WO2013146489A1

WO2013146489A1 - Combustion control device and combustion state detection device in incinerator

Info

Publication number: WO2013146489A1
Application number: PCT/JP2013/057952
Authority: WO
Inventors: 誠藤吉; 馨川端; 秀友市橋; 達也堅多
Original assignee: 日立造船株式会社
Priority date: 2012-03-30
Filing date: 2013-03-21
Publication date: 2013-10-03
Also published as: JP2013210108A; JP5755171B2; MY174994A

Abstract

The present invention is configured from: a combustion region extraction unit (24) that extracts a combustion region comprising a primary combustion stage portion with regards to a combustion fire grate and a post-combustion stage portion with regards to a post-combustion fire grate from image data obtained from a captured image acquisition unit (21); a combustion region compartmentalization unit (25) that compartmentalizes the combustion region extracted by the combustion region extraction unit into first compartments, and second compartments and third compartments larger than the first compartments to obtain three sizes of image data; a combustion state analysis unit (26) that uses the fuzzy c-means to analyze the state of combustion in the image data of each compartment obtained by the combustion region compartmentalization unit; and a combustion state detection unit (27) that detects the combustion state on the basis of the analysis data obtained by the combustion state analysis unit.

Description

Combustion state detection device and combustion control device in incinerator

The present invention relates to a combustion state detection device for detecting a combustion state in a combustion chamber in a waste incinerator and a combustion control device for controlling the combustion of waste based on the combustion state detected by the combustion state detection device.

Conventionally, in a waste incinerator, when the combustion state of waste in the furnace is recognized, that is, when the combustion state is detected, the combustion chamber in the incinerator is photographed with a photographing camera, and the photographed image is subjected to image processing and Based on the luminance value, for example, a burnout position or the like has been detected (for example, see Patent Document 1).

No. 2004-85094

However, in the case of detecting the conventional combustion state, the detection accuracy is not good for the one that performs image processing on the captured image and detects the combustion state such as the burnout position by the luminance value, It is influenced by disturbances and cannot be used for control. Therefore, there is a problem that it is not possible to sufficiently cope with the deterioration of the combustion state and the change of the flame.

Therefore, the present invention is capable of accurately detecting a combustion state such as a burnout position of garbage, and capable of sufficiently dealing with deterioration of the combustion state, change in flame, and the like, and a combustion state detection device in a waste incinerator and the combustion state An object of the present invention is to provide a combustion control device using a detection device.

In order to solve the above-mentioned problem, according to a first aspect of the present invention, there is provided a combustion state detecting device in an incinerator in which a combustion grate is disposed on a bottom wall portion of a combustion chamber and a post-combustion grate is disposed on the lower side thereof. A combustion state detection device for detecting the combustion state of waste in an incinerator,
A captured image acquisition unit that acquires captured images taken by the camera from the lower side in the combustion chamber at predetermined time intervals;
A combustion zone extraction unit for extracting a combustion zone consisting of a main combustion stage portion provided with a combustion grate and a rear combustion stage portion provided with a post combustion grate from image data acquired by the captured image acquisition unit; ,
The combustion zone extracted by the combustion zone extraction unit is divided into a first zone and a second zone and a third zone that are larger than the first zone to obtain image data of three different sizes. And
A combustion state analysis unit that analyzes the combustion state with respect to the image data of the first section, the second section, and the third section obtained by the combustion zone section using a fuzzy c average method;
The combustion state detection unit is configured to detect a combustion state based on each analysis data obtained by the combustion state analysis unit.

Moreover, the 2nd side surface of this invention is a combustion state analysis part in the combustion state detection apparatus of the 1st side surface,
A combustion discriminating unit that inputs image data of the first zone and discriminates, for each of the first zones, whether it is in a combustion state or an unburned state by a discriminator using a fuzzy c-average method;
An unburned mass discriminating unit that inputs the image data of the third zone and discriminates the presence or absence of the unburned mass by a classifier using a fuzzy c average method for each of these third zones,
The image data of the second section is input, and for each of these second sections, a flame discriminating unit that discriminates the intensity of the flame by a discriminator using a fuzzy c average method is configured.

Moreover, the 3rd side surface of this invention is a combustion state detection part in the combustion state detection apparatus of the 1st side surface,
A burnout position detector for detecting the first section of the combustion state obtained by the combustion identification unit of the combustion state analyzer to detect the burned mass and detecting the burnout position by detecting the lower end position of the burned mass When,
Based on the burnout position detected by the burnout position detection unit and the unburned lump section data obtained by the unburned lump identification unit of the combustion state analysis unit, the degree and position of the unburned lump are detected. An unburned lump detection unit;
This is composed of a flame region detection unit that detects the intensity and position of the flame based on the flame division data obtained by the flame identification unit of the combustion state analysis unit.

According to a fourth aspect of the present invention, there is provided any one of the combustion state detection devices described in the first to third aspects, a burnout position detected by the combustion state detection device, and an unburned garbage lump. The fuzzy operation unit which inputs the degree and its position, the flame intensity and its position data, performs fuzzy inference based on these data, and outputs the control command of the supply amount of combustion air and dust, and the combustion state An air control target position detection unit for inputting the position data of the unburned garbage lump and the flame position detected by the detection device and detecting the control target position of the combustion air output from the fuzzy calculation unit; Is.

Further, according to a fifth aspect of the present invention, there is provided a fuzzy arithmetic unit in the combustion control device of the fourth aspect.
A fuzzy inference data creation unit for obtaining an average value of a plurality of data input in time series and a rate of change of the average value;
The average value and change rate obtained by the fuzzy inference data creation unit are input, and a predetermined fuzzy rule is applied to the average value and change rate to obtain the output value for controlling the supply amount of combustion air and dust. It is composed of the fuzzy reasoning part to be sought.

According to the configuration of the combustion state detecting device, the combustion zone is extracted from the combustion chamber consisting of the main combustion stage portion related to the combustion grate and the rear combustion stage portion related to the post-combustion grate. The extracted combustion zone is divided into first, second and third sections of different sizes to obtain three types of image data, and the obtained first, second and third sections are obtained. Since the combustion state is analyzed with respect to the image data of the sections using the fuzzy c average method, the combustion state in the combustion chamber can be detected with high accuracy.

Further, according to the configuration of the combustion control device, the data indicating the combustion state detected by the combustion state detection device, that is, the combustion state such as the burnout position, the degree of unburned litter, and the strength of the flame is input. In addition, since the inference is performed by applying the average value and the rate of change to a predetermined fuzzy rule, it is possible to perform a control that takes into account the tendency of combustion, and thus to the deterioration of the combustion state, the change of the flame, etc. I can cope with it enough.

It is a schematic diagram which shows schematic structure of the waste incinerator which concerns on the Example of this invention. It is a block diagram which shows schematic structure of the combustion control apparatus in the waste incinerator concerning the Example. It is a block diagram which shows schematic structure of the FCM discrimination device used with the combustion control apparatus. It is explanatory drawing of the fuzzy clustering by the FCM discriminator. It is a graph which shows the fuzzy set for input used in the fuzzy reasoning part of the combustion control device. It is a graph which shows the fuzzy set for output used in the fuzzy reasoning part of the combustion control device. It is a table | surface which shows the fuzzy rule used in the same fuzzy inference part. It is a table | surface which shows the fuzzy rule used in the same fuzzy inference part. It is a table | surface which shows the fuzzy rule used in the same fuzzy reasoning part. It is a table | surface which shows the fuzzy rule used in the same fuzzy inference part. It is a table | surface which shows the fuzzy rule used in the same fuzzy inference part. It is a figure which shows the combustion state which extracted the combustion step part of the picked-up image in the combustion chamber in the refuse incinerator. It is a figure which shows the combustion state which extracted the combustion step part of the picked-up image in the combustion chamber in the refuse incinerator. It is a figure which shows the combustion state which extracted the combustion step part of the picked-up image in the combustion chamber in the refuse incinerator. It is a figure which shows the combustion state which extracted the combustion step part of the picked-up image in the combustion chamber in the refuse incinerator. It is a figure which shows the combustion state which extracted the combustion step part of the picked-up image in the combustion chamber in the refuse incinerator.

Hereinafter, a combustion state detection device and a combustion control device in a waste incinerator according to an embodiment of the present invention will be described.

In this embodiment, after detecting the combustion state of the waste (waste) in the waste incinerator, the detected data is used to determine the combustion in the incinerator, that is, at least the waste supply amount and the combustion air supply amount. Therefore, the combustion control device including the combustion state detection device will be described.

First, a schematic configuration of a waste incinerator in which the combustion control device according to the present invention is used will be described.

As shown in FIG. 1, the waste incinerator is a stoker furnace, which is disposed in the lower part of the furnace body 1 and is provided with a waste input port 2 at the front part and an incineration residue outlet 3 at the rear part. A combustion chamber 4 which is combusted and provided with a freeboard space above the front side, a first flue 6 for guiding the exhaust gas generated in the combustion chamber 4 to the outside, and a boiler portion 7 for heat recovery are arranged. And a second flue 8. Air supply nozzles 9 for blowing combustion air to the left and right sides of the combustion chamber 4 are provided on the left and right side walls of the combustion chamber 4 at the front of the furnace body 1.

At the bottom of the combustion chamber 4, a movable combustion grate 10 (10A) and a post combustion stage (also referred to as a post combustion stage part) B constituting a main combustion stage (also referred to as a main combustion stage part) A are configured. A post-combustion grate (10B) is provided. The combustion chamber 4 is supplied with dust and is also supplied with combustion air from the left and right air supply nozzles 9. These supply amounts are controlled by the combustion control device 11 according to the present invention. The rear wall 1a of the combustion chamber 4 is provided with a photographing camera 12 (for example, an industrial camera such as a CCD camera is used) for photographing the front. The captured image, that is, image data is input to the combustion control device 11.

Hereinafter, the combustion control device 11 will be described. In brief, the inside of the combustion chamber 4 is photographed from the rear wall side (incineration residue outlet side) to the front wall side (garbage inlet side). Is applied to the combustion chamber 4 to detect the combustion state such as the presence or absence of combustion, the strength of the flame, that is, the strength of the flame, and the like, and the detected data is used to supply the combustion chamber 4 so that optimum combustion is performed. It controls the supply amount of combustion air and garbage.

Hereinafter, the combustion control device 11 will be described with reference to FIG.

In this combustion control device 11, since fuzzy clustering using a fuzzy c-means (hereinafter also referred to as FCM discrimination method) is performed, an identifier based on the fuzzy c average method (hereinafter also referred to as an FCM discriminator). It is configured by software).

Clustering by this FCM discriminator is performed at a plurality of locations, and the application location is different, but the discrimination method itself is the same and will be described here.

Here, an identification method used for detection of the combustion state, particularly used in the combustion identification unit described later, will be described. However, when this fuzzy c-average method is applied, an actual identification is performed with a preliminary process for selecting various parameters. There is a process.

In the preliminary process, first, the training data is used to classify the combustion state (referred to as class 1) and the unburned state (referred to as class 2), and each class includes, for example, two clusters. Classify into clusters, and optimize each parameter of membership function (described later) using data such as cluster center position (hereinafter referred to as cluster center), variance-covariance matrix, and evaluation data obtained at this time Is planned. Note that the training data is image data of the combustion state in the combustion chamber taken in advance, and the evaluation data is also image data of the combustion state in the combustion chamber taken by the photographing camera.

Next, the fuzzy c averaging method will be described.

Since this fuzzy c-average method uses an iteratively weighted least square method, the objective function J _i is set as shown in the following equation (1).

In the equation (1), _uki is a membership, and D is a Mahalanobis distance, which is represented by the following equation (2).

S _i is a variance-covariance matrix (fuzzy variance-covariance matrix), and v _i is an average value of data, that is, the center of cluster (i), and is represented by the following formulas (3) and (4), respectively. Note that k is a data number.

Further, the mixing ratio α of the cluster (i) is expressed by the following equation (5).

Clustering by the FCM discriminator repeatedly obtains the above equations (2) to (5) until the objective function value indicated by equation (1) converges.

In the preliminary process, the cluster phase for calculating the Mahalanobis distance and the first phase of clustering the training data for each class to obtain the variance-covariance matrix, and the optimization of the membership function parameters are performed. Therefore, a second phase for classifying the evaluation data for each class is provided.

By the way, the membership (value) (tilde u _qk ) to the class q of the k-th data x _k is expressed by the following equation (6). In the formula, a tilde symbol (˜) is added to the head of u, and this “tilde u” is expressed as “* u” in the text. The membership function is expressed as “u ^* ” as shown below.

In the equation (6), u _qkj ^* is a membership function for the cluster j and is represented by the following equation (7).

π _q is a mixing ratio of class q (mixing ratio of training data), that is, a prior probability, c is the number of clusters j for each class, and is 2 (c = 2) in this embodiment.

The membership function u _qkj ^* in the above equation (7) includes a plurality of parameters, and these parameters are optimal using training data and evaluation data (only training data may be used). Is done.

And, when performing clustering, semi-hardware clustering is performed in order to shorten the calculation time. In this semi-hardware clustering, discrete values (for example, 0.6, 0.8, etc.) are used as membership.

In this semi-hardware clustering, as described above, the class is divided into two and each class is divided into two clusters. This is because experience has shown that dividing each class into two clusters allows efficient clustering. The number of clusters is not limited to two, and may be one or three or more, for example. For example, when the discrete value is 0.5, the number of clusters is 1, and when the discrete value is 1.0, two hard (not fuzzy) clusters are obtained. Of course, the number of classes is not limited to two, and may be three or more.

In the first phase, semi-hardware clustering is performed for each class, that is, for the image data of the combustion state, using the training data.

That is, in order to perform semi-hard clustering between hardware and fuzzy, membership u _ki is set as in the following equation (8).

Hereinafter, a specific procedure will be described (the number of clusters (i) is two). Note that β is specified by the discrete value described above.

First, the initial membership (u _k1 , u _k2 ) for each cluster is set as in the following equations (9) and (10).

In the above formula, f _k is a main component score of one class of image data x _k .

When membership is given by the above equation, v is obtained from equation (4), S is obtained from equation (3), and Mahalanobis distance D is obtained from equations (2) and (3), (8) Update membership u with formula. This update is repeated until membership has converged.

Next, clustering is performed for each class (classes 1 and 2) using the training data. That is, each class is divided into two clusters.

At this time, semi-hardware clustering is performed using the objective function J.

That is, the membership u for each training data that minimizes the objective function J is obtained. Thus, the center v of each cluster and its membership u are obtained.

Then, clustering is performed for each class based on the obtained membership u.

In the second phase, clustering is performed using the evaluation data.

That is, the Mahalanobis distance D is obtained for the center v of each cluster obtained in the first phase. At this time, the membership u obtained in the first phase is used.

Then, by applying (using) the obtained Mahalanobis distance D to the membership function u ^* shown by the predetermined equation (6), each parameter (m, γ, ν) in the membership function u ^* is used. , Α; these are also called free parameters or hyper parameters). In this optimization, a particle swarm optimization method (PSO) is used.

Here, this particle swarm optimization method will be briefly described.

In this particle swarm optimization method, the cluster mixing ratio α, which is the same parameter as the three parameters (m, γ, ν) in the membership function u ^* , is optimized.

That is, in this particle swarm optimization method, the search for the best position of the particle swarm is performed by the following update formulas (11) and (12).

In the above formula, Para is a parameter indicating the position of the particle and is a vector composed of α, m, γ, and ν. Velo is the velocity vector of the particle. Rand is a diagonal matrix of random numbers between “0” and “1”. w ₀ , c ₁ , and c ₂ are scalar constants. pbset and gbest are position vectors of pbset and gbest, respectively.

These free parameters are determined so as to minimize the misidentification rate of the evaluation data.

That is, optimization is performed so that the membership * u using the membership function u ^* belongs to the correct class, in other words, the correct class membership * u is larger.

In the above-described preparation process, the membership function u ^* is obtained.

Next, the actual process when the combustion state is detected using the FCM discriminator will be briefly described. Here, the case where the combustion state in the combustion chamber 4 and an unburned state are identified is demonstrated.

First, the Mahalanobis distance D with respect to the center v of each cluster obtained in the preparation step of the image data that has been photographed and subjected to predetermined processing (described later) is obtained, and this Mahalanobis distance is expressed by the above equation (6). The membership function value u ^* obtained by substitution is substituted into the above equation (5) to obtain membership * u for each class cluster. Then, by adding the membership * u, which is determined for each cluster, determine the membership * u _i of each class, in which, and the membership * u ₂ of the members of the class 1 ship * u ₁ and class 2 In comparison, the larger value is determined as the class to which the image data belongs.

Specifically, when the class 1 membership * u ₁ is large, it is determined to be in a burning state, and when the class 2 membership * u ₂ is large, it is determined to be in an unburned state. .

Also, when determining whether or not there is an unburned lump, the degree of unburned lump, that is, the probability, is output based on the value of membership * u.

Hereinafter, the combustion control device 11 including the combustion state detection device 13 that executes the above-described preparation step and actual state of the combustion state will be described.

In the following description, the free parameters (m, γ, ν, α), the cluster center v, the variance-covariance matrix S, the class mixture ratio π, and the like are elements related to identification, and are also identification parameters.

As shown in FIG. 2, the combustion control device 11 captures a captured image (a still image) in the combustion chamber 4 captured from the rear wall 1a side by the imaging camera 12 at a first predetermined time interval. For example, the captured image acquisition unit 21 that acquires (inputs) at intervals of 1 second and the balance of the combustion state in the captured image, that is, the image data acquired by the captured image acquisition unit 21 (such as the burnout point of garbage) are relatively clear. The degree of understanding can be said to be well-balanced). The combustion pass / fail judgment unit 22 provided with an FCM discriminator for judging, and the image data judged by the combustion pass / fail judgment unit 22 is a second predetermined value. As shown in FIG. 12, an image data averaging processing unit 23 that averages at a time interval, for example, an interval of 10 seconds, and image data averaged by the image data averaging processing unit 23 are input. The main combustion grate 10 (10A) and the post-combustion grate 10 (10B), that is, the combustion zone, that is, the portions other than the main combustion stage portion A and the rear combustion stage portion B are masked with black so that the combustion zone is The combustion zone extraction unit 24 to be extracted, the combustion zone partition unit 25 that is divided (blocked) into three types of sizes with respect to the combustion zone extracted by the combustion zone extraction unit 24, and the combustion zone partition unit 25 The three types of image data obtained in the above are input, and the combustion state analysis unit 26 that analyzes the combustion state using an FCM discriminator, and the analysis data analyzed by the combustion state analysis unit 26 and the FCM identification A combustion state detection unit 27 that detects a combustion state using a gas generator, and the detection data detected by the combustion state detection unit 27 are input and fuzzy inference is performed to thereby supply a combustion air supply amount (hereinafter referred to as air). A fuzzy calculation unit 28 that outputs a control command (control signal) of a supply amount of waste) and a supply amount of waste (hereinafter referred to as a waste supply amount), that is, a correction command (correction signal) for the current command value, and the combustion state detection unit 27, the position of the unburned lump and the position of the flame detected in 27 is input, and the air to be detected, that is, the right or left air supply nozzle 9, is detected from the fuzzy computing unit 28. And a control target position detection unit 29.

In the combustion quality determination unit 22, the FCM discriminator is in a state where the combustion state is generally good and the burnout position which is the burnout point is clear (balance is good), the combustion state is badly seen and the burnout position is bad. It is determined whether or not the combustion state cannot be visually confirmed (determination is impossible) due to a state in which the combustion state is not clear (balance imbalance), or ascending ash, filling with water vapor, or the like.

Then, in the combustion quality determination unit 22, for example, after determining 10 captured images, if more than half of the images are poorly balanced, a balance abnormality alarm is output, and there is no one with good balance. The subsequent processing is stopped.

In the image data averaging processing unit 23, RGB component values (luminance values) are determined for each pixel with respect to the image data (for example, 480 × 640 pixels) determined to be in good balance by the combustion quality determination unit 22. An average value is obtained.

In the combustion zone partition unit 25, the combustion zone extracted by the combustion zone extraction unit 24 has three types of sizes, that is, the first zone (about 10 × 10 pixels; it can be said that the square is a small square). It is divided into a second section (about 30 × 80 pixels; a medium-long section of a medium rectangle with a long width) and a third section (about 60 × 60 pixels; a large square section with a large width). The Briefly, the combustion zone is divided into a square first section, a rectangular second section having an area larger than the first section, and a square third section having an area larger than the first section. ing. The difference between the second section and the third section is the shape, and the size of the area of both sections is selected according to the state of the flame, so it cannot be determined which is larger. . Accordingly, when viewed in the combustion zone, as shown in FIG. 13, the first section (D1) is about 500 pieces, and as shown in FIG. 14, the third section (D3: 1 section is shown by hatching) is 14, As shown in FIG. 15, 40 second sections (D2: 1 section is indicated by hatching) are obtained. The number of these sections and the number of pixels for each section can be changed as necessary.

By the way, about the 3rd division (D3) and the 2nd division (D2), when dividing, it is the distance of the half of the division in the horizontal direction and the up-and-down direction, for example so that the adjacent divisions may overlap a part. So they are staggered. The reason why the adjacent sections are shifted from each other in this manner is to identify the combustion state in a narrow range as much as possible so that the combustion state can be accurately determined while being a large section.

In the combustion state analysis unit 26, the FCM discriminator identifies the combustion / unburned state, the presence / absence of unburned mass, that is, the degree of unburned mass, the strength of the flame, and the like.

That is, as shown in FIG. 2, the combustion state analyzing unit 26 divides the image data of each first section obtained by the combustion zone section 25 into two classes of combustion / unburned by the FCM discriminator. The combustion discriminating unit 41, the unburned mass discriminating unit 42 for determining the presence / absence of an unburned mass with the FCM discriminator using the image data of each third zone obtained by the combustion zone dividing unit 25, and the combustion zone dividing unit A flame discriminating unit 43 for discriminating the image data of the second section obtained in No. 25 into four classes according to the strength of the flame, for example, strong flame, medium flame, small flame, no flame, is provided. Has been.

In the combustion state detection unit 27, the burnout position of the burned garbage, the degree of unburned lump (the probability of being unburned litter) and its position, and the intensity of the flame, that is, a strong flame area (for example, strong fire) Flame) and its position is detected.

As shown in FIG. 2, the combustion state detection unit 27 detects a combustion mass based on the first section of the combustion state obtained by the combustion identification unit 41 of the combustion state analysis unit 26 and based on this combustion mass. A burnout position detection unit 51 that detects a burnout position by detecting the lower end position of the combustion lump, and a burnout position detected by the burnout position detection unit 51 and an unburned lump identification unit of the combustion state analysis unit 26 Based on the unburned lump data obtained at 42, the unburned lump detection unit 52 that detects the degree and position of the unburned lump and the flame obtained by the flame identification unit 43 of the combustion state analysis unit 26 And a flame region detection unit 53 for detecting a region with a strong flame based on the data.

The burnout position detection unit 51 uses the data of the first section of the combustion state obtained by the combustion identification unit 41 to extract a combustion mass that is a burning mass and scan the combustion mass from above, for example. Then, the position at which the lump has disappeared is determined as the burnout position L (conversely, the position at which the lump appeared by scanning from below may be determined as the burnout position) (see FIG. 16). .

In addition, about a combustion lump, a labeling process is performed with respect to a 1st division, and it detects when the 1st division which is burning continues eight or more (8 connection labeling process).

In the unburned lump detection unit 52, based on the data of the third zone obtained by the unburned lump identification unit 42, that is, when there is a third zone that is not burned above the burnout position, A certain degree and its position (position of the third section) are detected.

Further, in the flame area detection unit 53, similarly to the unburned litter detection unit 52, the intensity and position of the strongly burning flame (strong flame) based on the data of the second section obtained by the flame identification unit 43. (The position of the second section) is detected.

Note that, here, a numerical value of 0 to 100 (%) is used as the flame intensity based on the membership value (which is also a probability representing the probability of identification) when identified by the FCM classifier. Further, when there is an adjacent third section or second section, the position where each section overlaps is output as the position. That is, both can identify a narrow portion while being a relatively large section.

The fuzzy computing unit 28 is a fuzzy inference data creation unit (hereinafter referred to as a data creation unit) that calculates a plurality of, for example, 10 average values of combustion state detection data input in time series and a rate of change of the average value. ) 61 and the data obtained by the data creation unit 61 are input, and this data is applied to a fuzzy rule to make an inference, thereby controlling the supply amount of combustion air and dust as an output (correction command) ) To obtain a fuzzy reasoning unit 62.

First, the data creation unit 61 will be described.

That is, data indicating the burnout position, unburned lump detection data (degree of unburned lump), and flame detection data (flame intensity) are input to the data creation unit 61 in time series, and each 10 The average value E of the individual data is obtained, and the difference between the two average values E before and after the obtained data is obtained, and the rate of change [(E _k −E _k−1 ) / Δt] is obtained.

Next, the fuzzy inference unit 62 will be described.

Although not shown in the figure, the fuzzy inference unit 62 uses an average value of data obtained by the data creation unit 61 (referred to as a current value in the fuzzy set shown below) and a rate of change thereof for each input An inference unit 62a having a fuzzy rule (described later) for inferring an output membership function (corresponding to the antecedent part) from a membership function (corresponding to the antecedent part), and an output obtained by the inference part 62a A non-fuzzification unit 62b is provided for defuzzifying (numerizing) the output from the membership function for use by, for example, the mini-max method. Hereinafter, the membership function will be described as MSF.

For example, as the fuzzy set for input, as shown in FIG. 5, a set of three MSFs [VS (small), ME (medium), VB (large)] is used, and an output fuzzy set is used. As shown in FIG. 6, the one composed of five MSFs [VS (very little), MS (somewhat little), ME (zero), MB (somewhat much), VB (very much)] is used. It is done. A graph representing a specific fuzzy set will be described later.

Hereinafter, inference contents performed by the fuzzy inference unit 62 will be described in detail.
(1) First, the case where the amount of waste supply is determined will be described.

ご Regarding the amount of waste supplied, the degree of unburned waste lump, the strength of the flame, and the burnout position are considered.

For example, when the degree of unburned litter is 50% (value in the range of 0 to 100%) and the rate of change is -5% (value in the range of -50 to + 50%), it is shown in FIG. Thus, the input MSFs are ME (corresponding to 50%) and ME (corresponding to −5%), respectively, and ME (zero) is selected as the output MSF.

In this case, the smaller one of the two input MSF grades is selected, and this grade is applied to the output MSF. For example, when the output MSF (ME) has a triangular shape, a trapezoidal graph in which the upper portion is cut from the grade value is output.

Similarly, the MSF for output is also required for the flame strength and burnout position.

Then, the three MSFs obtained in this way are synthesized, and the center of gravity is obtained to obtain an output value.

As this output value, when there is an unburned garbage lump, the supply speed (that is, the feed rate of the grate) is slowed according to the degree. This is to dry the garbage. Further, when the flame strength is high, the supply speed is increased.

FIG. 8 shows a fuzzy rule for the flame strength, and FIG. 9 shows a fuzzy rule for the burnout position.
(2) Next, the case where the air supply amount is obtained will be described.

∙ Regarding the air supply amount, each position is considered in addition to the degree of unburned litter and the strength of the flame.

Hereafter, although it is the same procedure as the case of waste supply amount, it explains.

That is, when the degree of unburned lump is 50% (value in the range of 0 to 100%) and the rate of change is -5% (value in the range of -50 to + 50%), it is shown in FIG. Thus, the input MSFs are ME (corresponding to 50%) and ME (corresponding to −5%), respectively, and ME (zero) is selected as the output MSF.

Similarly, MSF for output is also required for the flame strength.

Then, the two MSFs for the unburned garbage lump and the flame strength obtained in this way are synthesized, and for example, the center of gravity is obtained, and an output value, that is, an air supply amount as a control command is obtained. As the output value, when the flame is strong, the air supply amount is decreased, and when there is an unburned garbage lump, the air supply amount is increased. In addition, the fuzzy rule regarding the intensity | strength of a flame is shown in FIG.

In the air control target position detection unit 29, as described above, when the air supply amount is controlled by the fuzzy calculation unit 28, the position data of the third section output from the unburned lump detection unit 52 and The position data of the second section output from the flame region detection unit 53 is input, and it is determined (selected) whether to control the air supply amount from the left or right air supply nozzle 9.

Therefore, a command for selecting the air supply nozzle 9 to be controlled is output together with a control command for the air supply amount.

Here, the configuration of the FCM discriminator described above will be described.

Here, the description will be made on the assumption that the combustion identification unit 41 of the combustion state analysis unit 26 identifies whether the combustion state is the combustion state or the unburned state.

As shown in FIG. 3, the FCM discriminator is a normalization unit that partitions the image data of the first section input from the combustion zone section 25 into 1024 (32 sections in the vertical and horizontal directions to have the same number of data). 71, a vectorization unit 72 that converts the image data normalized by the normalization unit 71 into a 1024-dimensional vector, and the dimension of the vector data vectorized by the vectorization unit 72 is, for example, 50 dimensions by principal component analysis. A dimensional compression unit 73 that compresses the data into the data, a Mahalanobis distance calculation unit 74 that obtains a Mahalanobis distance with respect to the center of the cluster that has been dimensionally compressed by the dimensional compression unit 73, and a Mahalanobis distance calculation unit 74. Based on the calculated Mahalanobis distance, the above equation (7) for determining the membership function u ^* and membership * u and ( 6) Based on the membership calculation unit 75 provided with the equation and the membership * u obtained by the membership calculation unit 75, the class to which the image data belongs, that is, the combustion state or the unburned state It is comprised from the state judgment part 76 which judges these. In the calculation by the dimension compression unit 73, the Mahalanobis distance calculation unit 74, the membership calculation unit 75, and the state determination unit 76, appropriate combustion chamber position information, image compression coefficients, and identification data are obtained from a database unit (not shown). Parameters are read and used. In actual operation, the identification parameter read is a value that is optimally adjusted by training or the like (the description is omitted, but a parameter adjustment unit is provided).

Specifically, the Mahalanobis distance calculation unit 74 obtains the Mahalanobis distance D with respect to the cluster center obtained in advance of the image data, and the membership calculation unit 75 obtains the membership function u ^* (function) based on the Mahalanobis distance D. Value), and membership * u is obtained based on this membership function u ^* . That is, membership * u ₁ , * u ₂ is obtained for each cluster. Then, for each class, membership is added, and memberships * u _c1 and * u _c2 in the class are obtained.

Then, the state determination unit 76 compares the obtained memberships * u _c1 and * u _{c2 in} each class, and determines that this image data belongs to the larger value.

For example, the membership in each class at the time of this determination is shown in FIG. 4, the two Class 1 and Class 2, in which respectively form two clusters 1 and cluster 2, respectively, the class 1 side, there is a cluster 2 clusters 1 and v ₁₂ of v _11, also On the class 2 side, there is also a cluster 1 of v ₂₁ and a cluster 2 of v ₂₂ . In FIG. 4, if the point of X is image data to be determined, the membership of X for one cluster 1 in class 1 is u ₁₁ , the membership of X for the other cluster 2 is u ₁₂ , and the class 2, the membership of X to class 1 is u ₂₁ , and the membership of X to the other cluster 2 is u ₂₂ , the membership u ₁ to class 1 of X is u ₁₁ + u ₁₂ , membership _{u 2} for the class _2, the _u 21 ₊ u _22. That is, the total value of membership of each cluster in each class (which is the sum of membership heights for both clusters) becomes the membership for each class. Of the memberships u ₁ and u ₂ , they belong to the larger class. If it belongs to class 1, it is in a burning state, and if it belongs to class 2, it indicates an unburned state.

Note that free parameters are searched for by the above-described particle swarm optimization method so that the misidentification rate is minimized. For example, when the cluster mixing ratio α is changed, the cluster region (shown by contour lines) in FIG. Will change, and therefore the boundaries of the two classes will also change non-linearly.

Next, an overall control method of the combustion state by the combustion control device 11 will be described.

In the state in which various identification parameters are determined in advance by the training data and the evaluation data in the combustion state detection device 13, the state in the combustion chamber 4 is captured by the captured camera 12 by the captured image acquisition unit 21. When it is taken in, first, it is determined whether or not the combustion in the combustion chamber 4 is properly performed by the combustion quality determination unit 22. The combustion quality determination unit 22 applies the FCM identification method to the captured image data of the combustion chamber 4 to determine whether combustion is normally performed. The normal case is a state in which the combustion can be controlled, for example, a case where the burnout position in the combustion chamber 4 is relatively clearly shown. On the other hand, when it is not normal, that is, when it is abnormal, it is in a state where combustion cannot be controlled. Say such a case. When such a state continues for 10 images (about 10 seconds), an alarm indicating that the combustion state is abnormal (unbalanced) is output.

When the combustion quality determination unit 22 determines that the combustion is not abnormal but normal, the image data is input to the image data averaging processing unit 23, and the image data is averaged. The RGB data for each pixel is averaged.

The averaged image data is input to the combustion zone extraction unit 24, and the main combustion stage portion A and the post combustion stage portion B in the combustion chamber 4 are extracted.

Next, the extracted combustion area is input to the combustion area section 25 and divided into three types of sections, that is, image data of the first section, the second section, and the third section. The first section and the third section are performed for the

entire combustion zone

10A, 10B, while the second section is performed for the main combustion zone 10A.

Next, these segmented image data are input to the combustion state analysis unit 26.

That is, the image data of the first section is input to the combustion identification unit 41, and it is determined for each first section whether it is in a combustion state or an unburned state by the FCM identifier.

Further, the image data of the third section is input to the unburned mass identification unit 42, and it is determined for each third section whether the combustion state is burning by the FCM identifier or the unburned state is not burning. .

Further, the image data of the second section is input to the flame discriminating unit 43, and is classified into four in accordance with the intensity of the flame, that is, the presence / absence of the flame and the strength of the flame for each second section by the FCM discriminator. That is, it is classified into four types: strong flame, medium flame, weak flame, and no flame.

Next, the combustion state detection unit 27 detects the combustion state.

That is, in the burnout position detection unit 51, the combustion / unburned data for each first section obtained by the combustion identification unit 41 is input and, for example, an 8-connected labeling process is performed on the burning first section. By performing the above, a combustion mass is detected. In addition, since it is thought that it is the state which is partially burning about the lump less than 8 connection, it does not consider.

Then, the region determined to be a combustion mass is scanned, for example, for each row of the first section from above, and the position of the lower end row is obtained. This position is detected as a burnout position.

Next, in the unburned lump detection unit 52, the burnout position detected by the burnout position detection unit 51 is input, and the third zone data of the unburned lump detected by the unburned lump detection unit 42 is input. It is determined whether or not the position of the third section is above the burnout position, and if it is above, it is determined that there is an unburned garbage lump, and the position of the third section Is detected. As the position of the third section, a value expressed as a numerical value from the front side with the total length of the combustion zone being 0 to 100 (%) is used. Also, the degree of unburned litter is output as a numerical value in the range of 0 to 100%. As this numerical value, the membership value obtained when the FCM discriminator classifies the third section as burned / unburned is used.

Next, the flame area detection unit 53 inputs the identification results identified by the flame identification unit 43 into the four types of strong flame, medium flame, weak flame and no flame, and the second zone having the highest flame intensity. The position (in this case also, the total length of the combustion zone is expressed as a numerical value from the front side with 0 to 100 (%)) and its intensity are detected as numerical values.

Next, the value obtained above, that is, burnout position data (a numerical value when the combustion range in the front-rear direction is in the range of 0 to 100%), the degree of unburned litter (in the range of 0 to 100%) ) And the flame intensity (numerical values in the range of 0 to 100%) are input to the fuzzy computing unit 28.

Then, as described above, the data creation unit 61 of the fuzzy calculation unit 28 obtains, for example, an average value for 10 pieces and a change rate thereof as fuzzy inference data based on the input data.

Next, these average values and rate of change are inferred by applying fuzzy rules prepared for each, and then de-fuzzified to obtain the waste supply amount and air supply amount, and the air control target position The air supply nozzle 9 that is the air control target detected by the detection unit 29 is selected. Then, the determined dust supply amount and air supply amount and the position of the air supply nozzle 9 to be controlled are output to the dust supply control unit and the combustion air control unit in the incinerator, and an optimum combustion state is obtained. To be controlled. For example, when the burnout position is closer to the front, the amount of waste supply is increased, that is, the waste supply speed is increased, and when there is an unburned waste lump, the air supply from the corresponding air supply nozzle 9 is supplied. When the amount is increased and the flame is strong, the corresponding air supply amount from the air supply nozzle 9 is decreased.

According to the configuration of the combustion state detection device described above and the combustion control device using the combustion state detection device, a combustion region including a main combustion stage portion related to the combustion grate and a rear combustion stage portion related to the post combustion grate is extracted in the combustion chamber. In addition, the combustion zone extracted by the combustion zone extraction unit is divided into a first zone, a second zone, and a third zone having different sizes and shapes to obtain three types of image data. Since the combustion state is analyzed using the fuzzy c average method for the image data of the first, second and third sections, the combustion state in the combustion chamber can be detected with high accuracy. .

In addition, each data indicating the detected combustion state, that is, the burnout position, the degree of unburned garbage lump, the intensity of the flame, and the like are input, and the average value and the change rate are input to a predetermined fuzzy rule. Since inference is performed by applying the control, it is possible to perform control in consideration of the tendency of combustion, and therefore, it is possible to sufficiently cope with deterioration of the combustion state, change of flame, and the like. For example, even when an unburned garbage lump is generated in the incinerator, it is possible to immediately perform optimum combustion that can eliminate the lump.

More specifically, conventionally, when detecting the combustion state, it is only necessary to objectively analyze the image in the combustion chamber, for example, to determine the presence or absence of a flame based on the magnitude of the brightness value and detect the burnout position. Therefore, there is a problem that small flames and the like that do not affect many are detected, and the detection accuracy is low. On the other hand, in this embodiment, by using an FCM discriminator incorporating an algorithm for judging the combustion state using the experience of experienced operators, it is possible to determine the position of burnout of garbage in an incinerator, particularly in a stoker furnace. The combustion state is accurately detected. That is, the combustion state in the combustion chamber is very complicated due to the burning of dust, and cannot be reproduced based on data obtained from sensors such as calculations and flow rate / temperature, but an FCM discriminator is used. Thus, the flame state can be directly learned and the judgment model can be self-generated, so that the combustion state can be detected with high accuracy. In addition, in the conventional method using luminance values, human beings give some numerical values, and the appropriate values change depending on the garbage quality and driving conditions. Leading to a decline.

The configuration of the above-described FCM discriminator will be generally and briefly described as follows.

This FCM discriminator includes a normalization unit that obtains and normalizes a plurality of representative values from captured image data (section image data) captured by the imaging camera, and a representative value obtained by the normalization unit. A vectorization unit that vectorizes the image data to be obtained, a dimension compression unit that compresses the dimension of the vector data vectorized by the vectorization unit, and training data of the vector data compressed by the dimension compression unit in advance. The Mahalanobis distance calculation unit for obtaining the Mahalanobis distance to the obtained cluster center, the Mahalanobis distance obtained by the Mahalanobis distance calculation unit are substituted into the membership function (u ^* ) shown in the following equation (13), and the function value is (14) Membership calculation section for substituting into equation (tilde u) and membership calculation And a state determination unit that divides the photographed image into two classes (either a burning state or a non-burning state) using the membership obtained in .

In the above formula, α _qj is the mixing ratio of cluster j in class q, S is the variance-covariance matrix, and π _q is the mixing ratio of class q.

Also, the principal component analysis method is used when the dimension is compressed by the dimension compression unit.

Furthermore, a particle swarm optimization method is used when optimizing each parameter (m, γ, ν, α) in the membership function (u ^* ) expressed by equation (13).

Claims

A combustion state detection device for detecting a combustion state of garbage in an incinerator in which a combustion grate is disposed on a bottom wall portion of a combustion chamber and a rear combustion grate is disposed on the lower side thereof,
A captured image acquisition unit that acquires captured images taken by the camera from the lower side in the combustion chamber at predetermined time intervals;
A combustion zone extraction unit for extracting a combustion zone consisting of a main combustion stage portion provided with a combustion grate and a rear combustion stage portion provided with a post combustion grate from image data acquired by the captured image acquisition unit; ,
The combustion zone extracted by the combustion zone extraction unit is divided into a first zone and a second zone and a third zone that are larger than the first zone to obtain image data of three different sizes. And
A combustion state analysis unit that analyzes the combustion state with respect to the image data of the first section, the second section, and the third section obtained by the combustion zone section using a fuzzy c average method;
A combustion state detection device in an incinerator characterized by comprising a combustion state detection unit for detecting a combustion state based on each analysis data obtained by the combustion state analysis unit.
The combustion state analyzer
A combustion discriminating unit that inputs image data of the first zone and discriminates, for each of the first zones, whether it is in a combustion state or an unburned state by a discriminator using a fuzzy c-average method;
An unburned mass discriminating unit that inputs the image data of the third zone and discriminates the presence or absence of the unburned mass by a discriminator using a fuzzy c average method for each of these third zones,
The image data of the second section is input, and for each of these second sections, a flame discriminating unit that discriminates the intensity of the flame with a discriminator using a fuzzy c-average method is provided. Apparatus for detecting the state of combustion in an incinerator.
Combustion state detector
A burnout position detector for detecting the first section of the combustion state obtained by the combustion identification unit of the combustion state analyzer to detect the burned mass and detecting the burnout position by detecting the lower end position of the burned mass When,
Based on the burnout position detected by the burnout position detection unit and the unburned lump section data obtained by the unburned lump identification unit of the combustion state analysis unit, the degree and position of the unburned lump are detected. An unburned lump detection unit;
The combustion in an incinerator according to claim 1, characterized by comprising: a flame region detection unit for detecting the intensity and position of the flame based on the flame division data obtained by the flame identification unit of the combustion state analysis unit. State detection device.
The combustion state detection device according to any one of claims 1 to 3, the burnout position detected by the combustion state detection device, the degree and position of unburned lump, and the intensity of the flame and data of the position And fuzzy inference based on these data, and output a control command for the supply amount of combustion air and waste,
An air control target position detection unit for detecting the control target position of the combustion air output from the fuzzy calculation unit by inputting the data of the position of the unburned garbage lump and the flame position detected by the combustion state detection device; A combustion control device in an incinerator characterized by comprising:
Fuzzy operation unit
A fuzzy inference data creation unit for obtaining an average value of a plurality of data input in time series and a rate of change of the average value;
The average value and change rate obtained by the fuzzy inference data creation unit are input, and a predetermined fuzzy rule is applied to the average value and change rate to obtain the output value for controlling the supply amount of combustion air and dust. The combustion control device for an incinerator according to claim 4, comprising: a fuzzy reasoning unit to be obtained.