TW201923286A - In-furnace state quantity estimation device, estimation model creation device, and program and method for same - Google Patents

Abstract

An in-furnace state quantity estimation device, provided with: a feature quantity extraction unit for extracting feature quantities from a captured image of the interior of a furnace; a feature quantity selection unit for selecting one or more feature quantities from the extracted feature quantities; a learning data generation unit for generating learning data by associating the state quantity of the furnace corresponding to the combustion state indicated by the image and the selected feature quantity; an estimation model creation unit for creating an estimation model for estimating the combustion state from an image of the interior of the furnace using the learning data; and an estimation state quantity calculation unit for estimating, upon acquiring a captured image of the interior of the furnace, a state quantity corresponding to the combustion state indicated by the acquired image using the estimation model. The feature quantity is selected on the basis of the degree of contribution with respect to the state quantity to be estimated.

Description

Furnace state quantity estimation device, estimation model creation device, program and method thereof

The present disclosure relates to a monitoring technique of a combustion state based on image information of a furnace in which a combustion furnace is captured.

For example, Patent Documents 1 and 2 disclose a method for determining a combustion state based on image information in a combustion furnace. Specifically, Patent Document 1 discloses a method for detecting an abnormal combustion state based on a surface temperature distribution of a burner flame obtained from a combustion image (video information). In addition, Patent Document 2 discloses that in a combustion furnace, image processing is performed on an image captured by a television camera (camera device) provided directly above a primary combustion range of fuel for increasing flame combustion, and the obtained image is processed based on the image processing. The amount of change in the brightness of the image is predicted by a fuzzy inference method. In addition, there is a method of visually detecting the shape of a flame from an image captured by a TV camera to monitor a burning state (see Patent Document 1).

[Prior technical literature]
[Patent Literature]

[Patent Document 1] Japanese Patent Laid-Open No. 4-143515
[Patent Document 2] Japanese Patent Laid-Open No. 2001-4116

[Questions to be Solved by the Invention]

Patent Document 1 is to obtain the surface temperature distribution of flame, and it is necessary to indicate the flame in the image information, so it is difficult to apply it to the image information because of the influence of exhaust gas during combustion or the installation position of the camera in the furnace Show fire. On the other hand, in Patent Document 2, it is considered that the prediction is based on the amount of change in the brightness of the image obtained by image processing, so there is a possibility that the flame does not appear even if the flame does not appear in the image information. The type of combustion furnace It is easier to apply because the restrictions such as the placement position of the camera are small. However, for example, it is necessary to collect and analyze judgment methods based on the experience of a skilled operator for regularization of control by fuzzy inference. Originally, the combustion state is changed due to various types of combustion objects (fuel) (garbage, coal, etc.), combustion ambient gases (air volume, furnace temperature, etc.), and the shape of the burner. It is difficult to make an appropriate determination of the combustion state without the experience of a skilled person, and it is impossible to deny that there is a possibility that a determination standard has not been recognized even by a skilled person.

In addition, for example, a method for determining the state of combustion by measuring state quantities such as unburned part, NOx concentration, and CO concentration in the ash during the operation of the combustion furnace is considered. However, it takes time to measure the state quantity. In the case of unburned parts such as ash, it is difficult to immediately use the determination result of the combustion state based on the measurement result in the operation control.

In view of the foregoing, at least one embodiment of the present invention is to provide an in-furnace state quantity estimation device. The in-furnace state quantity estimation device estimates the inside of a furnace quickly and accurately based on image information in the furnace. The state quantity of the burning state.

[Means for solving problems]

(1) The in-furnace state quantity estimation device according to at least one embodiment of the present invention includes: a feature quantity extraction unit that extracts a feature quantity from an image taken in the furnace;
A feature amount selecting unit that selects one or more of the feature amounts from the extracted feature amounts;
The learning material generating unit is configured to correlate the state quantity of the furnace corresponding to the combustion state shown in the foregoing image and the previously selected feature quantity, and thereby generate learning materials;
The estimation model creation unit creates an estimation model for estimating a combustion state from an image in the furnace using the learning materials, and the estimation state quantity calculation unit acquires and captures the image in the furnace, and estimates using the estimation model. The aforementioned state quantity corresponding to the combustion state shown in the acquired aforementioned image;
among them,
The feature amount is selected based on the degree of contribution to the state amount to be estimated.

According to the configuration of the above (1), in accordance with the combustion state in the furnace, for example, the state quantity such as the unburned part in the ash, the NOx concentration, and the CO concentration changes. The state quantity (measured value or estimated value) at the time of the image information is an estimated model made by mechanical learning corresponding to the related learning data. From the image information of the furnace when the combustion was taken, it is estimated to be generated in the combustion state. State amount (estimated state amount). Thereby, the state quantity can be estimated quickly and accurately from the image information in the furnace.

(2) In some embodiments, in the configuration of the above (1),
The estimation model creation unit performs mechanical learning using the learning materials, and thereafter, creates the estimation model.
According to the configuration of (2) above, learning information is created by correlating the image information during the combustion in the furnace and the state quantity when the image information was captured, and mechanical learning is performed from the created learning materials. , You can create a presumption model.

(3) In some embodiments, in the configurations (1) to (2),
The aforementioned contribution degree is an index indicating a magnitude relative to the aforementioned state quantity.
According to the configuration of the above (3), the contribution degree is an index indicating the magnitude of the relative relationship with the state quantity, and the feature quantity is selected according to the contribution degree, so that the differences in the image can be made prominent.

(4) In some embodiments, in the configurations (1) to (3),
The feature quantity selecting unit calculates a regression formula for calculating the state quantity from each of the feature quantities.
The contribution rate of each of the feature amounts in the calculated regression equation is defined as the contribution degree.
According to the configuration of the above (4), the contribution degree can be set based on the contribution rate of each feature amount in the regression formula.

(5) In some embodiments, in the configurations (1) to (4),
The feature amount extraction unit extracts the feature amount based on the brightness information obtained from the past image.
According to the configuration of (5) above, the brightness information is not obtained by detecting the flame itself, but is obtained from the image. For example, even if the flame cannot be captured due to the exhaust gas, such as when the camera is installed in the upper part of the furnace, You can also get image information. Thereby, the installation of the imaging device (camera camera) for imaging the inside of the furnace can be facilitated.

(6) In some embodiments, in the configurations (1) to (5),
The aforementioned state quantity is a state quantity related to exhaust gas or effluent generated during combustion in the aforementioned furnace;
The learning material generating unit generates the learning material by associating with the measured value measured at the measuring point of the state quantity after a specific time has elapsed since the image was captured.
According to the configuration of the above (6), the learning data is created in consideration of a time delay until the exhaust gas or the exhaust gas generated by the combustion state shown in the past image information reaches the measurement point of the state amount. Such a time delay may be different depending on the type of the combustion furnace or the location of the measurement site. By creating a learning material in consideration of the time delay, an estimation model with high estimation accuracy can be created.

(7) In some embodiments, in the configurations of (1) to (6),
The above-mentioned estimated model creation unit is used to create a plurality of above-mentioned estimated models using a plurality of mechanical learning techniques.
According to the configuration of the above (7), the state quantities can be estimated by using a plurality of estimation models, each of which is prepared based on a plurality of mechanical learning methods (algorithms). For example, comparing the estimation results caused by each of a plurality of mechanical learning methods with the state quantities (measurement values, etc.) constituting the learning data, after that, an estimation model with high estimation accuracy can be selected from the plurality of estimation models, and it can be selected and applied Estimated models based on the conditions of the learning materials (image size, number of learning materials, etc.) or the type of burner.

(8) In some embodiments, in the configurations (1) to (7),
It is further provided with a relearning determination unit that decides to regenerate the estimated model due to relearning when the difference between the estimated value of the state quantity and the measured value of the state quantity exceeds a specific threshold.
According to the configuration of the above (8), when it is confirmed that the estimation accuracy is lowered, it is determined that it is necessary to create an estimation model due to relearning. According to the judgment, the estimation model is re-created through re-learning, and a new estimation model is obtained. Based on this, it is possible to continue the estimation due to appropriate estimation accuracy. Therefore, it is possible to accurately estimate the state quantity from the input image information while following changes in the operating environment of the upper combustion furnace and the like.

(9) An estimated model creation device according to at least one embodiment of the present invention, including an estimated model creation unit,
The estimation model creation unit is configured to perform feature values obtained from images captured in the furnace, that is, past image information obtained from the previous images, and a state corresponding to the combustion state shown in the past images. The machine did the mechanical learning corresponding to the related learning materials, and from the input image information in the furnace was taken to create an estimation model for estimating the estimated state quantity.

According to the configuration of the above (9), similar to the above (2), the learning information is created by correlating the image information during the combustion in the furnace and the state quantity when the image information was captured, and learning from the learning that has been made The materials are machine-learned, and after that, an inferred model can be created. By using this estimation model, the state quantity can be quickly estimated from the image information (input image information) in the furnace during combustion.

In addition, according to the configuration of (9), the state quantity is quickly estimated from the image information (input image information) in the furnace during combustion, and as a result, the state quantities such as the unburned portion, the NOx concentration, and the CO concentration in the ash rise. In this case, an alarm is issued to notify the operator, or an operation for reducing the amount of these states is automatically performed, whereby the optimum combustion state due to the immediate operation control can be maintained.

(10) The in-furnace state quantity estimation program related to at least one embodiment of the present invention is for causing a computer to execute the following steps:
Feature amount extraction step, which extracts feature amounts from the image taken in the furnace;
The feature quantity selection step is to select one or more of the aforementioned feature quantities from the previously extracted feature quantities;
The step of generating learning materials is to correlate the state quantity of the furnace corresponding to the combustion state shown in the foregoing image and the previously selected feature quantity to generate learning materials;
The estimation model creation step is to use the aforementioned learning materials to create an estimation model for estimating the combustion state from the image in the furnace; and the estimation state quantity calculation step is to obtain the image of the furnace in which the image has been captured and use the estimation model to estimate The aforementioned state quantity corresponding to the combustion state shown in the acquired aforementioned image;
among them,
The feature amount is selected based on the degree of contribution to the state amount to be estimated.

According to the structure of said (10), the same effect as the said (1) is exhibited.

(11) In some embodiments, in the configuration of the above (10),
The step of creating the estimated model is to perform mechanical learning using the learning materials, and then to create the estimated model.
According to the structure of said (11), the same effect as the said (2) is exhibited.

(12) An in-furnace state quantity estimation method related to at least one embodiment of the present invention, comprising:
Feature amount extraction step, which extracts feature amounts from the image taken in the furnace;
The feature quantity selection step is to select one or more of the aforementioned feature quantities from the previously extracted feature quantities;
The step of generating learning materials is to correlate the state quantity of the furnace corresponding to the combustion state shown in the foregoing image and the previously selected feature quantity to generate learning materials;
The estimation model creation step is to use the aforementioned learning materials to create an estimation model for estimating the combustion state from the image in the furnace; and the estimation state quantity calculation step is to obtain the captured image in the furnace and use the estimation model to estimate The aforementioned state quantity corresponding to the combustion state shown in the acquired aforementioned image;
among them,
The feature amount is selected based on the degree of contribution to the state amount to be estimated.

According to the structure of said (12), the same effect as the said (1) is exhibited.

(13) In some embodiments, in the configuration of the above (12),
The step of creating the estimated model is to perform mechanical learning using the learning materials, and then to create the estimated model.
According to the structure of said (13), the same effect as the said (2) is exhibited.

[Inventive effect]

According to at least one embodiment of the present invention, a state-of-furnace quantity estimation device can be provided. The state-of-furnace quantity estimation device estimates a state quantity that can determine a combustion state in the furnace quickly and accurately based on image information in the furnace.

Hereinafter, some embodiments of the present invention will be described with reference to the drawings. However, the dimensions, materials, shapes, and relative arrangement of the component parts described in the embodiments or disclosed in the drawings are not intended to limit the scope of the present invention, but are merely illustrative examples.
For example, the expression "relative or absolute configuration" such as "in a certain direction", "along a certain direction", "parallel", "orthogonal", "center", "concentric", or "coaxial" is not only Such a configuration is shown rigorously, and it also indicates a state of relative displacement due to tolerances, or an angle or distance to the extent that the same function can be obtained.
For example, the expression of "equal", "equal", and "homogeneous" is equal to the state of things, not only the strict state of equality, but also the existence of tolerances, or the difference in the degree to which the same function can be obtained. status.
For example, the expression of a shape such as a quadrangular shape or a barrel shape is not only a shape such as a quadrangular shape or a barrel shape under a geometrically rigorous meaning, but also indicates that the same effect is included in a range including the same effect. Shapes such as irregularities and chamfers.
On the other hand, the expression of "constituting", "having", "having", "including" or "having" as one of the constituent elements is not an exclusive manifestation of excluding the existence of other constituent elements.

FIG. 1 is a diagram schematically showing a configuration of a combustion furnace 7 provided with an in-furnace state quantity estimation device 1 according to an embodiment of the present invention. The combustion furnace 7 according to the embodiment shown in FIG. 1 is a stoker type garbage incinerator that uses municipal waste or business waste as fuel Fg; the furnace state quantity estimation device 1 is an estimated garbage incinerator. Among them, for example, the state amount S such as the unburned portion of the ash, the NOx concentration, and the CO concentration. Hereinafter, the present invention will be described using a garbage incinerator having a combustion range 8 (combustion chamber) for burning garbage as an example of the combustion furnace 7, but the present invention is not limited to this embodiment. In several other embodiments, the in-furnace state quantity estimation device 1 may be configured to estimate combustion in a combustion furnace 7 provided with a combustion chamber that burns fuel Fg such as a boiler or a gasification furnace that gasifies coal in IGCC. The state quantity S that changes from state to state. A garbage incinerator, a boiler, a gas burner, etc. are common in that they have an upper portion 7c and a side wall portion 7s described later. When the burner 7 is a boiler or a gas burner, the burner 7 described below can be suitably used. Renamed as boiler or vaporizer.

The waste incinerator shown in FIG. 1 will be described. In the combustion furnace 7, the fuel Fg is pushed into the furnace from the fuel supply port 71 by the fuel squeeze device 72, and then on the grate 73 (motorized grate) located in the combustion range 8 The ash is dried, burned, and burned to become ash (burning ash), and the ash is discharged out of the furnace by the ash discharge port 74. Moreover, the combustion furnace 7 is in the furnace, and has a combustion range 8; the combustion range is composed of a primary combustion range 81 and a secondary combustion range 82; the primary combustion range is located on the grate 73 and is promoted by the fuel Fg The flames are composed of a main combustion range 81a in which the flame is vigorously burned and an embers combustion range 82b in which the embers are burned; the secondary combustion range is to burn the unburned fuel in the upper part of the grate 73. The combustion gas G of the fuel Fg is supplied into the furnace through a gas supply pipe 77, and a gas supply device 76 such as a blower is supplied from the lower part of the grate 73 to the primary combustion range 81 through a first gas flow regulating valve 78a. Alternatively, the second combustion flow range 82 is supplied from the side of the combustion furnace 7 through the second gas flow rate adjustment valve 78 b from the gas supply device 76. A typical example of the combustion gas G is air, but it may be a flammable gas. For example, a specific mixing ratio may be used to mix with the EGR gas (combustion exhaust gas) discharged from the primary combustion range 81 to generate a combustion gas. G. On the other hand, the exhaust gas E generated by burning the fuel Fg passes through the exhaust gas passage 91, passes through the exhaust gas processing device 92, and is discharged from the chimney 93.

As shown in FIG. 1, the combustion furnace 7 is provided with an imaging device 6 for imaging the inside of the furnace. The imaging device 6 is, for example, an in-furnace camera capable of taking at least one of a moving image and a still image. More specifically, the imaging device 6 is, for example, a digital video camera or a video camera, as long as it is a camera such as an infrared camera capable of detecting a specific wavelength. In the embodiment shown in FIG. 1, the imaging device 6 is disposed on the upper portion 7 c of the combustion furnace 7 positioned above the primary combustion range 81 (the upper portion in FIG. 1), and constitutes a photographing of the combustion state from directly above. However, the present invention is not limited to this embodiment. In several other embodiments, the imaging device 6 may be provided at a position other than the side wall portion 7s of the combustion furnace 7 and the like, and at the upper portion 7c of the combustion furnace 7. For example, the imaging device 6 may be provided at a portion where the secondary combustion range 82 is formed in the side wall portion 7s, or a portion thereof that is higher (a portion closer to the upper portion 7c), and the imaging device 6 may be used to capture the position of the primary combustion range 81 Underneath, it is set to face obliquely downward. The image V (combustion image) in the furnace during the combustion of the fuel Fg captured by the imaging device 6 in this manner is memorized (stored) in a memory device connected to the imaging device 6. In the embodiment shown in FIG. 1, the memory device 1 m constituting the state quantity estimation device 1 stored in the furnace is configured. In other embodiments, for example, a memory device included in an estimation model creation device 2 b described later, or The memory device or the like of another device may be a memory device that is different from the body state quantity estimation device 1.

Next, the state-of-furnace quantity estimation device 1 will be described with reference to FIG. 2. FIG. 2 is a block diagram showing a function of the in-furnace state quantity estimation device 1 according to an embodiment of the present invention. As shown in FIG. 2, the state-of-furnace estimation device 1 includes an estimation model acquisition unit 3, an input image information acquisition unit 4, and an estimation state quantity calculation unit 5. In the embodiment shown in FIG. 2, the in-furnace state quantity estimation device 1 further includes an estimation model creation unit 2 configured to perform the above-mentioned functional units using an estimation model M created by the estimation model creation unit 2 (3 to 5). Estimation of the resulting state quantity S.
The functional units described above will be described separately.

In addition, the in-furnace state quantity estimation device 1 is configured by a computer, and includes a CPU (processor), a ROM of a ROM or a RAM (not shown), or a memory device 1m such as an external memory device. Next, the CPU (operation of data calculation, etc.) is executed according to a command (program in the furnace state quantity estimation program and estimation model creation program) loaded in the memory (main memory device), thereby realizing the first state quantity estimation device in the furnace. Each of the above-mentioned functional units is provided. In other words, the above-mentioned program is software that enables a computer to implement the above-mentioned functional units.

The estimation model creation unit 2 performs image information I obtained based on the image V (hereinafter, referred to as image V) in the furnace in which the combustion furnace 7 was captured, which is the past image obtained from the past image V. The information Ip and the state quantity S corresponding to the combustion state shown in the past image information Ip are mechanically learned from the learning data D associated with it, and an estimated model M learning data D is created. The obtained past image information Ip (image information I) is data (learning configuration data) corresponding to the measured value Sr or estimated value of the state amount S when the past image V was captured. The image information I may be a captured image V of the furnace itself, or it may be a feature amount F that can be extracted from the image V as described later. The image V may be a still image or a frame (image) constituting a moving image. Next, the estimation model creation unit 2 executes the mechanical learning of the learning data D by using at least one of the known methods (algorithms) of mechanical learning, and then makes an estimation for calculating the state amount S from the image information I The model M stores the created estimated model M in a memory device 1m such as an external memory device.

In the embodiment shown in FIG. 2, the unburned part in the ash is sampled in synchronization with the shooting sequence of the past image V, or the NOx concentration or the CO concentration is measured with a sensor, and the state quantity S is obtained through this. Then, the state quantity S obtained in association with the past image information Ip obtained from the past image V is associated with each other, thereby creating each learning component data. The value of the unburned portion of the ash is an index of combustion efficiency, and can be obtained by a method (JIS M 8815) of heating the ash sampled at the furnace outlet with an analyzer and measuring the weight change at this time. The details of the creation of the estimation model M will be described later.

Next, the estimated model acquisition unit 3, the input image information acquisition unit 4, and the estimated state quantity calculation unit 5 described below are the image information I (estimated state quantity Se) of the estimated value of the state quantity S to be obtained (estimated state quantity Se). The input image information It) described later is input to the estimation model M created by the estimation model creation unit 2, and the state amount S corresponding to the input image information It is then estimated.

The estimation model M created by the estimation model acquisition part 3 and the estimation model creation part 2 is acquired. In the embodiment shown in FIG. 2, the estimated model acquisition unit 3 acquires the estimated model M by reading the estimated model M stored in an external memory device into a memory such as a RAM. However, the present invention is not limited to this embodiment. In some other embodiments, the estimation model creation device 2b which is different from the state-of-furnace state quantity estimation device 1 may include an estimation model creation unit 2 (estimated model creation program). In this case, the estimation model acquisition unit 3 of the in-furnace state quantity estimation device 1 can obtain an estimation made by the estimation model creation device 2b through a portable storage medium such as a communication network or a USB memory. Model M.

The input image information acquisition unit 4 obtains the above-mentioned image information I and becomes the input image information It input to the estimation model M. The input image information It is necessary to be the same kind of information as the past image information Ip. In the embodiment shown in FIG. 2, the input image information acquisition unit 4 is connected to the imaging device 6 provided in the combustion furnace 7 and can input an image V captured by the imaging device 6 in real time. The input image information acquisition unit 4 extracts a feature amount F (described later) described later from the video V.

The estimated state amount calculation unit 5 calculates an estimated state amount Se corresponding to the combustion state indicated by the input image information It obtained by the input image information acquisition unit 4 using the estimation model M. In other words, the input image information It is calculated based on the estimated model M, and the estimated state amount Se is output as a result thereof. In the embodiment shown in FIG. 2, a display device such as a display is displayed. At this time, together with the estimated state amount Se calculated this time, one or more estimated state amounts Se that have been calculated in the past, which are memorized in the memory device 1m, etc., can be displayed together, and the time transition of the estimated state amount Se can also be displayed. . Thereby, the operator or the like of the combustion furnace 7 can quantitatively grasp the state amount S corresponding to the combustion state indicated in the video information I by confirming the estimated state amount Se displayed on the display.

According to the above-mentioned configuration, the state quantity S such as the unburned part in the ash, the NOx concentration, and the CO concentration changes in accordance with the combustion state in the furnace, and the image information I during the combustion in the furnace is used, and the image is captured The state amount S (measured value Sr, estimated value, etc.) at the time of the image information I is an estimation model M created by mechanical learning corresponding to the associated learning data D, and is estimated from the image information I in the furnace when the combustion was taken The state amount S (estimated state amount Se) generated in the combustion state. Together, it is shown that the state quantity can be estimated quickly and accurately from the image information in the furnace.

Next, several embodiments related to the estimated model creation unit 2 will be described.
In several embodiments, as shown in FIG. 2, the estimated model creation unit 2 includes a feature amount extraction unit 21, a learning data generation unit 23, and a machine learning execution unit 24. The functional units described above will be described separately.

The feature amount extraction unit 21 extracts at least one feature amount F from a past image V. The feature amount F is an index that can capture the features of the image V. In other words, the feature amount F is an index that can quantify the differences from other images V. For example, the feature amount F is information such as the shape, size, color, intensity, brightness, temperature (temperature distribution), and wavelength (wavelength distribution) appearing in the image V or obtained from the image V, or such information. The amount of change in at least one of the information is sufficient. In a case where the feature amount F is determined as the above-mentioned change amount or the like, it may be a feature obtained through, for example, a cubic high-order local autocorrelation feature amount (CHLAC feature amount), a CNN (Convolution Neural Network), or an AE (Auto Encoder). The CHLAC feature has the advantage of being able to tightly represent the temporal and spatial variations that appear in the entire image V. CNN uses convolution processing or pooling processing. Furthermore, in AE, encoding is used to obtain features. This has the advantage that the overall features of image V can be effectively obtained and reduced.

In the embodiment shown in FIG. 2, the feature amount extraction unit 21 extracts the feature amount F based on the brightness information obtained from the past image V. More specifically, the feature amount F extracted based on the brightness information may be a brightness value (brightness value) itself obtained from the image V, or may be a statistical value such as an average of brightness values, a peak, and the like. Alternatively, the above-mentioned feature quantity F may be any one of the area, shape, and size of a portion having a brightness of a specific value or more. What is necessary is just the amount of change in the brightness value. Furthermore, the brightness information may include information that can be converted from a brightness value such as a temperature or a temperature distribution obtained by converting the brightness value. The feature amount extraction unit 21 extracts one or a plurality of feature amounts F from the past image V. The brightness information is not obtained by detecting the flame itself, but is obtained from the image V. For example, even when the flame cannot be captured due to the exhaust gas E, such as when the camera 6 is installed in the upper part of the furnace, image information can be obtained. I. Thereby, the installation of the imaging device 6 (camera camera) for imaging the inside of the furnace can be facilitated.

The learning material generating unit 23 associates at least one feature quantity F extracted by the feature quantity extracting unit 21 with a state quantity S, and generates a learning material D through this. That is, one or a plurality of feature quantities F respectively extracted from the plurality of past images V corresponds to the above-mentioned past image information Ip correspondingly associated with each state quantity S. In addition, the learning material generating unit 23 may perform arbitrary preprocessing such as averaging processing on the extracted feature quantity F, and associate the preprocessed feature quantity F with the state quantity S in correspondence.

The mechanical learning execution unit 24 executes mechanical learning based on the learning data D generated by the learning data generation unit 23. The mechanical learning execution unit 24 may perform mechanical learning using any one of known mechanical learning methods (algorithms) such as neural networks and generalized linear models.

According to the configuration described above, a learning material D corresponding to the feature quantity F and the state quantity S of the past video V is created, and the learning material D is used to perform mechanical learning. In this way, the machine learning is performed focusing on the relationship between the feature quantity F and the state quantity S, and thus an estimation model M with high estimation accuracy can be created.

Moreover, in the case where the feature quantity extraction unit 21 extracts a plurality of feature quantities F, in some embodiments, as shown in FIG. 2, the estimated model creation unit 2 may further include a feature quantity selection unit 22; The quantity selection unit selects N feature quantities F (N is an integer of 1 or more) from the plurality of feature quantities F extracted by the feature quantity extraction unit 21 based on the contribution degree C to the state quantity S. In this case, the aforementioned learning material generating unit 23 associates the N feature quantities F selected by the feature quantity selecting unit 22 with the state quantities S, and generates learning materials D through this. In the embodiment shown in FIG. 2, the feature quantity selecting unit 22 automatically selects the upper N (refined) large contribution degrees C based on the preset N number. However, the present invention is not limited to this embodiment. In some other embodiments, the feature amount F and the contribution degree C automatically selected by the feature amount selection unit 22 are expressed as understood by the operator, etc., and may also constitute operations such as the check of the check box. Selection operations (selection, deselection) by the operator. In this case, the feature quantity selection unit 22 selects (acquires) the feature quantity F based on the result of the selection operation by the acceptance operator. In several other embodiments, the feature quantity selecting unit 22 may display the feature quantity F and its contribution C in a list, and accept N selected by the operator's selection operation, thereby performing the feature quantity F select.

Here, the above-mentioned contribution degree C means an index indicating a magnitude related to the state quantity S. Therefore, the higher the contribution amount C of the feature amount F, the more the state amount S changes in response to the change. If the state amount S changes in response to a change in the image V (characteristic amount F) indicating the combustion state, the contribution degree C can also be referred to as an index indicating whether the change in the state amount S can be captured by the feature amount F. Therefore, the higher the feature amount F of the contribution degree C, the more the differences in the video V can be made more prominent.

In several embodiments, the contribution degree C may also be a contribution rate in regression analysis. Specifically, for each of the plural (n) feature quantities F extracted from one past image V, the corresponding state quantities S corresponding to the corresponding image V are associated with each other, thereby creating only the feature quantity F as a phase. Unique plural (n) data settings (set of n data settings = ｛(F ₁ , S _a ), (F ₂ , S _a ), ..., (F _n , S _a )｝. N is an integer , S _a is the value of the state amount S such as the unburned part in the ash). Based on this, a plurality of learning constituent materials relating to the constituent learning materials D are obtained. Next, a regression analysis is performed on a set constituted by using a plurality of data settings obtained from a plurality of learning constituent data and having, for example, the so-called same feature amount F of F _{1 to} calculate a state amount S from the feature amount F. Regression. Next, the off-potential of the actual value may be used to divide the off-potential of the predicted value of the state quantity S calculated by the regression formula using the above-mentioned data setting, thereby calculating the contribution ratio of the regression formula.

In other embodiments, the learning configuration data may be obtained from a plurality of pieces of data, and a data set composed of a plurality of feature amounts F and a state amount S (actual value) extracted from one piece of past image information Ip may be obtained. Regression analysis, to find the regression formula (S = Σ (b _i × F _i ) + c) to obtain the state quantity S from the plurality of feature quantities F, (i = 1, 2, 3, ..., n. N is 2 The above integer.), And the contribution degree C is calculated based on the magnitude of the coefficient (coefficient b _i ) in the regression equation, for example, by calculating the ratio.

According to the above configuration, the feature amount F is refined based on the contribution degree C, and thus the estimation accuracy of the state amount S caused by the estimation model M can be improved.

However, the present invention is not limited to the embodiments described above. In some other embodiments, the in-furnace state quantity estimation device 1 (estimated model creation unit 2) may not include a feature quantity extraction unit 21, a feature quantity selection unit 22, and a learning material generation unit 23. In this case, when the learning data D of the estimation model creation unit 2 is input to the state-of-furnace estimation device 1, the mechanical learning execution unit 24 executes mechanical learning using the input learning data D itself. In this case, the learning material D may be created by another device which is different from the state amount estimation device 1 in the furnace, or may be created by manpower.

Furthermore, in some embodiments, the estimation model creation unit 2 creates a plurality of estimation models M using a plurality of mechanical learning techniques. A well-known mechanical learning method (algorithm) may be used arbitrarily, and for example, an estimation model M such as a neural network or a generalized linear model may be separately prepared.

According to the above-mentioned configuration, it is possible to estimate the state amount S by using a plurality of estimation models M, each of which is prepared based on a plurality of mechanical learning methods (algorithms). For example, comparing the estimation results caused by each of a plurality of mechanical learning methods with the state quantity S (measurement value Sr, etc.) constituting the learning data D, and then, an estimation model with high estimation accuracy can be selected from the plurality of estimation models M. M, an estimation model M suitable for the conditions (image size, number of learning materials, and the like) of the learning materials D and the type of the combustion furnace 7 can be selected.

Next, several embodiments related to the creation of the learning material D (learning material generating unit 23) will be described.
In the case where the state quantity S is a state quantity S related to the exhaust gas E generated during combustion in the furnace, or an exhaust gas such as combustion ash contained in the exhaust gas E, in several embodiments, As shown in FIG. 2, the above-mentioned learning material generating unit 23 is for the above-mentioned past image information Ip, after a certain period of time has elapsed from the time of shooting to the image V which is the basis of the past image information Ip. The measurement value Sr measured at the measurement place is correlated with each other, thereby generating learning data D. The exhaust gas E generated by the combustion of the fuel Fg flows from the combustion range 8 toward the exhaust gas path 91 (see FIG. 1). However, from the shooting point (position) shown in the image V to the state amount S When there is a distance up to the measurement point, there is a time delay (specific time) until the emission point E generated in the combustion state shown in the image V reaches the measurement point of the state amount S. Therefore, in consideration of the time delay, the image V and the measurement value Sr of the state amount S are correlated with each other. Moreover, for example, the time delay in the waste incinerator is longer than the case where the fuel Fg such as a coal-fired boiler is finely pulverized and burned.

For example, the learning material generating unit 23 may be controlled to measure the state amount S such as NOx concentration or CO concentration after a specific time has passed from the shooting sequence of the image V, and associate the image V with the measured value Sr of the state amount S in correspondence. . In the case of the unburned part in the ash, sampling may be performed after a specific time has passed from the imaging sequence of the image V, and the image V and the analysis result (measurement value Sr) of the sample may be correlated with each other. Alternatively, you can retrieve time information from the set of images V and the set of measurement values Sr of the state amount S by referring to time information such as the shooting time of the image V and the measurement time of each state amount S. From the shooting time, the measurement value Sr having a measurement time that coincides with the time after a specific time elapsed in advance is found, and the two are associated with each other.

According to the above configuration, the learning material D is created in consideration of a time delay until the emission gas E generated in the combustion state indicated by the past image information Ip reaches the measurement point of the state amount S. Such a time delay may be different depending on the type of the combustion furnace or the location of the measurement site. By creating the learning material D in consideration of the time delay, an estimation model M with high estimation accuracy can be created.

Furthermore, in some embodiments, the learning material generating unit 23 described above may create the learning material D based on parameters related to the creation of a preset learning material D. Specifically, the parameter may be at least one of the number of pixels (image size) of the left and right image quality, data that should be included in the learning component D of the learning material D, and the content processed before the image V. By adjusting the parameters, the estimation accuracy caused by the estimation model M can be appropriately set.

In addition, in several embodiments, as shown in FIG. 2, the state-of-furnace estimation device 1 further includes a relearning determination unit 13 that estimates the state quantity Se and the measured value Sr of the state quantity S. When the difference between the threshold and the threshold exceeds a certain threshold, it is decided to create an estimated model M caused by re-learning. The measured value Sr of the determined state quantity S used for relearning is measured at an arbitrary time sequence such as a regular time sequence longer than the estimation interval of the state quantity estimation device 1 in the furnace. Next, if the measured value Sr of the state amount S is input to the in-furnace state amount estimating device 1, the re-learning determination unit 13 calculates the difference between the estimated state amount Se and the measured value Sr of the state amount S, and determines the The need for learning. For example, the above-mentioned difference may be an absolute value of a result obtained by dividing the estimated state quantity Se and the measured value Sr of the state quantity S. Moreover, as long as the difference can be calculated, the calculation method is not required. In the embodiment shown in FIG. 2, when relearning is determined, the gist is notified to the estimated model creation unit 2, and the estimated model creation unit 2 uses the notification as an opportunity to rebuild the estimated model M.

According to the configuration described above, when it is confirmed that the estimation accuracy has decreased, it is determined that it is necessary to create an estimation model M due to re-learning. According to the determination, the estimation model M is re-created through re-learning, and a new estimation model M is obtained. Based on this, it is possible to continue the estimation due to appropriate estimation accuracy. Therefore, it is possible to accurately estimate the state amount S from the input image information It while following changes in the operating environment of the upper combustion furnace and the like.

Hereinafter, an in-furnace state quantity estimation method corresponding to a process executed by the above-mentioned in-furnace state quantity estimation device 1 (in-furnace state quantity estimation program) will be described with reference to FIGS. 3 to 5. Fig. 3 is a flowchart showing a method for estimating a state quantity in a furnace according to an embodiment of the present invention. FIG. 4 is a flowchart showing an estimated model creation step (S1) according to an embodiment of the present invention. 5 is a flowchart showing a relearning determination procedure according to an embodiment of the present invention. The state quantity estimation program in the furnace causes the computer to execute the following steps.

In several embodiments, as shown in FIG. 3, a method for estimating a state amount in a furnace includes an estimation model acquisition step (S2), an input image information acquisition step (S3), and an estimated state amount calculation step (S4). In this embodiment, in several embodiments, as shown in FIG. 3, the furnace state quantity estimation method further includes an estimation model creation step (S1). This estimation model creation step precedes the above-mentioned estimation model acquisition step (S2). carried out.
The method of estimating the state quantity in the furnace shown in FIG. 3 will be described in the order of steps in FIG. 3.

In step S1 in FIG. 3, an estimation model creation step is executed. The step of preparing the estimated model is a step of performing the mechanical learning of the learning material D described above and preparing the estimated model M described above. The estimation model creation step is the same as the processing performed by the estimation model creation unit 2 described above. The details will be described later. In addition, the step of creating the estimated model can also cause the computer to execute a program different from the state amount estimation program in the furnace (the estimated model creation program).

In step S2, an estimated model acquisition step is performed. The estimated model obtaining step is a step of executing the estimated model M described above. The estimated model acquisition step is the same as the processing performed by the estimated model acquisition unit 3 described above, and detailed portions are omitted.

In step S3, a step of obtaining input image information is performed. The step of obtaining input image information is a step of obtaining the above-mentioned input image information It. The input image information acquisition step is the same as the processing performed by the input image information acquisition unit 4 described above, and detailed portions are omitted.

In step S4, an estimated state amount calculation step is performed. The estimated state quantity calculation step is to use the estimated model M obtained in the estimated model acquisition step (S2) to calculate and express the input image information It obtained in the input image information acquisition step (S3) described above. The combustion state corresponding to the step of the estimated state amount Se. The estimated state quantity calculation step is the same as the processing performed by the estimated state quantity calculation unit 5 described above, and detailed portions are omitted.

According to the above configuration, the state amount S can be estimated quickly and accurately from the image information I in the furnace.

In several embodiments, as shown in FIG. 4, the estimated model creation step (S1) includes a feature amount extraction step (S11), a learning data generation step (S13), and a mechanical learning execution step (S14). In this embodiment, in several embodiments, as shown in FIG. 4, the estimated model creation step (S1) further includes a feature amount selection step (S12); the feature amount selection step is based on these feature amount extraction steps (S11) And the learning material generating step (S13).
The estimation model creation step (S1) shown in Fig. 4 will be described in the order of steps in Fig. 4.

In step S11 in FIG. 4, a feature amount extraction step is performed. The feature amount extraction step is a step of extracting at least one feature amount F from a past image V captured by the imaging device 6. The feature amount extraction step is the same as the processing performed by the feature amount extraction unit 21 described above, and detailed portions are omitted.

In step S12 of FIG. 4, a feature amount selection step is performed. The feature quantity selection step is to select N feature quantities based on the contribution to the state quantity S from the plurality of feature quantities in the case where a plurality of feature quantities F are extracted in the above-mentioned feature quantity extraction step (S11). (N represents an integer of 1 or more.) A step of a feature amount F. The feature amount selection step (S12) is the same as the processing performed by the feature amount selection unit 22 described above, and detailed portions are omitted.

In step S13 of FIG. 4, a learning material generating step is performed. The step of generating learning data is a step of generating learning data D by associating the above-mentioned past image information Ip, that is, at least one feature quantity F with a state quantity S in correspondence. The learning material generating step is the same as the processing performed by the learning material generating unit 23 described above, and detailed portions are omitted.

In step S14, a mechanical learning execution step is performed. The mechanical learning execution step is a step of executing the mechanical learning of the learning material D generated by the learning material generating step (S13) described above. The mechanical learning execution steps are the same as the processing executed by the above-mentioned mechanical learning execution unit 24, and detailed portions are omitted.

According to the configuration described above, the machine learning is performed while focusing on the relationship between the feature amount F and the state amount S. As a result, an estimation model M with high estimation accuracy can be created.

In some embodiments, as shown in FIG. 5, the method for estimating the state of the furnace further includes a relearning determination step (S51 to S55). The relearn decision step (S51 to S55) is the same as the processing performed by the relearn decision unit 13 described above. In the embodiment shown in FIG. 5, in step S51, the measurement value Sr of the state quantity S is acquired through a method of performing measurement at an arbitrary timing. In step S52, the estimated state amount Se calculated through the estimated state amount calculation step (S4) is obtained. In step S53, a difference (difference) is calculated from the measured value Sr of the state amount S by dividing the estimated state amount Se or the like. Next, if it is determined in step S54 that the above-mentioned difference exceeds a specific threshold value, re-learning is determined in step S55. On the contrary, when it is determined that the above-mentioned difference is equal to or less than a specific threshold value, it is decided that no further learning is required, and the relearning determination step is ended.

The present invention is not limited to the above-mentioned embodiment, and also includes a modified form added to the above-mentioned embodiment, or a form in which these forms are appropriately combined.

1‧‧‧furnace state quantity estimation device

1m‧‧‧memory device

13‧‧‧Re-learning decision

2‧‧‧ Presumed Model Creation Department

21‧‧‧Feature quantity extraction section

22‧‧‧Feature quantity selection department

23‧‧‧Learning Materials Generation Department

24‧‧‧ Machine Learning Executive

2b‧‧‧Presumed model creation device

3‧‧‧ Presumed model acquisition section

4‧‧‧ Input image information acquisition department

5‧‧‧Estimated state quantity calculation unit

6‧‧‧ Camera

7‧‧‧burner

7c‧‧‧ Upper part of the burner

7s‧‧‧ side wall of the burner

71‧‧‧ fuel supply port

72‧‧‧ Fuel squeeze device

73‧‧‧Grate (stoker)

74‧‧‧ash discharge

76‧‧‧Gas supply device

77‧‧‧Gas supply pipe

78a‧‧‧The first gas flow regulating valve

78b‧‧‧Second gas flow regulating valve

8‧‧‧burning range

81‧‧‧ primary combustion range

81a‧‧‧Main combustion range

82b‧‧‧ Ember burning range

82‧‧‧secondary combustion range

91‧‧‧ exhaust gas path

92‧‧‧Exhaust gas treatment device

93‧‧‧chimney

Fg‧‧‧ Fuel

G‧‧‧Combustion gas

E‧‧‧Exhaust gas

S‧‧‧Status

Measured value of Sr‧‧‧ state

D‧‧‧Learning materials

V‧‧‧Image

I‧‧‧Image Information

Ip‧‧‧Past image information

It‧‧‧ Enter image information

M‧‧‧ Presumed Model

Se‧‧‧Estimated state quantity

F‧‧‧Feature

C‧‧‧Contribution

a _i ‧‧‧ coefficient

b _i ‧‧‧ coefficient

[FIG. 1] A diagram schematically showing a configuration of a combustion furnace in which a state-of-furnace state quantity estimation device according to an embodiment of the present invention is provided.

FIG. 2 is a block diagram showing a function of an in-furnace state quantity estimation device according to an embodiment of the present invention.

3 is a flowchart showing a method for estimating a state quantity in a furnace according to an embodiment of the present invention.

4 is a flowchart showing an estimated model creation step (S1) according to an embodiment of the present invention.

5 is a flowchart showing a relearning decision procedure according to an embodiment of the present invention.

Claims (13)

A state quantity estimation device in a furnace, comprising: Feature quantity extraction unit, which extracts feature quantities from the image taken in the furnace; A feature amount selecting unit that selects one or more of the feature amounts from the extracted feature amounts; The learning material generating unit is configured to correlate the state quantity of the furnace corresponding to the combustion state shown in the foregoing image and the previously selected feature quantity, and thereby generate learning materials; An estimation model creation unit which creates an estimation model for estimating a combustion state from an image in the furnace using the learning materials; and The estimated state quantity calculation unit obtains the aforementioned image in the furnace and uses the aforementioned estimation model to estimate the aforementioned state quantity corresponding to the combustion state shown in the acquired aforementioned image; among them, The feature amount is selected based on the degree of contribution to the state amount to be estimated. For example, the in-furnace state quantity estimation device of claim 1, wherein: The estimation model creation unit performs mechanical learning using the learning materials, and thereafter, creates the estimation model. For example, the in-furnace state quantity estimation device of claim 1 or 2, wherein, The aforementioned contribution degree is an index indicating a magnitude relative to the aforementioned state quantity. For example, the in-furnace state quantity estimation device of claim 1 or 2, wherein, The aforementioned feature quantity selection unit is Calculate a regression formula that calculates the state quantity from each of the feature quantities, The contribution rate of each of the feature amounts in the calculated regression equation is defined as the contribution degree. For example, the in-furnace state quantity estimation device of claim 1 or 2, wherein, The feature amount extraction unit extracts the feature amount based on the brightness information obtained from the past image. For example, the in-furnace state quantity estimation device of claim 1 or 2, wherein, The aforementioned state quantity is a state quantity related to exhaust gas or effluent generated during combustion in the aforementioned furnace; The learning material generating unit generates the learning material by associating with the measured value measured at the measuring point of the state quantity after a specific time has elapsed since the image was captured. For example, the in-furnace state quantity estimation device of claim 1 or 2, wherein, The above-mentioned estimated model creation unit is used to create a plurality of above-mentioned estimated models using a plurality of mechanical learning techniques. For example, the in-furnace state quantity estimation device of claim 1 or 2, wherein, It is further provided with a relearning determination unit that decides to regenerate the estimated model due to relearning when the difference between the estimated value of the state quantity and the measured value of the state quantity exceeds a specific threshold. An estimation model creation device is provided with an estimation model creation unit, and the estimation model creation unit is configured to perform a feature quantity obtained from an image captured in a furnace, that is, past image information obtained from the foregoing image, and The state quantities corresponding to the combustion states shown in the previous images were subjected to mechanical learning corresponding to the associated learning data. From the input image information in the furnace, an estimation model for estimating the estimated state quantities was created. A furnace state quantity estimation program is used to make a computer perform the following steps: Feature amount extraction step, which extracts feature amounts from the image taken in the furnace; The feature quantity selection step is to select one or more of the aforementioned feature quantities from the previously extracted feature quantities; The step of generating learning materials is to correlate the state quantity of the furnace corresponding to the combustion state shown in the foregoing image and the previously selected feature quantity to generate learning materials; The estimation model creation step is to use the aforementioned learning materials to create an estimation model for estimating the combustion state from the image in the furnace; and The estimation state quantity calculation step is to acquire the aforementioned image in the furnace and use the estimation model to estimate the aforementioned state quantity corresponding to the combustion state shown in the acquired aforementioned image; among them, The feature amount is selected based on the degree of contribution to the state amount to be estimated. For example, in the furnace state quantity estimation program of item 10, The step of creating the estimated model is to perform mechanical learning using the learning materials, and then to create the estimated model. A method for estimating a state quantity in a furnace, including: Feature amount extraction step, which extracts feature amounts from the image taken in the furnace; The feature quantity selection step is to select one or more of the aforementioned feature quantities from the previously extracted feature quantities; The step of generating learning materials is to correlate the state quantity of the furnace corresponding to the combustion state shown in the foregoing image and the previously selected feature quantity to generate learning materials; The estimation model creation step is to use the aforementioned learning materials to create an estimation model for estimating the combustion state from the image in the furnace; and The estimation state quantity calculation step is to acquire the aforementioned image in the furnace and use the estimation model to estimate the aforementioned state quantity corresponding to the combustion state shown in the acquired aforementioned image; among them, The feature amount is selected based on the degree of contribution to the state amount to be estimated. For example, the method for estimating an in-furnace state quantity in claim 12, wherein: The step of creating the estimated model is to perform mechanical learning using the learning materials, and then to create the estimated model.