Learning program, estimation program, information processing device, learning method and estimation method

Abstract

To provide a learning program, an estimation program, an information processing device, a learning method and an estimation method capable of suppressing a burden of a manager associated with monitoring of an incinerator.SOLUTION: A learning program according to the present invention causes a computer to execute processing comprising the steps of: extracting feature amounts respectively from a plurality of learning image data on fire in an incinerator photographed at a plurality of continuous timings; calculating, for each of a predetermined number of the plurality of learning image data, aggregated feature amounts from feature amounts corresponding the respective learning image data; calculating aggregated operation information from operation information indicating operation states of the incinerator at timings corresponding respectively to the prescribed number of learning image data; generating a plurality of teacher data by adding the operation information indicating operations having been performed on the incinerator at the timings corresponding to the prescribed number of learning image data, to the aggregated feature amount and the aggregated operation information; and generating a learned model by subjecting the model to machine learning using the plurality of teacher data.SELECTED DRAWING: Figure 8

Description

[Technical field]

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

[0002] The manager of an incinerator (hereinafter simply referred to as the incinerator) that incinerates waste and other incinerated materials (hereinafter simply referred to as incinerated materials) monitors, for example, the incineration status of the incinerated materials in the incinerator. If the manager determines that the incineration status of the incinerated materials in the incinerator is not normal, the manager takes the necessary steps to normalize the status. Specifically, in this case, the manager adjusts, for example, the amount and temperature of air supplied to the incinerator (see Patent Documents 1 and 2). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-008991 [Patent Document 2] JP 2022-161090 A Summary of the Invention [Problem to be solved by the invention]

[0004] Here, in the case of the incinerator as described above, the burden on the manager may become large depending on, for example, the size of the incinerator, the operating hours, etc. Therefore, in the case of the incinerator as described above, it is desired to reduce the burden on the manager associated with monitoring the incinerator, for example.

[0005] Therefore, an object of the present invention is to provide a learning program, an estimation program, an information processing device, a learning method, and an estimation method that make it possible to reduce the burden on managers associated with monitoring incinerators. [Means for solving the problem]

[0006] In order to achieve the above object, the learning model program of the present invention extracts features from multiple training image data about a fire in an incinerator photographed at multiple consecutive timings, calculates an aggregate feature from the feature corresponding to each of the predetermined number of training image data for each of the multiple training image data, calculates aggregate operation information from operation information indicating the operating status of the incinerator at the timing corresponding to each of the predetermined number of training image data for each of the predetermined number of training image data, generates multiple teacher data by adding operation information indicating operations performed on the incinerator at the timing corresponding to the predetermined number of training image data to the aggregate feature corresponding to the predetermined number of training image data and the aggregate operation information corresponding to the predetermined number of training image data for each of the predetermined number of training image data, and generates a learning model by performing machine learning using the multiple teacher data. Effect of the Invention

[0007] According to the learning program, estimation program, information processing device, learning method and estimation method of the present invention, it is possible to reduce the burden on the manager associated with monitoring the incinerator. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of an incineration system 100 in the first embodiment. [Diagram 2] FIG. 2 is a diagram showing an example of a configuration of the information processing device 1 according to the first embodiment. [Diagram 3] FIG. 3 is a diagram for explaining an outline of the first embodiment. [Figure 4] FIG. 4 is a flowchart illustrating details of the learning process in the first embodiment. [Diagram 5] FIG. 5 is a flowchart illustrating details of the learning process in the first embodiment. [Figure 6]FIG. 6 is a flowchart illustrating details of the learning process in the first embodiment. [Figure 7] FIG. 7 is a flowchart illustrating details of the learning process in the first embodiment. [Figure 8] FIG. 8 is a diagram for explaining a specific example of the image data 131. As shown in FIG. [Figure 9] FIG. 9 is a diagram illustrating a specific example of the operation information 132. As shown in FIG. [Figure 10] FIG. 10 is a diagram illustrating a specific example of the operation information 133. As shown in FIG. [Figure 11] FIG. 11 is a diagram for explaining a specific example of the feature amount 134. As shown in FIG. [Figure 12] FIG. 12 is a diagram illustrating a specific example of the teacher data DT. [Figure 13] FIG. 13 is a diagram illustrating details of the learning process in the first embodiment. [Figure 14] FIG. 14 is a diagram illustrating a specific example of the process of S43. [Figure 15] FIG. 15 is a flowchart illustrating details of the estimation process in the first embodiment. [Figure 16] FIG. 16 is a diagram illustrating details of the estimation process in the first embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, the embodiments of the present disclosure will be described with reference to the drawings. However, such description should not be interpreted in a limiting sense, and does not limit the subject matter described in the claims. In addition, various changes, substitutions, and modifications can be made without departing from the spirit and scope of the present disclosure. In addition, different embodiments can be appropriately combined.

[0010] [Configuration example of incineration system 100 in the first embodiment] First, a configuration example of the incineration system 100 in the first embodiment will be described. Fig. 1 is a diagram showing a configuration example of the incineration system 100 in the first embodiment.

[0011] The incineration system 100 has an incinerator 10 that incinerates materials to be incinerated, such as garbage. The incinerator 10 is, for example, a stoker-type incinerator that has an incineration section (not shown) that incinerates the materials to be incinerated, and a post-incineration section (not shown) that further incinerates the materials to be incinerated after the incineration section has incinerated them.

[0012] The incinerator 10 is fitted with, for example, an imaging device (not shown) for imaging the interior of the incinerator 10. The incinerator 10 is also fitted with measuring devices such as a thermometer (not shown) for measuring the temperature inside the incinerator 10 and a concentration meter (not shown) for measuring the concentration of carbon dioxide in the exhaust gas discharged from the incinerator 10. Furthermore, a measuring device including, for example, a flowmeter (not shown) for measuring the flow rate of the air (incineration air) supplied to the incinerator 10 and a thermometer (not shown) for measuring the temperature of the air supplied to the incinerator 10 is fitted to a piping (not shown) for supplying air (incineration air).

[0013] The incineration system 100 also includes, for example, an information processing device 1 that controls the incineration of materials to be incinerated in the incinerator 10, and an operation terminal 2 through which a manager (hereinafter simply referred to as the manager) who monitors the incinerator 10 inputs necessary information. The information processing device 1 is connected to the incinerator 10 and each measuring device attached to a pipe that supplies air to the incinerator 10, for example, via a network NW such as the Internet. The information processing device 1 then acquires measurements from each measuring device at regular intervals, for example, every minute. An example configuration of the information processing device 1 will be described below.

[0014] [Configuration Example of Information Processing Device 1 in First Embodiment] FIG. 2 is a diagram showing an example of a configuration of the information processing device 1 according to the first embodiment.

[0015] The information processing device 1 is, for example, a physical machine or a virtual machine. The information processing device 1 performs, for example, a process of generating a learning model that estimates an operation that needs to be performed on the incinerator 10 (hereinafter, also referred to as a learning process) and a process of estimating an operation that needs to be performed on the incinerator 10 using the learning model generated by the learning process (hereinafter, also referred to as an estimation process).

[0016] 2, the information processing device 1 has, for example, a CPU 101 which is a processor, a memory 102, a communication interface 103, and a storage medium 104. Each unit is connected to each other via a bus 105.

[0017] The storage medium 104 has, for example, a program storage area (not shown) for storing a program (not shown) for performing a learning process or the like.

[0018] Furthermore, the storage medium 104 has, for example, a storage area (not shown) for storing information used when performing a learning process, etc. The storage medium 104 may be, for example, a hard disk drive (HDD) or a solid state drive (SSD).

[0019] The CPU 101 executes, for example, a program (not shown) loaded from the storage medium 104 to the memory 102 to perform learning processing and the like.

[0020] The communication interface 103 communicates, for example, via the network NW with the incinerator 10 and each measuring device attached to a pipe that supplies air to the incinerator 10. The communication interface 103 also communicates, for example, with the operation terminal 2. The operation terminal 2 may be, for example, a PC (Personal Computer).

[0021] [Outline of the first embodiment] Next, an outline of the first embodiment will be described with reference to Fig. 3. The first embodiment will be described with reference to Fig. 3.

[0022] 3, the information processing device 1 realizes various functions including, for example, an information management unit 111, a feature extraction unit 112, a data generation unit 113, a model generation unit 114, a data estimation unit 115, and a data output unit 116. Note that the information management unit 111, the feature extraction unit 112, the data generation unit 113, the model generation unit 114, and the second model generation unit 114 are functions that realize, for example, a learning process. Also, the information management unit 111, the feature extraction unit 112, the data estimation unit 115, and the data output unit 116 are functions that realize, for example, an estimation process.

[0023] In the following, the estimation process will be described on the assumption that it is executed in the information processing device 1, but is not limited thereto. Specifically, the estimation process may be executed in, for example, another information processing device (not shown) different from the information processing device 1.

[0024] First, the functions in the learning process will be described.

[0025] For example, when the information management unit 111 receives input of image data (hereinafter also referred to as learning image data) of a fire that has broken out inside the incinerator 10, the information management unit 111 stores the received input image data in the storage medium 104.

[0026] Furthermore, when the information management unit 111 receives, for example, input of information showing the operating status of the incinerator 10 (hereinafter also referred to as operation information), it stores the received input operation information in the storage medium 104. The operation information may be, for example, information including at least any of the temperature of the incineration unit, the temperature of the post-incineration unit, the carbon dioxide concentration in the exhaust gas discharged from the incinerator 10, the amount of air supplied to the incineration unit, the temperature of the air supplied to the incineration unit, the amount of air supplied to the post-incineration unit, and the temperature of the air supplied to the post-incineration unit.

[0027] Furthermore, when the information management unit 111 receives input of information indicating an operation performed by the manager on the incinerator 10 (hereinafter also referred to as operation information), the information management unit 111 stores the received operation information in the storage medium 104. The operation information may be information including, for example, at least one of adjustment of the amount of air supplied to the incineration unit, adjustment of the temperature of air supplied to the incineration unit, adjustment of the amount of air supplied to the post-incineration unit, and adjustment of the temperature of air supplied to the post-incineration unit.

[0028] The feature extraction unit 112 extracts features from the image data stored in the storage medium 104 by the information management unit 111, for example. The feature may be information including at least one of information indicating the size of the fire reflected in the image data, information indicating the position of the fire reflected in the image data, and information indicating the number of fires reflected in the image data.

[0029] The data generation unit 113 generates multiple pieces of teacher data each including, for example, the features extracted by the feature extraction unit 112, the operation information stored in the storage medium 104 by the information management unit 111, and the operation information stored in the storage medium 104 by the information management unit 111.

[0030] Specifically, the data generation unit 113 generates teacher data including, for example, feature amounts extracted from each image data, operation information at the timing when each image data was captured, and operation information at the timing when each image data was captured, for each of the multiple image data stored in the storage medium 104 by the information management unit 111. Then, the data generation unit 113 stores the multiple generated teacher data in the storage medium 104, for example.

[0031] The model generation unit 114 generates a learning model (hereinafter also simply referred to as a learning model) by, for example, performing machine learning using a plurality of pieces of teacher data stored in the storage medium 104 by the data generation unit 113.

[0032] Next, the functions in the estimation process will be described.

[0033] For example, when the information management unit 111 receives input of new image data (hereinafter also referred to as verification image data) that captures a fire that has broken out inside the incinerator 10, the information management unit 111 stores the new image data that has been received in input in the storage medium 104.

[0034] Furthermore, for example, when the information management unit 111 receives input of new operation information indicating the operation status of the incinerator 10, the information management unit 111 stores the received input of the new operation information in the storage medium 104.

[0035] The feature extraction unit 112 extracts new feature amounts from new image data that the information management unit 111 has stored in the storage medium 104, for example.

[0036] The data estimation unit 115 inputs, for example, new features extracted by the feature extraction unit 112 and new operation information stored in the storage medium 104 by the information management unit 111 to the learning model. Then, the data estimation unit 115 acquires new operation information output from the learning model in conjunction with the input of the new features and the new operation information.

[0037] The data output unit 116 outputs, for example, the new operation information acquired by the data estimation unit 115 to the operation terminal 2. After that, the manager, for example, refers to the operation information output to the operation terminal 2 and performs a necessary operation on the incinerator 10.

[0038] That is, the information processing device 1 in this embodiment generates a learning model by using multiple training data including, for example, new operation information indicating the operating status of the incinerator 10 and features indicating the size and position of the fire in the incinerator 10.

[0039] This allows the information processing device 1 in this embodiment to accurately estimate, for example, the operation that needs to be performed on the incinerator 10. In addition, the information processing device 1 can reduce the burden on the manager involved in monitoring the incinerator 10 by automatically estimating the operation that needs to be performed on the incinerator 10.

[0040] [Details of the learning process in the first embodiment] Next, the learning process in the first embodiment will be described in detail. Figures 4 to 7 are flow charts for explaining the learning process in the first embodiment in detail. Figures 8 to 14 are diagrams for explaining the learning process in the first embodiment in detail.

[0041] As shown in FIG. 4, the information management unit 111 waits until it receives input of image data 131 (learning image data 131) of an image of a fire occurring in the incinerator 10 (NO in S11).

[0042] Then, when the input of the image data 131 is accepted (YES in S11), the information management unit 111 stores, for example, the accepted input of the image data 131 in the storage medium 104 (S12). A specific example of the image data 131 will be described below.

[0043] [Example of Image Data 131] FIG. 8 is a diagram for explaining a specific example of the image data 131. As shown in FIG.

[0044] The photographing device attached to the incinerator 10 photographs, for example, the incineration status of the materials to be incinerated in the incinerator 10. Specifically, the photographing device photographs, for example, image data 131 showing the fire in the incinerator 10, as shown in FIG.

[0045] In this case, the photographing device may photograph image data 131 so that the fire in the incineration section and the fire in the post-incineration section are both shown. In this case, the photographing device may photograph image data 131 so that the fire in the incineration section is shown on the lower side of the image data 131 and the fire in the post-incineration section is shown on the upper side of the image data 131. In addition, the photographing device may photograph image data 131 showing the fire in the incineration section and image data 131 showing the fire in the post-incineration section, respectively.

[0046] Returning to FIG. 5, the information management unit 111 waits, for example, until it receives input of the operation information 132 indicating the operation status of the incinerator 10 (NO in S21).

[0047] Then, when the input of the operation information 132 is accepted (YES in S21), the information management unit 111, for example, stores the accepted input of the operation information 132 in the storage medium 104 (S22). A specific example of the operation information 132 will be described below.

[0048] [Specific example of operation information 132] FIG. 9 is a diagram illustrating a specific example of the operation information 132. As shown in FIG.

[0049] The operation information 132 shown in Fig. 9 has, for example, items such as "incinerator ID" in which the incinerator ID that identifies the incinerator 10 is set, and "time" in which the time at which each piece of information was measured by each measuring device is set. The operation information 132 shown in Fig. 9 also has, for example, items such as "air supply amount (incinerator)" in which the amount of air supplied to the incinerator at each time is set, "air temperature (incinerator)" in which the temperature of the air supplied to the incinerator at each time is set, "air supply amount (post-incinerator)" in which the amount of air supplied to the post-incinerator at each time is set, and "air temperature (post-incinerator)" in which the temperature of the air supplied to the post-incinerator at each time is set.

[0050] Specifically, in the operation information 132 shown in FIG. 9, the information in the first line has, for example, "001" set as the "incinerator ID," "time" set to "2 / 23 12:00," "air supply volume (incinerator section)" set to "100," "air temperature (incinerator section)" set to "400," "air supply volume (post-incinerator section)" set to "100," and "air temperature (post-incinerator section)" set to "450."

[0051] In the operation information 132 shown in Fig. 9, for example, "001" is set as the "incinerator ID", "2 / 23 12:02" is set as the "time", "100" is set as the "air supply amount (incinerator)", "400" is set as the "air temperature (incinerator)", "100" is set as the "air supply amount (post-incinerator)", and "460" is set as the "air temperature (post-incinerator)". Explanation of other information included in Fig. 9 will be omitted.

[0052] Returning to FIG. 6, the information management unit 111 waits, for example, until it receives input of operation information 133 indicating an operation performed by the manager on the incinerator 10 (NO in S31).

[0053] Then, when input of operation information 133 is accepted (YES in S31), the information management unit 111 stores the accepted input operation information 133 in the storage medium 104 (S32), for example. A specific example of the operation information 133 will be described below.

[0054] [Example of operation information 133] FIG. 10 is a diagram illustrating a specific example of the operation information 133. As shown in FIG.

[0055] The operation information 133 shown in FIG. 10 has, for example, the same items as the operation information 132 described with reference to FIG.

[0056] Specifically, in the operation information 133 shown in FIG. 10, the information in the first line has, for example, "001" set as the "incinerator ID," "2 / 23 12:00" set as the "time," and "-" set for each of the "air supply volume (incinerator section)," "air temperature (incinerator section)," "air supply volume (post-incinerator section)," and "air temperature (post-incinerator section)," indicating that no operation has been performed.

[0057] That is, the information in the first line of the operation information 133 shown in FIG. 10 indicates that, for example, at 12:00 on February 23rd, no operation was performed by the administrator.

[0058] In addition, in the operation information 133 shown in FIG. 10, the information on the third line has, for example, "001" set as the "incinerator ID," "time" set to "2 / 23 12:02," "-" set as each of "air supply volume (incinerator section)," "air temperature (incinerator section)," and "air supply volume (post-incinerator section)," and "UP" set for "air temperature (post-incinerator section)," indicating that an operation has been performed to increase the numerical value.

[0059] That is, the information on the third line in the operation information 133 shown in FIG. 10 indicates that, for example, at 12:02 on February 23rd, an operation was performed to increase the temperature of the air supplied to the post-incineration section.

[0060] In addition, in the operation information 133 shown in FIG. 10, the information on the 9th line, for example, has "001" set as the "incinerator ID," "time" set to "2 / 23 12:08," "-" set as each of "air temperature (incinerator section)," "air supply volume (post-incinerator section)," and "air temperature (post-incinerator section)," and "DOWN" set for "air supply volume (incinerator section)," indicating that an operation has been performed to decrease the numerical value.

[0061] That is, the information on the 9th line in the operation information 133 shown in Fig. 10 indicates that, for example, an operation was performed to reduce the amount of air supplied to the incineration unit at 12:08 on February 23. Explanation of the other information shown in Fig. 10 will be omitted.

[0062] 7, the feature extraction unit 112 waits, for example, until it is time to learn (NO in S41). The learning timing may be, for example, the timing when the administrator inputs, to the information processing device 1, information to the effect that the learning process is to be performed.

[0063] Then, when it is time to learn (YES in S41), the feature extraction unit 112 extracts feature amounts 134 from each of the multiple image data 131 stored in the storage medium 104 (S42).

[0064] Specifically, the feature extraction unit 112 specifies each piece of information about the position, contour, size, etc. of the fire included in each piece of image data 131, for example, according to the color and brightness of each part of each piece of image data 131. Then, from each piece of specified information, the feature extraction unit 112 extracts, as feature 134, at least one of the following: an average R value of the entire piece of image data, an average G value of the entire piece of image data, an average B value of the entire piece of image data, an average brightness of the entire piece of image data, a maximum size of one or more fires appearing in each piece of image data 131, a total value of the size of one or more fires appearing in each piece of image data 131, a bottom position of one or more fires appearing in each piece of image data 131, and a number of one or more fires appearing in each piece of image data 131.

[0065] The feature extraction unit 112 may, for example, exclude in advance blurred image data 131 (for example, image data 131 captured at a time when incineration ash is flying in the incinerator 10) from the multiple image data 131 stored in the storage medium 104. The feature extraction unit 112 may, for example, extract feature amounts 134 for each of the image data 131 other than the blurred image data 131 among the multiple image data 131 stored in the storage medium 104. Specific examples of the feature amounts 134 will be described below.

[0066] [Example of feature 134] FIG. 11 is a diagram for explaining a specific example of the feature amount 134. As shown in FIG.

[0067] The feature amount 134 shown in FIG. 11 has, for example, the items "incinerator ID" and "time" as in the operation information 132 described in FIG. 9. The feature amount 134 shown in FIG. 11 has, for example, the items "maximum fire size" in which the maximum size of one or more fire sizes included in each image data 131 is set, "total fire size" in which the total value of one or more fire sizes included in each image data 131 is set, "bottom end of fire" in which the bottom end position of one or more fire positions included in each image data 131 is set, and "number of fires" in which the number of one or more fires included in each image data 131 is set. Note that when the "number of fires" included in each image data 131 is "1", the "total fire size" may be set to the same value as the "maximum fire size".

[0068] Specifically, in feature 134 shown in FIG. 11, the information in the first line has, for example, "001" set as the "incinerator ID," "2 / 23 12:00" set as the "time," "780" set as the "maximum fire size," "780" set as the "total fire size," "820" set as the "bottom of the fire," and "1" set as the "number of fires."

[0069] 11, for example, "001" is set as the "incinerator ID", "2 / 23 12:03" is set as the "time", "770" is set as the "maximum fire size", "770" is set as the "total fire size", "820" is set as the "bottom of the fire", and "1" is set as the "number of fires". Explanation of other information included in FIG. 11 will be omitted.

[0070] The feature extraction unit 112 may extract values ​​set for the "maximum size of the fire", "total size of the fire", "bottom of the fire", and "number of fires" for each of the incineration unit and the post-incineration unit. Specifically, when a fire in the incineration unit is shown on the lower side of the image data 131 and a fire in the post-incineration unit is shown on the upper side of the image data 131, the feature extraction unit 112 may extract values ​​set for the "maximum size of the fire", "total size of the fire", "bottom of the fire", and "number of fires" for the incineration unit from the information of the fire shown on the lower side of the image data 131, and extract values ​​set for the "maximum size of the fire", "total size of the fire", "bottom of the fire", and "number of fires" for the post-incineration unit from the information of the fire shown on the upper side of the image data 131.

[0071] In addition, for example, when the feature extraction unit 112 identifies the number of one or more fires captured in each image data 131, it may not include fires that are smaller than a predetermined size among the one or more fires captured in each image data 131.

[0072] 7, the data generation unit 113 generates a plurality of pieces of teacher data DT (S43) by using, for example, at least a portion of the feature amount 134 extracted in the process of S42, at least a portion of the operation information 132 stored in the storage medium 104 in the process of S22, and at least a portion of the operation information 133 stored in the storage medium 104 in the process of S32. Then, the information management unit 111 stores, for example, the plurality of pieces of teacher data DT generated in the process of S43 in the storage medium 104. Specific examples of the teacher data DT will be described below.

[0073] [Specific examples of teacher data DT] FIG. 12 is a diagram illustrating a specific example of the teacher data DT.

[0074] The teacher data DT shown in Fig. 12 has items such as "maximum fire size" and "total fire size" similar to the feature amount 134 described in Fig. 11. The teacher data DT shown in Fig. 12 also has items such as "air supply volume (incinerator)" and "air temperature (incinerator)" similar to the operation information 132 described in Fig. 9. The teacher data DT shown in Fig. 12 also has items such as "operation information" in which information included in the operation information 133 described in Fig. 10 is set.

[0075] Specifically, in the teacher data DT shown in FIG. 12, the information in the first line includes, for example, at least a portion of the information in the first line of feature 134 described in FIG. 11, at least a portion of the information in the first line of operation information 132 described in FIG. 9, and at least a portion of the information in the first line of operation information 133 described in FIG. 10.

[0076] Furthermore, in the teacher data DT shown in Fig. 12, the information in the second line includes, for example, at least a part of the information in the second line in the feature amount 134 described in Fig. 11, at least a part of the information in the second line in the operation information 132 described in Fig. 9, and at least a part of the information in the second line in the operation information 133 described in Fig. 10. Descriptions of other information included in Fig. 12 will be omitted.

[0077] In addition, the data generation unit 113 may generate each teacher data DT, for example, so that a plurality of feature amounts 134 and a plurality of pieces of operation information 132 corresponding to each of a plurality of image data 131 are included in one teacher data DT.

[0078] This makes it possible for the information processing device 1 to improve the determination accuracy of the learning model even when generating a learning model that does not correspond to time-series data, for example.

[0079] 7, the model generation unit 114, for example, identifies teacher data DT to be excluded (hereinafter also referred to as excluded teacher data DT) from the teacher data DT generated in the process of S43 (S44). The teacher data DT to be excluded is, for example, teacher data DT corresponding to the time period immediately before the time period in which the administrator performed the operation.

[0080] Specifically, the teacher data DT from the third line to the seventh line in the teacher data DT shown in FIG. 12 is teacher data DT (hereinafter also referred to as operation teacher data DT) corresponding to the time period when the manager performed the operation to increase the air temperature in the post-incineration section. Therefore, the model generation unit 114, for example, identifies one or more teacher data DT corresponding to the time period immediately before the time period corresponding to these operation teacher data DT from the teacher data DT shown in FIG. 12 as excluded teacher data DT. Also, the teacher data DT from the ninth line to the eleventh line in the teacher data DT shown in FIG. 12 is operation teacher data DT corresponding to the time period when the manager performed the operation to reduce the air supply amount in the incineration section. Therefore, the model generation unit 114, for example, identifies the teacher data DT corresponding to the time period immediately before the time period corresponding to these operation teacher data DT from the teacher data DT shown in FIG. 12 as excluded teacher data DT.

[0081] More specifically, in this case, the model generation unit 114 specifies, for example, the teacher data DT in the first row and the teacher data DT in the second row, which correspond to the time period (e.g., 2 minutes) immediately before the time period corresponding to the teacher data DT (operation teacher data DT) from the third row to the seventh row, as excluded teacher data DT. Also, in this case, the model generation unit 114 specifies, for example, the teacher data DT in the seventh row and the teacher data DT in the eighth row, which correspond to the time period (e.g., 2 minutes) immediately before the time period corresponding to the teacher data DT (operation teacher data DT) from the ninth row to the eleventh row, as excluded teacher data DT.

[0082] That is, when the manager performs an operation on the incinerator 10, it is considered that the manager may have made a decision to operate the incinerator 10 based on, for example, the state of the incinerator 10 in the time period immediately prior to the time period when the manager actually performed the operation. In other words, in this case, the manager may have judged that, for example, the state of the incinerator 10 in the time period immediately prior to the time period when the manager actually performed the operation was no longer normal, and may have made a decision to operate the incinerator 10. Therefore, it can be determined that the teacher data DT (operation teacher data DT) corresponding to the time period immediately prior to the time period when the manager actually performed the operation may have been, for example, teacher data DT corresponding to the time period when the manager needed to perform the operation.

[0083] Therefore, the model generation unit 114 generates a learning model by using, for example, the teacher data DT other than the excluded teacher data DT among the multiple teacher data stored in the storage medium 104.

[0084] This enables the information processing device 1 to improve the determination accuracy of a learning model generated using the teacher data DT, for example.

[0085] Returning to FIG. 7, the model generation unit 114 generates a learning model by performing machine learning using, for example, the teacher data DT identified in the processing of S43, other than the excluded teacher data DT identified in the processing of S44 (S45).

[0086] Specifically, as shown in FIG. 13, the model generation unit 114 generates a learning model MD by performing machine learning using a plurality of pieces of teacher data DT, for example.

[0087] Then, the information management unit 111 stores, for example, the learning model generated in the process of S45 in the storage medium 104 (S46). A specific example of the process of S43 will be described below.

[0088] [Example of S43 processing] FIG. 14 is a diagram illustrating a specific example of the process of S43.

[0089] When generating multiple pieces of teacher data DT in the process of S43, the data generation unit 113 may, for example, divide the operation information 132 or the feature amount 134 into a predetermined number of pieces in chronological order as shown in Fig. 14. In this case, the data generation unit 113 may, for example, generate aggregated operation information 132a by aggregating a predetermined number of pieces of operation information 132 included in each group for each divided group.

[0090] Specifically, the data generator 113 may generate the aggregated performance information 132a according to, for example, the following formulas (1) and (2).

[0091]

number

[0092]

number

[0093] In the above formula (1) and formula (2), d i indicates the i-th operation information 132 in chronological order, and N A indicates the number (predetermined number) of operation information 132 included in each group, and f AG indicates a function for aggregating the operation information 132 included in each group, and d at indicates the t-th aggregated performance information 132a generated by aggregating the performance information 132 included in each group.

[0094] That is, the data generating unit 113 may generate the aggregated performance information 132a by calculating, for each divided group, the average, variance, or the like of a predetermined number of pieces of performance information 132 included in each group.

[0095] Similarly, in this case, the data generating unit 113 may generate, for each divided group, an aggregated feature 134a by aggregating a predetermined number of features 134 included in each group.

[0096] The data generation unit 113 may generate teacher data DT that includes, for example, aggregated operation information 132a and aggregated feature 134a, and further includes operation information 133 indicating operations performed by the administrator during the time period corresponding to the aggregated operation information 132a and the aggregated feature 134a.

[0097] In addition, when the administrator performs multiple operations during the time period corresponding to the aggregated operation information 132a and the aggregated feature 134a, the data generation unit 113 may include, for example, operation information 133 corresponding to any one of the multiple operations (for example, the operation having the highest predetermined priority) in the teacher data DT.

[0098] That is, for example, when the change in the operating status of the incinerator 10 is small, the data generation unit 113 may generate a lot of teacher data DT with little difference in content. In this case, overlearning may occur in the process of generating the learning model in the model generation unit 114, and the judgment accuracy of the generated learning model may decrease.

[0099] Therefore, the data generation unit 113 generates a plurality of pieces of teacher data DT including, for example, the aggregated operation information 132a and the aggregated feature amount 134a.

[0100] This enables the information processing device 1 to generate, for example, a plurality of teacher data DT each containing different information (in other words, a plurality of teacher data DT each expressing characteristics of the state of the incinerator 10 in the time period corresponding to each group). Therefore, the information processing device 1 can generate, for example, a learning model MD with high judgment accuracy.

[0101] In this case, the data generator 113 may generate each piece of teacher data DT such that a plurality of pieces of aggregated operation information 132a and a plurality of aggregated feature amounts 134a are included in one piece of teacher data DT.

[0102] Specifically, the data generation unit 113 may include a plurality of pieces of aggregated performance information 132a in the same teacher data DT according to, for example, the following formula (3).

[0103]

number

[0104] In the above formula (3), W indicates the number of aggregated performance information 132a included in the same teacher data DT, cancat indicates a function that combines the aggregated performance information 132a, and D indicates the teacher data DT.

[0105] Similarly, the data generation unit 113 may also include a plurality of aggregated feature amounts 134a in the teacher data DT in which a plurality of pieces of aggregated performance information 132a are included.

[0106] Furthermore, for example, when there is aggregated operation information 132a (hereinafter also referred to as excluded aggregated operation information 132a) and aggregate feature 134a (hereinafter also referred to as excluded aggregated feature 134a) corresponding to the time period immediately preceding the time period in which the administrator performed the operation, the data generation unit 113 may generate teacher data DT that includes multiple pieces of aggregated operation information 132a other than the excluded aggregated operation information 132a and multiple aggregated features 134a other than the excluded aggregated feature 134a.

[0107] Specifically, the data generation unit 113 may include a plurality of pieces of aggregated performance information 132a other than the excluded aggregated performance information 132a in the same teacher data DT, for example, according to the following formula (4).

[0108]

number

[0109] In the above formula (4), d aw is the exclusion aggregated performance information 132a corresponding to the timing immediately before the time period in which the administrator performed the operation, and B indicates the number of exclusion aggregated performance information 132a.

[0110] Similarly, the data generation unit 113 may, for example, include multiple aggregated features 134a other than the excluded aggregated feature 134a in the teacher data DT that includes multiple aggregated operation information 132a other than the excluded aggregated operation information 132a.

[0111] This enables the information processing device 1 to further improve the determination accuracy of a learning model generated using the teacher data DT, for example.

[0112] [Details of Estimation Processing in the First Embodiment] Next, the details of the estimation process in the first embodiment will be described. Fig. 15 is a flow chart for explaining the details of the estimation process in the first embodiment. Fig. 16 is a diagram for explaining the details of the estimation process in the first embodiment.

[0113] The information management unit 111 waits until it receives input of, for example, image data 131 (hereinafter also referred to as new image data 131) of a fire that has broken out inside the incinerator 10 and operation information 132 (hereinafter also referred to as new operation information 132) that indicates the operating status of the incinerator 10 (NO in S51).

[0114] Specifically, the information management unit 111 waits, for example, until new image data 131 captured by a photographing device of the incinerator 10 and new operation information 132 including each piece of information measured by each measuring device of the incinerator 10, etc., are input into the information processing device 1.

[0115] In addition, the information contained in the new image data 131 and the new operation information 132 may be, for example, acquired by the information management unit 111 from an imaging device of the incinerator 10 or each measuring device of the incinerator 10, etc.

[0116] Then, when input of new image data 131, etc. is accepted (YES in S51), the feature extraction unit 112 extracts, for example, feature 134 (hereinafter also referred to as new feature 134) from the new image data 131 accepted in the processing of S51 (S52).

[0117] Furthermore, the data estimation unit 115 acquires, for example, operation information 133 (hereinafter also referred to as new operation information 133) output from the learning model in response to the input of the new feature 134 extracted in the processing of S52 and the new operation information 132 received in the processing of S51 (S53).

[0118] Specifically, as shown in FIG. 16, the data estimation unit 115 acquires new operation information 133 output from the learning model MD (the learning model MD described in FIG. 13) generated by the learning process, for example.

[0119] Thereafter, the data output unit 116 outputs, for example, the new operation information 133 acquired in the process of S53 (S54).

[0120] For example, in the learning process, when a learning model MD is generated using teacher data DT including aggregated feature 134a and aggregated operation information 132a, the data estimation unit 115 may generate aggregated feature 134a (hereinafter also referred to as new aggregated feature 134a) from the multiple new feature 134 extracted in the process of S52, and generate aggregated operation information 132a (hereinafter also referred to as new aggregated operation information 132a) from the multiple new operation information 132 inputted in the process of S51. In this case, the data estimation unit 115 may acquire new operation information 133 by inputting the generated new aggregated feature 134a and new aggregated operation information 132a to the learning model MD.

[0121] In this way, the information processing device 1 in this embodiment extracts the feature amount 134 from each of the multiple learning image data 131 about the fire in the incinerator 10 captured at multiple consecutive timings in the learning process, for example. Then, the information processing device 1 calculates, for example, for each of the predetermined number of learning image data 131, the aggregated feature amount 134a from the feature amount 134 corresponding to each of the predetermined number of learning image data 131. Furthermore, the information processing device 1 calculates, for example, for each of the predetermined number of learning image data 131, aggregated operation information 132a from the operation information 132 indicating the operation status of the incinerator 10 at the timing corresponding to each of the predetermined number of learning image data 131. Furthermore, the information processing device 1 generates multiple teacher data DT by adding, for each of the predetermined number of learning image data 131, the aggregated feature amount 134a corresponding to the predetermined number of learning image data 131 and the aggregated operation information 132a corresponding to the predetermined number of learning image data 131. After that, the information processing device 1 generates a learning model MD by performing machine learning using, for example, a plurality of pieces of teacher data DT.

[0122] Moreover, in the estimation process, for example, the information processing device 1 in this embodiment extracts new feature amounts 134 from a plurality of verification image data 131 about the fire in the incinerator 10 captured at a plurality of consecutive timings. Then, for example, for each of the predetermined number of verification image data 131, the information processing device 1 calculates new aggregated feature amounts 134a from the new feature amounts 134 corresponding to each of the predetermined number of verification image data 131. Also, for example, for each of the predetermined number of verification image data 131, the information processing device 1 calculates new aggregated operation information 132a from new operation information 132 indicating the operation status of the incinerator 10 at the timing corresponding to each of the predetermined number of verification image data 131. After that, the information processing device 1 acquires, for example, for each of the predetermined number of verification image data 131, information output from the learning model MD in association with the input of the new aggregated feature amounts 134a and the new aggregated operation information 132a corresponding to the predetermined number of verification image data 131, as new operation information 133 indicating the operation performed on the incinerator 10 at the timing corresponding to the predetermined number of verification image data 131. Then, the information processing device 1 outputs, for example, each of the acquired new operation information 133.

[0123] This enables the information processing device 1 in this embodiment to generate, for example, multiple pieces of teacher data DT each containing different information (in other words, multiple pieces of teacher data DT each expressing characteristics of the state of the incinerator 10 in the time period corresponding to each group). Therefore, the information processing device 1 can prevent overlearning from occurring in the process of generating the learning model MD, and can generate a learning model MD with high judgment accuracy.

[0124] Furthermore, in the information processing device 1 in the present embodiment, when the multiple learning image data 131 about the fire in the incinerator 10 captured at multiple consecutive timings includes multiple learning image data 131 (hereinafter also referred to as multiple first learning image data 131) captured at multiple timings (hereinafter also referred to as multiple first timings) when no operation is performed on the incinerator 10, and multiple learning image data 131 (hereinafter also referred to as multiple second learning image data 131) captured at multiple timings (hereinafter also referred to as multiple second timings) that are later than the multiple first timings and when an operation is performed on the incinerator 10, the information processing device 1 specifies one or more learning image data 131 captured at a timing that is within a predetermined time until the first timing of the multiple second timings. Then, for example, the information processing device 1 extracts feature quantities 134 from the multiple learning image data 131 other than the one or more identified learning image data 131 and the multiple second learning image data 131. Furthermore, the information processing device 1 generates a plurality of training data DT by, for example, adding operation information 132 indicating the operation status of the incinerator 10 at the timing corresponding to each training image data 131 and operation information 133 indicating the operation performed on the incinerator 10 at the timing corresponding to each training image data 131 to the feature amount 134 extracted from each training image data 131 for each of the plurality of training image data 131 other than the one or more identified training image data 131 and the plurality of second training image data 131. Then, the information processing device 1 generates a learning model MD by, for example, performing machine learning using the plurality of training data DT.

[0125] As a result, the information processing device 1 in this embodiment can generate the learning model MD without using the teacher data DT that may be a factor in reducing the judgment accuracy of the learning model MD. Therefore, the information processing device 1 can improve the judgment accuracy of the learning model MD, for example. [Explanation of symbols]

[0126] 1: Information processing device 2: Operation terminal 10: Incinerator 100: Incineration system 101:CPU 102: Memory 103: Communication interface 104:Storage medium 105: Bus 111: Information Management Department 112: Feature extraction unit 113: Data generation unit 114: Model generation unit 115: Data Estimation Department 116: Data output section

Claims

1. Extracting feature amounts from a plurality of learning image data of a fire in an incinerator captured at a plurality of consecutive times; calculating an aggregate feature value from the feature values ​​corresponding to each of a predetermined number of pieces of learning image data among the plurality of learning image data; For each of the predetermined number of learning image data, aggregated operation information is calculated from operation information indicating the operation status of the incinerator at a timing corresponding to each of the predetermined number of learning image data; For each of the predetermined number of learning image data, operation information indicating an operation performed on the incinerator at a timing corresponding to the predetermined number of learning image data is added to the aggregate feature corresponding to the predetermined number of learning image data and the aggregate operation information corresponding to the predetermined number of learning image data, thereby generating a plurality of teacher data; generating a learning model by performing machine learning using the plurality of training data; A learning program that causes a computer to execute a process.

2. When a plurality of learning image data of a fire in an incinerator photographed at a plurality of consecutive timings includes a plurality of first learning image data photographed at a plurality of first timings at which no operation is performed on the incinerator, and a plurality of second learning image data photographed at a plurality of second timings that are later than the plurality of first timings and at which an operation is performed on the incinerator, one or more learning image data from the plurality of first learning image data photographed at a timing that is within a predetermined time until the first timing in the plurality of second timings is identified, Extracting features from each of the plurality of second learning image data and each of the plurality of learning image data other than the one or more learning image data among the plurality of first learning image data; For each of the plurality of first training image data other than the one or more training image data and the plurality of second training image data, a plurality of teacher data are generated by adding operation information indicating the operation status of the incinerator at a timing corresponding to each training image data and operation information indicating an operation performed on the incinerator at a timing corresponding to each training image data to the feature amount extracted from each training image data, generating a learning model by performing machine learning using the plurality of training data; A learning program that causes a computer to execute a process.

3. In claim 1 or 2, The feature amount corresponding to each of the plurality of training image data includes at least one of information corresponding to the size of the fire in each training image data, information indicating the position of the fire in each training image data, and information indicating the number of the fires reflected in each training image data. A learning program characterized by:

4. In claim 1 or 2, The operation information corresponding to each of the plurality of learning image data includes at least one of the amount of combustion air supplied to the incinerator and the temperature of the combustion air, A learning program characterized by:

5. In claim 1 or 2, The operation information corresponding to each of the plurality of learning image data includes information on an operation for at least one of the amount of combustion air supplied to the incinerator and the temperature of the combustion air, A learning program characterized by:

6. Extracting new features from a plurality of verification image data of a fire in an incinerator captured at a plurality of consecutive timings; calculating a new aggregate feature from the new feature corresponding to each of a predetermined number of pieces of verification image data among the plurality of pieces of verification image data; For each of the predetermined number of pieces of verification image data, new aggregated operation information is calculated from new operation information indicating the operation status of the incinerator at a timing corresponding to each of the predetermined number of pieces of verification image data; For each of the predetermined number of verification image data, information output from the learning model in association with the input of the new aggregate feature amount and the new aggregate operation information corresponding to the predetermined number of verification image data is acquired as new operation information indicating an operation performed on the incinerator at a timing corresponding to the predetermined number of verification image data; outputting each of the acquired new operation information; An estimation program that causes a computer to execute a process.

7. A feature extraction unit that extracts feature amounts from a plurality of learning image data of a fire in an incinerator captured at a plurality of consecutive timings; a data generating unit that calculates, for each of a predetermined number of learning image data among the plurality of learning image data, an aggregate feature from the feature corresponding to each of the predetermined number of learning image data, calculates, for each of the predetermined number of learning image data, aggregate operation information from operation information indicating the operation status of the incinerator at a timing corresponding to each of the predetermined number of learning image data, and generates, for each of the predetermined number of learning image data, a plurality of teacher data by adding, to the aggregate feature corresponding to the predetermined number of learning image data and the aggregate operation information corresponding to the predetermined number of learning image data, operation information indicating an operation performed on the incinerator at a timing corresponding to the predetermined number of learning image data; A model generation unit that generates a learning model by performing machine learning using the plurality of teacher data.

23. An information processing apparatus comprising:

8. a feature extraction unit that, when a plurality of pieces of training image data about a fire in an incinerator photographed at a plurality of consecutive timings includes a plurality of first training image data photographed at a plurality of first timings at which no operation is performed on the incinerator, and a plurality of second training image data photographed at a plurality of second timings that are later than the plurality of first timings and at which an operation is performed on the incinerator, identifies one or more pieces of training image data among the plurality of first training image data photographed at a timing that is within a predetermined time until the first timing among the plurality of second timings, and extracts features from training image data other than the one or more pieces of training image data among the plurality of first training image data and the plurality of second training image data; a data generating unit that generates a plurality of teacher data by adding operation information indicating the operation status of the incinerator at a timing corresponding to each of the training image data and operation information indicating an operation performed on the incinerator at a timing corresponding to each of the training image data to the feature amount extracted from each of the training image data other than the one or more training image data among the plurality of first training image data and each of the plurality of second training image data; A model generation unit that generates a learning model by performing machine learning using the plurality of teacher data.

23. An information processing apparatus comprising:

9. A feature extraction unit that extracts new feature values ​​from a plurality of verification image data of a fire in an incinerator captured at a plurality of consecutive timings; a data estimation unit that calculates new aggregate features from the new features corresponding to each of the predetermined number of verification image data for each of the plurality of verification image data, calculates new aggregate operation information from new operation information indicating the operation status of the incinerator at a timing corresponding to each of the predetermined number of verification image data for each of the predetermined number of verification image data, and acquires, for each of the predetermined number of verification image data, information output from a learning model in association with input of the new aggregate features and the new aggregate operation information corresponding to the predetermined number of verification image data as new operation information indicating the operation performed on the incinerator at a timing corresponding to the predetermined number of verification image data; a data output unit that outputs the acquired new operation information, 23. An information processing apparatus comprising:

10. Extracting feature amounts from a plurality of learning image data of a fire in an incinerator captured at a plurality of consecutive times; calculating an aggregate feature value from the feature values ​​corresponding to each of a predetermined number of pieces of learning image data among the plurality of learning image data; For each of the predetermined number of learning image data, aggregated operation information is calculated from operation information indicating the operation status of the incinerator at a timing corresponding to each of the predetermined number of learning image data; For each of the predetermined number of learning image data, operation information indicating an operation performed on the incinerator at a timing corresponding to the predetermined number of learning image data is added to the aggregate feature corresponding to the predetermined number of learning image data and the aggregate operation information corresponding to the predetermined number of learning image data, thereby generating a plurality of teacher data; generating a learning model by performing machine learning using the plurality of training data; A learning method characterized in that the processing is executed by a computer.

11. When a plurality of learning image data of a fire in an incinerator photographed at a plurality of consecutive timings includes a plurality of first learning image data photographed at a plurality of first timings at which no operation is performed on the incinerator, and a plurality of second learning image data photographed at a plurality of second timings that are later than the plurality of first timings and at which an operation is performed on the incinerator, one or more learning image data from the plurality of first learning image data photographed at a timing that is within a predetermined time until the first timing in the plurality of second timings is identified, Extracting features from each of the plurality of second learning image data and each of the plurality of learning image data other than the one or more learning image data among the plurality of first learning image data; For each of the plurality of first training image data other than the one or more training image data and the plurality of second training image data, a plurality of teacher data are generated by adding operation information indicating the operation status of the incinerator at a timing corresponding to each training image data and operation information indicating an operation performed on the incinerator at a timing corresponding to each training image data to the feature amount extracted from each training image data, generating a learning model by performing machine learning using the plurality of training data; A learning method characterized in that the processing is executed by a computer.

12. Extracting new features from a plurality of verification image data of a fire in an incinerator captured at a plurality of consecutive timings; calculating a new aggregate feature from the new feature corresponding to each of a predetermined number of pieces of verification image data among the plurality of pieces of verification image data; For each of the predetermined number of pieces of verification image data, new aggregated operation information is calculated from new operation information indicating the operation status of the incinerator at a timing corresponding to each of the predetermined number of pieces of verification image data; For each of the predetermined number of verification image data, information output from the learning model in association with the input of the new aggregate feature amount and the new aggregate operation information corresponding to the predetermined number of verification image data is acquired as new operation information indicating an operation performed on the incinerator at a timing corresponding to the predetermined number of verification image data; outputting each of the acquired new operation information; An estimation method characterized in that the processing is executed by a computer.

Images