DE10341762A1 - Handling the feasibility of restrictions and limits in an optimizer for process control systems - Google Patents

Abstract

An optimization technique for use in operating a process plant control device such as a model prediction control device uses an organized, systematic, but computationally simple method of loosening or redefining control, control and / or auxiliary variable restrictions if there is no feasible optimal solution within previously set limits, thereby developing an achievable solution for use by the controller. The optimization routine uses penalized margin variables and / or redefines the restriction model in conjunction with the use of penalty variables to develop a new objective function, and then uses the new objective function to determine a control solution that best adheres to the original restriction limits.

Description

This application is a partial continuation application to US patent application serial number 10 / 241,350 and entitled "Integrated Model Predictive Control and Optimization Within a Process Control System", filed September 11 2002 and the disclosure of which is expressly incorporated by reference herein.

This patent relates generally Process control systems and especially the handling of feasibility of restrictions and limits in an optimizer for process control systems when used, for example, with a model prediction controller to control a process plant.

Process control systems, for example distributed or scalable process control systems like those in chemical, petrochemical or other processes commonly used one or more process control devices, which are connected via analog, digital or combined analog / digital buses communicating with each other, with at least one host or operator workstation and with one or several plant facilities are coupled. The plant facilities, which for example valves, valve actuating mechanisms, switches and Encoders (e.g. temperature, pressure and flow sensors) can lead in the process Functions like opening or closing of valves and measuring process parameters. The process control device receives Signals which process measurements made by the plant equipment report, and / or other information related to the facility, uses this information to implement a control routine and then generates control signals which are sent to the facility equipment via the buses be sent to control the operation of the process. information of the plant facilities and the control facility are usually one or more applications run by the operator workstation available made to enable an operator to perform a desired function in terms of of the process, for example displaying the current status of the process, the changing the operational flow of the process, etc.

Process control devices are common programmed for various algorithms, subroutines or control loops (which all are control routines) for each of a number of different ones defined for a process or loops contained in a process, for example flow control loops, temperature control loops, Pressure control loops etc. Generally, each contains such a control loop one or more input blocks such as an analog input (AI) function block, a single output control block like one Proportional-integrating-differentiating (PID) or a fuzzy logic control function block, and a single output block such as an analog output (AO) function block. These control loops usually lead single input / single output control by the control block generates a single control output that is used to control a single one Process input such as a valve position etc. is used. In Certain cases but is the use of a number of independently operating single input / single output control loops not very effective because the controlled process variables by more be influenced as a single process input and everyone did Process input affect the state of many process outputs can. An example of this situation could be found in a Process with a container occur, which is filled by two input lines and is emptied through a single output line, with each line is controlled by another valve and at what temperature, Pressure and throughput of the container can be controlled so that they are at or close to desired Values. As indicated above, control of throughput, the temperature and pressure of the container by means of a separate one Throughput control loop, a separate temperature control loop and a separate pressure control loop. In this situation However, the operation of the temperature control loop can be changed by changing the Setting one of the input valves to control the temperature in the container cause the pressure in the container increases, which causes the pressure control loop, for example, that Open exhaust valve, to lower the pressure. This measure can then cause the throughput control loop to close one of the inlet valves, which is related to the temperature affects and the temperature control loop causes any other measure to take. In this example it is understood that the single input / single output control loops the process outputs (in this case, throughput, temperature and pressure) cause themselves behave in an unacceptable way where the outputs vibrate, without ever having a stationary To achieve state.

Model Predictive Control (MPC) or other types of sophisticated controls have been used to perform process control in situations where changes to a single controlled process variable affect more than one process variable or output. Many successful implementations of model prediction control have been reported since the late 1970s, and MPC is the primary form of a sophisticated one Become multi-variable control in the process industry. In addition, MPC has been implemented in distributed control systems as a multi-layer software for a distributed control system.

In general, MPC is a multi-input / multi-output control strategy, in which the impact of changing each one of a number of process inputs on each one Number of process outputs are measured and these measured responses are then used to create a control matrix or model of the process. The process model or the control matrix (which or which in usually the stationary Operation of the process defined) is inverted mathematically and then in or as a multi-input / multi-output controller used to make the changes made to the process inputs based process outputs to control. In some cases the process model is used as a process output response characteristic (usually a step response characteristic) for each of the process inputs shown, and these characteristics can be based on an each of the process inputs supplied series of, for example, pseudorandom jump changes be generated. These response characteristics can be used to Model the process in known ways. MPC is according to the state known in the art, and therefore their properties are described herein not described. MPC is generally described in Qin, S. Joe and Thomas A. Badgwell, "An Overview of Industrial Model Predictive Control Technology ", AKNE Conference, 1996. About it U.S. Patent Nos. 4,616,308 and 4,349,869 also describe generally MPC controllers used in a process control system can be.

MPC has been considered a very effective one and usable control method recognized and in connection with Process optimization used. To optimize a process that MPC uses, minimizes, or maximizes an or an optimizer several process input variables determined by the MPC routine, to make the process run at an optimal point. During this Computationally possible procedure is, it is to the process from an economic point of view From the point of view of optimization, it is necessary to select which process variables for example considerable Impact on the improvement of economic operations of the process (e.g. process throughput or quality). From a financial or economic point of view usually requires operating the process at an optimal point controlling many process variables in conjunction with each other, not just a single process variable.

Optimizing using quadratic Programming procedures or more common procedures such as procedures with within points has been proposed as a solution to dynamic Achieve optimization with MPC. In this process, a optimization solution determined, and the optimizer provides the controller with movements in the control device outputs (i.e. in the process control variables), whereby he process dynamics, current restrictions and optimization goals are taken into account. However, this approach involves an enormous amount of computing effort and cannot be practically implemented at the current technological level.

In most cases, when using MPC the number of control variables available in the process (i.e. of the control outputs the MPC routine) larger than the number of control variables of the process (i.e. the number of Process variables that need to be controlled in order to act on a particular Setpoint). So there are usually more degrees of freedom, those for optimizing and handling restrictions to disposal stand. Theoretically, process variables, restrictions, Limits and economic factors expressed an optimal working point values defining the process are calculated to such a Perform optimization. In many cases these process variables are restricted variables because they are with physical properties of the process to which they belong, related boundaries within which these variables must be kept. To the An example is a process variable that represents a tank level to which Area between that in the container physically maximum and minimum achievable level is limited. An optimization function can do that with any of the restricted Calculate variables or auxiliary variables linked costs and / or revenues, to work at a level that maximizes the proceeds the costs are minimized, etc. measurements of these auxiliary variables can then as inputs delivered to the MPC routine and by the MPC routine as control variables that have a setpoint equal to that of the optimization routine defines the working point for the auxiliary variable is.

US patent application serial number 10 / 241,350 and entitled "Integrated Model Predictive Control and Optimization Within a Process Control System", which was filed on September 11, 2002 and which is assigned to the right holder of this application and the disclosure of which is hereby expressly disclosed Reference, incorporated herein, discloses a method and apparatus for achieving online optimization with a sophisticated control block, such as an MPC block, in which the optimization is an integral part of the multi-variable model prediction control. This method is largely successful because, during normal operation, the MPC control device provides a future forecast of the process outputs up to a steady state, which creates the conditions necessary for the optimizer to function reliably. This optimization approach to finding However, a solution does not always work because in some situations some of the controller outputs or the predicted controller outputs (referred to herein as control variables) or some of the process outputs (referred to herein as controlled or control variables) are associated with each of the possible optimized solutions , lie outside of predefined limits or limits previously set for these variables and the solution therefore lies in a previously defined, unrealizable area. It is desirable, and in many control situations, that the optimizer, when integrated with the operation of the MPC or other control device, always find a solution and at the same time avoid to the best possible extent to work in an unrealizable area.

If an optimizer finds that there is no feasible optimal solution which all process control outputs or inputs within predetermined limits or limits, loosens the optimizer currently usually one or more restrictions or limits to one acceptable solution to find. This recovery process is commonly used priorities associated with each of the control and control variables to to determine which restriction limit to loosen first. Here's a simple approach the restrictions lowest priority drop to allow the optimizer to find a solution find what the restrictions higher priority comply. However, this procedure is not necessarily the most appropriate and way, with unrealizable solutions with restrictions deal because dropping lower priority restrictions results may have these restrictions more than push their limits, just a minimal adjustment to one or more restrictions higher priority to enable. Moreover it is usually necessary estimate how many restrictions have to be dropped a feasible solution to find. Naturally is it desirable the minimum number of restrictions drop that is required to comply with the restrictions higher priority to enable. To determine the appropriate number of restrictions to drop however, it may be necessary to determine on the available degrees of freedom offline calculations based on the system perform or to have the optimizer drop an iterative constraint and determine if the new reduced set of constraints a feasible solution exists until a feasible solution is found. Real time- or online optimization systems, offline calculations work unfortunately not good, and the iterative approach is usually unlimited and can therefore cause an unacceptable delay in finding a realizable one solution cause. In other words, the process of developing one optimal solution through successive dropping of restrictions may have to be repeated until a solution is found, and such a time-limited iterative Process is not in most real-time optimization applications desirable.

Several other people became known approaches to solve the optimization in the presence of unrealizable solutions. For example, Tyler, M.L. and Morari M., "Propositional Logic in Control and Monitoring Problems, "Proceedings of European Control Conference '97, pages 623-628, Brussels, Belgium, June 1997, using integer variables to weight to manage something, being the size of the violation by loosening a sequence of mixed integer optimization problems is minimized. Alternatively, Vada, J., Slupphaug, O. and Foss, B.A. discusses "Infeasibility Handling in Linear MPC subject to Prioritized Constraints ", Preprints IFAC '99, 14th World Congress, Beijing, China, July 1999, an algorithm that can be used to a series of linear programming (LP) problems or problems quadratic programming (QP) to solve the violations of restrictions, which cannot be adhered to. however are these approaches both computationally demanding and for fast real-time applications usually not sufficient or acceptable. Alternatively disclosed Vada, J., Slupphaug, 0. and Johansen, T.A., "Efficient Infeasibility Handling in Linear MPC subject to Prioritized Constraints, "ACC2002 Proceedings, Anchorage, Alaska, May 2002, an offline algorithm that calculates LP weights, which the calculated limit violations optimize. While This procedure takes the burden of excessive computing effort into one Moving offline environment, it requires dealing with an additional one Offline optimization problem and is for online or real time optimization procedures not very useful.

An integrated optimization process which is used to optimize a sophisticated control procedure how a model prediction control procedure can be used uses an organized, systematic but computationally simple process of loosening or redefining restrictions or limits for Control or control variables if there is no realizable optimal solution within the limits or limits previously set, to thereby achieve an achievable optimal solution for use in the Find tax procedure. In particular, the optimization routine uses if they with a limitation set within previously unrealizable optimizer solution is faced with penalized margin variables and / or redefining of the restriction model, robust and reliable Procedures to systematically choose the best solution. Because the method disclosed is computationally simple, this method is suitable for online implementation a real-time optimizer.

In one embodiment, the optimizer use an objective function, the penalized margin variable contains in the presence of one or more control or control variables, which about preset limits are pushed out, an optimal solution too Find. In another embodiment the optimizer can use the constraint model redefine to thereby accept the acceptable limits for or control variables that are outside of boundaries to redefine, and then one in the new constraint model use defined objective function with punitive variables to the outside limits with higher priorities without further violations of restrictions with restrictions lower priority to push against previously set limits. In a further embodiment can the optimizer the use of penalized margin variables with redefining the constraint model merge to create a flexible and efficient method of handling of restrictions to accomplish.

1 Figure 4 is a block diagram of a process control system including a control module with a sophisticated controller function block that integrates an optimizer with an MPC controller;

2 Fig. 3 is a block diagram of the advanced controller functional block from 1 , in which an optimizer and an MPC control device are integrated;

3 FIG. 10 is a flowchart outlining one way of creating and setting up the integrated optimizer and MPC controller function block 2 illustrated;

4 Fig. 3 is a flowchart showing the operation of the integrated optimizer and the MPC controller 2 illustrated during online process operation;

5 FIG. 10 is a graphical representation showing a manner of loosening a variable restriction or limit using margin variables in a target function of the optimizer 2 illustrated;

6 FIG. 3 is a graphical representation showing a manner of loosening areas linked to a target value by means of margin variables in the target function of the optimizer 2 illustrated;

7 FIG. 12 is a graphical representation showing a manner of loosening two sets of ranges associated with a setpoint using lower and higher penalized margin variables in the optimizer target function 2 illustrated;

8th Figure 11 is a graphical representation showing a way of redefining the constraint model or constraints for a variable and using penalty variables associated with the new constraint model in the optimizer's objective function 2 illustrated;

9 FIG. 12 is a graphical representation illustrating the use of both margin variables and redefining a constraint model to perform the optimization when one or more out-of-bounds variables exist; and

10 is a graphical representation illustrating the use of margin variables for optimizing when there is an ideal rest value of a control variable.

As in 1 shown includes a process control system 10 a process control device 11 which with a data logger 12 and with one or more host workstations or computers 13 (which can be any type of personal computers, workstations, etc.), each with a display screen 14 have communicative connection. The control device 11 is about input / output (I / O) cards 26 and 28 also with plant facilities 15 - 22 connected. The data logger 12 can be any desired type of data acquisition device having any desired type of memory and any desired or known software, hardware or firmware for storing data, and can be from the workstations 13 separated (as in 1 shown) or part of one of the workstations 13 his. The control device 11 , which can be the Delta VTM control device sold by Emerson Process Management, for example, via an Ethernet connection or any other desired communication network 29 with the host computers 13 and the data logger 12 communicatively connected. The communication network 29 can be a local area network (LAN), wide area network (WAN), telecommunications network, etc. implemented using wired or wireless technology. The control device 11 is with the plant equipment using any desired hardware and software and / or any intelligent communication protocol such as the FOUNDATION ^™ Fieldbus protocol (Fieldbus), the HART ^™ protocol, etc., such as associated with standard 4-20 mA devices 15-22 communicatively connected.

The plant facilities 15-22 can be any type of devices such as sensors, valves, encoders, actuators etc. while the I / O cards 26 and 28 any kind of to any can be compatible I / O devices in a desired communication or controller protocol. In the in 1 The embodiment shown are the plant facilities 15 - 18 Standard 4-20mA devices that use analog lines with the I / O card 26 communicate while the facility facilities 19 - 22 are intelligent devices such as Fieldbus system devices that communicate with the I / O card 28 by means of Fieldbus protocol communication via a digital bus. Of course, the facility facilities could 15-22 be compatible with any other desired standards or protocols, including any standards or protocols developed in the future.

The control device 11 which is one of many distributed control devices within the plant 10 and which contains at least one processor, implements or monitors one or more process control routines, which can contain control loops stored therein or otherwise linked to them. The control device 11 also communicates with the facilities 15 - 22 , the host computers 13 and the data logger 12 to control a process in any desired way. It should be noted that control routines or elements described herein may have parts which, if so desired, are implemented or executed by other types of control devices or other devices. Likewise, those described herein can be used in the process control system 10 control routines or elements to be implemented take any form including software, firmware, hardware, etc. For the purpose of this discussion, a process control element may be any part of a process control system including, for example, a routine, block, or module stored on any computer readable medium. Control routines, which can be modules or any part of a control procedure such as a subroutine, parts of a subroutine (such as lines of code), etc., can be in any desired software format, such as ladder logic, sequence function tables, block diagrams, object oriented programming or another software programming language or of another software development paradigm. Likewise, the control routines can be hard coded into, for example, one or more EPROMs, EEPROMs, application specific integrated circuits (ASICs) or any other hardware or firmware elements. Furthermore, the control routines can be developed using any development tools, including graphical development tools, or any other type of software, hardware, or firmware programming or development tool. The control device can therefore 11 configured to implement a control strategy or routine in any desired manner.

In one embodiment, the controller implements 11 using what is commonly referred to as functional blocks, a control strategy in which each functional block is part or object of an overall control routine and interacts with other functional blocks (via connections referred to as communication) to work in the process control system 10 Implement process control loops. Usually function blocks either have an input function such as that linked to a transmitter, a sensor or another process parameter measuring device, a control function such as that linked to a control routine that carries out PID, fuzzy logic control etc., or an output function that controls the operation of any device, such as a valve, in the process control system 10 perform any physical function through. Of course there are hybrid and other types of functional blocks. Function blocks can be in the control device 11 stored and executed by, which is usually the case when these functional blocks are used for or linked to standard 4-20 mA devices and some types of intelligent plant devices such as HART devices, or can be stored in the plant devices themselves and implemented by them themselves, which can be the case with Fieldbus devices. Although the description of the control system herein is done using a function block control strategy that uses an object-oriented programming paradigm, the control strategy or control loops or modules could also be implemented using other conventions such as ladder logic, sequence function tables, etc., or using any other desired programming language or any other desired programming paradigm can be implemented or developed.

Like through the spread block 30 in 1 can be seen, the control device 11 contain a number of single loop control routines, shown here as routines 32 and 34 , and implement one or more sophisticated control loops, shown here as a control loop 36 , Each such loop is commonly referred to as a control module. The single loop control routines shown 32 and 34 perform single-loop control using a single-input / single-output fuzzy logic control block or a single-input / single-output PID control block, which are connected to suitable analog input (AI) and analog output (AO) function blocks connected to Process control devices such as valves, with measuring devices such as temperature and pressure sensors or with any other device in the process control system 10 can be linked. The sophisticated control loop shown 36 contains a sophisticated control block 38 , whose inputs communicate with numerous AI function blocks and whose outputs communicate with numerous AO function blocks are tied, although the inputs and outputs of the sophisticated control block 38 can be communicatively connected to any other desired function blocks or controls to receive other types of inputs and provide other types of control outputs. As will be further described, the sophisticated control block 38 a control block that integrates a model prediction control routine with an optimizer to perform optimized control of the process or part of the process. Although the sophisticated control block 38 in the description given herein containing a model prediction control (MPC) block, the sophisticated control block can 38 any other multi-input / multi-output control routine or procedure, such as a neural network modeling or control routine, a multi-variable fuzzy logic control routine, etc. It is understood that the in 1 functional blocks shown including the advanced control block 38 from the control device 11 can be executed or that they can alternatively be in any other processing facility such as one of the workstations 13 or even in one of the plant facilities 19 - 22 located and can be executed by this.

As in 1 shown contains one of the workstations 13 a sophisticated control block generation routine 40 which is used to control the sophisticated control block 38 to generate, download and implement. During the generation routine for a sophisticated control block 40 in a memory in the workstation 13 stored and executed by a processor included therein, this routine (or any part thereof) may, if desired, additionally or alternatively, in any other facility in the process control system 10 saved and executed by this. Generally, the generation routine for a sophisticated control block contains 40 a control block generation routine 42 , which creates a sophisticated control block and incorporates this sophisticated control block into the process control system, a process modeling routine 44 which generates a process model for the process or part of the process based on data captured by the sophisticated control block, a control logic parameter generation routine 46 which generates control logic parameters for the advanced control block from the process model and which stores or downloads these control logic parameters for use in controlling the process in the advanced control block, and an optimizer routine 48 which creates an optimizer for use with the sophisticated control block. It is understood that the routines 42 . 44 . 46 and 48 may consist of a number of different routines, such as a first routine which creates a sophisticated control element with control inputs suitable for receiving process outputs and for providing control signals suitable for supplying control signals to process inputs, a second routine which enables a user to use the sophisticated control element in the Process Control Routine (which can be any desired configuration routine) to download and connect communicatively, a third routine that uses the sophisticated control to apply excitation waveforms to each of the process inputs, a fourth routine that uses the advanced control to respond to each one of the process outputs to capture data reflecting the excitation waveforms, a fifth routine that selects a set of inputs for the sophisticated control block or allows a user to do so The sixth routine that creates a process model, a seventh routine that develops sophisticated control logic parameters from the process model, an eighth routine that inserts the sophisticated control logic and, if necessary, the process model into the sophisticated control to enable the sophisticated control to control the process and a ninth routine which is an optimizer for use in the sophisticated control block 38 selects or allows a user to do so.

2 shows a more detailed block diagram of an embodiment of the with a process 50 communicatively coupled sophisticated control blocks 38 , it being understood that the sophisticated control block 38 produces a set of control variables that are supplied to other function blocks, which in turn are used with process control inputs 50 are connected. As in 2 shown, contains the sophisticated control block 38 an MPC controller block 52 , an optimizer 54 , a target conversion block 55 , a step response model or a control matrix 56 and an input processing / filter block 58 (input processing / filter block). The MPC controller 52 can be any standard M-by-M square (where M can be any number greater than or equal to one) MPC routine or procedure that has the same number of inputs as outputs. The MPC controller 52 receives as inputs a set of N control and auxiliary variables CV and AV (which are vectors of values) as in the process 50 measured, a set of disturbance variables DV, which are known or expected, at some point in the future on the process 50 changes or disturbances being applied, and a set of from the target conversion block 55 delivered stationary target value control and auxiliary variables CV _T and AV _T. The MPC controller 52 uses these inputs to generate the set of M variables MV (in the form of control signals) and provides the variable (MV) signals to control the process 50 ,

The MPC controller also calculates 52 a set of predicted stationary tax variable CV _SS and auxiliary variables AV _SS as well as a set of predicted stationary control variables MV _SS , which represent the predicted values of the control variables CV, the auxiliary variables AV and the control variables MV on the prediction horizon (for the CVs and AVs) and on the control horizon (for the MVs) , and delivers them to the input processing / filter block 58 , The input processing / filter block 58 processes the determined predicted steady-state values of the control, auxiliary and actuating variables CV _SS , AV _SS and MV _SS in order to reduce the effects of noise and unpredictable disturbances on these variables. It is understood that the input processing / filter block 58 may include a low pass filter or any other input _processing that reduces the effects of noise, modeling errors, and disturbances on these values and the filtered control, auxiliary and manipulated variables CV _SSfil , AV _SSfil and MV _SSfil to the optimizer 54 supplies.

The optimizer 54 in this example is an optimizer that works with linear programming (LP) and uses an objective function (OF) that is generated by a selection block 62 can be delivered to carry out the process optimization. Alternatively, the optimizer could 54 be an optimizer using quadratic programming, which is an optimizer with a linear model and a quadratic objective function. In general, the objective function OF defines costs or revenues associated with each of a number of control, auxiliary and actuating variables (which are generally referred to as process variables) and is determined by the optimizer 54 by finding a set of process variables that maximize or minimize the objective function, target values for these variables. The selection block 62 can the to the optimizer 54 delivered target function OF as one of a set of pre-stored target functions 64 choose which mathematically different types, the optimal operation of the process 50 to define, represent. For example, one of the pre-stored target functions 64 Configured to maximize system revenue, another of the target functions 64 configured to minimize the consumption of a particular resource that is in short supply while still another of the target functions 64 to maximize the quality of the process 50 manufactured product can be configured. In general, the objective function uses a cost or revenue value associated with each movement of a control, auxiliary and actuating variable in order to find the optimal process operating point within the set of acceptable points, such as the setpoints or ranges of the control variables CV and the limits of the auxiliary and Control variables AV and MV defined to determine. Of course, any desired objective function may be used in place of or in addition to that described herein, including objective functions that optimize each one to a certain extent from a number of aspects such as resource consumption, profitability, etc.

A user or operator can use one of the target functions 64 select by using an operator or user terminal (e.g. one of the workstations 13 out 1 ) an indication of the target function to be used 64 makes what choice about an input 66 to the selection block 62 is given. In response to the entrance 66 delivers the selection block 62 the selected target function OF to the optimizer 54 , Of course, the user or operator can change the objective function to be used during the operation of the process. If desired, a default objective function can be used in cases where the user does not specify or select an objective function. A possible target-objective function is discussed in more detail below. Although they are part of the sophisticated tax block 38 The various target functions can be shown in the operator terminal 13 out 1 and can be one of these objective functions during the creation or creation of the sophisticated control block 38 be delivered to them.

In addition to the target function OF, the optimizer receives 54 as inputs a set of control variable setpoints (which are usually setpoints set by the operator for the control variable CV of the process 50 and can be changed by the operator or another user) and an area associated with each of the control variables CV as well as an associated weight or an associated priority. The optimizer also receives 54 a set of ranges or limitation limits and a set of weights or priorities for the auxiliary variables AV and a set of limits for the control variables MV, which are used to control the process 50 be used. The optimizer 54 can also receive setpoints, preferred work areas, ideal idle values or other limits that are linked to one or more of the process variables. In general, the areas for the auxiliary and control variables define the limits (usually based on the physical properties of the plant) for the auxiliary and control variables, whereas the areas for the control variables provide an area in which the control variables can work for satisfactory control of the process , The weights for the control, auxiliary and actuating variables can determine the relative importance of the control variables, the auxiliary variables and the actuating variables with respect to one another during the optimization process and can in some circumstances be used to assist the optimizer 54 to enable a control target value solution to be created when some of the constraints or limits associated with these variables must be exceeded.

During operation, the optimizer 54 use a linear programming (LP) method to perform the optimization. As is well known, linear programming is mathematical method for solving a set of linear equations and inequalities that maximizes or minimizes a certain additional function called an objective function. As discussed above, the objective function can express an economic value such as cost or revenue, but instead of or in addition to economic objectives, it can express other objectives. In addition, as discussed above, the objective function can be tailored to include costs or penalties associated with breaking boundaries or boundaries to allow the objective function to be used to determine an acceptable solution, if there is no solution, all of them Keeps CVs, AVs and MVs within the preset limits.

It is understood that the matrix the stationary reinforcements the stationary reinforcement for any couple the control variables and the control or auxiliary variables. In other words, the stationary gain matrix defines the steady state gain in each control and auxiliary variable for a unit change in each of the control and disturbance variables. This matrix of the stationary reinforcements is usually an N-by-M matrix, where N is the number of tax and auxiliary variables and M is the number of control variables which be used in the optimizer routine. In general, N can be greater than be equal to or less than M, where N is usually greater than Damn.

The optimizer goes using any known or standard LP algorithm or method 54 usually iteratively to determine the set of target value control variables MV _T (as determined from the matrix of the stationary gains) which maximize or minimize the selected objective function OF, if possible achieving process operation that takes place within the Control variable (CV) setpoint limits, auxiliary variable (AV) restriction limits and actuating variable (MV) limits remain. In one embodiment, the optimizer determines 54 the change in the control variables and uses the indication of the predicted steady-state control variables, auxiliary variables and control variables CV _SSfil , AV _SSfil and MV _SSfil to determine the changes in process mode from its current mode, _i.e. during the process of reaching the target value or optimal process operating point to determine the dynamic operation of the MPC control routine. This dynamic operation is important because it is desirable to ensure that none of the restriction limits are crossed during the movement from the current working point to the target working point.

In one embodiment, the LP optimizer 54 be designed to minimize an objective function of the following form: Q = P t * A * delta MV + C t ΔMV (1)
in which:
Q = total cost / revenue;
P = revenue vector associated with AVs and CVs;
C = cost vector associated with MVs;
A = gain matrix; and
ΔMV = vector for calculated change in MVs.

The revenue values are generally positive and the cost values are generally negative numbers, so that their influence on the goal can be seen. The LP optimizer calculates using this or another objective function 54 the changes in the control variables MV which minimize the objective function and at the same time ensure that the control variables CV remain within a range around their target setpoint, that the auxiliary variables AV are within their upper and lower limit limits and that the control variables MV are within their upper limits - and lower limits.

vorhergesagte Änderungen in Ausgängen am Ende des Vorhersagehorizonts (t + p) angibt,

die m-mal-n-Verstärkungsmatrix des stabilen Zustands des Prozesses ist, und

Änderungen in Stellvariablen am Ende des Steuerhorizonts (t + c) angibt.In an optimization procedure that can be used, incremental values of actuating variables are used at the current time (t) and a sum of increments of actuating variables is used over the prediction horizon, with incremental values of control and auxiliary variables being determined at the end of the prediction horizon instead of as in LP applications common, current position values. Of course, the LP algorithm for this modification can be changed accordingly. In any case, the LP optimizer 54 use a stationary model and consequently a steady state is usually required for its application. With a prediction horizon, as is normally used when designing an MPC, a future steady state is guaranteed for a self-regulating process. A possible equation of the stable state of a predicted process for an m-by-n input-output process with prediction horizon p, control horizon c, expressed in incremental form is: ΔCV (t + p) = A * ΔMV (t + c) (2)
in which:

indicates predicted changes in outputs at the end of the forecast horizon (t + p),

is the m-by-n gain matrix of the stable state of the process, and

Indicates changes in control variables at the end of the control horizon (t + c).

The vector ΔMV (t + c) represents the sum of the changes made by each control device output mv _i over the control horizon, so that:

The changes should preferably meet the limits of both the MV control variable and the CV control variable (auxiliary variables are treated here as control variables), so that: MV min ≤ MV current + ΔMV (t + c) ≤ MV Max (4) CV min ≤ CV predicted + ΔCV (t + p) ≤ CV Max (5)

wobei:
UCV der Kostenvektor für eine Einheitsänderung im Steuervariablen- (CV-) Prozesswert ist; und
UMV der Kostenvektor für eine Einheitsänderung im Stellvariablen- (MV-) Prozesswert ist.In this case, the objective function to maximize the product value and minimize the raw material costs can be defined together as:

in which:
UCV is the cost vector for a unit change in the control variable (CV) process value; and
UMV is the cost vector for a unit change in the control variable (MV) process value.

Using equation (1) above, the objective function can be expressed in the form of control variables MV as:

To find an optimal solution, the LP algorithm calculates the objective function for an initial vertex in the range defined by equation (7) and improves the solution in every next step until the algorithm reaches the vertex with the maximum (or minimum) value of the objective function determined as the optimal solution. The determined optimal control variable values are then applied to or delivered to the control device as the target value control variables MV _T to be reached within the control horizon.

Generally, applying the LP algorithm to the prepared matrix gives three possible results. First, there is a clear solution for the MV _T target variable. Second, the solution is un limits what should not happen if each control and auxiliary variable has an upper and a lower limit. Third, there is no solution that adheres to the limits of the process variables, which means that the limits or limitations of the process variables are too narrow. To deal with the third case, the overall constraints can be relaxed and the optimizer can be run again with the relaxed constraints to get a solution. The basic assumption is that the limits of control variables (upper / lower limits) cannot be changed by the optimizer, although this assumption does not have to apply. The same assumption can also be used for restrictions or limits of the auxiliary variables (upper / lower limits). However, the optimizer can switch from driving the control variable CV against the specified setpoints (from the CV setpoint control) to driving the control variable against any of the values within a range around the setpoint (for the CV range control). In this case, the values of the control variables may lie within a range instead of an exact setpoint. If there are several auxiliary variables AV that exceed their restrictions and switching from CV setpoint control to CV range control does not provide a solution, it is also possible to relax or ignore the restrictions of the auxiliary variables based on the respective weights or priority information. In one embodiment, a solution could be found by minimizing the squared error of the auxiliary variables that allowed each to exceed their respective constraints, or by successively dropping constraints on the lowest priority auxiliary variables. As will be described in more detail below, other ways of dealing with out-of-bounds solutions include modifying the objective function to include margin variables or penalty variables that impose a cost or penalty for each process variable that exceeds a preset limit or limitation , and re-executing the optimizer with the new objective function and / or redefining the limits for one or more out-of-bound process variables, modifying the objective function to include penalties for those process variables within the new limits around those variables to push against the original limits, and re-run the optimizer with the new objective function to find the optimal solution within these newly defined limits. First, however, a method of initially selecting a target function will be described.

As mentioned above, the target function OF can be selected or by the program for generating control blocks 40 can be set as default. Although it is desirable to provide the ability to optimize, many situations may only require that control variable setpoints be maintained in a manner that still respects the operational constraints of the auxiliary and manipulated variables. For these applications, the block 38 configured so that it works exclusively as an MPC function block. In order to make this user-friendliness possible, a default "operating" target function with the various variables assigned predetermined costs in addition to the predetermined auxiliary variable (AV) weights can be generated automatically. These specifications can set all costs for the auxiliary variables AV and the actuating variables MV equally or create a different predetermined allocation of costs to the auxiliary and actuating variables AV and MV. If an option for experienced users is selected, the user can generate additional optimization selections and the associated costs for the various objective functions 64 define. The experienced user can also be allowed to change the auxiliary variable (AV) and control variable (CV) default weights of the default objective function, as well as the limit or setpoint violation costs, which are described in more detail below and which can be used by the optimizer if the optimizer cannot find a feasible solution initially.

In one embodiment, the objective function, if, for example, economic factors for process configuration are not defined, created automatically from the MPC configuration become. In general, the objective function can be done using the following formula to be created.

The variables C _j and p _j can be defined from the configuration settings. In particular, assuming that the control variable (CV) setpoint can only be defined at a lower limit (LL) or at an upper limit (HL), the value p _j can be defined in the following way:
p _j = -1 if the setpoint is defined at LL or "minimize" is selected; and
p _j = 1 if the setpoint is defined at HL or "Maximize" is selected.

Assuming that no configuration information has been entered for the auxiliary variables AV, p _j = 0 for all auxiliary variables AV. Accordingly, the value C _j for the control variables MV depends on whether the preferred control variable target value MV _{T is} defined or not. Where the preferred target variable value MV _{T is} defined:
C _j = 1 when MV _T is HL or "Maximize" is selected;
C _j = -1 when MV _T is LL or "minimize" is selected; and
C _j = 0 if MV _{T is} not defined.

If desired, the choice of using the optimizer 54 in connection with the MPC control device 52 be adjustable in order to create a degree of optimization. In order to carry out this function, that of the control device 52 change used in the control variables MV by applying different weights to those by the MPC control device 52 and the optimizer 54 certain changes in the control variables MV can be modified. Such a weighted combination of the control variables MV is referred to herein as an effective MV (MV _eff ). The effective MV _eff can be determined as: delta MV eff = ΔMV mpc (1 + α / S) + ΔMV opt (1 - α) 0 <α <1 (9) where S is chosen arbitrarily or heuristically. Usually S is greater than one and can be in the range of ten. Here the optimizer contributes to the effective output at α = 1, as was set during generation. If α = 0, the control device only delivers dynamic MPC control. Of course, the range between 0 and 1 provides different contributions from the optimizer and MPC control.

The default objective function described above can be used to set the optimizer's operation during various possible modes of operation thereof. In particular, if the number of control variables CVs matches the number of control variables MVs, the expected behavior in the default setting is that the control variable (CV) setpoints are maintained as long as the auxiliary variables AVs and the control variables MVs are expected to be within their limits , If it is predicted that an auxiliary variable or a control variable will exceed its limit, the operating setpoints of the control variables can be changed within their ranges to prevent these limits from being exceeded, if possible. If in this case the optimizer 54 cannot find a solution which, while keeping the control variables within their range, meets the auxiliary and control variable limits, the control variables can be kept within their range while allowing the auxiliary and / or control variables to deviate from their limitation limits. When finding the best solution, those auxiliary variables AVs and control variables MVs that are likely to cross a limit can be treated equally and their mean limit deviation can be minimized. To achieve this behavior, the target cost / revenue value used by the objective function can be automatically set to assign a 1 revenue to the CV control variables if the range is defined to allow a deviation below the setpoint, and proceeds of –1 if the range is defined to allow deviation from the target. The auxiliary variables AVs within the limits can be assigned a revenue of 0 and the control variables MVs can be assigned a cost value of 0. Alternatively, a revenue of 1 or -1 can be assigned to the auxiliary variables and a cost value of 0.1 to the control variables.

If the number of control variables CVs is smaller than the number of control variables MVs, the additional Degrees of freedom can be used to compare the (MV) end rest position to satisfy related requirements. Here you can the control variable setpoints (if any control variables- (CV) setpoints are defined) as long as the auxiliary and control variables are expected to be within their limits become. The mean deviation of the control variables from the configured one End-rest position is minimized. If for one or more of the auxiliary and control variables is predicted, that they cross their borders can be the operating setpoints of the control variables within their ranges changed to prevent these limits from being exceeded. Under this Condition can, if there are several solutions to control solution used be the solution which is the mean deviation of the control variables from the configured one End rest position minimized.

In normal situations it is up to the optimizer 54 Target setpoints developed in the manner described above within acceptable ranges and the actuating, auxiliary and control variables remain within the preset limits or ranges. However, if the perturbations are too severe to be compensated for within the constraint limits, some constraints will be exceeded, in which case the optimizer 54 may not be able to find a feasible solution (that is, a solution in which none of the restrictions or limits will be exceeded) using the setpoint loosening methods described above using the predetermined or selected objective function. Possible ways of optimizing under these circumstances include refraining from any optimization measure and dropping lower priority constraints.

While avoiding optimization is the easiest way to deal with an unrealizable solution, it is not the best way to handle constraints because if there is no viable solution within the limits, in most cases it is still possible to under consideration to minimize optimization violations. Furthermore, dropping lower priority constraints is an extreme measure that, while allowing compliance with higher priority constraints, can result in excessive uncontrolled violations of the boundaries of the dropped constraints, which may be unacceptable. Another problem with dropping lower priority constraints is to estimate how many lower priority constraints need to be dropped to get a solution. Such an estimate is based on the degrees of freedom available in the system, which estimate must be calculated before the optimization procedure or determined during the procedure. In the latter case, the process of developing an optimal solution by successively dropping constraints may need to be repeated until a solution is found, and such an indefinite iterative process is not desirable in most real-time optimization applications.

However, it was discovered that it still possible in some situations is an acceptable solution too find if the objective function is originally configured that no viable solution exists by, depending on the priority of the variable and the extent of the violation, in the objective function a penalty for each violation of the restriction assigned (which is usually referred to as using margin variables in the objective function will) by limiting limits for one or several process variables are redefined and the target function is redefined to the process variable using the pushing redefined borders against the previous border while others Process variables are not allowed to be their original restrictions or to cross borders or by a combination of these two methods.

In particular, they cannot be implemented Optimizer solutions handle using the well-known concept of margin variables. This procedure assigns penalties based on the priority of the violators becoming restriction or on the extent of border transgression or be based on both. Although common in the prior art it was suggested that it was possible is in the presence of violations of tax facility restriction limits The inventors have used the use of margin variables when optimizing no concrete revelation about known a way of doing this. So that was described below Processes developed by the inventors to adjust travel variables use an objective function used by an optimizer redefine if developed in the using the original objective function solution necessarily violations of restrictions available.

Generally, there are scope variables defined as the value or extent by which or which the predicted process variable the violated Limit (over or is below this). To minimize the margin variable (and thereby crossing the border to minimize), the objective function used by the optimizer redefined so they have a penalty for each margin variable value, that is not null contains which essentially prompts the optimizer to find a solution find which one in conjunction with the others from the objective function defined goals such as economic Aiming the scope variable minimized. According to this procedure, if the optimizer encounters a situation in which he just doesn't feasible solutions can deliver, redefine the objective function so that it has a Cost value or contains a penalty which one or which with a limit violation for each process variable (e.g. control, auxiliary or control variable) or at least for each process variable, which are likely to exceed a preset limit is linked is. If desired, can for each of the process variables the same or a different one Punishment can be defined so that the objective function is a margin variable contains which defines a cost value or a penalty, which one or which from every border crossing for every Process variable which is an associated Crossing the border, results. The margin variable cost value or penalty value in The newly defined objective function can depend both on the extent of the border crossing as well as the process variable in which the border crossing occurs, depend. Unit costs or penalties can be, for example, for process variable limits higher priority higher and for process variable limits lower priority be lower. In any case, the newly defined objective function used to minimize (or maximize) the newly defined Objective function in the presence of both of the previously defined economic ones Factors (proceeds and costs) as well as with the existence of margin variables, which are not zero, associated costs or penalties for one or more of the process variables to an optimal solution Find.

In particular, linear variable vectors S _max ≥ 0 and S _min ≥ 0 can be used in linear programming as follows: CV predicted + A * ΔMV (t + c) = CV Max + S min (10) CV predicted + A * ΔMV (t + c) = CV Max - p Max (11)

It is understood here that AVs in the CV terms are included in these equations (ie the Va Variable CVs cover all outputs where control variables are auxiliary variables). Equality is required in equations (10) and (11) for the linear programming model, and the range variables S _min and S _max are only used to support formal parameters of no particular importance in application. In this application, additional latitude variables S ⁺ and S ^- are used to extend the range boundaries of the margin vector, using S ⁺ ≥ 0 to increase the upper limit and S ^- ≥ 0 to lower the lower limit. Effectively, equations (10) and (11) can be redefined or reformulated as: CV predicted + A * ΔMV (t + c) = CV min + S min - p - (12) CV predicted + A * ΔMV (t + c) = CV Max - p Max + S + (13) where the values S ^- and S ^{+ are} the values by which the predicted process variable exceeds the lower and upper limits of the restriction. In this definition, the new latitude variables S ^- and S ⁺ are used to define penalty factors in the target function in order to obtain an LP solution within the areas or with only minimal exceeding of the areas.

und

erweitert werden, so dass:

wobei:
PS_– der Spielraumvariablen-Strafenvektor für das Übertreten von Untergrenzen ist;
PS₊ der Spielraumvariablen-Strafenvektor für das Übertreten von Obergrenzen ist; und
PS_– >> UCV und PS₊ >> UCV (wobei das Symbol >> "viel größer als" bedeutet).As an example, the objective function can be obtained by adding the terms

and

be expanded so that:

in which:
PS _{- is} the margin variable penalty vector for crossing lower bounds;
PS _{+ is} the margin variable penalty vector for violating ceilings; and
PS _- >> UCV and PS ₊ >> UCV (where the symbol >> means "much larger than").

The newly defined objective function can then be used in a standard manner so that by minimizing the newly defined objective function the values S ^- and S ^{+ are} minimized as defined by the costs associated with these values in order to obtain a solution which is optimal in the It makes sense that it optimally reduces the border crossings with regard to the other goals of the objective function.

In general, the penalties associated with the S ^- and S ⁺ scope variables should be significantly higher than the economic costs or revenues optimized in or linked to the objective function. So all components of the vectors PS. and PS + (which define the penalties associated with the various margin variables) are considerably greater than those of the economic cost / revenue vectors used in the objective function. It is reasonable to assume that the smallest component of the PS _- and PS ₊ vectors should generally be several times larger than the largest component of the UCV vector. Thus, the margin variables in the objective function, as described herein, should be punished with high penalties compared to the economic costs, proceeds, etc.

5 illustrates graphically the use of scope variables to deal with violation of a process variable (without associated setpoint). Illustrated in particular 5 the values for S _max (i), S _min (i) for the cases in which ( 1 ) the predicted process variable within the (by the lines 202 and 204 defined) predefined restriction limits, as indicated by the point 206 illustrated ( 2 ) the predicted process variable above the restriction upper limit 202 lies as through the point 208 illustrated, and ( 3 ) the predicted process variable below the restriction lower limit 204 lies as through the point 210 illustrated. In the first (with the point 206 connected) case, the values of S ⁺ (i) and S ^- (i) are zero because the upper limit and lower limit are not exceeded. In this case, no penalty is linked to or assigned to the process variable in the target function. In the second (through the point 208 illustrated case) but the value of S ⁺ (i) is not zero because of the point 208 the upper limit 202 exceeds. The bold dashed line illustrates the value of the margin variable S ⁺ (i), which is the value that is multiplied by a unit cost value associated with exceeding the upper limit for this process variable in the objective function. Accordingly, in (by the point 210 illustrated) third case the value of S ^- (i) not zero because of the point 210 the lower limit 204 exceeds. The bold dashed line here illustrates the value of the Margin variable S ^- (i), which is the value multiplied by a unit cost value associated with violating the lower limit for this process variable in the objective function. Illustrated of course 5 a single point in time associated with a signal process variable, and it is understood that the newly defined objective function minimizes the violation of restrictions over time up to the tax horizon.

In addition to minimizing (or instead of minimizing) violations of restriction limits as described above, the optimizer can be penalized margin variables use to setpoint optimization either in the case in which an unrealizable solution is found, or in the case where it is desired for other reasons, to optimize to a target value. In particular, with Penalized travel variables are used to, in response for an unrealizable solution or from any other desired Reason, the loosening of setpoints within a preselected setpoint (e.g. one supplied by the operator or any other source Setpoint) to allow defined acceptable ranges to others To enable process variables their associated Keeping boundaries or getting closer to them. The setpoint ranges could be one high range (within which the setpoint is allowed, within deviate from a range above the preselected setpoint), a low range (in which the setpoint is allowed, deviate within a range below the preselected setpoint) or include both. The setpoint ranges used in this procedure can be one-sided or be two-sided. Unilateral areas can be minimized or be linked to a maximization objective function which is used for the deviation outside of one preselected Setpoint range defines a penalty in the objective function, however which for a deviation from the setpoint within the setpoint range only economic costs imposed in the objective function. two-sided Areas on the other hand typically have not economic goals, but are used to find the optimal one solution as close as possible at the selected Setpoint within a preferred range, which is determined by use of travel variables outside of which are subject to high penalties of the area can be expanded. If the preferred If the setpoint range is zero, the situation is essentially the same like a setpoint with the margin variables that are subject to penalties around him.

The 6 respectively 7 illustrate the use of travel variables without and with the use of an extended range. In 6 illustrates a point 215 a setpoint for a process variable within a through the lines 217 and 219 defined area. With the deviation from the target value within the through the lines 217 and 219 limited range is not associated with a margin variable penalty. As for a point 221 is shown, however, if the process variable exceeds the upper range limit 217 a punished margin variable S ⁺ (i) is used to impose a penalty for this deviation in the objective function, similar to the use of margin variables in the constraint violation situations discussed above. Correspondingly, as for a point 223 shown when the process variable is the lower range limit 219 a punished margin variable S ^- (i) is used to impose a penalty for this deviation in the objective function, similar to the use of margin variables in the constraint violation situations described above.

7 illustrates the use of margin variables in an extended range situation in which a first margin variable penalty for deviations from the target within a first range is imposed and high penalty penalties (with high penalty costs) are used to cover the possible range expand the setpoint outside the first range. One point in particular illustrates 230 the situation when the predicted process variable is above a preselected setpoint 231 , but within a preset allowable, through the lines 232 and 234 defined range. Here a variable S (i) _is used _above to match the deviation of the point 230 from the preset setpoint 231 to define associated punishment. As indicated by the dashed line, the non-zero value of the variable S (i) at the _{top of} the objective function is punished. On the other hand, the point illustrates 236 the use of margin variables with high penalties if the actuating variable linked to the setpoint exceeds the preset upper range limit 232 by a value S ⁺ (i). Here, the particularly bold dashed line indicates that the variable S ⁺ (i) in the objective function has a high penalty compared to the variable S (i) _above , by deviations outside the range limit 232 to punish much higher than deviations from the target value 231 within the range limit 232 ,

In a similar way, a point illustrates 238 the situation when the predicted process variable is below the preselected setpoint 231 , but within a preset allowable, through the lines 232 and 234 defined range. Here a variable S (i) _{below is} used to match the deviation of the point 238 from the preselected setpoint 231 to define associated punishment. As indicated by the dashed line, the non-zero value of the variable S (i) _is punished in the target function _below . It also illustrates a point 240 the use of margin variables with high penalties, if the process variable linked to the setpoint is the preset lower range limit 234 by a value S ^- (i). Here, the particularly bold dashed line again indicates that the variable S ^- (i) in the objective function is punished with a high penalty compared to the variable S (i) _below , to reject checks outside the range limit 234 to punish much higher than deviations from the target value 231 within the range limit 234 ,

The equations for the setpoint control with the two-sided areas can be represented in the following form: CV predicted + A * ΔMV (t + c) = SP - S below + S above (15) CV predicted + A * ΔMV (t + c) = CV min + S min - p - (16) CV predicted + A * ΔMV (t + c) = CV Max - p Max + S + (17)

sollten zur Zielfunktion hinzuaddiert werden, wobei:

die Einheitsstrafe für die Lösung unter dem Sollwert ist; und

die Einheitsstrafe für die Lösung über dem Sollwert ist.Here S _below and S _above are vectors of travel variables for the solutions below and above the target values, and the terms

should be added to the objective function, where:

the unit penalty for the solution is below the target; and

the unit penalty for the solution is above the target value.

It is understood that the objective function also includes the margin variable penalties for the variables S ^- and S ⁺ as defined above.

This procedure for handling restrictions with scope variables that are subject to penalties considerable flexibility when handling unrealizable situations. In particular through The optimizer can apply penalties always an optimal solution Find the total cost of constraint violations as from the objective function defined minimized even if the solution is outside the pre-selected limits or restrictions lies. However, can some process variable outputs, which ones before running of the newly defined optimizer with the scope variables that are subject to penalties are within the limits due to the solution with margin variables to cross the borders. Moreover becomes the amount of border crossing for one certain process variables not defined quantitatively before generating the solution. These two characteristics are possible not desirable in many applications, because some applications may require the process variables lower priority, which initially are within the range limits, not pushed beyond the limits be higher to process variables priority to push the limits. moreover can some applications have well-defined solution limits for all process variables require. These two goals can be met by using the restriction model new in response to an unrealizable situation or solution is defined.

The restriction model itself can be redefined, in particular in order to handle non-realizations in a way that forces precisely defined degrees of border crossings. Of course, this redefinition takes place after the first optimization run if penalized margin variables are not used and there is no solution within original limits or if penalized margin variables are used but the solution with penalized margin variables is not acceptable. Generally, the newly defined limits are set as the value of the predicted process variable (e.g., a control variable CV) that crosses an original limit and the original limit minus any nominal value. These newly defined limits are then used in a second optimization run in order to find a solution that does not exceed the newly defined limits while minimizing or maximizing the target function. This redefinition of boundaries prevents the process variable that crosses an original boundary from deteriorating, and at the same time prevents other process variables that did not exceed their original preset limits from doing so in the second optimization run. However, the process variables are still within the original limits (if the predicted value of the process variable did not exceed the original limit) or within the newly defined limits (if the predicted value of the process variable was the original border was exceeded) to create a limited, optimized solution in the event of border violations.

This method can be mathematically defined for a control variable CV as follows: If there is no solution and an upper limit C ^{HL is} exceeded, new limits for the CV are defined as: CV HL ' = CV V orhersage CV LL ' = CV HL - Δ where Δ = 1-3% to avoid a solution exactly on the original limit.

Accordingly, when a lower limit CV "is exceeded, the new limits are defined as: CV LL ' = CV V orhersage CV HL ' = CV LL - Δ

Of course, appropriate limits for others Process variables such as control variables MV and auxiliary variables AV defined become.

These newly defined limits are common in 8th shown. In particular, the original CV ^HL and CV ^LL restrictions are on the left hand side of 8th through the lines 250 and 252 and are two CV predictions that exceed these constraints (and which would be generated by the optimizer's first pass using the original objective function) through the points 254 and 256 shown. Two sets of newly defined limits CV ^{HL '} and CV ^LL' are through the sets of lines 258 . 260 and 262 . 264 shown. It is understood that the set of newly defined limits 258 and 260 the new boundaries for the point 254 corresponds and by the point 254 on the high side and through the original border 250 minus a delta function is limited on the low side. The set of newly defined limits corresponds accordingly 262 and 264 the new boundaries for the point 256 and is on the high side (the more positive side) by the point 256 and on the low side (the less positive side) through the original boundary 252 plus a delta function limited. These new sets of limits will function as new limits for the points 254 or 256 (whichever it is) and the new objective function is not allowed to find a solution outside of these limits. You can also see the distance between the new border 260 (in the case of the point 254 ) or the new border 262 (in the case of the point 256 ) and the new CV value based high penalty factor and used in the new objective function to the new CV with the new limits against the new lower limit 260 (in the case of the point 254 ) or against the new upper limit 262 (in the case of the point 256 ) to drive. In particular after redefining the restriction handling model or the limits, the penalty vector can be recalculated for all CVs lying outside the ranges or limits in a way that drives these CVs in the direction of the exceeded limits. In order to achieve this goal, the penalty for crossing the border should go beyond economic criteria and should consequently be the penalties associated with the new borders with high penalties.

In this way, new frontiers for every Process variables (e.g. CV) for which predictions are made that they're outside the limits will be around a limited range for this Creating process variables becomes a goal function Punishment used the process variable in the second pass of the optimizer against the original Push the limit, thereby avoiding the process variables any limits (either redefined limits in the case of process variables outside of borders or more original Limits in the case of process variables which original Do not cross borders) to optimize. In this case, however, the limits (either the original boundaries for process variables, which are their original Limits were not exceeded in the first pass of the objective function or the newly defined limits for the process variables, which their original Limits exceeded in the first pass of the objective function) and the optimizer is able to find a solution within these limits to find.

A general form of an LP target function that uses newly defined limits and the penalty factors for these limits can be expressed as: (P T * A + C T ) * (MV t - MV t-1 ) (18) where the cost of the process outputs can be expressed by the following vector:

The resulting vector is:

In order to give priority to the restriction handling, an additional penalty v _i can be defined for the outputs with exceeded restrictions if the process variable (e.g. a CV) exceeds an upper limit as a negative value and if the process variable exceeds a lower limit as a positive value become. A contribution of the additional penalty v _i to the cost vector is:

wobei: r_i die Prioritäts-/Rangnummer der neu definierten CV ist; r_max die maximale Prioritäts-/Rangnummer für die niedrigste Priorität/den niedrigsten Rang ist; und r_min die minimale Prioritäts-/Rangnummer für die höchste Priorität/den höchsten Rang ist.So that the restriction handling takes precedence over economic factors, each component of this vector should be larger than the corresponding one of the vector CM ^T. Therefore:

where: r _{i is} the priority / rank number of the newly defined CV; r _{max is} the maximum priority / rank number for the lowest priority / lowest rank; and r _{min is} the minimum priority / rank number for the highest priority / rank.

Calculations can be simplified using the high estimate of | v _i | as:

For practical purposes, | a _ij | > 0.05 assumed to exclude extremely low process gains from the calculations. After calculating the penalty v for all process variables (e.g. CVs) that lie outside the preset limits or outside the preset ranges, the penalties can be set depending on the priority of the process variable as:

After calculating the costs for the individual process variables that exceed their respective restrictions according to equation (23), the total cost of all restrictions for all control variables can be calculated from equation (19). The penalty calculation procedure should be performed sequentially for all CVs that violate the constraints by applying Equation (23) starting with the lowest priority constraint violated (greatest r _i ) and working towards the highest priority constraint violated. For practical purposes, the effective penalty vector can be normalized. One possible normalization method is to divide all vector components by a maximum component and multiply them by 100.

If desired, the concept of constraint handling can be extended to a situation that integrates the two approaches described above, and in particular by combining the use of penalized margin variables and redefining the model. If there are no violations of the restrictions or if the opti If the solution is acceptable, only the margin variables with penalties are used in this integrated approach. However, if limits are violated and the solution with penalized margin variables is not acceptable, the process exit limits are redefined. In this case, the new output limits will be the same as the predictions for the predicted process outputs that cross the limits as in 8th shown. However, the original limit is still used to define penalties for travel variables, as in the Travel Variables application described earlier.

9 illustrates this integrated use of margin variables and redefinition of boundaries. One point in particular illustrates 270 a predicted process variable or CV within the through the lines 272 and 274 shown initial limits CV ^HL and CV ^LL . One point 276 illustrates a predicted process variable or CV, which is the upper limit CV ^HL 272 transgresses. In this case, the limits are redefined as CV ^{HL '} and CV ^LL' , as above regarding the redefinition of the limits in 8th generally discussed. In addition, margin variables S ' _max and S ⁺ (i) are used to impose a penalty for violating the original limit by means of margin variable penalties as discussed above. In a similar way, a point illustrates 280 a predicted process variable or CV, which is the lower limit CV ^LL 274 transgresses. In this case, the limits are redefined as CV ^{HL '} and CV ^LL' , as above regarding the redefinition of the limits in 8th generally discussed. In addition, margin variables S ' _min and S ^- (i) are used to impose a penalty for violating the original limit using margin variable penalties as discussed above.

This method can be expressed using the following equations: CV predicted + A * ΔMV (t + c) = CV LL + S min - p - (25) CV predicted + A * ΔMV (t + c) = CV HL - p Max + S + (26) CV predicted + A * ΔMV (t + c) = CV LL ' + S ' min (27) CV predicted + A * ΔMV (t + c) = CV HL ' - S ' Max (28)

Here are the newly defined limit values CV ^{LL '} and CV ^HL' upper and lower limits, which the optimizer cannot exceed. The limits are set as out of bounds CV predictions or out of bounds values with a wider range than the original bounds. Integrated equations for CV range control and bilateral CV range control can be developed in a similar way. In addition, integrated equations for unilateral range control are identical to equations (25) - (28). Two-sided range control can be implemented using the following equation: CV predicted + A * ΔMV (t + c) - SP - S below + S above (29) which is identical to equation (15).

It is understood that this integrated approach allows restriction handling to be accomplished in a more flexible way. At present, entrances have only hard or fixed limits. Using the same approach, it is possible to define soft constraints contained in hard constraints for some of the inputs. Introducing penalized margin variables for the soft restrictions makes it easy to define a penalized area for the MV. Equations for these soft areas can be expressed as:

Finally, the same approach can be used to get the optimizer to become a pro drive variable such as a control variable MV against a preferred input value or the so-called "ideal idle value". This approach is in 9 shown in which the ideal rest value through the line 290 is illustrated and the upper and lower limits of the MV by the lines 292 respectively 294 are illustrated. The points 296 and 298 illustrate the situation when the predicted value is above or below the ideal resting value 290 lies. In these cases, the penalty variables S (i) _above and S (i) _below are used to define the cost or penalty value associated with the deviation from the ideal rest value in the objective function. Equations for an MV with the ideal rest value can be expressed as: IRV - S below + S above = MV current + ΔMV (t + c) (32) MV min + S min = MV current + ΔMV (t + c) (33) MV Max - p Max = MV current + ΔMV (t + c) (34)

It is understood that the components of the penalty vectors S ^- , S ⁺ , S _below , S _above e.g. the unit cost value or penalty value, in the objective function as a function of the variable priorities, variable unit costs and / or other factors such as the degree border crossing can be set or selected.

In tests using this Proceedings were violated achieved by high-level interference were applied or by lower and upper limits of the inputs and outputs were set in a way that caused these limits to be exceeded were. These tests confirmed the Usability of the procedures. Worked particularly during these tests the optimizer flawlessly and effectively, being the violation of restrictions high priority without further violation of restrictions lower priority improved. Therefore, the handling of unrealizable LP solutions by application of travel variables and redefinition of the model inherently extreme flexible and effective approach.

In addition, the constraint handling principles discussed above can be used to develop a number of changes to the constraint models to meet more specific requirements. Likewise, the approach can be used for other real-time optimization applications without using MPC control, such as optimizing a gasoline blending process. Moreover, the optimizer can 54 use the methods described above to implement ideal rest values in situations in which one or more realizable solutions are possible, range control. So the optimizer can 54 use these methods when either using an objective function in situations where a feasible solution cannot be found, or when using an objective function in situations where one or more feasible solutions are possible, but it is desired to be ideal Tried to optimize the rest value, within a setpoint range etc., to optimize.

How again 2 shows, the optimizer delivers 54 after determining a solution for the set of control variable target values, the set of optimal or target value control variables MV _T to the target value conversion block 55 , which uses the matrix of the stationary gains to determine the stationary target value control and control variables which result from the target value control variables MV _T. This conversion is computationally simple, since the matrix of the stationary gains defines the relationships between the control variables and the control and auxiliary variables and can therefore be used to calculate the target value control and auxiliary variables CV _T and AV _T from the defined (stationary) target value -Uniquely determine actuating variables MV _T.

After they are determined, at least a subset of N of the target control and auxiliary variables CV _T and AV _{T are} input to the MPC controller 52 which, as mentioned above, uses these target values CV _T and AV _T to determine a new set of stationary control variables (over the control horizon) MV _SS which the current control and control variables CV and AV against the target values CV _T and AV _T drives at the end of the tax horizon. Of course, the MPC control device changes the control variables when trying to reach the stationary values for these variables MV _SS , which theoretically are those of the optimizer 54 certain target value control variables MV _T will be known, gradually. Because the optimizer 54 and the MPC controller 52 during each process query, as described above, the target values of the control variables MV _T can change from query to query and consequently the MPC control device can never really reach a specific one of these sets of target value control variables MV _T , especially not when there is noise, unexpected disturbances, changes in the process 50 etc. However, the optimization works 54 always on the control device 52 so that it moves the control variables MV against an optimal solution.

As is known, the MPC control device contains 52 a control prediction process model 70 , which can be an N-by-M + D step response matrix (where N is the number of control variables CV plus the number of auxiliary variables AV, M is the number of control variables MV and D is the number of disturbance variables DV). The control prediction process model 70 produced at an exit 72 a previously calculated prediction for each of the control and auxiliary variables CV and AV, and a vector summer 74 subtracts these predicted values for the current point in time from the measured actual values of the control and auxiliary variables CV and AV um at the input 76 to produce an error or correction vector.

The control prediction process model 70 then uses the N-by-M + D step response matrix to calculate CV and AV for each of the control variables and auxiliary variables across the control horizon based on the other inputs of the control prediction process model 70 delivered fault and control variables to predict a future control parameter. The control prediction process model 70 also provides the predicted steady-state values of the control variables and the auxiliary variables CV _SS and AV _SS to the input processing / filter block 58 ,

A control target block 80 determined using one for the block 38 previously set up trajectory filter 82 a control target vector for each of those from the target conversion block 55 N target value control and auxiliary variables CV _T and AV _T supplied to it. In particular, the trajectory filter provides a unit vector which defines the manner in which control and auxiliary variables are to be driven against their target values over time. The control target value block 80 uses this unit vector and the target value variables CV _T and AV _T to produce a vector of the dynamic control target values for each of the control and auxiliary variables, which changes the changes in the target value variables CV _T and AV _T over that of the control horizon time defined period of time defined. A vector summer 84 Subtract the future control parameter vector from the dynamic control vectors for each of the control and auxiliary variables CV and AV to define an error vector for each of the control and auxiliary variables CV and AV. The future error vector for each of the CV and AV control and auxiliary variables is then provided to the MPC algorithm, which operates to select the variable (MV) steps that, for example, minimize the smallest squared error across the control horizon , Of course, the MPC algorithm or the MPC controller uses an M-by-M process model or an M-by-M control matrix, which is based on the relationships between those in the MPC controller 52 entered N control and auxiliary variables and those from the MPC control device 52 output M control variables was developed.

In particular, the pursues with the Optimizer working MPC algorithm two main goals. First tried the MPC algorithm, the CV control error with minimal MV movements within the operating restrictions to minimize, and secondly, he tries to create the optimizer optimal stationary MV values and the CV target values calculated directly from the optimal steady-state MV values to reach.

wobei:
CV(k) der Vektor der "p-Schritt-vorwärts"-Vorhersagen der gesteuerten Ausgänge ist;
R(k) der Vektor der "p-Schritt-vorwärts"-Bezugstrajektorie (des "p-Schritt-vorwärts"-Sollwerts) ist;
ΔMV(k) der Vektor der inkrementalen "c-Schritt-vorwärts"-Steuerbewegungen ist;

eine Matrix der Strafen auf die Fehler der gesteuerten Ausgänge ist;

eine Matrix der Strafen auf die Steuerbewegungen ist;
p der Vorhersagehorizont (Anzahl von Schritten) ist;
c der Steuerhorizont (Anzahl von Schritten) ist; und
Γ⁰ eine Strafe auf den Fehler der Summe der Ausgangsbewegungen der Steuereinrichtung über den Steuerhorizont bezüglich der vom Optimierer definierten optimalen Zielwert-Änderung von MV ist.To meet these goals, the original unrestricted MPC algorithm can be expanded to include MV target values in the smallest quadratic solution. The objective function for this MPC control device is:

in which:
CV (k) is the vector of the "p-step forward" predictions of the controlled outputs;
R (k) is the vector of the "p-step forward" reference trajectory (the "p-step forward"setpoint);
ΔMV (k) is the vector of the incremental "c-step forward" control movements;

is a matrix of the penalties for the errors of the controlled outputs;

is a matrix of penalties on tax movements;
p is the prediction horizon (number of steps);
c is the tax horizon (number of steps); and
Γ ^{0 is} a penalty for the error in the sum of the output movements of the control device over the control horizon with regard to the optimal change in target value of MV defined by the optimizer.

To simplify the presentation is the target function for Einzeleingangs- / Einzelausgangs- (SISO) control shown.

It is understood that the first two terms the objective function for the unlimited MPC controller are while the third term sets up an additional condition, which is the sum the output movements of the control device equal to the optimal ones Makes target values. In other words, put the first two terms Goals for the dynamic operation of the control device during the third term goals for the stationary Optimization.

It should be noted that the general solution for this controller, similar to that for the unrestricted MPC controller, can be expressed as: ΔMV (k) = (S bDC Γ T ΓS u + Γ bDC Γ u ) -1 SuTΓ T ΓE P + 1 (k) = K ompC e p + 1 (k) (36) in which:
ΔMV (k) is the change in the output of the MPC controller at time k;
K _{ompc is} the optimized gain of the MPC _controller ; and
S ^{u is} the matrix of process dynamics.
S ^u can be made from the step responses of size p × c for a SISO model and
p * n × c * m for a multi-input / multi-output (MIMO) model with m control inputs and n control outputs.

For an optimized MPC, the dynamic matrix is expanded to the size (p + 1) × m for a SISO model and (p + m) * n × c * m for a MIMO model to reflect the MV error adapt. _{EP + 1} (k) is the CV error vector over the prediction horizon and the error of the sum of the output movements of the control device over the control horizon with regard to the optimal change in target value of MV. The matrix T combines the matrices Γ ^y and Γ ⁰ and is a quadratic matrix of size (p + 1) for a SISO control device and [n (p + m)] for the multi-variable control device. The uppercase T denotes a transposed matrix.

It was found that because of optimizing 54 optimized on the basis of all control and auxiliary variables CV and AV in order to determine a target value set of control variables MV _T , which defines a clear optimal working point, without the importance of the MPC control device 52 only works with a subset of the control and auxiliary variables CV and AV in their control matrix in order to produce the control variable (MV) output therefrom because if the control device 52 drives the selected subset of control and auxiliary variables CV and AV against their associated target values, the rest of the complete set of control and auxiliary variables will also lie on their target values. As a result, it was found that a square (M by M) MPC controller with an M by M control matrix with an optimizer that uses a rectangular (N by M) process model to perform the process optimization can be. This makes it possible to use standard MPC control methods with standard optimization methods without having to invert a non-quadratic matrix in a control device with the approximations and risks associated with such conversion methods.

In one embodiment, when the MPC controller is square, that is, when the number of control variables MV is equal to the number of control variables CV, the target variable (MV) value can be effectively achieved by changes in CV values as follows: ΔCV = A * ΔMVT (37) where: ΔMVT is the optimal target change of MV; and
ΔCV is the CV change to achieve an optimal MV.

Of course there will be a CV change implemented by handling CV setpoints.

Optimizing creates and updates in operation 54 the steady-state target values for the unrestricted MPC control device with each query. The MPC control device thus performs 52 the unlimited algorithm. Because the target values CV _T and AV _{T are set} in a way that takes restrictions into account as long as there is a feasible solution, the controller operates within the limits of the restrictions. Therefore, optimization is an integral part of the MPC control device.

The 3 and 4 show a flow chart 90 , which illustrates the steps used to perform integrated model prediction control and optimization. The flow chart 90 is in two sections 90a ( 3 ) and 90b ( 4 ) divides which functions occur before process operation ( 90a ), and functions that occur during process operation ( 90b ), for example with every query of process mode. Before process operation, an operator or operator takes a number of steps to get the sophisticated control block 38 including an integrated MPC control device and an optimizer. Especially in one block 92 can be a sophisticated Control template for use as an advanced control block 38 to be selected. The template can be found in a library in a configuration application on the user interface 13 stored and copied therefrom, and it can perform the general math and logic functions of the MPC controller routine 52 and the optimizer 54 without the respective MPC, process models and stationary gain or control matrices and the respective target function. This sophisticated control template can be used in other blocks such as to communicate with facilities within the process 50 configured input and output blocks or other types of functional blocks such as control blocks including PID, neuron network and fuzzy logic control block modules. It goes without saying that in one embodiment all blocks in a module are objects in an object-oriented programming paradigm, the inputs and outputs of which are connected to one another in order to carry out the communication between the blocks. In operation, the processor executing the module, using the inputs of the blocks, sequentially executes each of the blocks at a different time to produce the outputs of the blocks, which are then delivered to the inputs of other blocks as by the established communication links between the blocks defined.

In one block 94 the operator defines those in the block 38 individual control variables, control variables, restricted variables and fault variables to be used. If desired, the user can use a configuration program such as the program 40 out 1 view the control template, select inputs and outputs to be named and configured, browse the configuration environment with a standard browser to find the actual inputs and outputs in the control system, and select these actual control variables as the input and output control variables for the control template.

After selecting the inputs and outputs of the advanced control function block, the user can define the setpoints associated with the control variables, the ranges or limits associated with the control variables, the auxiliary variables and the actuating variables, and the weights associated with each of the control, auxiliary and actuating variables , Of course, some of this information, such as restriction limits or ranges, may already be linked to these variables when these variables are selected or found in the process control system configuration environment. In one block 96 out 3 If desired, the operator can configure the one or more target functions to be used in the optimizer by specifying the unit costs and / or revenues for each of the control variables, the control variables and the auxiliary variables. Of course, at this point the operator can choose to use the default target function as described above. Moreover, the user can with each of the latitude variables or criminal variables S ^+, S ^-, S _{n unte} etc. associated costs or penalties for each of the control, auxiliary and manipulated variables define S _above. If desired, the user can determine the exact manner in which the optimizer should handle or deal with unrealizable solutions as defined by the original constraint limits, such as by using margin variables, by redefining constraint limits, or by any combination of these both options.

After the inputs (control, auxiliary and disturbance variables) and the outputs (control variables) are named and connected to the sophisticated control template and the weights, limits and setpoints are linked to them, in a block 98 out 3 downloaded the sophisticated control template as a function block to be used for control in a selected control device in the process. The general nature of the control block and the manner in which it is configured is described in US Pat. No. 6,445,963 entitled "Integrated Advanced Control Blocks in Process Control Systems" which is assigned to the assignee of the present application and which is hereby expressly incorporated is incorporated herein by reference. Although this patent describes the manner of creating an MPC controller in a process control system and does not discuss the manner in which an optimizer can be connected to this controller, it is understood that the general ones used to connect and configure the controller Steps for the control block described herein 38 can be used, the template all herein for the control block 38 contains logic elements discussed instead of only those described in the reference patent.

In any case, after the sophisticated control template is downloaded to the controller, the operator can in one block 100 decide to run a test phase of the control template to generate the step response matrix and process model to be used in the MPC controller algorithm. As described in the patent referred to above, the control logic provides in the sophisticated control block 38 during the test phase, a series of pseudorandom waveforms as the process variables to the process and observes the changes in the control and auxiliary variables (which are essentially treated as control variables by the MPC controller). If desired, both the control and fault variables as well as the control and auxiliary variables can be obtained from the logging device 12 out

1 can be detected and the operator can use the configuration program 40 ( 1 ) set up to this data from the logger 12 and to somehow perform a trend analysis on this data to obtain or determine the matrix of step responses, each step response responding over time to one of the control or auxiliary variables to a unit change in one (and only one) of the position - and control variables. This unit change is usually a jump change, but could be another type of change, such as a pulse or ramp change. On the other hand, the control block 38 , if desired, the step response matrix in response to applying the pseudorandom waveforms to the process 50 generate captured data and then these waveforms to the operator interface 13 deliver which from the highly developed control block 38 generating and setting up operator or user is used.

After the jump response matrix is generated, in the event that the control and auxiliary variables outnumber the manipulated variables, the jump response matrix is used to select the subset of control and auxiliary variables which, in the MPC algorithm, is the M times -M process model or the M-by-M control matrix, which one or which in the MPC control device 52 to be inverted and used, will be used. This selection process can be done manually by the operator or automatically through a routine, for example in the user interface 13 which have access to the step response matrix. In general, a single one of the control and auxiliary variables is identified as being the most closely related to a single one of the control variables. So a single and unique (i.e. special) of the control or auxiliary variables (which are inputs of the process control device) is linked to each of the various control variables (which are the outputs of the process control device), so that the MPC algorithm is one on one M times -M set of step responses generated process model can be established.

In one embodiment, which is for manufacturing Using a heuristic approach to a pair selects the automatic routine or the operator when trying to be the only one Select control or auxiliary variables that indicate a change in unit in a certain of the control variables a combination of the highest reinforcement and having the fastest response time, the set of M (where M is the number of control variables) control and auxiliary variables and pairs these two variables. Of course, in some cases certain control or auxiliary variables with several control variables a high gain and have a fast response time. Here this tax or auxiliary variable with any of the associated control variables can be paired and can actually are paired with a control variable which is not the highest gain and produces the fastest response time because the control variable which the lower gain or slower response time, overall no other control or auxiliary variable influenced to an acceptable degree. The pairs of control variables on the one hand and control or auxiliary variables on the other hand selected to be in a general Sense the control variables with the subset of the control and auxiliary variables, which are the control variables most responsive to the control variables represent pair. moreover it doesn't matter if all control variables are not considered one of the subsets of M control and auxiliary variables can be selected and therefore the MPC controller does not have all of the control variables than their inputs receives because the set of control and auxiliary variable target values from the optimizer so chosen is that it forms a working point of the process in which the not selected Tax (as well as those not selected Auxiliary) variables at their setpoint or within their intended Work area.

Of course, because there can be dozens or even hundreds of control and auxiliary variables on the one hand and dozens or hundreds of control variables on the other hand, it can be difficult to select the set of control variables and auxiliary variables that provides the best response to each of the various control variables, at least from View point of view of the visualization. To overcome this problem, the generation routine for a sophisticated control block 40 in the operator interface 13 include or present a set of on-screen displays to the user or operator to assist and enable the operator to make appropriate selections of the control and auxiliary variables which are considered to be used during operation in the MPC controller 52 intended subset of control and auxiliary variables should be used.

Thus, the operator can in one 3 shown block 120 are presented with a screen in which the operator can view the response of each of the control and auxiliary variables to a specific or selected one of the control variables. The operator can scroll through the control variables, one by one, view the step responses of each of the control and auxiliary variables to each of the various control variables and, during the process, select the control or auxiliary variable which is the most responsive to that control variable. Usually the operator will try to select the control or control variable which has the best combination of the highest stationary gain and the fastest response time for the control variable.

It is understood that such display screens allow the operator to display and select the subset of the M control and auxiliary variables used as inputs to the MPC control algorithm, which is particularly useful when there are numerous of these variables. Of course you can the one in the block 74 certain set of control and restriction variables is automatically or electronically selected based on some predetermined criteria or a selection routine that can select the input variables to be used based on any combination of gain response and time delay as determined from the step responses for the control variables and the control variables ,

In another embodiment An automatic selection process can be started by selecting one Input / output matrix based on the state number of the matrix determine a tax matrix, e.g. by minimizing the state number up to a desired one Amount and subsequent Develop a controller configuration from the control matrix.

In this example, the condition number of matrix A ^T A can be determined to test the controllability of the matrix for a process gain matrix A. A smaller number of states usually means better controllability, whereas a higher number of states means less controllability and more control steps or movements during the operation of the dynamic control. There are no strict criteria for defining an acceptable level of controllability, and therefore the state number can be used as a relative comparison of various possible control matrices and as a test for poorly sourced matrices. As is known, a state number for a poorly procured matrix goes against "infinity". Mathematically speaking, bad nature occurs in the case of collinear process variables - that is, due to collinear rows or columns in the control matrix. As a result, cross correlation between matrix rows and columns is a significant factor affecting the state number and controllability. By carefully selecting the input / output variables in the control matrix, problems with the nature can be reduced. In practice, there is cause for concern if the status number of a tax matrix is hundreds (e.g. 500) or even higher. Such a matrix results in extremely excessive movements of the control device actuating variables.

As discussed above, the control matrix solves the problem of dynamic control while the LP optimizer solves the stationary optimization problem, and the control matrix must be a square input / output matrix be even if the MPC controller block is an odd one Number of MVs (including AVs) and CVs can have. To select the inputs and outputs for the tax matrix to start using to create the controller, are common all available MVs as control device outputs included or selected. After selecting of the outputs (of the MVs) must the process output variables (i.e. the CVs and AVs) which are included in the dynamic control matrix are included, selected so that they produce a quadratic tax matrix, which they don't is badly made.

Now a procedure for automatic or manual selection of the CVs and AVs as inputs in the tax matrix discussed it being understood that other methods are also used can.

Step 1 - CVs are selected until, if possible, the number of CVs equal to the number of MVs (i.e. the number of -Controller outputs) is. In case there are more CVs than MVs, the CVs can be in any Order and on any desired criteria such as priority, reinforcement or phase response, user input, etc. can be selected based. If the whole possible Number of CVs is equal to the number of MVs, go to step 4, the resulting state number of the quadratic control matrix to test for suitability. If the number of CVs is less than the number of MVs can AVs are selected as described in step 2. If no CVs are defined, select the AV with maximum gain with respect to an MV and go to step 2.

Step 2 - One by one the State number for every possible to the already selected, through the previously selected Calculate added AVs to CVs and AVs defined control matrix. It understands yourself that by the selected CVs defined matrix for each selected CV and AV contain one line, which is the stationary reinforcement for this CV or AV defined for each of the previously selected MVs.

Step 3 - The one determined in step 2 AV, which is in the minimum state number for the resulting matrix results, select and the matrix as the previous matrix with the addition of the chosen Define AV. If the number of MVs is now equal to the number of chosen CVs plus the number of selected ones AVs is (that is, if the matrix is now square), go to step 4. Otherwise, return to step 2.

Step 4 - Calculate the state number for the generated square control matrix A _c . If desired, the calculation of the state number for the matrix A _{c can be used} instead of the matrix A _c ^T A _c , since the state numbers for these different matrices are related to each other as the square root of the others.

Step 5 - If the one calculated in step 4 State number is acceptable by selecting the CV or AV with the maximum gain in terms of link each CV and selected AV with a MV to a certain MV until the mating completely is. At this point the selection process is complete.

If, on the other hand, the state number larger than is the minimum acceptable state number, the last one to the control matrix added Remove AV / CV and perform the wrapping procedure from step 6.

Step 6 - Perform a wrapping procedure for each of the selected MVs, one after the other, and calculate the state number of the matrix resulting from each wrapping procedure. Essentially, a wrapping procedure is performed by sequentially inserting a unit response for each of the various MVs in place of the removed AV (or CV). The unit response is a unit on one of the digits in the matrix row and zero everywhere else. Essentially, each of the individual MVs in this case is used as an input and an output instead of the AV to form a well-designed quadratic control matrix. As an example of a 4 by 4 matrix, the combinations 1000, 0100, 0010, and 0001 are inserted into the row of the removed AV in the gain matrix A _c .

Step 7 - After performing one Wrapping procedure for each of the MVs select the combination that has the minimum state number results. If there is no improvement, the original one Keep matrix. At this point by selecting the CV or AV with maximum reinforcement in terms of a particular MV, except the MV, which are used to control their is used itself (i.e. the MV that was wrapped), each selected CV and each selected Link AV to MV.

Of course, both can be done by this procedure defined control matrix as well as the resulting one State number can be submitted to the user and the user the defined control matrix for use in generating the control device accept or reject.

It should be noted that in the above described automatic procedure for the purpose of improvement controllability at most only one MV was selected to control itself (i.e. wrapped). In the manual procedure, the number of MVs wrapped arbitrarily chosen become. The MVs you choose to control your own step through that Lack of a corresponding output variable selection in the controller configuration out. You can also use more MVs as breaks to control if the number of MVs is larger than the number of all CVs plus AVs. In this way, it eventually becomes a square Control matrix supplied to the control device, which each one of the MVs as outputs Has. It is understood that the process of performing and Using breaks means the number of for selected the tax matrix CVs and AVs can be less than the number generated by the controller controlled MVs, the difference in the number of as inputs of the Tax matrix wrapped MVs exists. You can also use this wrapping procedure be used in a process where there are fewer CVs plus AVs as MVs out there.

Of course, the state number above by means of the stationary reinforcements is calculated, and therefore the tax matrix defines controllability essentially for the stationary Status. Also process dynamics (dead time, delay, etc.) and model uncertainty affect dynamic steerability, and this impact can by changing the priority of process variables (e.g. control and auxiliary variables) are taken into account be what because of the impact they have on the dynamic Have control, require their inclusion in the tax matrix can.

It is also possible for others to improve both the stationary as well as the heuristic provided for dynamic controllability Procedures to use. Such a procedure would typically be a number of heuristic criteria, including some that may be incompatible are, which are applied in several phases to develop a tax matrix and thereby a suitable sentence of control device inputs select which bring about some improvements to the tax matrix. In a such heuristic procedure, the CVs and AVs are based the relationship with the highest reinforcement grouped by MV. For each MV grouping then becomes the one process output with the fastest Dynamic and considerable reinforcement selected. This selection process can take into account the area of trust and CVs opposite Prefer AVs (where everyone else is the same). The routine then used to generate the process model during the Generation of the MPC control from each group of selected parameters. Because for each MV only one parameter is selected the answer matrix is square and can be inverted.

Definitely creates a block 124 out 3 after selecting the subset of M (or fewer) control and auxiliary variable inputs of the MPC control device from the determined quadratic control matrix, the process model or the control device, which or which in the MPC control algorithm 86 out 2 is to be used. As is known, this step to generate the control device is a computation-intensive procedure. A block 126 then loads this MPC process model (already including the control matrix itself) or this control device and, if necessary, the step responses and the matrix of the stable step response gains into the control block 38 down, and this data is put into operation in the control block 38 added. Right now is the control block 38 ready for online operation in the process 50 ,

If desired, the process step responses can be reconfigured or formed in a manner other than by generating these step responses. For example, the step responses can be copied from various models stored in the system in order to determine the step response of a specific control or auxiliary variable to a control or fault variable. In addition, a user can set the parameters for a step response, such as the stationary gain, the response time, the time constant first Order and the squared error can be defined or set and, if desired, the user can view and change the properties of a step response by setting other parameters such as a different gain or time constant if so desired. If the user specifies a different gain or other parameters, the step response model can be mathematically regenerated to include this new parameter or set of parameters. This function is useful if the user knows the parameters of the step response and has to change the step response generated in order to achieve agreement with these parameters or to comply with these parameters.

In 4 are now the general ones during each cycle of operation or query of the sophisticated control block 38 steps performed, such as using the flowchart 90a out 3 generated, represented during the process 50 works online. In one block 150 receives and processes the MPC controller 52 ( 2 ) the measured values of the control and auxiliary variables CV and AV. In particular, the control prediction process model processes the CV, AV and DV measurements or inputs to produce the future control parameter vector as well as the predicted steady-state control and auxiliary variables CV _SS and AV _SS .

In one block 152 processes or filters the input processing / filter block 58 ( 2 ) then that of the MPC controller 52 developed predicted control and auxiliary and actuating variables CV _SS , AV _SS and MV _SS and delivers these filtered values to the optimizer 54 , In one block 154 leads the optimizer 54 Standard LP methods to determine the set of M variable target values MV _T , which maximize or minimize the selected objective function or the default objective function, with none of the limits of the auxiliary and actuating variables being exceeded and with the control variables on their setpoint or within the specified ranges for these variables. Generally the optimizer calculates 54 a target value control variable solution MV _T by forcing each of the control variables and auxiliary variables to their limits. As mentioned above, in many cases there is a solution in which each of the control variables is at its set point (which can initially be treated as an upper limit of the control variable) while each of the auxiliary variables remains within their respective limits. If this is the case, the optimizer needs 54 only output the specific control variable target values MV _T , which produce an optimal result for the target function.

In some cases, however, it may be impossible to find a working point at which all control variables are at their setpoint and all auxiliary variables are within their respective limitation limits because there is no such solution because of narrow restrictions with some or all auxiliary or control variables. In these cases, using the margin variable methods and / or redefinition methods described above, the optimizer develops 54 a new objective function and / or selects a new set of limits, as through the block 155 illustrated. Then the block leads 154 the optimizer 54 again with the newly developed objective function and / or the newly developed limits with or without penalty variables in order to determine an optimal solution. If there is still no viable solution or if an otherwise unacceptable solution is found, the block can 155 again take effect to a new set of limits or a new objective function for use in another pass of the optimizer 54 define. For example, the block 155 in a first run with a solution that is not feasible, decide to use scope variables and thereby a new target function for use in the second run of the optimizer 54 to deliver. If after the second round an unrealizable or otherwise unacceptable solution is found, the block can 155 redefine the restriction limits either with or without penalty variables and these new restriction limits and possibly a new objective function for use in a third pass to the block 154 deliver. Of course the block can 155 use a combination of margin variables and limit redefinitions and a new objective function and new limits for use in any run of the optimizer 54 deliver. The block 155 can, of course, use any other desired strategy for selecting a new objective function or for redefining restriction limits in the situation in which no feasible solution is found by means of the original objective function. If desired, the in block 154 Executed original objective function include margin variable penalties that can be used to drive the process variables to a limit, setpoint, or ideal idle value during the optimizer's first pass (or any subsequent pass). In addition, the in the block 154 the original objective function used includes the margin variable constraint handling in the first place, and this can be used the first time the optimizer is run. In subsequent runs, if necessary, the redefinition of bounds can then be used either alone or in conjunction with penalty variables. Likewise, it is possible to use margin variables or limit redefinition, or both, to drive process variables against limits, against a setpoint, against an ideal idle value, etc. during a subsequent run of the optimizer using one of the methods described above.

Definitely uses a block 156 after the block 154 a feasible or acceptable Has found a solution using the target conversion block 55 ( 2 ) the gain matrix of the step responses of the stable equilibrium in order to determine the target values of the control and auxiliary variables CV _T and AV _T from the target values for the control variables MV _T , and supplies the selected subset of N (where N is equal to or less than M) these values as target value inputs to the MPC control device 52 , In one block 158 uses the MPC controller 52 the control matrix, or logic derived therefrom, to operate as an unrestricted MPC controller as described above to determine the future CV and AV vector for these target values and vector subtracts the future control parameter vector to obtain the to produce future error vector. The MPC algorithm works in a known manner to determine the stationary control variable MV _SS on the basis of the process model developed from the M by M step responses and supplies these values MV _SS to the input processing / filter block 58 ( 2 ). In one block 160 The MPC algorithm also determines the process 50 MV steps to be output and gives the first of these steps to the process in a suitable manner 50 out.

During operation, one or more can be in, for example, one of the interfaces 13 monitoring applications running either directly or via the logging device 12 Obtain information from the sophisticated control block or other function blocks communicatively connected thereto and provide the user or operator with one or more display or diagnostic screens for displaying the operating status of the advanced control block. The function block technology has cascaded inputs (CAS_IN) and cascaded remote inputs (RCAS_IN) as well as corresponding recalculation outputs (BKCAL_OUT and RCAS_OUT) on both control and output function blocks. By means of these connecting devices, it is possible to mount a supervised, optimized MPC control strategy above the existing control strategy, and this supervised control strategy can be displayed by means of one or one or more display screens or displays. Likewise, target values for the optimized MPC control device can also be changed based on a strategy, if so desired.

Although the advanced function block in the illustration given herein has an optimizer that is in the same function block and therefore is executed in the same facility as the MPC controller, it is also possible to implement the optimizer in a separate facility. In particular, the optimizer can be located in another facility, for example in one of the user workstations 13 , are located and can be used with the MPC control device as in connection with 2 described communicate during each execution or query of the control device in order to calculate the target value control variables (MV _T ) or the subset of the control and auxiliary variables (CV and AV) determined therefrom and to deliver them to the MPC control device. Of course, a special interface, for example a known OPC interface, can be used to form the communication interface between the control device or the function block containing the MPC control device and the workstation or another computer which implements or executes the optimizer , As for the 2 In the described embodiment, the optimizer and the MPC control device still have to communicate with one another during each polling cycle in order to provide integrated, optimized MPC control.

Of course, other desired types of optimizers, such as known or standard real-time optimizers that may already exist in a process control environment, can use the unreachability techniques described herein. This feature can also be used to advantage if the optimization problem is non-linear and the solution requires non-linear programming methods. Moreover, the optimizer can 54 , although used in the description given herein to develop target variables for an MPC routine, use the unreachability methods described herein to use target values or other variables for use by other types of controllers such as PID control. Control devices, fuzzy logic control devices, etc. to produce.

Although the sophisticated control block and the other blocks and routines described herein are used in connection with Fieldbus and standard 4-20 mA devices in the description herein, they can of course be implemented using any other process control communication protocol or programming environment and programming environment can be used with any other type of device, function block or control device. Although the sophisticated control blocks and associated generation and test routines described herein are preferably implemented as software, they can be implemented as hardware, firmware, etc., and can be executed by any other processor associated with a process control system. Hence the routine described herein 40 implemented in a standard multi-purpose CPU or on specially developed hardware or firmware such as ASICs, if desired. When implemented as software, the software can be in any computer readable memory such as a magnetic disk, laser disk, optical disk, or other storage medium um, stored in a RAM or ROM of a computer or processor, etc. Likewise, this software can be implemented by any known or desired delivery method including, for example, on a computer readable disc or other portable computer storage mechanism or in a modulated form over a communication channel such as a telephone line, the Internet, etc. (which is considered the same as or interchangeable will be delivered to a user or to a process control system against delivery of such software via a portable storage medium).

Although the present invention has been described with reference to specific examples which only as illustrative and in no way as limiting the invention are understood, it will be obvious to every average specialist, that changes additions or deletions to the disclosed application forms can be without departing from the spirit and scope of the invention.

Claims (53)

Method for controlling a process, comprising: the To run an optimizer that uses an objective function to Develop a set of target-defining solutions; the Determine whether the solution is realizable with respect to a set of process variable restrictions; if the solution is not feasible; (1) redefine process variable constraints for at least a process variable to a new upper and lower process variable restriction limit for the define a process variable; (2) developing one new objective function by adding a penalty variable to the objective function for the a process variable, the penalty variable being based on the amount, by which the one process variable from one of the new process variable restriction limits for the a process variable deviates; and (3) rerunning the Optimizers with the new objective function to find a new solution develop a new set of target values with the new ones Process variable constraint limits for the defines a process variable; delivering the target values or the new target values to a control device; and running the Control device using the target values or the new one Target values to a set of control signals to control the process to develop. The method of claim 1, further comprising the Creating a set of scope variables in the objective function, by one on the amount by which the at least one of the process variables violates an associated limitation, based punishment for to define at least one of the process variables. A method according to claim 2, including adjusting a unit penalty associated with the set of margin variables, which considerably is bigger than the unit cost variables in the objective function, which one define one or more economic cost values. The method of claim 1, further comprising the Creating a set of scope variables in the objective function, by one on the amount by which the at least one of the process variables deviates from an ideal rest value, based punishment for at least to define one of the process variables. A method according to claim 4, including adjusting a unit penalty associated with the set of margin variables, which considerably is bigger than the unit cost variables in the objective function, which one define one or more economic cost values. The method of claim 1, further comprising the Creating a set of scope variables in the objective function, by one on the amount by which the at least one of the process variables deviates from a target value, based penalty for at least one of the process variables define. The method of claim 6, including adjusting a unit penalty associated with the set of margin variables, which considerably is bigger than the unit cost variables in the objective function, which one define one or more economic cost values. The method of claim 1, including creating a standard sentence for the penalty variable, which is considerable is bigger as one or more unit cost variables in the objective function, which define one or more economic cost values. The method of claim 1, including creating a standard sentence for the penalty variable, which is larger than all unit cost variables in the objective function, which economic Define cost values. The method of claim 1, in which performing the Control device executing a multi-input / multi-output controller. The method of claim 10, in which performing the Multi-input / multi-output controller performing one Control device of the model prediction control type. System for use in controlling a process, which for that is designed to be implemented with a processor comprising: on computer readable medium; one on the computer readable medium stored optimizer routine, which is suitable, on the processor accomplished to become a target function to develop a solution to use which defines a set of target values; a handling routine stored on the computer readable medium the feasibility which is suitable to be executed on the processor, to determine if the solution in terms of a set of process variable constraints is realizable, and if the solution is not feasible; redefine process variable constraints for at least a process variable to a new upper and lower process variable restriction limit for the define a process variable; developing a new one Objective function by adding a penalty variable for one Process variable for the objective function, the punitive variable the objective function with one on the amount by which a process variable of one of the new process variable restriction limits for one Process variable deviates, based on punishment based; and the run again the optimizer routine with the new objective function to find a new solution develop a new set of targets with the new ones Process variable constraint limits for the defines a process variable; and one on the computer readable Medium-stored control device routine, which is suitable executed on the processor to use the target values or the new target values to develop a set of control signals to control the process. The system of claim 12, in which the objective function includes a set of margin variables, one on the amount, by which the at least one of the process variables passes an associated restriction, based punishment for Define at least one of the process variables. The system of claim 13, in which the objective function includes a unit penalty associated with the set of margin variables, which considerably is bigger as unit cost variables in the objective function, which one or more economic cost values define. The system of claim 12, in which the objective function includes a set of margin variables, one on the amount, by which the at least one of the process variables from an ideal Quiescent value deviates, based punishment for at least one of the process variables define. The system of claim 15, in which the objective function includes a unit penalty associated with the set of margin variables, which considerably is bigger as unit cost variables in the objective function, which one or more economic cost values define. The system of claim 12, in which the objective function includes a set of margin variables, one on the amount, by which the at least one of the process variables from a setpoint deviates, based punishment for Define at least one of the process variables. The system of claim 17, in which the objective function includes a unit penalty associated with the set of margin variables, which considerably is bigger as unit cost variables in the objective function, which one or more economic cost values define. The system of claim 12, in which the objective function a standard sentence for includes the penalty variable, which is considerable is bigger as one or more unit cost variables in the objective function, which define one or more economic cost values. The system of claim 12, in which the objective function implements a unit penalty for the penalty variable concludes which is greater than all unit cost variables in the target function that define economic cost values. The system of claim 12, in which the controller routine a multi-input / multi-output controller routine includes. The system of claim 21, in which the multi-input / multi-output controller routine is one Control device routine of the type prediction control includes. Method for controlling a process, comprising: the Define at least one with each one from a set of Process variables linked restriction; the Define an objective function so that it has one or more linked to the process variables economic unit cost values and one with one of the process variables, which is the limitation for the one of the process variables linked Punitive variable includes the penalty variable has a unit penalty which is greater as each of the economic unit cost values; and the Use the objective function to generate a set of process control signals to develop for use in controlling the process. The method of claim 23, in which defining of the at least one restriction defining a set associated with one of the process variables of range boundaries. The method of claim 23, in which defining of the at least one restriction defining a setpoint associated with one of the process variables includes. The method of claim 23, in which defining of the at least one restriction defining an ideal associated with one of the process variables Includes resting values. The method of claim 23, in which defining of the at least one restriction defining a first one associated with one of the process variables Set of range boundaries and one with one of the process variables linked second set of range boundaries and in which defining an objective function is to define one when the first sentence is crossed linked by area boundaries first penalty variable and one with the violation of the second sentence linked by area boundaries second punitive variable, one with the first Penal variables linked first unit penalty less than one with the second penalty variable linked second unit penalty is. The method of claim 23, in which defining of the at least one restriction defining a first one associated with one of the process variables Value and a record linked to one of the process variables of range boundaries and in which defining an objective function defines one with one of the process variables, which is from the first value deviates, linked first punitive variables and one with one of the process variables, which crosses the set of range boundaries linked second Includes punitive variables, whereby a first unit penalty linked to the first penalty variable is smaller as a second unit penalty linked to the second penalty variable is. The method of claim 28, in which the first Value is a setpoint. The method of claim 28, in which the first Value is an ideal resting value. The method of claim 23 including using the objective function to get an optimal set of control signals over a Time horizon to develop during one of the process variables is likely to exceed the limit for that process variable becomes. A method according to claim 23, comprising determining when the one of the process variables that with the one of the process variables linked restriction expected to be exceeded and redefining the constraint on one of the process variables, redefining the objective function and using the newly defined Objective function to determine the set of process control signals if one of the process variables is the one of the process variables linked restriction is likely to be exceeded. 33. The method of claim 32, wherein redefining the objective function adds one includes further punitive variables for the objective function for the one of the process variables, the further punitive variable providing the objective function with a penalty based on the amount by which the one of the process variables deviates from the newly defined restriction for the one of the process variables. The method of claim 33, in which adding of the other punitive variables, setting one with the other Penal variables linked further standard punishment, which is greater than the economic one Unit cost values in the target function, which one or more define economic cost values, includes. The method of claim 33, in which adding of the other punitive variables, setting one with the other Punishment linked includes another standard sentence, which is greater than the uniform penalty associated with the punitive variable in the Objective function. The method of claim 33, in which adding the further punitive variable the setting one with the other Punishment linked includes further standard punishment, which is considerable is bigger than the uniform penalty associated with the punitive variable in the Objective function. System for controlling a process for use with a processor containing: a computer readable medium; a first routine stored on the computer-readable medium, which is capable of running on the processor at least a constraint associated with each one of a set of process variables to save; a second one stored on the computer readable medium Routine which is suitable to be executed on the processor, to save a target function that one or more with linked to the process variables economic unit cost values and one with at least one the process variables that exceed the limit for one of the process variables, linked Penalty variable defined, where the penalty variable is a unit penalty which is larger as the economic unit cost values; and a third routine stored on the computer readable medium, which is suitable is executed on the processor to become the objective function to use a set of process control signals to develop for use in controlling the process. The system of claim 37, in which the at least a limitation for the one of the process variables is a set linked to one of the process variables of range boundaries. The system of claim 37, in which the at least a limitation for the one of the process variables is a setpoint linked to one of the process variables includes. The system of claim 37, in which the at least a limitation for the one of the process variables one ideal linked to one of the process variables Includes rest value. The system of claim 37, in which the at least a limitation for the one of the process variables a first linked to one of the process variables Set of range limits and one with one of the process variables linked second set of range boundaries and in which the objective function a first associated with the first set of range boundaries Penalty variable and one with the second set of range boundaries linked includes second penalty variable, whereby a first unit penalty linked to the first penalty variable is smaller as a second unit penalty linked to the second penalty variable is. The system of claim 37, in which the at least a limitation for the one of the process variables one with the one of the process variables linked first value and a record linked to one of the process variables of range boundaries and in which the objective function is one that deviates from the first value linked one of the process variables first penalty variable and one with the one exceeding the set of range limits linked one of the process variables includes second penalty variable, whereby a first unit penalty linked to the first penalty variable is smaller as a second unit penalty linked to the second penalty variable is. The system of claim 42, in which the first value is a setpoint. The system of claim 42, in which the first value is an ideal resting value. The system of claim 37, in which the third routine is suitable to use the objective function to get an optimal sentence of process control signals via develop a time horizon in which one of the process variables the limitation for the one of the process variables is likely to be exceeded. The system of claim 37, including a fourth routine stored on the computer readable medium, which is suitable is executed on the processor to determine when one of the process variables is the one with one of the process variable restrictions is likely to be exceeded, and about the limitation for the redefine one of the process variables to the target function redefine and initiate the third routine, the redefined objective function to use the set of process control signals to determine if one of the process variables is the one with the one of the process variables restriction expected to be exceeded becomes. The system of claim 46, in which the fourth routine is suitable, the target function by adding another Penalty variable for the objective function for redefine one of the process variables, with the other Punitive variable the objective function with one on the amount by which one of the process variables from the newly defined restriction for one of the process variable deviates, based on punishment. The system of claim 47, in which the fourth routine is suitable, another linked with the other penalty variable To set a uniform penalty, which is greater than the economic one Unit cost values in the objective function. The system of claim 47, in which the fourth routine a further unit penalty linked to the further punishment is suitable to set which is larger than the uniform penalty associated with the punitive variable in the Objective function. The system of claim 47, in which the fourth routine a further unit penalty linked to the further punishment is suitable adjust which is considerable is bigger than the uniform penalty associated with the punitive variable in the Objective function. The system of claim 37, in which the third routine is an optimizer routine which is suitable for a set of Target values for to develop a control device. The system of claim 37, in which the third routine is a control routine. The system of claim 37, in which the third routine a model prediction control type control routine is.