A system and method for monitoring a flame front of a boiler based on infrared light flow

Abstract

This invention provides a system and method for monitoring the flame front of a counter-current boiler based on infrared optical flow, belonging to the field of boiler combustion monitoring and control technology, which can partially solve the problem of difficulty in online quantitative monitoring of the counter-current combustion front. The system includes: a pair of symmetrically arranged infrared monitoring modules at the same preset elevation on both sides of the furnace sidewalls, used to acquire infrared image sequences of the counter-current combustion area; an image processing and analysis module, used for blackbody calibration to obtain a radiation intensity distribution map, performing image preprocessing, initial threshold segmentation, and convolutional neural network semantic segmentation, extracting the front contour and calculating its two-dimensional position parameters in the furnace longitudinal section coordinate system, obtaining the motion vector field based on dense optical flow and extracting the fluctuation rate / frequency, offset, and oscillation parameters; and an early warning and combustion adjustment decision module, used for stability assessment, early warning classification, and outputting suggestions for adjusting the secondary damper and pulverized coal distribution.

Description

Technical Field

[0001] This invention belongs to the field of boiler combustion monitoring and control technology, specifically relating to a counter-current boiler flame front monitoring system and method based on infrared optical flow. Background Technology

[0002] Coal-fired boilers remain one of the most important heat energy conversion devices in thermal power units. The economy and safety of their combustion process directly determine the unit's coal consumption for power generation, load response capability, and long-term reliable operation level. With increasingly stringent requirements for energy conservation, emission reduction, and pollutant control, boiler combustion operation is gradually shifting from "experience-based adjustment" to "refined, data-driven, and intelligent" operation. Especially under conditions such as low-NOx combustion and deep peak shaving, even slight deviations in the furnace aerodynamic field and fuel / air ratio can lead to unstable combustion, localized high temperatures, or enhanced reducing atmosphere, resulting in increased nitrogen oxide production, fluctuating fly ash carbon content, and exacerbated slagging / ash accumulation. This can even cause abnormal localized heat loads on water-cooled walls and safety risks. Therefore, obtaining key visual information that reflects the true combustion organization within the furnace and using it for operation optimization and early warning is a crucial foundation for the digital operation and intelligent control of thermal power boilers.

[0003] In large coal-fired boilers, counter-current combustion is widely used: burners are arranged on both sides of the furnace (such as the front and rear walls or left and right walls), and the pulverized coal jets ejected from both sides collide, penetrate, mix, and burn against each other in the central area of ​​the furnace, forming a dynamically changing flame front structure. The position, symmetry, and stability of the flame front in space essentially reflect the balance between fuel and secondary air organization on both sides of the furnace and the quality of the formation of the central mixing combustion zone, serving as a direct window for judging whether the aerodynamic field is reasonable. If the front is consistently biased to one side, it often indicates an imbalance in pulverized coal concentration, air volume distribution, or swirl intensity on both sides, which may lead to problems such as poor local burnout, localized high-temperature adhesion to the wall, and increased thermal deviation of the heating surface. If the front exhibits periodic oscillations with a specific frequency or large amplitude, it may indicate combustion instability or abnormal recirculation structure, increasing operational risks. Therefore, online monitoring and quantitative assessment of the boiler flame front are of great significance for improving combustion efficiency, reducing pollutants (especially nitrogen oxides), and ensuring the safety of water-cooled walls.

[0004] Current combustion monitoring of power plant boilers largely relies on indirect measurement signals such as flue gas composition, furnace negative pressure, and bed / furnace temperature, or on limited flame presence and local image information provided by devices such as flame detectors and flame television. However, for the critical area of ​​opposed combustion boilers—the opposing mixing and front formation zone in the center of the furnace—existing monitoring methods have significant shortcomings: on the one hand, in-furnace flame television usually does not cover the area where the opposed burner jets oppose each other in the center; on the other hand, the monitoring field of view of flame detectors is small, making it difficult to form imaging information that can be used for analysis. Therefore, specialized monitoring methods for the flame front of opposed boilers are relatively lacking, failing to provide intuitive evidence for combustion organization diagnosis. In addition, under the conditions of high furnace temperature, high dust, strong radiation, and complex background, even if some images or signals are obtained, problems such as information lag or missing information, insufficient imaging anti-interference ability, and inaccurate extraction of flame boundary / interaction area features are still common, making it difficult to stably, reliably, and quantitatively identify and track the front position. Furthermore, existing methods often only provide "flame / no flame" or rough brightness changes, making it difficult to simultaneously characterize the static spatial offset and dynamic fluctuation characteristics of the flame front (such as oscillation amplitude and frequency). Therefore, it is difficult to achieve early identification of combustion instability signs and to support the requirements of closed-loop combustion fine adjustment and automatic optimization control. To address this, we propose an infrared optical flow-based system and method for monitoring the flame front of counter-current boilers. Summary of the Invention

[0005] The present invention aims to at least solve one of the technical problems existing in the prior art, and provides a system and method for monitoring the flame front of a counter-current boiler based on infrared optical flow.

[0006] This invention provides, in one aspect, an infrared optical flow-based flame front monitoring system for counter-current boilers, comprising: An infrared monitoring module, comprising at least a pair of infrared monitoring units symmetrically arranged on opposite side walls of the furnace of the opposed boiler at the same preset elevation, for acquiring infrared image sequences of the opposed combustion zone of the furnace; An image processing and analysis module, communicatively connected to the infrared monitoring module, is used to perform radiation intensity calibration and image preprocessing on the infrared image sequence. It then performs pixel-level segmentation on the preprocessed infrared images to obtain segmentation results including at least a background area, a first-side flame body, a second-side flame body, and a flame front interaction area. Based on the segmentation results, it extracts the contour of the flame front and calculates the two-dimensional position parameters of the flame front in the longitudinal section coordinate system of the furnace. Finally, it performs optical flow calculation on consecutive frames of infrared images containing the flame front interaction area to obtain the flame front motion vector field and outputs the front dynamic parameters. The early warning and combustion adjustment decision module is communicatively connected to the image processing and analysis module. It is used to quantitatively evaluate the stability of the flame front based on the two-dimensional position parameters and / or the front dynamic parameters, and output corresponding early warning information according to preset grading criteria.

[0007] Furthermore, the installation elevation of the infrared monitoring module is set to the elevation corresponding to the center line of the burner nozzle of the counter-flow boiler or within a preset range above and below it, and the infrared monitoring modules on the side walls of the counter-flow boiler are symmetrically arranged in the horizontal position and in the pitch angle, so that the flame areas collected on both sides form a corresponding observation relationship in space.

[0008] Specifically, the infrared monitoring module is a mid-wave infrared imaging device, and the imaging band range of the infrared monitoring module is 3 to 5 micrometers.

[0009] Specifically, the infrared monitoring module is equipped with a cooling and protection kit to adapt to the furnace environment. The infrared monitoring module is synchronously triggered and collected by a central processing unit or a background server to ensure the temporal consistency of the infrared image sequence.

[0010] Preferably, the image processing and analysis module includes a grayscale-radiance intensity calibration unit, which establishes a mapping relationship between the pixel grayscale values ​​of the infrared image and the target radiation intensity through blackbody calibration, so as to convert the original grayscale image of the infrared image into a radiation intensity distribution map.

[0011] Specifically, the image preprocessing includes at least non-uniformity correction, bad pixel repair, and background noise suppression. After the image preprocessing is completed, an adaptive threshold segmentation algorithm is used on the radiation intensity distribution map to perform preliminary classification of each pixel in the radiation intensity distribution map according to a preset radiation intensity interval threshold, so as to obtain preliminary segmentation results of the background area, unburned coal powder jet area, and high-temperature combustion area in the radiation intensity distribution map. The preliminary segmentation results are used as prior information and / or training input for subsequent flame semantic segmentation.

[0012] Furthermore, the pixel-level segmentation is implemented using a flame semantic segmentation model based on a convolutional neural network. The flame semantic segmentation model takes the infrared image after preprocessing a single frame or a time series composed of multiple preprocessed infrared images as input, performs semantic classification on the pixels in the input infrared image, and outputs the category label and / or category probability of each pixel in the infrared image belonging to a preset category set. The preset category set includes at least the background area, the first side flame main body area, the second side flame main body area, and the flame front interaction area.

[0013] Further, extracting the contour of the flame front based on the segmentation result includes performing edge detection on the boundary of the flame front interaction area based on the pixel-level segmentation result to obtain the front contour line; calculating the two-dimensional position parameters of the flame front in the longitudinal section coordinate system of the furnace includes performing symmetrical matching on the corresponding front contour lines in the images acquired from the side walls on both sides of the furnace, and calculating the two-dimensional coordinates of each point on the front contour line in the longitudinal section coordinate system of the furnace accordingly.

[0014] Furthermore, the frontal dynamic parameters include at least the frontal fluctuation rate and the frontal fluctuation frequency, which are obtained by performing dense optical flow calculations on continuous frame flame frontal images or frontal interaction zone images; the two-dimensional position parameters include at least the average frontal center offset, which is calculated based on the center position of the boundary frontal contours of the first side flame body and the flame frontal interaction zone and the boundary frontal contours of the second side flame body and the flame frontal interaction zone, and the frontal oscillation amplitude and oscillation frequency are determined according to the change of the average frontal center offset; the early warning and combustion adjustment decision module sets multi-level early warning thresholds based on the average frontal center offset and its oscillation amplitude / oscillation frequency, and outputs combustion adjustment suggestions for correcting the aerodynamic field when the control trigger conditions are met, the combustion adjustment suggestions include symmetrical adjustments to the secondary air dampers of the burners on both sides of the opposed boiler and / or adjustments to the pulverized coal concentration distribution on both sides of the opposed boiler.

[0015] Another aspect of the present invention provides a method for monitoring the flame front of a counter-current boiler based on infrared optical flow. The method is implemented using the aforementioned infrared optical flow-based counter-current boiler flame front monitoring system and includes the following steps: S1: The infrared monitoring units are symmetrically arranged at the same preset elevation on the opposite side walls of the furnace to collect infrared image sequences of the opposing combustion zones of the furnace. S2: Establish the mapping relationship between pixel grayscale values ​​and radiation intensity through blackbody calibration, and convert the infrared image sequence into a radiation intensity distribution map; S3: Perform non-uniformity correction, bad pixel repair and background noise suppression on the radiation intensity distribution map, and use adaptive threshold segmentation to obtain the preliminary classification results of the background area, unburned pulverized coal jet area and high-temperature combustion area; S4: Perform pixel-level flame semantic segmentation on the preprocessed radiation intensity distribution map to obtain a segmentation result that includes at least the background area, the first side flame body, the second side flame body, and the flame front interaction area; S5: Extract the flame front contour line based on the segmentation result, and perform symmetrical matching on the corresponding front contour lines in the two side images to calculate the two-dimensional coordinates of the front contour line in the longitudinal section coordinate system of the furnace, thereby obtaining the average offset of the front center. S6: Perform dense optical flow calculation on consecutive frame images containing the flame front interaction area to obtain the motion vector field of the flame front interaction area, and calculate the fluctuation rate and fluctuation frequency of the flame front based on the motion vector field; and determine the oscillation amplitude and oscillation frequency of the flame front according to the curve of the average offset of the front center over time. S7: The stability of the flame front is quantitatively evaluated based on the average offset of the front center, the oscillation amplitude, and the oscillation frequency, and the corresponding warning level is output according to the preset grading criteria; when the average offset of the front center and / or the oscillation amplitude and the oscillation frequency meet the preset control trigger threshold conditions, a combustion adjustment suggestion is output, which includes symmetrical adjustment of the secondary air dampers on both sides of the counter-flow boiler and / or adjustment of the pulverized coal concentration distribution on both sides of the counter-flow boiler.

[0016] The beneficial effects of this invention are as follows: An infrared monitoring module is set up to extract the frontal contour through bilateral symmetrical infrared imaging and reconstruct its two-dimensional position in the longitudinal section of the furnace, solving the problem of the front being difficult to monitor directly. The motion vector field is obtained by using dense optical flow to extract the fluctuation rate / frequency, and the oscillation amplitude / frequency is obtained by combining the offset time series, realizing the joint characterization of "position + fluctuation". Based on the offset and oscillation index, a graded early warning is issued, and after triggering, adjustment suggestions such as secondary air damper / pulverized coal distribution are given, and the effect can be verified through subsequent monitoring. Attached Figure Description

[0017] Figure 1 The diagram shows the structural connection of an infrared optical flow-based counter-current boiler flame front monitoring system according to a specific embodiment of the present invention. Figure 2 The flowchart illustrates the steps of a method for monitoring the flame front of a counter-current boiler based on infrared optical flow, according to a specific embodiment of the present invention. Figure 3 Infrared images of power plant burner flames, representing a specific embodiment of the present invention, of a counter-current boiler flame front monitoring method based on infrared optical flow. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] like Figure 1As shown in the figure, a specific embodiment of the present invention provides a flame front monitoring system for a counter-current boiler based on infrared optical flow, comprising: an infrared monitoring module, wherein the infrared monitoring module includes at least a pair of infrared monitoring units symmetrically arranged on opposite side walls of the counter-current boiler furnace at the same preset elevation, for acquiring infrared image sequences of the counter-current combustion zone of the furnace; and an image processing and analysis module, communicatively connected to the infrared monitoring module, for performing radiation intensity calibration and image preprocessing on the infrared image sequences, and performing pixel-level segmentation on the preprocessed infrared images to obtain segmentation results including at least a background area, a first-side flame body, a second-side flame body, and a flame front interaction area. Furthermore, based on the segmentation results, the contour of the flame front is extracted and the two-dimensional position parameters of the flame front in the longitudinal section coordinate system of the furnace are calculated. Optical flow calculation is performed on continuous frame infrared images containing the flame front interaction zone to obtain the flame front motion vector field and output the front dynamic parameters. A warning and combustion adjustment decision module is also included, which is communicatively connected to the image processing and analysis module. It is used to quantitatively evaluate the stability of the flame front based on the two-dimensional position parameters and / or the front dynamic parameters, and output corresponding warning information according to preset grading criteria. When the two-dimensional position parameters and / or the front dynamic parameters meet the preset control trigger threshold conditions, combustion adjustment suggestions are output.

[0020] Specifically, the external monitoring module is used to acquire real-time infrared imaging information of the opposing combustion zone. The raw infrared grayscale image sequence acquired by the infrared monitoring module is uniformly triggered and received by the background server or central processor located in the control room, and transmitted in real time to the image processing and analysis module to ensure that the images on both sides are aligned and comparable in the time dimension.

[0021] Furthermore, the image processing and analysis module can be executed sequentially according to the process: preprocessing and blackbody calibration of the original infrared grayscale image sequence to convert the grayscale image into a standardized radiation intensity distribution map; performing adaptive threshold segmentation on the radiation intensity distribution map to obtain preliminary segmentation results; then calling a deep learning semantic segmentation model to perform fine pixel-level segmentation of the flame and front regions; subsequently, extracting the front contour based on the segmentation results and calculating the two-dimensional spatial position parameters of the front in the longitudinal section coordinate system of the furnace by combining the bilateral symmetrical observation relationship; finally, performing dense optical flow analysis on continuous frame images to obtain front dynamic parameters (such as fluctuation rate, fluctuation frequency, etc.), thereby forming a set of quantitative features that can be used for stability assessment and early warning decision-making.

[0022] Furthermore, the early warning and combustion adjustment decision module can compare the above quantitative results with the built-in stability assessment model / expert rules, output the early warning level and combustion adjustment suggestions; and can automatically lift the early warning after the offset and oscillation index return to the normal range, realizing closed-loop optimization of "monitoring-diagnosis-guidance-re-monitoring verification".

[0023] Based on the above basic implementation method, the installation elevation of the infrared monitoring module is set to the elevation corresponding to the center line of the burner nozzle of the counter-flow boiler or within a preset range above and below it. The infrared monitoring modules on the side walls of the counter-flow boiler are symmetrically arranged in terms of horizontal position and pitch angle so that the flame areas collected on both sides form a corresponding observation relationship in space. The infrared monitoring module is a mid-wave infrared imaging device, and the imaging band range of the infrared monitoring module is 3 to 5 micrometers.

[0024] Furthermore, to ensure the accuracy of subsequent three-dimensional reconstruction and quantitative comparison, the infrared monitoring units on both side walls can be installed in a strictly symmetrical manner. In specific implementation, the horizontal installation position can be calibrated by means of laser positioning, so that the symmetrical installation parameters meet the requirements of subsequent symmetrical registration / matching, and the optical axis of the infrared monitoring unit is pointed to the center area of ​​the furnace to enhance the effective coverage of the opposing combustion zone.

[0025] In one specific implementation, the infrared monitoring module is equipped with a cooling and protection kit to adapt to the furnace environment. The infrared monitoring module is synchronously triggered to collect data by a central processing unit or a background server to ensure the temporal consistency of the infrared image sequence.

[0026] In this embodiment, to adapt to the high temperature and high dust environment of the furnace, the infrared monitoring module can be equipped with a cooling protection and lens cleaning device, such as using compressed air for continuous cleaning and cooling to ensure lens cleanliness and equipment safety.

[0027] Furthermore, the infrared monitoring module can acquire raw infrared grayscale images at a fixed frame rate and transmit them in real time. For example, it can acquire a sequence of raw infrared grayscale images of 1024×768 pixels at 25 frames per second and transmit them to the backend server in real time through a gigabit fiber optic network to meet the timeliness requirements of real-time monitoring and algorithm calculation.

[0028] In another specific embodiment, the image processing and analysis module includes a grayscale-radiation intensity calibration unit. The grayscale-radiation intensity calibration unit establishes a mapping relationship between the pixel grayscale values ​​of the infrared image and the target radiation intensity through blackbody calibration, so as to convert the grayscale image of the original infrared image into a radiation intensity distribution map. Image preprocessing includes at least non-uniformity correction, bad pixel repair, and background noise suppression. After image preprocessing is completed, an adaptive threshold segmentation algorithm is used for the radiation intensity distribution map. Based on the preset radiation intensity interval threshold, each pixel in the radiation intensity distribution map is initially classified to obtain the preliminary segmentation results of the background area, the unburned coal powder jet area, and the high-temperature combustion area in the radiation intensity distribution map. The preliminary segmentation results are used as prior information and / or training input for subsequent flame semantic segmentation.

[0029] In this embodiment, the image processing and analysis module can use a standard blackbody calibration source for online or periodic calibration to establish a mapping relationship between image grayscale values ​​and absolute radiation intensity, thereby converting the original infrared grayscale image into a standardized radiation intensity distribution map, providing a foundation for subsequent classification and recognition based on physical characteristics.

[0030] Specifically, after obtaining the radiation intensity distribution map, image preprocessing includes at least non-uniformity correction, bad pixel repair, and background noise suppression to reduce the interference of uneven device response, bad pixels, and strong noise background on the segmentation results.

[0031] Furthermore, after image preprocessing, an adaptive threshold segmentation algorithm can be applied to the radiation intensity distribution map. Different radiation intensity range thresholds are set according to the physical process of pulverized coal combustion to initially classify each pixel in the radiation intensity distribution map, obtaining preliminary segmentation results for the background area, unburned pulverized coal jet area, and high-temperature combustion area. The background area corresponds to regions with extremely low radiation intensity (such as furnace walls or low-temperature flue gas), the unburned pulverized coal jet area corresponds to darker areas with moderate radiation intensity (such as volatile matter release and the initial combustion stage, where temperatures are relatively low), and the high-temperature combustion area corresponds to bright areas with high radiation intensity (such as intense gas-phase combustion and char particle combustion). These preliminary segmentation results can serve as prior information and / or training input for subsequent deep learning semantic segmentation to improve the robustness of segmentation under complex boundaries and interconnected conditions.

[0032] In another specific embodiment, pixel-level segmentation is achieved using a flame semantic segmentation model based on a convolutional neural network. The flame semantic segmentation model takes a pre-processed single-frame infrared image or a time series composed of multiple pre-processed infrared images as input, performs semantic classification on the pixels in the input infrared image, and outputs the category label and / or category probability of each pixel in the infrared image belonging to a preset category set. The preset category set includes at least the background area, the first side flame main body area, the second side flame main body area, and the flame front interaction area.

[0033] Furthermore, since flame morphology is complex and fluctuates violently, simple threshold segmentation is difficult to accurately distinguish the edge of the flame front. This invention introduces a flame semantic segmentation model based on convolutional neural networks to perform pixel-level fine segmentation. The model outputs a high-precision pixel-level classification map, which can accurately identify the main flames on both sides and the frontal areas formed by mutual penetration and mixing.

[0034] Specifically, in one embodiment, a U-Net semantic segmentation model with an encoder-decoder structure can be used: taking a single frame or multiple consecutive frames of time-series images acquired synchronously as input, and outputting a probability map or category label map of each pixel belonging to a preset category set; the preset category set can at least include "background", "first side flame body", "second side flame body", and "flame front interaction area". In the scenario of a counter-flow boiler with front and rear walls, the first side flame body and the second side flame body can be further mapped to "front wall flame" and "rear wall flame", and can be expanded to include categories such as "pulverized coal jet"; the model can be trained by labeled samples of the same type of counter-flow boiler, for example, using more than 5,000 pixel-level labeled images for training.

[0035] In another specific embodiment, extracting the flame front contour based on the segmentation results includes edge detection of the flame front interaction zone boundary based on pixel-level segmentation results to obtain the flame front contour line; calculating the two-dimensional position parameters of the flame front in the furnace longitudinal section coordinate system includes symmetrical matching of the corresponding flame front contour lines in the images acquired from the side walls of the furnace, and calculating the two-dimensional coordinates of each point on the flame front contour line in the furnace longitudinal section coordinate system accordingly. Symmetrical matching includes transforming the images of both sides to a unified coordinate system based on the symmetrical installation parameters of the infrared monitoring modules on both sides and then performing registration matching; to achieve two-dimensional position quantization of the flame front in the furnace longitudinal section coordinate system, geometric calibration and coordinate conversion processing can be performed on the infrared monitoring units on both sides. The geometric calibration can use conventional calibration methods in the art to obtain the imaging parameters and installation parameters of the infrared monitoring units, for example, using calibration targets with known geometric dimensions, furnace structural feature points, or preset reference marks to establish the correspondence between pixel coordinates and furnace longitudinal section coordinates; and the relative installation relationship of the infrared monitoring units on both sides (such as horizontal installation position, pitch angle, optical axis pointing, etc.) can be parameterized to form symmetrical installation parameters. Based on the calibration results, the infrared images acquired from both sides can be transformed to a unified coordinate system and then registered, ensuring that the same furnace space region has a consistent spatial correspondence in the images from both sides. Furthermore, based on the geometric relationship of the bilateral observations, the pixel coordinates of the frontal contour line are mapped to the furnace longitudinal section coordinate system, obtaining the two-dimensional coordinates (X, Y) of each point on the frontal contour line in the furnace longitudinal section coordinate system. The above calibration and coordinate conversion process can be implemented using conventional geometric mapping / binocular measurement / registration methods in this field, and is not limited to specific mathematical models or solution methods.

[0036] Furthermore, based on the semantic segmentation results, the boundary between the "flame front interaction zone" and the main flame areas on both sides can be extracted as the flame front contour line; and a binocular vision model is established using the observation relationship of symmetrical arrangement on both sides, and the corresponding front contour lines extracted from the images on both sides are symmetrically matched to calculate the two-dimensional coordinates of each point on the front contour line in the longitudinal section coordinate system of the furnace, thereby reconstructing the front curve.

[0037] Specifically, symmetrical matching may include: transforming the images on both sides to a unified coordinate system based on the symmetrical installation parameters of the infrared monitoring units on both sides (such as horizontal installation position, pitch angle, optical axis pointing, etc.) and then performing registration matching; after registration is completed, using binocular vision / geometric relationship to spatially locate the front contour line and obtain the two-dimensional position parameters of the front curve in the longitudinal section coordinate system of the furnace.

[0038] In another specific embodiment, the frontal dynamic parameters include at least the frontal fluctuation rate and the frontal fluctuation frequency, which are obtained by performing dense optical flow calculations on continuous frame flame front images or frontal interaction zone images; the two-dimensional position parameters include at least the average frontal center offset, which is calculated based on the center position of the boundary frontal contours of the first side flame body and the flame frontal interaction zone and the boundary frontal contours of the second side flame body and the flame frontal interaction zone, and the frontal oscillation amplitude and oscillation frequency are determined according to the change of the average frontal center offset; the early warning and combustion adjustment decision module sets multi-level early warning thresholds based on the average frontal center offset and its oscillation amplitude / oscillation frequency, and outputs combustion adjustment suggestions for correcting the aerodynamic field when the control trigger conditions are met, including symmetrical adjustment of the secondary air dampers of the burners on both sides of the opposed boiler and / or adjustment suggestions for the pulverized coal concentration distribution on both sides of the opposed boiler.

[0039] Furthermore, the system can calculate the average offset index Do of the front center: for example, the average X-coordinate of the center of the two flame front curves can be calculated separately, and the coordinate direction can be established with the geometric center of the furnace as the origin (positive for the direction towards the front wall and negative for the direction towards the rear wall). Then, the average X-coordinate of the center of the front curves on both sides (or the corresponding front and rear walls) can be averaged to obtain the average offset Do. The sign of Do reflects the direction of the front deflection, and the absolute value reflects the degree of deflection. The change of Do over time can be used to characterize the dynamic characteristics of front oscillation, where the extreme value of the change of Do can be defined as the front oscillation amplitude, and the frequency of change can be defined as the oscillation frequency. Define the frontal center average offset index Do: The calculation formula is as follows:

[0040] In the formula, Represents the coordinates of the flame front on the front wall. The coordinates of the flame front on the rear wall are represented by the formula, which represents the average position of the flame front with the front wall of the opposing boiler as the reference plane. If the position of Do is biased towards the front wall, it means that the flame front is biased towards the front wall, and vice versa. The absolute value of the offset reflects the severity of the bias. The change in the position of Do reflects the oscillation dynamic characteristics of the front. The extreme value of the change in the position of Do is defined as the amplitude of the front oscillation, and the frequency of the change is defined as the oscillation frequency. After obtaining the pixel-level segmentation results, the image processing and analysis module can perform edge detection, contour tracking, or morphological processing on the boundary between the "flame front interaction area" and the main flame areas on both sides to extract the flame front contour line. The flame front contour line can be represented by a point set composed of multiple contour points. Based on the frontal contour point sets obtained by matching the images on both sides, the average offset Do of the frontal center can be calculated. For example, statistical values ​​(such as mean or median) of the lateral coordinates of the first and second frontal contour point sets can be taken to characterize their respective center positions. The center positions on both sides can be combined (such as by averaging or differencing) to obtain Do, reflecting the direction and degree of the frontal offset relative to the geometric center of the furnace. At the same time, the frontal oscillation amplitude and oscillation frequency can be determined based on the time-varying sequence of Do. The oscillation amplitude can be characterized by the peak-to-valley difference or statistical amplitude of the Do sequence, and the oscillation frequency can be obtained by periodic statistics or frequency domain analysis of the Do sequence. When calculating the dynamic parameters of the front, dense optical flow calculation can be performed on continuous frame images containing the flame-front interaction zone to obtain the motion vector field. The optical flow vectors in the region of interest corresponding to the flame-front interaction zone can be statistically summarized (such as taking the statistics of the vector magnitude or principal direction component) to obtain the frontal fluctuation rate. Furthermore, the fluctuation frequency can be obtained by periodic statistics or frequency domain analysis on the time-varying sequence of the fluctuation rate. The above parameter calculation process is only an example. This invention does not limit the specific statistical method, frequency domain analysis method or optical flow algorithm type, as long as it can obtain the rate and frequency indicators that reflect the frontal fluctuation characteristics based on the motion information of continuous frame images.

[0041] Furthermore, frontal dynamic parameters can also be obtained through dense optical flow: for example, dense optical flow calculation is performed on consecutive frames of frontal images or consecutive frames of images containing the interaction zone between flame fronts to obtain the motion vector field of frontal pixels; and dynamic indicators such as the frontal fluctuation rate v and fluctuation frequency f are quantified based on the motion vector field, thereby realizing the joint monitoring of the static position and dynamic fluctuation of the front.

[0042] In one specific implementation, such as Figure 2 As shown, this invention provides a method for monitoring the flame front of a counter-flow boiler based on infrared optical flow. The method is implemented using the aforementioned infrared optical flow-based counter-flow boiler flame front monitoring system and includes the following steps: S1: Infrared monitoring units are symmetrically arranged at the same preset elevation on the opposite side walls of the furnace to collect infrared image sequences of the opposing combustion zones of the furnace. S2: Establish the mapping relationship between pixel grayscale values ​​and radiation intensity through blackbody calibration, and convert the infrared image sequence into a radiation intensity distribution map; S3: The radiation intensity distribution map is preprocessed by non-uniformity correction, bad spot repair and background noise suppression, and adaptive threshold segmentation is used to obtain preliminary classification results of background area, unburned coal powder jet area and high temperature combustion area; S4: Perform pixel-level flame semantic segmentation on the preprocessed radiation intensity distribution map to obtain a segmentation result that includes at least the background area, the first side flame body, the second side flame body, and the flame front interaction area. S5: Extract the flame front contour line based on the segmentation results, and perform symmetrical matching on the corresponding front contour lines in the two side images to calculate the two-dimensional coordinates of the front contour line in the longitudinal section coordinate system of the furnace, thereby obtaining the average offset of the front center. S6: Perform dense optical flow calculation on continuous frame images containing the flame front interaction zone to obtain the motion vector field of the flame front interaction zone, and calculate the fluctuation rate and fluctuation frequency of the flame front based on the motion vector field; and determine the oscillation amplitude and oscillation frequency of the flame front according to the curve of the average offset of the front center over time. S7: Quantitatively assess the stability of the flame front based on the average offset of the front center, oscillation amplitude, and oscillation frequency, and output the corresponding warning level according to the preset grading criteria; when the average offset of the front center and / or oscillation amplitude and oscillation frequency meet the preset control trigger threshold conditions, output combustion adjustment suggestions, which include symmetrical adjustment of the secondary air dampers on both sides of the counter-flow boiler and / or adjustment of the pulverized coal concentration distribution on both sides of the counter-flow boiler.

[0043] Specifically, the early warning and combustion adjustment decision module can establish a frontal stability assessment model based on the frontal dynamic characteristic data (such as position standard deviation, oscillation amplitude, oscillation frequency, etc.) obtained from long-term series monitoring, and set multiple early warning thresholds, such as: normal level, early warning level and control level; among which the control level can correspond to situations such as the average position of the front deviating significantly from the centerline or experiencing large-amplitude, low-frequency periodic oscillations, and will trigger active control when judged to be at the control level.

[0044] Furthermore, when the system determines that the control trigger conditions have been met, in addition to issuing an alarm, it can also provide combustion adjustment suggestions based on the front's deviation and oscillation characteristics. For example, when the front continuously deviates to one side, it can prompt symmetrical adjustment of the secondary air dampers on both burners and check the pulverized coal concentration on both sides, reducing the pulverized coal concentration and secondary air volume on the side opposite to the deviation direction. When a specific frequency oscillation is detected, it can prompt fine-tuning of the secondary air volume or frequency to interfere with the oscillation formation. In one example, when the front center is detected to continuously deviate by more than 0.1 meters and is accompanied by oscillations at a characteristic frequency of 0.5–2 Hz, the system can determine that there is a risk of combustion instability and pop up a warning, while providing specific suggestions for adjusting the secondary air damper opening. After the operators adjust according to the suggestions, the front center can return to the normal range, the oscillation amplitude will decrease, and the warning will be automatically lifted.

[0045] In one specific implementation, this monitoring method was applied to a 600MW supercritical pulverized coal-fired boiler with opposed combustion. This boiler employs a front-and-rear wall opposed combustion configuration, with three layers of swirl burners arranged on each wall. During implementation, two sets of mid-wave infrared thermal imagers were symmetrically installed on the left and right walls at the elevation of the middle layer of burners in the boiler. Their operating wavelength was 3-4.8 micrometers to adapt to the high-temperature, high-dust environment inside the furnace. The horizontal installation position of each thermal imager was laser-calibrated to ensure that the corresponding equipment on the front and rear walls were on the same axis, with their optical axes pointing horizontally towards the center of the furnace. Continuous purging and cooling with 0.6MPa compressed air was provided to ensure lens cleanliness and equipment safety. All thermal imagers were hardwired to synchronize with a backend server located in the central control room, acquiring raw infrared grayscale images of 1024×768 pixels at a rate of 25 frames per second, and transmitting them in real time via a gigabit fiber optic network.

[0046] In this embodiment, such as Figure 3As shown, the software system on the backend server first preprocesses the original image sequence. Online calibration is performed periodically using a standard blackbody calibration source to establish a mapping relationship between image grayscale values ​​and absolute radiation intensity, thus converting the image into a standardized radiation intensity distribution map. Non-uniformity correction, bad pixel repair, and background noise suppression are then performed on the original infrared image. An adaptive threshold segmentation algorithm is developed on the radiation intensity distribution map. The system calls a pre-trained U-Net deep learning model for flame semantic segmentation. This model is trained using over 5000 labeled images (pixel-level labels: "background," "front wall flame," "rear wall flame," "front reaction zone," and "pulverized coal jet") of the same type of counter-current boiler. The model uses continuously acquired images of the front and rear walls simultaneously as input, extracts spatiotemporal features through an encoder-decoder structure, and finally outputs a probability map of each pixel belonging to the aforementioned five categories. The complex frontal region formed by the mutual impact and entanglement of the two flames on the front and rear walls of the counter-flow boiler is highlighted. Based on the segmentation results, the system extracts the boundary between the "frontal reaction zone" and the flames on both sides as the frontal contour line. Using a binocular vision model symmetrically arranged on the front and rear walls, the actual spatial coordinates (X,Y) of each point on the frontal contour in the longitudinal section plane coordinate system of the furnace are calculated, thereby reconstructing the frontal curve.

[0047] Specifically, the system employs an algorithm to calculate in real-time the average X-coordinate of the center of the frontal curves of the two flame jets, with the geometric center of the furnace as the origin, positive for the direction towards the front wall and negative for the direction towards the rear wall. The average X-coordinates of the centers of the two frontal curves are then averaged again to obtain the average offset index. In a power plant, the obtained flame image is as follows: Figure 2As shown. After processing, the detected value was -0.11 meters, which quantifies that the flame front continuously deviates 0.11 meters behind the furnace. Simultaneously, the system applies a dense optical flow algorithm to the segmented images of consecutive frames to calculate the motion vector field of the front region, analyzing that the current front exhibits a periodic oscillation with an amplitude of approximately 0.05 meters at a frequency of 0.8 Hz. The monitoring system compares the above quantification results (position offset, oscillation frequency, and amplitude) with the built-in expert knowledge base. When the system detects a continuous deviation of the front center exceeding 0.1 meters accompanied by characteristic frequency oscillations of 0.5-2 Hz, it determines that there is a risk of combustion instability. The operator station immediately displayed an "early warning" message and provided specific suggestions: "A flame front has been detected to be continuously deviating forward by 0.15 meters, accompanied by a low-frequency oscillation of 0.8 Hz. It is recommended to symmetrically reduce the opening of the secondary air damper on the front wall A layer by 3%, and symmetrically increase the opening of the secondary air damper on the rear wall A layer by 3% to correct the aerodynamic field." After the operators made the adjustments accordingly, the monitoring interface showed that the coordinates of the front center gradually returned to the normal range of -0.02 meters to +0.02 meters within a few minutes, and the oscillation amplitude also significantly decreased, automatically deactivating the warning. This process fully demonstrates the entire process of this method, from image acquisition, intelligent recognition, quantitative analysis to precise guidance for combustion adjustment, achieving transparent monitoring and closed-loop optimization of the invisible combustion process.

[0048] To aid in a better understanding of the present invention, a more comprehensive and specific embodiment is described. In this embodiment, the present invention provides an infrared optical flow-based flame front monitoring system for opposed-flow boilers, comprising: an infrared monitoring module, which includes at least a pair of infrared monitoring units symmetrically arranged on opposite sidewalls of the boiler furnace at the same preset elevation for acquiring infrared image sequences of the opposed combustion zones within the furnace; and an image processing and analysis module, communicatively connected to the infrared monitoring module, for performing radiation intensity calibration and image preprocessing on the infrared image sequences, and performing pixel-level segmentation on the preprocessed infrared images to obtain at least a background area, a first-side flame body, a second-side flame body, and a flame front. The system includes a segmentation result of the flame front interaction zone, which extracts the flame front contour based on the segmentation result and calculates the two-dimensional position parameters of the flame front in the longitudinal section coordinate system of the furnace. Optical flow calculation is performed on continuous frame infrared images containing the flame front interaction zone to obtain the flame front motion vector field and output the front dynamic parameters. A warning and combustion adjustment decision module, communicatively connected to the image processing and analysis module, is used to quantitatively evaluate the stability of the flame front based on the two-dimensional position parameters and / or the front dynamic parameters, and output corresponding warning information according to preset grading criteria. When the two-dimensional position parameters and / or the front dynamic parameters meet preset control trigger threshold conditions, combustion adjustment suggestions are output.

[0049] In this embodiment, the installation elevation of the infrared monitoring module is set to the elevation corresponding to the center line of the burner nozzle of the counter-flow boiler or within a preset range above and below it. The infrared monitoring modules on both side walls of the counter-flow boiler are symmetrically arranged in terms of horizontal position and elevation angle, so that the flame areas collected on both sides form a corresponding observation relationship in space. The infrared monitoring module is a mid-wave infrared imaging device with an imaging band range of 3–5 micrometers. The infrared monitoring module is equipped with a cooling and protection kit to adapt to the furnace environment. The infrared monitoring module is synchronously triggered for acquisition by a central processing unit or a back-end server to ensure the temporal consistency of the infrared image sequence. The image processing and analysis module includes a grayscale-radiance intensity calibration unit. The grayscale-radiance intensity calibration unit establishes a mapping relationship between pixel grayscale values ​​and target radiance intensity in the infrared image through blackbody calibration, converting the original infrared image's grayscale image into a radiance intensity distribution map. Image preprocessing includes at least non-uniformity correction, bad pixel repair, and background noise suppression. After image preprocessing, an adaptive threshold segmentation algorithm is used for the radiance intensity distribution map. Based on a preset radiance intensity interval threshold, each pixel in the radiance intensity distribution map is initially classified to obtain preliminary segmentation results for the background area, unburned coal powder jet area, and high-temperature combustion area. The preliminary segmentation results are used as prior information and / or training input for subsequent flame semantic segmentation. Pixel-level segmentation employs... A flame semantic segmentation model based on a convolutional neural network is implemented. The model takes a pre-processed single-frame infrared image or a time series composed of multiple pre-processed infrared images as input. It performs semantic classification on the pixels in the input infrared image and outputs the category label and / or category probability of each pixel belonging to a preset category set. The preset category set includes at least the background area, the first side flame main body area, the second side flame main body area, and the flame front interaction area. Based on the segmentation results, the contour of the flame front is extracted, including edge detection of the flame front interaction area boundary based on the pixel-level segmentation results to obtain the front contour line. The two-dimensional position parameters of the flame front in the longitudinal section coordinate system of the furnace are calculated. This includes symmetrical matching of the frontal contour lines in the images acquired from the sidewalls of the furnace, and calculating the two-dimensional coordinates of each point on the frontal contour line in the longitudinal section coordinate system of the furnace accordingly; the frontal dynamic parameters include at least the frontal fluctuation rate and the frontal fluctuation frequency, which are obtained by performing dense optical flow calculations on continuous frame flame frontal images or frontal interaction zone images; the two-dimensional positional parameters include at least the average offset of the frontal center, which is calculated based on the center position of the boundary frontal contour of the first side flame body and the flame frontal interaction zone and the boundary frontal contour of the second side flame body and the flame frontal interaction zone, and the frontal oscillation amplitude and oscillation frequency are determined according to the change of the average offset of the frontal center.The early warning and combustion adjustment decision module sets multi-level early warning thresholds based on the average offset of the front center and its oscillation amplitude / frequency. When the control trigger conditions are met, it outputs combustion adjustment suggestions to correct the aerodynamic field. These suggestions include symmetrical adjustments to the secondary air dampers of the burners on both sides of the opposed boiler and / or adjustments to the pulverized coal concentration distribution on both sides of the opposed boiler.

[0050] Specifically, another aspect of the present invention provides a method for monitoring the flame front of a counter-current boiler based on infrared optical flow. The method is implemented using the aforementioned infrared optical flow-based counter-current boiler flame front monitoring system and includes the following steps: S1: Infrared monitoring units are symmetrically arranged at the same preset elevation on the opposite side walls of the furnace to collect infrared image sequences of the opposing combustion zones of the furnace. S2: Establish the mapping relationship between pixel grayscale values ​​and radiation intensity through blackbody calibration, and convert the infrared image sequence into a radiation intensity distribution map; S3: The radiation intensity distribution map is preprocessed by non-uniformity correction, bad spot repair and background noise suppression, and adaptive threshold segmentation is used to obtain preliminary classification results of background area, unburned coal powder jet area and high temperature combustion area; S4: Perform pixel-level flame semantic segmentation on the preprocessed radiation intensity distribution map to obtain a segmentation result that includes at least the background area, the first side flame body, the second side flame body, and the flame front interaction area. S5: Extract the flame front contour line based on the segmentation results, and perform symmetrical matching on the corresponding front contour lines in the two side images to calculate the two-dimensional coordinates of the front contour line in the longitudinal section coordinate system of the furnace, thereby obtaining the average offset of the front center. S6: Perform dense optical flow calculation on continuous frame images containing the flame front interaction zone to obtain the motion vector field of the flame front interaction zone, and calculate the fluctuation rate and fluctuation frequency of the flame front based on the motion vector field; and determine the oscillation amplitude and oscillation frequency of the flame front according to the curve of the average offset of the front center over time. S7: Quantitatively assess the stability of the flame front based on the average offset of the front center, oscillation amplitude, and oscillation frequency, and output the corresponding warning level according to the preset grading criteria; when the average offset of the front center and / or oscillation amplitude and oscillation frequency meet the preset control trigger threshold conditions, output combustion adjustment suggestions, which include symmetrical adjustment of the secondary air dampers on both sides of the counter-flow boiler and / or adjustment of the pulverized coal concentration distribution on both sides of the counter-flow boiler.

[0051] In summary, the embodiments disclosed herein have at least the following technical effects: By acquiring information about the opposing combustion zone through symmetrically arranged infrared imaging on both sides, it is possible to extract the flame front profile and reconstruct its two-dimensional position in the longitudinal section coordinate system of the furnace, thereby achieving quantitative monitoring of the "offset and shape" of the front, rather than relying solely on inference from indirect parameters. Dense optical flow calculations are performed on the frontal interaction zone image to obtain the frontal motion vector field, and dynamic indicators such as fluctuation rate and fluctuation frequency are further extracted. The oscillation amplitude and oscillation frequency are determined by combining the average offset of the frontal center with the change over time, thereby achieving a joint representation of "position + fluctuation". Blackbody calibration is used to convert grayscale images into radiation intensity distribution maps, and preprocessing such as non-uniformity correction, bad pixel repair, and noise suppression is performed. Then, combined with adaptive threshold initial segmentation and convolutional neural network semantic segmentation, the flame body and front interaction zone can be more stably distinguished in furnace environments with high temperature, strong radiation, and complex dust background, reducing the risk of missegmentation. Based on offset and oscillation characteristics, a stability assessment and multi-level early warning threshold are established. When the triggering conditions are met, adjustment suggestions for secondary air dampers, pulverized coal concentration distribution, etc. are output. This can be used to identify combustion instability signs in advance and guide operation intervention, thereby reducing the safety and economic risks caused by combustion deviation and oscillation. After adjustment, the offset and oscillation parameter regression can be continuously monitored and observed, supporting early warning cancellation and effect verification, facilitating long-term operation optimization and early fault prevention.

[0052] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A flame front monitoring system for counter-current boilers based on infrared optical flow, characterized in that, include: An infrared monitoring module, comprising at least a pair of infrared monitoring units symmetrically arranged on opposite side walls of the furnace of the opposed boiler at the same preset elevation, for acquiring infrared image sequences of the opposed combustion zone of the furnace; An image processing and analysis module, communicatively connected to the infrared monitoring module, is used to perform radiation intensity calibration and image preprocessing on the infrared image sequence. It then performs pixel-level segmentation on the preprocessed infrared images to obtain segmentation results including at least a background area, a first-side flame body, a second-side flame body, and a flame front interaction area. Based on the segmentation results, it extracts the contour of the flame front and calculates the two-dimensional position parameters of the flame front in the longitudinal section coordinate system of the furnace. Finally, it performs optical flow calculation on consecutive frames of infrared images containing the flame front interaction area to obtain the flame front motion vector field and outputs the front dynamic parameters. The early warning and combustion adjustment decision module is communicatively connected to the image processing and analysis module to quantitatively evaluate the stability of the flame front based on the two-dimensional position parameters and / or the front dynamic parameters, and output corresponding early warning information according to preset grading criteria.

2. The infrared optical flow-based flame front monitoring system for opposed-flow boilers according to claim 1, characterized in that, The installation elevation of the infrared monitoring module is set to be at the same level as the center line of the burner nozzle of the counter-flow boiler or within a preset range above and below it. The infrared monitoring modules on the side walls of the counter-flow boiler are symmetrically arranged in terms of horizontal position and pitch angle, so that the flame areas collected by the infrared monitoring modules form a relative observation relationship in space.

3. The infrared optical flow-based flame front monitoring system for counter-current boilers according to claim 1, characterized in that, The infrared monitoring module is a mid-wave infrared imaging device, and the imaging band range of the infrared monitoring module is 3 to 5 micrometers.

4. The infrared optical flow-based flame front monitoring system for counter-current boilers according to claim 1, characterized in that, The infrared monitoring module is equipped with a cooling and protection kit to adapt to the furnace environment. The infrared monitoring module is synchronously triggered to collect data by a central processing unit or a background server to ensure the temporal consistency of the infrared image sequence.

5. The infrared optical flow-based flame front monitoring system for counter-current boilers according to claim 1, characterized in that, The image processing and analysis module includes a grayscale-radiance intensity calibration unit. The grayscale-radiance intensity calibration unit establishes a mapping relationship between the pixel grayscale values ​​of the infrared image and the target radiation intensity through blackbody calibration, so as to convert the original grayscale image of the infrared image into a radiation intensity distribution map.

6. The infrared optical flow-based flame front monitoring system for counter-current boilers according to claim 5, characterized in that, The image preprocessing includes at least non-uniformity correction, bad pixel repair, and background noise suppression. After the image preprocessing is completed, an adaptive threshold segmentation algorithm is used on the radiation intensity distribution map to perform preliminary classification of each pixel in the radiation intensity distribution map according to a preset radiation intensity interval threshold, so as to obtain preliminary segmentation results of the background area, unburned coal powder jet area and high-temperature combustion area in the radiation intensity distribution map. The preliminary segmentation results are used as prior information and / or training input for subsequent flame semantic segmentation.

7. The infrared optical flow-based flame front monitoring system for opposed-flow boilers according to claim 1, characterized in that, The pixel-level segmentation is achieved using a flame semantic segmentation model based on a convolutional neural network. The flame semantic segmentation model takes the infrared image after preprocessing a single frame or a time series composed of multiple preprocessed infrared images as input, performs semantic classification on the pixels in the input infrared image, and outputs the category label and / or category probability of each pixel in the infrared image belonging to a preset category set. The preset category set includes at least the background area, the first side flame main body area, the second side flame main body area, and the flame front interaction area.

8. The infrared optical flow-based flame front monitoring system for counter-current boilers according to claim 1, characterized in that, Extracting the contour of the flame front based on the segmentation result includes performing edge detection on the boundary of the flame front interaction area based on the pixel-level segmentation result to obtain the flame front contour line; calculating the two-dimensional position parameters of the flame front in the longitudinal section coordinate system of the furnace includes performing symmetrical matching on the corresponding flame front contour lines in the images acquired from the side walls of the furnace, and calculating the two-dimensional coordinates of each point on the flame front contour line in the longitudinal section coordinate system of the furnace accordingly.

9. The infrared optical flow-based flame front monitoring system for counter-current boilers according to any one of claims 1 to 8, characterized in that: The frontal dynamic parameters include at least the frontal fluctuation rate and the frontal fluctuation frequency, which are obtained by performing dense optical flow calculations on continuous frame flame frontal images or frontal interaction zone images. The two-dimensional position parameters include at least the average frontal center offset, which is calculated based on the center position of the boundary frontal contours of the first-side flame body and the flame frontal interaction zone and the boundary frontal contours of the second-side flame body and the flame frontal interaction zone. The frontal oscillation amplitude and oscillation frequency are determined according to the change of the average frontal center offset. The early warning and combustion adjustment decision module sets multi-level early warning thresholds based on the average frontal center offset and its oscillation amplitude / oscillation frequency, and outputs combustion adjustment suggestions for correcting the aerodynamic field when the control trigger conditions are met. The combustion adjustment suggestions include symmetrical adjustments to the secondary air dampers of the burners on both sides of the opposed boiler and / or adjustments to the pulverized coal concentration distribution on both sides of the opposed boiler.

10. A method for monitoring the flame front of a counter-current boiler based on infrared optical flow, characterized in that, The method is implemented using the infrared optical flow-based counter-current boiler flame front monitoring system according to any one of claims 1 to 9, and includes the following steps: S1: The infrared monitoring units are symmetrically arranged at the same preset elevation on the opposite side walls of the furnace to collect infrared image sequences of the opposing combustion zones of the furnace. S2: Establish the mapping relationship between pixel grayscale values ​​and radiation intensity through blackbody calibration, and convert the infrared image sequence into a radiation intensity distribution map; S3: Perform non-uniformity correction, bad pixel repair and background noise suppression on the radiation intensity distribution map, and obtain preliminary classification results of the background area, unburned pulverized coal jet area and high-temperature combustion area; S4: Perform pixel-level flame semantic segmentation on the preprocessed radiation intensity distribution map to obtain a segmentation result that includes at least the background area, the first side flame body, the second side flame body, and the flame front interaction area; S5: Extract the flame front contour line based on the segmentation result, and perform symmetrical matching on the corresponding front contour lines in the two side images to calculate the two-dimensional coordinates of the front contour line in the longitudinal section coordinate system of the furnace, thereby obtaining the average offset of the front center. S6: Perform dense optical flow calculation on consecutive frame images containing the flame front interaction area to obtain the motion vector field of the flame front interaction area, and calculate the fluctuation rate and fluctuation frequency of the flame front based on the motion vector field; and determine the oscillation amplitude and oscillation frequency of the flame front according to the curve of the average offset of the front center over time. S7: Quantitatively evaluate the stability of the flame front based on the average offset of the front center, the oscillation amplitude, and the oscillation frequency, and output the corresponding warning level according to the preset grading criteria; when the average offset of the front center and / or the oscillation amplitude and the oscillation frequency meet the preset control trigger threshold conditions, output the combustion adjustment suggestion for the counter-current boiler.

Images