EP0559942B1 - Control of an electrofilter - Google Patents

Description

The present invention relates to a control for an electrostatic filter, in which filter breakdowns are continuously evaluated on the basis of evaluation rules and on the basis of the evaluation results new, optimized control parameters, i.e. at least one breakdown waiting time and one deionization time for which the electrostatic filter is determined and given to the electrostatic filter.

Such controls are widely used. As a rule, from time to time, e.g. at intervals of several minutes, filter characteristics recorded and displayed on a screen. An operator then evaluates this filter characteristic based on their experience and specifies new parameters for the control. This type of parameter specification has the disadvantage that the control parameters supplied by the operator are not always optimal. In addition, an operator must always be present.

A generic control is also known from EP 0 103 950 A2. With this control, the frequency of filter breakdowns is determined and control parameters are changed if the breakdown rate exceeds certain frequencies. The control of this font is already carrying out a certain self-optimization.

The object of the present invention is to provide a further control for an electrostatic filter which optimizes and monitors itself.

The object is achieved by the characterizing features of claim 1.

The allocation to the breakdown types is preferably weighted, the sum of the weightings always giving the value 1. Furthermore, an incoming breakdown is advantageously assigned to at most two types of breakdown at the same time.

The incoming breakdown and a predeterminable number of immediately preceding breakdowns are preferably evaluated. In principle, however, it is also possible to work with a continuously acting forgetfulness factor.

Self-optimization is particularly optimal if the filter current is also optimized, and only after the deionization time has been optimized.

In the event that high-resistance dusts are to be separated, it is advantageous to avoid back-spraying effects if a filter characteristic curve is recorded at preselectable time intervals. In the event that the filter voltage has a maximum as a function of the filter current, the filter current at which this filter voltage maximum is reached is then predefined as a filter current setpoint for the electrostatic filter.

If the control has an optimization unit designed as a neural network, the control shows self-learning behavior. This makes it possible to optimize not only the control parameters, but also the evaluation rules themselves.

If the controller has a display device located in a control room, e.g. a monitor that is assigned can be displayed by means of the display device, process states, process data and / or the control parameters. This makes it possible to monitor the electrostatic filter and, if necessary, to take corrective action. It is also possible to manually enter new parameters into the control system and to observe their effects. As a result, the control can be optimized even further than the self-optimizing control itself can.

If the controller is at least partially implemented in an independent subsystem of a networked automation system of a technical system, i.e. the self-optimizing controller is used as an independent subsystem of the automation system, the self-optimizing controller and the rest of the networked automation system have as little effect possible. Such a networked automation system is described, for example, in the older DE application P 41 25 374.4.

FIG 1

ein Blockdiagramm des Aufbaus einer Elektrofilterstromversorgung,

FIG 2

schematisch die Reaktion bei einem Durchschlag,

FIG 3

eine Verteilungsfunktion von selbstverlöschenden Durchschlägen,

FIG 4

das Prinzip der Zuordnung von Durchschlägen zu bestimmten Durchschlagsarten,

FIG 5

einen Korrekturfaktorverlauf und

FIG 6

eine Filterkennlinie.

Further advantages and details emerge from the following description of an exemplary embodiment, using the drawings and in conjunction with the further subclaims. Show:

FIG. 1

1 shows a block diagram of the structure of an electrostatic filter power supply,

FIG 2

schematically the reaction to a breakdown,

FIG 3

a distribution function of self-extinguishing breakthroughs,

FIG 4

the principle of assigning carbon to certain types of carbon,

FIG 5

a correction factor curve and

FIG 6

a filter characteristic.

According to FIG 1 there is the power supply device of an electrostatic filter, e.g. a power plant, from a power unit 1, by means of which the electrostatic filter (not shown for the sake of clarity) is supplied with electricity. The power section 1 can be, for example, an intermediate circuit converter, as is generally known, e.g. from DE-OS 35 22 569 or DE-GM 90 03 125.

The power section 1 is controlled via a control device 2, which controls the power section 1 on the basis of the specifications of the further control levels 3 to 5. The technical limits of power unit 1 are of course taken into account when driving power unit 1.

The control device 2 receives its setpoints from the monitoring unit 3. The setpoints are, for example, the current setpoint I * for the filter current I _F and an enable signal. Conversely, the control device 2 reports to the monitoring unit 3 if it cannot control the power section 1 in accordance with requirements, for example because a wire break has occurred or because one of the semiconductor power elements of the power section 1 is defective.

Furthermore, the filter voltage U _{F is} continuously evaluated in the monitoring unit 3, and certain filter states, for example filter breakdown, short circuit or other errors, are inferred therefrom. Furthermore, the monitoring unit 3 records a filter characteristic from time to time, for example every minute, the meaning and purpose of which will be explained later in connection with FIG. 6.

The monitoring unit 3 receives data both from the optimization unit 4 and from the user interface 5. The user interface 5 provides the monitoring unit 3 with, for example, the operating mode (pulse operation, DC voltage operation, mixed operation, etc.) and the time interval in which a filter characteristic curve is recorded again and again shall be.

The further parameters are initially specified by the user interface 5 to the optimization unit 4, which is typically a fuzzy control unit. The optimization unit 4 specifies to the monitoring unit 3 the further control parameters which the monitoring unit 3 requires to control the control device. The control parameters will be explained in more detail later in FIG. 2. At this point it should only be mentioned that the filter breakthroughs are detected by the monitoring unit 3 and corresponding messages are transmitted to the optimization unit 4. The optimization unit 4 automatically evaluates these messages on the basis of empirical values, determines new, optimized control parameters on the basis of the evaluation results and specifies these parameters to the monitoring unit 3. The control parameters are now explained in more detail below in connection with FIG. 2.

2 shows the current I _F flowing in the electrostatic filter and the time t to the right. A filter breakdown occurs at time T ₁ . If the filter breakdown has not gone out by itself by the time T ₂ , the monitoring unit 3 blocks the control of the power section 1 via the release line 6. The power section 1 consequently no longer feeds energy into the electrostatic filter. The current I _F in the electrostatic precipitator suddenly drops to zero due to a lack of further energy supply. At time T ₃ , the monitoring unit 3 releases the control device 2 again via the release line 6 and also gives the control device 2 the value I _O as the new current setpoint. The filter current I _F is then continuously increased to the value I *, which is reached at time T ₄ , unless a new breakdown takes place beforehand.

Typical control parameters for the power supply device of the electrostatic filter are now the breakdown waiting time t _DS , the deionization time t _E and the desired filter current I *. The breakdown waiting time t _DS is given by the difference between the times T ₂ and T ₁ . The deionization time t _E is given by the difference between the times T ₃ and T ₂ . Further possible control parameters are the rise time, ie the difference between the times T ₄ and T ₃ as well the current reduction after a breakdown, i.e. the difference between I _O and I _DS .

• Die Meldung, daß ein Durchschlag erfolgt ist,
• ob der Durchschlag selbstverlöschend war,
• die Zeit t, die seit dem letzten Durchschlag verstrichen ist,
• der Filterstrom I_DS zum Zeitpunkt des Durchschlages und
• die Löschzeit t_L, wenn der Filterdurchschlag ein selbstverlöschender Durchschlag war.

The monitoring unit 3 continuously registers whether a breakdown has occurred. If a breakthrough has occurred, a distinction is made between self-extinguishing and non-self-extinguishing punctures. So that the optimization unit 4 can optimize the control parameters breakdown waiting time t _DS , deionization time t _E and filter current I _F , the following values are transferred from the monitoring unit 3 to the optimization unit 4 with each breakdown:

• The message that a breakthrough has occurred
• whether the breakthrough was self-extinguishing
• the time t that has elapsed since the last breakdown,
• the filter current I _DS at the time of the breakdown and
• the extinguishing time t _L if the filter breakdown was a self-extinguishing breakdown.

In the optimization unit 4, the data of, for example, the last 100 copies are kept constantly available. The optimization unit 4 first optimizes the breakdown waiting time t _DS . The optimization of the breakdown waiting time t _DS is based on the fact of experience that self-extinguishing filter breakthroughs generally have a certain minimum duration on the one hand, but on the other hand no longer extinguish themselves above a certain maximum duration. To optimize the breakdown waiting time t _DS , the relative frequency of self-extinguishing filter breakdowns as a function of the extinguishing time t _L is consequently calculated from the data stored in the optimization unit 4. Then we try to find a suitable distribution curve, For example, to fit a parabola or a Gaussian curve into the H values, as is shown schematically in FIG. In FIG 3, a parabola was fitted into the H values.

, wenn M der Zeitwert ist, bei dem die gefittete H-Kurve ihr Maximum erreicht.Based on the fitted curve, a new, optimized breakdown waiting time t _DS is then calculated, for example ${\text{t}}_{\text{DS}}\text{= 1.5 x M}$

, if M is the time value at which the fitted H curve reaches its maximum.

der Löschzeiten t_L ermittelt und mit einem geeigneten Faktor F multipliziert. Der Faktor F ergibt sich aufgrund einer einfachen Überlegung. Wenn der optimale Zustand bereits erreicht ist, soll dieser ja nicht mehr verändert werden. Man nimmt also an, daß man die gefittete H-Kurve gemessen habe und daß die Meßwerte bei der gewünschten, optimierten Durchschlagswartezeit t_DS enden. Sodann errechnet man den hypothetischen Mittelwert dieser H-Kurve. Der optimale Faktor F ergibt sich dann als Quotient von optimierter Durchschlagswartezeit t_DS und diesem Mittelwert. Im obenstehenden Beispiel (Fitten einer Parabel, ${\text{t}}_{\text{DS}}\text{= 1,5 x M}$

) ergäbe sich beispielsweise ein optimaler Faktor F von 12/7.If this type of calculation of the optimal breakdown waiting time t _{DS is} too complicated, the determination of the optimized breakdown waiting time t _{DS can} also be carried out more easily. In this case it simply becomes the mean $\overline{{\text{t}}_{\text{L}}}$

the extinguishing times t _{L are} determined and multiplied by a suitable factor F. The factor F results from a simple consideration. If the optimal state has already been reached, this should not be changed. It is therefore assumed that the fitted H curve has been measured and that the measured values end at the desired, optimized breakdown waiting time t _DS . The hypothetical mean value of this H curve is then calculated. The optimal factor F then results as the quotient of the optimized breakdown waiting time t _DS and this mean value. In the example above (fitting a parabola, ${\text{t}}_{\text{DS}}\text{= 1.5 x M}$

) would result in an optimal factor F of 12/7, for example.

• Wenn die Zeit t kleiner als die Zeit t₁ ist, wird der Durchschlag mit der Wahrscheinlichkeit G_F = 1 als Folgedurchschlag gewertet.
• Wenn die Zeit t zwischen t₁ und t₂ liegt, sinkt die Wahrscheinlichkeit, mit der der Durchschlag als Folgedurchschlag zu werten ist, auf Null ab. Entsprechend sinkt auch die Gewichtung G_F, mit der der Durchschlag als Folgedurchschlag gewichtet wird, linear auf Null ab. Damit die Summe der Gewichtungen 1 bleibt, muß folglich der Gewichtungsfaktor G_R, mit der der Durchschlag als repräsentativer Durchschlag gewichtet wird, linear von Null auf 1 ansteigen.
• Wenn die Zeit t seit dem letzten Durchschlag zwischen t₂ und t₃ liegt, wird der Durchschlag mit dem Gewichtungsfaktor G_R = 1 als repräsentativer Durchschlag gewichtet.
• Oberhalb der Zeit t₃ sinkt der Gewichtungsfaktor G_R für einen repräsentativen Durchschlag linear auf Null ab. Dementsprechend steigt der Gewichtungsfaktor G_Z, mit dem der Durchschlag als zufälliger Durchschlag gewichtet wird, linear an.
• Ab der Zeit t₄ wird der Durchschlag dann ausschließlich als zufälliger Durchschlag gewichtet, d.h. G_Z = 1.

After calculating the optimized breakdown waiting time t _DS , a new, optimized deionization time t _{E is} calculated. For this purpose, the registered breakthroughs are assigned to the "follow-through", "representative" or "random" types. The selection criterion for assigning a breakdown to one of the breakdown types is the time since the previous one Breakthrough has passed. The optimization of the deionization time t _E is based on the fact that empirical breakthroughs occur continuously if the deionization time t _{E is} too short. On the other hand, since follow-through breakdowns also occur continuously if the breakdown waiting time t _{DS is} too small, the deionization time t _E can only be optimized after the breakdown waiting time t _{DS has been} optimized.

• If the time t is less than the time t ₁ , the breakdown is evaluated as a subsequent breakdown with the probability G _F = 1.
• If the time t lies between t ₁ and t ₂ , the probability with which the breakdown is to be regarded as a subsequent breakdown drops to zero. Correspondingly, the weighting G _F , with which the breakdown is weighted as a subsequent breakdown, drops linearly to zero. So that the sum of the weights remains 1, the weighting factor G _R , with which the breakdown is weighted as a representative breakdown, must increase linearly from zero to 1.
• If the time t since the last breakdown is between t ₂ and t ₃ , the breakdown is weighted with the weighting factor G _R = 1 as a representative breakdown.
• Above time t ₃ , the weighting factor G _R drops linearly to zero for a representative breakdown. Accordingly, the weighting factor G _Z , with which the breakdown is weighted as a random breakdown, increases linearly.
• From time t ₄ the breakdown is then weighted exclusively as a random breakdown, ie G _Z = 1.

Instead of the weighting functions G _F , G _R and G _Z shown in FIG. 4, other weighting functions can of course also be used be used. It should only be ensured that the sum of the selected weighting functions is always 1 and that the weighting functions G _F and G _Z do not overlap, or in other words that t _{3 is} always greater than or at most equal to t ₂ .

In order to keep the influence of individual, random "outliers" as low as possible, not only the last breakthrough is evaluated, but, as before when optimizing the breakthrough waiting time t _DS , all the saved breakthroughs are evaluated. The number of copies to be saved can be specified as required. The number can be 50, for example, but can also be 500.

From the stored punctures the optimization unit 4 calculates the quotient of the number N _R of representative punches and the number N _F of secondary breakdowns. The number N _R of representative breakdowns is given by the equation $${\text{N}}_{\text{R}}{\text{= Σ G}}_{\text{R}}$$

The number N _F of follow-through breakdowns through the equation is analogous $${\text{N}}_{\text{F}}{\text{= Σ G}}_{\text{F}}$$

The deionization time t _E is optimal when the ratio of representative carbon to secondary carbon has a value that can be selected by the user, for example 10. If the ratio is greater than 10, the deionization time t _{E is} too long and must therefore be reduced. Conversely, the deionization time t _E must be increased if too many follow-throughs occur. The deionization time t _E is optimized in the following way.

First, a correction factor k _E for the deionization time is calculated from the quotient of representative breakthroughs to follow-through breakthroughs based on a predetermined function curve. k _E must be zero if the ratio of representative breakdowns to random breakdowns is reached. The correction factor k _E must be greater than zero if the ratio is undershot and less than zero if the ratio is exceeded.

5 shows a particularly simple possibility for calculating the correction factor k _E.

The new, optimized deionization time t _E then results in $${\text{t}}_{\text{E}}{\text{: = (1 + k}}_{\text{E}}{\text{). t}}_{\text{E}}$$

berechnet.The filter current I _F is then optimized last. For this purpose, use is made of the knowledge that with an optimized breakdown waiting time t _DS and deionization time t _E, practically only random filter breakdowns occur if the filter current I _{F is} too low. Conversely, the ratio of representative breakdowns to random breakdowns increases when the filter current I _F approaches its breakdown limit. Analogous to the optimization of the deionizing time t _E, therefore, except for the number N _R of representative punches nor the number N _Z of random punches, according to the equation $${\text{N}}_{\text{Z.}}{\text{= Σ G}}_{\text{Z.}}$$

calculated.

A correction factor k _I for the filter current setpoint I * then results from the ratio of N _R to N _Z. The new setpoint I * results from the equation $${\text{I *: = (1 + k}}_{\text{I.}}\text{). I *}$$

The slopes of the correction functions k _E or k _I can of course also run differently than shown in FIG. 5. In particular, they can be steeper or flatter if this is required for reasons of stability, and in particular the correction functions k _E and k _{I can} also be different from one another.

The new, optimized control parameters t _DS , t _E and I * determined in this way are specified by the optimization unit 4 of the monitoring unit 3. The monitoring unit 3 therefore controls the electrostatic precipitator with these optimized control parameters t _DS , t _E and I * until the operating behavior of the electrostatic filter changes and a new correction of the control parameters t _DS , t _E and I * is necessary.

From time to time; For example, every minute or every 5 minutes, the optimization unit 4 triggers the recording of a filter characteristic by the monitoring unit 3. The monitoring unit 3 receives then a filter characteristic of the filter voltage U _F ie, as a function of the filter current I _F is determined.

If the filter characteristic curve is essentially monotonous, as indicated for curve 1 in FIG. 6, the monitoring unit 3 is used as the setpoint I * for the filter current I _F the maximum current I _max , minus a safety discount of, for example, 5%, is predetermined and the filter current I _F is optimized as described above.

If, on the other hand, the filter characteristic curve has a maximum, as shown in curve 2, the monitoring unit 3 is given the setpoint I * for the filter current I _{F as} the current at which the voltage _maximum U _{max is} reached. In this case, the optimization of the filter current described above is omitted. The optimization of the filter current would be pointless in this case, since above the voltage _maximum U _max a back spray would occur, which would greatly reduce the efficiency of the electrostatic filter.

In conclusion, it should be mentioned that the control device 2, the monitoring device 3, the optimization unit 4 and the user interface 5 can of course also be implemented in the form of a computer program, a hardware configuration is not necessary. It should also be mentioned that the present invention can of course also be used with pulsed electrostatic filters. In this case, the pulse height and duration as well as the pulse repetition frequency are available for optimization in addition to the breakdown waiting time t _DS and deionization time t _E.

The optimization unit can of course not only understand the human, fuzzy logic, so it is not just a fuzzy control unit, but can also be designed as a neural network with self-learning behavior. This makes it possible not only to optimize the control parameters, but also the evaluation rules themselves. This self-learning behavior is generally known as "supervised learning by back propagation" of a neural network. The neural network can be formed in a manner known per se.

Claims (12)

1. Control for an electrostatic filter, where filter breakdowns are continuously evaluated on the basis of evaluation rules and, on the basis of the evaluation results, new, optimized control parameters (t_DS, t_E), i.e. at least a breakdown waiting time (t_DS) and a deionizing time (t_E), are determined for the electrostatic filter and are specified for the electrostatic filter, characterized in that the evaluation rules comprise the following steps, carried out in this sequence:

   - on the basis of breakdown-specific characteristic data, the allocation of each detected filter breakdown to at least one of the breakdown types: self-extinguishing breakdown, sequential breakdown and typical breakdown;

   - determination of an optimized breakdown waiting time (t_DS) from the relative frequency of the self-extinguishing filter breakdowns as a function of the extinguishing times (t_L); and

   - determination of an optimized deionizing time (t_E) from the statistical ratio between sequential breakdowns and typical breakdowns.
2. Control according to claim 1, characterized in that the breakdown-specific characteristic data (t,I_DS,t_L) include the time (t) since the last breakdown.
3. Control according to claim 1 or 2, characterized in that the allocation to the breakdown types is weighted, whereby the sum ( ${\text{G}}_{\text{F}}{\text{+G}}_{\text{R}}{\text{+G}}_{\text{Z}}$

   

   ) of the weightings (G_F,G_R,G_Z) always results in the value "one".
4. Control according to claim 1, 2 or 3, characterized in that an occurring breakdown is allocated to at most two breakdown types at the same time.
5. Control according to one of the above claims,
   characterized in that to optimize the control parameters (t_DS,t_E) the occurring breakdown and a specifiable number of immediately preceding breakdowns are evaluated.
6. Control according to one of the above claims,
   characterized in that to evaluate the breakdowns mean values ( $\overline{{\text{t}}_{\text{L}}}$

   

   ) are used.
7. Control according to one of the above claims,
   characterized in that the optimized control parameters (t_DS, t_E) are loaded with safety allowances or safety reductions.
8. Control according to one of the above claims,
   characterized in that the filter current (I*) is also optimized, whereby the filter current (I*) is only optimized after the optimization of the deionizing time (t_E).
9. Control according to one of the above claims,
   characterized in that at preselectable intervals a filter characteristic is recorded and in that for the case where the filter voltage (U_F) as a function of the filter current (I_F) has a maximum (U_max), the filter current, where this filter voltage maximum (U_max) is reached, is firmly specified for the electrostatic filter as desired filter-current value (I*).
10. Control according to one of the above claims,
    characterized in that the control (1-5) has an optimization unit (4) which is preferably constructed as a neuronal system.
11. Control according to one of the above claims,
    characterized in that there is associated with the control a display unit arranged in a control room, for example a monitor, so that by means of the display unit process states, process data (for example $\overline{{\text{t}}_{\text{L}}}$

    

    ) and/or the control parameters (t_DS,t_E,I*) can be represented.
12. Control according to one of the above claims,
    characterized in that the control (1-5) is implemented at least partially (2-4) in an independent subsystem of a cross-linked automation system of a technical installation.

Images