Control device, remote monitoring system, and control method for a wet flue gas desulfurization system.

Abstract

To provide a control apparatus improved in table accuracy for determining a control target value of a wet flue-gas desulfurization device.SOLUTION: A control apparatus 15 in a wet flue-gas desulfurization of a combustion device exhaust gas comprises: a learning model construction unit; a predicted value correction unit; a table generation unit; and a control target value determination unit. The learning model construction unit constructs a learning model for a relationship between an explanatory variable including a load of a combustion device and sulfur dioxide concentration at an absorption tower outlet which is an objective variable. The predicted value correction unit corrects a predicted value acquired through the learning model on the basis of a difference from an actual measured value. The table generation unit generates a table showing a relationship between the load of the combustion device and an absorbent concentration target value and an absorbent circulation quantity target value for satisfying a reference value by a corrected predicted value. The control target value determination unit calculates the absorbent concentration target value on the basis of the table to determine control target values of an input amount of an absorbent and a circulation flow rate of an absorbent. The predicted value correction unit includes a first connection part for correcting the predicted value on the basis of the load.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present disclosure relates to a control device for a wet flue gas desulfurization device, a remote monitoring system, and a control method.

Background Art

[0002] In a wet flue gas desulfurization device, exhaust gas generated by a combustion device such as a boiler is introduced into an absorption tower of the desulfurization device and brought into gas-liquid contact with an absorption liquid circulating in the absorption tower. In the process of gas-liquid contact, the absorbent (for example, calcium carbonate) in the absorption liquid reacts with sulfur dioxide (SO2) in the exhaust gas, so that SO2 in the exhaust gas is absorbed by the absorption liquid and SO2 is removed from the exhaust gas (the exhaust gas is desulfurized). On the other hand, the absorption liquid that has absorbed SO2 falls and is stored in a storage tank below the absorption tower. An absorbent is supplied to the storage tank, and the absorption liquid whose absorption performance has been restored with the supplied absorbent is supplied above the absorption tower by a circulation pump and used for gas-liquid contact (absorption of SO2) with the exhaust gas. Since the circulation pump for circulating the absorption liquid consumes a large amount of power, conventionally, for the purpose of suppressing power consumption, the circulation flow rate of the absorption liquid required based on the flow rate of the exhaust gas flowing into the absorption tower, the SO2 concentration in the exhaust gas, etc. is calculated, and the number of operating units of the circulation pump is controlled.

[0003] Patent Document 1 discloses a technique for appropriately adjusting the operating conditions of a circulation pump for circulating an absorption liquid in an absorption tower of such a wet flue gas desulfurization device. In this document, using the operating data obtained from a combustion device such as a boiler and a wet flue gas desulfurization device, the correlation between the operating data and the SO2 concentration at the outlet of the absorption tower, and the correlation between the operating data and the absorbent concentration contained in the absorption liquid are respectively modeled by machine learning, and the circulation flow rate of the absorption liquid and the absorbent concentration are controlled to be optimized based on the tables obtained by these two learning models.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

[0005] In the above-mentioned Patent Document 1, the control target values ​​for the circulating flow rate of the absorbent solution and the absorbent concentration are calculated based on a table obtained by a learning model. Therefore, if the prediction accuracy of the learning model decreases due to some factor (if the predicted values ​​of the learning model deviate from the measured values), the control target values ​​calculated based on the table may deviate from the optimal values, potentially leading to a decrease in control accuracy. To suppress such a decrease in control accuracy, it is conceivable to improve the accuracy by rebuilding the learning model through retraining when the prediction accuracy of the learning model decreases. However, retraining a learning model requires advanced processing power from the processing unit such as the computer that performs the calculations, resulting in high implementation costs. Another method to improve the prediction accuracy of the learning model is to perform correction processing on the predicted values ​​of the learning model without retraining the learning model. However, the accuracy of the predicted values ​​obtained using the learning model depends on the state of the wet flue gas desulfurization system. Therefore, even when improving prediction accuracy by correcting the predicted values ​​of the learning model, how to perform the correction processing remains a challenge.

[0006] At least one embodiment of this disclosure has been made in view of the above circumstances and aims to provide a control device, remote monitoring system, and control method for a wet flue gas desulfurization apparatus that can improve the accuracy of a table for determining control target values ​​by correcting the predicted values ​​of a learning model. [Means for solving the problem]

[0007] A control device for a wet flue gas desulfurization apparatus according to at least one embodiment of the present disclosure solves the above problems, A control device for a wet flue gas desulfurization system that performs desulfurization by bringing exhaust gas generated in a combustion device into gas-liquid contact with an absorbent liquid inside an absorption tower, A learning model construction unit for constructing a learning model by machine learning regarding the relationship between at least one parameter included in the operating data of at least one of the combustion apparatus or the wet flue gas desulfurization apparatus, which is an explanatory variable that includes at least the load of the combustion apparatus, and a target variable that is the sulfur dioxide concentration at the future outlet of the absorption tower. A prediction value correction unit for correcting the predicted value of the sulfur dioxide concentration by the learning model using a correction coefficient calculated based on the difference between the predicted value and the measured value of the sulfur dioxide concentration at the outlet of the absorption tower, A table creation unit for creating a table showing the relationship between the load of the combustion device and the target value of the absorbent concentration and the target value of the absorbent circulation amount of the absorbent liquid so that the corrected predicted value, corrected by the predicted value correction unit, satisfies the reference value. A control target value determination unit calculates the target absorbent concentration value and the target absorbent circulation rate value corresponding to the operating data based on the table, and determines the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution corresponding to the target absorbent concentration value and the target absorbent circulation rate value, Equipped with, The predicted value correction unit includes a first correction unit that corrects the predicted value using a first correction coefficient calculated based on the load as the correction coefficient.

[0008] A control method for a wet flue gas desulfurization apparatus according to at least one embodiment of the present disclosure solves the above problems. A control method for a wet flue gas desulfurization system that performs desulfurization by bringing exhaust gas generated in a combustion device and an absorbent liquid into gas-liquid contact within an absorption tower, A step of constructing a learning model by machine learning regarding the relationship between at least one parameter included in the operating data of at least one of the combustion apparatus or the wet flue gas desulfurization apparatus, which is an explanatory variable that includes at least the load of the combustion apparatus, and a target variable that is the sulfur dioxide concentration at the future outlet of the absorption tower. A step of correcting the predicted value of the sulfur dioxide concentration by the learning model using a correction coefficient calculated based on the difference between the predicted value and the measured value of the sulfur dioxide concentration at the outlet of the absorption tower, A step of creating a table showing the relationship between the load of the combustion device and the target value of the absorbent concentration and the target value of the absorbent circulation amount of the absorbent liquid so that the corrected predicted value, corrected by the predicted value correction unit, satisfies the reference value. A step of calculating the target absorbent concentration value and the target absorbent circulation volume value corresponding to the operating data based on the table, and determining the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution corresponding to the target absorbent concentration value and the target absorbent circulation volume value, Equipped with, In the step of correcting the predicted value, the predicted value is corrected using a first correction coefficient calculated based on the load as the correction coefficient. [Effects of the Invention]

[0009] According to at least one embodiment of this disclosure, a control device, remote monitoring system, and control method for a wet flue gas desulfurization apparatus can be provided that improve the accuracy of a table for determining control target values ​​by correcting the predicted values ​​of a learning model. [Brief explanation of the drawing]

[0010] [Figure 1] This is a diagram illustrating the configuration of a wet flue gas desulfurization apparatus according to one embodiment. [Figure 2] Figure 1 is a block diagram of the control device. [Figure 3] Figure 2 is a processing flow diagram of the requirements determination unit. [Figure 4] Figure 2 is a processing flow diagram of the first correction condition determination unit. [Figure 5] Figure 2 is a processing flow diagram of the load-specific correction signal generation unit. [Figure 6] This is a processing flow diagram of the first load-specific correction coefficient calculation unit in the load-specific correction coefficient calculation unit shown in Figure 2. [Figure 7] Figure 2 is a processing flow diagram of the first correction coefficient calculation unit. [Figure 8] Figure 2 is a processing flow diagram of the second correction condition determination unit. [Figure 9] Figure 2 is a processing flow diagram of the second correction coefficient calculation unit. [Figure 10] It is a flowchart of the table creation unit in FIG. 2. [Figure 11] It is a flowchart of the correction of the control target value by the control target value determination unit in FIG. 2. [Figure 12] It is a flowchart of the circulation pump adjustment unit in FIG. 2. [Figure 13A] It is a block configuration diagram showing another aspect of FIG. 2. [Figure 13B] It is a block configuration diagram showing another aspect of FIG. 2.

Mode for Carrying Out the Invention

[0011] Hereinafter, some embodiments of the present invention will be described with reference to the drawings. However, the scope of the present invention is not limited to the following embodiments. The dimensions, materials, shapes, relative arrangements, etc. of the configurations described in the following embodiments are not intended to limit the scope of the present invention thereto, but are merely illustrative examples.

[0012] FIG. 1 is a configuration diagram of a wet flue gas desulfurization apparatus 10 according to an embodiment. The wet flue gas desulfurization apparatus 10 is a device for desulfurizing the exhaust gas generated in the combustion apparatus 1. The combustion apparatus 1 is, for example, a boiler for generating steam. The steam generated in the combustion apparatus 1 is supplied to, for example, a steam turbine (not shown), and when the steam turbine is driven by the steam, electricity is generated by a generator 5 connected to the output shaft of the steam turbine. The wet flue gas desulfurization apparatus 10 includes an absorption tower 11 that is in communication with the combustion apparatus 1 via piping 2, a plurality of circulation pumps 12a, 12b, 12c, ... (in Figure 1, three circulation pumps are typically shown as examples, but the number is not limited. Also, when referring to them collectively, they will be appropriately called "circulation pumps 12") provided in the circulation piping 3 for the absorbent liquid circulating inside the absorption tower 11, an absorbent slurry supply unit 13 for supplying a slurry of calcium carbonate (CaCO3), which is an absorbent contained in the absorbent liquid, into the absorption tower 11, and a gypsum recovery unit 14 for recovering gypsum in the absorbent liquid. The absorption tower 11 is equipped with an outlet pipe 16 for the exhaust gas, which has been desulfurized by the operation described later, to flow out of the absorption tower 11 as an outflow gas, and the outlet pipe 16 is equipped with a gas analyzer 17 for measuring the SO2 concentration in the outflow gas.

[0013] The absorbent slurry supply unit 13 includes an absorbent slurry manufacturing facility 21 for manufacturing absorbent slurry, an absorbent slurry supply pipe 22 connecting the absorbent slurry manufacturing facility 21 and the absorption tower 11, and an absorbent slurry supply amount control valve 23 for controlling the flow rate of absorbent slurry flowing through the absorbent slurry supply pipe 22. The gypsum recovery unit 14 includes a gypsum separator 25, a gypsum slurry extraction pipe 26 connecting the gypsum separator 25 and the absorption tower 11, and a gypsum slurry extraction pump 27 provided in the gypsum slurry extraction pipe 26.

[0014] The wet flue gas desulfurization apparatus 10 is equipped with a control device 15 for controlling the wet flue gas desulfurization apparatus 10. The control device 15 acquires various operating data from the combustion apparatus 1 and the wet flue gas desulfurization apparatus 10 (for example, temperature and pressure at various parts, flow rates of various fluids, etc.) and performs various controls.

[0015] The hardware configuration of the control device 15 consists, for example, a CPU (Central Processing Unit), RAM (Random Access Memory), ROM (Read Only Memory), and a computer-readable storage medium. A series of processes for realizing various functions are stored in the storage medium in the form of a program, for example. The CPU reads this program into the RAM and performs information processing and calculations to realize the various functions. The program may be pre-installed in ROM or other storage media, provided in a state where it is stored in a computer-readable storage medium, or distributed via wired or wireless communication. Computer-readable storage media include magnetic disks, magneto-optical disks, CD-ROMs, DVD-ROMs, semiconductor memory, etc.

[0016] Figure 2 is a block diagram of the control device 15 in Figure 1. In this example, the control device 15 is composed of a higher-level control device 15A and a lower-level control device 15B. The higher-level control device 15A is the main unit that constitutes the control device 15, and each functional block is implemented by being described by a program. The lower-level control device 15B is a control device that is under the management of the higher-level control device 15A, and is, for example, a distributed control system (DCS) for each piece of equipment that constitutes the wet flue gas desulfurization system 10 (see Figure 1), and each functional block is implemented by being described by control logic. These higher-level control device 15A and lower-level control device 15B can communicate with each other and cooperate to control the wet flue gas desulfurization system 10.

[0017] The control device 15 includes an operation data receiving unit 30, a learning model construction unit 38, a table creation unit 31, a predicted value correction unit 36, a control target value determination unit 32, a circulation pump adjustment unit 33, and an absorbent slurry supply control unit 34. Each of these functional blocks of the control device 15 can send and receive data from each other and cooperate to realize the control method described later. In this embodiment, of these functional blocks, the operation data receiving unit 30, the learning model construction unit 38, and the table creation unit 31 are located in the higher-level control device 15A, while the predicted value correction unit 36, the control target value determination unit 32, the circulation pump adjustment unit 33, and the absorbent slurry supply control unit 34 are located in the lower-level control device 15B.

[0018] Furthermore, the control device 15 includes a relearning device 15C in addition to the upper-level control device 15A and the lower-level control device 15B. The relearning device 15C is configured to improve the prediction accuracy of the learning model M constructed by the learning model construction unit 38 by relearning the learning model M based on a relearning algorithm when the prediction accuracy of the learning model M deteriorates. When the prediction accuracy of the learning model M deteriorates, this can be addressed by correcting the predicted value Vp by the predicted value correction unit 36, as described later. However, if the correction of the predicted value Vp by the predicted value correction unit 36 ​​is insufficient, the learning model M can be relearned by the relearning device 15C. Since such relearning is not performed frequently, providing the relearning device 15C as a separate configuration from the upper-level control device 15A and the lower-level control device 15B effectively reduces the processing load on the upper-level control device 15A and the lower-level control device 15B.

[0019] The operation data receiving unit 30 is configured to receive operation data acquired by the operation data acquisition unit 20. This operation data includes various parameters acquired by the combustion device 1 and the wet flue gas desulfurization device 10, and in particular includes at least the load of the combustion device 1 and the SO2 concentration at the outlet of the absorption tower measured by the gas analyzer 17.

[0020] The learning model construction unit 38 is configured to construct a learning model M using machine learning to understand the relationship between various operational data received by the operational data reception unit 30 and the future SO2 concentration at the absorption tower outlet. The learning model M is constructed as a regression model using regression methods such as multiple regression, ridge regression, lasso regression, or Elastic Net. In this embodiment, as an example of the learning model M, a regression model expressed as a linear polynomial is constructed as shown in the following equation. SO2 concentration at the outlet of the absorption tower = k1 × explanatory variable 1 + k2 × explanatory variable 2 + ... + kn × explanatory variable n + b (1) By using a linear polynomial as the learning model M in this way, the explainability (interpretability) is higher compared to complex simulation models, and the computational load can be effectively reduced. Note that n is an arbitrary natural number, k1 to kn are coefficients, and b is an arbitrary intercept.

[0021] The learning model M obtained by machine learning is constructed as a model that shows the correlation between explanatory variables, which consist of multiple parameters included in the operating data received by the operating data receiving unit 30, and the future SO2 concentration at the absorption tower outlet, which is the dependent variable. Here, the combination of multiple parameters included in the explanatory variables of the learning model M can be arbitrarily selected from the following candidates. i) Output command value for generator 5 (external output command value) ii) Output of generator 5 iii) Supply air flow rate to combustion device 1 or exhaust gas flow rate from combustion device 1 iv) SO2 concentration at the inlet of absorption tower 11 or SO2 concentration at the outlet of combustion device 1 v) SO2 concentration at the outlet of absorption tower 11 vi) CaCO3 concentration or pH of the absorption solution vii) Control value of the number of operating circulation pumps 12 or the discharge flow rate viii) O2 concentration at the outlet of combustion device 1 or at the inlet of absorption tower 11 x) Desulfurization rate in absorption tower 11 (= 100% - (sulfur concentration at outlet of absorption tower 11 / sulfur concentration at inlet of absorption tower 11) × 100%) These candidates are parameters that can typically be measured conventionally in many wet flue gas desulfurization systems 10, and may also include other parameters.

[0022] In this embodiment, the explanatory variables of the learning model M are selected from the above candidates to include at least one of the following: i) the output command value for the generator 5 (external output command value), iii) the supply air flow rate to the combustion device 1 or the exhaust gas flow rate from the combustion device 1, or iv) the SO2 concentration at the inlet of the absorption tower 11 or the SO2 concentration at the outlet of the combustion device 1. More preferably, the explanatory variables of the learning model are selected from the above candidates to include iii) the supply air flow rate to the combustion device 1 or the exhaust gas flow rate from the combustion device 1, and iv) the SO2 concentration at the inlet of the absorption tower 11 or the SO2 concentration at the outlet of the combustion device 1. This makes it possible to construct a learning model M that can predict the future SO2 concentration at the outlet of the absorption tower, which is the objective function, with good accuracy. Furthermore, by selecting some parameters as explanatory variables from the operation data received by the operation data receiving unit 30 in this way, the data to be learned can be efficiently narrowed down, and the computational burden of machine learning can be reduced.

[0023] The table creation unit 31 is configured to create a table Tb that defines the relationship between the output of the generator 5 (load of the combustion device 1) and the target value of the absorbent concentration of the absorbent liquid so that the predicted value Vp of the future SO2 concentration at the outlet of the absorption tower meets the standard value, using the learning model M constructed by the learning model construction unit 38. The specific method of creating table Tb by the table creation unit 31 will be described later, but the accuracy of table Tb created by the table creation unit 31 depends on the accuracy of the predicted value Vp calculated based on the learning model M. Therefore, if the prediction accuracy of the learning model M is low (i.e., there is a discrepancy between the predicted value Vp and the measured value Vm), the prediction value correction unit 36 ​​corrects the predicted value Vp of the learning model M. By creating table Tb using the predicted value Vp (hereinafter referred to as "corrected predicted value Vp'"), the table creation unit 31 can improve the accuracy of calculating the control target value.

[0024] The control target value determination unit 32 is configured to determine the target absorbent concentration value corresponding to the load included in the operation data received by the operation data receiving unit 30, based on the table Tb created by the table creation unit 31, and to determine the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution, respectively, corresponding to the target absorbent concentration value.

[0025] Furthermore, the circulation pump adjustment unit 33 is configured to control the circulation pumps 12a to 12c based on the control target value for the circulation flow rate of the absorbent solution determined by the control target value determination unit 32. The absorbent slurry supply control unit 34 is configured to control the amount of absorbent supplied based on the control target value for the amount of absorbent to be added determined by the control target value determination unit 32.

[0026] Next, the configuration of the predicted value correction unit 36 ​​will be explained in detail. The predicted value correction unit 36 ​​includes a necessity condition determination unit 39 for determining whether or not correction processing is necessary for the predicted value Vp of the learning model M. The necessity condition determination unit 39 is configured to determine whether or not correction is necessary for the predicted value Vp of the learning model M constructed by the learning model construction unit 38. The learning model M contains a certain amount of prediction error, and if the difference ΔV between the predicted value Vp and the measured value Vm becomes large, the predicted value Vp by the learning model M is corrected by at least one of the first correction unit 40 or the second correction unit 50, which will be described later. The necessity conditions are defined as conditions for determining whether or not such correction processing is necessary for the predicted value Vp of the learning model M.

[0027] Figure 3 is a processing flow diagram of the requirement determination unit 39 in Figure 2. In Figure 3, as an example of the configuration of the requirement determination unit 39, the processing flow is shown when a determination is made based on the following seven requirements. Specifically, the requirement determination unit 39 determines that the requirements are met when all of the following seven conditions (conditions 1 to 7) are satisfied. (Condition 1) The measured value Vm (SO2 concentration at the outlet of the absorption tower) corresponding to the predicted value Vp of the learning model M is greater than the preset reference value Vmref. (Condition 2) The load L of the combustion device 1 is greater than the preset reference value Lref. (Condition 3) There are no abnormalities in the parameters that are correlated with the explanatory variables of the learning model M. (Condition 4) The bypass damper is in the closed position. (Condition 5) The wet flue gas desulfurization unit 10 is in a state where smoke is flowing. (Condition 6) There is no abnormality in the upper-level control unit 15A. (Condition 7) The initialization of the control device 15 is completed.

[0028] Condition 1 is a condition for determining whether it is necessary to correct the predicted value Vp of the learning model M based on whether the measured value Vm of the SO2 concentration at the outlet of the absorption tower, which is the target of prediction by the learning model M, is greater than the reference value Vref. Condition 2 is a condition for determining whether it is necessary to correct the predicted value Vp of the learning model M based on whether the load L of the combustion device 1 is greater than the reference load value Lref. Furthermore, since conditions 1 and 2 are similar in nature and serve to determine whether or not correction is necessary, either one may be omitted.

[0029] Condition 3 is a condition for determining whether the operating conditions are suitable for correcting the predicted value Vp of the learning model M. The "parameters correlated with the explanatory variables of the learning model M" may be, for example, parameters included in the explanatory variables (such as the SO2 concentration at the outlet of the absorption tower, the exhaust gas flow rate of the combustion device 1, and the absorbent concentration), or they may be parameters used to monitor abnormalities in the measuring instruments used to measure these parameters.

[0030] Condition 4 is a condition for determining whether the bypass damper of the wet flue gas desulfurization unit 10 is in the closed state. The bypass damper is a damper valve that bypasses the boiler exhaust gas directly to the chimney without passing it through the desulfurization unit. It is in the closed state when the power plant is in normal operation, but in emergencies, such as when some kind of abnormality occurs, it is configured to be fully opened to protect the equipment of the desulfurization unit. Note that in recent years, some plants are not equipped with bypass dampers, in which case Condition 4 can be omitted from the requirements.

[0031] Condition 5 is a condition for determining whether the wet flue gas desulfurization unit 10 is in a flue gas state, and is determined, for example, based on whether the desulfurization fan of the wet flue gas desulfurization unit 10 is in operation, or whether the IDF is started.

[0032] Condition 6 is a condition for determining whether or not there is a malfunction in the upper-level control unit 15A of the control unit 15, and condition 7 is a condition for determining whether or not the initialization of the control unit 15 has been completed.

[0033] The requirement determination unit 39 determines that the requirement conditions have been met if all of these conditions are satisfied, and turns on the requirement condition fulfillment flag. When the requirement condition fulfillment flag is turned on, the first correction unit 40 and the second correction unit 50 can begin correcting the predicted value Vp.

[0034] The first correction unit 40 is configured to correct the predicted value Vp from the learning model M according to the load L of the combustion device 1. The first correction unit 40 comprises a first correction condition determination unit 41, a load-specific correction signal generation unit 42, a load-specific correction coefficient calculation unit 43, and a first correction coefficient calculation unit 44. The configurations of the first correction unit 40 will be described in order below with reference to Figures 4 to 7.

[0035] Figure 4 is a processing flow diagram of the first correction condition determination unit 41 in Figure 2. The first correction condition determination unit 41 is configured to determine whether the first correction condition is met. The first correction condition is a condition for determining whether or not to perform the first correction by the first correction unit 40 on the predicted value Vp of the learning model M. In Figure 4, as an example of the first correction condition, the determination is made based on the following four conditions. (Condition 8) The difference ΔV between the corrected predicted value Vp' (predicted value Vp before initial correction) and the measured value Vm is greater than the reference value ΔVref. (Condition 9) The flag indicating that the necessary condition has been met must be ON. (Condition 10) The first correction frequency signal Sa1 is input. (Condition 11) The first correction permission flag is ON.

[0036] Under condition 8, the absolute value (ABS) of the difference ΔV between the corrected predicted value Vp' (which is the predicted value Vp of the learning model M before the initial correction) and the measured value Vm is calculated, and it is determined whether this absolute value is greater than the reference value ΔVref. In Figure 4, the measured value Vm, which is compared with the corrected predicted value Vp' of the learning model M, is corrected for measurement delay. Measurement delay correction is a process to compensate for the time lag (measurement delay) between the corrected predicted value Vp' and the measured value Vm obtained by the gas analyzer 17, because obtaining the measured value Vm by the gas analyzer 17 takes a considerable amount of measurement time.

[0037] Condition 9 is a condition for determining whether the necessary condition fulfillment flag, which can be switched by the output of the aforementioned necessary condition determination unit 39, is ON or OFF. By including condition 9 in the first correction condition in this way, the first correction by the first correction unit 40 is performed on the premise that the aforementioned necessary conditions are met.

[0038] Condition 10 is a condition for adjusting the frequency of implementation by the first correction unit 40 based on the input first correction frequency signal Sa1. The first correction frequency signal Sa1 is a rectangular wave pulse signal that repeats between an ON time when the first correction is enabled and an OFF time when the first correction is disabled. When such a first correction frequency signal Sa1 is input, condition 10 is determined to be true during the ON time and false during the OFF time. With condition 10, the frequency of implementation of the first correction can be adjusted by changing the ratio of the ON time and OFF time of the input first correction frequency signal Sa1. The ON time and OFF time of such first correction frequency signal Sa1 can be set by the user.

[0039] Condition 11 is a condition for determining whether the first correction permission flag, which determines whether the first correction is permitted, is ON. For example, the control device 15 is equipped with an operation button (not shown) that allows the user to select whether or not to perform the first correction, and when this operation button is pressed ON, the first correction permission flag is turned ON. This allows the user to select whether or not to perform the first correction.

[0040] The first correction condition determination unit 41 determines that the first correction condition is met when all of conditions 8 to 11 are satisfied, and outputs the first correction signal S1. As mentioned above, the first correction frequency signal Sa1, which is a rectangular wave pulse signal, is input as condition 10 to the first correction condition, so the first correction signal S1 is output as a rectangular wave pulse signal (flicker signal) in which ON time and OFF time are repeated in accordance with the first correction frequency signal Sa1.

[0041] The load-specific correction signal generation unit 42 is configured to generate a load-specific correction signal Sl corresponding to the load L of the combustion device 1. Figure 5 is a processing flow diagram of the load-specific correction signal generation unit 42 shown in Figure 2.

[0042] In this embodiment, the load range that the combustion device 1 can take is divided into a first load range Lr1 to a fourth load range Lr4 (Lr1 < Lr2 < Lr3 < Lr4), and the load-specific correction signal generation unit 42 includes first load range determination units 42a1 to fourth load range determination units 42a4 corresponding thereto. The first load range determination unit 42a1 is a logic circuit that outputs an ON signal when the input load L is included in the first load range Lr1, and outputs an OFF signal when it is not included in the first load range Lr1. The second load range determination unit 42a2 is a logic circuit that outputs an ON signal when the input load L is included in the second load range Lr2, and outputs an OFF signal when it is not included in the second load range Lr2. The third load range determination unit 42a3 is a logic circuit that outputs an ON signal when the input load L is included in the third load range Lr3, and outputs an OFF signal when it is not included in the third load range Lr3. The fourth load range determination unit 42a4 is a logic circuit that outputs an ON signal when the input load L is included in the fourth load range Lr4, and outputs an OFF signal when it is not included in the fourth load range Lr4.

[0043] The first load-specific correction signal output unit 42b1 outputs a first load-specific correction signal Sl1 to command the correction corresponding to the first load range Lr1 when an ON signal is output from the first load range determination unit 42a1 and the first correction signal S1 (flicker signal) from the first correction condition determination unit 41 is in the ON state. The second load-specific correction signal output unit 42b2 outputs a second load-specific correction signal Sl2 to command the correction corresponding to the second load range Lr2 when an ON signal is output from the second load range determination unit 42a2 and the first correction signal S1 (flicker signal) from the first correction condition determination unit 41 is in the ON state. The third load-specific correction signal output unit 42b3 outputs a third load-specific correction signal Sl3 to command the correction to be performed for the third load range Lr3 when an ON signal is output from the third load range determination unit 42a3 and the first correction signal S1 (flicker signal) from the first correction condition determination unit 41 is in the ON state. The fourth load-specific correction signal output unit 42b4 outputs a fourth load-specific correction signal Sl4 to command the correction to be performed for the fourth load range Lr4 when an ON signal is output from the fourth load range determination unit 41a4 and the first correction signal S1 (flicker signal) from the first correction condition determination unit 41 is in the ON state.

[0044] The load-specific correction coefficient calculation unit 43 is configured to calculate load-specific correction coefficients Y1 to Y4 corresponding to the load-specific correction signals Sl1 to Sl4 generated by the load-specific correction signal generation unit 42. Here, Figure 6 is a processing flow diagram of the first load-specific correction coefficient calculation unit 43a of the load-specific correction coefficient calculation unit 43 shown in Figure 2. The first load-specific correction coefficient calculation unit 43a is configured to calculate the first load-specific correction coefficient Y1 based on the first load-specific correction signal Sl1 of the load-specific correction coefficient calculation unit 43.

[0045] Furthermore, the load-specific correction coefficient calculation unit 43 includes, in addition to the first load-specific correction coefficient calculation unit 43a, a second load-specific correction coefficient calculation unit 43b for calculating a second load-specific correction coefficient Y2 based on a second load-specific correction signal Sl2, a third load-specific correction coefficient calculation unit 43c for calculating a third load-specific correction coefficient Y3 based on a third load-specific correction signal Sl3, and a fourth load-specific correction coefficient calculation unit 43d for calculating a fourth load-specific correction coefficient Y4 based on a fourth load-specific correction signal Sl4. Although not shown in the figures, the second load-specific correction coefficient calculation unit 43b, the third load-specific correction coefficient calculation unit 43c, and the fourth load-specific correction coefficient calculation unit 43d are the same as the first load-specific correction coefficient calculation unit 43a described below, unless otherwise specified.

[0046] In the first load-specific correction coefficient calculation unit 43a, when the difference ΔV between the corrected predicted value Vp' (the predicted value Vp of the learning model M itself before the initial correction) and the measured value Vm is input, if the difference ΔV is negative (i.e., the predicted value Vp is smaller than the measured value Vm), the first switch T1 is switched to select the upward correction value A1. The upward correction value A1 is set as a coefficient of 1 or more, obtained by adding the upward correction range to the reference value "1", for example, "1.01". The upward correction value A1 output from the first switch T1 is multiplied by the correction gain K (typically set to "1.0", but can be changed as appropriate) that is set in advance by the correction gain adjustment unit P, and then output to the second switch T2 at the timing when the first correction signal S1 output from the first correction condition determination unit 41 is in the ON state (note that when the first correction signal S1 input to the second switch T2 is in the OFF state, the second switch T2 outputs the default value "1"). The output of the second switch T2 is multiplied by the previously stored value (first load-specific correction coefficient) to output the first load-specific correction coefficient Y1. This output first load-specific correction coefficient Y1 is stored in a memory unit (not shown) and used as the previous value in the next calculation cycle.

[0047] When the difference ΔV is negative, the first load-specific correction coefficient Y1 increases with each calculation cycle by the amount of the upward correction range set in the upward correction value A1. As a result, if the corrected predicted value Vp' is smaller than the measured value Vm, the first load-specific correction coefficient Y1 is calculated so that the corrected predicted value Vp' increases and approaches the measured value Vm.

[0048] On the other hand, if the difference ΔV is positive (i.e., the corrected predicted value Vp' is greater than the measured value Vm), the first switch T1 is switched to select the downward correction value A2. The downward correction value A2 can be set as a coefficient less than 1 obtained by adding the downward correction range (negative value) to the reference value "1", for example, "0.99". The downward correction value A2 output from the first switch T1 is multiplied by the correction gain K (typically set to "1", but can be changed as appropriate) set in advance by the correction gain adjustment unit P, and then output to the second switch T2 at the timing when the first correction signal S1 output from the first correction condition determination unit 41 is in the ON state (note that when the first correction signal input to the second switch T2 is in the OFF state, the second switch T2 outputs the default value "1"). The output of the second switch T2 is multiplied by the previously stored value (first load-specific correction coefficient) and output as the first load-specific correction coefficient Y1. The first load-specific correction coefficient Y1, which is output in this manner, is stored in a memory unit (not shown) and used as the previous value in the next calculation cycle.

[0049] When the difference ΔV is positive, the first load-specific correction coefficient Y1 decreases with each calculation cycle by the amount of the downward correction range set in the downward correction value A2. As a result, if the corrected predicted value Vp' is greater than the measured value Vm, the first load-specific correction coefficient Y1 is calculated so that the corrected predicted value Vp' decreases and approaches the measured value Vm.

[0050] The first correction coefficient calculation unit 44 is configured to correct the predicted value Vp from the learning model M using a first correction coefficient Y corresponding to the load L. Here, Figure 7 is a processing flow diagram of the first correction coefficient calculation unit 44 in Figure 2.

[0051] The first correction coefficient calculation unit 44 receives the first load-specific correction coefficients Y1 to the fourth load-specific correction coefficients Y4 for each load range, which are calculated by the load-specific correction coefficient calculation unit 43. When a load L included in the operating data is input, one of the first load-specific correction coefficients Y1 to the fourth load-specific correction coefficients Y4 corresponding to the load range to which the load L belongs is calculated as the first correction coefficient Y. The first correction coefficient Y calculated in this way is a value corresponding to the magnitude of the load L.

[0052] The first correction coefficient Y calculated by the first correction coefficient calculation unit 44 is output from the third switch T3 when the first correction permission flag is turned ON, and is multiplied by the predicted value Vp of the learning model M to obtain the corrected predicted value Vp' after the first correction has been applied. This first correction is performed by multiplying the predicted value Vp by the first correction coefficient Y, which is calculated in accordance with whether the load L belongs to the first load range Lr1 to the fourth load range Lr4. Therefore, the amount of correction applied to the predicted value Vp can be finely set according to the value of the load L, and the prediction accuracy can be effectively improved without rebuilding the learning model M.

[0053] The second correction unit 50 is configured to perform a correction on the predicted value Vp by the learning model M according to the total load L of the combustion device 1 (i.e., unlike the aforementioned first correction which depends on the load L, this correction does not depend on the magnitude of the load L). The second correction unit 50 comprises a second correction condition determination unit 51 and a second correction coefficient calculation unit 52.

[0054] Figure 8 is a processing flow diagram of the second correction condition determination unit 51 in Figure 2. The second correction condition determination unit 51 is configured to determine whether the second correction condition is met. The second correction condition is a condition for determining whether or not to perform the second correction by the second correction unit 50 on the predicted value Vp of the learning model M. In Figure 8, as an example of the second correction condition, a determination is made based on the following four conditions. (Condition 12) The difference ΔV between the corrected predicted value Vp' (predicted value Vp before initial correction) multiplied by the overall adjustment coefficient and the measured value Vm is greater than the reference value ΔVref. (Condition 13) The flag indicating that the necessary condition has been met must be ON. (Condition 14) The second correction frequency signal Sa2 is input. (Condition 15) The second correction permission flag must be ON.

[0055] Under condition 12, the absolute value (ABS) of the difference ΔV between the corrected predicted value Vp' (which is the predicted value Vp of the learning model M before the initial correction) and the measured value Vm is calculated, and it is determined whether this absolute value is greater than the reference value ΔVref. In Figure 8, the measured value Vm, which is compared with the corrected predicted value Vp' of the learning model M, is subject to measurement delay correction. Measurement delay correction is a process to compensate for the time lag (measurement time delay) between the corrected predicted value Vp' and the measured value Vm obtained by the gas analyzer 17, since obtaining the measured value Vm by the gas analyzer 17 requires a considerable amount of measurement time.

[0056] Condition 13 is a condition for determining whether the necessary condition fulfillment flag, which can be switched by the output of the aforementioned necessary condition determination unit 39, is ON or OFF. By including condition 13 in the second correction condition in this way, the second correction by the second correction unit 50 is performed on the premise that the aforementioned necessary conditions are met.

[0057] Condition 14 is a condition for adjusting the frequency of implementation by the second correction unit 50 based on the input second correction frequency signal Sa2. The second correction frequency signal Sa2 is a rectangular wave pulse signal that repeats between an ON time when the second correction is enabled and an OFF time when the second correction is disabled. When such a second correction frequency signal Sa2 is input, condition 14 is determined to be true during the ON time and false during the OFF time. With condition 14, the frequency of implementation of the second correction can be adjusted by changing the ratio of the ON time and OFF time of the input second correction frequency signal Sa2. The ON time and OFF time of such a second correction frequency signal Sa2 can be set by the user.

[0058] Condition 15 is a condition for determining whether the first correction permission flag, which determines whether the second correction is permitted, is ON. For example, the control device 15 is equipped with an operation button (not shown) that allows the user to select whether or not to perform the second correction, and when this operation button is pressed ON, the second correction permission flag is turned ON. This allows the user to select whether or not to perform the second correction.

[0059] The second correction condition determination unit 51 determines that the second correction condition is met when all of conditions 12 to 15 are satisfied, and outputs the second correction signal S2. As mentioned above, the second correction frequency signal Sa2, which is a rectangular wave pulse signal, is input as condition 14 to the second correction condition, so the second correction signal S2 is output as a rectangular wave pulse signal (flicker signal) in which ON time and OFF time are repeated in accordance with the second correction frequency signal Sa2.

[0060] The second correction coefficient calculation unit 52 is configured to calculate the second correction coefficient based on the second correction signal S2 output from the second correction condition determination unit 51. Figure 9 is a processing flow diagram of the second correction coefficient calculation unit 52 in Figure 2.

[0061] In the second correction coefficient calculation unit 52, when the difference ΔV between the corrected predicted value Vp' (the predicted value Vp of the learning model M itself before the initial correction) and the measured value Vm is input, if the difference ΔV is negative (i.e., the predicted value Vp is less than the measured value Vm), the fourth switch T4 is switched to select the upward correction value A3. The upward correction value A3 is set as a coefficient of 1 or more, obtained by adding the upward correction range to the reference value "1", for example, "1.01". The upward correction value A3 output from the fourth switch T4 is multiplied by the correction gain K (typically set to "1.0", but can be changed as appropriate) which is set in advance by the correction gain adjustment unit P, and then output to the fifth switch T5 at the timing when the second correction signal S2 output from the second correction condition determination unit 51 is in the ON state (note that when the second correction signal S2 input to the fifth switch T5 is in the OFF state, the fifth switch T5 outputs the default value "1"). The output of the fifth switch T5 is multiplied by the previously stored value (second correction coefficient) and output as the second correction coefficient Z. Furthermore, the second correction coefficient Z, which is output in this manner, is stored in a memory unit (not shown) and used as the previous value in the next calculation cycle.

[0062] When the difference ΔV is negative, the second correction coefficient Z increases with each calculation cycle by the amount of the upward correction range set in the upward correction value A3. As a result, if the corrected predicted value Vp' is smaller than the measured value Vm, the second correction coefficient Z is calculated so that the corrected predicted value Vp' increases and approaches the measured value Vm.

[0063] On the other hand, if the difference ΔV is positive (i.e., the corrected predicted value Vp' is greater than the measured value Vm), the fourth switch T4 is switched to select the downward correction value A4. The downward correction value A4 can be set as a coefficient less than 1 obtained by adding the downward correction range (a negative value) to the reference value "1", for example, "0.99". The downward correction value A4 output from the fourth switch T4 is multiplied by the correction gain K (typically set to "1", but can be changed as appropriate) that is set in advance by the correction gain adjustment unit P, and then output to the fifth switch T5 at the timing when the second correction signal S2 output from the second correction condition determination unit 51 mentioned above is turned ON (note that when the second correction signal S2 input to the fifth switch T5 is turned OFF, the fifth switch T5 outputs the default value "1"). The output of the fifth switch T5 is multiplied by the previously stored second correction coefficient Z and output as the second correction coefficient Z. Furthermore, the second correction coefficient Z, which is output in this manner, is stored in a memory unit (not shown) and used as the previous value in the next calculation cycle.

[0064] When the difference ΔV is positive, the second correction coefficient Z decreases with each calculation cycle by the amount of the downward correction range set in the downward correction value A4. As a result, if the corrected predicted value Vp' is greater than the measured value Vm, the second correction coefficient Z is calculated so that the corrected predicted value Vp' is corrected in a way that decreases it and approaches the measured value Vm.

[0065] The second correction coefficient Z calculated by the second correction coefficient calculation unit 52 is output from the sixth switch T6 when the second correction permission flag is turned ON, and is multiplied by the predicted value Vp of the learning model M to obtain the corrected predicted value Vp' after the second correction has been applied. This second correction is performed by multiplying the predicted value Vp by the second correction coefficient Z regardless of whether the load L belongs to the first load range Lr1 to the fourth load range Lr4. Therefore, the predicted value Vp can be corrected overall regardless of the value of the load L, and the prediction accuracy can be effectively improved without rebuilding the learning model M.

[0066] Next, the method for creating table Tb by the table creation unit 31 will be explained with reference to Figure 10. Figure 10 is a processing flow diagram of the table creation unit 31 in Figure 2.

[0067] The creation of table Tb begins by inputting each parameter included in the operating data received by the operating data receiving unit 30 into the learning model M to calculate the learning model M's predicted value Vp (predicted value of SO2 concentration at the absorption tower outlet). The obtained predicted value Vp is corrected by at least one of the first correction unit 40 or the second correction unit 50 according to the success or failure status of the aforementioned necessary conditions, first correction condition, and second correction condition. This yields a corrected predicted value Vp' with higher prediction accuracy than the learning model M's predicted value Vp. Then, the target value of the absorbent concentration in the absorbent solution is calculated as the optimal value for the corrected predicted value Vp' to satisfy a preset reference value (for example, to be below the reference value). Figure 10 shows how table Tb, which defines the relationship between load L and the target absorbent concentration M, is created when such combinations of load L1, L2, ... and the optimal absorbent concentration M1, M2, ... are input.

[0068] The control target value determination unit 32 then inputs the load L included in the operating data into the table Tb created by the table creation unit 31 to determine the target absorbent concentration M so that the corrected predicted value Vp' satisfies a preset standard value, and determines the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution corresponding to the target absorbent concentration M.

[0069] Here, the control target value determination unit 32 may determine the control target value for the amount of absorbent to be added by further correcting the absorbent concentration target value M obtained from table Tb using a feedforward component. Figure 11 is a processing flow diagram of the control target value determination process by the control target value determination unit 32 in Figure 2.

[0070] The control target value determination unit 32 first inputs the difference Δ between the absorbent concentration target value M and its measured value Mm to the PI controller. The PI control value output from the PI controller is corrected using a feedforward correction value calculated based on the operating data. Specifically, the feedforward correction value ff is obtained by adding a first feedforward correction value ff1 corresponding to the exhaust gas flow rate of the combustion device 1 and the SO2 concentration at the absorption tower inlet, and a second feedforward correction value ff2 obtained by converting the predicted input amount using a conversion logic. The feedforward correction value ff obtained in this way is added to the PI control value to determine the control target value for the amount of absorbent to be added. Since the control target value determined in this way enables pre-control using the feedforward correction value, the control accuracy of the control device 15 can be further improved.

[0071] In the control device 15 described above, as previously mentioned with reference to Figure 2, the predicted value Vp of the learning model M is corrected by the predicted value correction unit 36, a table Tb is created using the corrected predicted value Vp', and the control target value determination unit 32 determines the control target value using the table Tb. Such a control device 15 can be modified or added to by adjusting the parameters of the predicted value correction unit 36 ​​and the control target value determination unit 32. For example, the predicted value correction unit 36 ​​can change the reference value of ΔVref, the ON / OFF time in the second correction frequency signal Sa2 of the first correction condition 10 and the second correction condition 14, the upper correction values ​​A1 and A3, the lower correction values ​​A2 and A4, the correction gain K which can be adjusted by the correction gain adjustment unit P, and the upper and lower limits of the first load-specific correction coefficient Y1 and the second correction coefficient Z. Furthermore, the control target value determination unit 32 can change or add PI parameters (parameters related to the PI controller in Figure 11, such as the proportionality constant K and integration time T), the output ff1 of the function Fx in Figure 11, and the output ff2 of the conversion logic, which are the amounts of absorbent to be added.

[0072] If, hypothetically, all the functional blocks of the control device 15, including the predicted value correction unit 36 ​​and the control target value determination unit 32, were to be placed in the higher-level control device 15A, then in the higher-level control device 15A, the functional blocks would be described by a program. Therefore, when changing or adding processes, at least two personnel would be required, including a program engineer to handle the higher-level control device 15A and a control logic engineer to handle the lower-level control device 15B.

[0073] In contrast, with the above configuration, as mentioned above with reference to Figure 2, the predicted value correction unit 36 ​​and the control target value determination unit 32 are located in the lower-level control device 15B, which means that only one engineer (an engineer who can understand the control logic in the lower-level control device 15B) is needed to handle the process, effectively reducing the number of personnel required to change or add processes.

[0074] Figure 12 is a processing flow diagram of the circulation pump adjustment unit 33 in Figure 2. The circulation pump adjustment unit 33 adjusts the number of operating circulation pumps based on the control target value for the circulation flow rate of the absorbent liquid determined by the control target value determination unit 32. The circulation pump adjustment unit 33 adjusts the number of operating circulation pumps by issuing a command to increase the number of operating circulation pumps when, for example, the following four conditions are met. (Condition 16) If the difference Δ between the target value and the measured value Vm of SO2 concentration at the absorption tower outlet changes in the positive direction beyond the allowable value (i.e., if emergency activation of the circulation pump is necessary to maintain the target value of SO2 concentration at the absorption tower outlet), the difference Δ between the optimal number of circulation pumps and the currently operating number of pumps must be less than the standard value (e.g., -0.5 pumps). (Condition 17) After starting one circulation pump, the additional startup prohibition period before it is reflected in the SO2 concentration at the absorption tower outlet must not be in effect. (Condition 18) There must be a pump that can be started. (Condition 19) The optimal driving AI must be in use.

[0075] Furthermore, the circulation pump adjustment unit 33 adjusts the number of operating circulation pumps by issuing a command to reduce the number of operating circulation pumps when, for example, the following four conditions are met. (Condition 16) If the difference Δ between the target value and the measured value Vm of SO2 concentration at the absorption tower outlet falls below the permissible value and changes in the negative direction (i.e., the difference between the target value and the measured value Vm of SO2 concentration at the absorption tower outlet is large and the number of pumps can be reduced), then the difference Δ between the optimal number of circulation pumps and the current number of operating pumps must be greater than the standard value (e.g., 0.5 units). (Condition 20) The system must not be under the "stop after startup" restriction (to prevent frequent starting and stopping of the circulation pump). (Condition 21) The system must not be under a "stop after stop" restriction (to prevent frequent starting and stopping of the circulation pump). (Condition 22) There must be a pump that can be stopped.

[0076] Next, referring to Figure 13, variations in the configuration of the control device 15 described above will be explained. Figures 13A to 13B are block diagrams showing other embodiments of Figure 2.

[0077] In the embodiment shown in Figure 13A, the functional blocks provided by the higher-level control device 15A in Figure 2 (operation data receiving unit 30, learning model construction unit 38, table creation unit 31) are integrated into the lower-level control device 15B. That is, the lower-level control device 15B, excluding the retraining device 15C, includes the operation data receiving unit 30, learning model construction unit 38, table creation unit 31, predicted value correction unit 36, control target value determination unit 32, circulation pump adjustment unit 33, and absorbent slurry supply control unit 34, and all of these functional blocks are realized by control logic. Some of these functional blocks provided by the lower-level control device 15B may be implemented in a programming language that can be handled by the lower-level control device 15B.

[0078] In the configuration shown in Figure 13A, the functional blocks (operation data receiving unit 30, learning model construction unit 38, table creation unit 31) that were described by program by being located in the higher-level control device 15A in the configuration of Figure 2 are realized by being described as control logic in the lower-level control device 15B. As a result, by consolidating the various functions of the control device 15 into the lower-level control device 15B, it becomes unnecessary to install the higher-level control device 15A at the site where the wet flue gas desulfurization system 10, which is the target of control, is located, for example, thereby reducing the management burden of the control device 15.

[0079] In the embodiment shown in Figure 13B, an edge server 15D is provided along with the upper-level control device 15A and the lower-level control device 15B. The edge server 15D is equipped with a data relay unit 60 and is positioned between the upper-level control device 15A and the lower-level control device 15B, enabling data transmission and reception between the upper-level control device 15A and the lower-level control device 15B, which are geographically separated from each other. In this case, the upper-level control device 15A can be applied to an embodiment in which it functions as a remote monitoring system for remotely monitoring the control status of the wet flue gas desulfurization apparatus 10 from a remote location far from the site where the lower-level control device 15B is located (the location where the wet flue gas desulfurization apparatus 10, which is the target of control, is installed).

[0080] As shown in Figure 13B, the wet flue gas desulfurization apparatus 10 may also be configured as a remote monitoring system consisting of, for example, an information processing device 18 that can communicate with a higher-level control device 15A. In this remote monitoring system, remote monitoring is possible by displaying information regarding the control status of the wet flue gas desulfurization apparatus 10 on a display unit 70 provided in the information processing device 18. Furthermore, in such a system configuration, the system may also include a configuration in which the control device 15 executes each processing flow in response to a request from the information processing device 18 via the display unit 70 of the information processing device 18.

[0081] As described above, according to each of the embodiments described above, the predicted values ​​of the learning model used to create a table used to determine the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent liquid are corrected using a correction coefficient. The correction of the predicted values ​​is performed by taking into account the characteristics with respect to the load by using a first correction coefficient calculated based on the load of the combustion device, so that even when the difference between the predicted value and the measured value changes depending on the magnitude of the load, for example, a corrected predicted value with good accuracy can be obtained. By creating a table using such a corrected predicted value, the control target value can be determined with high accuracy, and as a result, good control accuracy can be obtained even when there is a difference between the predicted value of the learning model and the measured value.

[0082] Furthermore, it is possible to replace the components in the above-described embodiments with well-known components as appropriate, without departing from the spirit of this disclosure, and the above-described embodiments may also be combined as appropriate.

[0083] The contents described in each of the above embodiments can be understood, for example, as follows:

[0084] (1) A control device for a wet flue gas desulfurization system according to one embodiment is: A control device for a wet flue gas desulfurization system that performs desulfurization by bringing exhaust gas generated in a combustion device into gas-liquid contact with an absorbent liquid inside an absorption tower, A learning model construction unit for constructing a learning model by machine learning regarding the relationship between at least one parameter included in the operating data of at least one of the combustion apparatus or the wet flue gas desulfurization apparatus, which is an explanatory variable that includes at least the load of the combustion apparatus, and a target variable that is the sulfur dioxide concentration at the future outlet of the absorption tower. A prediction value correction unit for correcting the predicted value of the sulfur dioxide concentration by the learning model using a correction coefficient calculated based on the difference between the predicted value and the measured value of the sulfur dioxide concentration at the outlet of the absorption tower, A table creation unit for creating a table showing the relationship between the load of the combustion device and the target value of the absorbent concentration and the target value of the absorbent circulation amount of the absorbent liquid so that the corrected predicted value, corrected by the predicted value correction unit, satisfies the reference value. A control target value determination unit calculates the target absorbent concentration value and the target absorbent circulation rate value corresponding to the operating data based on the table, and determines the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution corresponding to the target absorbent concentration value and the target absorbent circulation rate value, Equipped with, The predicted value correction unit includes a first correction unit that corrects the predicted value using a first correction coefficient calculated based on the load as the correction coefficient.

[0085] According to the embodiment described in (1) above, the predicted values ​​of the learning model used to create a table used to determine the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent liquid are corrected using a correction coefficient. The correction of the predicted values ​​is performed by taking into account the characteristics with respect to the load by using a first correction coefficient calculated based on the load of the combustion device, so that even when the difference between the predicted value and the measured value changes depending on the magnitude of the load, for example, a corrected predicted value with good accuracy can be obtained. By creating a table using such a corrected predicted value, the control target value can be determined with high accuracy, and as a result, good control accuracy can be obtained even when there is a difference between the predicted value of the learning model and the measured value.

[0086] (2) In other embodiments, in the embodiment of (1) above, The first correction coefficient is calculated such that the difference decreases for each load range to which the load belongs.

[0087] According to the embodiment of (2) above, the first correction coefficient is calculated such that the difference between the predicted value of the learning model and the measured value decreases within the load range to which the load of the combustion device belongs. This makes it possible to make fine-grained corrections to the predicted value of the learning model according to the magnitude of the load of the combustion device, and to create a table for determining the control target value with high accuracy.

[0088] (3) In other embodiments, in the embodiment of (1) or (2) above, The first correction unit calculates the first correction coefficient as the correction coefficient when the first correction condition, which includes the difference between the corrected predicted value and the measured value, is greater than the reference value, is met.

[0089] According to the embodiment described in (3) above, the correction of the predicted value using the first correction coefficient is performed when the first correction condition is met.

[0090] (4) In other embodiments, in any one embodiment of (1) to (3) above, The correction coefficient is calculated to change at a predetermined rate of change so that the difference decreases with each calculation cycle.

[0091] According to the embodiment of (4) above, the correction coefficient is calculated to change at a predetermined rate of change, thereby correcting the predicted value so that the difference between the predicted value and the actual value of the learning model is suitably reduced. Furthermore, the rate of change of the correction coefficient can be set by, for example, a parameter that can be set in advance by the user, so that the degree of correction to the predicted value can be arbitrarily adjusted.

[0092] (5) In other embodiments, in any one embodiment of (1) to (4) above, The predicted value correction unit includes a second correction unit that corrects the predicted value using a second correction coefficient that is independent of the load as the correction coefficient.

[0093] According to the embodiment of (5) above, the correction of the predicted value of the learning model is performed using a second correction coefficient that is independent of the load, in addition to a first correction coefficient calculated based on the load described above. By performing a correction by combining the first and second correction coefficients in this way, the difference between the predicted value of the learning model and the actual value can be reduced more effectively.

[0094] (6) In other embodiments, in the embodiment of (5) above, The second correction unit calculates the second correction coefficient as the correction coefficient when a second correction condition is met, which includes the fact that the difference between the corrected predicted value and the result obtained by multiplying the second correction coefficient is greater than the reference value.

[0095] According to the embodiment described in (6) above, the correction of the predicted value using the second correction coefficient is performed when the second correction condition is met.

[0096] (7) In other embodiments, in any one embodiment of (1) to (6) above, The control target value determination unit determines the control target value for the amount of absorbent to be added by correcting the PI control value corresponding to the difference between the absorbent concentration target value calculated based on the table and the measured absorbent concentration of the absorbent solution using a feedforward correction value calculated based on the operation data.

[0097] According to the embodiment of (7) above, when calculating the control target value for the amount of absorbent to be added based on the absorbent concentration of the absorbent solution calculated based on the table, a correction is made using a feedforward correction value calculated based on the operating data, thereby enabling the calculation of a control target value with high accuracy.

[0098] (8) In other embodiments, in any one embodiment of (1) to (7) above, A higher-level control unit in which functional blocks are composed of programs, The system comprises a lower-level control unit that cooperates with the aforementioned higher-level control unit and whose functional blocks are composed of control logic, At least one of the predicted value correction unit or the control target value determination unit is provided in the lower-level control device.

[0099] According to the embodiment of (8) above, at least one of the functional blocks of the control device, either the predicted value correction unit or the control target value determination unit, is configured with control logic in the lower-level control device. When changing or adding processes by the aforementioned control device, adjustments such as changing the set values ​​in the predicted value correction unit or the control target value determination unit are necessary. However, by providing the predicted value correction unit or the control target value determination unit in the lower-level control device, which is configured with control logic, it becomes possible to handle these changes without requiring personnel familiar with the programs handled by the higher-level control device. Furthermore, by reducing the number of functional blocks in the higher-level control device, the machine specifications required for the higher-level control device can also be reduced.

[0100] (9) In other embodiments, in the embodiment of (8) above, The above-ground control unit is the main control unit. The lower-level control device is under the control of the higher-level control device and is a distributed control device for controlling the components of the wet flue gas desulfurization apparatus.

[0101] According to the embodiment of (9) above, it is suitable for a control device of a wet flue gas desulfurization system that uses a distributed control system (DCS) as a lower-level control device.

[0102] (10) In other embodiments, in the embodiment of (8) or (9) above, The learning model construction unit, the predicted value correction unit, the table creation unit, and the control target value determination unit are provided in the lower-level control device.

[0103] According to the embodiment of (10) above, these functional blocks may be aggregated in a lower-level control unit so that they are all composed of control logic.

[0104] (11) In other embodiments, in any one embodiment of (8) to (10) above, The system includes a relearning device electrically connected to at least one of the higher-level control device or the lower-level control device for relearning the learning model.

[0105] According to the embodiment described in (11) above, if the accuracy is not sufficiently improved by correcting the predicted values ​​of the learning model, this can be addressed by retraining the learning model using a retraining device. In this case, the retraining device is configured separately from the aforementioned higher-level control device and lower-level control device, and is electrically connected to these control devices, thereby reducing the machine specifications required for the higher-level control device and lower-level control device.

[0106] (12) In other embodiments, in any one embodiment of (8) to (11) above, The aforementioned higher-level control device is a remote control device for remotely monitoring the lower-level control device installed at the site where the wet flue gas desulfurization apparatus is located. The lower-level control unit can communicate with the upper-level control unit via an edge server capable of relaying data between the upper-level control unit and the lower-level control unit.

[0107] According to the embodiment described in (12) above, data transmission and reception between the upper-level control unit and the lower-level control unit becomes possible through data relay by the edge server. Therefore, by locating the upper-level control unit and the lower-level control unit in distant locations, a remote control unit for flue gas desulfurization can be realized.

[0108] (13) A remote monitoring system relating to one aspect is: A remote monitoring system comprising a control device and a communication-enabled information processing device for a wet flue gas desulfurization system that performs desulfurization by gas-liquid contact between exhaust gas generated in a combustion device and an absorbent liquid inside an absorption tower, The control device is A learning model construction unit constructs a learning model by machine learning regarding the relationship between at least one parameter included in the operating data of at least one of the combustion apparatus or the wet flue gas desulfurization apparatus, which is an explanatory variable that includes at least the load of the combustion apparatus, and a target variable which is the sulfur dioxide concentration at the future outlet of the absorption tower. A prediction value correction unit corrects the predicted value of the sulfur dioxide concentration by the learning model using a correction coefficient calculated based on the difference between the predicted value and the measured value of the sulfur dioxide concentration at the outlet of the absorption tower. A table creation unit creates a table showing the relationship between the load of the combustion device and the target value of the absorbent concentration and the target value of the absorbent circulation amount of the absorbent liquid so that the corrected predicted value, corrected by the predicted value correction unit, satisfies the reference value. A control target value determination unit calculates the target absorbent concentration value and the target absorbent circulation rate value corresponding to the operating data based on the table, and determines the target control values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution corresponding to the target absorbent concentration value and the target absorbent circulation rate value. Equipped with, The predicted value correction unit includes a first correction unit that corrects the predicted value using a first correction coefficient calculated based on the load as the correction coefficient.

[0109] According to the embodiment of (13) above, remote monitoring can be suitably performed by displaying information regarding the control status of the wet flue gas desulfurization apparatus 1 according to each embodiment on a display unit provided in the information processing device.

[0110] (14) A control method for a remote monitoring system according to one embodiment is: A control method for a remote monitoring system comprising a control device and a communication-enabled information processing device for a wet flue gas desulfurization apparatus that performs desulfurization by gas-liquid contact between exhaust gas generated in a combustion device and an absorbent liquid inside an absorption tower, wherein the system performs desulfurization. The control device is A learning model construction step involves constructing a learning model by machine learning for the relationship between an explanatory variable, which is at least one parameter included in the operating data of at least one of the combustion apparatus or the wet flue gas desulfurization apparatus, and which includes at least the load of the combustion apparatus, and a target variable, which is the sulfur dioxide concentration at the future outlet of the absorption tower. A prediction value correction step, in which the predicted value of the sulfur dioxide concentration by the learning model is corrected using a correction coefficient calculated based on the difference between the predicted value and the measured value of the sulfur dioxide concentration at the outlet of the absorption tower, A table creation step involves creating a table that shows the relationship between the load of the combustion device and the target value of the absorbent concentration and the target value of the absorbent circulation amount of the absorbent liquid so that the corrected predicted value, corrected by the predicted value correction step, satisfies the reference value. A control target value determination step involves calculating the target absorbent concentration value and the target absorbent circulation rate value corresponding to the operating data based on the table, and determining the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution corresponding to the target absorbent concentration value and the target absorbent circulation rate value. Execute, The predicted value correction step includes a first correction step of correcting the predicted value using a first correction coefficient calculated based on the load as the correction coefficient.

[0111] According to the embodiment of (14) above, remote monitoring can be suitably performed by displaying information regarding the control status of the wet flue gas desulfurization apparatus 1 according to each embodiment on a display unit provided in the information processing device.

[0112] (15) A control method for a wet flue gas desulfurization apparatus according to one embodiment is: A control method for a wet flue gas desulfurization system that performs desulfurization by bringing exhaust gas generated in a combustion device and an absorbent liquid into gas-liquid contact within an absorption tower, A step of constructing a learning model by machine learning regarding the relationship between at least one parameter included in the operating data of at least one of the combustion apparatus or the wet flue gas desulfurization apparatus, which is an explanatory variable that includes at least the load of the combustion apparatus, and a target variable that is the sulfur dioxide concentration at the future outlet of the absorption tower. A step of correcting the predicted value of the sulfur dioxide concentration by the learning model using a correction coefficient calculated based on the difference between the predicted value and the measured value of the sulfur dioxide concentration at the outlet of the absorption tower, A step of creating a table showing the relationship between the load of the combustion device and the target value of the absorbent concentration and the target value of the absorbent circulation amount of the absorbent liquid so that the corrected predicted value, corrected by the predicted value correction unit, satisfies the reference value. A step of calculating the target absorbent concentration value and the target absorbent circulation volume value corresponding to the operating data based on the table, and determining the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution corresponding to the target absorbent concentration value and the target absorbent circulation volume value, Equipped with, In the step of correcting the predicted value, the predicted value is corrected using a first correction coefficient calculated based on the load as the correction coefficient.

[0113] According to the embodiment of (15) above, the predicted values ​​of the learning model used to create a table used to determine the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent liquid are corrected using a correction coefficient. The correction of the predicted values ​​is performed by taking into account the characteristics with respect to the load by using a first correction coefficient calculated based on the load of the combustion device, so that even when the difference between the predicted value and the measured value changes depending on the magnitude of the load, for example, a corrected predicted value with good accuracy can be obtained. By creating a table using such a corrected predicted value, the control target value can be determined with good accuracy, and as a result, good control accuracy can be obtained even when there is a difference between the predicted value of the learning model and the measured value. [Explanation of Symbols]

[0114] 1. Combustion device 2 Piping 3 Circulation piping 5 Generators 10 Wet flue gas desulfurization equipment 11 Absorption Tower 12 Circulation pump 13 Absorbent slurry supply unit 14. Gypsum Recovery Section 15 Control device 15A Higher-level control unit 15B Lower Control Unit 15C Retraining Device 15D Edge Server 16 Outlet piping 17 Gas analyzer 18 Information Processing Devices 20. Operation data acquisition unit 21 Absorbent Slurry Manufacturing Equipment 22 Piping for supplying absorbent slurry 23 Absorbent slurry supply control valve 25 Gypsum separator 26. Piping for extracting gypsum slurry 27. Pump for extracting gypsum slurry 30. Operation data receiving unit 31 Table Creation Section 32 Control target value determination unit 33 Circulation pump adjustment unit 34 Absorbent slurry supply control unit 36 Prediction value correction unit 38. Learning Model Construction Department 39 Necessary condition determination section 40. First Amendment Section 41 First correction condition determination section 42 Load-specific correction signal generation unit 42a1~42a4 1st~4th Load Range Determination Unit 42b1~42b4 First to fourth load-specific correction signal output section 43 Load-Specific Correction Coefficient Calculation Unit 43a~43d Calculation unit for first to fourth load-specific correction coefficients 44. First Correction Coefficient Calculation Unit 50 Second Correction Section 51 Second correction condition determination section 52 Second Correction Coefficient Calculation Unit 60 Data Relay Unit 70 Display section M Learning Model

Claims

1. A control device for a wet flue gas desulfurization system that performs desulfurization by bringing exhaust gas generated in a combustion device into gas-liquid contact with an absorbent liquid inside an absorption tower, A learning model construction unit for constructing a learning model by machine learning regarding the relationship between at least one parameter included in the operating data of at least one of the combustion apparatus or the wet flue gas desulfurization apparatus, which is an explanatory variable that includes at least the load of the combustion apparatus, and a target variable that is the sulfur dioxide concentration at the future outlet of the absorption tower. A prediction value correction unit for correcting the predicted value of the sulfur dioxide concentration by the learning model using a correction coefficient calculated based on the difference between the predicted value and the measured value of the sulfur dioxide concentration at the outlet of the absorption tower, A table creation unit for creating a table showing the relationship between the load of the combustion device and the target value of the absorbent concentration and the target value of the absorbent circulation amount of the absorbent liquid so that the corrected predicted value, corrected by the predicted value correction unit, satisfies the reference value. A control target value determination unit calculates the target absorbent concentration value and the target absorbent circulation rate value corresponding to the operating data based on the table, and determines the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution corresponding to the target absorbent concentration value and the target absorbent circulation rate value, Equipped with, The control device for a wet flue gas desulfurization apparatus includes a first correction unit that corrects the predicted value using a first correction coefficient calculated based on the load as the correction coefficient.

2. The control device for a wet flue gas desulfurization apparatus according to claim 1, wherein the first correction coefficient is calculated such that the difference decreases for each load range to which the load belongs.

3. The control device for a wet flue gas desulfurization apparatus according to claim 1 or 2, wherein the first correction unit calculates the first correction coefficient as the correction coefficient when a first correction condition is met, which includes the difference between the corrected predicted value and the measured value being greater than a reference value.

4. The control device for a wet flue gas desulfurization apparatus according to claim 1 or 2, wherein the correction coefficient is calculated to change at a preset rate of change such that the difference decreases with each calculation cycle.

5. The control device for a wet flue gas desulfurization apparatus according to claim 1 or 2, wherein the predicted value correction unit includes a second correction unit that corrects the predicted value using a second correction coefficient that is independent of the load as the correction coefficient.

6. The control device for a wet flue gas desulfurization apparatus according to claim 5, wherein the second correction unit calculates the second correction coefficient as the correction coefficient when a second correction condition is met, which includes the condition that the difference between the corrected predicted value and the result obtained by multiplying the second correction coefficient is greater than a reference value.

7. The control device for a wet flue gas desulfurization apparatus according to claim 1 or 2, wherein the control target value determination unit determines the control target value for the amount of absorbent to be added by correcting the PI control value corresponding to the difference between the absorbent concentration target value calculated based on the table and the measured absorbent concentration of the absorbent solution using a feedforward correction value calculated based on the operation data.

8. A higher-level control unit in which functional blocks are composed of programs, The system comprises a lower-level control unit that cooperates with the aforementioned higher-level control unit and whose functional blocks are composed of control logic, The control device for a wet flue gas desulfurization apparatus according to claim 1 or 2, wherein at least one of the predicted value correction unit or the control target value determination unit is provided in the lower control device.

9. The above-ground control unit is the main control unit. The control device for a wet flue gas desulfurization apparatus according to claim 8, wherein the lower-level control device is under the control of the higher-level control device and is a distributed control device for controlling the components of the wet flue gas desulfurization apparatus.

10. The control device for a wet flue gas desulfurization apparatus according to claim 8 or 9, wherein the learning model construction unit, the predicted value correction unit, the table creation unit, and the control target value determination unit are provided in the lower-level control device.

11. A control device for a wet flue gas desulfurization apparatus according to claim 8 or 9, comprising a relearning device electrically connected to at least one of the higher-level control device or the lower-level control device for relearning the learning model.

12. The aforementioned higher-level control device is a remote control device for remotely monitoring the lower-level control device installed at the site where the wet flue gas desulfurization apparatus is located. The control device for a wet flue gas desulfurization apparatus according to claim 8 or 9, wherein the lower-level control device can communicate with the upper-level control device via an edge server capable of relaying data between the upper-level control device and the lower-level control device.

13. A remote monitoring system comprising a control device and a communication-enabled information processing device for a wet flue gas desulfurization system that performs desulfurization by bringing exhaust gas generated in a combustion device and an absorbent liquid into gas-liquid contact within an absorption tower, The control device is A learning model construction unit constructs a learning model by machine learning regarding the relationship between at least one parameter included in the operating data of at least one of the combustion apparatus or the wet flue gas desulfurization apparatus, which is an explanatory variable that includes at least the load of the combustion apparatus, and a target variable which is the sulfur dioxide concentration at the future outlet of the absorption tower. A prediction value correction unit corrects the predicted value of the sulfur dioxide concentration by the learning model using a correction coefficient calculated based on the difference between the predicted value and the measured value of the sulfur dioxide concentration at the outlet of the absorption tower. A table creation unit creates a table showing the relationship between the load of the combustion device and the target value of the absorbent concentration and the target value of the absorbent circulation amount of the absorbent liquid so that the corrected predicted value, corrected by the predicted value correction unit, satisfies the reference value. A control target value determination unit calculates the target absorbent concentration value and the target absorbent circulation rate value corresponding to the operating data based on the table, and determines the target control values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution corresponding to the target absorbent concentration value and the target absorbent circulation rate value. Equipped with, The remote monitoring system for a wet flue gas desulfurization apparatus includes a first correction unit that corrects the predicted value using a first correction coefficient calculated based on the load as the correction coefficient.

14. A control method for a remote monitoring system comprising a control device and a communication-enabled information processing device for a wet flue gas desulfurization apparatus that performs desulfurization by gas-liquid contact between exhaust gas generated in a combustion device and an absorbent liquid inside an absorption tower, wherein the system performs desulfurization. The control device is A learning model construction step involves constructing a learning model by machine learning for the relationship between an explanatory variable, which is at least one parameter included in the operating data of at least one of the combustion apparatus or the wet flue gas desulfurization apparatus, and which includes at least the load of the combustion apparatus, and a target variable, which is the sulfur dioxide concentration at the future outlet of the absorption tower. A prediction value correction step, in which the predicted value of the sulfur dioxide concentration by the learning model is corrected using a correction coefficient calculated based on the difference between the predicted value and the measured value of the sulfur dioxide concentration at the outlet of the absorption tower, A table creation step involves creating a table that shows the relationship between the load of the combustion device and the target value of the absorbent concentration and the target value of the absorbent circulation amount of the absorbent liquid so that the corrected predicted value, corrected by the predicted value correction step, satisfies the reference value. A control target value determination step involves calculating the target absorbent concentration value and the target absorbent circulation rate value corresponding to the operating data based on the table, and determining the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution corresponding to the target absorbent concentration value and the target absorbent circulation rate value. Execute, A control method for a remote monitoring system, wherein the predicted value correction step includes a first correction step of correcting the predicted value using a first correction coefficient calculated based on the load as the correction coefficient.

15. A control method for a wet flue gas desulfurization system that performs desulfurization by bringing exhaust gas generated in a combustion device and an absorbent liquid into gas-liquid contact within an absorption tower, A step of constructing a learning model by machine learning regarding the relationship between at least one parameter included in the operating data of at least one of the combustion apparatus or the wet flue gas desulfurization apparatus, which is an explanatory variable that includes at least the load of the combustion apparatus, and a target variable that is the sulfur dioxide concentration at the future outlet of the absorption tower. A step of correcting the predicted value of the sulfur dioxide concentration by the learning model using a correction coefficient calculated based on the difference between the predicted value and the measured value of the sulfur dioxide concentration at the outlet of the absorption tower, A step of creating a table showing the relationship between the load of the combustion device and the target value of the absorbent concentration and the target value of the absorbent circulation amount of the absorbent liquid so that the corrected predicted value, corrected by the predicted value correction unit, satisfies the reference value. A step of calculating the target absorbent concentration value and the target absorbent circulation rate value corresponding to the operating data based on the table, and determining the control target values ​​for the amount of absorbent to be added and the circulation flow rate of the absorbent solution corresponding to the target absorbent concentration value and the target absorbent circulation rate value, Execute, A control method for a wet flue gas desulfurization apparatus, wherein in the step of correcting the predicted value, the predicted value is corrected using a first correction coefficient calculated based on the load as the correction coefficient.

Images