Monitoring program and monitoring device

Abstract

To detect abnormalities early and make facilities energy-efficient.SOLUTION: A monitoring program for monitoring flares that occur when excess gas at a facility is incinerated causes a computer to function as a classification unit 132 that classifies the state of the flare by inputting image data of an area where a flare is generated into a classification model 104, an estimation unit 134 that estimates parameters of the excess gas by inputting the image data into an estimation model 106, and an output unit 136 that outputs the flare state classified by the classification unit 132 and the parameters of the excess gas estimated by the estimation unit 134.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present disclosure relates to a monitoring program and a monitoring device. [Background technology]

[0002] At refineries and other facilities, excess gas emitted during the refining process is incinerated in equipment called flare stacks. Flames (flares) and black smoke are emitted from the tip of the flare stack as a result of the combustion of excess gas. Early detection of flare abnormalities and the generation of black smoke is important for the operation of the flare stack, so technology that can accurately determine the flare state is required.

[0003] In this regard, Patent Document 1 discloses a monitoring device that uses a trained model to determine flare conditions such as flare abnormalities, misfires, and black smoke generation. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2020-160794 Summary of the Invention [Problem to be solved by the invention]

[0005] However, with the monitoring device described in Patent Document 1, even though the flare stack operator can grasp the state of the flare, he or she cannot grasp the parameters of the excess gas. This makes it impossible to accurately determine future course of action, and it is not possible to detect abnormalities early or to achieve energy conservation at the facility.

[0006] In view of the above-mentioned problems, an object of the present disclosure is to provide a monitoring program and a monitoring device that can detect abnormalities and the like early and achieve energy conservation in facilities. [Means for solving the problem]

[0007] In order to solve the above problems, a monitoring program according to one aspect of the present disclosure is a monitoring program that monitors flares that occur when excess gas at a facility is incinerated, and causes a computer to function as an acquisition unit that acquires image data of an area where the flare is generated, a classification unit that classifies the state of the flare by inputting the image data acquired by the acquisition unit into a classification model, an estimation unit that estimates parameters of the excess gas by inputting the image data acquired by the acquisition unit into an estimation model, and an output unit that outputs the state of the flare classified by the classification unit and the parameters of the excess gas estimated by the estimation unit.

[0008] Furthermore, a monitoring device according to another aspect of the present disclosure is a monitoring device that monitors flares that occur when excess gas at a facility is incinerated, and includes a computer; an acquisition unit that acquires image data of an area where the flare is generated; a classification unit that classifies the state of the flare by inputting the image data acquired by the acquisition unit into a classification model; an estimation unit that estimates parameters of the excess gas by inputting the image data acquired by the acquisition unit into an estimation model; and an output unit that outputs the state of the flare classified by the classification unit and the parameters of the excess gas estimated by the estimation unit.

[0009] The present disclosure may be realized as a semiconductor integrated circuit that implements part or all of a monitoring program, as an information processing device, or as a system including an information processing device. [Effects of the Invention]

[0010] The monitoring program and monitoring device according to the present disclosure enable early detection of abnormalities and energy conservation in facilities. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a diagram schematically illustrating an example of the overall configuration of a monitoring system according to a first embodiment of the present disclosure. [Figure 2]FIG. 2 is a block diagram illustrating an example of a hardware configuration of a monitoring device. [Figure 3] FIG. 2 is a block diagram showing an example of a functional configuration of a learning device. [Figure 4] FIG. 2 is a block diagram showing an example of a functional configuration of a monitoring device. [Figure 5] FIG. 4 is a diagram showing an example of a monitoring screen displayed on a display device of the monitoring device. [Figure 6] 10 is a flowchart illustrating an example of a monitoring process of a monitoring device. [Figure 7] 7 is a flowchart showing an example of a subroutine process in the analysis process of step SP18 shown in FIG. 6. [Figure 8] 10 is a flowchart illustrating an example of an analysis process according to a second embodiment of the present disclosure. [Figure 9] 10 is a flowchart illustrating an example of an analysis process according to a third embodiment of the present disclosure. [Figure 10] 10 is a flowchart illustrating an example of an analysis process according to a fourth embodiment of the present disclosure. [Figure 11] FIG. 13 is a block diagram illustrating an example of a functional configuration of a monitoring device according to a fifth embodiment of the present disclosure. [Figure 12] 13 is a flowchart illustrating an example of an analysis process according to a fifth embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0012] First, some aspects of the disclosure will be described.

[0013] A monitoring program according to a first aspect of the present disclosure is a monitoring program for monitoring flares that occur when excess gas in a facility is incinerated, and causes a computer to function as an acquisition unit that acquires image data of an area where the flare is generated, a classification unit that classifies the state of the flare by inputting the image data acquired by the acquisition unit into a classification model, an estimation unit that estimates parameters of the excess gas by inputting the image data acquired by the acquisition unit into an estimation model, and an output unit that outputs the state of the flare classified by the classification unit and the parameters of the excess gas estimated by the estimation unit.

[0014] In the monitoring program according to the second aspect of the present disclosure, the classification model classifies the flare state into the state of flare misfire, the state of the flare size, and / or the state of black smoke that may be generated from the flare, and the estimation model estimates the flow rate of the excess gas and / or the calories of the excess gas as parameters of the excess gas.

[0015] In the monitoring program according to the third aspect of the present disclosure, the acquisition unit acquires operating data of the facility, and the estimation unit estimates parameters of the excess gas by inputting the image data and the operating data into the estimation model.

[0016] In the monitoring program according to a fourth aspect of the present disclosure, the acquisition unit acquires operation data of the facility, and the classification unit classifies the state of the flare by inputting the image data and the operation data into the classification model.

[0017] A monitoring program according to a fifth aspect of the present disclosure further causes the computer to function as a determination unit that determines a course of action based on the state of the flare and parameters of the excess gas.

[0018] In the monitoring program according to a sixth aspect of the present disclosure, the classification model classifies the size of the flare as the flare state, the estimation model estimates the flow rate of the excess gas as a parameter of the excess gas, and the determination unit determines, based on the size of the flare and the flow rate of the excess gas, as the response policy, whether to inject steam to extinguish the flare or whether to increase or decrease the amount of steam injected.

[0019] In the monitoring program according to a seventh aspect of the present disclosure, the classification model classifies, as the state of the flare, the state of black smoke that may be generated from the flare, the estimation model estimates, as parameters of the excess gas, the calories of the excess gas and / or the flow rate of the excess gas, and the determination unit determines, as the response policy, whether to inject steam to extinguish the flare or whether to increase or decrease the amount of steam injected, based on the state of the black smoke that may be generated from the flare and the calories of the excess gas and / or the flow rate of the excess gas.

[0020] In the monitoring program according to an eighth aspect of the present disclosure, the classification model classifies a state of flare misfire as the state of the flare, the estimation model estimates the flow rate of the excess gas and / or the calories of the excess gas as parameters of the excess gas, and the determination unit determines, as the response policy, whether or not to re-learn the classification model or the estimation model, based on the state of flare misfire and the flow rate of the excess gas or the calories of the excess gas.

[0021] In the monitoring program according to a ninth aspect of the present disclosure, the classification model classifies the size of the flare as the state of the flare, the estimation model estimates the flow rate of the excess gas and / or the calories of the excess gas as parameters of the excess gas, and the determination unit determines, as the response policy, whether or not to re-learn the classification model or the estimation model, based on the size of the flare and the flow rate of the excess gas or the calories of the excess gas.

[0022] In the monitoring program according to a tenth aspect of the present disclosure, the computer is caused to function as a memory unit that stores the state of the flare previously classified by the classification unit, and a judgment unit that judges the state of the flare based on the state of the flare newly classified by the classification unit and the state of the flare previously classified by the classification unit.

[0023] In the monitoring program according to an eleventh aspect of the present disclosure, the memory unit stores the parameters of the excess gas previously estimated by the estimation unit, and the judgment unit judges the parameters of the excess gas based on the parameters of the excess gas newly estimated by the estimation unit and the parameters of the excess gas previously estimated by the estimation unit.

[0024] In the monitoring program according to the twelfth aspect of the present disclosure, the computer is made to function as a memory unit that stores image data previously acquired by the acquisition unit, a reception unit that receives a selection of at least one image data from the image data stored in the memory unit, and a learning unit that re-learns the classification model or the estimation model based on the image data received by the reception unit.

[0025] In a monitoring program according to a thirteenth aspect of the present disclosure, the classification unit performs classification before the estimation unit performs estimation, and the estimation unit determines whether to perform the estimation based on the state of the flare classified by the classification unit, and performs the estimation if the determination is affirmative, or does not perform the estimation if the determination is negative.

[0026] In a monitoring program according to a fourteenth aspect of the present disclosure, the classification unit classifies the flare misfire state into one of a plurality of states including a state in which the flare is misfiring or a state in which the flare is not misfiring, and the estimation unit estimates the flow rate of the excess gas and / or the calories of the excess gas when the flare state is classified as the non-misfiring state, and does not estimate the flow rate of the excess gas and / or the calories of the excess gas when the flare state is classified as the misfiring state.

[0027] In the monitoring program according to a fifteenth aspect of the present disclosure, the classification unit classifies the state of flare misfire into one of a plurality of states including a state in which the flare is misfiring or a state in which the flare is not misfiring, and then classifies the state of the size of the flare and / or the state of black smoke that may be generated from the flare.

[0028] In a monitoring program according to a sixteenth aspect of the present disclosure, the classification unit classifies the flare size state into one of a plurality of states including a normal state and an abnormal state, and the estimation unit estimates the flow rate of the excess gas and / or the calories of the excess gas when the flare size state is classified into the abnormal state, and does not estimate the flow rate of the excess gas and / or the calories of the excess gas when the flare size state is classified into the normal state.

[0029] In a monitoring program according to a seventeenth aspect of the present disclosure, when the classification unit is a first classification unit, the computer is made to function as a second classification unit that classifies the state of the flare by a method different from the classification method of the first classification unit, and a determination unit that determines the state of the flare based on the state of the flare classified by the first classification unit and the state of the flare classified by the second classification unit.

[0030] In the monitoring program according to the eighteenth aspect of the present disclosure, when the estimation unit is a first estimation unit, the computer is made to function as a second estimation unit that estimates the parameters of the excess gas using a method different from the estimation method of the first estimation unit, and the determination unit determines the parameters of the excess gas based on the parameters of the excess gas estimated by the first estimation unit and the parameters of the excess gas estimated by the second estimation unit.

[0031] A monitoring device according to a 19th aspect of the present disclosure is a monitoring device that monitors flares that occur when excess gas at a facility is incinerated, and includes a computer; an acquisition unit that acquires image data of an area where the flares are generated; a classification unit that classifies the state of the flare by inputting the image data acquired by the acquisition unit into a classification model; an estimation unit that estimates parameters of the excess gas by inputting the image data acquired by the acquisition unit into an estimation model; and an output unit that outputs the state of the flare classified by the classification unit and the parameters of the excess gas estimated by the estimation unit.

[0032] Hereinafter, first to fifth embodiments of the present disclosure will be described with reference to the accompanying drawings. To facilitate understanding of the description, the same components and steps in each drawing will be denoted by the same reference numerals as much as possible, and duplicate explanations will be omitted. Furthermore, when terms such as "first" and "second" are used in this specification or claims, unless otherwise specified, they do not represent any order or importance, but are used to distinguish one configuration from another.

[0033] First Embodiment First, the first embodiment will be described.

[0034] ---Overall structure--- FIG. 1 is a diagram schematically illustrating an example of the overall configuration of a monitoring system 2 according to a first embodiment of the present disclosure.

[0035] As shown in FIG. 1, the monitoring system 2 is a system for monitoring a flare F that occurs when excess gas G in a facility 4 is burned.

[0036] The facility 4 is not particularly limited as long as it is a facility that generates surplus gas G, and examples thereof include a crude oil mining facility, a gas processing facility, a refinery, a chemical facility, a thermal power plant, and a waste incineration facility. The facility 4 is equipped with a flare stack 6, which is equipment for incinerating surplus gas G generated within the facility 4. In order to suppress the generation of black smoke due to incomplete combustion of the surplus gas G, steam S, the flow rate of which is controlled by a control member 8 such as a flow control valve, is injected into the flare stack 6. Note that examples of the control member 8 include, in addition to a flow control valve for steam S, flow control valves for other gases, on / off switches, alarms, etc.

[0037] The monitoring system 2 includes a camera 10, a learning device 20, a monitoring device 30, and a control device 40.

[0038] Camera 10 captures images of the burner portion (the tip portion that dissipates flare F) at the tip of flare stack 6 and its surroundings. In other words, camera 10 captures the area where flare F is generated. Camera 10 directly or indirectly transmits image data 10A obtained by capturing the images to monitoring device 30. Camera 10 is preferably an imaging device that captures video, but may also be one that captures still images at predetermined time intervals. Camera 10 may also be an infrared camera. When steam S is injected into flare stack 6, flare F becomes smaller, but water vapor is released instead. An infrared camera can detect this water vapor and more accurately estimate the flow rate of excess gas G. An infrared camera can also detect hydrogen gas when it is released. Preferably, multiple cameras 10 are installed to capture images from different directions. This allows a three-dimensional view of flare F to be captured even when the flare F is blown by the wind.

[0039] The learning device 20 is an information processing device that generates a trained model 20A for analyzing the state of the flare F and the parameters of the excess gas G and transmits the trained model 20A to the monitoring device 30. The functions of the learning device 20 may be incorporated into the monitoring device 30. In this case, the learning device 20 can be omitted.

[0040] The monitoring device 30 is an information processing device that monitors the flare F. The monitoring device 30 acquires image data 10A from the camera 10, operation data 10B of the facility 4 from the facility 4, and a trained model 20A from the learning device 20. The monitoring device 30 analyzes the state of the flare F and parameters of the surplus gas G using the acquired image data 10A, operation data 10B, and trained model 20A. Examples of the operation data 10B include the flow rate of the surplus gas G, the composition of the surplus gas G, and the flow rate of steam S.

[0041] The control device 40 is an information processing device that outputs a control signal C for controlling the control member 8 based on the monitoring results of the monitoring device 30. The functions of the control device 40 may be incorporated into the monitoring device 30. In this case, the control device 40 can be omitted.

[0042] ---Hardware configuration--- FIG. 2 is a block diagram showing an example of the hardware configuration of the monitoring device 30. As shown in FIG.

[0043] As shown in FIG. 2, the monitoring device 30 includes a CPU (Central Processing Unit) 30A, a memory 30B, a storage device 30C, an input device 30D, a display device 30E, and a communication device 30F.

[0044] The CPU 30A executes a monitoring program stored in the memory 30B or the storage device 30C, etc., and thereby functions as various functional components described below to monitor the flare F that occurs when the facility's excess gas G is incinerated.

[0045] The memory 30B and the storage device 30C store various programs and information required for executing processes in the monitoring device 30, as well as information on the results of the processes. The memory 30B and the storage device 30C are non-transitory, tangible computer-readable storage media. The non-transitory, tangible computer-readable storage media are any storage media that can be accessed by a computer, a CPU 30A, an MPU (Micro Processing Unit), etc.

[0046] The input device 30D is a device that accepts input operations by the operator.

[0047] Display device 30E displays a monitoring screen of flare F based on image data 10A from camera 10. Note that the display device that displays the monitoring screen may be provided separately from display device 30E of monitoring device 30.

[0048] The communication device 30F is configured with a communication interface for communicating with the camera 10, the learning device 20, and the control device 40, and the like.

[0049] The monitoring device 30 can be realized using an information processing device such as a dedicated or general-purpose server computer. The monitoring device 30 may be configured from a single information processing device or multiple information processing devices distributed over a communication network. FIG. 2 shows only a portion of the main hardware configuration of the monitoring device 30, and the monitoring device 30 may also have other components that are generally included in a server. The hardware configuration of the learning device 20 and the control device 40 may also have a configuration similar to that of the monitoring device 30.

[0050] ――Functional configuration―― FIG. 3 is a block diagram showing an example of the functional configuration of the learning device 20. As shown in FIG.

[0051] 3, the learning device 20 includes a functional configuration including a storage unit 100. The storage unit 100 is realized by a memory and / or a storage device. The storage unit 100 stores training data 102, a classification model 104, and an estimation model 106.

[0052] The training data 102 includes data in which a plurality of image data 10A is associated with a correct label (normal, abnormal, etc.) and also includes data in which a plurality of image data 10A is associated with a region image of the flare F portion.

[0053] The classification model 104 is a type of trained model 20A for classifying the state of a flare F. When image data 10A is input to this classification model 104, it outputs a degree of certainty that the image data 10A belongs to a plurality of pre-prepared classes. This classification model 104 includes a misfire classification model 104A, a size classification model 104B, and a black smoke classification model 104C. Misfire classification model 104A is a model for classifying the misfire state of the Flare F. The size classification model 104B is a model for classifying the size state of the flare F. The black smoke classification model 104C is a model for classifying the state of black smoke that may be generated from a flare F. These classification models 104 may output confidence levels that belong to two classes, such as whether or not a misfire is occurring and whether or not the size is normal, or may output confidence levels that belong to three or more classes, including, for example, a class called "alert."

[0054] The estimation model 106 is a type of trained model 20A for estimating parameters of surplus gas G when image data 10A and operating data 10B are input. The estimation model 106 includes a flow rate estimation model 106A and a calorie estimation model 106B. The flow rate estimation model 106A is a model for estimating the flow rate of the surplus gas G as a parameter. The calorie estimation model 106B is a model for estimating the calorie of the surplus gas G as a parameter.

[0055] Furthermore, the learning device 20 includes, as functional components, a preprocessing unit 110 and a learning unit 112. These functional components are realized by the CPU executing a program stored in a storage device.

[0056] The preprocessing unit 110 performs normalization, standardization, noise removal, and augmentation of the training data 102 as preprocessing for machine learning. Augmentation adds transformations such as affine transformations to the training data 102, enabling accurate analysis even when, for example, the camera 10 is displaced, the color tone changes due to aging, or the monitored object changes. It is also possible to reuse the trained model 20A for a flare stack in a different location than the flare stack 6.

[0057] The learning unit 112 performs machine learning based on the teacher data 102 processed by the preprocessing unit 110, and generates the classification model 104 and the estimation model 106. The learning unit 112 may also re-learn the classification model 104 or the estimation model 106 based on new image data 10A. Specifically, the learning unit 112 may add the new image data 10A received by and from the monitoring device 30 to the teacher data 102, and re-generate the classification model 104 or the estimation model 106 based on the teacher data 102.

[0058] FIG. 4 is a block diagram showing an example of the functional configuration of the monitoring device 30. As shown in FIG.

[0059] 4, the monitoring device 30 includes, as a functional component, a storage unit 120. The storage unit 120 is realized by a memory 30B and / or a storage device 30C.

[0060] The storage unit 120 stores the classification model 104 and the estimation model 106 acquired by the monitoring device 30 from the learning device 20. The storage unit 120 may also store the state of a flare F previously classified by the classification unit 132. The storage unit 120 may also store parameters of excess gas G previously estimated by the estimation unit 134. The storage unit 120 may also store image data 10A previously acquired by the acquisition unit 130.

[0061] The monitoring device 30 has, as its functional components, an acquisition unit 130, a classification unit 132, an estimation unit 134, an output unit 136, a determination unit 138, and a reception unit 140. These functional components are realized by the CPU 30A executing a monitoring program stored in the storage device 30C.

[0062] Acquisition unit 130 acquires image data 10A of an image of an area where flare F occurs and operation data 10B of flare stack 6 in facility 4. Specifically, acquisition unit 130 acquires image data 10A from camera 10 or another device (not shown). Examples of the other device include a video acquisition device, a video distribution device, and an external storage device connected to camera 10. Acquisition unit 130 also acquires operation data 10B from a communication device (not shown) of facility 4.

[0063] The classification unit 132 classifies the state of the flare F by inputting the image data 10A acquired by the acquisition unit 130 into the classification model 104. Specifically, the classification unit 132 classifies the state of the flare F based on the confidence level of each class and a threshold value output by inputting the image data 10A into the classification model 104. For example, if the confidence level of "abnormal" for the input image data 10A is 0.90, the confidence level of "normal" is 0.10, and the threshold value is 0.5, the classification unit 132 classifies the image data 10A as "abnormal." The classification unit 132 inputs the image data 10A into the misfire classification model 104A to classify the state of the flare F as a misfire state or not. The classification unit 132 also inputs the image data 10A into the size classification model 104B to classify the state of the flare F as a normal state or an abnormal state. Furthermore, the classification unit 132 inputs the image data 10A to the black smoke classification model 104C to classify the state of the flare F as whether or not black smoke is being emitted from the flare F. The classification unit 132 may input not only the image data 10A but also the operational data 10B to the classification model 104. For example, because black smoke is likely to be generated when heavy oil is treated, the classification unit 132 can improve the accuracy of classifying the state of flare F as whether or not black smoke is being generated by inputting the composition of excess gas G as the operational data 10B to the black smoke classification model 104C.

[0064] The estimation unit 134 estimates parameters of the surplus gas G by inputting the image data 10A acquired by the acquisition unit 130 into the estimation model 106. Specifically, the estimation unit 134 estimates parameters of the surplus gas G based on image data output by inputting the image data 10A into the estimation model 106. For example, the estimation unit 134 estimates the flow rate of the surplus gas G by inputting the image data 10A acquired by the acquisition unit 130 into the flow rate estimation model 106A. Furthermore, the estimation unit 134 estimates the calories of the surplus gas G by inputting the image data 10A acquired by the acquisition unit 130 into the calorie estimation model 106B. The estimation unit 134 may input not only the image data 10A but also the operating data 10B to the estimation model 106. For example, the estimation unit 134 can improve the accuracy of estimating the flow rate of the surplus gas G by inputting the flow rate of the surplus gas G and the flow rate of steam S to the flow rate estimation model 106A as the operating data 10B. Furthermore, because calories are significantly affected by the composition of the surplus gas G, the estimation unit 134 can improve the accuracy of estimating the calories of the surplus gas G by inputting the composition of the surplus gas G to the calorie estimation model 106B as the operating data 10B.

[0065] The output unit 136 outputs the state of the flare F classified by the classification unit 132 and the parameters of the surplus gas G estimated by the estimation unit 134. Examples of output destinations of the output unit 136 include the display device 30E, memory 30B, storage device 30C, and communication device 30F of the monitoring device 30. In the first embodiment, the output destination of the output unit 136 is the display device 30E, and the output unit 136 controls the display of the state of the flare F classified by the classification unit 132 and the parameters of the surplus gas G estimated by the estimation unit 134 on the monitoring screen of the display device 30E.

[0066] FIG. 5 is a diagram showing an example of a monitoring screen 150 displayed on the display device 30E of the monitoring device 30. As shown in FIG.

[0067] As shown in FIG. 5 , the monitoring screen 150 includes an image display field 152, a first information display field 154, a second information display field 156, a third information display field 158, and a fourth information display field 160. The image display field 152 displays an image based on image data 10A acquired from the camera 10 in real time. The first information display field 154 displays the size of the flare F classified by the classification unit 132. The second information display field 156 displays the flow rate and calories of the excess gas G estimated by the estimation unit 134. The third information display field 158 displays, for example, “injection,” “caution,” or “unnecessary” regarding the injection of steam S. The fourth information display field 160 displays, for example, “recommended” or “unnecessary” regarding the relearning of the trained model 20A.

[0068] 4, the determination unit 138 determines one of a plurality of countermeasures predetermined for the operator or the control device 40 based on the state of the flare F classified by the classification unit 132 and the parameters of the excess gas G estimated by the estimation unit 134. Specific examples of the determination made by the determination unit 138 will be described in the following (1) to (4).

[0069] (1) The determination unit 138 determines, based on the size of the flare F and the flow rate of the excess gas G, a course of action as to whether or not to inject steam S to extinguish the flare. Specifically, as shown in Table 1, the determination unit 138 determines the countermeasure policy to be “steam injection” when the size of the flare F is abnormal and the flow rate of the excess gas G is equal to or greater than the first threshold. When this countermeasure policy is determined, the output unit 136 controls the third information display field 158 of the monitoring screen 150 to display “injection” as the countermeasure policy, and the control device 40 controls the control element 8 to inject steam S into the flare stack 6. Alternatively, after the operator makes a final decision on whether to inject steam S into the flare stack 6 based on the determination result of the countermeasure policy “injection” displayed in the third information display field 158, the control device 40 controls the control element 8 to inject steam S into the flare stack 6. In this case, the determination unit 138 may calculate the injection amount of steam S to be injected or the control amount of the control element 8 based on the size of the flare F and the flow rate of the excess gas G. Furthermore, the determination unit 138 may calculate the injection amount of steam S to be injected or the control amount of the control element 8 for each determination of the target policy, “steam injection.” In this way, the determination unit 138 may determine whether to increase or decrease the injection amount of steam S based on the size of the flare F and the flow rate of the excess gas G. Furthermore, when the size of the flare F is abnormal and the flow rate of the excess gas G is less than the first threshold, the determination unit 138 determines the response policy to be "alert." When this response policy is determined, the output unit 136 controls the display of "alert" in the third information display field 158 on the monitoring screen 150. Furthermore, when the size of the flare F is normal and the flow rate of the excess gas G is equal to or greater than the first threshold, the determination unit 138 determines the response policy to be "alert." When this response policy is determined, the output unit 136 controls the display of "alert" in the third information display field 158 on the monitoring screen 150. Furthermore, when the size of the flare F is normal and the flow rate of the excess gas G is less than the first threshold, the determination unit 138 determines that the response policy is "no alert required." When this response policy is determined, the output unit 136 controls the third information display field 158 on the monitoring screen 150 to display "no alert required."

[0070] [Table 1]

[0071] (2) The determination unit 138 determines a course of action, based on the state of the black smoke and the calorie of the excess gas G, as to whether or not to inject steam S to extinguish the flare F. Specifically, as shown in Table 2, when black smoke is being generated from the flare F and the calories of the excess gas G are equal to or greater than the second threshold, the determination unit 138 determines the appropriate response policy as “steam injection.” Once this response policy is determined, the output unit 136 controls the third information display field 158 on the monitoring screen 150 to display “injection,” and the control device 40 controls the control element 8 to inject steam S into the flare stack 6. Alternatively, after an operator makes a final decision on whether to inject steam S into the flare stack 6 based on the determination result of the response policy “injection” displayed in the third information display field 158, the control device 40 controls the control element 8 to inject steam S into the flare stack 6. In this case, the determination unit 138 may calculate the amount of steam S to be injected or the control amount of the control element 8 based on the state of black smoke and the calories of the excess gas G. Furthermore, the determination unit 138 may calculate the amount of steam S to be injected or the control amount of the control element 8 for each determination of the target policy as “steam injection.” In this way, the determination unit 138 may determine whether to increase or decrease the amount of steam S injected based on the state of the black smoke and the calorie of the excess gas G. Furthermore, when black smoke is being emitted from the flare F and the calories of the excess gas G are less than the second threshold, the determination unit 138 determines the response policy to be "alert." When this response policy is determined, the output unit 136 controls the display of "alert" in the third information display field 158 on the monitoring screen 150. Furthermore, when black smoke is not being generated from the flare F and the calories of the excess gas G are equal to or greater than the second threshold, the determination unit 138 determines the response policy to be "alert." When this response policy is determined, the output unit 136 controls the display of "alert" in the third information display field 158 on the monitoring screen 150. Furthermore, when black smoke is not being generated from the flare F and the calories of the excess gas G are less than the second threshold, the determination unit 138 determines that the response policy is "no vigilance necessary." In other words, the determination unit 138 determines that the state of the flare F in terms of black smoke is normal and that injection of steam S is unnecessary. When this response policy is determined, the output unit 136 controls the third information display field 158 on the monitoring screen 150 to display "no vigilance necessary." In the above description, the determination unit 138 determines a countermeasure based on the state of black smoke and the calories of the excess gas G. However, the present invention is not limited to this. The determination unit 138 may determine a countermeasure using the flow rate of the excess gas G instead of the calories of the excess gas G. That is, the determination unit 138 may determine a countermeasure similar to the countermeasures shown in Table 2 based on the state of black smoke and the flow rate of the excess gas G. In this case, the determination unit 138 may calculate the injection amount of steam S to be injected or the control amount of the control member 8 based on the state of black smoke and the flow rate of the excess gas G. Furthermore, the determination unit 138 may calculate the injection amount of steam S to be injected or the control amount of the control member 8 for each determination of the target policy of "steam injection." In this way, the determination unit 138 may determine whether to increase or decrease the injection amount of steam S based on the state of black smoke and the flow rate of the excess gas G. Furthermore, the determination unit 138 may use the second threshold as the threshold corresponding to the flow rate of the excess gas G, or may use any threshold other than the second threshold. The determination unit 138 may determine the target policy by combining a determination based on the state of black smoke and the calories of the excess gas G with a determination based on the state of black smoke and the flow rate of the excess gas G. That is, the determination unit 138 may determine the target policy based on the state of black smoke, the calories of the excess gas G, and the flow rate of the excess gas G. The same applies to the determination by the determination unit 138 of whether to increase or decrease the amount of steam S injected. The user may appropriately set how to determine the target policy or the amount of steam S injected based on the combined results of both determinations. For example, the user may set the target policy to be determined by giving priority to one determination result, or may set the target policy to be determined by calculating the average of both determination results for the amount of steam S injected or the control amount of the control member 8 and determining whether to increase or decrease the amount of steam S injected.

[0072] [Table 2]

[0073] (3) The determination unit 138 determines whether or not to recommend re-learning of the classification model 104 or the estimation model 106 based on the state of misfire of the flare F and the flow rate of the excess gas G or the calorie of the flare F. Specifically, as shown in Table 3, when flare F is in a misfire state (misfire occurs) and the flow rate of excess gas G is equal to or greater than the third threshold, determination unit 138 determines the corrective action to be "re-learning recommended." When this corrective action is determined, output unit 136 controls to display "recommended" in fourth information display field 160 on monitoring screen 150. Furthermore, when flare F is in a misfire state (misfire occurs) and the flow rate of excess gas G is less than the third threshold, determination unit 138 determines that the corrective action policy is "relearning not required." When this corrective action policy is determined, output unit 136 controls fourth information display field 160 on monitoring screen 150 to display "not required." Furthermore, when flare F is not in a misfire state (no misfire) and the flow rate of excess gas G is equal to or greater than the third threshold, determination unit 138 determines that the corrective action policy is "no relearning required." When this corrective action policy is determined, output unit 136 controls fourth information display field 160 on monitoring screen 150 to display "no relearning required." Furthermore, when flare F is not in a misfire state (no misfire) and the flow rate of excess gas G is less than the third threshold, determination unit 138 determines the corrective action policy to be "relearning not required." When this corrective action policy is determined, output unit 136 controls fourth information display field 160 on monitoring screen 150 to display "not required."

[0074] [Table 3]

[0075] Similarly, as shown in Table 4, when flare F is in a misfire state (misfire occurs) and the calories of excess gas G are equal to or greater than the fourth threshold, determination unit 138 determines the corrective action to be "re-learning recommended." When this corrective action is determined, output unit 136 controls the display of "recommended" in fourth information display field 160 on monitoring screen 150. Furthermore, when flare F is in a misfire state (misfire occurs) and the calories of excess gas G are less than the fourth threshold, determination unit 138 determines that the corrective action policy is "relearning not required." When this corrective action policy is determined, output unit 136 controls fourth information display field 160 on monitoring screen 150 to display "not required." Furthermore, when flare F is not in a misfire state (no misfire) and the calories of excess gas G are equal to or greater than the fourth threshold, determination unit 138 determines that the corrective action policy is "relearning not required." When this corrective action policy is determined, output unit 136 controls fourth information display field 160 on monitoring screen 150 to display "not required." Furthermore, when flare F is not in a misfire state (no misfire) and the calories of excess gas G are less than the fourth threshold, determination unit 138 determines that the corrective action policy is "no relearning required." When this corrective action policy is determined, output unit 136 controls fourth information display field 160 on monitoring screen 150 to display "no relearning required." [Table 4]

[0076] (4) The determination unit 138 determines whether or not to recommend re-learning of the classification model 104 or the estimation model 106 based on the size of the flare F and the flow rate of the excess gas G or the calorie of the flare F. Specifically, as shown in Table 5, when the size of the flare F is abnormal and the flow rate of the excess gas G is equal to or greater than the fifth threshold, the determination unit 138 determines that the corrective action is "relearning not required." When this corrective action is determined, the output unit 136 controls the fourth information display field 160 on the monitoring screen 150 to display "not required." Furthermore, when the size of the flare F is abnormal and the flow rate of the excess gas G is less than the fifth threshold, the determination unit 138 determines the corrective action policy to be "re-learning recommended." When this corrective action policy is determined, the output unit 136 controls the display of "recommended" in the fourth information display field 160 on the monitoring screen 150. Furthermore, when the size of the flare F is normal and the flow rate of the excess gas G is equal to or greater than the fifth threshold, the determination unit 138 determines that the corrective action policy is "recommended to re-learn." When this corrective action policy is determined, the output unit 136 controls the display of "recommended" in the fourth information display field 160 on the monitoring screen 150. Furthermore, when the size of the flare F is normal and the flow rate of the excess gas G is less than the fifth threshold, the determination unit 138 determines that the corrective action policy is "relearning not required." When this corrective action policy is determined, the output unit 136 controls the fourth information display field 160 on the monitoring screen 150 to display "not required."

[0077] [Table 5]

[0078] Similarly, as shown in Table 6, when the size of the flare F is abnormal and the calories of the excess gas G are equal to or greater than the sixth threshold, the determination unit 138 determines that the corrective action is "relearning not required." When this corrective action is determined, the output unit 136 controls the fourth information display field 160 on the monitoring screen 150 to display "not required." Furthermore, if the size of the flare F is abnormal and the calories of the surplus gas G are less than the sixth threshold, the determination unit 138 determines that the corrective action policy is "recommended to re-learn." When this corrective action policy is determined, the output unit 136 controls the fourth information display field 160 on the monitoring screen 150 to display "recommended." Furthermore, when the size of the flare F is normal and the calories of the surplus gas G are equal to or greater than the sixth threshold, the determination unit 138 determines that the corrective action policy is "recommended to re-learn." When this corrective action policy is determined, the output unit 136 controls the fourth information display field 160 on the monitoring screen 150 to display "recommended." Furthermore, when the size of the flare F is normal and the calories of the excess gas G are less than the sixth threshold, the determination unit 138 determines that the corrective action policy is "relearning not required." When this corrective action policy is determined, the output unit 136 controls the fourth information display field 160 on the monitoring screen 150 to display "not required." [Table 6]

[0079] In addition to determining the above-described countermeasure policies (1) to (4), the determination unit 138 may determine the state of the flare F based on the state of the flare F newly classified by the classification unit 132 and the state of the flare F previously classified. For example, if the number of times the flare F has been classified as an abnormal state during the past predetermined time is equal to or less than a predetermined value, the determination unit 138 determines that the state of the flare F is normal even if the newly classified state is classified as an abnormal state. Furthermore, if the number of times the flare F has been classified as an abnormal state during the past predetermined time is greater than a predetermined value, the determination unit 138 determines that the state of the flare F is abnormal even if the newly classified state is classified as a normal state. Then, the output unit 136 controls the display of the state of the flare F determined by the determination unit 138, rather than the state of the flare F newly classified by the classification unit 132, in the first information display field 154 of the monitoring screen 150. Furthermore, the determination unit 138 may determine that the state of flare F is abnormal if the rate of change in the confidence level for classifying as abnormal or normal over the past predetermined time period is equal to or greater than a predetermined value. Furthermore, the determination unit 138 may determine that the state of flare F is abnormal if the confidence level for classifying as abnormal or normal over the past predetermined time period monotonically increases. Furthermore, the determination unit 138 may determine that the state of flare F is abnormal if the confidence level for classifying as abnormal or normal over the past predetermined time period monotonically decreases.

[0080] Furthermore, the determination unit 138 determines the parameters of the excess gas G based on the state of the flare F newly classified by the estimation unit 134 and the parameters of the excess gas G previously estimated by the estimation unit 134. For example, if the difference between the average value of the parameters over a predetermined time period in the past and the parameters of the excess gas G newly estimated by the estimation unit 134 is equal to or greater than a threshold, the determination unit 138 determines the average value of the parameters over the predetermined time period in the past or the upper limit value over the predetermined time period as the parameters of the excess gas G. Then, the output unit 136 controls the display of the parameters of the excess gas G determined by the determination unit 138, rather than the parameters of the excess gas G newly estimated by the estimation unit 134, in the second information display field 156 of the monitoring screen 150. Furthermore, the determination unit 138 may determine that the state of the flare F is abnormal if the rate of change of the parameter estimated by the estimation unit 134 over the past predetermined time period is equal to or greater than a predetermined value. Furthermore, the determination unit 138 may determine that the state of the flare F is abnormal if the parameter estimated by the estimation unit 134 over the past predetermined time period monotonically increases. Furthermore, the determination unit 138 may determine that the state of the flare F is abnormal if the parameter estimated by the estimation unit 134 over the past predetermined time period monotonically decreases.

[0081] The reception unit 140 receives, from the operator operating the input device 30D, a selection of at least one image data 10A from the image data 10A stored in the memory unit 120 and the state of flare F. In response to this, the reception unit 140 transmits the received image data 10A and the state of flare F to the learning device 20. The learning device 20 adds the received image data 10A and the state of flare F to the training data 102, and re-trains the classification model 104 and / or the estimation model 106 based on the added training data 102.

[0082] ---Processing flow--- 6 is a flowchart showing an example of the monitoring process of the monitoring device 30. This monitoring process is started when an operator operates the input device 30D to press a monitoring start button (not shown), when a predetermined schedule is reached, at random timing, or the like.

[0083] (Step SP10) The acquisition unit 130 acquires image data 10A of an area where flare F is generated from the camera 10, and stores the acquired image data 10A in the storage unit 120. Then, the process proceeds to the process of step SP12.

[0084] (Step SP12) Acquisition unit 130 acquires operating data 10B of flare stack 6 from a communication device (not shown) installed in facility 4, and stores the acquired operating data 10B in storage unit 120. Then, the process proceeds to step SP14.

[0085] (Step SP14) The acquisition unit 130 performs pre-processing on the acquired image data 10A. This pre-processing includes noise removal, sharpening, etc. Then, the process proceeds to step SP16.

[0086] (Step SP16) The acquisition unit 130 performs pre-processing on the acquired driving data 10B. This pre-processing may include normalization, standardization, missing value removal, outlier removal, etc. Then, the process proceeds to step SP18.

[0087] (Step SP18) The classification unit 132 and the estimation unit 134 perform an analysis process to analyze the image data 10A.

[0088] FIG. 7 is a flowchart showing an example of a subroutine process in the analysis process of step SP18 shown in FIG.

[0089] (Step SP50) The classification unit 132 inputs the image data 10A to a misfire classification model 104A, a size classification model 104B, and a black smoke classification model 104C, which serve as the classification models 104. In the first embodiment, the classification unit 132 does not input the driving data 10B to the misfire classification model 104A, the size classification model 104B, and the black smoke classification model 104C. In addition, the estimation unit 134 inputs the image data 10A and the driving data 10B to a flow rate estimation model 106A and a calorie estimation model 106B, which serve as the estimation model 106. Then, the processing proceeds to steps SP52 and SP54. Steps SP52 and SP54 are processed in parallel.

[0090] (Step SP52) As shown in the following steps SP52A to SP52C, the classification unit 132 classifies the state of the flare F based on the confidence level of each class output from the classification model 104 and a threshold value.

[0091] (Step SP52A) The classification unit 132 classifies whether the flare F is in a misfire state based on the confidence level of each class output from the misfire classification model 104A and a threshold value.

[0092] (Step SP52B) The classification unit 132 classifies the size of the flare F as being in a normal state or an abnormal state based on the confidence level of each class output from the size classification model 104B and a threshold value.

[0093] (Step SP52C) The classification unit 132 classifies whether or not black smoke is being emitted from the flare F based on the confidence level of each class output from the black smoke classification model 104C and a threshold value.

[0094] (Step SP54) The estimation unit 134 estimates parameters based on the regional image of the flare F output from the estimation model 106, as shown in the following steps SP54A and SP54B.

[0095] (Step SP54B) The estimation unit 134 calculates the area of ​​the regional image of the flare F from the regional image output from the flow rate estimation model 106A. Then, the estimation unit 134 substitutes the calculated area into the relational expression between the area of ​​the regional image and the flow rate, and estimates the flow rate of the excess gas G.

[0096] (Step SP54C) The estimation unit 134 calculates the area of ​​the regional image of the flare F output from the calorie estimation model 106B. Next, the estimation unit 134 substitutes the calculated area into the relational expression between the area of ​​the regional image and the calories, and estimates the calories of the excess gas G.

[0097] Returning to FIG. 6, the process proceeds to step SP20.

[0098] (Step SP20) The determination unit 138 executes post-processing. Specifically, the determination unit 138 determines the state of the flare F based on the state of the flare F newly classified by the classification unit 132 and the state of the flare F previously classified by the classification unit 132. The determination unit 138 also determines the parameters of the excess gas G based on the parameters of the excess gas G newly estimated by the estimation unit 134 and the parameters of the excess gas G previously estimated by the estimation unit 134. Then, the processing proceeds to the processing of step SP22.

[0099] (Step SP22) The determination unit 138 determines one of a plurality of predetermined response policies based on the determined state of the flare F and the determined parameters of the surplus gas G. Then, the process proceeds to step SP24.

[0100] (Step SP24) The output unit 136 outputs the state of the flare F determined by the determination unit 138 and the parameters of the excess gas G determined by the determination unit 138. The output unit 136 also controls the display device 30E and the communication device 30F based on the response policy determined by the determination unit 138. For example, the output unit 136 controls the display of "alert" or the like on the monitoring screen 150 based on the response policy. The output unit 136 may output the state of the flare F classified by the classification unit 132 and the parameters of the excess gas G newly estimated by the estimation unit 134 together with or separately from the state of the flare F determined by the determination unit 138 and the parameters of the excess gas G determined by the determination unit 138. Then, the processing proceeds to the processing of step SP26.

[0101] (Step SP26) The reception unit 140 determines whether or not an instruction to execute the countermeasure policy determined by the determination unit 138 has been received from the operator. If the determination is affirmative, the process proceeds to step SP28, and if the determination is negative, the process proceeds to step SP30.

[0102] (Step SP28) Output unit 136 executes control in accordance with the response policy. For example, output unit 136 transmits an instruction to control steam S to control device 40. In response to this, control device 40 controls control member 8 to inject steam S into flare stack 6. Then, the process proceeds to step SP30.

[0103] (Step SP30) The reception unit 140 determines whether or not the termination condition for the analysis process is met. If the determination is affirmative, the process proceeds to step SP32, and if the determination is negative, the process returns to step SP10.

[0104] (Step SP32) The reception unit 140 determines whether or not the operator has issued an instruction to execute relearning. If the determination is affirmative, the process proceeds to step SP34, and if the determination is negative, the series of monitoring processes shown in FIG. 6 ends.

[0105] (Step SP34) The receiving unit 140 receives a selection of at least one image data 10A from the image data 10A stored in the memory unit 120. The output unit 136 transmits a re-learning instruction including the image data 10A selected by the receiving unit 140 to the learning device 20. In response to this, the learning device 20 adds the selected image data 10A to the teacher data 102 and re-trains the trained model 20A based on the teacher data 102. The receiving unit 140 may also accept a change to the correct label of image data in the teacher data 102. For example, if an operator wishes to tighten the criteria for abnormalities, the receiving unit 140 accepts an instruction to change the correct label of image data that is classified as normal image data and that the operator wishes to change to abnormal. The output unit 136 transmits the instruction to change the correct label accepted by the receiving unit 140 to the learning device 20. In response to this, the learning device 20 changes the correct label of the image data in the teacher data 102. Subsequently, the learning device 20 re-trains the trained model 20A based on the changed teacher data 102.

[0106] Then, the series of monitoring processes shown in FIG. 6 ends.

[0107] ---Action and effect--- As described above, the monitoring device 30 according to the first embodiment includes an acquisition unit 130 that acquires image data 10A of an image of an area where a flare F is generated, and a classification unit 132 that classifies the state of the flare F by inputting the image data 10A acquired by the acquisition unit 130 into a classification model 104. The monitoring device 30 further includes an estimation unit 134 that estimates parameters of excess gas G by inputting the image data 10A acquired by the acquisition unit 130 into an estimation model 106, and an output unit 136 that outputs the state of the flare F classified by the classification unit 132 and the parameters of the excess gas G estimated by the estimation unit 134.

[0108] According to this configuration, not only the state of the flare F but also the parameters of the excess gas G are output, so that the operator can combine both the state of the flare F and the parameters of the excess gas G to accurately determine a future course of action. For example, by combining both the state of the flare F and the parameters of the excess gas G, the operator can determine, as a course of action, whether or not to relearn the classification model 104 or the estimation model 106. Therefore, if the relearning is performed early, the classification accuracy of the state of the flare F can be improved, thereby enabling early detection of abnormalities, etc. Furthermore, by combining both the state of the flare F and the parameters of the excess gas G, the operator can determine, as a course of action, whether or not to inject steam S to extinguish the flare F, thereby enabling energy conservation at the facility.

[0109] Furthermore, in the monitoring device 30 according to the first embodiment, the classification model 104 classifies the state of the flare F into the state of misfire of the flare F, the state of the size of the flare F, and / or the state of black smoke that may be generated from the flare F. Furthermore, the estimation model 106 estimates the flow rate of the excess gas G and / or the calories of the excess gas G as parameters of the excess gas G.

[0110] According to this configuration, the operator can determine whether or not to re-learn the classification model 104 or the estimation model 106 by combining both the state of the flare F misfire and the flow rate or calorie of the excess gas G, thereby enabling early detection of abnormalities, etc. Furthermore, the operator can determine whether or not to re-learn the classification model 104 or the estimation model 106 based on the state of the size of the flare F and the flow rate or calorie of the excess gas G, thereby enabling early detection of abnormalities, etc. Furthermore, the operator can determine whether or not to inject steam S to quench the flare F by combining both the size of the flare F and the flow rate of the excess gas G, thereby achieving energy conservation in the facility. Also, the operator can determine whether or not to inject steam S to quench the flare F by combining both the state of whether black smoke is being emitted from the flare F and the calories of the excess gas G, thereby achieving energy conservation in the facility 4.

[0111] Furthermore, in the monitoring device 30 according to the first embodiment, the acquisition unit 130 acquires operating data 10B of the facility 4, and the estimation unit 134 estimates parameters of the excess gas G by inputting the image data 10A and the operating data 10B into the estimation model 106.

[0112] According to this configuration, the accuracy of estimating the parameters of the surplus gas G can be improved. For example, the estimation unit 134 can improve the accuracy of estimating the flow rate of the surplus gas G by inputting the flow rate of the surplus gas G and the flow rate of steam S into the flow rate estimation model 106A as the operating data 10B. Furthermore, since calories are significantly affected by the composition of the surplus gas G, the estimation unit 134 can improve the accuracy of estimating the calories of the surplus gas G by inputting the composition of the surplus gas G into the calorie estimation model 106B as the operating data 10B. Note that if the flow rate of excess gas G is acquired as the operating data 10B, there are cases where it is not necessary to estimate the flow rate of excess gas G; however, in reality, the flow meter for excess gas G has a limited range, and accuracy may decrease except at high flow rates. For this reason, estimating the flow rate of excess gas G allows for a more accurate understanding of the flow rate. Also, since the flare F appears smaller due to steam S, estimating the flow rate of excess gas G from only the image data 10A tends to result in an underestimation. For this reason, by considering the flow rate of steam S as the operating data 10B, the accuracy of estimating the actual flow rate of excess gas G can be improved.

[0113] Furthermore, in the monitoring device 30 according to the first embodiment, the acquisition unit 130 acquires operating data 10B of the facility 4, and the classification unit 132 classifies the state of the flare F by inputting the image data 10A and the operating data 10B into the classification model 104.

[0114] According to this configuration, the accuracy of classifying the state of the flare F can be improved. For example, black smoke is likely to be generated when heavy oil is processed, so the classification unit 132 can improve the accuracy of classifying whether or not black smoke is being generated from the flare F by inputting the composition of excess gas G as operation data 10B into the black smoke classification model 104C.

[0115] Furthermore, the monitoring device 30 according to the first embodiment includes a determination unit 138 that determines a course of action based on the state of the flare F and the parameters of the excess gas G.

[0116] According to this configuration, the determining unit 138 determines the course of action, and the operator can easily make a final decision on the course of action based on the determination result.

[0117] Furthermore, in the monitoring device 30 according to the first embodiment, the classification model 104 classifies the size of the flare F as the state of the flare F, the estimation model 106 estimates the flow rate of the excess gas G as a parameter of the excess gas G, and the judgment unit 138 judges, based on the size of the flare F and the flow rate of the excess gas G, as a response policy, whether to inject steam S to extinguish the flare F or whether to increase or decrease the amount of steam S introduced.

[0118] According to this configuration, the judgment unit 138 judges whether or not to inject steam S to extinguish the flare F or whether to increase or decrease the amount of steam S introduced, so that based on the judgment result, the operator can easily make a final decision on whether or not to inject steam S or whether to increase or decrease the amount of steam S introduced.

[0119] Furthermore, in the monitoring device 30 according to the first embodiment, the classification model 104 classifies the state of black smoke that may be generated from the flare F as the state of the flare F, the estimation model 106 estimates the calories of the excess gas G and / or the flow rate of the excess gas G as parameters of the excess gas G, and the judgment unit 138 judges, as a response policy, whether to inject steam S to extinguish the flare F or whether to increase or decrease the amount of steam S introduced, based on the state of whether black smoke is being generated in the flare F and the calories of the excess gas G and / or the flow rate of the excess gas G.

[0120] According to this configuration, the judgment unit 138 judges whether or not to inject steam S to extinguish the flare F, and based on the judgment result, the operator can easily make a final decision on whether or not to inject steam S to extinguish the flare F or whether to increase or decrease the amount of steam S introduced.

[0121] Furthermore, in the monitoring device 30 according to the first embodiment, the classification model 104 classifies the state of misfire of the flare F as the state of the flare F, the estimation model 106 estimates the flow rate of the excess gas G and / or the calories of the excess gas G as parameters of the excess gas G, and the judgment unit 138 judges, as a response policy, whether or not the classification model 104 or the estimation model 106 should be re-learned based on the state of misfire of the flare F and the flow rate of the excess gas G or the calories of the excess gas G.

[0122] According to this configuration, the judgment unit 138 judges whether or not the classification model 104 or the estimation model 106 should be re-learned, and based on the judgment result, the operator can make a final decision on whether or not to re-learn the classification model 104 or the estimation model 106.

[0123] Furthermore, in the monitoring device 30 according to the first embodiment, the classification model 104 classifies the size of the flare F as the state of the flare F, the estimation model 106 estimates the flow rate of the surplus gas G and / or the calories of the surplus gas G as parameters of the surplus gas G, and the judgment unit 138 judges, as a response policy, whether or not the classification model 104 or the estimation model 106 should be re-learned based on the size of the flare F and the flow rate of the surplus gas G or the calories of the surplus gas G.

[0124] According to this configuration, the judgment unit 138 judges whether or not the classification model 104 or the estimation model 106 should be re-learned, and based on the judgment result, the operator can make a final decision on whether or not to re-learn the classification model 104 or the estimation model 106.

[0125] In addition, the monitoring device 30 according to the first embodiment includes a memory unit 100 that stores the state of flare F previously classified by the classification unit 132, and a judgment unit 138 that judges the state of flare F based on the state of flare F newly classified by the classification unit 132 and the state of flare F previously classified by the classification unit 132.

[0126] According to this configuration, the state of the flare F can be determined with high accuracy by taking into consideration the state of the flare F classified in the past.

[0127] In addition, in the monitoring device 30 according to the first embodiment, the memory unit 100 stores the parameters of the excess gas G previously estimated by the estimation unit 134, and the judgment unit 138 judges the parameters of the excess gas G based on the parameters of the excess gas G newly estimated by the estimation unit 134 and the parameters of the excess gas G previously estimated by the estimation unit 134.

[0128] According to this configuration, by taking into consideration the parameters of the surplus gas G that have been estimated in the past, the parameters of the surplus gas G can be determined with high accuracy.

[0129] In addition, the monitoring device 30 according to the first embodiment includes a memory unit 100 that stores image data 10A previously acquired by the acquisition unit 130, a reception unit 140 that receives a selection of at least one image data 10A from the image data 10A stored in the memory unit 100, and a learning unit 112 that re-learns the classification model 104 or the estimation model 106 based on the image data 10A received by the reception unit 140.

[0130] According to this configuration, by re-learning the classification model 104, it is possible to accurately determine the state of the flare F. Furthermore, by re-learning the estimation model 106, it is possible to accurately determine the parameters of the excess gas G.

[0131] Second Embodiment Next, a second embodiment will be described.

[0132] The second embodiment differs from the first embodiment in that the classification unit 132 performs classification before estimation by the estimation unit 134. Note that the configuration and processing of the monitoring system 2 that are not described in the second embodiment are the same as those in the first embodiment.

[0133] FIG. 8 is a flowchart showing an example of the analysis process according to the second embodiment of the present disclosure.

[0134] (Step SP200) The classification unit 132 inputs the image data 10A to each classification model 104 and classifies the state of the flare F based on the certainty factor and threshold value of each class output from each classification model 104. Specifically, the classification unit 132 inputs the image data 10A to a misfire classification model 104A and classifies whether the flare F is in a misfire state based on the certainty factor and threshold value of each class output from the misfire classification model 104A. The classification unit 132 also inputs the image data 10A to a size classification model 104B and classifies whether the size of the flare F is in a normal state or an abnormal state based on the certainty factor and threshold value of each class output from the size classification model 104B. The classification unit 132 also inputs the image data 10A to a black smoke classification model 104C and classifies whether black smoke is being emitted from the flare F based on the certainty factor and threshold value of each class output from the black smoke classification model 104C.

[0135] (Step SP202) Estimation unit 134 determines whether to perform estimation by determining whether a predetermined condition is satisfied with respect to the state of flare F classified by classification unit 132. Specifically, estimation unit 134 determines whether the condition that the state of flare F is a state in which flare F is misfiring is satisfied. If estimation unit 134 judges this determination to be positive, it determines not to perform estimation, and the series of analysis processes shown in FIG. 8 ends. If estimation unit 134 judges this determination to be negative, it determines to perform estimation, and proceeds to the processing of step SP204.

[0136] (Step SP204) The estimation unit 134 inputs the image data 10A to the estimation model 106, and estimates parameters based on the regional image of the flare F output from the estimation model 106. Then, the series of analysis processes shown in FIG. 8 ends.

[0137] As described above, in the monitoring device according to the second embodiment, the classification unit 132 performs classification before the estimation by the estimation unit 134, and the estimation unit 134 determines whether or not to perform estimation based on the state of the flare F classified by the classification unit 132, and performs estimation if the determination is affirmative, and does not perform estimation if the determination is negative.

[0138] According to this configuration, the estimation by the estimation unit 134 is omitted depending on the state of the flare F classified by the classification unit 132, so that the analysis process can be performed at a high speed.

[0139] In addition, in the monitoring device of the second embodiment, the classification unit 132 classifies the misfire state of the flare F into one of multiple states including a state in which the flare F is misfiring or a state in which the flare F is not misfiring, and the estimation unit 134 estimates the flow rate of the excess gas G and / or the calories of the excess gas when the state of the flare F is classified as a state in which the flare F is not misfiring, and does not estimate the flow rate of the excess gas G and / or the calories of the excess gas G when the state of the flare F is classified as a state in which the flare F is misfiring.

[0140] According to this configuration, if the state of flare F classified by classification unit 132 is a misfire state, the estimation by estimation unit 134 is omitted, so the conditions for omission can be simplified.

[0141] Third Embodiment Next, a third embodiment will be described.

[0142] The third embodiment differs from the second embodiment in that the classification unit 132 performs other classifications after classifying the misfire state of the flare F. Note that the configuration and processing of the monitoring system 2 that are not described in the third embodiment are the same as those of the first embodiment.

[0143] FIG. 9 is a flowchart showing an example of the analysis process according to the third embodiment of the present disclosure.

[0144] (Step SP300) The classification unit 132 inputs the image data 10A to the misfire classification model 104A, and classifies whether the flare F is in a misfire state or not based on the confidence level of each class and the threshold value output from the misfire classification model 104A. Then, the process proceeds to step SP302.

[0145] (Step SP302) The determination unit 138 determines whether or not the condition that the flare F is in a misfire state is satisfied with respect to the state of the flare F. If the determination is affirmative, the series of analysis processes shown in Fig. 9 ends, and if the determination is negative, the process proceeds to steps SP304 and SP306.

[0146] (Step SP304) The classification unit 132 performs the remaining classifications other than the misfire state. That is, the classification unit 132 inputs the image data 10A to the size classification model 104B, and classifies whether the size of the flare F is normal or abnormal based on the certainty factor and threshold value of each class output from the size classification model 104B. The classification unit 132 also inputs the image data 10A to the black smoke classification model 104C, and classifies whether black smoke is being emitted from the flare F based on the certainty factor and threshold value of each class output from the black smoke classification model 104C. Then, the series of analysis processes shown in FIG. 9 ends.

[0147] (Step SP306) The estimation unit 134 inputs the image data 10A to the estimation model 106, and estimates the parameters (the flow rate and calories of the excess gas G) based on the regional image of the flare F output from the estimation model 106. Then, the series of analysis processes shown in FIG. 9 ends.

[0148] As described above, in the monitoring device of the third embodiment, the classification unit 132 classifies the misfire state of the flare F into one of multiple states including a state in which the flare F is misfiring or a state in which the flare F is not misfiring, and then classifies the size state of the flare F and / or the state of black smoke that may be generated from the flare F.

[0149] According to this configuration, classification unit 132 performs other classifications after classifying the misfire state of flare F, so that it is possible to eliminate unnecessary training data, for example, when a misfire has occurred in training data 102 for size classification model 104B or training data 102 for black smoke classification model 104C, and improve the accuracy of classifying the size state of flare F and / or the state of black smoke that may be generated from flare F.

[0150] <Fourth embodiment> Next, a fourth embodiment will be described.

[0151] The fourth embodiment differs from the third embodiment in that the estimation unit 134 performs estimation after the classification unit 132 classifies the state of the size of the flare F. Note that the configuration and processing of the monitoring system 2 that are not described in the fourth embodiment are the same as those in the first embodiment.

[0152] FIG. 10 is a flowchart showing an example of the analysis process according to the fourth embodiment of the present disclosure.

[0153] (Step SP400) The classification unit 132 inputs the image data 10A to the misfire classification model 104A, and classifies whether the flare F is in a misfire state based on the confidence level of each class and the threshold value output from the misfire classification model 104A. Then, the process proceeds to step SP402.

[0154] (Step SP402) Determination unit 138 determines whether the condition that the flare F is in a misfire state is met regarding the misfire state of the flare F. If the determination is affirmative, the series of analysis processes shown in FIG. 10 ends, and if the determination is negative, the process proceeds to step SP404.

[0155] (Step SP404) The classification unit 132 inputs the image data 10A to the size classification model 104B, and classifies the size of the flare F as either normal or abnormal based on the confidence level of each class and a threshold value output from the size classification model 104B. Then, the process proceeds to step SP406.

[0156] (Step SP406) The determination unit 138 determines whether or not the condition that the size of the flare F is normal is satisfied with respect to the size state of the flare F. If the determination is affirmative, the series of analysis processes shown in Fig. 10 ends, and if the determination is negative, the process proceeds to step SP408.

[0157] (Step SP408) The estimation unit 134 inputs the image data 10A into the flow rate estimation model 106A to estimate the flow rate of the surplus gas G. Then, the process proceeds to step SP410.

[0158] (Step SP410) The determination unit 138 determines whether or not it is necessary to inject steam S to extinguish the flare F, based on the size of the flare F and / or the flow rate of the excess gas G. If the determination is affirmative, the process proceeds to the process of SP412, and if the determination is negative, the series of analysis processes shown in FIG. 10 ends.

[0159] (Step SP412) The determination unit 138 instructs the control device 40 to open the flow rate adjustment valve more. In response to this, the control device 40 controls the flow rate adjustment valve to open more, thereby increasing the injection amount of steam S. Then, the processing proceeds to the processing of step SP414.

[0160] (Step SP414) The estimation unit 134 inputs the image data 10A into the calorie estimation model 106B to estimate the calories of the surplus gas G. Then, the series of analysis processes shown in FIG.

[0161] As described above, in the monitoring device of the fourth embodiment, the classification unit 132 classifies the size state of the flare F into one of a plurality of states including a normal state and an abnormal state, and the estimation unit 134 estimates the flow rate of the excess gas G and / or the calories of the excess gas G when the size state of the flare F is classified as an abnormal state, and does not estimate the flow rate of the excess gas G and / or the calories of the excess gas G when the size state of the flare F is classified as a normal state.

[0162] According to this configuration, when the state of the flare F is classified as a normal state, the estimation by the estimation unit 134 is omitted, thereby speeding up the analysis process.

[0163] Fifth Embodiment Finally, a fifth embodiment will be described.

[0164] The fifth embodiment differs from the first embodiment in that the state of the flare F is classified by a method other than the classification method described in the first embodiment, and the parameters of the excess gas G are estimated by a method other than the estimation method described in the first embodiment. Note that the configuration and processing of the monitoring system 2 not described in the fifth embodiment are the same as those of the first embodiment.

[0165] FIG. 11 is a block diagram illustrating an example of a functional configuration of a monitoring device according to a fifth embodiment of the present disclosure.

[0166] As shown in FIG. 11, the monitoring device 300 includes an acquisition unit 130, an output unit 136, a determination unit 138, as well as a first classification unit 132A, a second classification unit 132B, a first estimation unit 134A, and a second estimation unit 134B.

[0167] The first classification unit 132A inputs the image data 10A acquired by the acquisition unit 130 into the classification model 104, thereby classifying the state of the flare F.

[0168] The second classification unit 132B classifies the state of the flare F using a method different from the classification method of the first classification unit 132A. For example, the second classification unit 132B classifies a plurality of pixels included in the image data 10A acquired by the acquisition unit 130 into first pixels indicating the flare F and second pixels different from the first pixels, and classifies the state of the flare F based on the number of pixels classified into the first pixels. The second classification unit 132B may also classify the state of the flare F based on a first infrared radiation intensity of a resonant radiation wavelength specific to the flare F and a second infrared radiation intensity of a wavelength different from the first infrared radiation and not including the resonant radiation wavelength, both measured by a measuring device. The second classification unit 132B may also classify a state in which black smoke has been generated from the flare F into first pixels indicating black smoke generated from the flare F and second pixels different from the first pixels, and classify the state in which black smoke has been generated from the flare F based on the number of pixels classified into the first pixels.

[0169] The first estimation unit 134A estimates parameters of the excess gas G by inputting the image data 10A acquired by the acquisition unit 130 into the estimation model 106.

[0170] The second estimation unit 134B estimates the parameters of the excess gas G using a method different from the estimation method of the first estimation unit 134A. For example, the second estimation unit 134B classifies a plurality of pixels included in the image data 10A acquired by the acquisition unit 130 into first pixels that indicate a flare F and second pixels that are different from the first pixels, and estimates the parameters of the excess gas G based on the number of pixels classified into the first pixels.

[0171] FIG. 12 is a flowchart showing an example of the analysis process according to the fifth embodiment of the present disclosure.

[0172] (Step SP500) The determination unit 138 determines whether a predetermined condition is satisfied. Specifically, the determination unit 138 determines whether the condition that a predetermined period of time has elapsed since the generation of the trained model 20A is satisfied. If the determination is affirmative, the process proceeds to step SP502, and if the determination is negative, the process proceeds to steps SP504 and SP506. In addition to the fact that a predetermined period of time has passed since the trained model 20A was generated, the specified conditions include a condition that an operator has performed an operation, a condition that a predetermined period of time has passed since the monitoring device 30 monitored the facility 4, a condition that the trained model 20A has been re-trained a predetermined number of times, and a condition that the flare stack 6 is equipment that is prone to generating black smoke or misfires.

[0173] (Step SP502) The monitoring device 300 performs analysis using the trained model 20A. Specifically, the first classification unit 132A inputs the image data 10A acquired by the acquisition unit 130 into the classification model 104, thereby classifying the state of the flare F. Furthermore, the first estimation unit 134A inputs the image data 10A acquired by the acquisition unit 130 into the estimation model 106, thereby estimating the parameters of the excess gas G. Then, the series of analysis processes shown in FIG. 12 ends.

[0174] (Step SP504) The monitoring device 300 executes analysis using the trained model 20A. Specifically, the first classification unit 132A inputs the image data 10A acquired by the acquisition unit 130 into the classification model 104 to classify the state of the flare F. Furthermore, the first estimation unit 134A inputs the image data 10A acquired by the acquisition unit 130 into the estimation model 106 to estimate parameters of the excess gas G. Then, the process proceeds to step SP508.

[0175] (Step SP506) The monitoring device 300 performs analysis using a method other than that using the trained model 20A. Specifically, the second classification unit 132B classifies the state of the flare F using a method different from that used by the first classification unit 132A. The second estimation unit 134B estimates the parameters of the excess gas G using a method different from that used by the first estimation unit 134A. Then, the process proceeds to step SP508.

[0176] (Step SP508) The determination unit 138 determines the state of the flare F based on the state of the flare F classified by the first classification unit 132A and the state of the flare F classified by the second classification unit 132B. The determination unit 138 also determines the parameters of the excess gas G based on the parameters of the excess gas G estimated by the first estimation unit 134A and the parameters of the excess gas G estimated by the second estimation unit 134B. In other words, the determination unit 138 determines the state of the flare F and the parameters of the excess gas G based on the analysis results by the trained model 20A and the analysis results by other methods that do not use the trained model 20A. For example, as shown in Table 7, the determination unit 138 determines the state of the flare F as normal if the analysis results by the trained model 20A are normal and the analysis results by other methods are normal, and determines the state to be on alert if the analysis results by the trained model 20A are normal and the analysis results by other methods are abnormal. In addition, the judgment unit 138 judges that there is an alert if the analysis result by the trained model 20A is abnormal and the analysis result by other methods is normal, and judges that there is an abnormality if the analysis result by the trained model 20A is abnormal and the analysis result by other methods is abnormal.

[0177] [Table 7]

[0178] As described above, the monitoring device 300 according to the fifth embodiment includes a first classifying unit 132A that classifies the state of the flare F by inputting image data 10A acquired by the acquisition unit 130 into the classification model 104. The monitoring device 300 also includes a second classifying unit 132B that classifies the state of the flare F using a method different from the classification method of the first classifying unit 132A. The monitoring device 300 also includes a determination unit 138 that determines the state of the flare F based on the state of the flare F classified by the first classifying unit 132A and the state of the flare F classified by the second classifying unit 132B.

[0179] According to this configuration, the accuracy of determining the state of the flare F can be improved.

[0180] In addition, the monitoring device 300 according to the fifth embodiment includes a second estimation unit 134B that estimates the parameters of the excess gas using a method different from the estimation method of the first estimation unit 134A, and a determination unit 138 that determines the parameters of the excess gas G based on the parameters of the excess gas G estimated by the first estimation unit 134A and the parameters of the excess gas G estimated by the second estimation unit 134B.

[0181] According to this configuration, the accuracy of determining the parameters of the excess gas G can be improved.

[0182] <Modification> The present disclosure is not limited to the above-described embodiments. In other words, variations of the above-described embodiments, which are appropriately modified by a person skilled in the art, are also included within the scope of the present disclosure as long as they include the features of the present disclosure. Furthermore, the elements of the above-described embodiments and the modifications described below can be combined to the extent technically possible, and such combinations are also included within the scope of the present disclosure as long as they include the features of the present disclosure.

[0183] For example, in the first embodiment, the classification model 104 classifies the size state of the flare F, but the color state, shape state, etc. of the flare F may also be classified.

[0184] Furthermore, in the second embodiment, a case has been described in which the estimation by estimation unit 134 is omitted when the state of flare F is classified as a misfire state. However, when reception unit 140 receives an operation for speeding up processing, etc., the estimation by estimation unit 134 may be omitted even when the state of flare F is classified as a non-misfire state.

[0185] Furthermore, in the second embodiment, the classification unit 132 performs classification before the estimation unit 134 performs estimation. However, the classification unit 132 may perform classification after the estimation unit 134 performs estimation.

[0186] In the first embodiment, one each of the misfire classification model 104A, the size classification model 104B, the black smoke classification model 104C, the flow rate estimation model 106A, and the calorie estimation model 106B is provided. However, a plurality of these models may be provided. In this case, the classification unit 132 may select at least one model from the plurality of misfire classification models 104A based on the operating data 10B. The classification unit 132 may also select at least one model from the plurality of size classification models 104B based on the operating data 10B. The classification unit 132 may also select at least one model from the plurality of black smoke classification models 104C based on the operating data 10B. Similarly, the estimation unit 134 may select one model based on the operating data 10B. This improves the accuracy of the analysis. For example, the classification unit 132 may select at least one size classification model 104B from the plurality of size classification models 104B based on the flow rate of steam S as the operating data 10B. Since the flare F appears smaller due to the steam S, it is better to prepare a size classification model 104B for each flow rate of the steam S in order to improve classification accuracy.

[0187] In addition, in the first embodiment, the judgment unit 138 has been described as determining whether or not to inject steam S based on the size of the flare F and the flow rate of excess gas G, but it may also determine whether or not to inject steam S based only on the size of the flare F or only on the flow rate of excess gas G.

[0188] Similarly, in the first embodiment, the judgment unit 138 determines whether to inject steam S based on whether black smoke is being generated in the flare F and the calories of the excess gas, but the judgment unit 138 may also determine whether to inject steam S based only on whether black smoke is being generated in the flare F or only on the calories of the excess gas G.

[0189] Similarly, in the first embodiment, the judgment unit 138 determines whether or not to re-learn the classification model 104 or the estimation model 106 based on the misfire state of the flare F and the flow rate of excess gas G or the calories of excess gas G. However, the judgment unit 138 may determine whether or not to re-learn the classification model 104 or the estimation model 106 based only on the misfire state of the flare F, only on the flow rate of excess gas G, or only on the calories of excess gas G.

[0190] Similarly, in the first embodiment, the judgment unit 138 has been described as judging whether or not the classification model 104 or the estimation model 106 should be re-learned based on the size of the flare F and the flow rate of excess gas G or the calories of excess gas G, but it may also be determined whether or not the classification model 104 or the estimation model 106 should be re-learned based only on the size of the flare F, only on the flow rate of excess gas G, or only on the calories of excess gas G. [Explanation of symbols]

[0191] 4: Facilities 10A: Image data 10B: Operation data 30: Monitoring device 100: Storage section 104: Estimation model 104: Classification model 106: Estimation model 112: Learning Department 120: Storage section 130: Acquisition section 132: Classification section 132:First classification section 132A: First classification section 132B:Second classification section 134:Estimation Department 134A:First Estimation Department 134B:Second Estimation Department 136: Output section 138: Judgment section 140: Reception 300: Monitoring device F: Flare G: Excess gas S: Steam

Claims

1. A monitoring program that monitors flares that occur when excess gas at a facility is burned, Computer, an acquisition unit that acquires image data of an area where the flare is generated; a classification unit that classifies the state of the flare by inputting the image data acquired by the acquisition unit into a classification model; an estimation unit that estimates parameters of the excess gas by inputting the image data acquired by the acquisition unit into an estimation model; an output unit that outputs the flare state classified by the classifying unit and the parameters of the excess gas estimated by the estimating unit; A monitoring program that acts as a

2. the classification model classifies, as the flare state, a state of misfire of the flare, a state of size of the flare, and / or a state of black smoke that may be generated from the flare; The estimation model estimates a flow rate of the excess gas or / and a calorie of the excess gas as a parameter of the excess gas. The monitoring program according to claim 1 .

3. the acquisition unit acquires operation data of the facility, the estimation unit estimates parameters of the excess gas by inputting the image data and the operating data into the estimation model; 3. The monitoring program according to claim 1.

4. the acquisition unit acquires operation data of the facility, the classification unit classifies the flare state by inputting the image data and the operating data into the classification model.

3. The monitoring program according to claim 1.

5. The computer a determination unit that determines a course of action based on the flare state and the parameters of the excess gas; 2. The monitoring program according to claim 1, further functioning as:

6. the classification model classifies the state of the flare according to the magnitude of the flare, the estimation model estimates a flow rate of the excess gas as a parameter of the excess gas; the determination unit determines, as the countermeasure, whether to inject steam to extinguish the flare or whether to increase or decrease the amount of steam injected, based on the size of the flare and the flow rate of the excess gas. The monitoring program according to claim 5.

7. the classification model classifies, as the flare state, a state of black smoke that may be generated from the flare; the estimation model estimates a calorie of the excess gas or / and a flow rate of the excess gas as a parameter of the excess gas; the determination unit determines, as the countermeasure, whether to inject steam to extinguish the flare or whether to increase or decrease the amount of steam injection, based on the state of the black smoke and the calories of the excess gas and / or the flow rate of the excess gas. The monitoring program according to claim 5.

8. the classification model classifies a flare misfire state as the flare state; the estimation model estimates a flow rate of the excess gas and / or a calorie of the excess gas as a parameter of the excess gas; the determination unit determines, as the countermeasure, whether or not the classification model or the estimation model should be re-learned, based on a state of misfire of the flare and the flow rate of the excess gas or the calorie of the excess gas. The monitoring program according to claim 5.

9. the classification model classifies the state of the flare according to the magnitude of the flare, the estimation model estimates a flow rate of the excess gas and / or a calorie of the excess gas as a parameter of the excess gas; the determination unit determines, as the response policy, whether or not the classification model or the estimation model should be re-learned, based on the size of the flare and the flow rate of the excess gas or the calorie of the excess gas. The monitoring program according to claim 5.

10. The computer a storage unit that stores the flare states previously classified by the classification unit; a determination unit that determines the state of the flare based on the state of the flare newly classified by the classification unit and the state of the flare previously classified by the classification unit; 3. The monitoring program according to claim 1, wherein the monitoring program functions as:

11. the storage unit stores parameters of the excess gas previously estimated by the estimation unit, the determination unit determines the parameters of the excess gas based on the parameters of the excess gas newly estimated by the estimation unit and the parameters of the excess gas previously estimated by the estimation unit. The monitoring program according to claim 10.

12. The computer a storage unit that stores image data previously acquired by the acquisition unit; a receiving unit that receives a selection of at least one image data item from the image data items stored in the storage unit; a learning unit that re-learns the classification model or the estimation model based on the image data accepted by the accepting unit; 3. The monitoring program according to claim 1, wherein the monitoring program functions as:

13. The classification unit performs classification before the estimation by the estimation unit, the estimation unit determines whether or not to perform the estimation based on the state of the flare classified by the classification unit, and performs the estimation when a positive determination is made, and does not perform the estimation when a negative determination is made. The monitoring program according to claim 2 .

14. the classifying unit classifies the state of the flare misfire into one of a plurality of states including a state in which the flare misfires and a state in which the flare does not misfire; the estimation unit estimates the flow rate of the excess gas and / or the calories of the excess gas when the flare state is classified as the non-misfire state, and does not estimate the flow rate of the excess gas and / or the calories of the excess gas when the flare state is classified as the misfire state. The monitoring program according to claim 13.

15. the classifying unit classifies the state of flare misfire into one of a plurality of states including a state in which the flare misfires or a state in which the flare does not misfire, and then classifies the state of the size of the flare and / or the state of black smoke that may be generated from the flare. The monitoring program according to claim 13.

16. the classifying unit classifies the state of the magnitude of the flare into one of a plurality of states including a normal state and an abnormal state, the estimation unit estimates the flow rate of the excess gas and / or the calories of the excess gas when the size of the flare is classified as the abnormal state, and does not estimate the flow rate of the excess gas and / or the calories of the excess gas when the size of the flare is classified as the normal state.

16. The monitoring program according to any one of claims 13 to 15.

17. When the classification unit is a first classification unit, The computer a second classifying unit that classifies the state of the flare by a method different from the classification method of the first classifying unit; a determination unit that determines the state of the flare based on the state of the flare classified by the first classification unit and the state of the flare classified by the second classification unit; 3. The monitoring program according to claim 1, wherein the monitoring program functions as:

18. When the estimation unit is a first estimation unit, The computer a second estimation unit that estimates the parameters of the excess gas by a method different from the estimation method of the first estimation unit; a determination unit that determines the parameters of the excess gas based on the parameters of the excess gas estimated by the first estimation unit and the parameters of the excess gas estimated by the second estimation unit; 3. The monitoring program according to claim 1, wherein the monitoring program functions as:

19. A monitoring device that monitors flares that occur when excess gas in a facility is burned, Computer, an acquisition unit that acquires image data of an area where the flare is generated; a classification unit that classifies the state of the flare by inputting the image data acquired by the acquisition unit into a classification model; an estimation unit that estimates parameters of the excess gas by inputting the image data acquired by the acquisition unit into an estimation model; an output unit that outputs the flare state classified by the classification unit and the parameters of the excess gas estimated by the estimation unit; A monitoring device comprising:

Images