A method and device for identifying a boiler flame

Abstract

The application relates to a boiler flame identification method and device, which comprises the following steps: obtaining operation data from a discrete control system based on computational fluid dynamics, obtaining a flame combustion video, and extracting flame images of several working condition frames; obtaining a basic combustion physical field corresponding to the working condition based on the operation data, and obtaining flame image regions of the working condition frames and their corresponding contours based on region segmentation of the flame images; mining the basic combustion physical field and the flame image regions and their corresponding contours to obtain numerical features and graph features, wherein the numerical features comprise physical field features at different height layers of the boiler, and the graph features comprise multilayer features; performing heterogeneous feature fusion combustion state analysis according to the numerical features and the graph features to determine the advantages and disadvantages and the combustion stability of the boiler combustion state. The application has the effects that the boiler combustion analysis result is more materialized and can be explained, and more accurate boiler operation adjustment strategies are provided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of boiler combustion analysis, in particular to a boiler flame identification method and device. BACKGROUND

[0002] Under the current energy form, new energy is developing continuously. In order to fully consume the power generated by new energy and maintain the stability of the power grid, traditional thermal power generation undertakes the role of peak shaving and frequency modulation. In order to adapt to the needs of participating in peak shaving and frequency modulation, many researches on deep peak shaving and unit flexibility modification of thermal power generating units are carried out. In the peak shaving operation, the combustion condition of the boiler equipment directly affects the operation safety of the whole unit. Therefore, timely online analysis and judgment of the combustion condition of the boiler can provide accurate adjustment basis for the operation mode of the boiler.

[0003] The commonly used online combustion analysis method of the boiler equipment is based on machine learning, which trains a large amount of operation data to form an intelligent algorithm model to analyze and optimize the combustion state of the boiler. Although this method can quickly and simply form a big data analysis model, due to the black box characteristics of the intelligent algorithm model, the analysis result lacks materialized explainability, and then lacks substantial basis for the adjustment strategy of the boiler operation.

[0004] For the related technologies in the above, the inventors find that the existing online combustion analysis method of the boiler equipment has the problem of lacking materialized explainability of the analysis result, which leads to lacking substantial basis for the adjustment strategy of the boiler operation. SUMMARY

[0005] In order to make the combustion analysis result of the boiler more materialized and explainable, and to assist in providing more accurate adjustment strategy of the boiler operation, the present application provides a boiler flame identification method and device.

[0006] In a first aspect, the present application provides a boiler flame identification method.

[0007] The present application is realized by the following technical solutions:

[0008] A boiler flame identification method, comprising the following steps,

[0009] obtaining operation data from a discrete control system based on computational fluid dynamics, and

[0010] obtaining a flame combustion video and extracting flame images of a plurality of working condition frames;

[0011] obtaining a basic combustion physical field corresponding to the working condition based on the operation data, and

[0012] performing region segmentation based on the flame images to obtain flame image regions of the working condition frames and their corresponding contours;

[0013] mining the basic combustion physical field and the flame image region and the corresponding contour thereof, to obtain numerical features and graph features, wherein the numerical features include physical field features at different height layers of the boiler, and the graph features include multi-layer features;

[0014] According to the numerical features and the graph features, performing heterogeneous feature fusion combustion state analysis to determine the pros and cons of the boiler combustion state and the combustion stability.

[0015] In a preferred example, the application can be further configured to: the step of obtaining the basic combustion physical field corresponding to the working condition based on the operation data includes,

[0016] extracting current working condition operation parameters and historical operation parameters in a preset time range of the discrete control system, to obtain an initial data set of boiler operation parameters in the current working condition and a historical operation data set;

[0017] comparing the initial data set and the historical operation data set with a pre-constructed multi-layer index structure full working condition physical field database, to obtain the basic combustion physical field corresponding to the working condition.

[0018] In a preferred example, the application can be further configured to: the step of mining the basic combustion physical field includes,

[0019] According to the physical and chemical process characteristics of the coal powder entering the boiler furnace, the basic combustion physical field corresponding to the working condition is divided into a main combustion zone, a reburning zone and a burnout zone;

[0020] According to the flow path and variation characteristics of the coal powder in the height direction of the boiler, the main combustion zone, the reburning zone and the burnout zone are respectively subjected to chromatography slicing, to obtain two-dimensional field distributions at different heights;

[0021] Based on the two-dimensional field distributions, temperature combustion features and oxygen combustion features are extracted,

[0022] combining the changes in different heights and different depths, gradient features of combustion distribution changes along the longitudinal direction are extracted, to obtain the basic field features of the adjacent working condition.

[0023] In a preferred example, the application can be further configured to: the step of mining the flame image region and the corresponding contour thereof includes,

[0024] using a multi-layer feature extraction model to extract features of the flame image region and the corresponding contour thereof, to obtain basic features and high-order features;

[0025] The basic features include geometric features and color features, and based on the basic features, area, flame contour length, flame center, circularity, gray scale distribution and texture features are calculated respectively;

[0026] Based on the high-order features, center displacement degree, area change rate and brightness change rate between the current working condition frame and its previous several frames are calculated.

[0027] In a preferred example, the application can be further configured to perform heterogeneous feature fusion combustion state analysis according to the numerical features and the image features, and the step of determining the pros and cons of the boiler combustion state includes,

[0028] Based on the correlation between combustion working condition efficiency and pollutant emission, the pros and cons weight coefficient is optimized to determine;

[0029] According to the numerical features, flame image filling degree and combustion offset degree, combined with the pros and cons weight coefficient, the combustion state pros and cons index is calculated.

[0030] In a preferred example, the application can be further configured to perform heterogeneous feature fusion combustion state analysis according to the numerical features and the image features, and the step of determining the pros and cons of the boiler combustion state includes,

[0031] Based on the field combustion stability, the stability index is optimized to determine;

[0032] According to the image features, area change rate and brightness change rate, combined with the stability index, the combustion stability index is calculated.

[0033] In a preferred example, the application can be further configured to, after the step of obtaining the operation data from the discrete control system based on computational fluid dynamics,

[0034] The Random Forest algorithm is used to filter and clean the operation data to remove abnormal items in the operation data, form an effective operation data set, and replace the operation data with the effective operation data set.

[0035] In a preferred example, the application can be further configured to, after the step of extracting the flame image of the several working condition frames,

[0036] The flame image is resized to obtain a first image set;

[0037] The first image set is filtered and denoised to obtain a second image set;

[0038] The second image set is image enhanced to obtain a third image set, and the third image set is used to replace the flame image.

[0039] The application can be further configured in a preferred example, and the step of performing region segmentation on the flame image based on the flame image to obtain the flame image region of the working condition frame and the corresponding contour thereof comprises,

[0040] The Mask R-CNN model is called to perform region segmentation on the flame image to obtain the flame image region of the working condition frame and the corresponding contour thereof.

[0041] In a second aspect, the application provides a boiler flame identification device.

[0042] The application is realized through the following technical solutions:

[0043] A boiler flame identification device comprises,

[0044] A boiler data module is configured to obtain operation data from a discrete control system based on computational fluid dynamics, and obtain a flame combustion video and extract flame images of several working condition frames;

[0045] A contrast segmentation module is configured to obtain a basic combustion physical field corresponding to a working condition based on the operation data, and perform region segmentation on the flame image to obtain a flame image region of a working condition frame and a corresponding contour thereof;

[0046] A mining module is configured to mine the basic combustion physical field and the flame image region and the corresponding contour thereof to obtain numerical features and graph features, wherein the numerical features comprise physical field features at different height layers of the boiler, and the graph features comprise multi-level features;

[0047] A fusion module is configured to perform heterogeneous feature fusion combustion state analysis according to the numerical features and the graph features to determine the advantages and disadvantages of the combustion state and the combustion stability of the boiler.

[0048] In a third aspect, the application provides a boiler flame identification system.

[0049] The application is realized through the following technical solutions:

[0050] A boiler flame identification system comprises,

[0051] A camera is configured to shoot a flame combustion video;

[0052] A database server is configured to obtain operation data from a discrete control system based on computational fluid dynamics;

[0053] A system algorithm processing workstation is in communication connection with the database server and the camera, and is configured to obtain a basic combustion physical field corresponding to a working condition based on the operation data, and obtain a flame image region of a working condition frame and a corresponding contour thereof; the basic combustion physical field and the flame image region and the corresponding contour thereof are mined to obtain numerical features and graph features, wherein the numerical features include physical field features at different height layers of the boiler, and the graph features include multi-layer features; combustion state analysis is performed according to the numerical features and the graph features, and the advantages and disadvantages of the boiler combustion state and the combustion stability are determined.

[0054] In a fourth aspect, the present application provides a computer device.

[0055] The present application is implemented by the following technical solutions:

[0056] The computer device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of any one of the above boiler flame identification methods when executing the computer program.

[0057] In a fifth aspect, the present application provides a computer readable storage medium.

[0058] The present application is implemented by the following technical solutions:

[0059] The computer readable storage medium stores a computer program, and the computer program implements the steps of any one of the above boiler flame identification methods when executed by a processor.

[0060] In summary, compared with the prior art, the technical solutions provided by the present application have at least the following beneficial effects:

[0061] The operation data is obtained from a discrete control system based on computational fluid dynamics, and the flame combustion video is obtained, and the flame images of several working condition frames are extracted, so as to take the DCS operation data and the flame images as the data basis for subsequent boiler combustion analysis; based on the operation data, the basic combustion physical field corresponding to the working condition is obtained, and based on the flame image, the flame image region of the working condition frame and the corresponding contour are obtained through region segmentation, so as to form a combustion evaluation system based on three-dimensional combustion field and real-time image, and the real-time and dynamic of the boiler combustion analysis is improved; the basic combustion physical field and the flame image region and the corresponding contour are mined to obtain numerical features and image features, wherein the numerical features include physical field features at different height layers of the boiler, and the image features include multi-level features, so as to make full use of the numerical features and the image features as heterogeneous information for the boiler combustion analysis, so that the data dimension for the boiler combustion analysis is more rich; the combustion state analysis is carried out according to the heterogeneous feature fusion of the numerical features and the image features, the advantages and disadvantages of the boiler combustion state and the combustion stability are judged, so that the analysis mode is more material and can be explained, which is beneficial to assisting to provide more accurate boiler operation adjustment strategy, and meanwhile, without additional increase of multi-angle or multi-position emission / receiving equipment, the installation mode and operation and maintenance mode of the boiler combustion analysis system are more simple. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 A boiler flame identification method based on CFD and flame image multi-source heterogeneous data fusion is provided for an example embodiment of the present application.

[0063] Figure 2 A hardware topology diagram of a boiler flame identification system is provided for an example embodiment of the present application.

[0064] Figure 3 An operation data outlier cleaning schematic diagram of a boiler flame identification method is provided for another example embodiment of the present application.

[0065] Figure 4 A physical field slicing diagram of a boiler flame identification method is provided for another example embodiment of the present application.

[0066] Figure 5 A CFD field feature parameter height distribution trend diagram of a boiler flame identification method is provided for an example embodiment of the present application.

[0067] Figure 6 A flame image core region segmentation schematic diagram of a boiler flame identification method is provided for an example embodiment of the present application.

[0068] Figure 7 A structure block diagram of a boiler flame identification device is provided for an example embodiment of the present application. DETAILED DESCRIPTION

[0069] The specific embodiments are only illustrative of the present application, and are not intended to limit the present application. Those skilled in the art can make modifications to the embodiments without creative contribution after reading the present specification, and the modifications are protected by the patent law as long as they are within the scope of the claims of the present application.

[0070] To make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative contribution are within the scope of protection of the present application.

[0071] In addition, the term "and / or" in the present document is only to describe the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the three cases of existence of A alone, existence of A and B together, and existence of B alone. In addition, the character " / " in the present document generally represents an "or" relationship between the associated objects unless otherwise specified.

[0072] The existing online combustion analysis method of the boiler equipment can be divided into two categories. One is a big data analysis method based on DCS (Distributed Control System) operation data. This method combines machine learning methods to train massive operation data to form an intelligent algorithm model to analyze and optimize the boiler combustion state. Although this method can quickly and simply form a big data analysis model, due to the "black box" characteristics of the intelligent algorithm model, the analysis results of the model lack physical explainability, and thus lack substantial basis for the adjustment strategy of the boiler operation. The other is a medium-based online analysis method of the boiler combustion temperature measurement, mainly including a radiation image method and an acoustic method. Although this method has strong immediacy and dynamic characteristics, if the three-dimensional distribution of the combustion flame in the boiler furnace needs to be fully obtained, multiple emission / receiving devices at different angles and different positions need to be added, which leads to a more complex installation and operation and maintenance mode of the boiler combustion analysis system.

[0073] To this end, the application proposes a multi-source heterogeneous data fusion boiler combustion analysis method based on CFD (Computational Fluid Dynamics) and flame images based on the idea of combining numerical and graphical heterogeneous data, which can fully exploit the value of the original industrial flame television system flame image information and DCS operation data, provide strong physical and chemical basis for boiler combustion analysis, form a three-dimensional combustion field of the boiler furnace, greatly improve the physicochemical explainability of the boiler combustion analysis, and provide more accurate boiler operation adjustment strategy. At the same time, without increasing the cost of equipment installation and operation and maintenance, a more comprehensive three-dimensional combustion field of the boiler furnace can be obtained, and the immediacy and dynamic characteristics of the boiler combustion analysis are ensured.

[0074] The embodiments of the application will be further described in detail below with reference to the accompanying drawings.

[0075] The embodiments of the application provide a boiler flame identification method, and main steps of the method are described as follows.

[0076] Obtaining operation data from a discrete control system based on computational fluid dynamics, and

[0077] Obtaining a flame combustion video and extracting flame images of several working condition frames;

[0078] Based on the operation data, obtaining a basic combustion physical field corresponding to the working condition, and

[0079] Based on the flame image, performing region segmentation to obtain a flame image region of the working condition frame and a corresponding contour thereof;

[0080] Mining the basic combustion physical field and the flame image region and the corresponding contour thereof to obtain numerical features and graphical features, wherein the numerical features include physical field features at different height layers of the boiler, and the graphical features include multi-level features;

[0081] According to the numerical features and the graphical features, performing heterogeneous feature fusion combustion state analysis to determine the pros and cons of the boiler combustion state and the combustion stability.

[0082] Reference Figure 1Specifically, by obtaining DCS operation data points from a discrete control system based on computational fluid dynamics, obtaining operation data, and obtaining flame combustion video from a monitoring system and extracting flame images of several working condition frames, based on the operation data, a basic combustion physical field corresponding to the working condition is obtained, and a number of features, i.e., a basic field feature of a near neighbor working condition, are mined from the basic combustion physical field, and based on the flame image, a flame image region and its corresponding contour of the working condition frame are obtained, and a graph feature, i.e., a core region feature, is mined, and finally the number of features and the graph feature are fused for boiler combustion state analysis to determine the pros and cons of the boiler combustion state and the combustion stability.

[0083] The number of features is a series of physical field parameters such as temperature and oxygen content information of different height layers obtained from the combustion field simulated by CFD in a chromatography manner.

[0084] The graph feature is the feature information of the image obtained after image processing of the flame image frame, such as area and contour length.

[0085] The number of features and the graph feature are related to the mechanism of the boiler combustion, the data structures of the number of features and the graph feature are different, and the characteristics of the boiler combustion are reflected from different data dimensions for comprehensive analysis of the boiler combustion condition.

[0086] The advantage is that the graph feature is more intuitive, and the number of features is more stable, fast and accurate.

[0087] By the three parts of DCS operation data corresponding to the basic field feature acquisition based on CFD, multi-layer feature extraction of the combustion flame based on image processing, and heterogeneous feature fusion combustion state analysis, the number of features and the graph feature are fully utilized as heterogeneous information for boiler combustion analysis to form a combustion evaluation system based on three-dimensional combustion field and real-time image, so that the data dimension for boiler combustion analysis is more abundant, the boiler combustion analysis method is more material and can be explained, which is beneficial to auxiliary provide more accurate boiler operation adjustment strategy, at the same time, without additional increase multi-angle or multi-position transmitting / receiving equipment, the installation mode and operation and maintenance mode of the boiler combustion analysis system are more simple, also improve the real-time and dynamic of the boiler combustion analysis.

[0088] In an embodiment, the step of obtaining a basic combustion physical field corresponding to the working condition based on the operation data comprises,

[0089] Extracting the current working condition operation parameters and the historical operation parameters in a preset time range of the discrete control system to obtain an initial data set of the boiler operation parameters of the current working condition and a historical operation data set, respectively.

[0090] The initial data set and the historical operation data set are compared with a pre-constructed full-condition physical field database of a multi-level index structure to obtain a basic combustion physical field corresponding to a condition.

[0091] The full-condition physical field database can be indexed according to condition parameters, and is obtained through offline CFD numerical simulation of historical operation conditions.

[0092] The data is distributed in a tree structure according to the boiler input parameters as index features. For example, the boiler input parameters include coal quality, coal supply amount, air volume, air damper opening degree, air damper swing angle, etc. The parameters are classified according to the feature parameters, and are layered according to the importance of the feature parameters. In this embodiment, the coal quality can be preferentially divided into the first layer, and then the coal amount, the air volume, etc. The basic combustion field distribution corresponding to the condition is obtained in the full-condition physical field database of the multi-level index structure according to the data set of the furnace operation parameters, so as to more objectively determine the basic combustion physical field corresponding to the condition, and further materialize the explainability of the boiler combustion analysis.

[0093] In an embodiment, the step of mining the basic combustion physical field comprises,

[0094] According to the physical and chemical process characteristics of the pulverized coal entering the boiler furnace, the basic combustion physical field corresponding to the condition is divided into a main combustion zone, a reburning zone and a burnout zone.

[0095] Based on the two-dimensional field distribution, temperature combustion characteristics and oxygen combustion characteristics are extracted, and

[0096] Combining the changes in different heights and different depths, the gradient characteristics of the combustion distribution changes along the longitudinal direction are extracted, and the basic field characteristics of the near-neighbor condition are mined.

[0097] According to the physical and chemical process characteristics of the pulverized coal entering the boiler furnace, the boiler is divided into three zones according to the pulverized coal combustion, reburning and burnout states.

[0098] According to the flow path and change characteristics of the pulverized coal in the height direction, the main combustion zone and the reburning zone are divided according to the center height of each layer of the nozzle, and the burnout zone is divided according to every 3m interval, so as to obtain the two-dimensional field distribution at different heights through chromatography slicing.

[0099] From the two-dimensional combustion field distribution at different heights, the temperature and oxygen content directly related to combustion are extracted as combustion characteristics, and the gradient characteristics of the combustion distribution changes along the longitudinal and depth directions are extracted, i.e. the basic field characteristics of the near-neighbor condition, so as to mine the data value of the basic combustion physical field, and make the materialization and explainability of the boiler combustion analysis better.

[0100] The height direction and the depth direction of the boiler can be Figure 4 The coordinate shown in the middle is taken as an example, that is, the z coordinate direction in the figure refers to the height direction of the boiler, and the gradient feature of the combustion distribution change along the longitudinal direction can be extracted with reference to the z coordinate direction; the y coordinate direction in the figure refers to the depth direction of the boiler, and the gradient feature of the combustion distribution change along the depth direction can be extracted with reference to the y coordinate direction.

[0101] In an embodiment, the step of mining the flame image region and the contour corresponding thereto comprises,

[0102] The multi-level feature extraction model is used to extract features of the flame image region and the contour corresponding thereto, to obtain basic features and high-order features.

[0103] The basic features include geometric features and color features, and based on the basic features, the area of the region, the length of the flame contour, the center of the flame, the circularity, the gray scale distribution and the texture feature are calculated respectively.

[0104] Based on the high-order features, the center displacement degree, the area change rate and the brightness change rate between the current working condition frame and the previous several frames are calculated.

[0105] The multi-level feature extraction model includes a basic feature module and a high-order feature module, the multi-level feature extraction model is used to extract features of the flame image region and the contour corresponding thereto, to obtain basic features and high-order features, and the basic features and the high-order features are mined, so as to more directly represent the boiler combustion characteristics, and the physical and chemical explainability of the boiler combustion analysis is better.

[0106] The basic feature module includes geometric features and color features, and the area H of the region, the length L of the flame contour, the center W of the flame, the circularity C, the gray scale distribution G and the texture feature M are calculated respectively, and the specific formula is as follows:

[0107] Wherein, F(x, y) is a feature contour vector function, and P and Q are two components of the F vector.

[0108] Wherein, F(x, y) is a feature contour vector function.

[0109] The center of the flame includes Wherein, Wx is the x coordinate of the center, and Wy is the y coordinate of the center.

[0110] Wherein, H is the area of the region, L is the length of the contour, and π is the circular ratio.

[0111] G(r) = n(r) / N, wherein n(r) is the number of pixels with the gray scale value r, and N represents the total number of image pixels.

[0112] The texture feature is expressed by a gray level co-occurrence matrix (GLCM), wherein G1, G i are gray values of two pixel pairs, and θ is an angle of texture scanning.

[0113] The high-order feature is to calculate the center displacement degree O_deviation, the area change rate S_rate, and the brightness change rate B_rate between the current working condition frame and, for example, the previous 10 frames, and the specific formula is as follows:

[0114] wherein x1 and y1 are the center coordinates of the first frame, x i and y i are the center coordinates of the i-th frame, and n is the number of frames selected for sampling.

[0115] wherein H1 is the area of the first frame, H i is the area of the i-th frame, and n is the number of frames selected for sampling.

[0116] wherein B1 is the average brightness of the first frame, B i is the average brightness of the i-th frame, and n is the number of frames selected for sampling.

[0117] In an embodiment, according to the numerical features and the graphical features, the step of performing heterogeneous feature fusion combustion state analysis to determine the pros and cons of the boiler combustion state comprises,

[0118] Based on the correlation between the combustion working condition efficiency and the pollutant emission, the pros and cons weight coefficient is optimized and determined;

[0119] According to the numerical features, the flame image fullness, and the combustion deviation, the pros and cons weight coefficient is combined to calculate the combustion state pros and cons index.

[0120] In an embodiment, according to the numerical features and the graphical features, the step of performing heterogeneous feature fusion combustion state analysis to determine the stability of the boiler combustion comprises,

[0121] Based on the field combustion stability, the stability index is optimized and determined;

[0122] According to the graphical features, the area change rate, and the brightness change rate, the stability index is combined to calculate the combustion stability index.

[0123] In an embodiment, after the step of obtaining the operation data from the discrete control system based on computational fluid dynamics, the method further comprises,

[0124] The running data is screened and cleaned by using a Random Forest algorithm to remove abnormal items in the running data, to form an effective running data set, and the effective running data set is used to replace the running data.

[0125] The data abnormal item refers to a value error or a missing value caused by measurement error or monitoring failure when the DCS system collects running data. These errors or missing values are invalid for subsequent processing and analysis, so it is necessary to monitor and clean the abnormal items. By preprocessing the running data, an effective running data set is obtained, which is beneficial for more accurate boiler combustion analysis, and the analysis result is more accurate.

[0126] In an embodiment, after the step of extracting the flame images of the working condition frames, the method further comprises,

[0127] The flame images are resized to obtain a first image set;

[0128] The first image set is filtered and denoised to obtain a second image set;

[0129] The second image set is image-enhanced to obtain a third image set, and the third image set is used to replace the flame images.

[0130] By preprocessing the flame images, an effective flame image is obtained, which is beneficial for more accurate boiler combustion analysis, and the analysis result of the boiler combustion is more accurate.

[0131] In an embodiment, a Mask R-CNN model is called to perform region segmentation on the flame images to obtain flame image regions of the working condition frames and their corresponding contours.

[0132] By calling the Mask R-CNN model, the preprocessed flame images are region segmented to obtain a clearer flame core region morphology, and the segmentation method is simple and convenient.

[0133] Taking a certain four-corner tangential circle boiler equipment as an example, the specific description of each embodiment is as follows.

[0134] Referring to Figure 2 , DCS running data and flame video images are collected by using a DCS running system of a boiler equipment and an industrial flame monitoring system of the boiler equipment.

[0135] The algorithm processing and analysis module built can be carried in a system algorithm processing workstation of an electronic room.

[0136] Specifically, the boiler operation data generated by the boiler SIS system is transmitted to the system algorithm processing workstation in the electronic room through the firewall and the switch, and data communication is performed with the database server to obtain or store target data.

[0137] In addition, the video of the boiler furnace combustion flame is captured by a camera and an optical lens, and is uploaded to the hard disk recorder in the control room, and then the data is transmitted to the system algorithm processing workstation in the electronic room.

[0138] Meanwhile, the captured video of the boiler furnace combustion flame is segmented by a video segmenter and displayed on a video monitor.

[0139] Furthermore, the client in the control room directly communicates and interacts with the system algorithm processing workstation in the electronic room, which is more convenient to operate.

[0140] In the process of online operation of the DCS operation system of the boiler equipment and the industrial flame monitoring system of the boiler equipment, the system algorithm processing workstation processes and analyzes the two heterogeneous data of the real-time collected boiler operation data and flame images, and outputs the current combustion analysis diagnosis result to the client in the control room for visual display.

[0141] The technical solution of the present application mainly includes three parts: obtaining the basic field characteristics corresponding to the DCS operation data based on CFD, extracting the multi-layer characteristics of the combustion flame based on image processing, and analyzing the combustion state by fusing heterogeneous characteristics.

[0142] In the obtaining of the basic field characteristics corresponding to the DCS operation data based on CFD, the DCS operation measurement point parameters of the current working condition frame are extracted, the operation data are preprocessed by screening and cleaning to obtain an effective operation data set, the current effective operation data set is compared with the multi-level index structure full-working-condition physical field database constructed by pre-offline CFD calculation to obtain a three-dimensional basic combustion physical field, and the distribution characteristics are mined from the three-dimensional combustion physical field to obtain numerical characteristics.

[0143] Based on the above hardware architecture, the system algorithm module is built, and the DCS operation measurement point parameters and the flame image signals of the flame monitoring system are collected according to the above steps to preprocess the two multi-source heterogeneous data of number and image.

[0144] The DCS operation parameters are processed: during the online operation of the boiler, the real-time operation parameters and a period of time of the equipment are collected from the DCS system. For example, the historical operation parameters of 5 minutes are debugged and calibrated, which are transmitted to the system algorithm processing workstation through Ethernet to form the initial data set D0 of the boiler operation parameters of the current working condition and the historical operation data set D his .

[0145] The Random Forest algorithm is used to detect and eliminate data outliers, and valid items are retained to form a cleaned data set D R , as shown in Figure 3 .

[0146] In the multi-layer feature extraction process of the combustion flame based on image processing, the flame video is collected by the industrial flame television system and the current single-frame flame image is obtained by interception, and the frame flame image is preprocessed; the Mask R-CNN model is called to perform region segmentation on the preprocessed flame image, to obtain a clear flame core region morphology, and to perform region morphology feature mining to obtain a graph feature.

[0147] The flame image signal is processed: during the operation of the boiler equipment in the embodiment, the flame combustion video is obtained from the industrial flame monitoring system, the current working condition frame and the previous 10 working condition frame flame images are extracted to form an initial flame image data set P O (N0,N1,N2,…,N 10 ).

[0148] The size of the initial flame image data set sample is normalized to 1024x1024x3 to form an image data set P S (N0,N1,N2,…,N 10 ).

[0149] Gaussian noise reduction filtering is performed on each pixel point of the image set sample to obtain the filtered P G (N0,N1,N2,…,N 10 ) data set, and then histogram equalization is performed on the image set sample to obtain P H (N0,N1,N2,…,N 10 ).

[0150] Mask R-CNN is used to segment the core combustion region of the enhanced flame image P H (N0,N1,N2,…,N 10 ), to obtain the effective flame image region P R-CNN (N0,N1,N2,…,N 10 ) of the working condition frame and its corresponding effective contour L R-CNN (N0,N1,N2,…,N 10 ).

[0151] ​After the above pretreatment, an effective flame image set is obtained, and a multi-level feature extraction model is used on the effective image set to extract features of the effective flame image area, to obtain the basic features of the real-time frame of the flame image: area H, flame contour length L, flame center W, circularity C, gray scale distribution G, and texture feature M. High-order features: center displacement O_deviation, area change rate S_rate, and brightness change rate B_rate.

[0152] In the heterogeneous feature fusion combustion state analysis, the numerical features and image features from steps 1 and 2 are received, and the combustion analysis comprehensive index is calculated: the combustion state advantage and disadvantage index I Comb and the combustion stability index I Stat .

[0153] The DCS operation data based on CFD corresponds to the basic field feature acquisition step, which includes:

[0154] Step 101, during the online operation of the boiler, real-time operation parameters and historical operation parameters of a period of time (5 minutes for example debugging and calibration) are collected from the DCS system, transmitted to the system algorithm processing workstation through Ethernet, and form the initial data set D0 of the boiler operation parameters of the current working condition and the historical operation data set D his .

[0155] Step 102, pre-process the initial data set D0, and compare D his The Random Forest algorithm (RF algorithm) is used to detect data outliers, remove outliers in the data set, and retain valid items to form the cleaned data set D R .

[0156] The RF algorithm outlier detection is a non-supervised learning method, which detects outliers within a certain range by inputting a large number of DCS historical operation parameters. By setting the base evaluator number n=100, the maximum sample number max_samples=the number of input historical operation parameters, and the outlier threshold contamination=0.2 parameters, the outliers satisfying the above settings are calculated and obtained.

[0157] Step 103, arrange each working condition parameter in the effective data set, and obtain the basic combustion field distribution F base of the corresponding working condition according to the effective working condition parameters in the multi-layer index structure of the full working condition physical field database. The full working condition physical field database is indexed according to the working condition parameters, and is obtained by offline CFD numerical simulation of historical operation working conditions.

[0158] The sorting rule is to layer according to the importance of the feature parameters, such as coal quality first, then coal quantity, air quantity, etc.

[0159] Step 104, obtaining the basic combustion field distribution F base is divided into a main combustion zone, a reburning zone and a burnout zone, and the basic combustion field distribution F slice is respectively sliced in different zones to obtain two-dimensional field distribution F n at different heights.

[0160] Step 105, extracting the temperature and oxygen content directly related to combustion as combustion features from the combustion field distribution F slice (s1, s2, …, s n ) at different heights, and combining the changes in different heights and depths to extract the longitudinal and deep combustion distribution gradient features λ1, λ2, …, λ n , i.e. the near-neighbor working condition basic field features.

[0161] The working condition parameters in the data set are arranged, and the corresponding working condition combustion field distribution F base is obtained from the multi-layer index structure database according to the effective working condition parameters.

[0162] Referring to Figure 4 , the combustion field distribution is divided into a main combustion zone, a reburning zone and a burnout zone, and the basic combustion field distribution F base is divided into different slice distributions F slice (s1, s2, …, s n ) according to different zones, and combustion feature parameters are extracted from each slice distribution to form longitudinal and deep combustion distribution gradient features λ1, λ2, …, λ n , i.e. numerical features, as shown in Figure 5 .

[0163] The multi-layer feature extraction process of the combustion flame based on image processing includes:

[0164] Step 201, during the operation of the boiler equipment, obtaining flame combustion video from an industrial flame television, and extracting flame images of the current working condition frame N0 and the previous 10 working condition frames N1, N2, …, N 10 , denoted as initial flame images P O (N0, N1, N2, …, N 10 ).

[0165] Step 202, performing size normalization processing on the initial flame images P O (N0, N1, N2, …, N 10 ) in step 201 to obtain P S (N0, N1, N2, …, N 10). Wherein, the size regulation can be 1024x1024 pixel standard.

[0166] Step 203, performing noise reduction filtering processing on the regulation flame image P S (N0, N1, N2, …, N 10 ) two-dimensional Gaussian filter noise reduction processing: Obtain P G (N0, N1, N2, …, N 10 )。

[0167] Step 204, performing image enhancement processing on the noise reduction flame image P G (N0, N1, N2, …, N 10 ) histogram equalization strengthening to obtain P H (N0, N1, N2, …, N 10 )。

[0168] Step 205, using Mask R-CNN to perform core combustion area segmentation on the strengthened flame image P H (N0, N1, N2, …, N 10 ) to obtain the effective flame image area P R-CNN (N0, N1, N2, …, N 10 ) and its corresponding effective contour L R-CNN (N0, N1, N2, …, N 10 ) of the working condition frame. Wherein, the flame image core area segmentation is as shown in Figure 6 .

[0169] Step 206, using a multi-level feature extraction model to perform feature extraction processing on the effective flame image area, which includes a basic feature module and a high-order feature module. The basic feature module includes geometric features and color features, which respectively calculate the area H, flame contour length L, flame center W, circularity C, gray distribution G, and texture feature M. The high-order feature is to calculate the center displacement degree O_deviation between the current working condition frame and the previous 10 frames, the area change rate S_rate, and the brightness change rate B_rate.

[0170] Among them, the heterogeneous feature fusion combustion state analysis includes:

[0171] Step 301, receiving the numerical features and image features obtained by the foregoing steps, and performing combustion analysis comprehensive index calculation of heterogeneous feature fusion. Calculate the combustion state advantage and disadvantage index I Comb = a(λ1, λ2...λ n)+bF+cD, F is combustion fullness, D is combustion deviation, coefficients a, b, c are quality weight coefficients, which are determined by optimization combining with the correlation of combustion condition efficiency and pollutant emission; the combustion stability index I is calculated Stat = d1S_rate + d2B_rate, the stability index is quantified by adjusting according to the experience of field combustion stability.

[0172] Step 302, the boiler combustion state is analyzed according to the combustion analysis comprehensive index, wherein the combustion state quality index I is evaluated Comb related to physical field number characteristics, flame image fullness F (H, L, C, G, M) and deviation D (W, O_deviation), the current frame state of the flame is qualitatively and quantitatively judged, and the state trend change curve of multiple frames is recorded. The flame stability index represents the combustion fluctuation based on the change rate, feeds back the flame combustion stability level, and records the trend change curve of multiple frames of indexes.

[0173] Finally, the heterogeneous data acquisition and processing module in the workstation sends the current working condition number heterogeneous data characteristics of the above processing steps to the state analysis calculation processing module, fuses the above number characteristics and image characteristics, calculates the combustion analysis comprehensive index combustion state quality index I Comb = a(λ1, λ2,.. λ n )+bF+cD, the combustion stability index I is calculated Stat = d1S rate +d2B rate , the state analysis result calculation of the current period is completed, and the record display is performed in the client part.

[0174] Combining the combustion analysis of multiple periods, the trend change curve of multiple frames of combustion states is drawn, the flame combustion change is analyzed, the state reference and real-time monitoring are performed for the operation adjustment of wind and coal parameters.

[0175] To sum up, the method for identifying a boiler flame obtains operation data from a discrete control system based on computational fluid dynamics, obtains a flame combustion video, and extracts flame images of several working condition frames, so as to take the DCS operation data and the flame images as a data basis for subsequent boiler combustion analysis; based on the operation data, a basic combustion physical field corresponding to a working condition is obtained, and region segmentation is performed based on the flame images to obtain flame image regions of the working condition frames and their corresponding contours, thereby forming a combustion evaluation system based on a three-dimensional combustion field and real-time images, and the real-time and dynamic nature of the boiler combustion analysis is improved; the basic combustion physical field and the flame image regions and their corresponding contours are mined to obtain numerical features and graph features, wherein the numerical features include physical field features at different height layers of the boiler, and the graph features include multi-level features, so as to make full use of the numerical features and the graph features as heterogeneous information for the boiler combustion analysis, and make the data dimension for the boiler combustion analysis more abundant; combustion state analysis is performed according to the numerical features and the graph features, and the advantages and disadvantages and the stability of the combustion state of the boiler are determined, so that the analysis method is more physically interpretable, and it is beneficial to assist in providing more accurate boiler operation adjustment strategies, and meanwhile, without the need to additionally increase emission / reception devices at multiple angles or positions, the installation mode and the operation and maintenance mode of the boiler combustion analysis system are more convenient.

[0176] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0177] With reference to Figure 7 The embodiments of the present application also provide a boiler flame identification device, which corresponds to the above-mentioned boiler flame identification method one-to-one. The boiler flame identification device comprises,

[0178] A boiler data module is configured to obtain operation data from a discrete control system based on computational fluid dynamics, obtain a flame combustion video, and extract flame images of several working condition frames.

[0179] A contrast segmentation module is configured to obtain a basic combustion physical field corresponding to a working condition based on the operation data, and perform region segmentation based on the flame images to obtain flame image regions of the working condition frames and their corresponding contours.

[0180] A mining module is configured to mine the basic combustion physical field and the flame image regions and their corresponding contours to obtain numerical features and graph features, wherein the numerical features include physical field features at different height layers of the boiler, and the graph features include multi-level features.

[0181] A fusion module is configured to perform heterogeneous feature fusion combustion state analysis based on the numerical features and the image features, and determine the pros and cons of the boiler combustion state and the combustion stability.

[0182] The boiler flame recognition device further includes,

[0183] A preprocessing module is configured to filter and clean the operation data by using a Random Forest algorithm to remove abnormal items in the operation data, form an effective operation data set, and replace the operation data with the effective operation data set, and perform size normalization on the flame image to obtain a first image set; perform filtering and noise reduction processing on the first image set to obtain a second image set; perform image enhancement processing on the second image set to obtain a third image set, and replace the flame image with the third image set.

[0184] The specific limitations of the boiler flame recognition device can be referred to the limitations of the boiler flame recognition method described above, which will not be repeated here.

[0185] The modules of the above-mentioned boiler flame recognition device can be all or partially realized by software, hardware, and combinations thereof. The modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the modules.

[0186] In one embodiment, a boiler flame recognition system is provided, including,

[0187] A camera is configured to shoot a flame combustion video;

[0188] A database server is configured to obtain operation data from a discrete control system based on computational fluid dynamics;

[0189] A system algorithm processing workstation is communicatively connected to the database server and the camera, configured to obtain a basic combustion physical field corresponding to a working condition based on the operation data, and obtain a flame image area and a corresponding contour of a working condition frame; mine the basic combustion physical field and the flame image area and the corresponding contour to obtain numerical features and image features, wherein the numerical features include physical field features at different height layers of the boiler, and the image features include multi-level features; perform heterogeneous feature fusion combustion state analysis based on the numerical features and the image features, and determine the pros and cons of the boiler combustion state and the combustion stability.

[0190] In one embodiment, a computer device is provided, which can be a server. The computer device comprises a processor, a memory, a network interface and a database connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement any of the above-mentioned boiler flame identification methods.

[0191] In one embodiment, a computer readable storage medium is provided, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the following steps:

[0192] obtaining operation data from a discrete control system based on computational fluid dynamics, and

[0193] obtaining a flame combustion video and extracting flame images of several working condition frames;

[0194] based on the operation data, obtaining a basic combustion physical field corresponding to the working condition, and

[0195] based on the flame images, performing region segmentation to obtain flame image regions of the working condition frames and their corresponding contours;

[0196] mining the basic combustion physical field and the flame image regions and their corresponding contours to obtain numerical features and graph features, wherein the numerical features include physical field features at different height layers of the boiler, and the graph features include multi-level features;

[0197] according to the numerical features and the graph features, performing heterogeneous feature fusion combustion state analysis to determine the pros and cons of the boiler combustion state and the combustion stability.

[0198] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0199] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of functional units and modules is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the system is divided into different functional units or modules to complete all or part of the above-described functions.

Claims

1. A method of boiler flame recognition, characterized by, The method comprises the following steps, obtaining operation data from a discrete control system based on computational fluid dynamics, and obtaining flame combustion videos and extracting flame images of several working condition frames; based on the operation data, obtaining the basic combustion physical field corresponding to the working condition, and based on the flame image, region segmentation is performed to obtain the flame image region of the working condition frame and the corresponding contour thereof; mining the basic combustion physical field and the flame image region and the corresponding contour thereof to obtain numerical features and graph features, wherein the numerical features include physical field features at different height layers of the boiler, and the graph features include multi-level features; according to the numerical features and the graph features, performing heterogeneous feature fusion combustion state analysis to determine the advantages and disadvantages of the boiler combustion state and the combustion stability; wherein the step of mining the basic combustion physical field comprises, according to the physical and chemical process characteristics of the coal powder entering the boiler furnace, the basic combustion physical field corresponding to the working condition is divided into a main combustion zone, a reburning zone and a burnout zone; according to the flow path and variation characteristics of the coal powder in the height direction of the boiler, the main combustion zone, the reburning zone and the burnout zone are respectively subjected to chromatography slicing to obtain two-dimensional field distribution at different heights; based on the two-dimensional field distribution, temperature combustion features and oxygen combustion features are extracted, and combining the changes at different heights and different depths, gradient features of combustion distribution changes along the longitudinal direction are extracted to mine the basic field features of the adjacent working condition.

2. The boiler flame identification method according to claim 1, characterized in that, The step of obtaining the basic combustion physical field corresponding to the working condition based on the operation data comprises, extracting the current working condition operation parameters of the discrete control system and the historical operation parameters within a preset time range to obtain an initial data set of the boiler operation parameters of the current working condition and a historical operation data set; comparing the initial data set and the historical operation data set with a pre-constructed multi-level index structure full-working-condition physical field database to obtain the basic combustion physical field corresponding to the working condition.

3. The boiler flame identification method according to claim 1, characterized by, The step of mining the flame image region and the corresponding contour thereof comprises, using a multi-level feature extraction model to extract features from the flame image region and the corresponding contour thereof to obtain basic features and high-order features; the basic features include geometric features and color features, and based on the basic features, the area, flame contour length, flame center, circularity, gray scale distribution and texture features are calculated respectively; based on the high-order features, the center displacement degree, area change rate and brightness change rate between the current working condition frame and the previous several frames are calculated.

4. A boiler flame recognition method according to any one of claims 1-3, characterized in that, The step of performing heterogeneous feature fusion combustion state analysis to determine the advantages and disadvantages of the boiler combustion state according to the numerical features and the graph features comprises, based on the correlation between combustion working condition efficiency and pollutant emission, the advantage and disadvantage weight coefficient is optimized and determined; combustion state advantage and disadvantage indexes are calculated according to the numerical features, flame image fullness and combustion offset degree in combination with the advantage and disadvantage weight coefficient.

5. The boiler flame identification method according to claim 4, characterized in that, The step of performing heterogeneous feature fusion combustion state analysis to determine the combustion stability of the boiler according to the numerical features and the graph features comprises, based on the field combustion stability, the stability index is optimized and determined; According to the graph feature, the area change rate and the brightness change rate, and in combination with the stability index, a combustion stability index is calculated.

6. The boiler flame identification method according to claim 4, characterized by, After the step of extracting the flame images of the several working condition frames, the method further comprises, The flame images are resized to obtain a first image set; The first image set is filtered and denoised to obtain a second image set; The second image set is subjected to image enhancement processing to obtain a third image set, and the third image set replaces the flame images.

7. A boiler flame recognition device, characterized by The method comprises, a boiler data module configured to acquire operation data from a discrete control system based on computational fluid dynamics, and acquire a flame combustion video and extract flame images of several working condition frames; a contrast segmentation module configured to obtain a basic combustion physical field corresponding to a working condition based on the operation data, and perform region segmentation based on the flame images to obtain flame image regions of the working condition frames and their corresponding contours; a mining module configured to mine the basic combustion physical field and the flame image regions and their corresponding contours to obtain numerical features and graph features, wherein the numerical features comprise physical field features at different height layers of the boiler, and the graph features comprise multi-level features; the step of mining the basic combustion physical field comprises: dividing the basic combustion physical field corresponding to the working condition into a main combustion zone, a reburning zone and a burnout zone according to physical and chemical process characteristics of coal powder entering the furnace of the boiler; performing tomographic sectioning on the main combustion zone, the reburning zone and the burnout zone respectively to obtain two-dimensional field distributions at different heights according to a flow path and variation characteristics of the coal powder in the height direction of the boiler; extracting temperature combustion features and oxygen combustion features based on the two-dimensional field distributions, and extracting gradient features of combustion distribution changes along the longitudinal direction in combination with variations at different heights and different depths to mine proximate working condition basic field features; a fusion module configured to perform heterogeneous feature fusion combustion state analysis according to the numerical features and the graph features to determine a pros and cons of a boiler combustion state and combustion stability.

8. A computer device, comprising: The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

Images