural network | |
| Emaletdinova et al. | Algorithms of constructing a neural network model for a dynamic object of control and adjustment of PID controller parameters | |
| CN113743784A (en) | Production time sequence table intelligent generation method based on deep reinforcement learning | |
| CN114330119A (en) | Deep learning-based pumped storage unit adjusting system identification method | |
| Inanc et al. | Long short-term memory for improved transients in neural network adaptive control | |
| CN112379601A (en) | MFA control system design method based on industrial process | |
| Kudlačák et al. | Artificial neural network for adaptive PID controller | |
| Narayanan et al. | Optimal event-triggered control of nonlinear systems: a min-max approach |
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 |