JP7162560B2 - Information processing device, information processing method, information processing program, and garbage incineration plant - Google Patents

Description

The present invention relates to an information processing apparatus and the like for plant control using machine learning.

Conventionally, the development of technology related to the automatic operation of plants such as power plants and waste incineration facilities has been advanced. For example, in Patent Document 1 below, machine learning is performed using history data of intervention operations performed by operators on equipment operating in a plant to generate rule information including conditions for performing intervention operations. A technique for doing so is disclosed.

Japanese Patent Application Laid-Open No. 2017-49801 (published on March 9, 2017)

The technique of Patent Document 1 is based on the premise that the operator's intervention operation is appropriate. However, not all interventions are appropriate, even for experienced operators. In addition, when an inappropriate intervention operation is included in the history data, the inappropriate intervention operation is learned, so there is a problem that effective machine learning cannot be performed.

One aspect of the present invention has been made in view of the above problems, and an object thereof is to provide an information processing apparatus and the like that can select learning data that enables effective machine learning. That's what it is.

In order to solve the above problems, an information processing apparatus according to an aspect of the present invention provides a plurality of pieces of history data indicating a history of plant control by manual operation. A compatibility determination unit that determines whether or not the state of the plant at the time when the manual operation is performed is compatible, and the history data determined to be compatible by the compatibility determination unit, the control to be executed in the plant. is used as learning data for machine learning.

Further, an information processing method according to an aspect of the present invention is an information processing method using an information processing apparatus, wherein for each of a plurality of history data indicating a history of plant control by manual operation, the manual operation indicated in the history data is performed. A determination step of determining whether or not the operation conforms to the state of the plant at the time when the manual operation was performed; and a step of using the power control as learning data for machine learning.

According to one aspect of the present invention, history data of manual operations suitable for the state of the plant can be used as learning data, so that effective machine learning can be performed.

1 is a block diagram showing an example of a main configuration of an information processing apparatus according to Embodiment 1 of the present invention; FIG. FIG. 10 illustrates an example of an operating set filter corresponding to the "Increase dust dispenser speed" classification; It is a figure explaining the outline|summary of the suitability determination in the said information processing apparatus. It is a flow chart which shows an example of processing which the above-mentioned information processor performs. 4 is a flow chart showing another example of processing executed by the information processing apparatus; FIG. 10 is a block diagram showing an example of a main configuration of an information processing apparatus according to Embodiment 2 of the present invention; FIG. 4 illustrates an example method for determining which operating set filter to use; It is a flow chart which shows an example of processing which the above-mentioned information processor performs. 4 is a flow chart showing another example of processing executed by the information processing apparatus; FIG. 11 is a diagram illustrating an outline of suitability determination in an information processing apparatus according to Embodiment 3 of the present invention;

[Embodiment 1]
[Waste incineration plant]
The configuration of the refuse incineration plant 5 of this embodiment will be described based on FIG. In this embodiment, a garbage incineration plant will be described, but the present invention can be applied to an incineration plant that incinerates any object to be incinerated. For example, in addition to general waste incineration plants, it can also be applied to industrial waste incineration plants and sludge incineration plants. These incineration plants may use the heat generated by incineration to generate electricity. In addition, a plant such as a biomass power plant that incinerates mainly for power generation is also included in the category of the incineration plant. Also, the plant is not limited to an incineration plant, and may be any plant that uses at least one machine or device to perform a predetermined treatment (predetermined treatment for a predetermined object, manufacture of a predetermined article, etc.).

The garbage incineration plant 5 is a plant for collecting and incinerating garbage, and as shown in FIG.

The plant equipment 3 includes an incinerator, a dust feeder for supplying waste into the furnace, a grate for moving waste within the furnace, a boiler, a generator, and the like. The plant equipment 3 supplies the heat generated by incinerating the waste to the boiler, and the steam generated from the boiler rotates the turbine of the generator to generate power. In addition, the plant equipment 3 includes a valve for adjusting the flow rate of combustion air, a heating device for adjusting the temperature of the combustion air, a valve for adjusting the flow rate of air supplied to the furnace (for example, a primary air adjustment valve and A valve for secondary air control), etc. are provided.

Information on the operating status of the plant equipment 3, such as the temperature in the combustion furnace, the amount of garbage supplied to the combustion furnace, the amount of steam generated from the boiler, the oxygen concentration contained in the exhaust gas, the flow rate/temperature of the combustion air, etc. The information is detected by sensors or the like provided at various locations of the plant equipment 3 and output to at least one of the plant control unit 106 of the plant control device 2 and the information processing device 1 .

The plant control device 2 includes at least a plant control unit 106 that controls operations of various devices included in the plant equipment 3, and a communication unit 107 that allows the plant control device 2 to communicate with other devices.

In the garbage incineration plant 5, in order to generate power efficiently, it is desired to generate steam in a stable manner. On the other hand, in an incinerator, the combustion state of the incinerator changes due to various factors such as fluctuations in the quality and amount of garbage to be incinerated and fluctuations in the amount of operation of the incinerator. Therefore, it is necessary to control an incinerator that can stably generate steam while responding to changes in the combustion state of the incinerator.

In a conventional general refuse treatment plant, an operator performs various controls on the plant equipment by manual operation while considering the values of various instruments and conditions inside the furnace. On the other hand, in the waste incineration plant 5, it is possible to automatically control the combustion furnace using a learned model created by performing machine learning using history data of manual operations by plant operators. there is

Specifically, the information processing apparatus 1 performs machine learning in advance using history data of manual operations by plant operators as learning data to generate the learned model 115 . Then, the plant control unit 106 of the plant control device 2 performs the control determined to be executed by the generated learned model 115 on the plant equipment 3 .

However, not all manual operations performed by operators are necessarily appropriate, and inappropriate manual operations may be included in the history data. Therefore, in order to perform machine learning efficiently, it is necessary to exclude inappropriate history data and select appropriate history data.

History data is history data of plant control by manual operation, and includes, for example, data such as details of manual operation performed on equipment operating in the plant, amount of operation, operation time, and the like. The content of manual operation is the content of manual operation performed by the operator on equipment operating in the plant. , combustion air flow rate increase/decrease, combustion air temperature increase/decrease, and secondary air flow rate increase/decrease.

Further, the history data includes data input or transmitted to the information processing device 1 by the operator who performed the manual operation, data transmitted to the information processing device 1 from the equipment of the plant equipment 3 manually operated, and 3 or data transmitted to the information processing device 1 via the plant control device 2, or a combination of these data.

[Information processing device]
The configuration of the information processing device 1 will be described in detail based on FIG. FIG. 1 is a block diagram showing an example of the main configuration of an information processing apparatus 1. As shown in FIG.

The information processing device 1 includes a control unit 10 that controls and controls each unit of the information processing device 1, a storage unit 11 that stores various data used by the information processing device 1, an input unit 12 that receives input operations, an information and a communication unit 14 for communicating with other devices.

The control unit 10 includes a history data acquisition unit 101 , a classification unit 102 , a matching determination unit 103 , a learning data selection unit 104 and a machine learning execution unit 105 . The storage unit 11 also stores a history database (DB) 111, an operation set filter 112, a plant state DB 113, a learning DB 114, and a learned model 115. FIG.

The history data acquisition unit 101 acquires multiple pieces of history data from the history DB 111 . The history DB 111 is a database that stores in advance history data of plant control by manual operation performed by an operator. History data is, as described above, data that includes, for example, details of manual operations performed on devices operating in a plant, amounts of operations, operation times, and the like.

The classification unit 102 classifies a plurality of pieces of history data acquired by the history data acquisition unit 101 according to manual operation details indicated in each piece of history data.

For each of a plurality of history data indicating the history of plant control by manual operation, the suitability determination unit 103 determines whether the manual operation indicated in the history data is suitable for the state of the plant at the time when the manual operation was performed. Determine whether or not Note that the plurality of pieces of history data to be determined are the history data classified by the classification unit 102 .

Details of the determination will be described later with reference to FIGS. When the predetermined conditions that are satisfied when manual operation is to be performed are satisfied, the manual operation shown in the history data is considered to be compatible with the state of the plant at the time when the manual operation was performed. You can judge. In addition, the state of the plant in this embodiment means the temperature in the combustion furnace, the amount of garbage supplied to the combustion furnace, the amount of steam generated from the boiler, the oxygen concentration contained in the exhaust gas, etc. Combustion of the incinerator It's about the state.

In this case, the suitability determination unit 103 reads the state data from the plant state DB 113 . The plant state DB 113 is a database that stores data indicating the state of the plant at the time of manual operation. Also, the conformity determination unit 103 reads out the predetermined condition from the manipulation set filter 112 . The operation set filter 112, which will be described in detail later with reference to FIG. 2, indicates predetermined conditions that are satisfied when the plant is in a state in which a predetermined manual operation should be performed.

The learning data selection unit 104 uses the history data determined by the compatibility determination unit 103 to be suitable as learning data for machine learning of the control to be executed in the plant. Specifically, the history data determined by the conformity determination unit 103 to be unsuitable for the plant state are excluded, the history data determined to be suitable are selected as learning data, and stored in the learning DB 114 . The learning DB 114 is a database that stores the selected history data as learning data. The details of the sorting will be described later with reference to FIGS. 3 to 7. FIG.

The machine learning execution unit 105 executes machine learning using the learning data selected by the learning data selection unit 104, and generates a learned model 115 for determining control to be executed in the plant. The generated trained model 115 is stored in the storage unit 11 .

According to the above configuration, it is possible to extract history data suitable for the state of the plant. By using the extracted history data as learning data, the learned model 115 can be generated by efficient machine learning. Experiments by the inventors of the present invention have shown that the accuracy of a trained model constructed using learning data screened based on the manipulation set filter 112 greatly exceeds the accuracy of a trained model constructed without such screening. Outperforming results are obtained.

The garbage incineration plant 5 includes the information processing device 1 and is controlled by the plant control unit 106 . Since the control by the plant control unit 106 is performed based on the highly accurate learned model 115 as described above, the waste incineration plant 5 can perform appropriate plant control corresponding to changes in the combustion state of the incinerator. becomes.

[Details of compliance judgment]
<Operation set filter>
Suitability determination by the suitability determination unit 103 is performed based on whether or not the manual operation of the history data is suitable for the state of the plant at the time when the manual operation was performed. Specifically, the operation set filter 112 is preset with predetermined conditions corresponding to the content of the manual operation. If the state of the plant at the time when the manual operation was performed satisfies this predetermined condition, the conformity determination unit 103 determines the state of the plant at the time when the manual operation was performed according to the operation set filter 112 . If not, it is determined that the plant conditions are not met.

Predetermined conditions corresponding to the manual operation details regarding the garbage incineration plant 5 are, for example, "conditions regarding the temperature in the incinerator", "conditions regarding the amount of steam generated from the incinerator", and "exhaust gas from the incinerator". at least one of "conditions regarding oxygen concentration" and "conditions regarding the amount of garbage in the incinerator".

Any of the above conditions are plant conditions that should be considered in properly determining the manual operations to be performed in a refuse incineration plant. Therefore, according to the above configuration using at least one of these conditions, it is possible to appropriately determine whether or not the manual operation indicated in the history data matches the state of the plant at the time when the manual operation was performed. can do.

Here, the classification unit 102 classifies each history data according to the content of the manual operation shown in each history data, for example, data indicating an increase in speed of the dust feeder and data indicating a reduction in speed of the grate. The suitability determination unit 103 uses the operation set filter 112 according to the classification to determine suitability for each classification.

Based on FIG. 2, the predetermined conditions shown in the manipulation set filter 112 will be specifically described. FIG. 2 is a diagram showing an example of an operating set filter 112 corresponding to the "increase dust dispenser speed" classification. The suitability determination unit 103 uses the operation set filter 112 of FIG. 2 to determine the suitability of each history data classified as "increase in dust feeder speed" with the state of the plant.

In the operation set filter 112 of FIG. 2, filter items are associated with plant states. A combination of this filter item and the state of the plant indicates the "predetermined condition". For example, a combination of the filter item "furnace temperature" and the state "lowering tendency" indicates a "predetermined condition" that the furnace temperature is on the decline.

In addition to this, the operation set filter 112 of FIG. 2 indicates a predetermined condition that the "manipulated amount of generated steam amount control" is "increasing trend". The manipulated variable of the generated steam amount control is the manipulated variable of the control for bringing the measured value (PV) of the generated steam amount closer to the set value (SV). For example, the manipulated variable for control to increase or decrease the amount of refuse to be supplied to the incinerator and the manipulated variable for control to increase or decrease the amount of combustion air supplied correspond to the manipulated variable for controlling the amount of generated steam. Note that this manipulated variable may be a controlled variable in PID control (Proportional-Integral-Differential Controller). Furthermore, for the "exhaust gas O ₂ concentration", a predetermined condition is indicated that the measured value (PV) is greater than the set value (SV) or that it is "increasing". As for the "amount of dust", a predetermined condition is indicated that the measured value (PV) is smaller than the set value (SV) or that it is "decreasing".

All of these conditions are conditions that are satisfied in a plant state in which it is appropriate to perform manual operation of "acceleration of the dust feeder speed". That is, when the furnace temperature tends to decrease, when the manipulated variable of the generated steam amount control tends to be effective, when the exhaust gas O ₂ concentration is PV>SV or tends to increase, and when the amount of dust is PV <SV or decreasing trend, it is appropriate to perform manual operation of "acceleration of dust feeder speed". Therefore, it can be said that the manual operation of "increasing the speed of the dust feeder" is appropriate when as many of these conditions as possible are satisfied.

The suitability determination unit 103 determines whether the state of the plant when the manual operation shown in the history data was performed satisfies each of these conditions. For example, when making a judgment on history data indicating that a manual operation of "increase the speed of the dust feeder" was performed at a certain time, the suitability judgment unit 103 determines whether the temperature in the furnace at that time was on a downward trend. judge. Similarly, the suitability determination unit 103 also determines the “manipulated amount of generated steam amount control”, “exhaust gas O ₂ concentration”, and “dust amount” at that time.

The “tendency” of each condition is set for a predetermined period starting from the time of manual operation shown in the history data, and various numerical values (furnace temperature, etc.) indicating the plant state during that period are obtained from the plant state DB 113. It can be determined by reading.

<Conformity determination of historical data>
The suitability determination unit 103 performs suitability determination of history data based on the determination result for each condition shown in the operation set filter 112 . FIG. 3 is a diagram illustrating an outline of compatibility determination. FIG. 3 shows information indicated in history data and determination results for each filter item set in the operation set filter 112 . The "operation time", "manual operation content", and "manipulation amount" in the figure are the information shown in each history data, and are the time when the manual operation was performed, the manual operation content, and the manual operation amount, respectively. shows the manipulated variable of "Filter Adaptability Judgment Result" in the figure indicates the judgment result of the suitability between the manual operation and the state of the plant for each filter item. "O" is entered for those determined to be compatible, and "X" is entered for those determined to be incompatible.

For example, history data 1 indicates that manual operation to increase the speed of the dust feeder was performed at time 12:15 with an operation amount of +10% (manual operation to increase the speed of the dust feeder by 10% was performed). there is As for the judgment result of filter suitability, "X" (not suitable) is entered for furnace temperature, and "○" (suitable) is entered for operation amount of generated steam amount control, exhaust gas _O2 concentration, and amount of dust. .

The suitability determination unit 103 determines suitability of history data based on each filter suitability determination result. For example, suitability may be determined by the number of "O" or "X" in the filter suitability determination result. If the number of "o"s is large (for example, 3 or more), it may be determined that there is compatibility, and if the number of "crosses" is large (eg, 2 or more), it may be determined that there is no compatibility. History data determined to be incompatible is excluded from history data to be learned by the learning data selection unit 104 . On the other hand, history data determined to be compatible is used as learning target history data.

Further, the rightmost column in FIG. 3 indicates "time until next operation". The “time until the next operation” is the time from when one manual operation is performed until when the next manual operation is performed. Here, it can be seen that the time from the operation of the history data 4 to the operation of the history data 5 is as short as 5 minutes. Also, both the operation of the history data 4 and the operation of the history data 5 are "increase the speed of the dust feeder", and the same operation content is performed multiple times within a short period of time for the same operation target. become.

If the operation target and operation content were the same, it should have been possible to complete the operation with a single manual operation. At least, it cannot be said that the operation efficiency was good.

Therefore, if the same operation target is manually operated multiple times within a predetermined period of time, the history data of those manual operations is considered to be inappropriate as learning data. During the predetermined period, the manual operation with the same operation content was performed multiple times on the same operation target, but the same effect can be obtained even if one manual operation is performed instead of the multiple manual operations. The period should be such that it can be regarded as having been obtained. The predetermined period can be arbitrarily set according to the applied plant and manual operation content. Also, the predetermined period may be determined by the operator, the manager of the plant equipment, or the like. For example, if the predetermined period is 20 minutes, history data 4 and history data 5 are excluded by learning data selection unit 104 and do not become learning data.

In this way, the learning data selection unit 104 does not set history data of a plurality of manual operations performed within a predetermined period of time for the same control target as learning data. As a result, history data indicating inefficient manual operations is not used as learning data, so that more effective machine learning can be performed.

<Process flow>
Next, based on FIG. 4, the flow of processing for determining history data to be used as learning data by the information processing apparatus 1 will be described. FIG. 4 is a flowchart showing an example of processing (information processing method) executed by the information processing device 1 .

In S<b>1 , the history data acquisition unit 101 acquires multiple pieces of history data from the history DB 111 . Then, in S2, the classification unit 102 classifies the history data according to the content of the manual operation indicated in the history data.

In S3, the compatibility determination unit 103 determines the compatibility between the manual operation indicated in the history data and the state of the plant at the time when the manual operation was performed, according to a predetermined condition corresponding to a predetermined operation content for each classification. judge. Specifically, the suitability determination unit 103 determines suitability based on how many of a plurality of predetermined conditions indicated by the manipulation set filter 112 corresponding to the classification are satisfied.

In S4, the learning data selection unit 104 excludes the history data judged to have low suitability (non-suitability) among the history data judged in S3 from the object of learning data. In other words, the learning data selection unit 104 uses the history data determined to be suitable in S3 as learning data for machine learning of the control to be executed in the plant. However, the history data determined to be suitable in S3 are processed in S5, and the history data excluded in this process do not become learning data.

In S5, the learning data selection unit 104 excludes history data in which the same operation is performed multiple times on the same controlled object within a short period of time from the history data determined to be suitable in S3. do. Note that the process of S5 may be performed prior to S4. Finally, in S6, the learning data selection unit 104 determines history data to be used as learning data. In other words, the learning data selection unit 104 determines to use history data that has not been excluded in either S4 or S5 as learning data.

For example, in the example of FIG. 3, if it is determined that there is compatibility when there are three or more "o"s in the determination result of filter compatibility, history data 2 is excluded in S4. Also, if the time (predetermined period) used as the criterion for determination in S5 is assumed to be 20 minutes, history data 4 and 5 are excluded in S5. Therefore, in this case, in S6, history data 1 and 3 are determined as learning data to be used in machine learning.

[Addition of manipulated variable in history data]
When the same operation content is performed multiple times on the same operation target within a predetermined period, it is possible that the manual operation was appropriate for the control target but inappropriate for the amount of operation. . In this case, if the operation amount of the manual operation performed first is appropriate, it can be said that the operation can be performed only once, which is efficient.

For this reason, the learning data selection unit 104 sets the history data of a plurality of manual operations with the same controlled object performed within a predetermined period as an operation amount obtained by adding the operation amount to the controlled object in the plurality of manual operations. may be included in the learning data after being changed to history data of one manual operation. As a result, history data in which the amount of operation was inappropriate can also be effectively utilized as learning data.

<Process flow>
Next, another example of the flow of processing in the information processing apparatus 1 will be described with reference to FIG. FIG. 5 is a flowchart showing another example of the flow of processing in the information processing apparatus 1. As shown in FIG. Note that the processing from S1 to S4 and S6 is the same as the processing from S1 to S4 and S6 in FIG. 4, so the description will not be repeated. After these processes classify the acquired historical data and determine suitability, data with low suitability are excluded.

In S5a, the learning data selection unit 104 adds the operation amount to the history data in which the same controlled object has been operated multiple times within a predetermined period, and changes it to one manual operation. Include in training data. For example, if the predetermined time is 20 minutes in the example of FIG. +5%", the operation amount "+12%" of the history data 5 is added, the operation amount is corrected to "+17%", and the history data changed to one manual operation is selected as learning data.

[Embodiment 2]
Other embodiments of the invention are described below. For convenience of explanation, the same reference numerals are given to the parts having the same functions as those explained in the first embodiment, and the explanation thereof will not be repeated. The same applies to the third and subsequent embodiments.

<Device configuration>
FIG. 6 is a block diagram showing an example of the main configuration of the information processing apparatus 1 according to the second embodiment. As illustrated, the control unit 10 of the second embodiment includes an operation intention determination unit 151 that determines to which of a plurality of predetermined operation intentions the operation intention of the manual operation indicated in each history data corresponds. It differs from the control unit 10 of the first embodiment. The operation intention is the operator's intention when he/she manually operates the equipment operating in the plant. For example, "I want to increase the amount of garbage supplied to the incinerator", "I want to promote combustion in the incinerator", etc. are examples of operational intentions.

The operation intention may be input and recorded in the history DB 111 when the operator performs a manual operation, for example. In this case, the operation intention determination unit 151 can determine the operation intention of the manual operation indicated by each history data by referring to the history DB 111 . Further, the operation intention determination unit 151 may determine the operation intention of the manual operation based on the plant state when the manual operation is performed. In this case, if the correspondence relationship between the plant state and the operation intention is modeled in advance, the operation intention determination unit 151 can use the model to determine the operation intention of the manual operation indicated by each history data.

The classification unit 102 of the second embodiment classifies history data according to the operation intention determined as described above by the operation intention determination unit 151 . Therefore, in the second embodiment, the operation set filter 112 created in advance for each operation intention is stored in the storage unit 11 .

<Determination of operation set filter based on operation intention>
A method for determining the operation set filter 112 to be used by the conformity determination unit 103 will be described with reference to FIG. 7 . The table in FIG. 7 associates operation intentions, filter numbers, and filtering targets.

In the table of FIG. 7, there are five types of operation intentions: "I want to increase the amount of garbage", "I want to reduce the amount of garbage", "I want to promote combustion", "I want to suppress combustion", and "I want to reduce CO". exemplified. Since filter numbers 1 to 5 are associated with these operational intentions, conformity determination section 103 refers to this table to specify the filter number corresponding to the classification result of classification section 102. can be done. A filter number is a number that identifies an operating set filter 112 (for example, a combination of a plurality of predetermined conditions used for one-time suitability determination for one piece of history data, as in the example of FIG. 2). As described above, in the present embodiment, the predetermined condition used for determining suitability is set in advance for each of a plurality of operational intentions.

In addition, in the table of FIG. 7, a filter target is associated with each operation intention. The conformity determination unit 103 applies the operation set filter 112 to the manual operation of the content indicated by this filtering target. For example, the suitability determination unit 103 selects the history data classified as the operation intention of “want to increase the amount of dust” to filter “increase the speed of the dust dispenser” or “increase the speed of the grate”. manual operation history data, the operation set filter 112 with the filter number “1” is used.

<Process flow>
Next, based on FIG. 8, the flow of processing executed by the information processing apparatus 1 according to the second embodiment will be described. FIG. 8 is a flow chart showing an example of the flow of processing executed by the information processing apparatus 1 of the second embodiment.

In S<b>21 , the history data acquisition unit 101 acquires multiple pieces of history data from the history DB 111 . In subsequent S22, the operation intention determination unit 151 determines the operation intention of each history data. Then, in S23, the classification unit 102 classifies the history data based on the operation intention determined in S22.

In S24, the compatibility determination unit 103 uses the operation set filter 112 corresponding to the classification in S23 to determine compatibility between the state of the plant and the operation content for each history data acquired in S21. For example, when using the table of FIG. 7, the conformity determination unit 103 refers to the table to specify the operation set filter 112 corresponding to the classification of S23. Then, using the specified operation set filter 112, the compatibility determination unit 103 determines the compatibility of the history data of the manual operations that are filtered in the above table.

Note that the suitability determination unit 103 determines that history data in which the operation intention and the filter target do not match is incompatible. For example, in the example of FIG. 7, the operation intention corresponds to "I want to increase the amount of dust", but the performed manual operation is a manual operation other than "increase the speed of the dust feeder" and "increase the speed of the fire grate". Conformance determination unit 103 determines that certain history data is not conforming. The processing from S25 to S27 is the same as the processing from S4 to S6 in FIG. 4, so the description is omitted.

As described above, the information processing apparatus 1 of the present embodiment includes the operation intention determination unit 151 that determines which of a plurality of predetermined operation intentions the operation intention of the manual operation indicated in the history data corresponds to. . Further, as described with reference to FIG. 7, the predetermined condition is set in advance for each of a plurality of operational intentions. Then, the compatibility determination unit 103 uses a predetermined condition corresponding to the operation intention to determine whether or not the manual operation indicated in the history data is compatible with the state of the plant at the time when the manual operation was performed. judge. According to this configuration, since the determination is made using the condition corresponding to the operation intention, it is possible to use the history data of the manual operation whose operation intention matches the plant state as the learning data.

<Processing flow (addition of operation amount in history data)>
Also in the present embodiment, as in the example of FIG. 5, the operation amount in the history data may be added and then added to the learning data. This will be described with reference to FIG. FIG. 9 is a flowchart showing another example of the flow of processing in the information processing device 1 of the second embodiment. The description of the processing other than S26a, which is different from the flowchart of FIG. 8, will be omitted.

As shown in FIG. 9, in S26a, the learning data selection unit 104 adds the operation amount of data in which multiple operations are performed on the same control object within a predetermined period to obtain a single manual operation. , and include it in the training data. According to this configuration, history data that is inappropriate as learning data is corrected before being used as learning data, so that history data in which the amount of operation is inappropriate can also be effectively used as learning data.

[Embodiment 3]
Still another embodiment of the present invention will be described based on FIG. The learning data selection unit 104 of the third embodiment differs from the learning data selection unit 104 of the first embodiment in that history data with a short time to the opposite direction operation is excluded from the learning data.

Here, the operation in the opposite direction means a manual operation in which the content of the operation is the same as that of the manual operation that has already been performed on the same operation target, but the amount of operation is changed in the opposite direction. To give a specific example, for example, the operation in the opposite direction to the operation for increasing the speed of the dust feeder means the operation for decelerating the speed of the dust feeder.

In FIG. 10B, "operation time" indicates the time when the operation was performed, "manual operation details" indicates the details of the manual operation, and "operation amount" indicates the direction and amount of the manual operation. . Here, it can be seen that the time from manual operation for history data 4 to manual operation for history data 5 is as short as 5 minutes. The operation target of both the history data 4 and the history data 5 is "the speed of the dust feeder", but the operation amount of the history data 4 is "+5%", that is, the speed increase. The amount is "-7%" or deceleration. Therefore, the manual operation of history data 5 is the operation in the opposite direction to the manual operation of history data 4 .

If an operation in the opposite direction is performed within a predetermined period of time after a certain manual operation was performed, it is considered that the second manual operation was performed because the direction of the first manual operation was inappropriate. Therefore, the learning data selection unit 104 does not select history data of manual operations as learning data when the opposite direction operation is performed within a predetermined period.

In this way, history data indicating manual operations whose operation direction is determined to be inappropriate is excluded as inappropriate learning data, so that more effective machine learning can be performed. As will be described below, when an operation in the opposite direction is performed within a predetermined period after a certain manual operation, the learning data selection unit 104 adds the operation amounts to each of the operation amounts for one manual operation. The operation history data may be changed and then included in the learning data. As a result, history data of operations in the opposite direction can also be effectively utilized as learning data.

<Process flow>
The flow of processing executed by the information processing apparatus 1 according to the third embodiment will be described with reference to FIG. 10(a). FIG. 10A is a flowchart showing an example of the flow of processing executed by the information processing apparatus 1 of the third embodiment. Note that the processes from S31 to S34 and S37 are the same as the processes from S1 to S4 and S6 in FIG. 4, so description thereof will be omitted.

In S35, the learning data selection unit 104 selects, from among the history data determined to be suitable in S33, history data in which the same operation of the same content is performed multiple times within a short period of time on the same controlled object. exclude. Alternatively, the learning data selection unit 104 adds the operation amount of history data in which the same operation is performed multiple times on the same control object within a short period of time, and changes the operation to one manual operation. May be included in learning data.

In S36, the learning data selection unit 104 excludes the history data with a short time to the opposite direction operation from among the history data determined to be suitable in S33 and not excluded in S34. Alternatively, the learning data selection unit 104 may add the operation amount of the opposite direction operation to the operation amount of a certain manual operation, change the operation amount to one manual operation, and then include the operation amount in the learning data. For example, the learning data selection unit 104 adds the operation amount "-7%" of the history data 5 to the operation amount "+5%" of the history data 4, and corrects the operation amount "-2%" once. History data changed to manual operation may be used as learning data.

[Example of realization by software]
The control blocks of the information processing apparatus 1 (especially each block included in the control unit 10 in FIG. 1) may be realized by a logic circuit (hardware) formed in an integrated circuit (IC chip) or the like, or by software. may be realized.

In the latter case, the information processing apparatus 1 is provided with a computer that executes instructions of a program, which is software that implements each function. This computer includes, for example, one or more processors, and a computer-readable recording medium storing the program. In the computer, the processor reads the program (information processing program) from the recording medium and executes it, thereby achieving the object of the present invention. As the processor, for example, a CPU (Central Processing Unit) can be used. As the recording medium, a "non-temporary tangible medium" such as a ROM (Read Only Memory), a tape, a disk, a card, a semiconductor memory, a programmable logic circuit, or the like can be used. In addition, a RAM (Random Access Memory) for developing the above program may be further provided. Also, the program may be supplied to the computer via any transmission medium (communication network, broadcast wave, etc.) capable of transmitting the program. Note that one aspect of the present invention can also be implemented in the form of a data signal embedded in a carrier wave in which the program is embodied by electronic transmission.

[Modification]
Each process described in each of the above embodiments can be changed as appropriate. For example, in the first embodiment, after classifying the history data according to the operation content, the operation set filter 112 selects suitable history data.

For example, in the example of the second embodiment, the matching determination unit 103 may apply all the history data to the operation set filter 112 with the operation intention of "I want to increase dust", and extract history data with a high matching rate. Then, the learning data selection unit 104 selects, among the extracted history data, those whose operation contents match the operation intention of "I want to increase the amount of dust" ) may be selected and used as learning data.

Further, the conformity determination unit 103 determines whether or not the manual operation indicated in the history data conforms to the state of the plant at the time when the manual operation was performed. It can be done based on the situation.

For example, the suitability determination unit 103 may determine that the manual operation is suitable for the state of the plant if the period during which the combustion state in the furnace after the manual operation is normal is longer than or equal to a predetermined time. In this case, if the period during which the combustion state in the furnace after the manual operation is normal is less than the predetermined time, the suitability determination unit 103 determines that the manual operation is not suitable for the state of the plant. Such manual operation history data is excluded without being treated as learning data.

Whether or not the combustion state is normal may be determined based on the plant state DB 113 . For example, the conditions for the temperature in the furnace, the deviation of the amount of generated steam, and the CO concentration of the exhaust gas when the combustion state is normal are set in advance, and the compatibility determination unit 103 determines that the combustion state is normal when these conditions are satisfied. It may be determined that As a specific numerical example, the compatibility determination unit 103 determines that the furnace temperature is 900 to 950° C., the generated steam amount is |PV-SV|/SV<0.05, and the exhaust gas CO concentration is If it is less than 5 ppm, it may be determined that the combustion state is normal. |PV-SV|/SV<0.05 means that the difference between PV and SV is less than 5% of SV.

Moreover, the determination based on the period in which the combustion state is normal and the determination based on the operation set filter 112 may be used together. For example, the learning data selection unit 104 may preferentially use, as learning data, history data that has been determined to be compatible based on the operation set filter 112 and that has a longer period in which the combustion state is normal.

It should be noted that the system configuration of FIG. 1 is an example, and even if the system configuration is changed, it is possible to realize the same function as that of the refuse incineration plant 5 . For example, a system in which another information processing apparatus has the function of the machine learning execution unit 105 may be used. In this case, the information processing apparatus 1 performs up to selection of learning data, and another information processing apparatus performs machine learning using the selected data. Further, for example, the information processing device 1 may be incorporated into the plant control device 2 to form one device.

The present invention is not limited to the above-described embodiments, but can be modified in various ways within the scope of the claims, and can be obtained by appropriately combining technical means disclosed in different embodiments. is also included in the technical scope of the present invention.

1 information processing device 103 suitability determination unit 104 learning data selection unit

Claims (11)

an operation intention determination unit that determines which of a plurality of predetermined operation intentions the operation intention of the manual operation indicated in the history data indicating the history of control of the plant by manual operation corresponds to;
a suitability determination unit that determines, for each of the plurality of history data, whether or not the manual operation indicated in the history data matches the state of the plant at the time when the manual operation was performed;
a learning data selection unit that uses the history data determined by the compatibility determination unit to be compatible as learning data for machine learning of the control to be executed in the plant ,
The conformity determination unit determines that the state data indicating the state of the plant at the time when the manual operation indicated in the history data is performed is a predetermined condition that is satisfied when the plant is in a state in which the manual operation should be performed. is satisfied, the manual operation shown in the history data is determined to be compatible with the state of the plant at the time the manual operation was performed,
The predetermined condition is set in advance for each of the plurality of operation intentions,
The suitability determination unit uses the predetermined condition corresponding to the operation intention to determine whether the manual operation indicated by the history data is suitable for the state of the plant at the time when the manual operation was performed. An information processing apparatus characterized by determining The plant is a garbage incineration plant,
The above-mentioned predetermined conditions include conditions related to the temperature in the incinerator, conditions related to the amount of steam generated from the incinerator, conditions related to the concentration of oxygen contained in exhaust gas from the incinerator, and conditions related to the amount of garbage in the incinerator. 2. An information processing apparatus according to claim 1 , comprising at least one of: For each of a plurality of history data indicating the history of plant control by manual operation, it is determined whether the manual operation indicated in the history data matches the state of the plant at the time when the manual operation was performed. a conformity determination unit to
a learning data selection unit that uses the history data determined by the compatibility determination unit to be compatible as learning data for machine learning of the control to be executed in the plant,
The information processing apparatus, wherein the learning data selection unit does not use history data of a plurality of manual operations performed within a predetermined period of time for the same control target as the learning data. For each of a plurality of history data indicating the history of plant control by manual operation, it is determined whether the manual operation indicated in the history data matches the state of the plant at the time when the manual operation was performed. a conformity determination unit to
a learning data selection unit that uses the history data determined by the compatibility determination unit to be compatible as learning data for machine learning of the control to be executed in the plant,
The learning data selection unit selects the history data of a plurality of manual operations performed within a predetermined period of time for the same control target as one of the operation amounts obtained by adding the operation amounts for the control target in the plurality of manual operations. An information processing apparatus characterized in that the data is changed to history data of manual operations performed one time and then included in the learning data. A machine learning execution unit that generates a learned model for determining control to be executed in the plant by machine learning using the history data that the learning data selection unit uses as learning data. The information processing apparatus according to any one of claims 1 to 4 . 6. The information processing apparatus according to claim 5 , further comprising a plant control unit that performs the control determined to be executed based on the learned model generated by the information processing apparatus for the plant. An information processing method by an information processing device,
an operation intention determination step of determining to which of a plurality of predetermined operation intentions the operation intention of the manual operation indicated in the history data indicating the history of plant control by manual operation corresponds;
a suitability determination step for determining, for each of the plurality of history data, whether or not the manual operation indicated in the history data is suitable for the state of the plant at the time when the manual operation was performed;
a learning data selection step of using the history data determined to be compatible in the compatibility determination step as learning data for machine learning the control to be executed in the plant ,
In the suitability determination step, the state data indicating the state of the plant at the time when the manual operation indicated in the history data is performed is a predetermined condition that is satisfied when the plant is in a state in which the manual operation should be performed. If the conditions are satisfied, determine that the manual operation shown in the history data is compatible with the state of the plant at the time when the manual operation was performed,
The predetermined condition is set in advance for each of the plurality of operation intentions,
In the suitability determination step, using the predetermined condition corresponding to the operation intention, whether or not the manual operation shown in the history data is suitable for the state of the plant at the time when the manual operation was performed. An information processing method characterized by determining whether An information processing method by an information processing device,
For each of a plurality of history data indicating the history of plant control by manual operation, it is determined whether the manual operation indicated in the history data matches the state of the plant at the time when the manual operation was performed. a conformance determination step for
a learning data selection step of using the history data determined to be compatible in the compatibility determination step as learning data for machine learning the control to be executed in the plant,
The information processing method, wherein in the learning data selection step, history data of a plurality of manual operations performed within a predetermined period of time with the same controlled object is not used as the learning data. An information processing method by an information processing device,
For each of a plurality of history data indicating the history of plant control by manual operation, it is determined whether the manual operation indicated in the history data matches the state of the plant at the time when the manual operation was performed. a conformance determination step for
a learning data selection step of using the history data determined to be compatible in the compatibility determination step as learning data for machine learning the control to be executed in the plant,
In the learning data selection step, the history data of a plurality of manual operations performed within a predetermined period of time for the same control object is the operation amount obtained by adding the operation amount for the control object in the plurality of manual operations. An information processing method characterized in that the history data of one manual operation is changed and then included in the learning data. 2. An information processing program for causing a computer to function as the information processing apparatus according to claim 1, the information processing program for causing the computer to function as the operation intention determination unit, the suitability determination unit, and the learning data selection unit. . A waste incineration plant,
including the information processing device according to claim 6 ,
A garbage incineration plant controlled by the plant control unit provided in the information processing device.

Images