JPH05279980A - Apparatus for controlling recovery boiler - Google Patents

Abstract

PURPOSE:To provide an apparatus for controlling a recovery boiler and capable of realizing completely automated operation of the boiler. CONSTITUTION:The image information obtained from a combustion furnace is analyzed by a neural network 22 for real-line analysis to recognize the patterns of the char bed form deposited on the bottom of the furnace and the distribution of the suspension at the upper part of the furnace. The recognition information is inputted together with a black liquor flow rate signal and an exhaust gas concentration signal into a preset value varying part 30, in which the preset values of the black liquor temperature and the amount of combustion air are varied based on a fuzzy inference to approach the optimum operation condition. Separately, the apparatus is provided with an automatic operation controlling part 35 to inspect the image information and the recognition information of the neural net work for real line, to switch the operation mode according to the command inputted by an operator and to perform the learning operation by a data storage part 37 for learning and estimation, a neural network 38 for learning and estimation and a learning and estimation managing part 39.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a control device for controlling a recovery boiler which is a power plant of a pulp and paper plant.

[0002]

2. Description of the Related Art As is well known, in a paper pulp manufacturing process, black liquor (cooking waste liquor) is generated in the cooking process. This black liquor contains combustible components (organic components of raw wood) and also valuable chemicals (caustic soda) used in the cooking process. Therefore, in a pulp and paper plant, normally, the black liquor is guided to a recovery boiler and burned, and the thermal energy obtained by this combustion is used to obtain steam for on-site power generation and to recover the contained chemicals. There is. FIG. 12 shows a schematic configuration of such a recovery boiler.

The black liquor is sprayed into the combustion furnace 2 from the black liquor gun 1, is dried in the vicinity of 4 directly above the charvet 3, which is a deposit of black liquor, and lands on the charvet 3 while igniting. The air required for combustion is the primary low air flow 1L, the primary high air flow 1H, the secondary low air flow 2L, and the secondary high air flow 2H from the blower 5.
Given as.

As the reduction reaction progresses in the charbet 3, the chemical 7 is recovered in the recovery tank 6 as a reduced product. On the other hand, the high temperature gas generated by the combustion passes through the heat exchangers 8 to 11 and exchanges heat with the water 12 flowing in these heat exchangers. Then, the gas that has passed through the heat exchanger is discharged from the discharge port 13. Further, the steam 14 generated by the heat exchange is guided to a turbine or the like (not shown).
When operating such a recovery boiler, it is necessary to pay attention to the following three points.

(1) The recovery rate of the chemical 7 largely depends on the shape of the pile of the charbed 3. That is, the better the shape, the higher the recovery rate obtained. Therefore, it is necessary to always improve the mountain shape of the charbed 3 by some means.

(2) When the black liquor floating near the area directly above 4 rises together with the combustion gas and adheres to the heat exchanger, the heat exchange efficiency is significantly reduced. In the worst case, a dust trouble will occur and the boiler must be stopped. Therefore, it is necessary to avoid the floating state as much as possible. (3) SO ₂ and N contained in exhaust gas from the standpoint of environmental problems
It is necessary to keep the amount of Ox below the reference value.

Taking these points into consideration, in the operation of the recovery boiler described above, the concentration of the exhaust gas component and the flow rate of the black liquor are measured by the concentration measuring device 15 and the flow meter 16, and the mountain shape of the charbed 3 and The suspended matter in the upper space of the combustion furnace 2 is photographed by the ITV cameras 17a and 17b,
Based on these information, the operator makes full use of his experience for a long time, and the value measured by the concentration measuring device 15 is below the reference value,
In addition, a method of manually controlling the blower 5 and the black liquor temperature regulator 16 is adopted so that the peak shape of the charbet 3 is the best and the suspended matter in the upper space of the combustion furnace 2 is minimized.

However, in such a control system,
Since it depends only on the intuition of the operator, there is a problem that it is not possible to always carry out optimal control and a great deal of labor is required.

In order to solve such a problem, recently, a proposal has been made to combine a neural network and fuzzy inference to completely automate the control of the recovery boiler. However, this proposal was not perfect, and it was insufficient to integrate it into the actual system.

[0010]

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a recovery boiler control device capable of fully automating the control by combining a neural network and fuzzy inference and operating in accordance with an actual system. There is.

[0011]

In order to achieve the above object, the control device according to the present invention is provided with a means for obtaining image information in the combustion furnace, and the image information obtained by this means is analyzed to make a furnace. A real line neural network is provided which recognizes and outputs to which pattern among a plurality of patterns the shape of the charbed deposited on the inner bottom and the distribution of suspended matter in the upper part of the furnace belong. By inputting the recognition information of this real line neural network, the black liquor flow rate information and the exhaust gas concentration information, the set values of the black liquor temperature and the combustion air amount are changed to values that can approach the optimum operation based on fuzzy inference. A setting value changing means is provided. on the other hand,
An operation mode in which the recognition information obtained by the neural network for a real line is introduced into the set value changing means for operation and a manual operation mode in which the recognition information is manually given to the set value changing means based on image information is switched to the operation mode. A switching means is provided. Further, learning / evaluation data storage means for storing the image information input while operating in the manual operation mode and the recognition information given by the operator at that time, and the learning / evaluation having the same configuration as for the real line. Learning / evaluation in which the data stored in the learning / evaluation data storage means is introduced into the learning / evaluation neural network for learning, and the connection weights as learning results are transferred to the real line neural network. The management means, the function of monitoring at least the image information and the recognition information of the real line neural network, and the operation of the operation mode switching means, the learning / evaluation data storage means, and the learning / evaluation management means in response to a request from the operator. The operation management means for managing is provided.

[0012]

[Function] Since the operation mode switching means for switching between the automatic operation mode and the manual operation mode is provided, even if a problem occurs in the image information processing system or the real line neural network, the operation is immediately switched to the manual operation mode. I can continue. Also, the operation mode is switched to the manual operation mode, the input / output data at that time is taken into the learning / evaluation data storage means, and the learning / evaluation neural network is made to learn by using this data, and the connection weight as the learning result is used for the real line. Since it can be transferred to a neural network, the same control as when an experienced operator controls it can be performed automatically.

[0013]

Embodiments will be described below with reference to the drawings. FIG. 1 shows a control device according to an embodiment of the present invention.

In this control device, the ITV shown in FIG.
The image information Aa and Ab obtained by 17a and 17b are introduced into the pre-processing unit 21, and the pre-processing unit 21 extracts the contour,
It is introduced into the real line neural network 22 after being subjected to noise reduction processing.

Real line neural network 22
Is composed of two systems, a system for determining the mountain shape of the charbet 3 and a system for determining the suspended matter distribution in the upper part of the combustion furnace 2. The system 22a for determining the mountain shape of the charbed 3 includes an input layer 23, an intermediate layer 24, and an output layer 25, as shown in FIG. Then, the contour data Aaa indicating the mountain shape of the charbed 3 processed by the preprocessing unit 21 is introduced into the input layer 23, which pattern the contour belongs to is analyzed, and the analysis result is output layer 25.
To "0" and "1" in combination. In this example, as shown in FIGS. 3 (a) to 3 (g), seven reference patterns are predetermined and it is determined which of these patterns they belong to. In the recovery boiler, the pattern that can maximize the recovery rate of chemicals is shown in Fig. 3.
This is the pattern shown in (a).

As shown in FIG. 4, the system 22b for determining the distribution of suspended solids in the upper part of the combustion furnace 2 has an input layer 26,
The intermediate layer 27 and the output layer 28 are provided. Then, the image data Sbb indicating the distribution state of the floating matter processed by the preprocessing unit 21 is introduced into the input layer 26, which pattern the distribution state of the floating matter belongs to is analyzed, and the analysis result is output layer 28. To "0" and "1" in combination. In this example, as shown in FIGS. 5 (a) to 5 (c), three reference patterns are predetermined, and it is determined which of these patterns belongs. In the recovery boiler, the pattern that can maximize the exchange efficiency of the heat exchanger is the pattern shown in Fig. 5 (a).

The recognition information Ba, Bb output from the real line neural network 22 is given to the set value changing unit 30 via the changeover switch 29. As shown in FIG. 6, the set value changing unit 30 includes a fuzzy inference unit 31 and an air flow rate setting unit 32.

The fuzzy inference unit 31 recognizes the recognition information Ba, B.
b and SO measured by the concentration measuring device 15 shown in FIG.
₂ , the concentration signal indicating NOx and the adjustment signal P from the parameter adjuster 33 are introduced, the concentration of SO ₂ , NOx is below the reference value, the chemical recovery rate can be maximized, and the heat exchange efficiency can be maximized. A possible set value X of air flow distribution and a set value Y of black liquor temperature are calculated. Then, the set value X is given to the air flow rate setting unit 32.

The air flow rate setting unit 32 introduces the set value X and the black liquor flow rate signal F measured by the flow meter 16 to obtain the flow rate shown in FIG.
Primary low air flow 1 blown from the blower 5 shown in FIG.
The flow rates of L, primary high air flow 1H, secondary low air flow 2L, and secondary high air flow 2H are determined. The flow rate setting signal and the black liquor temperature setting signal determined in this way are transferred to the changeover switch 3
4 to the blower 5 and the black liquor temperature adjuster 18 shown in FIG.

On the other hand, reference numeral 35 in FIG. 1 denotes an automatic operation management unit, 36 denotes a console under the control of an operator, 37 denotes a learning / evaluation storage unit, and 38 denotes a real line neural network. 22 shows a learning / evaluation neural network having the same configuration as that of 22, where 39 is a learning / evaluation neural network.
The evaluation management part is shown.

The automatic driving management unit 35 is provided with the image information Aa, A
b, the image data processed by the preprocessing unit 21, the recognition information Ba and Bb of the real line neural network 22, and the switches 29 and 34 on the console 36 side, that is, automatically in response to a request from the operator. Switching control from the operation mode to the manual operation mode, storage (erasure) control of input / output information in the learning / evaluation data storage unit 37 in response to a request from an operator, learning / evaluation management unit 39
To issue an operation command, etc., and specifically operates as described later. It should be noted that reference numerals 40, 41, and 42 in FIG. 1 denote switches that operate according to a command from the automatic driving management unit 35.
Next, the operation of the control device configured as described above will be described with reference to the flow diagrams shown in FIGS. 7 to 11.

First, when the real line neural network 22 is not learned (f1), the operation is performed in the manual operation mode. At this time, the automatic driving management unit 35 sets the changeover switch 29 to the console 36 side according to the instruction of the operator.
Switch the changeover switch 34 to the set value changer 30 side (f
2).

The operator sets recognition information by referring to the image of the current charbet 3 or floating object displayed on the display section of the automatic operation management section 35 and the image data processed by the preprocessing section 21. It is given to the value changing unit 30.

This series of input / output data is stored and held as learning material in the learning / evaluation storage unit 37 in response to a command from the automatic driving management unit 35 (f3). Input / output data for evaluation is stored in advance in the learning / evaluation storage unit 37.

When a sufficient number of input / output data are stored and held in the learning / evaluation storage unit 37 (f4), the automatic driving management unit 35 notifies the operator that learning can be started.
(f5). In response to this, the operator instructs to start learning.
(f6), the automatic driving management unit 35 sends a learning execution command to the learning / evaluation management unit 39.

The learning / evaluation management unit 39 gives the learning input data (output of the preprocessing unit 21) of the data stored in the learning / evaluation storage unit 37 to the learning / evaluation neural network 38, and at the same time. The learning output data (recognition result by the operator) is given as a teacher signal for learning.

When the learning is completed, the fact that the learning is completed is notified to the automatic driving management section 35. On the contrary,
The automatic driving management unit 35 copies the connection weight, which is the learning result of the learning / evaluating neural network 38, to the operator for the real line neural network 22, or evaluates the learning result of the learning / evaluating neural network 38. It is instructed to select either (1) or hold the learning result as it is (f7).

If the operator selects "hold", the result of the learned connection weight is held and the process is terminated (f8). When "evaluation" is selected, the input data of the evaluation data held in the learning / evaluation data storage unit 37 is given to the learning / evaluation neural network 38, and the recognition result is learned / evaluated. It is sequentially displayed on the automatic driving management unit 35 together with the input data and the output data via the evaluation management unit 39.

If learning is insufficient with reference to the evaluation result, the operator appropriately instructs the re-learning by stopping the evaluation by designating the evaluation data with poor recognition result as the learning data. Or (f9, f10, f11). If it is determined that the learning is sufficient, "copy" is selected (f12). When "copy" is instructed, the connection weight of the learning / evaluation neural network 38 is changed through the learning / evaluation management unit 39 to the real line neural network 2
It is copied to that of 2 (f13). When this copying is completed, a display to that effect is displayed on the automatic operation management unit 35, and the input / output display of the automatic operation management unit 35 is switched to that of the real line neural network 22. The initial setting is completed by the steps so far.

Next, the operator checks the recognition result of the real line neural network 22 with respect to the actual pre-processing result while checking the recognition capability (f14) while confirming the content displayed on the automatic operation management unit 35. When the recognition ability is insufficient, the input / output data having a poor recognition result is stored and held in the learning / evaluation data storage unit 37 in the same manner as described above (f15, f16). At this time, conversely, it is possible to delete the learning data currently stored and held as inappropriate. Even if this deletion is performed, if new learning data for that amount is additionally registered, relearning (f17, f1
8) becomes possible. Further, the operator can give the re-learning command again to the automatic operation management unit 35 (f20) at the time when the operator finishes checking the recognition ability of the real line neural network 22 (f19). At this time,
As in the previous case, the automatic driving management unit 35 uses the learning / evaluation management unit 39 to learn / evaluate the neural network 38.
Is given a re-learning command, and the result is fed back to the real line neural network 22. This cycle of “re-learning ← → evaluation of evaluation data → evaluation of actual data → re-learning” is repeated as necessary, and finally it is determined that the real line neural network 22 has acquired sufficient recognition ability. At this stage, the operator instructs the automatic operation management unit 35 to switch from the manual operation mode to the automatic operation mode (f21). Note that the process up to this point is referred to as an operation mode level 1 for convenience.

In response to a command to switch to the automatic operation mode, the automatic operation management unit 35 switches the changeover switch 29 from the console 36 side to the real line neural network 22 side. Therefore, the automatic operation mode (operation mode level 2) is switched from this point.

When the automatic operation mode is entered, the real line neural network 22 recognizes the image data Aaa, Bbb output from the preprocessor 21 online, and the recognition information Ba, Bb is set via the changeover switch 29. It is given to the value changing unit 30. Therefore, the system will operate autonomously. At this time, the image information Aa, A
b, preprocessed image data Aaa, Bbb, recognition information Ba, Bb of the real line neural network 22.
Is shown to the operator through the display unit of the automatic driving management unit 35. Therefore, the operator can check whether or not there is any modulation in the automatic driving centering on the recognition status of the real line neural network 22.

Further, the automatic operation management unit 35 has a function of monitoring the recognition result of the real line neural network 22, and has a function of issuing an alarm to the operator to prompt a check when the recognition situation is unstable. (F22, f
twenty three) . As a method of determining whether or not the recognition situation is unstable, (1) a large number of units in the output unit group of the real line neural network 22 output the same level of output, (2) The output state of the real line neural network 22 is examined such that a plurality of output units corresponding to patterns having a low degree of similarity with each other in terms of goodness of fit output a large output.

The above (1) and (2) correspond to the errors of the first and second types, and in the case of (1), the classification and recognition are insufficient because of insufficient recognition and classification ability. It is a state in which no decision can be made, and in the case of (2), it indicates that there is a possibility that incorrect classification remains in the learning process for acquiring cognitive ability.

In response to the occurrence of such an alarm, the operator issues an "input / output data storage request" to learn the input / output data or image information determined to be an unstable recognition result for learning or evaluation. Re-learning of the learning / evaluation neural network 37, which is a copy of the real-line neural network 38, is stored as a background job for online recognition by holding it in the evaluation storage unit 37 and further issuing a "re-learning execution request". Do (f24)
The learning result can be fed back to the real line neural network 22 (that is, the connection weight can be replaced).

Further, when the recognition state of the real line neural network 22 is seriously deteriorated and there is a problem in continuing the automatic operation, a "operation mode switching request" is instructed (f2
5) It is also possible to cancel the automatic operation mode, return to the manual operation mode, and then relearn the real line neural network 22 by a "relearning execution request" command.

The automatic operation management unit 35 includes a set value changing unit 30.
Is also displayed, and the operator can further issue a command to switch to the manual operation mode (operation mode level 0) when necessary according to the setting contents. Upon receiving this command, the automatic driving management unit 35 switches the switch 34 from the set value changing unit 30 side to the console 36 side. Therefore, the operator can operate in a mode in which the set value is directly given.

By repeating the adjustment centering on the re-learning of the real line neural network 22 against the background of the operation including the operation mode switching, the real line neural network 22 finally obtains the recognition processing capability. As a result, long-term self-sustaining operation is possible.

[0039]

As described above, according to the present invention, the recovery boiler can be automatically operated with a high chemical recovery rate and a high heat exchange efficiency while taking environmental problems into consideration.

[Brief description of drawings]

FIG. 1 is a block configuration diagram of a control device according to an embodiment of the present invention,

FIG. 2 is a schematic diagram of a chevette mountain shape recognition system in a real line neural network incorporated in the control device;

FIG. 3 is a diagram showing an example of a mountain pattern classification pattern of charvet,

FIG. 4 is a schematic diagram of a suspended matter distribution recognition system in a real line neural network incorporated in the control device;

FIG. 5 is a view showing an example of a classification pattern of floating material distribution,

FIG. 6 is a block configuration diagram of a set value changing unit incorporated in the control device;

FIG. 7 is a flow chart for explaining the operation of the control device,

FIG. 8 is a flow chart for explaining the operation of the control device,

FIG. 9 is a flow chart for explaining the operation of the control device,

FIG. 10 is a flow chart for explaining the operation of the control device,

FIG. 11 is a flowchart for explaining the operation of the control device,

FIG. 12 is a diagram for explaining a conventional control method in a recovery boiler.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Black liquor gun 2 ... Combustion furnace 3 ... Charvet 5 ... Blower 7 ... Chemicals 8-11 ... Heat exchanger 15 ... Concentration measuring instrument 16 ... Flowmeter 17a, 17b ... ITV camera 18 ... Black liquor temperature regulator 21 ... Pre-processing unit 22 ... Real line neural network 29, 34 ... Changeover switch 30 ... Setting value changing unit 35 ... Automatic operation management unit 36 ... Console 37 ... Learning / evaluation storage unit 38 ... Learning / evaluation neural network 39 ... Learning and evaluation management department.

Claims (1)

[Claims] 1. A control unit for controlling a recovery boiler that recovers thermal energy by burning black liquor containing a combustible component generated in a paper pulp manufacturing process, and supplies it to a combustion furnace. A temperature adjusting means for adjusting the temperature of the black liquor, an air amount adjusting means for adjusting the flow rate of the combustion air supplied to each part in the combustion furnace, a means for obtaining image information in the combustion furnace, and By analyzing the image information obtained by the means, it is output by recognizing which of the multiple patterns the shape of the charbed deposited on the bottom of the furnace and the distribution of suspended matter in the upper part of the furnace belong to Real line neural network, concentration measuring means for measuring the concentration of combustion exhaust gas components, flow rate measuring means for measuring the flow rate of black liquor supplied to the combustion furnace, the flow rate measuring means and the concentration measurement Input the measurement information obtained in the step and the recognition information obtained by the real line neural network, and set values of the temperature adjusting means and the air amount adjusting means to a value that can approach optimum operation based on fuzzy reasoning. The setting value changing means for changing the setting value, the automatic operation mode in which the recognition information obtained by the real line neural network is introduced into the setting value changing means, and the setting value changing means is manually operated based on the image information. For learning / evaluation, which stores operation mode switching means for switching to a manual operation mode for giving recognition information to the user, image information input when operating in the manual operation mode, and recognition information given by the operator at that time. Data storage means, a learning / evaluation neural network having the same configuration as the real line neural network, and the learning / evaluation neural network Learning / evaluation management means for introducing the data stored in the valuation data storage means into the learning / evaluation neural network for learning, and transferring the connection weight as a learning result to the real line neural network; The operation mode switching means, the learning / evaluation data storage means, the learning / monitoring means for monitoring the image information and the recognition information of the real line neural network and responding to a request from an operator.
A recovery boiler control device comprising: an operation management unit that manages the operation of the evaluation management unit.