Method for controlling a distributed control system
By combining the first principle steady-state model and distributed control system data of the chemical plant, the black box model is trained and iteratively optimized, the problem of difficult to accurately determine the control input in the chemical plant is solved, and more efficient conversion and reducing the deterioration effect is achieved.
Patent Information
- Application Number
- CN202380075549.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-11-25
- Filing Date
- 2023-11-21
- Publication Date
- 2025-06-06
AI Technical Summary
In distributed control systems in chemical plants, it is difficult to accurately determine the appropriate control input without performing a large number of experiments to achieve certain desired conditions such as maximizing conversion efficiency or minimizing degradation effects, such as scaling in heat exchangers.
The first principle-based steady-state model is used to combine with the data of the distributed control system to generate data points for training the black box model. Through the iterative optimization process, the mismatch between the black box model and the distributed control system is used to modify the objective function and variable constraints, and the optimal set point is gradually determined.
It is possible to more accurately determine the control input of the distributed control system without conducting a large number of experiments, improve the conversion efficiency of the chemical plant and reduce the deterioration effect, such as scaling in the heat exchanger.
Smart Images
Figure CN120112866A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for controlling a distributed control system, the distributed control system being configured for controlling a chemical plant. Background Art
[0002] In the operation of a chemical plant, which is then controlled by a corresponding interface represented as a distributed control system, it is usually necessary to either maximize positive effects or minimize negative effects. For example, it is usually desirable to maximize the conversion efficiency of some conversion processes, or minimize the degradation effects of individual components of a chemical plant, such as fouling in a heat exchanger, which may be caused by, for example, a suboptimal steam pressure distribution. This optimization usually needs to be real-time. For such a process for which optimization is sought, there is usually an analytical representation of the process, thereby revealing positive or negative effects. That is, the basic mechanism of the process is analytically known, and therefore it is known in principle which variable values on the local scale of the chemical plant will lead to better or worse results. However, since the chemical plant as a whole is a very complex and large-scale system, the variables in this local scale are usually not independently controlled. In other words, the dependence of local scale variables on the (control) input of the distributed control system of the chemical plant and the mutual dependence of these variables on each other are still complex, so that not all variables can be set to any value at the same time. Therefore, appropriate inputs to a chemical plant distributed control system suitable for optimizing a desired effect remain elusive but can be determined in real-time optimization (RTO). Summary of the invention
[0003] It is therefore an object of the present invention to provide a method for controlling a distributed control system which allows, without extensive experimentation, to determine more accurately (control) inputs to the distributed control system which are suitable for achieving some desired condition in a plant controlled by the distributed control system. Such desired condition may be related to specific material properties of a substance produced by the plant or to some other quantity associated with the plant operation, such as the amount and quality of waste generated by the plant operation, or a certain degradation rate of the plant itself due to the plant operation. Such inputs may vary over time due to uncontrolled influences (disturbances) on the plant.
[0004] With regard to a method for controlling a distributed control system, the object of the invention is achieved by a method for controlling a distributed control system, the object of which is to determine an optimal set point for a chemical plant controlled by the distributed control system, the method having the features of claim 1.
[0005] The present invention is based on the recognition that a steady-state model based on first principles of specific chemical or physical relationships describing the process can be combined with data provided by a distributed control system of a chemical plant (for a real chemical plant, or for a control system of a simulated plant in the form of an operator training simulator (OTS)) to obtain a solution to the optimization problem. Compared with conventional RTO schemes, the present invention does not directly apply a steady-state model, because the solution time of large-scale models is usually too high, and sometimes convergence to a solution cannot be guaranteed, which excludes the use of RTO. In contrast, a steady-state model is adopted to generate data points for training a black-box model. This is more advantageous than generating data points from a real chemical plant or OTS, because it is less time-consuming, and the production process and training activities are not interrupted. In addition, an objective function and constraints on variables reflecting physical or chemical boundary conditions are provided, thereby defining the optimization problem together with the black-box model. The optimization process is iterative, and in each iteration, the variables provided by the distributed control system, more precisely, the variables provided by the entity to which the distributed control system provides an interface, are used for the optimization problem, and thereafter the variables determined in the optimization (i.e., set points) are applied to the distributed control system. Since the black box model and the entity behind the distributed control system (i.e. the real plant or the simulated plant) are not identical, a mismatch in the variables between the two occurs at each iteration step. This mismatch is used not only to modify the objective function underlying the optimization, but also to modify the constraints defined on the (process) variables. On the other hand, the black box model is not adapted in the iteration.
[0006] The method according to the invention is a distributed control system, the purpose of which is to determine an optimal set point of a distributed control system, wherein the distributed control system is configured for controlling a chemical plant, wherein the chemical plant is configured for a chemical production process. The chemical production process may be any kind of chemical production process. In particular, the chemical production process may be a process for producing methylene diphenyl diisocyanate (MDI).
[0007] In the method according to the present invention, a training set of data points is generated, wherein the training set of data points generated includes applying at least one analytical function, which represents a steady-state model and links the variables of a chemical production process. In a steady-state model, all state variables (that is, all variables related to the system state) are assumed to be constants in time. Preferably, at least one analytical function is based on a corresponding first principle. Further, at least one analytical function can represent a corresponding strict first principle steady-state model. This first principle presents the basic equation describing a physical and / or chemical reaction or phenomenon. In other words, at least one analytical function is based on a first principle because it is not based on an analytical representation of a function of measured data fitted to, for example, a production process from a chemical plant. However, the analytical function can be an analytical function measured independently of the production process in, for example, a specific laboratory environment. For example, this applies to data points of specific substances from textbooks, databases, etc. The variables of a chemical production process can be any number, parameter, or setting associated with a chemical production process having a chemical, physical, or technical property.
[0008] At least one analysis function links variables because the variables consist of a domain or codomain of at least one shared function. In principle, there can be any number of analysis functions. Due to the complexity of chemical production processes, at least one analysis function may include 100 or more analysis functions, each of which partially represents a steady-state model and links the variables of the chemical production process. Therefore, all analysis functions together represent a steady-state model. In particular, it is possible that at least one analysis function may include at least 1,000 analysis functions, or even at least 100,000 analysis functions, each of which partially represents a steady-state model and links the variables of the chemical production process. At least one analysis function may be included in a commercial software product for simulating a steady-state process model, such as AVEVA Process Simulation (AVEVA Group) or Aspen Plus (AspenTech).
[0009] Preferably, at least one analysis function is also linked to a constant of the chemical production process. Such a constant may include a physical or chemical constant, such as Young's modulus or electrical resistance, Henry's law constant, or activation energy in the Arrhenius equation.
[0010] Further, in the method according to the present invention, the data point corresponds to the variable value of the linked variable. In other words, the data point represents a set of variable values of the linked variable that conforms to at least one analysis function. Therefore, the data point can be generated by at least one analysis function.
[0011] In the method according to the invention, a black box model is trained using a training set of data points. A black box model is a mathematical model implemented in software that has stimulus inputs and output responses. It is a black box model in the sense that the internal workings are not explainable and cannot be directly modified except by training. The behavior of the black box model is determined by training the black box model using training data, and the model is modified thereby. The training data is generated by solving a steady-state model comprising at least one analytical function, preferably a large number of analytical functions, i.e., the training data for the black box can be calculated data provided by commercial process simulator software.
[0012] In the method according to the present invention, an objective function is defined on optimization variables, which are at least partially composed of linked variables, and the objective function defines the optimal value of the optimization variables, wherein variable constraints applied to the optimization variables are defined. The objective function can also be referred to as a loss function or a cost function, and allows the degree of optimization achieved to be quantified. In other words, the purpose of optimization is to minimize or maximize the result of the objective function, depending on the definition of the objective function. The optimization variables are a subset of the linked variables. The optimization variables can also be the same as the linked variables. The variable constraints present the linked variables, and therefore also the boundary conditions that at least some of the optimization variables must comply with. In principle, the variable constraints can correspond to any kind of boundary conditions of the linked variables. In particular, variable constraints and boundary conditions are not limited to defining the value range of specific variables. They can also be more complex, and for example specify that the sum, product or other calculation results of a set of variables are within a value range. The interval of the value range can then depend on the variables. Other kinds of constraints can also be considered. The objective function together with the variable constraints can be understood as presenting the optimization problem together.
[0013] Further, in the method according to the present invention, the following iterative steps are repeated to implement input variables for controlling a distributed control system, the purpose of which is to determine an optimal set point: applying the input variables to the distributed control system, obtaining process values of at least some of the linked variables from the distributed control system, determining a mismatch between the obtained process values and the values of the linked variables based on a black box model, modifying the objective function based on the determined mismatch, modifying variable constraints based on the determined mismatch, and generating new input variables based on the modified objective function and the modified variable constraints.
[0014] A distributed control system is an interface for controlling a chemical plant. When input variables are applied, the distributed control system provides process values of at least some of the linked variables. It is possible that the process values of all linked variables are obtained from the distributed control system. Preferably, the process value is a dynamic value. In other words, for the process value, the state variable is considered to be variable in time, and therefore not constant. When input variables are applied, the distributed control system particularly provides the dynamic process value of the linked variable. Preferably, when the chemical production process has reached a steady state, the plant process variable is obtained from the distributed control system. The distributed control system can be composed of a real chemical plant or an operator training simulator, which is a computer-based simulation of the chemical plant. In other words, even if the distributed control system is an interface to a dynamic process, before obtaining the plant process variable, it is necessary to wait until the dynamic process reaches a steady state. In general, a distributed control system, whether as part of a real chemical plant or as part of an operator training simulator, is pre-existing and does not need to be developed specifically for optimizing the objective function (i.e., for solving the purpose of the present invention). The black box model only needs to represent the steady state, and is therefore not very computationally intensive. Therefore, black-box models can be applied for optimization and applied iteratively with pre-existing distributed control systems.
[0015] By comparing the obtained process values and the values of the linked variables, a mismatch between the obtained process values and the values of the linked variables according to the black box model can be determined. If process values are not obtained for all linked variables, the mismatch is only determined for those linked variables for which process values are actually obtained.
[0016] In principle, the iteration steps can be repeated any number of times. In particular, they can be repeated until the objective function crosses a predefined boundary, or until the difference of the objective function from one iteration to the next crosses a predefined boundary or is below a predefined threshold. New input variables can be generated by any optimization algorithm based on the modified objective function and the modified variable constraints. Preferably, the new input variables are obtained by an interior point method. Any other suitable method may also be used.
[0017] In a preferred embodiment, the black box model comprises a neural network. It has been found that such a neural network is a particularly useful black box model for representing non-linear relationships between variables.
[0018] In a further preferred embodiment, the distributed control system is represented by an operator training simulator. The operator training simulator also comprises a dynamic process model for the chemical plant, wherein the dynamic process model involves at least some of the linked variables, and wherein the values of at least some of the linked variables are obtained from the operator training simulator. The use of the operator training simulator makes it possible to perform the method according to the invention in a pure computing environment, without having to rely on an actual chemical plant. However, the results can be transferred to the chemical plant represented by the operator training simulator.
[0019] A further preferred embodiment is characterized in that the distributed control system is connected to a real chemical plant and that the values of at least some of the linked variables are obtained from the real chemical plant.The process values obtained from the real chemical plant represent more valid data than data from a computer simulation.
[0020] A further preferred embodiment is characterized in that modifying the objective function comprises adding a target gradient correction term to the objective function.
[0021] In a preferred embodiment, modifying the variable constraint includes adding a constraint bias term to the variable constraint. Preferably, modifying the variable constraint also includes adding a constraint gradient term to the variable constraint.
[0022] A further preferred embodiment is characterized in that the objective function is configured to minimize the steam pressure. This minimization may be related to any number of positions at which the steam pressure is minimized. The minimization can be based on any metric for taking into account the steam pressure at different positions in principle, for example by adding the steam pressure values at those positions. Preferably, the objective function is configured to minimize the steam pressure and thus minimize the fouling in the heat exchanger, in particular the fouling in the heat exchanger of a chemical plant. Therefore, the metric forming the basis of the minimization can be based on an equation that describes the fouling in the heat exchanger as a function of the steam pressure.
[0023] In a preferred embodiment, the objective function is configured to distribute the steam energy input, such as to balance the fouling. It is known that a higher steam energy input corresponds to a higher fouling rate. Therefore, when the current fouling at the predefined position set is equal, the steam energy input at these positions is also equal. However, if there is more fouling, and in particular severe fouling, at some positions in the predefined position set, the steam energy input at these positions with more or severe fouling is reduced so as to achieve a balanced resulting fouling at all positions in the predefined position set. Then, the remaining steam energy input can be evenly distributed on the remaining positions in the predefined position set. In this way, the time until excessive fouling needs to be resolved can be maximized. Preferably, the objective function is configured to distribute the steam energy input, such as to balance the fouling within the process stage of the chemical plant. Therefore, the predefined position sets for which fouling balance is sought are all located in the same process stage.
[0024] A preferred embodiment is characterized in that the linked variables (preferably optimized variables) include steam pressure variables, and in particular steam pressure variables of heat exchangers of a chemical plant. Such steam pressure variables may in particular include specific steam pressure values at specific locations. For example, these steam pressure variables may include steam pressure values within a specific heat exchanger of a chemical plant. Each location for which steam pressure is minimized may be at a corresponding heat exchanger of the chemical plant.
[0025] It is also possible that the linked variables, preferably the optimization variables, include heat transfer coefficients. In particular, these may be heat transfer coefficients of heat exchangers. These heat transfer coefficients may vary, in particular depending on the current fouling state of the respective heat exchanger.
[0026] In a further preferred embodiment, the linked variables include process temperature variables. Such process temperature variables may correspond to respective temperatures at respective locations within the chemical plant.
[0027] A preferred embodiment is characterized in that the linked variables comprise a heat duty variable, in particular a heat duty variable of a heat exchanger. Such a heat duty variable describes the amount of heat flow.
[0028] In a further preferred embodiment, the variable constraints include at least one of the following: a steam pressure limit, a constraint that steam can only be used for heating, and at least one limit on the total amount of heat transferred. The steam pressure limit defines the maximum steam pressure that must not be exceeded at one or more specific locations. The constraint that steam can only be used for heating means that steam cannot be used for cooling. The limit on the total amount of heat transferred defines the upper and lower limits of the total steam energy input at one or more specific locations.
[0029] In theory, the black box model can be adapted from one step of the iteration to the next. However, a further preferred embodiment is characterized in that the black box model remains constant through repeated iteration steps. This reduces the computational load of each iteration.
[0030] In a preferred embodiment, the linked variables are divided into input variables and output variables of the black box model. In the case of a neural network, it is further preferred that the neural network includes a separate neural network model for each output variable. This can lead to a higher overall accuracy with respect to the neural network part.
[0031] In principle, each iteration step, and therefore also the entire iteration, may require an arbitrarily long amount of computing time. A preferred embodiment is characterized in that the iteration steps are performed to obtain input variables for controlling a distributed control system with the aim of obtaining an optimal set point and preferably also for controlling a chemical plant in real time.
[0032] Thus, the computation speed of each iterative step is fast enough to allow real-time determination of the optimal set points for the distributed control system and / or chemical plant. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Further advantageous and preferred features are discussed in the following description with respect to the accompanying drawings. In the following it is shown: Figure 1 is a schematic representation of a chemical plant and its distributed control system on which the method according to the invention is carried out, Figure 2 on which the method according to the invention is respectively performed Figure 1 Schematic representation of a chemical plant and its distributed control system and an operator training simulator and its distributed control system, Figure 3 is a schematic representation of how the iterative steps of the method according to the invention are carried out, and Figure 4a , 4b is a schematic representation of variable constraints on linked variables according to the method of the invention before and after modification. DETAILED DESCRIPTION
[0034] exist Figure 1In FIG. 1 , a chemical plant 1 is shown which is used for the production of methylene diphenyl diisocyanate (MDI). A single process stage 17 of the chemical plant 1 is shown in more detail. In the simplified description of the process stage 17, the precursor MDA is fed in parallel to six towers 2a to 2f of the chemical plant 1 for phosgenation. Each tower is heated by a tower jacket 3a to 3f of the chemical plant 1, in which the steam is condensed for heat transfer. The respective outlet of each tower 2a to 2f is fed to a buffer tank 4 of the chemical plant 1, from which it is further fed to a distillation column 5 with a reboiler of the chemical plant 1, in which the steam is also condensed for heat transfer in the reboiler jacket 6 of the distillation column 5. The tower jackets 3a to f and the column jackets 6 present the heat exchangers 7a to 7g of the chemical plant 1. In order to minimize the fouling effect, it is desirable to minimize the steam pressure in these heat exchangers 7a to g. This leads to the question of what the corresponding temperature set points at the respective outlets of the columns 2a to 2f and at the distillation column 5 should be in order to achieve this minimized vapor pressure.
[0035] The relationship between the temperature set point and the steam pressure at steady state is called a set of equations, which correspond to a set of analytical functions that represent Figure 3 Steady-state model 8 is shown. The steam pressure value, the heat load variable, the heat transfer coefficient of the heat exchanger, and the temperature set point can be understood as presenting variables linked by the steady-state model 8. The applicable constants of these equations are also known, such as the thermal conductivity of the equipment involved (which is also linked by the steady-state model 8) and the various constraints on the variables linked by these equations. The minimization of the steam pressure value is formulated as an objective function, where the steam pressure value is the optimization variable. The heat load variable represents the output variable of the steady-state model 8. Based on the analytical function of the steady-state model 8, a training set 9 of data points is generated, and then the training set 9 of time points is used to train a neural network representing the black box model 10. After training, the black box model 10 remains unchanged.
[0036] The actual operation of the chemical plant 1 is controlled by the distributed control system 13 of the chemical plant 1, such as Figure 2 In addition to the chemical plant 1, there is also an operator training simulator 7, such as Figure 2As shown, it provides a simulation for the operation of a chemical plant and also includes a distributed control system 13 as an interface. In order to maximize its fidelity, the operator training simulator 7 is controlled by its distributed control system 13 in the same way as the real chemical plant 1. The above-mentioned temperature set points are input variables 16 of the distributed control system 13. In other words, the desired temperature set points can be input to the distributed control system 13 by the operator, and then the appropriate components of the real chemical plant 1 will be adjusted to achieve these desired temperature set points, or in the case of the operator training simulator 7, the operator training simulator 7 will simulate the actual chemical plant 1 so that the desired temperature set points are achieved. In addition, the distributed control systems 13 of both the chemical plant 1 and the online training simulator 7 will also provide a certain amount of process values 11 as output. These process values 11 also include at least some of the linked variables mentioned above, in particular the steam pressure and heat load variables.
[0037] exist Figure 3 In the following steps illustrated, the distributed control system 13 of the chemical plant 1 may always be used, or, alternatively, the distributed control system 13 of the operator training simulator 7 may always be used, even though both options continue to be referenced.
[0038] In a first step, an initial set of input variables 16 is applied to the distributed control system 13 of the chemical plant 1 or operator training simulator 7. Process values 11 are obtained from the distributed control system 13 once the chemical plant 1 or operator training simulator 7 has reached a steady state.
[0039] These process values 11 are now compared with the linked variables 14 obtained from the black box model 10, which are also based on the input variables 16 fed to the black box model 10. Since the relationships between the different quantities from the chemical plant 1 or the online training simulator 7 do not exactly match the relationships defined by the black box model 10, there will be a mismatch between the process values 11 and the corresponding values from the black box model 10. Based on this mismatch, both the objective function and the constraints are modified in a modification step 18 in the manner defined below.
[0040] In the publication "Iterative set-point optimization of batch chromatography", W. Gao, S. Engell; Computer & Chemical Engineering 29 (2005) pp. 1401-1409 [Gao 2005], iterative gradient modified optimization (IGMO) for iterative set-point optimization is introduced. It is proposed here to adapt not only the cost function but also the constraint function additionally by means of bias and gradient modification terms. Thus, the required plant gradient is estimated by using past measurements and finite differences. The method is applied to batch chromatography.
[0041] In the article "A reliable modifier-adaptation strategy for real-time optimization", W. Gao, S. Wenzel, S. Engell; Computer & Chemical Engineering 91 (2016) pp. 318-328 [Gao 2016], modifier adaptation with quadratic approximation is introduced as an extension of IGMO. It combines IGMO with quadratic approximation of gradient estimation and further concepts of derivative-free optimization (DFO), such as trust region and optimization based on quadratic surrogate models. The method is applied to two case studies taken from this paper.
[0042] In the review article, “Modifier Adaptation for Real-Time Optimization—Methods and Applications,” AGMarchetti, G. Francois, T. Faulwasser, D. Bonvin; Processes 2016, 4, 55 [Marchetti 2016], an analysis and proof of convergence and optimality of modifier adaptation methods are presented in general. It reviews reported variations of modifier adaptation, such as the use of higher-order modifiers, and discusses important properties and limitations. Finally, it highlights further research directions.
[0043] When the objective function to be minimized is given as , the modified objective function is defined in each iteration as follows: The index "ad" stands for adaptation. and The gradient of the objective function with respect to the input at iteration k is plotted. Herein, the subscripts m and p denote the available model and the true plant function, respectively. u is the optimized input, where u (k) is the last input applied to the system. In addition, an affine modification term is added to the nominal objective function J m (u, θ). This modification includes a bias correction term (i.e., ) and the gradient correction term (i.e., ) to compensate for the mismatch between the model and the plant function [Gao 2005]. Here and in the following, u is the temperature set point and θ is the other parameters.
[0044] Similarly, when the variable constraints are given as , the modified constraint is defined as follows: Here, the bias term is and the gradient term is The pilot plant gradients can be estimated by using finite differences or quadratic approximation and Finite differences and quadratic approximations can be generated based on measurements of the plant / OTS. Additional constraints can be added to enforce the trust region and restrict the solution of the optimization to be within the validity domain of the quadratic approximation. [Gao2016] Under certain assumptions and noise-free conditions, the modified term adaptation is guaranteed to match the optimal point of the real plant. Convergence can be guaranteed by similarity to the trust region method [Marchetti 2016].
[0045] Figure 4a The variable constraints 12a in the space of linked variables 14 that are initially set (ie set before the first iteration of the method steps) are graphically illustrated. Note that for illustration purposes only two dimensions are shown. Figure 4b The same variable constraints 12b according to the operator training simulator 7 are shown in the case where the operator training simulator 7 is used instead of the real chemical plant 1. Note that the respective areas represented by the variable constraints 12a, 12b in the graphical representation represent the space of linked variables 14 that are not allowed according to the respective variable constraints 12a, 12b. Therefore, the respective space of allowed values 15 corresponds to the space outside the space represented by each variable constraint 12a, 12b, i.e., the remaining space. Figure 4a and 4b Each contour line 20 in corresponds to a constant value of the objective function produced by the variables on that contour line. As can be seen, the space of allowed values 15 of the linked variables 14 is Figure 4a and Figure 4b As mentioned above, each modification of a constraint will result in Figure 4b, i.e., implemented by the operator training simulator 7 in a local "cut" region around the current set of input variables 16 by the modified constraints. The same situation occurs when a real chemical plant 1 is used instead of the operator training simulator 7, after necessary adjustments.
[0046] Using the modified objective function and the modified variable constraints, a new set of input variables 16 is implemented by solving the optimization problem consisting of the objective function and the constraints by applying the interior point method 19. The steps given above are repeated until the stopping criteria are met, such as the objective function converges to a specific value or the value of the objective function is lower than a predefined limit value.
Claims
1. A method for controlling a distributed control system (13), the distributed control system (13) being configured for controlling a chemical plant (1), wherein the chemical plant (1) is configured for a chemical production process, A training set (9) of data points is generated, wherein generating a training set (9) of data points comprises applying at least one analytical function representing a steady-state model (8) and linking variables (14) of the chemical production process, wherein the data points correspond to variable values of the linked variables (14), wherein the black box model (10) is trained using a training set (9) of data points, wherein an objective function is defined on optimization variables, the optimization variables at least partially consisting of the linked variables (14), the objective function defining the optimal value of the optimization variables, wherein variable constraints (12a, 12b) applied to linked variables (14) are defined, The following iterative steps are repeated to obtain input variables (16) for controlling the distributed control system (13): Applying the input variables (16) to the distributed control system (13), obtaining process values (11) of at least some of the linked variables (14) from a distributed control system (13), determining a mismatch between the obtained process value (11) and the value of the linked variable (14) according to the black box model (10) based on the same input variable (16), Modifying the objective function based on the determined mismatch, modifying the variable constraints based on the determined mismatch, New input parameters are generated based on the modified objective function and the modified variable constraints.
2. The method according to claim 1, It is characterized in that The black box model (10) includes a neural network.
3. The method according to claim 1 or 2, It is characterized in that The distributed control system (13) is represented by an operator training simulator (7), which includes a dynamic process model of the chemical plant (1), the dynamic process model involving at least some of the linked variables (14), and the values of at least some of the linked variables (14) are obtained from the operator training simulator (7).
4. The method according to claim 1 or 2, It is characterized in that The distributed control system (13) is connected to the real chemical plant (1), and the values of at least some of the linked variables (14) are obtained from the real chemical plant (1).
5. The method according to one of claims 1 to 4, It is characterized in that Modifying the objective function includes adding an objective gradient correction term to the objective function.
6. The method according to one of claims 1 to 5, It is characterized in that Modifying the variable constraints (12a, 12b) includes adding a constraint bias term to the variable constraints (12a, 12b), and modifying the variable constraints (12a, 12b) includes adding a constraint gradient term to the variable constraints (12a, 12b).
7. The method according to one of claims 1 to 6, It is characterized in that The objective function is configured to minimize the steam pressure, preferably thereby minimizing fouling in the heat exchanger.
8. The method according to one of claims 1 to 7, It is characterized in that The objective function is configured to distribute steam energy input, such as to balance fouling, preferably within a process stage (17) of a chemical plant (1).
9. The method according to one of claims 1 to 8, It is characterized in that The linked variables (14), preferably the optimized variables, include steam pressure variables, in particular steam pressure variables of heat exchangers (7a-7g) of the chemical plant (1), and further preferably, the linked variables (14) include heat transfer coefficients.
10. The method according to one of claims 1 to 9, It is characterized in that The linked variables (14) include a process temperature variable, and preferably, the at least one analysis function is also linked to a constant of the chemical production process.
11. The method according to one of claims 1 to 10, It is characterized in that The linked variables (14) include heat load variables, in particular heat load variables of the heat exchangers (7a-7g).
12. The method according to one of claims 1 to 11, It is characterized in that The variable constraints (12a, 12b) include at least one of: a steam pressure limit, a constraint that steam can only perform heating, and at least one limit on the total amount of heat transferred.
13. The method according to one of claims 1 to 12, It is characterized in that The black box model (10) is kept constant through repeated iteration steps.
14. The method according to one of claims 1 to 13, It is characterized in that The linked variables (14) are divided into input variables (16) and output variables of the black box model (10), and preferably the neural network includes a separate neural network model for each output variable.
15. The method according to one of claims 1 to 14, It is characterized in that The iteration steps are repeated to obtain input variables for real-time control of the distributed control system (13), and preferably also of the chemical plant (1).