Information processing device, information processing method, program, and information processing system

Abstract

To efficiently learn an estimation model for use in estimation of a content of a material in gas.SOLUTION: An information processing apparatus according to the present invention has a processing unit. Based on a change amount of a content at each of a plurality of points in time when a plurality of first input data having both a plurality of explanatory variables and the content of a first material included in gas is acquired, the processing unit determines that each of the plurality of points in time is determined as any one of a first point in time and a second point in time with a smaller change amount than that of the first point in time. The processing unit selects a learning point in time as a point in time of learning data for use in learning a first estimation model for use in estimation of the content based on the plurality of explanatory variables from each of the first point in time and the second point in time. The processing unit learns the first estimation model with use of the first input data acquired at the learning point in time as the learning data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an information processing device, an information processing method, a program, and an information processing system. [Background technology]

[0002] Exhaust gases emitted from combustion devices such as boilers and gas turbines used in thermal power plants and diesel engines can contain substances such as nitrogen oxides (NOx), sulfur oxides (SOx), and carbon monoxide (CO).

[0003] For example, in thermal power plants, denitration control is performed to remove NOx from exhaust gas in order to keep the amount of NOx contained in the exhaust gas within regulatory limits. Denitration control is a process in which a reducing substance (reducing agent) such as ammonia water or urea water is injected into the exhaust gas, causing NOx to react with NH3 on a catalyst and decomposing it into nitrogen (N2) and water (H2O).

[0004] Regarding denitration control, a technology has been proposed that uses an estimation model obtained through machine learning to estimate the amount of NOx generated (content in exhaust gas), and then uses the estimated NOx amount to control the amount of ammonia injected. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Publication No. 2020-192501 Summary of the Invention [Problem to be solved by the invention]

[0006] An object of the present invention is to provide an information processing device, an information processing method, a program, and an information processing system that can more efficiently learn an estimation model for estimating the content of a substance in a gas. [Means for solving the problem]

[0007] An information processing device according to an embodiment includes a processing unit. The processing unit determines, based on an amount of change in content at each of a plurality of times at which a plurality of first input data, each including a plurality of explanatory variables and a content of a first substance contained in a gas, whether each of the plurality of times is a first time or a second time at which the amount of change is smaller than that at the first time. The processing unit selects, from each of the first time and the second time, a learning time, which is a time of learning data for learning a first estimation model that estimates the content from the plurality of explanatory variables. The processing unit learns the first estimation model using the first input data acquired at the learning time as learning data. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram of an information processing system according to an embodiment. [Figure 2] FIG. 10 is a diagram showing examples of important times, non-important times, and learning times. [Figure 3] FIG. 10 is a diagram showing an example of learning time. [Figure 4] 10 is a flowchart of a learning process according to an embodiment. [Figure 5] 4 is a flowchart of a control process according to the embodiment. [Figure 6] FIG. 1 is a hardware configuration diagram of an information processing apparatus according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of an information processing apparatus according to the present invention will be described in detail below with reference to the accompanying drawings.

[0010] In the following, an embodiment will be described in which the present invention is applied to denitration control for removing NOx from exhaust gas. However, applicable control is not limited to this, and the present invention can also be applied to control for removing other substances such as SOx or CO from gas.

[0011] The system (device) that discharges the exhaust gas is, for example, a combustion device such as a boiler, a gas turbine, or a diesel engine, but may be any other system (device). In the following, an example using a boiler used in a thermal power plant or the like will be described.

[0012] In denitrification control, feedback (FB) control is performed using the amount of NOx emitted (amount of emitted NOx). However, the measured value of the amount of emitted NOx is delayed by one to two minutes, so FB control alone is insufficient. Therefore, feedforward (FF) control is also performed using an estimated value of the amount of NOx generated (amount of generated NOx). For FF control, for example, a technology has been proposed in which the amount of generated NOx is estimated based on an empirical formula by the boiler manufacturer. However, because the combustion dynamics of a boiler are complex and it is difficult to obtain information, estimation of NOx amounts based on an empirical formula has low accuracy.

[0013] As a method that does not use an empirical formula, a technology has been proposed that estimates NOx amounts using an estimation model obtained by machine learning, as described above. To use this in FF control, it is necessary to estimate NOx amounts, for example, in units of seconds. However, if input data in units of seconds is used as is, the large number of samples means that it takes a very long time to train the estimation model. Furthermore, while most samples in the input data are steady-state, FF control requires accurate estimation in transient states, so using all the data makes it impossible to train efficiently.

[0014] The input data is data including, for example, a plurality of explanatory variables and the content of NOx (an example of a first substance) contained in the gas. For example, in a thermal power plant, process data can be used as input data. The process data includes, for example, the following data: Measurement data obtained from various sensors installed at various locations in thermal power generation equipment -Processed measurement data Data used to control the start and stop of thermal power generation equipment

[0015] Specifically, the process data is the following data: Rotation speed of thermal power generation equipment -Generator output of thermal power generation equipment ·Inlet NOx amount ·Fuel input amount Combustion air input Pressure and temperature at various points inside the furnace -Various quantities related to the start-up, stop-up and operating status of each mill

[0016] The inlet NOx amount is, for example, the measured amount of NOx at the inlet side of the denitration catalyst in the flue through which the exhaust gas passes. The direct measurement value is the NOx concentration, and the inlet NOx flow rate can be obtained by multiplying this by the exhaust gas flow rate. In other words, the inlet NOx amount corresponds to the amount of NOx contained in the exhaust gas. Hereinafter, the inlet NOx amount, which is the amount of NOx contained in the exhaust gas, will be referred to as the generated NOx amount. At least a portion of the process data other than the generated NOx amount (inlet NOx amount) is used as an explanatory variable. The estimation model is a model for estimating the amount of NOx generated in the exhaust gas from the explanatory variables. Hereinafter, this model will be referred to as the estimation model MA (first estimation model).

[0017] The information processing device according to this embodiment does not select process data at all times, but rather prioritizes process data at times during a transient state, for example, and trains the estimation model MA by selecting the process data as training data. This allows the estimation model MA for estimating the content of a substance in gas to be trained more efficiently. Furthermore, because training can be performed with an emphasis on times during a transient state, estimation in a transient state can be performed more accurately, making it possible to obtain estimation results that are more suitable for denitration control, for example.

[0018] 1 is a block diagram showing an example of the configuration of an information processing system (denitrification control system) according to this embodiment. The information processing system 10 of this embodiment includes an information processing device (denitrification control device) 100 and a power generation device 200.

[0019] The power generation system 200 is, for example, a thermal power generation system. In the power generation system 200, as exhaust gas discharged from a boiler passes through a flue, NOx contained in the gas is decomposed by reacting it with a reducing substance on a denitration catalyst in the flue, thereby sufficiently reducing (removing) the NOx, and the treated gas is released into the outside air from a chimney. The reducing substance for NOx is, for example, ammonia water and urea water.

[0020] The information processing device 100 controls the injection amount of the reducing substance to be injected into the exhaust gas of the power generation device 200. As described above, the information processing device 100 controls the injection amount of the reducing substance by FB control and FF control.

[0021] The following describes the internal configuration of the power generation device 200. The power generation device 200 includes a storage unit 221 and an injection unit 201.

[0022] The memory unit 221 stores various data used in the power generation device 200. For example, the memory unit 221 stores past data 231 and current data 232. The past data 231 includes a plurality of process data obtained in the past and the amount of NOx generated when the process data was obtained. The past data 231 corresponds to input data (first input data) that is mainly used as learning data when learning the estimation model MA. The current data 232 corresponds to input data (second input data) that is used when estimating the amount of NOx generated using the learned estimation model MA. For example, the current data 232 includes process data at the time (present) when the amount of NOx generated is estimated.

[0023] The storage unit 221 can be configured from any commonly used storage medium such as a flash memory, a memory card, a RAM (Random Access Memory), an HDD (Hard Disk Drive), and an optical disk.

[0024] The injection unit 201 injects the reducing substance into the exhaust gas in accordance with the control of the injection amount of the reducing substance by the information processing device 100.

[0025] Each of the above units (injection unit 201) is realized, for example, by one or more processors. For example, each of the above units may be realized by having a processor such as a CPU (Central Processing Unit) execute a program, i.e., by software. Each of the above units may be realized by a processor such as a dedicated IC (Integrated Circuit), i.e., by hardware. Each of the above units may be realized by a combination of software and hardware. When multiple processors are used, each processor may realize one of the units, or may realize two or more of the units.

[0026] The following describes the internal configuration of the information processing device 100. The information processing device 100 includes a storage unit 121, an acquisition unit 101, a determination unit 102, a time selection unit 103, a variable selection unit 104, a learning unit 105, a prediction unit 106, and a control unit 107.

[0027] 1 are elements for performing a process of controlling the injection amount of a reducing substance injected into exhaust gas from the power generation device 200, and other elements are omitted from Fig. 1. Furthermore, each element may be subdivided or combined.

[0028] The storage unit 121 stores various data used by the information processing device 100. For example, the storage unit 121 stores past data 231 and current data 232 received from the power generation device 200, as well as processing results from each unit.

[0029] The storage unit 121 can be configured from any commonly used storage medium such as a flash memory, a memory card, a RAM, a HDD, or an optical disk.

[0030] The acquiring unit 101 acquires various data used in the information processing device 100. For example, the acquiring unit 101 acquires past data 231 and current data 232 from the power generation plant 200. The past data 231 and current data 232 are acquired as time-series data associated with the time of acquisition. For example, the acquiring unit 101 acquires process data, which is the current data 232, in time series, associates the acquisition time with the data, and stores the data in the storage unit 121. The acquiring unit 101 may be configured to acquire the past data 231 or current data 232 associated with the time on the power generation plant 200 side.

[0031] The frequency of acquiring process data (past data 231 and current data 232) may be any value. In the following, an example in which the process data is acquired once per second will be described. The period for acquiring the past data 231 used for learning may be any value, but is preferably a predetermined period such as the past month or year.

[0032] When acquiring the current data 232, the acquiring unit 101 may acquire all explanatory variables included in the process data. Alternatively, the acquiring unit 101 may acquire only explanatory variables selected by a variable selecting unit 104 (described later) from among the explanatory variables included in the process data. This can reduce the time required to acquire the current data 232.

[0033] The determination unit 102 determines whether the multiple times at which the multiple process data included in the past data 231 were acquired are time TA (first time) or time TB (second time) based on the amount of change in the NOx content contained in the exhaust gas. Time TA is, for example, a time at which the amount of change is relatively larger than time TB, and time TB is, for example, a time at which the amount of change is relatively smaller than time TA. Hereinafter, time TA and time TB will be referred to as important time and non-important time, respectively. Also, hereinafter, the number of multiple times, i.e., the number of multiple process data included in the past data 231, will be referred to as K.

[0034] The critical time represents a time when the amount of change in the amount of generated NOx is relatively larger than that of non-critical times. For example, the critical time is a time in a transient state. The non-critical time represents a time when the amount of change in the amount of generated NOx is relatively smaller than that of the critical time. For example, the non-critical time is a time other than the critical time.

[0035] The determination unit 102 determines, for example, K times t k (1≦k≦K) For each, time t k The difference between the maximum and minimum values ​​of the amount of NOx generated at a plurality of times included in a period (hereinafter referred to as the calculation period) that includes at least one of the period before and the period after is calculated as the amount of change. In the following, the number of times included in the calculation period is defined as L, and each time included in the calculation period is defined as t l This is expressed as (1≦l≦L).

[0036] For example, the calculation period is time t k The calculation period for the six seconds is from three seconds before to three seconds after the time t k It includes seven time points: 1 second later, 2 seconds later, and 3 seconds later.

[0037] time t k If the time is set to "14:00:00", the determination unit 102 calculates the amount of change in the amount of generated NOx using the amount of generated NOx at seven points included in the six seconds from three seconds ago (13:59:57) to three seconds after (14:00:03). For example, if the maximum value of the amount of generated NOx within the calculation period is 60 and the minimum value is 20, the determination unit 102 calculates the amount of change in the amount of generated NOx by dividing the difference between the maximum and minimum values ​​by 40 at time t k The change in the amount of NOx generated at (14:00:00) is taken as the change in the amount of NOx generated at (14:00:00).

[0038] The amount of change may be the difference between the amount of generated NOx at the start and end of the calculation period. For example, the determination unit 102 may calculate the difference between the amount of generated NOx at the start (13:59:57) and the amount of generated NOx at the end (14:00:03) at time t k The change in the amount of NOx generated at (14:00:00) is taken as the change in the amount of NOx generated at (14:00:00).

[0039] The determination unit 102 determines K times t kAmong these, a certain percentage (for example, 1%) or a certain number of times are determined as important times in descending order of the amount of change, and times other than the important times are determined as non-important times. The percentage or number is specified, for example, by the user. The percentage or number may be determined, for example, based on the rate at which a transient state occurs.

[0040] The time selection unit 103 selects K times t k The time selection unit 103 selects learning times, which are times of process data to be used as learning data, from the important times and non-important times. For example, the time selection unit 103 randomly selects m important times (m is an integer of 2 or more) from the important times as learning times, and randomly selects n non-important times (n is an integer of 2 or more) from the non-important times as learning times. For example, m and n are set so that the proportion of important times among the learning times is large.

[0041] FIG. 2 is a diagram showing examples of important times, non-important times, and learning times. In FIG. 2, the horizontal axis represents time, and the vertical axis represents the amount of NOx generated. Line 251 represents the change in the amount of NOx generated. Note that the amount of NOx generated may be set at a different time to account for delays. Line 251 can be interpreted as an estimation target. Note that the time values ​​in FIG. 2 represent minutes. That is, FIG. 2 shows the change in the amount of NOx generated from 16:16 to 16:22.

[0042] Range 252 represents a range that includes the time determined to be the important time. Times that fall within a range other than range 252 are non-important times. In Figure 2, the range from approximately 16:17 to approximately 16:20:30 is shown as an example of range 252 that includes important times. Note that the range that includes important times is not necessarily a continuous range like range 252 in Figure 2.

[0043] The time corresponding to the circle is time t k2, in this embodiment, a large number of important times are selected as learning times, while only a small number of unimportant times are selected as learning times.

[0044] Figure 3 shows an example of study times selected over a longer period (15:50 to 16:50) than in Figure 2. Within the same important time range as in Figure 2, there are many circles indicating times selected as study times, while there are fewer circles for non-important times.

[0045] The variable selection unit 104 selects explanatory variables to be used for learning from among the multiple explanatory variables. For example, the variable selection unit 104 uses past data 231 at the learning time to construct an estimation model that estimates the amount of NOx generated from the multiple explanatory variables by a decision tree ensemble method. Hereinafter, this model will be referred to as an estimation model MB (second estimation model). The variable selection unit 104 selects explanatory variables to be used for learning from among the multiple explanatory variables based on the importance of each of the multiple explanatory variables output from the estimation model MB.

[0046] Examples of decision tree ensemble methods include Extra Trees regression, Random Forest regression, Gradient Boosting regression, and Light Gradient Boosting regression. When an estimation model MB is estimated by learning using a decision tree ensemble method, the importance of each of the multiple explanatory variables is also output at the same time. The importance of an explanatory variable in a decision tree ensemble method is calculated, for example, as the average importance of multiple decision trees. The sum of the importance of all explanatory variables is 1. An explanatory variable can be interpreted as being more important the larger its importance value. The importance of an explanatory variable in decision tree regression is calculated as the (normalized) total reduction in squared error brought about by that explanatory variable. The total reduction is the difference between the total squared error of a decision tree without a certain explanatory variable and the total squared error of a decision tree with that explanatory variable. The importance is the value normalized so that the sum of the total reduction in squared error of all explanatory variables is 1.

[0047] The variable selection unit 104, for example, sorts the multiple explanatory variables in order of importance and selects a certain number of explanatory variables with the highest importance.

[0048] The method for estimating the estimation model MB is not limited to the ensemble method of decision trees, but may be other methods that can obtain values ​​equivalent to importance, such as Lasso regression. Furthermore, if all selected variables are used for learning, the variable selection unit 104 may not be provided.

[0049] The learning unit 105 learns the estimation model MA using input data (past data 231) acquired at the learning time as learning data. For example, the learning unit 105 learns the estimation model MA so as to reduce the error between the output when an explanatory variable included in the past data 231 is input and the amount of generated NOx (objective variable) included in the past data 231.

[0050] The learning unit 105 may perform learning using all explanatory variables included in the past data 231 as learning data, or may perform learning using only the explanatory variables selected by the variable selection unit 104 as learning data.

[0051] The estimation model MA may be any model that inputs multiple explanatory variables and outputs an estimated value of the amount of generated NOx, which is the objective variable. Like the estimation model MB, the estimation model MA may be a model that is trained using the ensemble method of decision trees. Since there is a delay in the sensor for the amount of generated NOx, the amount of generated NOx shifted by the time of the delay may be used as the objective variable.

[0052] The determination unit 102, the time selection unit 103, the variable selection unit 104, and the learning unit 105 are mainly used for learning the estimation model MA. The prediction unit 106 and the control unit 107, which will be described below, are used for predicting (estimating) the amount of generated NOx using the learned estimation model MA, and for controlling the injection amount of reducing substances based on the estimated amount of generated NOx.

[0053] The prediction unit 106 predicts the amount of generated NOx by inputting the current data 232 into the estimation model MA. The prediction unit 106 may use the estimation model MA to predict the amount of generated NOx as well as an estimation error of the amount of generated NOx.

[0054] For example, in the decision tree ensemble method, a large number of decision trees are constructed, and the average value of the outputs of the large number of decision trees is output as the estimated value. The standard deviation of the decision tree output can be interpreted as representing the estimation error. In other words, the decision tree ensemble method can output not only the estimated value, but also the estimation error.

[0055] The control unit 107 controls the injection amount of the reducing substance injected by the injection unit 201 of the power generation device 200 through FB control and FF control. With regard to FF control, the control unit 107 controls the injection amount of the reducing substance using the amount of generated NOx predicted by the prediction unit 106. For example, the control unit 107 calculates the injection amount of the reducing substance required to reduce the amount of generated NOx to a predetermined target value, and controls so as to inject the calculated injection amount of the reducing substance.

[0056] The control unit 107 may control the injection amount of the reducing substance using the estimation error together with the predicted amount of generated NOx. For example, the control unit 107 sets a weighted sum of the predicted amount of generated NOx and the estimation error as a new estimated value of the amount of generated NOx, and calculates the injection amount of the reducing substance required to reduce this estimated value. By determining the injection amount of the reducing substance taking the estimation error into consideration, it is possible to suppress deterioration of control performance due to estimation errors.

[0057] The control unit 107 may change the calculation method for the injection amount of the reducing substance depending on whether the estimation error is equal to or greater than a threshold. For example, if the estimation error is equal to or greater than a threshold, the control unit 107 may control the injection amount without using the predicted amount of NOx generated. If the estimation error is equal to or greater than a threshold, the control unit 107 may adopt a method that uses a weighted sum of the amount of NOx generated and the estimation error, and if the estimation error is less than the threshold, may adopt a method that uses only the amount of NOx generated without using the estimation error.

[0058] At least a part of the above units (acquisition unit 101, determination unit 102, time selection unit 103, variable selection unit 104, learning unit 105, prediction unit 106, and control unit 107) may be realized by one processing unit. Each of the above units is realized, for example, by one or more processors. For example, each of the above units may be realized by having a processor such as a CPU execute a program, i.e., by software. Each of the above units may be realized by a processor such as a dedicated IC, i.e., by hardware. Each of the above units may be realized by using a combination of software and hardware. When multiple processors are used, each processor may realize one of the units, or may realize two or more of the units.

[0059] Next, the learning process of the estimation model performed by the information processing device 100 according to this embodiment will be described. Fig. 4 is a flowchart showing an example of the learning process according to this embodiment. The learning process may be performed at any timing. For example, the learning process may be performed during a period when power generation by the power generation device 200 is stopped.

[0060] The acquisition unit 101 acquires past process data and data on the amount of generated NOx (step S101). The determination unit 102 calculates the amount of change in the amount of generated NOx at each time from the acquired data on the amount of generated NOx, and determines whether each time is an important time or a non-important time based on the calculated amount of change at each time (step S102).

[0061] The time selection unit 103 selects a learning time to be used for learning based on the determined important time and non-important time (step S103). The variable selection unit 104 selects explanatory variables to be used for learning based on past process data at the learning time (step S104). The learning unit 105 learns the estimation model MA based on past process data for the selected explanatory variables at the selected learning time and data on the amount of NOx generated (step S105).

[0062] Next, a description will be given of a process for controlling the injection amount of the reducing substance using an estimation model by the information processing device 100 according to this embodiment. Fig. 5 is a flowchart showing an example of the control process according to this embodiment.

[0063] The acquisition unit 101 acquires current process data of the variables selected by the variable selection unit 104 (step S201). The prediction unit 106 uses the acquired current process data as input data and calculates an estimated value and estimation error of the amount of NOx generated using the trained estimation model (step S202). The control unit 107 determines the injection amount of the reducing substance based on the calculated estimated value and estimation error (step S203). The control unit 107 controls the power generation device 200 to inject the determined injection amount of the reducing substance (step S204).

[0064] As described above, the information processing device 100 of this embodiment can learn an estimation model in a short time even when using data in units of seconds. That is, an estimation model for estimating the content of a substance in a gas can be learned more efficiently. Furthermore, since learning can be performed with an emphasis on samples in a transient state, estimation results suitable for denitration control, which are accurate in estimation in a transient state, can be obtained.

[0065] Next, the hardware configuration of the information processing device according to this embodiment will be described with reference to Fig. 6. Fig. 6 is an explanatory diagram showing an example of the hardware configuration of the information processing device according to this embodiment.

[0066] The information processing device of this embodiment includes a control device such as a CPU 51, a storage device such as a ROM (Read Only Memory) 52 and a RAM 53, a communication I / F 54 that connects to a network and communicates, and a bus 61 that connects each part.

[0067] The programs executed by the information processing device according to this embodiment are provided in advance in the ROM 52 or the like.

[0068] The program executed by the information processing device according to this embodiment may be configured to be provided as a computer program product by being recorded in an installable or executable file format on a computer-readable recording medium such as a CD-ROM (Compact Disk Read Only Memory), a flexible disk (FD), a CD-R (Compact Disk Recordable), or a DVD (Digital Versatile Disk).

[0069] Furthermore, the program executed by the information processing device according to this embodiment may be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Also, the program executed by the information processing device according to this embodiment may be provided or distributed via a network such as the Internet.

[0070] The program executed by the information processing device according to this embodiment can cause a computer to function as each unit of the information processing device described above. In this computer, the CPU 51 can read the program from a computer-readable storage medium onto a main storage device and execute the program.

[0071] A configuration example of the embodiment will be described below. (Configuration example 1) determining whether each of the plurality of times is a first time or a second time at which the amount of change in the content of a first substance contained in the gas is smaller than the amount of change in the content of a first substance contained in the gas at each of the plurality of times at which a plurality of first input data each including a plurality of explanatory variables and the content of the first substance contained in the gas is acquired; selecting a learning time, which is a time of learning data for learning a first estimation model that estimates the content from the plurality of explanatory variables, from each of the first time and the second time; learning the first estimation model using the first input data acquired at the learning time as the learning data; Processing section An information processing device comprising: (Configuration example 2) The processing unit calculates, for each of the plurality of times, a difference between a maximum value and a minimum value of the content at a plurality of times included in a period that includes at least one of before and after the time, as the amount of change. The information processing device according to configuration example 1. (Configuration example 3) the processing unit determines a specified proportion or a specified number of the times in descending order of the amount of change among the plurality of times as the first time, and determines the times other than the first time as the second time. The information processing device according to configuration example 1 or 2. (Configuration Example 4) the processing unit randomly selects m (m is an integer of 2 or more) first times from the first times as the learning times, and randomly selects n (n is an integer of 2 or more) second times from the second times as the learning times; The information processing device according to any one of configuration examples 1 to 3. (Configuration Example 5) The processing unit constructing a second estimation model that estimates the content from a plurality of explanatory variables using the first input data by a decision tree ensemble method, and selecting an explanatory variable to be used for learning from the plurality of explanatory variables based on the importance of each of the plurality of explanatory variables output from the second estimation model; learning the first estimation model using the selected explanatory variables and the content amounts included in the first input data acquired at the learning time as learning data; The information processing device according to any one of configuration examples 1 to 4. (Configuration Example 6) The processing unit predicting the content by inputting second input data including a plurality of the explanatory variables into the first estimation model; using the predicted content to control an injection amount of a reducing substance for removing the first substance from the gas; 6. The information processing device according to any one of configuration examples 1 to 5. (Configuration Example 7) The processing unit Using the first estimation model, predict the content and the estimation error of the content; controlling the injection amount of the reducing substance using the predicted content and the estimation error; The information processing device according to configuration example 6. (Configuration Example 8) the processing unit sets a weighted sum of the predicted content and the estimation error as a new estimated value of the content, and calculates an injection amount of the reducing substance necessary to reduce the estimated value. The information processing device according to configuration example 7. (Configuration Example 9) the first time is an important time at which the amount of change is relatively larger than that of the second time, The second time is a non-important time at which the amount of change is relatively smaller than that of the first time. The information processing device according to any one of configuration examples 1 to 8. (Configuration Example 10) The processing unit a determination unit that determines whether each of the plurality of times is the first time or the second time; a time selection unit for selecting the learning time; a learning unit that learns the first estimation model; Equipped with The information processing device according to any one of configuration examples 1 to 9. (Configuration Example 11) An information processing method executed by an information processing device, a determining step of determining whether each of a plurality of times is a first time or a second time at which the amount of change in the content of a first substance contained in the gas is smaller than the amount of change in the content of the first substance contained in the gas at each of a plurality of times at which a plurality of first input data each including a plurality of explanatory variables and the content of the first substance contained in the gas is acquired; a time selection step of selecting a learning time, which is a time of learning data for learning a first estimation model that estimates the content from the plurality of explanatory variables, from each of the first time and the second time; a learning step of learning the first estimation model using the first input data acquired at the learning time as the learning data; An information processing method including: (Configuration Example 12) On the computer, a determining step of determining whether each of a plurality of times is a first time or a second time at which the amount of change in the content of a first substance contained in the gas is smaller than the amount of change in the content of the first substance contained in the gas at each of a plurality of times at which a plurality of first input data each including a plurality of explanatory variables and the content of the first substance contained in the gas is acquired; a time selection step of selecting a learning time, which is a time of learning data for learning a first estimation model that estimates the content from the plurality of explanatory variables, from each of the first time and the second time; a learning step of learning the first estimation model using the first input data acquired at the learning time as the learning data; A program to execute. (Configuration Example 13) An information processing system including an information processing device and a power generation device, the power generation device includes a storage unit configured to store a plurality of first input data each including a plurality of explanatory variables and an amount of a first substance contained in the gas; The information processing device includes: determining whether each of the plurality of times is a first time or a second time at which the amount of change in the content is smaller than that of the first time, based on the amount of change in the content at each of the plurality of times at which the plurality of first input data is acquired; selecting a learning time, which is a time of learning data for learning a first estimation model that estimates the content from the plurality of explanatory variables, from each of the first time and the second time; a processing unit that learns the first estimation model using the first input data acquired at the learning time as the learning data, Information processing system.

[0072] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be embodied in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, and are also included in the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0073] 100 Information processing device 101 Acquisition Department 102 Decision Section 103 Time selection section 104 Variable Selection Section 105 Learning Department 106 Prediction Department 107 Control Unit 121 Storage section 200 Power Generation Equipment 201 Injection part 221 Storage section

Claims

1. determining whether each of the plurality of times is a first time or a second time at which the amount of change in the content of a first substance contained in the gas is smaller than the amount of change in the content of a first substance contained in the gas at each of the plurality of times at which a plurality of first input data each including a plurality of explanatory variables and the content of the first substance contained in the gas is acquired; selecting a learning time, which is a time of learning data for learning a first estimation model that estimates the content from the plurality of explanatory variables, from each of the first time and the second time; learning the first estimation model using the first input data acquired at the learning time as learning data; Processing section An information processing device comprising:

2. The processing unit calculates, for each of the plurality of times, a difference between a maximum value and a minimum value of the content at a plurality of times included in a period including at least one of before and after the time, as the amount of change. The information processing device according to claim 1 .

3. the processing unit determines a specified percentage or a specified number of the times among the plurality of times in descending order of the amount of change as the first time, and determines the times other than the first time as the second time. The information processing device according to claim 1 .

4. the processing unit randomly selects m (m is an integer of 2 or more) first times from the first times as the learning times, and randomly selects n (n is an integer of 2 or more) second times from the second times as the learning times; The information processing device according to claim 1 .

5. The processing unit constructing a second estimation model that estimates the content from a plurality of explanatory variables using the first input data by a decision tree ensemble method, and selecting an explanatory variable to be used for learning from the plurality of explanatory variables based on the importance of each of the plurality of explanatory variables output from the second estimation model; learning the first estimation model using the selected explanatory variables and the content amounts included in the first input data acquired at the learning time as learning data; The information processing device according to claim 1 .

6. The processing unit predicting the content by inputting second input data including a plurality of the explanatory variables into the first estimation model; using the predicted content to control an injection amount of a reducing substance for removing the first substance from the gas; The information processing device according to claim 1 .

7. The processing unit Using the first estimation model, predict the content and the estimation error of the content; controlling the injection amount of the reducing substance using the predicted content and the estimation error; The information processing device according to claim 6 .

8. the processing unit sets a weighted sum of the predicted content and the estimation error as a new estimated value of the content, and calculates an injection amount of the reducing substance necessary to reduce the estimated value. The information processing device according to claim 7 .

9. the first time is an important time at which the amount of change is relatively larger than that of the second time, the second time is a non-important time at which the amount of change is relatively smaller than that of the first time; The information processing device according to claim 1 .

10. The processing unit a determination unit that determines whether each of the plurality of times is the first time or the second time; a time selection unit for selecting the learning time; a learning unit that learns the first estimation model; Equipped with The information processing device according to claim 1 .

11. An information processing method executed by an information processing device, a determining step of determining, based on an amount of change in the content of a first substance contained in a gas at each of a plurality of times at which a plurality of first input data, each of which includes a plurality of explanatory variables and the content of the first substance contained in the gas, is acquired, whether each of the plurality of times is a first time or a second time at which the amount of change is smaller than the first time; a time selection step of selecting a learning time, which is a time of learning data for learning a first estimation model that estimates the content from the plurality of explanatory variables, from each of the first time and the second time; a learning step of learning the first estimation model using the first input data acquired at the learning time as the learning data; An information processing method including:

12. On the computer, a determining step of determining, based on an amount of change in the content of a first substance contained in a gas at each of a plurality of times at which a plurality of first input data, each of which includes a plurality of explanatory variables and the content of the first substance contained in the gas, is acquired, whether each of the plurality of times is a first time or a second time at which the amount of change is smaller than the first time; a time selection step of selecting a learning time, which is a time of learning data for learning a first estimation model that estimates the content from the plurality of explanatory variables, from each of the first time and the second time; a learning step of learning the first estimation model using the first input data acquired at the learning time as the learning data; A program to execute.

13. An information processing system including an information processing device and a power generation device, the power generation device includes a storage unit configured to store a plurality of first input data each including a plurality of explanatory variables and an amount of a first substance contained in the gas; The information processing device includes: determining whether each of the plurality of times is a first time or a second time at which the amount of change in the content is smaller than the amount of change in the content at the first time, based on the amount of change in the content at each of the plurality of times at which the plurality of first input data is acquired; selecting a learning time, which is a time of learning data for learning a first estimation model that estimates the content from the plurality of explanatory variables, from each of the first time and the second time; a processing unit that uses the first input data acquired at the learning time as learning data to learn the first estimation model, Information processing system.

Images