PROCEDURE AND DEVICE FOR SUBOPTIMALEN THE REGULATION OF MEANS OF A SEARCH STRATEGY AND A PROCEDURE AND A DEVICE FOR THE GASZERLEGUNG, IN PARTICULAR FOR VERY LOW-TEMPERATURE LIQUID AIR SEPARATION

The invention relates to a method and a device for controlling according to & Nkulengu rail; in claims 1 or 10.

Besides [...] & Nkulengu rail; e, the [...] & Nkulengu rail; e and the mathematically mediating model ("first model") exists between them in the method, a [...] & Nkulengu rail; e, the acts to substantially known manner on the ("second model") [...] & Nkulengu rail; e.

The terms " [...] & Nkulengu rail; e", " [...] & Nkulengu rail; e" and " [...] & Nkulengu rail; e" are here used in a generalized meaning. You can describe not only a single parameter, but also in each case a set of two or more of such parameters, for example, physical parameters. In practice exactly one parameter at all three size often the consideration satisfies & Nkulengu rail; en ; the, the invention is readily applicable to systems, where the [...] & Nkulengu rail; e, the [...] e & Nkulengu rail; and/or the [...] & Nkulengu rail; e is formed by more than one parameter.

The [...] & Nkulengu rail within the framework of the method of adjustment, determined value of the; e are transferred either directly to a mechanical adjusting device can, or in the form of an analog or digital signal as [...] & Nkulengu rail; e are transferred to a further control system, the further physical parameter sets in turn, to ensure the corresponding value of the [...] & Nkulengu rail; e in the above meaning.

The [...] & Nkulengu rail; e lie-at least temporarily-information on their suspected future course before. In practice not all disturbances in their progression over time can be predicted. This information on the future is often erroneously or incompletely. Periods or time periods, where such information completely absent, it is assumed that the [...] & Nkulengu rail; e does not change; in this case the "prediction points" in the sense of the invention then a constant History on.

Also the system comprises a set of boundary conditions on, for example for an allowable range of values for the [...] & Nkulengu rail; e, a limit and a maximum adjusting speed of the [...] & Nkulengu rail; e and the like.

Classic rule strategies and known model method (model-MPC [...] control) to be expected in the future to take account of information on events au & Nkulengu rail; [...] of the system, as the prediction of the [...] & Nkulengu rail; e represents, as a rule, not.

The invention starts back up the fundamental, exploit this information for controlling the system, so as to allow a particularly advantageous operation of the technical system.

First it is an optimization problem, in addition to the information available in the past and present [...] & Nkulengu rail also the prediction of the above; e enters into for the future. For this can be resolved in principle equations established and numerically. The computational expense therefor is, however, very gro & Nkulengu rail ;; au & Nkulengu rail; the behaviour of the system under certain circumstances and for the unstable is Several units also comprise engineers but transparent.

Under the invention has proven itself that a, in a strictly mathematical meaning, "optimum" course [...] & Nkulengu rail for adjusting and/or by; e on many technical systems is not necessary. On the contrary, a rigorous optimization leads to frequent changes in the operation, which in themselves are undesirable. Therefore according to this invention is to exploit only voluntarily [...] & Nkulengu rail [...] course of the a; e. It Locating [...] & Nkulengu rail not the optimum value of the; e, but only any acceptable time course, for example, a set of predetermined boundary conditions not violated.

This suboptimal [...] & Nkulengu rail The amounts to future course of the; e a search strategy is now used, in which simplified assumptions about the possibilities of varying the [...] & Nkulengu rail; e are specified. Composed in a finite set of discrete possibilities of modification , wherein each element of this set is associated with a different priority value (that is to say, every scan). This possibilities of modification and their priority values are generally constant; there is, in principle, the [...] is & Nkulengu rail; e Method but also on time varying parameters of this type applicable.

In [...] & Nkulengu rail one-dimensional; e the possibilities of modification be formed by three values in the simplest case, for example, by

• [...] & Nkulengu rail an increase A + of the; e per unit of time,
• a reduction & Delta ;-the [...] & Nkulengu rail; e per unit time (e.g. & Delta ;-=-Δ+)and
• the value zero (constant [...] & Nkulengu rail; e).

The priority values are, for example, in the context of the technical facilities for [...] & Nkulengu rail; took determined. Such [...] & Nkulengu rail; took be used, when in the context of the implementation of a permitted limit of the possibilities of modification[...] & Nkulengu rail; e would risk being exceeded. Is [...] upon reaching the upper limit to be taken for example the & Nkulengu rail; nahme less expensive than the if there is a risk of a breach the lower limit to be used, the increase a higher priority than the reduction & Delta A, + ;-. The change 0 has the lowest priority. This can [...] & Nkulengu rail; en described be in tabular form:

The possibilities of modification allows the use of a simplified description of the search strategy now. This determined from a starting moment t_0, which as a rule is equal to the current real time, the future course of the [...] & Nkulengu rail; e to at a later date t_n, the [...]. This Search the be taken into account at the time the progress of the information available on 0 [...] & Nkulengu rail t; e in period t_{0 to sn.}

For the actual scheme is taken into account the initial value of the determined future [...] & Nkulengu rail temporal course of the; e. This as a digital or analog signal is adjusted in the technical system or output.

The entire search strategy Retrying point of real time For the next. Here [...] & Nkulengu rail the instantaneous measurement values can the adjusting and; e and an updated prediction for the future timing of the [...] & Nkulengu rail; e refer. In principle, this can in the renewed application of the search strategy [...]point of real time future course determined of the outcome of the for the previous & Nkulengu rail; e are taken into consideration; but preferably the search strategy is carried out until the new completely new [...] , solely on the basis of the current measured values and the current prediction of the [...] & Nkulengu rail; e. Generally point of real time[...] constant is the time interval between and held; in principle, however, it is possible within the scope of the invention, this distance to vary.

In implementing the search strategy is the interval between the current instant t [...] t_{n 0 and the into discrete}, equidistant in particular, [t_{0, 1 ton]} time steps, [t_{1, 2 t]} to [_{n -1 t,} t_n] divided ; (_{i = 1, ..., n) for each intermediate point is a value t} i for the determined [...] & Nkulengu rail; e and the [...] & Nkulengu rail; e. A predicted course for this size gives To & Nkulengu rail; en. For the current setting the [...] & Nkulengu rail; e although, as a rule, only the first value of that prediction is used. However, the entire future course is of adjusting and/or [...] & Nkulengu rail; e a valuable information, which may for this purpose are output graphically or numerically. This is particularly advantageous, when the operating personnel or further regulating devices can affect the future course. limit value injury advance the method such as a Does, the boundaries of the system is not avoidable because of physical, so the operator can be informed via the formation of an alarm, the appropriate Ma & Nkulengu rail initiates then took ;.

In many cases the technical system can [...] & Nkulengu rail changes in the desired; e not immediately implement, but requires a certain amount of time. This behaviour is described with a model for the timing of the change in the [...] & Nkulengu rail; e, the within the scope of the invention in the search strategy can be taken into account.

A measurement Should the [...] & Nkulengu rail; e be not be available or the measurement inaccurate, thus allows the use of an estimated value for the [...] & Nkulengu rail the method also; e or the measuring error.

The technical system can for example by a gas supply device with a gas separation unit, in particular a cryogenic air separation are formed. In the simplest case of one-dimensional size & Nkulengu rail; en the [...] is & Nkulengu rail; e for example by flow of the product fluid generated in the gas decomposition, the [...] & Nkulengu rail; e by the pressure in a reservoir and the said product fluid [...] & Nkulengu rail; e are formed by variations in consumption of the product fluid. The reservoir can, for example, by a pressure line (for example, a pipeline system) or by a dedicated pressure accumulator are formed or both.

The invention as well as further details of the invention are explained in more detail hereinafter with reference to an embodiment, the concerns the regulation of the production quantity of an air separation plant, in particular a cryogenic air separation plant.

Among the difficult tasks of the [...] -automation include rules, the production rates of the installations to a current consumption or the installations as Customize method that a predicted course of the consumption (Pipeline-Folgeregelung) is taken into account. The method is based on a numerical search algorithm described hereinafter, the production rates [...] a suboptimal course of the. [...] is to be understood in this context is avoided as far as possible via emergency supply systems that blowing off products or their supply, but no further [...] ng, for example, takes place in the sense of a smallest possible energy consumption of the production plant.

A [...] provides a gaseous product with the mass flow ( _{" [...] & Nkulengu rail; e") in} F I a pipeline. Consumers depend on the pipeline, the decrease in the sum the amount F_O. The pressure in the pipe at a future time (" [...] & Nkulengu rail; e") is_{calculated as follows t} n:

The symbols are used as follows:

P_g

measured pressure in the pipeline

P_para

[...] pressure in the pipeline

F_Ip

predicted production quantity of the plant

F_Op

[...] total consumption

F_err

systematic errors in the Quantity measures

T_p

pipeline printing time constant of the

t

Time

t₀

present time

t_n

future date

The operation of the plant is subject to restrictions, the can be as following formulate inequation secondary conditions , i.e. production quantities and the adjusting speed of the installation are limited: F ≤ F ≤ F F ≤ F dt ≤ F

Objective it is the operation of the plant, the pressure in the pipeline within pre-determined limits: P ≤ P ≤ P

Future consumption amounts as predicted time series (t) lie_F Op ("prediction for the future timing of the [...] & Nkulengu rail; e") before, the prediction must at any time be correctable. We Where to, the systematic error in the mass flow is approximately constant, so is F leaves_{err estimate as follows}:

The symbols are used as follows:

P_g

measured pressure in the pipeline

F_Ig

measured production quantity of the plant

F_Og

measured total consumption

F_err

systematic errors in the Quantity measures

T_p

pipeline printing time constant of the

t

Time

t₀

present time

t_m

Time from the past

Furthermore is given the characteristic, with which the plant may be adjusted in their production quantity. These load change program ALC is normally a delay due to the 1st order with the following differential equation:

The symbols are used as follows:

F_I

Production quantity of the plant

F_S

load change program theoretical quantity of the

f (_{F S)} ALC

ALC-function, usually 1st-order polynomial

T_{ALC (Fs/dt)}

Time constant of the ALC-programme

The time constant of the ALC-Programme T_{1 (Fs/dt) can} be dependent on the change direction of:

Is a future time course of F Wanted_{S with the} it is possible, to keep the pipeline-pressure within the predetermined limits.

In order to solve the above-stated problem are converted into the time-discrete form first of all the equations (Euler-Integration).

F ≤ F ≤ F P ≤ P ≤ P

quantity error :

ALC-Characteristic: F = a * b + F

The symbols are used as follows:

P_g

measured pressure in the pipeline

P_para

[...] pressure in the pipeline

F_Ip

predicted production quantity of the plant

F_Op

[...] total consumption

F_err

systematic errors in the Quantity measures

F_S

load change program theoretical quantity of the

[...]_I

of ALC calculated amount

T_p

pipeline printing time constant of the

T_ALC

Time constant of the ALC load change,

Δ t

Sampling Time

i

Indices of the time series

n

Index of the [...]

m

Index of the evaluation horizon , m is negative

0

present time

(I) is a time series F_{S Wanted,} with the the following quality criterion which is minimal:

The symbols are used as follows:

P_S

Pressure Setpoint Value

P_para

[...] pressure in the pipeline

i

Indices of the time series

n

[...]

inequation secondary conditions The above with respect to the minimum/maximum quantities and are Hard-Constraints quantity gradients (these are hard physical limits) and soft-Constraints with respect to the pipeline-pressure. The [...]. Can prevent overrun of the pressure is, if not it will be the one, by emergency supply or blowing prevented.

The index n is referred to as [...]. The index m has the function of a filter horizon or evaluation horizon.

Problem: When applying conventional optimization method must be very many parameters are optimized, i.e. the values of the time series (i) F_S. A way out offers the following search strategy, the provides a [...] , sufficient solution for the desired application.

To solve the problem of the pressure maintenance in the pipe it is not really necessary, an optimum solution for the time series (i) to find_F S. It is sufficient, to find any time series, with whose aid it is possible, to keep within the permitted limits the pipeline printing ("boundary conditions for the [...] & Nkulengu rail; e"). Release can be made with a search strategy. It the values are either not _{(i + 1) with respect to (i) from F S F} S is changed or Smax Smax + /-Δ F_Smax, i.e. Δ F_{S (i) either} 0, + Δ F or-Δ F . Furthermore, provided that the change in a direction has priority over an opposite change, e.g. increasing the production has priority over a reduction, as a back-up supply is more expensive than a blowing of the product. One possible search strategy would then read as follows:

1. 1. the future course Calculating pipeline printing P_{p (i) as long as} the, until the pressure a limit being violated. As a function thereof, whether the minimum or maximum limit has been violated, F_{S either increased or decreased to} a preceding time is. This Locating previous time.
2. 2. in the time series order k intervals back Go, at a maximum, up to the current interval 0 is oriented on the time constant of a. The number of intervals k ALC-load change. The Search Canceling but, as soon as a time is reached, at which a further adaptation of F_{S is not possible due to its} minimum or maximum limits (i). In latter case would draw with step 4 further.
3. 3. Falls Δ F_{S (i-n) already} equal to the desired Δ F_{S is}, so go further back in time, until a Δ F_{S not equal to} the desired & Delta (i-n); F_{S is} found, at a maximum, up to the current interval 0.
4. 4. in the interval starting from (i-k) the nearest (i-k) to i Search Δ F_{S (i -1),} which is infinitely Δ F_{S is}. Falls Δ F_{S has the} opposite sign as Δ F_{S (i-j) and} Δ F_{S (i-j) priority} over Δ F_S, so look for further.
5. 5. Falls no adjustment is possible_{until the time i} S of F, go at the next time and make further comprising step 1 (i.e. a further adaptation of the production rate is not possible).
6. 6. Replace Δ F_{S (i-j) by} Δ F_{S and i} expect a = i-j. Sham further comprising step 1, taking into account the modified i.
7. 7. the above procedure (i = 1 to n) until the complete time series repeat is calculated.

The calculation of the complete time series for the future [...] (" [...] future time course of the & Nkulengu rail; e") takes place in each time step newly. Changes in the predicted consumption and reducing measurement errors are taken into account immediately thereby. From the predicted time series only the first value is used, i.e. the required [...] with for the moment,. All future values of the time series will be discarded as a result of the recalculation in the subsequent time step.

By modified search strategies it is possible, to achieve the following further properties, to deal with uncertainty in the predicted consumption amounts to:

• To are carried out to achieve is to manipulate to manipulate is left to avoid is led towards that the [...] & Nkulengu rail; e and to close to a threshold value that the [...] & Nkulengu rail; e on an average desired value (in the absence of breakdown suggestion), it possible, either the used in the calculation and/or about the time limits of the pipeline-pressure, to the changes in the [...] & Nkulengu rail; e.
• Due to the allowable bandwidth for the pipeline-pressure is within certain limits with respect to the time of a a variation optionchange of desired value. This variation option can be used, so as to control the priority of leads desired value being holds. e.g. it is to be performed in the listed useful application case, increases and decreases in the nominal value as early as possible if possible late, resulting in a control behaviour, including the pipeline-pressure as high as possible.

Figure 1:

Example of a predictive output regulation according to & Nkulengu rail; of the invention

Figure 2:

Representation of the predictions of quantities and pressure

Figure 3:

Flowchart [...] & Nkulengu rail provided for carrying out the; en search strategy in a concrete application

Figure 1 shows a typical sample application for the described search strategy. The product provides cryogenic air separation plant ([...]) [...] A ([...] Oxygen) in a pipeline network that would improve. The air separation plant is of an automation system for load change (ALC-Automatic Load Change) guided. At relevant [...] & Nkulengu rail; en [...] and the pressure in the pipeline network that would improve the available are the production volume. If possible, should still another size & Nkulengu rail; e, the measured amount of consumption, are used. No measurement Represents the Usage Quantity available, so this is estimated with the above indicated equation for the measuring error F_err. (In the estimation equation for the measuring error of the measured total consumption F is_{Og then either} replaced by an estimated total consumption NULL set or on).

The predictive search algorithm is based on the described output regulation , hereinafter referred to as (Automatic Product adaptation) APA. The search algorithm includes the model of the pressure accumulator or the pipeline as a process model ("second model"), and, as the behaviour of the of ALC guided [...] ("first model"). That concludes from the point of view of a part of the process to be regulated the ALC-run [...] APA.

[...] & Nkulengu rail; en for APA pipeline printing are and the current production quantity of. Furthermore, made, e.g. from the plant driver , a prediction of the consumption amounts available. Based on these data a desired future course of the ALC-set point calculated APA. This calculation is the prediction of the pressure gradient as the pipeline by-product A under the conditions that the predicted consumption amounts are observed and the [...] the ALC-desired values follows.

In each time step, e.g. every 15 seconds, the prediction newly calculated APA. The current pipeline-pressure and changes in the prediction of the consumption amounts are along the so for an immediate correction of the predictions and ensure a recalculation.

The predicted curves representing consumption amounts, ALC-desired value and are valuable information for the pipeline printingplant driver and (Figure 2) are represented as plant drivertrend curve for the.

The flow chart of Figure 3 the outlet in a special application case with the following specific parameters shows the search strategy:

• Dividing the [...] into n equidistant time steps
• Three discrete possibilities of modification 0, & Delta ; +, & Delta ;-, wherein & Delta ;-=-& Delta ;+
• ... with the priority values: Prio (0) = 0 Prio (& Delta ;-) = (-1) = 1 Prio Prio (Δ+)= Prio (1) = 2
• Vector Δ F (i) with i = 1, ..., n [...] for the values of the changes in the & Nkulengu rail; e at the times with the corresponding index

The [...] & Nkulengu rail components of the; en-vector Δ F first of all, in principle, arbitrarily assigned a (i) are, for example [...] & Nkulengu rail; [...] & Nkulengu rail with the value 0. hinged; [...] time in the future is calculated from the end of the gradually [...] & Nkulengu rail; e [...] & Nkulengu rail (pressure) on the basis of the current; en-vector, the currently known [...] & Nkulengu rail; en-course [...] - [...] and of the model.

I = 1 is examined for each index i < n Starting with, whether the [...] & Nkulengu rail; e said predetermined levels limit value injury found a violated and the direction R of the:

1. R = 1, if the upper limit value is being violated
2. R =-1, if the lower limit value is being violated

limit value injury Occurs on a, in the time series is decreased, until a value Δ F (i-k) is found, the i is non-zero, or until the time 0 (k=)or a predetermined maximum value is achieved for the number of setbacks kmax. In said searched (i-k) are now taken into account the priority values time.

Falls the priority value (-R) the desired direction of change greater & Nkulengu rail Prio; he than the priority value of the currently calculated value of (δ F (i-k)) Prio [...] & Nkulengu rail; e is, this component is [...] & Nkulengu rail amended of the; en-vector corresponding to and i counted up.

This Is not the case, there is again a forwardly into the future and preceded Δ F sought (i-k), whose priority is smaller than that of the desired change-R, and there the modify. Remains also these Nothing found, the inevitable limit value injury is below the predetermined boundary conditions at this point and the calculation is now progressing at the next highest value for i further, without the the [...] & Nkulengu rail; en-vector had been modified.

Will i = n reaches, the calculation is of the future temporal course of adjusting and [...] & Nkulengu rail; e for the current time is terminated and the first value Δ F (1) the calculated [...] & Nkulengu rail; en-vector is predetermined in the physical system. The value Δ F (1) is output as a digital or analog signal or in the technical system adjusted.