Control system and control method

Abstract

To provide a control system and a control method capable of accurately causing a controlled variable to follow a target value in a device to be controlled where a response delay occurs in a change in the controlled variable relative to a change in a manipulated variable.SOLUTION: A prediction model construction unit 62 constructs a first prediction model 90 in which a derivative value of a controlled variable is used as an objective variable, and a derivative value of a manipulated variable offset by an offset time τ1 in the past relative to the controlled variable is included in an explanatory variable. A response delay time estimation unit 63 calculates an error between an output value of the first prediction model 90 and the derivative value of a measured value of the controlled variable, and estimates the offset time τ1 at which the error is minimized as a response display time. An optimal operation model construction unit 64 constructs an optimal operation model 92 which uses a model input value calculated on the basis of a predicted value after the response delay time of the controlled variable as input to the model and the manipulated variable making the controlled variable after the response delay time match the target value as output.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a control system and a control method for controlling a controlled device such as a combustion plant. [Background technology]

[0002] In a combustion plant, it takes time for the steam temperature (controlled variable) to change after the opening of the cooling valve (operated variable) that performs spray water injection is changed. For example, Patent Document 1 is a prior art document that discloses a control system or control method for controlling a controlled device in which a response delay occurs in the change in the controlled variable relative to a change in the operated variable.

[0003] Patent Document 1 describes a thermal power generation facility that includes a thermal power generation system that converts the reaction heat energy of fuel into electric power, a power generation prediction unit that calculates a predicted power generation output of a natural energy power generation system at a future set time, a target calculation unit that calculates a target power generation output of the thermal power generation system at the set time based on the predicted power generation output, a command calculation unit that calculates a command value for the thermal power generation system so that the power generation output of the thermal power generation system becomes the target power generation output, and a command output unit that outputs the command value to the thermal power generation system a response delay time of the thermal power generation system prior to the set time. According to the thermal power generation facility described in Patent Document 1, by controlling the thermal power generation system prior to the set time taking into account the response delay time of the thermal power generation system, fluctuations in the total power generation output of the natural energy power generation system and the thermal power generation system can be effectively suppressed. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2014 / 002274 Summary of the Invention [Problem to be solved by the invention]

[0005] Patent Document 1 does not describe a method for calculating a command value (operation variable) for a thermal power plant to match the power generation output (controlled variable) with a target power generation output. Therefore, a learning model based on a recurrent neural network (RNN) can be used to calculate an operation variable that matches the controlled variable, which has a response delay in response to changes in the operation variable, with the target value. RNNs alone can form a control model and are thought to be able to achieve control that is effective against response delays. However, this leads to very complex control models, which impose a heavy computational load during inference and learning of the control model, posing implementation challenges.

[0006] The present invention has been made in consideration of the above-mentioned problems, and its purpose is to provide a control system and a control method that can accurately make a controlled variable follow a target value in a controlled device in which a response delay occurs in a change in a controlled variable in response to a change in an manipulated variable. [Means for solving the problem]

[0007] In order to achieve the above object, the present invention provides a control system including an operating end for operating a component of a controlled device, a measurement end for measuring a state variable including a controlled variable of the controlled device, a learning unit for learning a change characteristic of the controlled variable with respect to a change in an operating variable of the component, and a control unit for determining the operating variable based on the measured value of the state variable and the change characteristic and outputting the operating variable to the operating end, wherein the learning unit includes a prediction model construction unit for constructing a first prediction model using a differential value of the controlled variable as a response variable and including a differential value of the operating variable offset in the past by an offset time with respect to the controlled variable as an explanatory variable, and ... controlled variable offset in the past by an offset time with respect to the controlled variable as an explanatory variable. and an optimal operation model construction unit that constructs an optimal operation model in which a model input value calculated based on a predicted value of the controlled variable after the response delay time is used as an input to the model and the operation value that makes the controlled variable after the response delay time match a target value, by learning a correspondence relationship between the measured value of the controlled variable and the manipulated variable offset in the past by the response delay time with respect to the measured value of the controlled variable.

[0008] Furthermore, the present invention provides a control method for determining manipulated variables for components of a controlled device based on measured values of state variables including a controlled variable of the controlled device, and controlling the controlled variable by manipulating the component in accordance with the manipulated variables, the control method comprising the steps of: a first step of constructing a first prediction model having a differential value of the controlled variable as a response variable and including, as an explanatory variable, a differential value of the manipulated variable offset in the past by an offset time relative to the controlled variable; a second step of calculating an error between an output value of the first prediction model and a differential value of the measured value of the controlled variable while changing the offset time of the first prediction model, and estimating, as a response delay time, the offset time when the error is minimized; and a third step of constructing an optimal operation model having, as an input to the model, a model input value calculated based on a predicted value of the controlled variable after the response delay time, and an output of the manipulated variable that makes the controlled variable after the response delay time match a target value, by learning a correspondence between the measured value of the controlled variable and the manipulated variable offset in the past by the response delay time relative to the measured value of the controlled variable. [Effects of the Invention]

[0009] According to the present invention, in a controlled device in which a response delay occurs in a change in a controlled variable relative to a change in a manipulated variable, it is possible to make the controlled variable accurately follow a target value. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a schematic diagram showing the overall configuration of a combustion plant. [Figure 2] FIG. 1 is a configuration diagram of a control system for controlling a combustion plant. [Figure 3] FIG. 10 is a flowchart showing the processing of a learning unit. [Figure 4] FIG. 4 is a flowchart showing the processing of the control unit. [Figure 5] FIG. 10 is a flowchart showing the processing of a prediction model construction unit. [Figure 6] FIG. 4 is a flowchart showing the processing of a response delay time estimation unit. [Figure 7]FIG. 4 is a conceptual diagram illustrating the processing of a response delay time estimation unit. [Figure 8] FIG. 10 is a flowchart showing the processing of an optimal operation model constructing unit. [Figure 9] FIG. 10 is a conceptual diagram showing the processing of an optimal operation model constructing unit. [Figure 10] FIG. 10 is a diagram illustrating an example of an operation screen displayed on a user input / output unit. [Figure 11] FIG. 10 is a diagram showing an example of a control result screen displayed on a user input / output unit. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In this embodiment, a combustion plant will be described as an example of a controlled device, but the present invention can be applied to any controlled device in which a response delay occurs in a change in a controlled variable in response to a change in a manipulated variable.

[0012] 1 is a schematic diagram showing the overall configuration of a combustion plant 100. The combustion plant 100 includes a combustion furnace 1 for burning waste as fuel, and a boiler 2 for recovering exhaust heat from the combustion furnace 1.

[0013] The combustion furnace 1 is provided with a main combustion chamber 3 (primary combustion chamber) and a secondary combustion chamber 4. The main combustion chamber 3 is provided with a stoker-type conveying mechanism 5. Downstream of the conveying mechanism 5, a discharge chute 6 is provided to discharge incineration ash from the main combustion chamber 3.

[0014] Primary combustion air 8 sent from blower 7 passes through a conveying mechanism 5 and is supplied to the main combustion chamber 3. Combustion gas 9 in the main combustion chamber 3 is mixed with secondary combustion air (not shown) blown out from the ceiling part on the downstream side of the main combustion chamber 3, and is completely combusted in the secondary combustion chamber 4.

[0015] The outlet of the secondary combustion chamber 4 of the combustion furnace 1 is connected to the boiler 2, and the combustion gas 9 from the combustion furnace 1 flows into the boiler 2. The boiler 2 is provided with a series of combustion gas flow paths consisting of a radiant chamber 10 (first flue), a second flue 11, and a third flue 12. A temperature sensor 13 that detects the temperature of the combustion gas 9 is provided near the inlet of the third flue 12. The combustion gas 9 that has passed through the boiler 2 is discharged from an exhaust port 14 provided in the third flue 12 to an exhaust path (not shown).

[0016] Water pipes (not shown) are laid throughout the walls of the radiant chamber 10 and the second flue 11. The heat recovery water flowing through these water pipes recovers heat from the radiant chamber 10 and the second flue 11, and a portion of it is vaporized into steam, which is returned to the boiler drum 15. The steam from the boiler drum 15 is sent to the superheater 17 via piping 16. The steam sent from the boiler drum 15 is further superheated to a high temperature and pressure as it passes through the superheater 17, and is then sent to the steam turbine (not shown). The steam used in the steam turbine is returned to the boiler drum 15 via a condenser and a condensate tank (not shown).

[0017] A temperature sensor 19 that measures the steam temperature is provided in the pipe 18 on the outlet side of the superheater 17. A cooling valve 20 (component) that cools the pipe 18 by spraying water is provided near the pipe 18. The steam temperature can be controlled by adjusting the amount of water sprayed by changing the opening (operation amount) of the cooling valve 20. The steam temperature, which is a controlled variable of the combustion plant 100, changes in response to changes in the operation amount of the cooling valve 20 and changes in the combustion gas temperature, but it takes time for changes in the operation amount of the cooling valve 20 or changes in the combustion gas temperature to appear as a change in the steam temperature. This time is called the response delay time.

[0018] 2 is a configuration diagram of a control system that controls the combustion plant 100. The control system 200 includes a measurement terminal 30 that measures the state of the combustion plant 100, an operation terminal 40 that operates the combustion plant 100, and a control device 50. In this embodiment, the measurement terminal 30 is composed of temperature sensors 13 and 19 (shown in FIG. 1). In this embodiment, the operation terminal 40 is composed of a drive device (not shown) for the cooling valve 20.

[0019] The control device 50 has a learning unit 60, a control unit 70, and a user input / output unit 80. The learning unit 60 has a database 61, a prediction model construction unit 62, a response delay time estimation unit 63, and an optimal operation model construction unit 64. The control unit 70 has a real-time prediction unit 71, a model input value calculation unit 72, and an operation amount calculation unit 73. The control device 50 is equipped with an arithmetic unit such as a CPU, storage devices such as ROM and RAM, an input / output interface for signal input / output with external devices, etc., and performs processing of each function by expanding and executing programs stored in ROM, etc. on RAM. The user input / output unit 80 is composed of a touch panel monitor, etc.

[0020] The database 61 stores the measured values of the combustion plant 100 measured by the measurement terminal 30 .

[0021] The prediction model construction unit 62 constructs a first prediction model 90 using the differential values of the measurement values stored in the database 61 as explanatory variables and the differential values of the controlled variables as objective variables, and transmits the first prediction model 90 to the response delay time estimation unit 63.

[0022] The response delay time estimation unit 63 estimates a response delay time using the first prediction model, and transmits the response delay time to the prediction model construction unit 62 and the optimal operation model construction unit 64.

[0023] The prediction model construction unit 62 also constructs a second prediction model 91 in which the measurement values (including feature values calculated based on the measurement values) stored in the database 61 are used as explanatory variables and the predicted value of the controlled variable after the response delay time is used as the objective variable, and transmits the second prediction model 91 to the real-time prediction unit 71.

[0024] The optimal operation model construction unit 64 constructs an optimal operation model 92 by a learning process using the measurement values of the operation variables and controlled variables stored in the database 61, in which the predicted value of the controlled variable after the response delay time (including the feature value calculated based on the predicted value) is used as an input to the model and the operation variable that makes the controlled variable match the target value after the response delay time is used as an output, and transmits the constructed optimal operation model to the operation variable calculation unit 73.

[0025] The real-time prediction unit 71 inputs the measurement value (including the feature calculated based on the predicted value) measured at the measurement end 30 into the second prediction model 91, thereby calculating the predicted value after the response delay time of the controlled variable and transmitting it to the model input value calculation unit 72.

[0026] The model input value calculation unit 72 calculates an input value (model input value) to the optimal operation model 92 based on the predicted value of the controlled variable after the response delay time, and transmits it to the operation variable calculation unit 73.

[0027] The operation amount calculation unit 73 inputs the model input value to the optimal operation model 92 to calculate the operation amount, and outputs it to the operation element 40. This makes it possible to make the controlled amount (steam temperature) after the response time of the combustion plant 100 coincide with the target value.

[0028] FIG. 3 is a flowchart showing the processing of the learning unit 60.

[0029] First, the learning unit 60 stores the measured values of the combustion plant 100 in the database 61 (step S101).

[0030] Following step S101, the learning unit 60 transmits the measurement values stored in the database 61 to each functional unit of the learning unit 60 (prediction model construction unit 62, response delay time estimation unit 63, and optimal operation model construction unit 64) (step S102).

[0031] Following step S102, the response delay time estimation unit 63 estimates the response delay time (step S103).

[0032] Following step S103, the response delay time estimation unit 63 transmits the response delay time to the prediction model construction unit 62 and the optimal operation model construction unit 64 (step S104).

[0033] Following step S104, the prediction model constructing unit 62 constructs a second prediction model, and the optimal operation model constructing unit 64 constructs the optimal operation model 92 (step S105).

[0034] Following step S105, the prediction model construction unit 62 transmits the second prediction model to the real-time prediction unit 71 of the control unit 70, and the optimal operation model construction unit 64 transmits the optimal operation model 92 to the operation amount calculation unit 73 of the control unit 70 (step S106).

[0035] FIG. 4 is a flowchart showing the processing of the control unit 70.

[0036] First, the real-time prediction unit 71 acquires the measured values of the combustion plant 100 (step S201).

[0037] Following step S201, the real-time prediction unit 71 inputs the measurement values and feature quantities of the combustion plant 100 into the second prediction model and calculates a predicted value after the response delay time of the controlled variable (step S202). The feature quantities here are defined as a combination of a differential value of the measurement value measured at the measurement end 30, a sampling difference value of the measurement value, an integral value of the measurement value, etc. The combination is specified by the user.

[0038] Following step S202, the real-time prediction unit 71 transmits the predicted value of the controlled variable after the response delay time to the model input value calculation unit 72 (step S203).

[0039] Following step S203, the model input value calculation unit 72 calculates an input value (model input value) to the optimal operation model based on the predicted value of the controlled variable after the response delay time (step S204). The model input value is defined as a combination of the predicted value of the controlled variable after the response delay time, the differential value of the predicted value, etc. The combination is specified by the user.

[0040] Following step S204, the operation amount calculation unit 73 inputs the model input value to the optimal operation model 92 to calculate the operation amount, and outputs it to the operation element 40 (step S205).

[0041] FIG. 5 is a flowchart showing the processing of the prediction model construction unit 62.

[0042] First, the prediction model construction unit 62 extracts the measurement values that affect the controlled variables, as specified by the user (step S301).

[0043] Following step S301, a feature is calculated based on the measurement value extracted in step S301 (step S302). The feature is defined here as a combination of the measurement value, a derivative of the measurement value, a sampling difference value of the measurement value, etc. The combination is specified by the user.

[0044] Following step S302, a multiple regression model is defined in which the feature quantities calculated in step S302 are used as explanatory variables and the predicted value of the response delay time of the controlled quantity is used as a response variable (step S303).

[0045] Following step S303, a second prediction model is generated by calculating coefficients of the multiple regression model using the least squares method (step S304).

[0046] FIG. 6 is a flow diagram showing the processing of the response delay time estimation unit 63, and FIG. 7 is a conceptual diagram showing the same processing.

[0047] The response delay time estimation unit 63 first extracts a measurement value that affects the controlled variable, which is specified by the user (step S401). In the example shown in Fig. 7, the combustion gas temperature is the measurement value that affects the controlled variable.

[0048] Following step S401, the differential values of the controlled variables, the measurement values extracted in step S401, and the manipulated variables are calculated (step S402).

[0049] Following step S402, the timestamps of the measurement values and manipulated variables extracted in step S401 for the controlled variables are offset backward by independent offset times (step S403). The combination of offset times for each timestamp is changed every time step S403 is executed. In the example shown in Fig. 7, the timestamp (time t) of the manipulated variables is offset backward by offset time τ1, and the timestamp of the combustion gas temperature is offset backward by offset time τ2.

[0050] Following step S403, a multiple regression model is defined in which the differential values of the measurement values and manipulated variables extracted in step S401 are used as explanatory variables and the differential value of the controlled variable is used as a response variable (step S404).

[0051] Following step S404, the coefficients of the multiple regression model defined in step S403 are calculated by the least squares method to construct the first prediction model 90 (step S405).

[0052] Following step S405, the error (model error of the first prediction model 90) between the predicted value of the differential value of the controlled quantity calculated using the first prediction model 90 and the differential value of the measured value of the controlled quantity is calculated, and the model error and the offset time τ1 of the manipulated variable are stored as a set in the database 61 (step S406).

[0053] Following step S406, it is determined whether the number of loops exceeds a set value (step S407). The number of loops here refers to the number of times steps S403 to S406 are repeatedly executed. By increasing the number of loops and reducing the range of change in the offset time in step S403, it is possible to improve the accuracy of estimating the response delay time.

[0054] If the determination result in step S407 is No, the process returns to step S403.

[0055] If the determination result in step S407 is Yes, the offset time τ1 stored in the database 61 together with the minimum model error is estimated as the response delay time (step S408).

[0056] FIG. 8 is a flow diagram showing the processing of the optimum operation model constructing unit 64, and FIG. 9 is a conceptual diagram showing the same processing.

[0057] The optimal operation model constructing unit 64 first associates a data sequence of the controlled variables with a data sequence of the operation variables offset in the past by the response delay time, and generates learning data (step S501). In the example shown in Fig. 9, learning data 93 is generated in which a data sequence of the controlled variables from time 15:11:00 onwards is associated with a data sequence of the operation variables from time 15:09:00 onwards, which is offset in the past by the response delay time (120 seconds).

[0058] Following step S501, the differential value of the controlled variable is calculated (step S502).

[0059] Following step S502, an optimal operation model 92 is constructed by an existing optimization method such as reinforcement learning using the learning data generated in step S501 (step S503). The optimal operation model 92 is a multiple regression model in which the controlled variable and the differential value of the controlled variable are used as explanatory variables and the manipulated variable that makes the controlled variable match the target value is used as a response variable.

[0060] 10 is a diagram showing an example of an operation screen displayed on the user input / output unit 80. The operation screen 300 has a control signal selection field 301 for selecting a control signal (controlled variable), an operation signal selection field 302 for selecting an operation signal (operation variable), a target value setting field 303 for setting a target value of the controlled variable, a check field 304 for specifying whether or not the current value (measured value) of the controlled variable is to be included in the input value (model input value) to the control model (optimal operation model 92) (the weighted average of the predicted value and the current value of the controlled variable is set as the model input value), a current value proportion setting field 305 for setting the weight (proportion) of the current value in the weighted average, a measurement signal selection field 306 for selecting a measurement signal related to (affecting) the controlled variable, a learning start button 307 for instructing the learning unit 60 to start a learning process, a control start button 308 for instructing the control unit 70 to start control, and a control stop button 309 for instructing the control unit 70 to stop control.

[0061] FIG. 11 illustrates an example of a control result screen displayed on the user input / output unit 80. The control result screen 400 includes a measurement value trend field 401 for displaying the trend of measurement values; a display signal selection field 402 for selecting the measurement signal whose trend is displayed in the measurement value trend field 401; a display width setting field 403 for setting the display width (maximum and minimum values of the measurement values) of the measurement value trend field 401; a measurement value field 404 for displaying each current measurement value; and a model output field 405 for displaying the output value of the second prediction model (the predicted value of the controlled variable after the response delay time and the derivative of that predicted value), the manipulated variable calculated by the optimal learning model, the response delay time calculated by the first prediction model, and so on. In the figure, "Signal D" selected in the display signal selection field 402 is the controlled variable (steam temperature), and the measurement value trend field 401 displays a trend 406 of the measurement value of signal D (steam temperature). Hovering a cursor 407 over the trend 406 displays a pop-up display of the measurement value and date and time at the cursor position.

[0062] (summary) In this embodiment, a control system 200 includes an actuation end 40 for operating a component 20 of a control target device 100, a measurement end 30 for measuring a state variable including a controlled variable of the control target device 100, a learning unit 60 for learning a change characteristic of the controlled variable with respect to a change in the actuation variable of the component 20, and a control unit for determining the actuation variable based on the measured value of the state variable and the change characteristic and outputting the actuation variable to the actuation end 40. The learning unit 60 includes a prediction model construction unit 62 for constructing a first prediction model 90 in which a differential value of the controlled variable is used as a response variable and the differential value of the actuation variable offset in the past by an offset time τ1 with respect to the controlled variable is used as an explanatory variable, and an offset time τ1 of the first prediction model 90. The present invention includes a response delay time estimation unit 63 that calculates an error between the output value of the first prediction model 90 and a differential value of the measurement value of the controlled variable while changing an offset time τ1, and estimates the offset time τ1 when the error is minimum as a response delay time, and an optimal operation model construction unit 64 that learns a correspondence relationship between the measurement value of the controlled variable and the manipulated variable offset by the response delay time in the past with respect to the measurement value of the controlled variable, and constructs an optimal operation model 92 in which a model input value calculated based on a predicted value of the controlled variable after the response delay time is used as an input to the model, and the manipulated variable that makes the controlled variable after the response delay time match a target value is used as an output.

[0063] Furthermore, in this embodiment, a control method for determining a manipulated variable for a component of a controlled device based on a measured value of a state variable including a controlled variable of the controlled device, and controlling the controlled variable by operating the component in accordance with the manipulated variable, includes the following steps: a first step of constructing a first prediction model, the first prediction model having a derivative of the controlled variable as a response variable and including, as an explanatory variable, a derivative of the manipulated variable offset in the past by an offset time τ1 with respect to the controlled variable; a second step of calculating an error between an output value of the first prediction model and a derivative of the measured value of the controlled variable while changing the offset time τ1 of the first prediction model, and estimating, as a response delay time, the offset time τ1 at which the error is minimized; and a third step of constructing an optimal operation model in which a model input value calculated based on a predicted value of the controlled variable after the response delay time is used as an input to the model, and the manipulated variable that makes the controlled variable after the response delay time match a target value is used as an output.

[0064] According to the present embodiment configured as described above, the controlled variable can be made to accurately track the target value by calculating the manipulated variable that matches the controlled variable after a response delay time using the optimal operation model 92, which is a learning model. Furthermore, by using the correspondence relationship between the measured value of the controlled variable and the manipulated variable offset by the response delay time from the measured value of the controlled variable in the past as learning data, the optimal operation model 92 can be expressed by a simple control model that does not include the influence of response delay, and therefore the calculation load during inference and learning of the optimal operation model 92 can be reduced.

[0065] Furthermore, in this embodiment, the prediction model construction unit 62 constructs a second prediction model in which feature quantities calculated based on the measurement values of the state quantities are used as explanatory variables and the predicted values are used as response variables, and the control unit 70 includes a real-time prediction unit 71 that calculates the predicted values using the second prediction model, a model input value calculation unit 72 that calculates the model input values based on the predicted values, and a manipulated variable calculation unit 73 that inputs the model input values to an optimal operation model 92 to calculate the manipulated variables. Similarly, the control method in this embodiment includes a fourth step of constructing a second prediction model in which feature quantities calculated based on the measurement values of the state quantities are used as explanatory variables and the predicted values are used as response variables, a fifth step of calculating the predicted values using the second prediction model, a sixth step of calculating the model input values based on the predicted values, and a seventh step of inputting the model input values to the optimal operation model 92 to calculate the manipulated variables. This makes it possible to improve the control accuracy of the controlled quantities by the control unit 70.

[0066] In addition, the second prediction model in this embodiment is configured as a regression model, and the feature quantity includes at least one of a differential value, a sampling difference value, and an integral value of the measurement value of the state quantity, thereby improving the accuracy of the second prediction model.

[0067] Furthermore, the model input value calculation unit 72 in this embodiment calculates a weighted average of the predicted value and the measured value of the controlled variable as the model input value, thereby reducing the risk that the model input value will deviate significantly from the actual controlled variable when the predicted value of the controlled variable calculated by the real-time prediction unit 71 is low in accuracy.

[0068] Furthermore, the optimal operation model constructing unit 64 in this embodiment constructs the optimal operation model 92 by value iteration or reinforcement learning, thereby making it possible to improve the accuracy of the optimal operation model 92.

[0069] Furthermore, the controlled device 100 in this embodiment is a combustion plant 100 including a combustion furnace 1 that incinerates waste as fuel, a boiler 2 that recovers exhaust heat from the combustion furnace 1, and a pipe 16 through which steam generated in the boiler 2 flows, and the component 20 is a cooling valve 20 that cools the pipe 16 by spraying water, the controlled variable is the temperature of the steam, and the manipulated variable is the opening of the cooling valve 20. This makes it possible to accurately make the steam temperature of the combustion plant 100 follow the target value.

[0070] It should be noted that the present invention is not limited to the above-described embodiment, and includes various modifications. For example, the above-described embodiment is shown to clearly explain the present invention, and the present invention is not necessarily limited to an embodiment having all of the described configurations. [Explanation of symbols]

[0071] 1...Combustion furnace, 2...Boiler, 3...Main combustion chamber, 4...Secondary combustion chamber, 5...Conveying mechanism, 6...Discharge chute, 7...Blower, 8...Primary combustion air, 9...Combustion gas, 10...Radiation chamber, 11...Second flue, 12...Third flue, 13...Temperature sensor, 14...Exhaust port, 15...Boiler drum, 16...Pipe, 17...Superheater, 18...Pipe, 19...Temperature sensor, 20...Cooling valve (component), 30...Measuring end, 40...Operating end, 50...Control device, 60...Learning unit, 61...Database, 62...Prediction model construction unit, 63...Response delay time estimation unit, 64...Optimal operation model construction unit, 70...Control unit, 71...Real-time prediction unit, 72...Model input value calculation unit, 73...Operating quantity calculation unit, 80... User input / output section, 90...first prediction model, 91...second prediction model, 92...optimal operation model, 93...learning data, 100...combustion plant (controlled device), 200...control system, 300...operation screen, 301...controlled signal selection field, 302...operation signal selection field, 303...target value setting field, 304...check field, 305...current value ratio setting field, 306...measurement signal selection field, 307...start learning button, 308...start control button, 309...stop control button, 400...control result screen, 401...measurement value trend field, 402...display signal selection field, 403...display width setting field, 404...measurement value field, 405...model output field, 406...trend, 407...cursor.

Claims

1. an operating terminal for operating a component of the device to be controlled; a measurement terminal for measuring a state quantity including a controlled quantity of the control target device; a learning unit that learns a change characteristic of the controlled variable relative to a change in the operation variable of the component; a control unit that determines the manipulated variable based on the measured value of the state variable and the change characteristic, and outputs the manipulated variable to the manipulated element, The learning unit a prediction model constructing unit that constructs a first prediction model in which a differential value of the controlled variable is used as a response variable and a differential value of the manipulated variable offset in the past by an offset time with respect to the controlled variable is used as an explanatory variable; a response delay time estimation unit that calculates an error between an output value of the first prediction model and a differential value of a measurement value of the controlled variable while changing the offset time of the first prediction model, and estimates, as a response delay time, the offset time when the error is minimum; and an optimal operation model constructing unit that constructs an optimal operation model in which a model input value calculated based on a predicted value of the controlled variable after the response delay time is used as an input to the model and the controlled variable that makes the controlled variable after the response delay time match a target value by learning a correspondence relationship between the measured value of the controlled variable and the manipulated variable offset in the past by the response delay time with respect to the measured value of the controlled variable, and outputs the optimal operation model. A control system comprising:

2. 2. The control system of claim 1, the prediction model construction unit constructs a second prediction model in which feature quantities calculated based on the measurement values of the state quantities are used as explanatory variables and the prediction values are used as objective variables; and The control unit a real-time prediction unit that calculates the predicted value using the second prediction model; a model input value calculation unit that calculates the model input value based on the predicted value; an operation amount calculation unit that inputs the model input value to the optimal operation model and calculates the operation amount; A control system comprising:

3. 3. The control system of claim 2, the second prediction model is a regression model; The feature quantity includes at least one of a differential value, a sampling difference value, and an integral value of the measurement value of the state quantity. A control system comprising:

4. 3. The control system of claim 2, The model input value calculation unit calculates a weighted average of the predicted value and the measured value of the controlled variable as the model input value. A control system comprising:

5. 2. The control system of claim 1, The optimal operation model construction unit constructs the optimal operation model by a value iteration method or reinforcement learning. A control system comprising:

6. 2. The control system of claim 1, The controlled device is a combustion plant including a combustion furnace for incinerating waste as fuel, a boiler for recovering exhaust heat from the combustion furnace, and piping through which steam generated by the boiler flows, The component is a cooling valve that cools the piping by spraying water, the controlled variable is the temperature of the steam; The manipulated variable is the opening degree of the cooling valve. A control system comprising:

7. A control method for determining an operation amount of a component of a control target device based on a measurement value of a state quantity including a controlled amount of the control target device, and controlling the controlled amount by operating the component in accordance with the operation amount, comprising: a first step of constructing a first prediction model in which a differential value of the controlled variable is used as a response variable and a differential value of the manipulated variable offset in the past by an offset time with respect to the controlled variable is used as an explanatory variable; a second step of calculating an error between an output value of the first prediction model and a differential value of a measurement value of the controlled variable while changing the offset time of the first prediction model, and estimating the offset time when the error is minimum as a response delay time; and a third step of constructing an optimal operation model in which a model input value calculated based on a predicted value of the controlled variable after the response delay time is used as an input to the model and the controlled variable that makes the controlled variable after the response delay time match a target value is used as an output by learning a correspondence relationship between the measured value of the controlled variable and the manipulated variable offset in the past by the response delay time with respect to the measured value of the controlled variable. A control method comprising:

8. 8. The control method according to claim 7, a fourth step of constructing a second prediction model using feature quantities calculated based on the measured values of the state quantities as explanatory variables and using the predicted values as objective variables; a fifth step of calculating the predicted value using the second prediction model; a sixth step of calculating the model input values based on the predicted values; a seventh step of inputting the model input value into the optimal operation model to calculate the operation amount; A control method comprising:

Images