Learning device and boiler control system

JP7595465B2Active Publication Date: 2024-12-06IHI CORP
View PDF 13 Cites 0 Cited by

Patent Information

Application Number
JP2020571299
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2019-02-08
Filing Date
2020-02-07
Publication Date
2024-12-06
Estimated Expiration
2040-02-07

AI Technical Summary

Benefits of technology

【0016】 以上説明したように、本開示によれば、発電指令値に対する発電量の追従性を向上させることができる。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007595465000001
    Figure 0007595465000001
  • Figure 0007595465000002
    Figure 0007595465000002
  • Figure 0007595465000003
    Figure 0007595465000003
Patent Text Reader

Abstract

This learning device (18, 18b) is a learning device that constructs a learning model for generating an advance acceleration command value for controlling a control target in advance when the load of a boiler (3) used for thermal power generation changes, and includes a learning unit (20, 20b) that generates the learning model by machine learning a data set including a power generation command value for the thermal power generation and the advance acceleration command value used in past operations of the boiler as learning data.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present disclosure relates to a learning device and a boiler control system. This application claims priority based on Japanese Patent Application No. 2019-021712, filed on February 8, 2019, the contents of which are incorporated herein by reference. [Background technology]

[0002] A boiler control device used in a thermal power plant is disclosed in the following Patent Document 1. This boiler control device controls a boiler so that the power generation amount E of the thermal power plant (steam turbine) follows a power generation command value MWD.

[0003] It is known that a delay in the response of a boiler occurs when the load of the boiler changes. Therefore, in order to compensate for the delay in the response, the boiler control device may, for example, increase the fuel of the boiler in advance based on a precedence acceleration command value called BIR (Boiler Input Rate) in addition to the power generation command value MWD when the load of the boiler changes. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2012-41889 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the BIR must be set by the user based on past experience. Therefore, the accuracy of the BIR may be greatly influenced by the user's experience, and the response delay may not be compensated for. As a result, the ability of the power generation amount E to follow the power generation command value MWD may be poor.

[0006] The present disclosure has been made in view of the above circumstances, and has an object to improve the ability of the amount of power generation to follow a power generation command value. [Means for solving the problem]

[0007] A learning device of a first aspect of the present disclosure is a learning device that constructs a learning model for generating an advance acceleration command value for controlling a controlled object in advance when the load of a boiler used for thermal power generation changes, and includes a learning unit that generates the learning model by machine learning a data set including a power generation command value for the thermal power generation and the advance acceleration command value used in past operations of the boiler as learning data.

[0008] A second aspect of the present disclosure is the learning device according to the first aspect, wherein the learning unit includes a process value of the controlled object obtained during past operation of the boiler in the learning data.

[0009] A third aspect of the present disclosure is directed to the learning device of the second aspect, wherein the process value of the controlled object is one or more of the flow rate of fuel supplied to the boiler, the flow rate of water supplied to the boiler, the flow rate of air supplied to the boiler, the ammonia flow rate for denitration, the spray flow rate for reheat steam temperature control, and the steam pressure.

[0010] A fourth aspect of the present disclosure is the learning device according to any one of the first to third aspects, wherein the learning unit uses the data set obtained during past operation of the boiler as learning data.

[0011] A fifth aspect of the present disclosure is a learning device according to any one of the first to fourth aspects, wherein the learning unit constructs the learning model based on the learning data, which inputs at least the power generation command value and outputs the anticipatory acceleration command value such that the tracking ability of the power generation amount E to the power generation command value exceeds a certain level.

[0012] A sixth aspect of the present disclosure is directed to the learning device of any one of the first to fourth aspects, wherein the learning unit includes in the data set the amount of power generated by the thermal power generation obtained during past operation of the boiler.

[0013] A seventh aspect of the present disclosure is a learning device of the sixth aspect, in which the learning unit constructs a learning model based on the learning data, which receives at least the power generation command value and the advance acceleration command value as input and outputs the power generation amount.

[0014] A boiler control system of an eighth aspect of the present disclosure includes the learning device of the fifth aspect, and a boiler control device that obtains the anticipatory acceleration command value by inputting at least the power generation command value into a learned model, which is the learning model constructed by the learning device, and when the load on the boiler changes, the boiler control device controls the controlled object based on a value obtained by adding the anticipatory acceleration command value to the power generation command value, thereby causing the power generation amount to follow the power generation command value.

[0015] A boiler control system of a ninth aspect of the present disclosure includes the learning device of the seventh aspect and a boiler control device that obtains the power generation amount by inputting at least the power generation command value into a trained model, which is the learning model constructed by the learning device, and the boiler control device determines the anticipatory acceleration command value that minimizes the deviation between the power generation command value input into the trained model and the power generation amount output from the trained model. Effect of the Invention

[0016] As described above, according to the present disclosure, it is possible to improve the ability of the amount of power generation to follow a power generation command value. [Brief description of the drawings]

[0017] [Figure 1] 1 is a diagram showing an example of a schematic configuration of a thermal power generation system according to a first embodiment. [Diagram 2] FIG. 4 is a diagram illustrating a pre-acceleration command value BIR. [Diagram 3]FIG. 2 is a functional block diagram of the learning device according to the first embodiment. [Figure 4] 1 is a functional block diagram of a boiler control device according to a first embodiment. [Diagram 5] FIG. 4 is a flowchart illustrating an example of the operation of the learning device according to the first embodiment. [Figure 6] FIG. 11 is a diagram showing an example of a schematic configuration of a thermal power generation system according to a second embodiment. [Figure 7] FIG. 11 is a functional block diagram of a learning device according to a second embodiment. [Figure 8] FIG. 5 is a functional block diagram of a boiler control device according to a second embodiment. [Figure 9A] FIG. 11 is a flowchart illustrating an example of the operation of the learning device according to the second embodiment. [Figure 9B] FIG. 11 is a flow diagram illustrating an example of an operation of a BIR generation unit according to the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0018] Hereinafter, a learning device and a boiler control system according to a first embodiment will be described with reference to the drawings.

[0019] (First embodiment) 1 is a diagram showing an example of a schematic configuration of a thermal power generation system A according to the first embodiment. As shown in FIG.

[0020] In the thermal power plant 1, heat from combustion gas generated by burning fuel is used to generate steam in a boiler 3. Then, in the thermal power plant 1, the steam is used to drive a steam turbine 4, and a desired amount of power generation E is obtained in a generator 5 directly connected to the steam turbine 4.

[0021] The plant control system 2 controls the power generation amount E generated by the generator 5 by controlling the boiler 3. Here, controlling the boiler 3 means controlling a parameter (hereinafter referred to as a "controlled object") required for the operation of the boiler 3. For example, the controlled object is one or more of the flow rate of fuel supplied to the boiler 3 (hereinafter referred to as a "fuel flow rate"), the flow rate of water supplied to the boiler 3 (hereinafter referred to as a "feedwater flow rate"), the flow rate of air supplied to the boiler 3 (hereinafter referred to as an "air flow rate"), the pressure of steam supplied from the boiler 3 to the steam turbine 4 (hereinafter referred to as a "main steam pressure") and the temperature of the steam (hereinafter referred to as a "main steam temperature"), the ammonia flow rate for denitration, and the flow rate of a spray for controlling the reheat steam temperature (hereinafter referred to as a "spray flow rate for controlling reheat steam temperature").

[0022] A schematic configuration of a thermal power plant 1 according to a first embodiment will be described below with reference to FIG.

[0023] The thermal power plant 1 includes a boiler 3, a steam turbine 4, a generator 5, a condenser 6, a fuel flow rate adjustment unit 7, an air flow rate adjustment unit 8, a steam flow rate regulation unit 9, a feedwater pump 10, a fuel flow rate sensor 11, an air flow rate sensor 12, a pressure sensor 13, a temperature sensor 14, a feedwater flow rate sensor 15 and a power sensor 16.

[0024] The boiler 3 takes in outside air and fuel such as pulverized coal and burns the fuel to generate combustion gas. The boiler 3 then heats water with the heat of the combustion gas to generate steam, and supplies the steam to the steam turbine 4. Here, pulverized coal is a fuel made by pulverizing coal. In the first embodiment, the fuel is coal (for example, pulverized coal), but the fuel is not limited to this. For example, the fuel may be petroleum such as heavy oil or light oil, liquefied natural gas (LNG), liquefied petroleum gas (LPG), natural gas such as methane hydrate or shale gas, or biomass.

[0025] The steam turbine 4 is directly connected to the generator 5. That is, the output shaft of the steam turbine 4 is connected to the rotating shaft of the generator 5. The steam turbine 4 is rotated by the steam generated in the boiler 3 to rotate the generator 5. The generator 5 is driven by the rotation of the steam turbine 4 to generate electricity.

[0026] The condenser 6 cools and condenses the steam that has rotated the steam turbine 4. The fuel flow rate adjusting unit 7 is controlled by the plant control system 2 to adjust the flow rate of fuel supplied to the boiler 3. For example, the fuel flow rate adjusting unit 7 is a flow rate adjustment valve, and the opening degree of the valve is controlled by the plant control system 2 to adjust the flow rate of fuel supplied to the boiler 3.

[0027] The air flow rate adjusting unit 8 is controlled by the plant control system 2 to adjust the flow rate of air supplied to the boiler 3. For example, the air flow rate adjusting unit 8 is a flow rate adjusting valve or a damper, and the opening degree of which is controlled by the plant control system 2 to adjust the flow rate of air supplied to the boiler 3. Here, the air may be primary air, secondary air, or both.

[0028] The steam flow rate adjustment unit 9 is controlled by the plant control system 2 to adjust the flow rate of steam supplied from the boiler 3 to the steam turbine 4. For example, the steam flow rate adjustment unit 9 is a flow rate adjustment valve, and the opening degree of which is controlled by the plant control system 2 to adjust the flow rate of steam supplied from the boiler 3 to the steam turbine 4.

[0029] The feedwater pump 10 supplies the water condensed by the condenser 6 to the boiler 3. The drive of the feedwater pump 10 is controlled by the plant control system 2. Therefore, the flow rate of the feedwater supplied from the feedwater pump 10 to the boiler 3 is controlled by the plant control system 2.

[0030] The fuel flow rate sensor 11 measures the flow rate of fuel supplied to the boiler 3 and outputs the measured fuel flow rate to the plant control system 2.

[0031] The air flow sensor 12 measures the flow rate of air supplied to the boiler 3 and outputs the measured air flow rate to the plant control system 2.

[0032] The pressure sensor 13 measures the pressure of the steam supplied from the boiler 3 to the steam turbine 4, and outputs the measured steam pressure to the plant control system 2.

[0033] The temperature sensor 14 measures the main steam temperature, which is the temperature of steam supplied from the boiler 3 to the steam turbine 4, and outputs the measured main steam temperature to the plant control system 2.

[0034] The feedwater flow rate sensor 15 measures the flow rate of the feedwater supplied to the boiler 3 , and outputs the measured feedwater flow rate to the plant control system 2 .

[0035] The power sensor 16 measures the amount of power generated E by the generator 5, and outputs the measured amount of power generated E to the plant control system 2.

[0036] Next, a schematic configuration of the plant control system 2 according to the first embodiment will be described. The plant control system 2 includes a boiler control device 17 and a learning device 18. In the first embodiment, the boiler control device 17 and the learning device 18 are configured as separate entities, but the present disclosure is not limited to this. The boiler control device 17 and the learning device 18 may be configured as an integrated device. For example, an information processing device may have the functions of the boiler control device 17 and the learning device 18. This information processing device is, for example, a computer. Each of the boiler control device 17 and the learning device 18 may be configured with a microprocessor such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit), or a microcontroller such as an MCU (Micro Controller Unit). An information processing device having the functions of the boiler control device 17 and the learning device 18 may be composed of a CPU, memory such as RAM (Random Access Memory) or ROM (Read Only Memory), storage devices such as SSD (Solid State Drive) or HDD (Hard Disk Drive), and input / output devices that exchange signals with external devices such as sensors.

[0037] The boiler control device 17 controls each control target based on a power generation command value MWD (Mega Watt Demand), which is a command value for the output of thermal power generation (generator 5), to make the power generation amount E follow the power generation command value MWD. ​​This power generation command value MWD is a target value for the power generation amount E generated by the generator 5. The boiler control device 17 may obtain the power generation command value MWD from an external device, or may calculate it using a known technique.

[0038] Here, it is known that a response delay occurs in the boiler 3 when the load of the boiler 3 changes. Therefore, in order to compensate for the response delay, the boiler control device 17 performs anticipatory control, for example, to increase fuel in the boiler 3 in anticipation, based on an anticipatory acceleration command value called BIR (Boiler Input Rate) (hereinafter referred to as "anticipatory acceleration command value BIR") in addition to the power generation command value MWD when the load of the boiler 3 changes.

[0039] 2, for example, it is assumed that the thermal power plant 1 according to the first embodiment is a 700 MW coal-fired power plant, and the load change is to increase the power generation amount E of the generator 5 from 350 MW to 700 MW. It is also assumed that the static characteristics of the boiler 3 are such that a fuel consumption amount (fuel flow rate) of 110 t / h to 220 t / h is required to increase the power generation amount E from 350 MW to 700 MW.

[0040] In this case, if the boiler control device 17 controls the fuel flow rate from 110 t / h to 220 t / h in one go to increase the power generation amount E from 350 MW to 700 MW, the drive of multiple devices (for example, a device for crushing coal) may not be able to keep up. Therefore, the boiler control device 17 does not increase the power generation amount E from 350 MW to 700 MW in one go, but increases the power generation amount E at a certain rate of change. That is, when there is a load change from the first power generation amount to the second power generation amount, the power generation command value MWD becomes a power generation command value that instructs the power generation amount E to be increased from the first power generation amount to the second power generation amount at a certain rate of change (FIG. 2(a)). However, on the other hand, there is a user's request to speed up this load change. Therefore, when the load of the boiler 3 changes, the boiler control device 17 performs a proactive control to increase the statically determined control target in advance, as shown in FIG. 2(c). Specifically, the boiler control device 17 increases the control target in advance based on the proactive acceleration command value BIR shown in FIG. 2(b). FIG. 2(a) is a diagram showing the relationship between time and the amount of power generation when the amount of power generation is increased at a certain rate of change, FIG. 2(b) is a diagram showing the relationship between time and the fuel flow rate when only the advance acceleration command value BIR is applied, and FIG. 2(c) is a diagram showing the relationship between time and the fuel flow rate when a fuel flow rate based on BIR is added to the fuel flow rate at the static state (the fuel flow rate when the amount of power generation is increased at the rate of change shown in FIG. 2(a)).

[0041] In the first embodiment, the boiler control device 17 will be described as an example of advanced control of the main steam temperature by proactively increasing the fuel flow rate. However, the advanced control of the boiler control device 17 is not limited to this, and the boiler control device 17 may proactively control the feedwater flow rate as advanced control of the main steam pressure. Furthermore, the boiler control device 17 may proactively control the air flow rate. For example, the boiler control device 17 may proactively control at least one of the above-mentioned control targets.

[0042] Here, one of the features of this embodiment is that the boiler control device 17 uses, for the advance control, an advance acceleration command value BIR generated using a learning model (trained model) constructed by the learning device 18 through machine learning. In the following description, the process of constructing the learning model may be referred to as a learning process, and the process of generating the advance acceleration command value BIR using the trained model may be referred to as a control process.

[0043] The learning device 18 constructs a learning model for generating an anticipatory acceleration command value BIR used in anticipatory control. First, the functional blocks of the learning device 18 according to the first embodiment will be described with reference to Fig. 3. Fig. 3 is a functional block diagram of the learning device 18 according to the first embodiment.

[0044] As shown in FIG. 3, the learning device 18 includes a learning data acquisition unit 19, a learning unit 20, and a learning model storage unit 21.

[0045] The learning data acquisition unit 19 acquires past operation data of the boiler 3 as learning data. This operation data is at least one of data necessary for the operation of the boiler 3 and data indicating the state of the boiler during the operation, and may include both data. For example, this operation data is a power generation command value MWD, an advance acceleration command value BIR, the power generation amount E, and a process value of a controlled object. The process value of a controlled object is, for example, one or more of a fuel flow rate, a feedwater flow rate, an air flow rate, a main steam pressure, ammonia flow rate for denitration, a spray flow rate for controlling reheat steam temperature, and a main steam temperature. Specifically, the learning data acquiring unit 19 acquires, as learning data, a data set including the power generation command value MWD and the advance acceleration command value BIR used in the past operation of the boiler 3. In addition, the learning data acquiring unit 19 may include the power generation amount E and the process values ​​of the controlled object obtained in the past operation of the boiler 3 in the above data set.

[0046] Furthermore, the learning data acquisition unit 19 may extract, from the past operation data of the boiler 3, a data set when the deviation between the power generation command value MWD used in the past operation of the boiler 3 and the power generation amount E obtained by the operation (hereinafter referred to as the "power generation deviation amount") is equal to or less than a threshold value as the learning data. The data set when the power generation deviation amount is equal to or less than a threshold value indicates a data set when the tracking ability of the power generation amount E with respect to the power generation command value MWD exceeds a certain level. The threshold value may be set according to a value that is acceptable as the power generation deviation amount during actual operation of the boiler 3.

[0047] In this way, the learning data acquisition unit 19 has an evaluation function for evaluating the tracking ability of the power generation amount E with respect to the power generation command value MWD, and may extract, as the learning data, the operation data when the evaluation of the tracking ability exceeds a certain level. However, the learning data acquisition unit 19 is not required to use, as the learning data, the operation data when the evaluation of the tracking ability exceeds a certain level, and may also extract, as the learning data, the operation data when the evaluation of the tracking ability does not exceed a certain level. In other words, the learning data acquisition unit 19 may use all operation data as the learning data, regardless of the tracking ability of the power generation amount E with respect to the power generation command value MWD. ​​For example, the learning data acquisition unit 19 may use, as the learning data, the operation data when the evaluation of the tracking ability exceeds a certain level and a correct answer label indicating that the tracking ability is good. Also, the learning data acquisition unit 19 may use, as the learning data, the operation data when the evaluation of the tracking ability does not exceed a certain level and a correct answer label indicating that the tracking ability is poor. The past operation may be a test run of the boiler performed in the past, an actual operation of the boiler performed in the past, or both. The learning data acquisition unit 19 may acquire the power generation amount E from the power sensor 16.

[0048] The learning data acquiring unit 19 may acquire the above-mentioned operating data as learning data from the boiler control device 17, or may acquire the data directly from the above-mentioned various sensors (fuel flow sensor 11, air flow sensor 12, pressure sensor 13, temperature sensor 14, feedwater flow sensor 15, etc.). Furthermore, the learning data acquiring unit 19 may acquire the above-mentioned operating data stored in a storage unit (not shown) provided outside or inside the learning device 18 by reading it as learning data. In this manner, the learning data acquiring unit 19 acquires, from among the operating data, a data set including the power generation command value MWD and the anticipatory acceleration command value BIR as learning data.

[0049] The learning unit 20 constructs a learning model for generating an anticipatory acceleration command value BIR by performing machine learning based on the learning data acquired by the learning data acquisition unit 19. This learning model is only required to input at least the power generation command value MWD and output an anticipatory acceleration command value BIR in which the tracking ability of the power generation amount E with respect to the power generation command value MWD exceeds a certain level, and the type of machine learning is not particularly limited. For example, this machine learning may be supervised learning such as a support vector machine (SVM), or may be reinforcement learning. Moreover, the above machine learning may be machine learning using a neural network, for example, deep learning.

[0050] For example, the learning unit 20 uses the power generation command value MWD of the learning data (data set) as input data, and the advance acceleration command value BIR as correct data for the input value. The learning unit 20 then inputs the input value to the learning model, and learns the parameters (weights) of the learning model so as to minimize the deviation between the output value output from the learning model and the correct data. Note that a process value of the controlled object may be used as the input value. The process value is, for example, one or more of the fuel flow rate, the feedwater flow rate, the air flow rate, the main steam pressure, the ammonia flow rate for denitration, the spray flow rate for controlling reheat steam temperature, and the main steam temperature.

[0051] In this way, the learning unit 20 determines a weighting that generates or predicts an anticipatory acceleration command value BIR at which the tracking ability of the power generation amount E with respect to the power generation command value MWD exceeds a certain level. For example, the learning unit 20 determines a weighting ω that indicates the influence of a predetermined explanatory variable on the objective variable. The explanatory variable in this case is the power generation command value MWD or the process value of the controlled object. Also, the objective variable is the anticipatory acceleration command value BIR.

[0052] The learning model storage unit 21 stores the learning model constructed by the learning unit 20. For example, the learning model storage unit 21 is configured with a HDD (Hard Disk Drive), a non-volatile memory, or the like.

[0053] Next, the functional blocks of the boiler control device 17 when the load of the boiler 3 according to the first embodiment changes will be described with reference to Fig. 4. Fig. 4 is a functional block diagram of the boiler control device 17 according to the first embodiment.

[0054] As shown in Fig. 4, the boiler control device 17 includes an advance control unit 22 and a feedback control unit 23. In the first embodiment, an example will be described in which the boiler control device 17 advances the fuel flow rate, but the invention can be similarly applied to cases in which other control targets, such as the main steam temperature, the feedwater flow rate, the main steam pressure, or the air flow rate, are advanced. In addition, a program for realizing all or part of the functions of the boiler control device 17 by a computer may be recorded in the computer-readable recording medium, and the program recorded in the recording medium may be read by the processor and executed to realize the functions.

[0055] The advance control unit 22 includes a function generator 221 , a BIR generation unit 222 , and an adder 223 .

[0056] The function generator 221 acquires the power generation command value MWD. ​​Then, the function generator 221 converts the power generation command value MWD into a command value for a fuel flow rate (hereinafter referred to as a "fuel flow rate command value") using a preset function. Then, the function generator 221 outputs the converted fuel flow rate command value to the adder 223.

[0057] The BIR generation unit 222 acquires the learning model (trained model) constructed by the learning device 18 from the learning device 18. For example, the BIR generation unit 222 reads out the trained model stored in the trained model storage unit 21. Then, the BIR generation unit 222 inputs the power generation command value MWD to the trained model that has been read out, thereby acquiring the anticipatory acceleration command value BIR from the trained model. For example, the BIR generation unit 222 acquires the anticipatory acceleration command value BIR from the trained model before the load of the boiler 3 actually changes. The BIR generation unit 222 outputs the anticipatory acceleration command value BIR acquired from the trained model to the adder 223.

[0058] The BIR generation unit 222 may obtain the anticipatory acceleration command value BIR from the read trained model by inputting the power generation command value MWD and a current process value (e.g., the current fuel flow rate, the main steam temperature, or both) to the trained model. For example, when the read trained model is trained using a process value, the BIR generation unit 222 may obtain the anticipatory acceleration command value BIR from the trained model by inputting the current value of the process value to the trained model in addition to the power generation command value MWD.

[0059] The adder 223 generates a first command value by adding the fuel flow rate command value output from the function generator 221 and the advance acceleration command value BIR output from the BIR generation unit 222. The adder 223 outputs the generated first command value to the feedback control unit 23.

[0060] The feedback control unit 23 includes a subtractor 231 , a multiplier 232 , a PI controller 233 , and an adder 234 .

[0061] Subtractor 231 obtains, as a process value of the controlled object, for example, a deviation ΔH between the main steam pressure measured by pressure sensor 13 and a preset set value of the process value.

[0062] The multiplier 232 multiplies the deviation ΔH calculated by the subtractor 231 by a coefficient K that is set based on the power generation command value MWD or the like. Then, the multiplier 232 outputs a value (ΔH×K) obtained by multiplying the deviation ΔH by the coefficient K to the PI controller 233.

[0063] The PI controller 233 generates a control command for eliminating the value (ΔH×K) by applying PI control to the value (ΔH×K) output from the multiplier 232. Then, the PI controller 233 outputs the generated control command to the adder 234.

[0064] The adder 234 generates a second command value by adding the control command output from the PI controller 233 and the first command value output from the adder 223. This second command value is an operation command for controlling the controlled object. The adder 234 outputs this second command value to a device for controlling the controlled object, for example, the fuel flow rate regulator 7.

[0065] Next, an example of the flow of the operation of the learning device 18 according to the first embodiment will be described with reference to Fig. 5. Fig. 5 is a flow diagram of an example of the operation of the learning device 18 according to the first embodiment.

[0066] First, the learning data acquiring unit 19 acquires past operation data of the boiler 3 as learning data (step S101). Here, as an example, the learning data acquiring unit 19 may extract, as learning data, operation data in which the tracking ability of the power generation amount E with respect to the power generation command value MWD exceeds a standard from among the past operation data of the boiler 3. For example, the learning data acquiring unit 19 extracts, as learning data, operation data in which the power generation deviation amount is equal to or less than a threshold from among the past operation data obtained for each control cycle of the boiler control device 17. In this way, the learning data acquiring unit 19 can extract, as learning data, operation data in which the tracking ability exceeds a standard. Note that, for example, the operation data is the power generation command value MWD, the advance acceleration command value BIR, and a process value (for example, the process value is one or more of the fuel flow rate, the feedwater flow rate, the air flow rate, the main steam pressure, the ammonia flow rate for denitration, the spray flow rate for controlling reheat steam temperature, and the main steam temperature). However, the learning data acquiring unit 19 may also extract, from the past operation data of the boiler 3, operation data in which the trackability of the power generation amount E with respect to the power generation command value MWD does not exceed a certain standard as learning data.

[0067] Next, the learning unit 20 inputs the power generation command value MWD and the process value from the learning data to the learning model as input values ​​to obtain an output value from the learning model. Then, the learning unit 20 learns the parameters (weights) of the learning model using, for example, an error backpropagation method so as to minimize the deviation between the output value from the learning model and the anticipatory acceleration command value BIR of the learning data (step S102).

[0068] Then, the learning unit 20 stores the constructed learning model (trained model) in the learning model storage unit 21 (step S103).

[0069] As described above, the learning device 18 of the above embodiment includes the learning unit 20 that constructs a learning model for generating an anticipatory acceleration command value BIR for controlling a control target in anticipation when the load of the boiler 3 used for thermal power generation changes. The learning unit 20 generates the learning model by performing machine learning on a data set including the power generation command value MWWD and the anticipatory acceleration command value BIR for thermal power generation used in past operations of the boiler 3 as learning data.

[0070] According to this configuration, the anticipatory acceleration command value BIR that exceeds a certain level of tracking can be generated without the user having to set the anticipatory acceleration command value BIR based on past experience, thereby improving the tracking of the power generation amount E with respect to the power generation command value MWD. Furthermore, the driving data during actual driving using the anticipatory acceleration command value BIR obtained from the learning model of the first embodiment is utilized as learning data. As a result, the learning device 18 can generate the anticipatory acceleration command value BIR with which the tracking ability exceeds a certain level with high accuracy as the actual driving is performed.

[0071] Second embodiment A thermal power generation system B according to the second embodiment will be described below. The thermal power generation system B according to the second embodiment differs from the thermal power generation system A according to the first embodiment in that the method of generating the advance acceleration command value BIR is different. In the drawings, the same or similar parts are denoted by the same reference numerals, and duplicated explanations may be omitted.

[0072] Fig. 6 is a diagram showing an example of a schematic configuration of a thermal power generation system B according to the second embodiment. As shown in Fig. 6, the thermal power generation system B includes a thermal power plant 1 and a plant control system 2b.

[0073] The plant control system 2b controls the power generation amount E generated by the generator 5 by controlling the boiler 3. Here, controlling the boiler 3 means controlling a control target, which is a parameter required for the operation of the boiler 3. This control target is the same as that in the first embodiment.

[0074] Next, a schematic configuration of a plant control system 2b according to a second embodiment will be described. The plant control system 2b includes a boiler control device 17b and a learning device 18b. In the second embodiment, the boiler control device 17b and the learning device 18b are configured as separate devices, but the present disclosure is not limited to this. The boiler control device 17b and the learning device 18b may be configured as an integrated device, as in the first embodiment. For example, an information processing device may have the functions of the boiler control device 17b and the learning device 18b. Each of the boiler control device 17b and the learning device 18b may be configured with a microprocessor such as a CPU or an MPU, or a microcontroller such as an MCU.

[0075] The boiler control device 17b controls each control target based on a power generation command value MWD, which is a command value for the output of thermal power generation (generator 5), to make the power generation amount E follow the power generation command value MWD. ​​This power generation command value MWD is a target value for the power generation amount E generated by the generator 5. The boiler control device 17b may obtain the power generation command value MWD from an external device, or may calculate it using a known technique.

[0076] When the load of the boiler 3 changes, the boiler control device 17b performs advance control, for example, to increase the fuel in the boiler in advance, based on the advance acceleration command value BIR. Note that, as in the first embodiment, the boiler control device 17b of the second embodiment will be described by taking as an example an advance control to advance the fuel flow rate as an advance control of the main steam temperature. However, the advance control of the boiler control device 17b is not limited to this, and the feedwater flow rate may be advanced as an advance control of the main steam pressure. Furthermore, the boiler control device 17b may perform advance control of the air flow rate.

[0077] The learning device 18b constructs a learning model for generating an advance acceleration command value BIR used in advance control. First, the functional blocks of the learning device 18b according to the second embodiment will be described with reference to Fig. 7. Fig. 7 is a functional block diagram of the learning device 18b according to the second embodiment.

[0078] As shown in FIG. 7, the learning device 18b includes a learning data acquiring unit 19b, a learning unit 20b, and a learning model storage unit 21b.

[0079] The learning data acquiring unit 19b acquires, as learning data, past operation data of the boiler 3 and the power generation amount E obtained during the operation when the operation data was obtained. For example, the learning data acquiring unit 19b acquires the operation data and the power generation amount E for each control cycle of the boiler control device 17b. This operation data is the same as that in the first embodiment. Specifically, the learning data acquiring unit 19b acquires, as learning data, a data set of operation data including a power generation command value MWD and an anticipatory acceleration command value BIR used in the past operation of the boiler 3 and the power generation amount E obtained during the operation.

[0080] The learning data acquiring unit 19b may acquire the learning data from the boiler control device 17b at regular intervals, or may acquire the learning data directly from the various sensors (fuel flow sensor 11, air flow sensor 12, pressure sensor 13, temperature sensor 14, feedwater flow sensor 15, power sensor 16, etc.) at regular intervals. Furthermore, the learning data acquiring unit 19b may acquire the learning data by reading the operating data and the power generation amount E that are stored in a storage unit (not shown) provided outside or inside the learning device 18b and are associated with each other.

[0081] The learning unit 20b constructs a learning model for generating the anticipatory acceleration command value BIR by machine learning based on the learning data acquired by the learning data acquisition unit 19b. For example, the learning unit 20b performs machine learning of the characteristics of the boiler 3 based on the learning data acquired by the learning data acquisition unit 19b.

[0082] For example, the learning unit 20b makes a learning model such as deep learning or a neural network learn the relationship between the power generation command value MWD, the anticipatory acceleration command value BIR, and the power generation amount E based on the learning data acquired by the learning data acquisition unit 19b. That is, the learning unit 20b constructs a learning model in which the power generation command value MWD and the anticipatory acceleration command value BIR are input in a control process to output the power generation amount E. As an example, the learning unit 20b constructs the learning model by learning the parameters (weights) of the learning model using, for example, an error backpropagation method so as to minimize the deviation between the output value output by inputting the power generation command value MWD and the anticipatory acceleration command value BIR of the learning data into the learning model as input data and the power generation amount E which is the learning data.

[0083] However, the learning unit 20b according to the present embodiment is not limited to this, and may machine-learn the relationship between the power generation command value MWD, the advance acceleration command value BIR, the process value, and the power generation amount E into a learning model based on the learning data acquired by the learning data acquisition unit 19b. That is, the learning unit 20b may construct a learning model by machine learning that outputs the power generation amount E when the power generation command value MWD, the advance acceleration command value BIR, and the process value are input in the control process. As an example, the learning unit 20b may construct the learning model by machine learning the parameters (weights) of the learning model using, for example, an error backpropagation method so as to minimize the deviation between the output value output by inputting the power generation command value MWD, the advance acceleration command value BIR, and the process value of the learning data into the learning model as input data and the power generation amount E of the learning data.

[0084] The learning model storage unit 21b stores the learning model constructed by the learning unit 20b. For example, the learning model storage unit 21b is configured with a HDD, a non-volatile memory, or the like.

[0085] Next, functional blocks of the boiler control device 17b according to the second embodiment when the load of the boiler 3 changes will be described with reference to Fig. 8. Fig. 8 is a functional block diagram of the boiler control device 17b according to the second embodiment.

[0086] As shown in Fig. 8, the boiler control device 17b includes an advance control unit 22b and a feedback control unit 23. In the second embodiment, an example will be described in which the boiler control device 17b advances the fuel flow rate, but the second embodiment can also be applied to cases in which the main steam pressure, the main steam temperature, the feedwater flow rate, the air flow rate, or the like is advanced. In addition, a program for realizing all or part of the functions of the boiler control device 17b by a computer may be recorded in the computer-readable recording medium, and the program recorded in the recording medium may be read by the processor and executed to realize the functions.

[0087] The advance control unit 22 b includes a function generator 221 , a BIR generation unit 222 b , and an adder 223 .

[0088] The BIR generating unit 222b acquires the learning model (trained model) constructed by the learning device 18b from the learning device 18b. For example, the BIR generating unit 222b reads out the learning model stored in the learning model storage unit 21b. Then, in the control process, the BIR generating unit 222b optimizes the anticipatory acceleration command value BIR using the read-out learning model. For example, in the control process, the BIR generating unit 222b inputs at least the power generation command value MWD obtained from an external device as an input value to the learning model, and acquires the power generation amount E as an output value from the learning model. Then, the BIR generating unit 222b obtains the anticipatory acceleration command value BIR that minimizes the deviation between the power generation command value MWD (input value) and the power generation amount E (output value). However, the BIR generating unit 222b according to the present embodiment is not limited to this, and it is sufficient to obtain the anticipatory acceleration command value BIR that minimizes the deviation between the power generation command value MWD (input value) and the power generation amount E (output value) even if the deviation between the power generation command value MWD (input value) and the power generation amount E (output value) is not minimized.

[0089] Here, in the learning process, if the learning data includes a process value, the BIR generation unit 222b may input the power generation command value MWD and the process value obtained from an external device as input values ​​to the learning model in the control process, and obtain the power generation amount E as an output value from the learning model. Then, the BIR generation unit 222b may obtain the anticipatory acceleration command value BIR that minimizes the deviation between the power generation command value MWD (input value) and the power generation amount E (output value).

[0090] It is preferable that the anticipatory acceleration command value BIR is set before the load on the boiler 3 changes.

[0091] The BIR generation unit 222 b calculates the anticipatory acceleration command value BIR using the learning model constructed by the learning device 18 b, and outputs the anticipatory acceleration command value BIR to the adder 223 .

[0092] The adder 223 generates a first command value by adding the fuel flow rate command value output from the function generator 221 and the advance acceleration command value BIR output from the BIR generation unit 222b. The adder 223 outputs the generated first command value to the adder 234. The adder 234 then generates a second command value by adding the control command output from the PI controller 233 and the first command value output from the adder 223. The adder 234 outputs this second command value to a device for controlling the controlled object, for example, the fuel flow rate adjustment unit 7.

[0093] An example of the flow of the operation of the learning device 18b according to the second embodiment will be described below with reference to Fig. 9A. Fig. 9A is a flow diagram of an example of the operation of the learning device 18b according to the second embodiment.

[0094] First, the learning data acquiring unit 19b acquires, as learning data, a data set including the power generation command value MWD, the advance acceleration command value BIR, and the power generation amount E used in the past operation of the boiler 3 (step S201).

[0095] Next, the learning unit 20b performs machine learning of the characteristics of the boiler 3 using a learning model based on the learning data acquired by the learning data acquisition unit 19b. For example, the learning unit 20b inputs the power generation command value MWD and the anticipatory acceleration command value BIR of the learning data as input values ​​to the learning model to acquire an output value output from the learning model. Then, the learning unit 20b learns the parameters (weights) of the learning model using, for example, an error backpropagation method so as to minimize the deviation between the output value from the learning model and the power generation amount E of the learning data (step S202).

[0096] Then, the learning unit 20b stores the constructed learning model (trained model) in the learning model storage unit 21b (step S203).

[0097] Next, an example of a method for generating the anticipatory acceleration command value BIR of the BIR generating unit 222b according to the second embodiment will be described with reference to Fig. 9B. Fig. 9B is a flow diagram of an example of the operation of the BIR generating unit 222b according to the second embodiment.

[0098] The BIR generating unit 222b reads out the learned model, which is the learning model constructed by the learning device 18b, from the learning model storage unit 21b (step S301). Next, the BIR generating unit 222b inputs the power generation command value MWD obtained from an external device as an input value to the read learning model (trained model) before the load change of the boiler 3 (step S302). Then, the BIR generating unit 222b calculates the anticipatory acceleration command value BIR that minimizes the deviation between the output value output from the learning model and the power generation command value MWD, which is the input value (step S303).

[0099] Here, in steps S302 and 303, the BIR generation unit 222b may input the power generation command value MWD and the process value obtained from an external device into the learning model as input values, and determine the anticipatory acceleration command value BIR that minimizes the deviation between the output value output from the learning model and the power generation command value MWD of the input value.

[0100] As described above, the learning device 18b according to the second embodiment includes the learning unit 20b that constructs a learning model for generating an anticipatory acceleration command value BIR for controlling a controlled object in anticipation when the load of the boiler 3 used for thermal power generation changes. The learning unit 20b generates the learning model by machine learning a data set including the power generation command value MWD for thermal power generation, the anticipatory acceleration command value BIR, and the power generation amount E used in the past operation of the boiler 3 as learning data.

[0101] More specifically, the learning unit 20b in the second embodiment learns the characteristics of the boiler 3 by machine learning the relationship between the power generation command value MWD, the advance acceleration command value BIR, and the power generation amount E based on the learning data acquired by the learning data acquisition unit 19b.

[0102] According to this configuration, the anticipatory acceleration command value BIR whose tracking ability exceeds a certain level can be generated without the user setting the anticipatory acceleration command value BIR based on past experience. For example, the boiler control device 17b can obtain the anticipatory acceleration command value BIR that minimizes the deviation between the power generation command value MWD input to the learned model, which is a learning model constructed by the learning device 18b, and the power generation amount E output from the learned model. Therefore, the tracking ability of the power generation amount E to the power generation command value MWD can be improved. Furthermore, the driving data during actual driving using the anticipatory acceleration command value BIR obtained from the learning model of the second embodiment is utilized as learning data. That is, the learning model is updated by the learning unit 20b based on the latest driving data and the power generation amount E. As a result, the learning device 18b can accurately generate the anticipatory acceleration command value BIR whose tracking ability exceeds a certain level as the actual driving is performed.

[0103] Although the embodiments of the present disclosure have been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and also includes designs within the scope of the gist of the present disclosure.

[0104] (Modification 1) In the above first and second embodiments, the boiler 3 is a pulverized coal boiler configured to burn pulverized coal as fuel, but is not limited thereto. For example, the boiler 3 may be a mixed combustion boiler that uses ammonia fuel in addition to fossil fuel as fuel and mixes the fossil fuel with the ammonia fuel.

[0105] (Modification 2) In the first and second embodiments, the advance acceleration command value BIR may be, for example, any of the items listed below. Main steam pressure command value Fuel flow command value Exhaust gas O2 command value -Gas damper opening command value for reheat steam temperature control Intermediate spray opening command value Air flow rate command value for two-stage combustion - Command value for ammonia control for denitration

[0106] The learning device of the first or second embodiment may be realized in whole or in part by a computer. In this case, the computer may include a processor such as a CPU or a GPU, and a computer-readable recording medium. The learning device may be realized by recording a program for realizing all or in part of the functions of the learning device by a computer in the computer-readable recording medium, and reading and executing the program recorded in the recording medium into the processor. Here, the "computer-readable recording medium" refers to a portable medium such as a flexible disk, an optical magnetic disk, a ROM, a CD-ROM, and a storage device such as a hard disk built into a computer system. Furthermore, the "computer-readable recording medium" may include a medium that dynamically holds a program for a short period of time, such as a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line, and a medium that holds a program for a certain period of time, such as a volatile memory inside a computer system that is a server or a client in that case. The program may be for realizing part of the functions described above, or may be a medium that can realize the functions described above in combination with a program already recorded in the computer system, or may be a medium that is realized using a programmable logic device such as an FPGA. [Industrial Applicability]

[0107] The present disclosure can be used in a learning device that constructs a learning model for generating a proactive acceleration command value for proactively controlling a controlled object when the load of a boiler used for thermal power generation changes, and in a boiler control system that is equipped with the learning device. [Explanation of symbols]

[0108] A,B Thermal power generation system 1. Thermal power plants 2,2b Plant control system 3. Boiler 17,17b Boiler control device 18,18b Learning device 20,20b Learning Section

Claims

1. A learning device for constructing a learning model for generating a precedence acceleration command value for controlling a controlled object in advance when a load change occurs in a boiler used for thermal power generation, comprising: A learning unit that generates the learning model by machine learning a data set including the power generation command value and the advance acceleration command value of the thermal power generation used in the past operation of the boiler as learning data, A command value of a main steam pressure or a command value of a fuel flow rate in the boiler is used as the advance acceleration command value, The learning unit is a learning device that constructs the learning model based on the learning data, inputs at least the power generation command value, and outputs the anticipatory acceleration command value that allows the tracking ability of the power generation amount E to exceed a certain level with respect to the power generation command value.

2. A learning device for constructing a learning model for generating a precedence acceleration command value for controlling a controlled object in advance when a load change occurs in a boiler used for thermal power generation, comprising: A learning unit that generates the learning model by machine learning a data set including the power generation command value and the advance acceleration command value of the thermal power generation used in the past operation of the boiler as learning data, A command value of a main steam pressure or a command value of a fuel flow rate in the boiler is used as the advance acceleration command value, The learning unit is a learning device that includes the amount of power generated by the thermal power generation obtained from past operations of the boiler in the data set.

3. A learning device as described in claim 1 or 2, wherein the learning unit includes process values ​​of the controlled object obtained during past operation of the boiler in the learning data.

4. The learning device described in claim 3, wherein the process value of the controlled object is one or more of the flow rate of fuel supplied to the boiler, the flow rate of water supplied to the boiler, the flow rate of air supplied to the boiler, the ammonia flow rate for denitration, the spray flow rate for reheat steam temperature control, and the steam pressure.

5. A learning device described in any one of claims 1 to 4, wherein the learning unit uses the data set obtained from past operations of the boiler as the learning data.

6. A learning device as described in Claim 2, wherein the learning unit constructs the learning model based on the learning data, using at least the power generation command value and the advance acceleration command value as input and outputting the power generation amount.

7. A learning device according to claim 6, A boiler control device that obtains the power generation amount by inputting at least the power generation command value into a trained model that is the learning model constructed by the learning device; and Equipped with The boiler control device determines the advance acceleration command value that minimizes the deviation between the power generation command value input to the trained model and the power generation amount output from the trained model.

Citation Information

Patent Citations

  • Method and device for process control

    JP1993053605A

  • Advance control method and advance control device

    JP1993265507A

  • Process control method

    JP1994195104A

  • Autotuning method for thermal power plant and thermal power plant controller utilizing the same

    JP1995210208A

  • Feed-forward controller for thermal power plant main control system

    JP1998063307A