Boiler combustion stability judgment and optimization adjustment method based on image recognition

Abstract

The embodiment of the invention relates to the field of boiler combustion control, and discloses a boiler combustion stability judgment and optimization adjustment method based on image recognition, which comprises the following steps: acquiring continuous multi-frame images of flame in a hearth; a flame area in each frame of image is divided into a root area, a center area, an edge area and a top area, multi-dimensional features of the flame area in each frame of image are extracted, deep features of each frame of image are extracted and fused with the multi-dimensional features to obtain a fused feature vector, and the multi-dimensional features include morphological features, color features, intensity features and motion features; obtaining stability scores of different dimensions based on the fusion feature vector of each frame of image, calculating a comprehensive stability score according to the stability scores of different dimensions, and judging a combustion state according to the comprehensive stability score; and when the combustion state is unstable, determining an instability reason and issuing a corresponding combustion parameter optimization adjustment strategy. According to the method disclosed by the invention, the problem of realizing closed-loop control of real-time monitoring and dynamic adjustment at low cost is solved.

Description

Technical Field

[0001] This invention relates to the field of boiler combustion control, and in particular to a method for judging and optimizing boiler combustion stability based on image recognition. Background Technology

[0002] Currently, the installed capacity of renewable energy power generation systems such as wind and solar power has increased significantly. The large-scale grid connection of renewable energy power generation has brought significant volatility and intermittency to the power system. This is because wind and solar power generation are significantly affected by weather conditions, resulting in large random fluctuations in power output. For example, solar power output may drop sharply to 10%-20% of its rated capacity during cloudy or rainy weather, while wind power output can fluctuate by more than 50% of its rated capacity when wind speed changes. Without external intervention, such power generation fluctuations can impact the grid during grid connection. Therefore, to ensure the safe and stable operation of the power system, sufficient flexible regulation resources must be allocated to balance the volatility of renewable energy power generation. In the existing power structure, thermal power units remain the most important flexible regulation resource, undertaking the important tasks of grid peak shaving, frequency regulation, voltage regulation, and reserve.

[0003] Traditional thermal power unit designs primarily consider stable operation under rated load or high load conditions. However, with the increasing penetration of renewable energy, thermal power units are required to have deeper peak-shaving capabilities and faster load-change speeds. In some regions, coal-fired units are required to reduce their minimum technical output to below 30% of their rated capacity to absorb surplus wind power. When the unit load drops below 50%, the boiler fuel and air volume decrease significantly, the furnace temperature drops, and the flame coverage decreases, easily leading to combustion instability phenomena, including flame blowout and violent flame flickering. These phenomena are often accompanied by decreased combustion efficiency and increased pollutant emissions. Therefore, developing diagnostic and optimization adjustment schemes for boiler furnace flame stability is essential, and this technology is also a crucial component of flexible peak-shaving technology for boilers.

[0004] In related technologies, flame stability monitoring schemes have high hardware requirements, high deployment and / or maintenance costs in industrial settings, and their reliability in industrial settings still needs improvement. How to achieve low-cost, real-time monitoring and dynamic adjustment of closed-loop control is a problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide at least one method for judging and optimizing boiler combustion stability based on image recognition, which can at least solve the problem of achieving closed-loop control with real-time monitoring and dynamic adjustment at low cost.

[0006] To address the aforementioned technical problems, at least one embodiment of this application provides a method for judging and optimizing boiler combustion stability based on image recognition, comprising: acquiring multiple consecutive frames of flame images inside the furnace of a target boiler; dividing the flame region in each frame image into a root region, a center region, an edge region, and a top region; extracting multidimensional features of the flame region in each frame image; extracting deep features of each frame image; and fusing these features with the multidimensional features to obtain a fused feature vector, wherein the multidimensional features include morphological features, color features, intensity features, and motion features; obtaining different dimensional stability scores of the flame region based on the fused feature vector of each frame image; calculating a comprehensive stability score based on the different dimensional stability scores; judging the combustion state based on the comprehensive stability score; and when the combustion state is unstable, determining the cause of instability and issuing corresponding combustion parameter optimization and adjustment strategies.

[0007] At least one embodiment of this application also provides a boiler combustion stability judgment and optimization adjustment device based on image recognition, comprising: an image acquisition module for acquiring multiple consecutive frames of images of the flame inside the furnace of a target boiler; an image processing module for dividing the flame region in each frame image into a root region, a center region, an edge region, and a top region, extracting multidimensional features of the flame region in each frame image, extracting deep features of each frame image, and fusing them with the multidimensional features to obtain a fused feature vector, wherein the multidimensional features include morphological features, color features, intensity features, and motion features; a stability judgment module for obtaining different dimension stability scores of the flame region based on the fused feature vector of each frame image, calculating a comprehensive stability score based on the different dimension stability scores, and judging the combustion state based on the comprehensive stability score; and an optimization adjustment module for determining the cause of instability and issuing corresponding combustion parameter optimization adjustment strategies when the combustion state is unstable.

[0008] At least one embodiment of this application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described image recognition-based boiler combustion stability judgment and optimization adjustment method.

[0009] At least one embodiment of this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for judging and optimizing boiler combustion stability based on image recognition.

[0010] The image recognition-based boiler combustion stability assessment and optimization method provided in this application achieves closed-loop control from stability assessment to optimization adjustment, and has practical engineering application value. On the one hand, it assesses combustion stability using continuous multi-frame images of the flame, eliminating the need for significant industrial site deployment costs. On the other hand, during the recognition and assessment process based on continuous multi-frame images, it comprehensively considers the morphological, color, intensity, and motion characteristics of different regions to conduct real-time and accurate stability evaluation. This enables timely detection of combustion instability, prompt determination of the causes, and adjustment of combustion parameters, thus achieving closed-loop control of real-time monitoring and dynamic adjustment in boiler combustion control and improving boiler combustion stability.

[0011] In some optional embodiments, based on the geometric features of the flame, the flame region in each frame image is divided into a root region, a center region, an edge region, and a top region. The root region includes the flame region extending from the burner outlet into the furnace within a first distance range. The center region includes the flame region with the highest brightness and the largest area. The edge region includes the region extending outward from the outer contour of the flame within a preset pixel range and intersecting with the background. The top region is the region extending within a second distance range below the highest position of the flame.

[0012] In some optional embodiments, a pre-trained convolutional neural network model based on spatial attention is used to extract deep features of each frame of the image and fuse them with the multi-dimensional features to obtain a fused feature vector. The convolutional neural network model based on spatial attention includes: a feature extraction layer for extracting deep features of each frame of the image to obtain a deep feature map; a spatial attention module connected to the feature extraction layer for generating a spatial attention feature map based on the deep feature map; a feature fusion layer connected to the spatial attention module for concatenating and fusing the multi-dimensional features with the spatial attention feature map to obtain a fused feature vector of each frame of the image; and an output layer connected to the feature fusion layer for outputting the fused feature vector of each frame of the image. The convolutional neural network model based on spatial attention is trained based on the spatial attention module assigning relatively high weights to the root region.

[0013] In some optional embodiments, the morphological features include the area of ​​the central region and / or the centroid position of the root region; the color features include the RGB three-channel mean and / or color temperature; the intensity features include grayscale value distribution; the motion features include optical flow motion vectors; based on the fused feature vectors of each frame image, different dimensional stability scores of the flame region are obtained, including: determining the morphological stability score based on the rate of change of morphological features in consecutive frames of images; determining the color stability score based on the fluctuation amplitude of the RGB three-channel mean and / or color temperature in consecutive frames of images; determining the intensity stability score based on the temporal stability of brightness in consecutive frames of images calculated according to the grayscale value distribution; and determining the motion stability score based on the uniformity of the distribution of optical flow motion vectors.

[0014] In some optional embodiments, a comprehensive stability score is calculated based on stability scores of different dimensions, and the combustion state is determined based on the comprehensive stability score, including: calculating a comprehensive stability score using stability scores of different dimensions and corresponding weighting coefficients; and determining the current combustion state based on a preset correspondence between a comprehensive stability score range and the combustion state.

[0015] In some optional embodiments, when the combustion state is unstable, the cause of instability is determined and a corresponding combustion parameter optimization and adjustment strategy is issued, including: using an expert knowledge base to determine the cause of instability and the corresponding combustion parameter optimization and adjustment strategy based on stability scores in different dimensions, and issuing the combustion parameter optimization and adjustment strategy; the expert knowledge base includes: diagnostic rules, including weight coefficients corresponding to stability scores in each dimension, and the mapping relationship between stability scores in different dimensions and the cause of instability; adjustment rules, including the mapping relationship between the cause of instability and the combustion parameter optimization and adjustment strategy; and constraint rules, including the optimization adjustment range and safety boundary of combustion parameters.

[0016] In some optional embodiments, the image recognition-based method for judging and optimizing boiler combustion stability further includes: after executing the combustion parameter optimization and adjustment strategy, continuing to acquire multiple consecutive frames of images of the flame inside the furnace of the target boiler to obtain a new comprehensive stability score, and judging the combustion state based on the new comprehensive stability score; if the combustion state is stable, recording that the executed combustion parameter optimization and adjustment strategy is effective, and adjusting the weight coefficients corresponding to the stability scores of different dimensions according to the stability scores of different dimensions, and updating them to the expert knowledge base. Attached Figure Description

[0017] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.

[0018] Figure 1This is a schematic flowchart of a boiler combustion stability judgment and optimization adjustment method based on image recognition provided in an embodiment of this application; Figure 2 This is a schematic diagram of flame area division provided in an embodiment of this application; Figure 3 This is a flowchart illustrating an application example provided in an embodiment of this application; Figure 4 The relevant data calculated based on the multidimensional features of the flame provided in the embodiments of this application are (a) the standard deviation of the central region area, (b) the temporal variation of the RGB three colors of the flame, (c) the mean value of the brightness gradient, and (d) the motion vector field between two adjacent frames. Figure 5 This is a schematic diagram of an image recognition-based boiler combustion stability judgment and optimization adjustment device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been provided in the various embodiments of this application to help readers better understand this application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and modifications based on the following embodiments. The division of the various embodiments below is for the convenience of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.

[0020] To facilitate understanding of the embodiments of this application, relevant content regarding flame stability monitoring technology will be introduced first.

[0021] In related technologies, flame stability is determined based on variational autoencoders, which integrate flame geometric parameters, physical parameters and flicker frequency. At the same time, an intelligent classification model is constructed, and a flame stability index is established by using VAE dimensionality reduction and Gaussian mixture clustering to monitor the stable and unstable states of the flame. Finally, real-time monitoring and adjustment are combined with the YOLO algorithm to realize flame area labeling and anomaly warning. However, VAE training and real-time image processing have demanding hardware requirements and high deployment costs in industrial settings.

[0022] Among related technologies, flame stability monitoring based on self-luminous images involves capturing flame images of OH* and CH* self-luminous groups using ultraviolet filters and identifying the critical point of coal combustion using normalized self-luminous intensity. However, the critical point threshold for coal is based on statistical experience, and its applicability to different types of coal is questionable. In addition, multi-level optical systems have high maintenance costs, and their reliability in industrial settings needs to be verified.

[0023] Therefore, there is an urgent need in this field to solve the problem of achieving closed-loop control with real-time monitoring and dynamic adjustment at low cost.

[0024] To address the aforementioned technical problem of achieving real-time monitoring and dynamic adjustment of closed-loop control at low cost, this invention proposes a boiler combustion stability judgment and optimization adjustment method based on image recognition. The implementation details of the boiler combustion stability judgment and optimization adjustment method based on image recognition in this embodiment are described below. The following content is only for ease of understanding and is not necessary for implementing this solution.

[0025] Example 1: The image recognition-based boiler combustion stability assessment and optimization method of this embodiment can be applied to electronic devices with communication, computing, and data storage capabilities. Its specific process can be as follows: Figure 1 As shown, it includes: Step 101: Obtain multiple consecutive frames of images of the flame inside the furnace of the target boiler.

[0026] Specifically, a high-speed visible light camera is used to acquire multiple consecutive RGB color images of the flame inside the furnace of the target boiler. These multiple consecutive RGB color images constitute an RGB color image sequence. Only a common high-speed visible light camera is used, eliminating the need for an ultraviolet camera, image intensifier, and multi-head fiber optic bundle. This results in low hardware costs and ease of implementation. Furthermore, the high-speed visible light camera is simple to maintain, extending the cleaning cycle of the quartz glass window to 1 to 2 months, and eliminating the need to replace expensive image intensifiers.

[0027] Step 102: Divide the flame region in each frame image into a root region, a center region, an edge region, and a top region. Extract multidimensional features of the flame region in each frame image, extract deep features of each frame image, and fuse them with the multidimensional features to obtain a fused feature vector. Specifically, the multidimensional features include morphological features, color features, intensity features, and motion features.

[0028] In the specific implementation, preprocessing is performed before dividing the flame area into regions, including Gaussian filtering to denoise multiple consecutive frames of images and size normalization.

[0029] Step 103: Based on the fusion feature vectors of each frame image, obtain the stability scores of different dimensions of the flame region, calculate the comprehensive stability score based on the stability scores of different dimensions, and determine the combustion state based on the comprehensive stability score.

[0030] Step 104: When the combustion state is unstable, determine the cause of the instability and issue the corresponding combustion parameter optimization and adjustment strategy.

[0031] In this embodiment, by acquiring multiple consecutive frames of flame images within the furnace of the target boiler, the flame region in each frame is divided into root, center, edge, and top regions. The morphological, color, intensity, and motion features of the flame region in each frame are extracted and then fused with the extracted deep features of each frame to obtain a fused feature vector. Based on the fused feature vectors of each frame, different dimensional stability scores of the flame region are obtained. A comprehensive stability score is calculated based on these scores to determine the combustion state. When the combustion state is unstable, the cause of instability is identified, and corresponding combustion parameter optimization and adjustment strategies are issued. This achieves closed-loop control from stability assessment to optimization and adjustment, demonstrating practical engineering application value. On one hand, the determination of combustion state stability is achieved through multiple consecutive frames of flame images, eliminating the need for significant industrial field deployment costs. On the other hand, the identification and judgment process based on multiple consecutive frames comprehensively considers the morphological, color, intensity, and motion features of different regions for real-time and accurate stability evaluation. This allows for timely detection of combustion instability, timely determination of the cause, and adjustment of combustion parameters, enabling closed-loop control of boiler combustion control through real-time monitoring and dynamic adjustment, thereby improving boiler combustion stability.

[0032] In some embodiments, based on the geometric features of the flame, the flame region in each frame image is divided into a root region, a central region, an edge region, and a top region. The root region includes the flame region extending from the burner outlet into the furnace within a first distance range; the central region includes the flame region with the highest brightness and the largest area; the edge region includes the region extending outward from the outer contour of the flame within a preset pixel range and intersecting with the background; and the top region is the region extending within a second distance range below the highest position of the flame.

[0033] For each frame of the image sequence, the flame is divided into four key regions based on its geometric shape: Root region: The flame area extending 200 pixels from the center of the burner outlet nozzle into the furnace in the image. The root region can be used to monitor whether the ignition point has deviated, such as... Figure 2 The area within the blue box; the central area: the flame region with the highest brightness and largest connected area in the image, defined as the main combustion reaction zone. (Japanese drama) Figure 2 The reddish-brown area; the edge area: the ring-shaped region in the image where the outer contour of the flame extends outward by 50 pixels and meets the black background, i.e. Figure 2 The green area in the image; the top area: the burning end region extending downwards from the highest point of the flame in the image, within a range of 150 pixels, such as... Figure 2 The area within the yellow box is for illustrative purposes only. Figure 2 Color-coded attached diagrams.

[0034] In some embodiments, morphological features include the area of ​​the central region and / or the centroid position of the root region; color features include the RGB three-channel mean and / or color temperature; intensity features include grayscale value distribution; and motion features include optical flow vectors, which include motion speed and direction. For each frame of the image, the extracted morphological features, color features, intensity features, and motion features are used to construct a feature vector. In subsequent steps, a pre-trained convolutional neural network model based on spatial attention mechanism is used to extract deep features of each frame of the image, and these features are fused with the feature vector constructed from the multi-dimensional features to obtain a fused feature vector for each frame of the image.

[0035] In other embodiments, the morphological features may also include at least one of the following: the perimeter of the central region, roundness, and eccentricity.

[0036] In some embodiments, a pre-trained convolutional neural network (CNN) model based on spatial attention is used to extract deep features from each frame of the image, and these features are fused with multi-dimensional features to obtain a fused feature vector. The CNN model based on spatial attention is trained by assigning relatively high weights to the root region in the spatial attention module. The lightweight network model has a fast inference speed, with a single frame processing time of <20ms and a total system response time of <1 second, which can meet the real-time monitoring requirements under rapid load changes.

[0037] Specifically, convolutional neural network models based on spatial attention mechanisms include: The feature extraction layer is used to extract deep features from each frame of the image using a depthwise separable convolutional structure to obtain a deep feature map. The spatial attention module, connected to the feature extraction layer, is used to generate a spatial attention feature map based on the deep feature map. Specifically, it performs max pooling and average pooling operations on the deep feature map to generate a spatial attention map, and then multiplies the spatial attention map element-wise with the original deep feature map to obtain the spatial attention feature map. The feature fusion layer, connected to the spatial attention module, concatenates and fuses multidimensional features with the spatial attention feature map to obtain the fused feature vector of each frame image. The output layer, connected to the feature fusion layer, is used to output a fused feature vector describing the high-dimensional features of each frame of the image. In its specific implementation, the above-mentioned image recognition-based method for judging and optimizing boiler combustion stability also includes: A lightweight convolutional neural network model based on spatial attention mechanism was pre-trained. Since the stability of the root region is most critical under low-load boiler conditions, the spatial attention module was configured to automatically assign relatively high weights to the root region during training to suppress background noise interference. Multiple image sequences composed of consecutive historical frames of the furnace flame were used to train the spatial attention-based convolutional neural network model, enabling it to extract deep features, generate spatial attention feature maps based on these deep feature maps, and concatenate and fuse multidimensional features with the spatial attention feature maps.

[0038] In some embodiments, morphological features include the area of ​​the central region and / or the centroid position of the root region; color features include the RGB three-channel mean and / or color temperature; intensity features include grayscale value distribution; motion features include optical flow vector. Based on the fused feature vectors of each frame image, stability scores for different dimensions of the flame region are obtained, including: The morphological stability score is determined based on the rate of change of morphological features in consecutive multi-frame images. The color stability score is determined based on the fluctuation range of the RGB three-channel mean and / or color temperature in consecutive frames of images. Intensity stability score is determined by calculating the temporal stability of brightness in multiple consecutive frames of images based on gray value distribution; The motion stability score is determined based on the uniformity of the distribution of optical flow motion vectors.

[0039] In some embodiments, determining a morphological stability score based on the rate of change of morphological features in consecutive multi-frame images includes: calculating the rate of change of morphological features, which includes the standard deviation of the area of ​​the central region and / or the centroid displacement of the root region in consecutive multi-frame images. The standard deviation of the area of ​​the central region can reflect the degree of fluctuation in the size of the flame, and the centroid displacement of the root region can reflect whether the flame drifts.

[0040] In some examples, the standard deviation of the central region area across multiple consecutive frames is used to assess the fluctuation (rate of change) of this morphological feature. For instance, if the standard deviation is ≤ 5% of the mean central region area across multiple consecutive frames, it is considered morphologically stable, with a morphological stability score of 1.0. If the standard deviation is > 5% and ≤ 15% of the mean central region area across multiple consecutive frames, the morphological stability score is 0.8. If the standard deviation is > 15% of the mean central region area across multiple consecutive frames, the morphological stability score is 0.5.

[0041] In some cases, the morphological characteristics (rate of change) of the centroid displacement of the root region across multiple consecutive frames can also be assessed. If the average centroid displacement of the root region is less than or equal to a preset root stability threshold, the root region is considered morphologically stable. If the average centroid displacement of the root region is greater than the preset root stability threshold, the root region is considered morphologically unstable, exhibiting significant drift. For example, with a preset root stability threshold of 8, if the average centroid displacement of the root region is 32.61 pixels, the morphological stability score is 0.5; if the average centroid displacement of the root region is 3.14 pixels, the morphological stability score is 1.0.

[0042] In other examples, the morphological fluctuations can be assessed simultaneously based on the standard deviation of the central region area and the average centroid displacement of the root region. Then, the morphological stability scores obtained solely based on the standard deviation of the central region area and the average centroid displacement of the root region are weighted and summed to obtain the final morphological stability score.

[0043] In some embodiments, a color stability score is determined based on the fluctuation range of the RGB three-channel mean and / or color temperature in consecutive multi-frame images, including: calculating the RGB three-channel mean of each frame in consecutive multi-frame images, and / or calculating the average color temperature based on the red-blue ratio; calculating the band amplitude of the RGB three-channel mean and / or average color temperature within the time window of consecutive multi-frame images; and determining the color stability score based on the band amplitude (fluctuation range).

[0044] In some examples, the average RGB three channels of each frame are extracted, and the red-to-blue ratio (R / B ratio) is calculated. If the R / B fluctuation range (maximum value - minimum value) is ≤0.1, the color stability score is 1.0; if the fluctuation range is ≤0.3 (0.1 < fluctuation range), the color stability score is 0.8; and if the fluctuation range is >0.3, the color stability score is 0.5. Furthermore, based on the determined uploaded color stability score, the trend of average brightness value changes across multiple consecutive frames can be determined to correct the above color stability score: if the average brightness decreases by ≤5%, the color stability score remains unchanged; if the average brightness decreases by >5%, the color stability score is reduced by 0.2 from the original score.

[0045] In some cases, if the proportion of the red channel gradually increases but the brightness decreases while calculating the color stability score, it indicates a decrease in combustion temperature.

[0046] In some embodiments, determining an intensity stability score based on the temporal stability of brightness in consecutive frames of images calculated according to the gray value distribution includes: obtaining a gray value distribution histogram for each frame of the image; calculating the average brightness of each frame of the image based on the gray value distribution histogram; calculating the absolute value of the difference between the average brightness of adjacent frames of the image, i.e., the brightness gradient; and calculating the mean brightness gradient of the current image sequence to evaluate the flickering frequency of the flame.

[0047] In some cases, if the mean luminance gradient is ≤3, the intensity stability score is 1.0; if the mean luminance gradient is >3 and ≤8, the intensity stability score is 0.8; and if the mean luminance gradient is >8, the intensity stability score is 0.5.

[0048] In some embodiments, determining a motion stability score based on the uniformity of the distribution of optical flow motion vectors includes: calculating the optical flow motion vectors between two adjacent frames using the optical flow method.

[0049] In some examples, the mean optical flow velocity and the distribution of optical flow direction in multiple consecutive frames are statistically analyzed. If the mean optical flow velocity is ≤1.5 pixels / frame and the variance of the optical flow direction is ≤0.05, the motion stability score is 1.0; if 1.5 < mean optical flow velocity ≤3 pixels / frame and 0.5 < variance of optical flow direction ≤0.1, the motion stability score is 0.8. If the mean optical flow velocity is >3 pixels / frame or the variance of the optical flow direction is >0.1, the motion stability score is 0.5. A large variance in the optical flow direction indicates severe airflow disturbance.

[0050] In some embodiments, a comprehensive stability score is calculated based on stability scores from different dimensions, and the combustion state is determined based on the comprehensive stability score, including: Calculate the overall stability score using stability scores from different dimensions and their corresponding weighting coefficients; The current combustion state is determined based on the pre-defined correspondence between the comprehensive stability score range and the combustion state.

[0051] Specifically, the sum of the weighting coefficients corresponding to the stability scores of different dimensions equals 1. The weighting coefficients corresponding to the stability scores of each dimension are dynamically adjusted according to factors such as the type and / or load conditions of the target boiler. For example, the correspondence between the boiler type and / or load conditions and the weighting coefficients corresponding to the stability scores of each dimension can be preset. When judging the combustion state, the weighting coefficients corresponding to the stability scores of each dimension can be determined based on the type and / or load conditions of the target boiler. Since the weighting coefficients can be dynamically adjusted (set) according to the type and load conditions (high load, medium load, low load), it is suitable for monitoring combustion stability across the entire operating range (30%-100% load) during deep peak shaving, and has strong adaptability.

[0052] Specifically, a pre-defined correspondence between the overall stability score range and the combustion state is used to determine the combustion state based on the overall stability score.

[0053] In some embodiments, when the combustion state is unstable, the cause of instability is determined and the corresponding combustion parameter optimization and adjustment strategy is issued, including: using an expert knowledge base, determining the cause of instability and the corresponding combustion parameter optimization and adjustment strategy based on stability scores of different dimensions, and issuing the combustion parameter optimization and adjustment strategy.

[0054] Specifically, the expert knowledge base is constructed in the form of a rule base, including: Diagnostic rules include the weight coefficients corresponding to the stability scores of each dimension, as well as the mapping relationship between the stability scores of different dimensions and the causes of instability; Adjustment rules, including the mapping relationship between the causes of instability and the optimization adjustment strategy for combustion parameters; Constraint rules include the optimal adjustment range and safety boundaries for combustion parameters.

[0055] The diagnostic rules include: If the morphological stability score is lower than the first preset threshold, the instability may be caused by uneven air distribution or uneven fuel distribution. The color stability score is lower than the second preset threshold. The reasons for instability include fluctuations in combustion temperature or changes in fuel quality. The strength stability score is lower than the third preset threshold. The reasons for instability include insufficient combustion intensity or an inappropriate excess air coefficient. If the motion stability score is lower than the fourth preset threshold, the instability may be caused by severe airflow disturbance or inappropriate primary wind speed.

[0056] The weighting coefficients corresponding to the stability scores of each dimension in the diagnostic rules can be dynamically adjusted based on the execution of the combustion parameter optimization and adjustment strategy. That is, after determining the weighting coefficients based on the correspondence between the boiler type and / or load condition and the weighting coefficients corresponding to the stability scores of each dimension, if the overall stability score indicates that the combustion state is unstable, then the combustion parameter optimization and adjustment strategy needs to be executed. If, after executing the combustion parameter optimization and adjustment strategy, the overall stability score indicates that the combustion state is stable, then the combustion parameter optimization and adjustment strategy is effective. At this time, the stability scores of at least some dimensions have improved. For example, if the intensity stability score improves to 1, then the intensity is stable. At this time, it is necessary to adjust the weighting coefficients corresponding to the current stability scores of different dimensions, reduce the weighting coefficients corresponding to the stability scores of dimensions that have not reached 1, and increase the weighting coefficients corresponding to the stability scores of dimensions that are below the corresponding preset thresholds. In this way, the determination of the overall stability score can focus more on the unstable dimensions, which is conducive to continuously improving combustion stability through dynamic adjustment of combustion parameters and avoiding the inaccuracy of the overall stability score due to the excessively high weighting coefficients of the already stable dimensions.

[0057] Combustion parameters include at least one of primary air volume, secondary air volume, fuel supply, and air distribution angle. Therefore, combustion parameter optimization and adjustment strategies include at least one of primary air volume adjustment, secondary air volume adjustment, fuel supply adjustment, and air distribution angle adjustment.

[0058] In some embodiments, the above-mentioned image recognition-based method for judging and optimizing boiler combustion stability further includes: after executing the combustion parameter optimization and adjustment strategy, continuing to acquire multiple consecutive frames of images of the flame inside the furnace of the target boiler to obtain a new comprehensive stability score, and judging the combustion state based on the new comprehensive stability score; if the combustion state is stable, recording that the executed combustion parameter optimization and adjustment strategy is effective, and adjusting the weight coefficients corresponding to the stability scores of different dimensions according to the stability scores of different dimensions, and updating them to the expert knowledge base.

[0059] Specifically, after executing the combustion parameter optimization and adjustment strategy, the next set of consecutive multi-frame images is acquired to perform a new round of comprehensive stability score calculation. If the new round of comprehensive stability score indicates that the combustion state is stable and there is no obvious drift phenomenon in the root region, then the executed combustion parameter optimization and adjustment strategy is effective. At this time, the weight coefficients corresponding to the dimensions whose stability scores have reached 1 (indicating that the threshold of 1 for the corresponding dimension is stable) are reduced, while the weight coefficients corresponding to the dimensions that have not reached 1 are increased, so that the determination of the comprehensive stability score can focus more on the unstable dimensions. The weight coefficients corresponding to the stability scores of each dimension in the diagnostic rules in the expert knowledge base are updated according to the adjusted weight coefficients for use in the next round of judgment. In the subsequent combustion stability monitoring process, the weight coefficients are dynamically adjusted according to the combustion stability, the execution of the combustion parameter optimization and adjustment strategy, etc., to achieve dynamic closed-loop feedback.

[0060] Example 2: This embodiment provides an application example of the above method embodiment.

[0061] The method described in the above embodiment is applied to a pilot-scale combustion test bench with a rated load of 300 kW. The test bench is in a deep peak-shaving phase at 40% of its rated load, resulting in a low furnace temperature and a high risk of combustion instability. The system hardware uses an industrial-grade visible light high-speed camera installed at the burner's viewing port, with a sampling frequency of 100 fps and a resolution of 1024 × 1024. The process of this application example is as follows: Figure 3 As shown.

[0062] Step 1: Image Preprocessing and Region Segmentation Image acquisition: Acquire 50 consecutive frames of RGB color images using a high-speed visible light camera.

[0063] Preprocessing: Gaussian filtering is applied to these 50 consecutive frames of images to remove noise, and size normalization is performed.

[0064] Region segmentation: For each frame of the image, it is divided into four key regions based on the geometric shape of the flame: Root region: The area extending 200 pixels into the furnace from the center of the burner outlet nozzle. Used to monitor whether the ignition point has deviated.

[0065] Central region: The region with the largest connected area and the highest brightness threshold in the image is defined as the main combustion reaction zone.

[0066] Edge region: Extract the outer contour of the flame, expand it outward by 50 pixels to form a ring-shaped area that intersects with the black background.

[0067] Top area: The area extending 150 pixels down from the highest pixel of the flame.

[0068] like Figure 2 As shown, the yellow box represents the top area, the blue box represents the root area, the green box represents the edge area, the area outside the green box (excluding the area inside the blue box) represents the background, and the red area represents the center area.

[0069] Step 2: Multidimensional Feature Extraction For this sequence of 50 images, the following four-dimensional features were extracted: Morphological characteristics: area of ​​the central region and location of the centroid of the root region; Color characteristics: Calculate the average RGB three-channel values ​​of each frame image and calculate the average color temperature based on the red-blue ratio.

[0070] Intensity features: Extract grayscale histograms to obtain the grayscale value distribution of each frame.

[0071] Motion characteristics: The optical flow motion vector between two adjacent frames is calculated using the optical flow method. The optical flow motion vector includes the optical flow velocity and the optical flow direction.

[0072] The extracted multidimensional feature data are used to construct a feature vector.

[0073] Step 3: Feature Fusion and Network Inference The original image is input into a pre-trained lightweight convolutional neural network model. In this model, the spatial attention mechanism automatically assigns higher weights to the root region (because root stability is most critical under low load) to suppress background noise interference. Feature fusion: The deep features extracted by the lightweight convolutional neural network model are concatenated with the feature vectors from step two at the feature fusion layer, outputting the final feature map describing the high-dimensional features.

[0074] Step 4: Characteristic Stability Evaluation Methods and Calculations First, based on the current load conditions, the corresponding weighting coefficients are automatically applied. Taking the current 50 consecutive frames of images being acquired under a 40% load on the test bench as an example, the weighting for morphological stability is 0.4; for color stability, it is 0.1; for intensity stability, it is 0.3; and for motion stability, it is 0.2.

[0075] Secondly, each feature is scored, and all scores are normalized to between 0 and 1: (a) Morphological stability score The area of ​​the central region with a grayscale value greater than 200 is extracted as a feature. The standard deviation of this feature is calculated for consecutive frames to evaluate its fluctuation characteristics. If the standard deviation is ≤ 5% of the average area of ​​the central region, it is considered morphologically stable, and the morphological stability score is 1.0. If the standard deviation is > 5% of the average area of ​​the central region and ≤ 15% of the average area of ​​the central region, the morphological stability score is 0.8. If the standard deviation is > 15% of the average area of ​​the central region, the morphological stability score is 0.5.

[0076] (b) Color stability score Extract the mean of the RGB three channels of each frame image and calculate the red-to-blue ratio (R / B Ratio). An R / B fluctuation range (maximum to minimum) ≤ 0.1 is considered color stable, with a color stability score of 1.0. A fluctuation range of 0.1 < ≤ 0.3 results in a color stability score of 0.8. A fluctuation range > 0.3 results in a color stability score of 0.5.

[0077] Average brightness value change trend: Average brightness decrease ≤5%: Color stability score remains unchanged. Average brightness decrease >5%: Color stability score decreases by 0.2 from the original score.

[0078] (c) Strength stability score Extract the grayscale histogram of each frame and calculate the absolute value of the average brightness difference between adjacent frames (brightness gradient).

[0079] Mean luminance gradient ≤ 3: Intensity stability score = 1.0. Mean luminance gradient > 3 and ≤ 8: Intensity stability score = 0.8. Mean luminance gradient > 8: Intensity stability score = 0.5.

[0080] (d) Motion stability score The optical flow velocity and direction of adjacent frames are calculated using optical flow. A velocity mean ≤ 1.5 pixels / frame and a direction variance ≤ 0.05 are considered motion-stable, with a motion stability score of 1.0. A velocity mean ≤ 3 pixels / frame and a direction variance ≤ 0.1 (1.5 < velocity mean ≤ 3 pixels / frame) are considered motion-stable, with a motion stability score of 0.8. A velocity mean > 3 pixels / frame or a direction variance > 0.1 is considered motion-stable, with a motion stability score of 0.5.

[0081] Relevant data calculated based on the multidimensional characteristics of flames, such as... Figure 4 As shown, (a) is the standard deviation of the central region area, (b) is the temporal variation of the RGB three colors of the flame, (c) is the mean of the brightness gradient, and (d) is the motion vector field between two adjacent frames.

[0082] Next, the overall stability score is calculated as follows: Overall stability score = (0.4 × morphological stability score) + (0.1 × color stability score) + (0.3 × brightness intensity stability score) + (0.2 × motion stability score).

[0083] The combustion state can be determined by referring to the table below:

[0084] In this application example, the standard deviation of the central area is 12% of the average area, and the morphological stability score is 0.8.

[0085] The average displacement of the root centroid is 32.61 pixels, and the morphological stability score is 0.5 (this score is used for calculation below).

[0086] R / B fluctuation range = 0, color stability score = 1. Average brightness decreases by 2%, color stability score remains unchanged.

[0087] Mean luminance gradient = 9.902, intensity stability score = 0.5.

[0088] Optical flow velocity = 2.084 pixels / frame, directional variance = 3.175, motion stability score = 0.5.

[0089] Calculate the overall stability score based on the weights: S=0.4×0.5+0.1×1+0.3×0.5+0.2×0.5=0.550.

[0090] Step 5: Fault Diagnosis and Optimization Decisions Step 1, Status Check: Based on the comprehensive stability score of 0.550 calculated in step four, compared with the preset stability threshold judgment table in Table 1, the current combustion state is determined to be "severely unstable". This is consistent with reality, because the combustion test stand was undergoing a rapid load increase process during the filming time, and the flame shape changed drastically in a short period of time, resulting in an unstable combustion state.

[0091] Step 2, troubleshooting: Analyze the multidimensional features extracted in step two, and combine them with the expert knowledge base, using the following criteria: Morphological characteristics: The average displacement of the centroid in the root region is 32.61 pixels, which exceeds the root stability threshold of 8 pixels, indicating that the root region is unstable and there is obvious drift phenomenon.

[0092] Motion characteristics: The variance of the optical flow vector direction is 0.125, which exceeds the upper limit of airflow disturbance of 0.10, indicating that the airflow disturbance is relatively large.

[0093] Intensity characteristics: The average brightness gradient is relatively high, at 2.084 pixels / frame, indicating that the flickering frequency of the flame is relatively fast.

[0094] Based on the above criteria, after matching the rules with the expert knowledge base, the system determined the main cause of the fault to be: Excessive wind speed caused the flame base to be blown off, and at the same time, the air distribution was deviated, causing a drop in temperature and fluctuations in brightness.

[0095] Step 3, Generation of Regulation Recommendations (Adjustment Strategies): Based on the fault diagnosis results, adjustment rules are extracted from the expert knowledge base, and control instructions are automatically generated, such as: Recommendation 1: Reduce the primary air pressure adjustment of the burner to decrease the air velocity and prevent the flame from being blown off.

[0096] Recommendation 2: Close a small number of perimeter wind deflectors to reduce airflow disturbance and concentrate the flame.

[0097] Recommendation 3: Continue to increase the amount of coal under this operating condition to improve temperature stability.

[0098] Step Six: Closed-Loop Feedback After the DCS (Distributed Control System) issues the above suggestions to the actuators and executes them, it continues to acquire 50 new consecutive frames of images. If the overall stability score in the new round of calculation improves to 0.85 and the morphology of the root region is stable, then the optimization adjustment strategy is recorded as effective, and the weight parameters in the expert knowledge base are updated to complete one closed-loop optimization.

[0099] This study uses a high-speed visible light camera to acquire RGB color image sequences of flames inside a boiler furnace. After preprocessing, the flame region is divided into root, center, edge, and top regions based on its geometric morphology. Morphological, color, brightness, and motion features are extracted. A lightweight convolutional neural network combined with spatial attention is then used to fuse these multi-dimensional features. Morphological stability scores, color stability scores, intensity stability scores, and motion stability scores are calculated, and a comprehensive stability score is obtained through weighted fusion. Based on the comprehensive stability score and feature analysis results, combustion parameter optimization suggestions are generated using an expert knowledge base. This method has low hardware costs, enables closed-loop control from stability assessment to optimization adjustment, and exhibits good practicality and interpretability.

[0100] Example 3: Another embodiment of this application relates to a boiler combustion stability judgment and optimization adjustment device based on image recognition. The implementation details of this embodiment's image recognition-based boiler combustion stability judgment and optimization adjustment device are described below. The following details are provided for ease of understanding and are not essential for implementing this solution. A schematic diagram of this embodiment's image recognition-based boiler combustion stability judgment and optimization adjustment device can be seen as follows: Figure 5 As shown, it includes an image acquisition module 301, an image processing module 302, a stability judgment module 303, and an optimization adjustment module 304.

[0101] The image acquisition module 301 is used to acquire multiple consecutive frames of images of the flame inside the furnace of the target boiler. Specifically, a high-speed visible light camera is used to acquire multiple consecutive RGB color images of the flame inside the furnace of the target boiler, and the multiple consecutive RGB color images constitute an RGB color image sequence. Only a common high-speed visible light camera is used, without the need for an ultraviolet camera, image intensifier, and multi-head fiber bundle, resulting in low hardware costs and ease of implementation. Furthermore, the high-speed visible light camera is simple to maintain, and the cleaning cycle of the quartz glass window is extended to 1 to 2 months, eliminating the need to replace the expensive image intensifier.

[0102] Image processing module 302 is used to divide the flame region in each frame image into root region, center region, edge region and top region, extract multi-dimensional features of the flame region in each frame image, extract deep features of each frame image, and fuse them with multi-dimensional features to obtain a fused feature vector. Multi-dimensional features include morphological features, color features, intensity features and motion features. In specific implementation, preprocessing is also performed before dividing the flame region into regions, including Gaussian filtering to denoise multiple consecutive frames of images and size normalization processing.

[0103] The stability judgment module 303 is used to obtain the stability scores of different dimensions of the flame area based on the fusion feature vector of each frame image, calculate the comprehensive stability score based on the stability scores of different dimensions, and judge the combustion state based on the comprehensive stability score. The optimization and adjustment module 304 is used to determine the cause of instability and issue corresponding combustion parameter optimization and adjustment strategies when the combustion state is unstable.

[0104] In this embodiment, by acquiring multiple consecutive frames of flame images within the furnace of the target boiler, the flame region in each frame is divided into root, center, edge, and top regions. The morphological, color, intensity, and motion features of the flame region in each frame are extracted and then fused with the extracted deep features of each frame to obtain a fused feature vector. Based on the fused feature vectors of each frame, different dimensional stability scores of the flame region are obtained. A comprehensive stability score is calculated based on these scores to determine the combustion state. When the combustion state is unstable, the cause of instability is identified, and corresponding combustion parameter optimization and adjustment strategies are issued. This achieves closed-loop control from stability assessment to optimization and adjustment, demonstrating practical engineering application value. On one hand, the determination of combustion state stability is achieved through multiple consecutive frames of flame images, eliminating the need for significant industrial field deployment costs. On the other hand, the identification and judgment process based on multiple consecutive frames comprehensively considers the morphological, color, intensity, and motion features of different regions for real-time and accurate stability evaluation. This allows for timely detection of combustion instability, timely determination of the cause, and adjustment of combustion parameters, enabling closed-loop control of boiler combustion control through real-time monitoring and dynamic adjustment, thereby improving boiler combustion stability.

[0105] In some embodiments, based on the geometric features of the flame, the flame region in each frame image is divided into a root region, a central region, an edge region, and a top region. The root region includes the flame region extending from the burner outlet into the furnace within a first distance range; the central region includes the flame region with the highest brightness and the largest area; the edge region includes the region extending outward from the outer contour of the flame within a preset pixel range and intersecting with the background; and the top region is the region extending within a second distance range below the highest position of the flame.

[0106] In some embodiments, morphological features include the area of ​​the central region and / or the centroid position of the root region; color features include the RGB three-channel mean and / or color temperature; intensity features include grayscale value distribution; and motion features include optical flow vectors, which include motion speed and direction. For each frame of the image, the extracted morphological features, color features, intensity features, and motion features are used to construct a feature vector. In subsequent steps, a pre-trained convolutional neural network model based on spatial attention mechanism is used to extract deep features of each frame of the image, and these features are fused with the feature vector constructed from the multi-dimensional features to obtain a fused feature vector for each frame of the image.

[0107] In other embodiments, the morphological features may also include at least one of the following: the perimeter of the central region, roundness, and eccentricity.

[0108] In some embodiments, a pre-trained convolutional neural network (CNN) model based on spatial attention is used to extract deep features from each frame of the image, and these features are fused with multi-dimensional features to obtain a fused feature vector. The CNN model based on spatial attention is trained by assigning relatively high weights to the root region in the spatial attention module. The lightweight network model has a fast inference speed, with a single frame processing time of <20ms and a total system response time of <1 second, which can meet the real-time monitoring requirements under rapid load changes.

[0109] Specifically, convolutional neural network models based on spatial attention mechanisms include: The feature extraction layer is used to extract deep features from each frame of the image using a depthwise separable convolutional structure to obtain a deep feature map. The spatial attention module, connected to the feature extraction layer, is used to generate a spatial attention feature map based on the deep feature map. Specifically, it performs max pooling and average pooling operations on the deep feature map to generate a spatial attention map, and then multiplies the spatial attention map element-wise with the original deep feature map to obtain the spatial attention feature map. The feature fusion layer, connected to the spatial attention module, concatenates and fuses multidimensional features with the spatial attention feature map to obtain the fused feature vector of each frame image. The output layer, connected to the feature fusion layer, is used to output a fused feature vector describing the high-dimensional features of each frame of the image. In its specific implementation, the above-mentioned image recognition-based method for judging and optimizing boiler combustion stability also includes: A lightweight convolutional neural network model based on spatial attention mechanism was pre-trained. Since the stability of the root region is most critical under low-load boiler conditions, the spatial attention module was configured to automatically assign relatively high weights to the root region during training to suppress background noise interference. Multiple image sequences composed of consecutive historical frames of the furnace flame were used to train the spatial attention-based convolutional neural network model, enabling it to extract deep features, generate spatial attention feature maps based on these deep feature maps, and concatenate and fuse multidimensional features with the spatial attention feature maps.

[0110] In some embodiments, morphological features include the area of ​​the central region and / or the centroid position of the root region; color features include the RGB three-channel mean and / or color temperature; intensity features include grayscale value distribution; motion features include optical flow vector. Based on the fused feature vectors of each frame image, stability scores for different dimensions of the flame region are obtained, including: The morphological stability score is determined based on the rate of change of morphological features in consecutive multi-frame images. The color stability score is determined based on the fluctuation range of the RGB three-channel mean and / or color temperature in consecutive frames of images. Intensity stability score is determined by calculating the temporal stability of brightness in multiple consecutive frames of images based on gray value distribution; The motion stability score is determined based on the uniformity of the distribution of optical flow motion vectors.

[0111] In some embodiments, determining a morphological stability score based on the rate of change of morphological features in consecutive multi-frame images includes: calculating the rate of change of morphological features, which includes the standard deviation of the area of ​​the central region and / or the centroid displacement of the root region in consecutive multi-frame images. The standard deviation of the area of ​​the central region can reflect the degree of fluctuation in the size of the flame, and the centroid displacement of the root region can reflect whether the flame drifts.

[0112] In some examples, the standard deviation of the central region area across multiple consecutive frames is used to assess the fluctuation (rate of change) of this morphological feature. For instance, if the standard deviation is ≤ 5% of the mean central region area across multiple consecutive frames, it is considered morphologically stable, with a morphological stability score of 1.0. If the standard deviation is > 5% and ≤ 15% of the mean central region area across multiple consecutive frames, the morphological stability score is 0.8. If the standard deviation is > 15% of the mean central region area across multiple consecutive frames, the morphological stability score is 0.5.

[0113] In some cases, the morphological characteristics (rate of change) of the centroid displacement of the root region across multiple consecutive frames can also be assessed. If the average centroid displacement of the root region is less than or equal to a preset root stability threshold, the root region is considered morphologically stable. If the average centroid displacement of the root region is greater than the preset root stability threshold, the root region is considered morphologically unstable, exhibiting significant drift. For example, with a preset root stability threshold of 8, if the average centroid displacement of the root region is 32.61 pixels, the morphological stability score is 0.5; if the average centroid displacement of the root region is 3.14 pixels, the morphological stability score is 1.0.

[0114] In other examples, the morphological fluctuations can be assessed simultaneously based on the standard deviation of the central region area and the average centroid displacement of the root region. Then, the morphological stability scores obtained solely based on the standard deviation of the central region area and the average centroid displacement of the root region are weighted and summed to obtain the final morphological stability score.

[0115] In some embodiments, a color stability score is determined based on the fluctuation range of the RGB three-channel mean and / or color temperature in consecutive multi-frame images, including: calculating the RGB three-channel mean of each frame in consecutive multi-frame images, and / or calculating the average color temperature based on the red-blue ratio; calculating the band amplitude of the RGB three-channel mean and / or average color temperature within the time window of consecutive multi-frame images; and determining the color stability score based on the band amplitude (fluctuation range).

[0116] In some examples, the average RGB three channels of each frame are extracted, and the red-to-blue ratio (R / B ratio) is calculated. If the R / B fluctuation range (maximum value - minimum value) is ≤0.1, the color stability score is 1.0; if the fluctuation range is ≤0.3 (0.1 < fluctuation range), the color stability score is 0.8; and if the fluctuation range is >0.3, the color stability score is 0.5. Furthermore, based on the determined uploaded color stability score, the trend of average brightness value changes across multiple consecutive frames can be determined to correct the above color stability score: if the average brightness decreases by ≤5%, the color stability score remains unchanged; if the average brightness decreases by >5%, the color stability score is reduced by 0.2 from the original score.

[0117] In some cases, if the proportion of the red channel gradually increases but the brightness decreases while calculating the color stability score, it indicates a decrease in combustion temperature.

[0118] In some embodiments, determining an intensity stability score based on the temporal stability of brightness in consecutive frames of images calculated according to the gray value distribution includes: obtaining a gray value distribution histogram for each frame of the image; calculating the average brightness of each frame of the image based on the gray value distribution histogram; calculating the absolute value of the difference between the average brightness of adjacent frames of the image, i.e., the brightness gradient; and calculating the mean brightness gradient of the current image sequence to evaluate the flickering frequency of the flame.

[0119] In some cases, if the mean luminance gradient is ≤3, the intensity stability score is 1.0; if the mean luminance gradient is >3 and ≤8, the intensity stability score is 0.8; and if the mean luminance gradient is >8, the intensity stability score is 0.5.

[0120] In some embodiments, determining a motion stability score based on the uniformity of the distribution of optical flow motion vectors includes: calculating the optical flow motion vectors between two adjacent frames using the optical flow method.

[0121] In some examples, the mean optical flow velocity and the distribution of optical flow direction in multiple consecutive frames are statistically analyzed. If the mean optical flow velocity is ≤1.5 pixels / frame and the variance of the optical flow direction is ≤0.05, the motion stability score is 1.0; if 1.5 < mean optical flow velocity ≤3 pixels / frame and 0.5 < variance of optical flow direction ≤0.1, the motion stability score is 0.8. If the mean optical flow velocity is >3 pixels / frame or the variance of the optical flow direction is >0.1, the motion stability score is 0.5. A large variance in the optical flow direction indicates severe airflow disturbance.

[0122] In some embodiments, a comprehensive stability score is calculated based on stability scores of different dimensions, and the combustion state is determined based on the comprehensive stability score, including: calculating a comprehensive stability score using stability scores of different dimensions and corresponding weighting coefficients; and determining the current combustion state based on a preset correspondence between a comprehensive stability score range and the combustion state.

[0123] Specifically, the sum of the weighting coefficients corresponding to the stability scores of different dimensions equals 1. The weighting coefficients corresponding to the stability scores of each dimension are dynamically adjusted according to factors such as the type and / or load conditions of the target boiler. For example, the correspondence between the boiler type and / or load conditions and the weighting coefficients corresponding to the stability scores of each dimension can be preset. When judging the combustion state, the weighting coefficients corresponding to the stability scores of each dimension can be determined based on the type and / or load conditions of the target boiler. Since the weighting coefficients can be dynamically adjusted (set) according to the type and load conditions (high load, medium load, low load), it is suitable for monitoring combustion stability across the entire operating range (30%-100% load) during deep peak shaving, and has strong adaptability.

[0124] Specifically, a pre-defined correspondence between the overall stability score range and the combustion state is used to determine the combustion state based on the overall stability score.

[0125] In some embodiments, when the combustion state is unstable, the cause of instability is determined and the corresponding combustion parameter optimization and adjustment strategy is issued, including: using an expert knowledge base, determining the cause of instability and the corresponding combustion parameter optimization and adjustment strategy based on stability scores of different dimensions, and issuing the combustion parameter optimization and adjustment strategy.

[0126] Specifically, the expert knowledge base is constructed in the form of a rule base, including: Diagnostic rules include the weight coefficients corresponding to the stability scores of each dimension, as well as the mapping relationship between the stability scores of different dimensions and the causes of instability; Adjustment rules, including the mapping relationship between the causes of instability and the optimization adjustment strategy for combustion parameters; Constraint rules include the optimal adjustment range and safety boundaries for combustion parameters.

[0127] The diagnostic rules include: If the morphological stability score is lower than the first preset threshold, the instability may be caused by uneven air distribution or uneven fuel distribution. The color stability score is lower than the second preset threshold. The reasons for instability include fluctuations in combustion temperature or changes in fuel quality. The strength stability score is lower than the third preset threshold. The reasons for instability include insufficient combustion intensity or an inappropriate excess air coefficient. If the motion stability score is lower than the fourth preset threshold, the instability may be caused by severe airflow disturbance or inappropriate primary wind speed.

[0128] The weighting coefficients corresponding to the stability scores of each dimension in the diagnostic rules can be dynamically adjusted based on the execution of the combustion parameter optimization and adjustment strategy. That is, after determining the weighting coefficients based on the correspondence between the boiler type and / or load condition and the weighting coefficients corresponding to the stability scores of each dimension, if the overall stability score indicates that the combustion state is unstable, then the combustion parameter optimization and adjustment strategy needs to be executed. If, after executing the combustion parameter optimization and adjustment strategy, the overall stability score indicates that the combustion state is stable, then the combustion parameter optimization and adjustment strategy is effective. At this time, the stability scores of at least some dimensions have improved. For example, if the intensity stability score improves to 1, then the intensity is stable. At this time, it is necessary to adjust the weighting coefficients corresponding to the current stability scores of different dimensions, reduce the weighting coefficients corresponding to the stability scores of dimensions that have not reached 1, and increase the weighting coefficients corresponding to the stability scores of dimensions that are below the corresponding preset thresholds. In this way, the determination of the overall stability score can focus more on the unstable dimensions, which is conducive to continuously improving combustion stability through dynamic adjustment of combustion parameters and avoiding the inaccuracy of the overall stability score due to the excessively high weighting coefficients of the already stable dimensions.

[0129] Combustion parameters include at least one of primary air volume, secondary air volume, fuel supply, and air distribution angle. Therefore, combustion parameter optimization and adjustment strategies include at least one of primary air volume adjustment, secondary air volume adjustment, fuel supply adjustment, and air distribution angle adjustment.

[0130] In some embodiments, the above-mentioned image recognition-based method for judging and optimizing boiler combustion stability further includes: after executing the combustion parameter optimization and adjustment strategy, continuing to acquire multiple consecutive frames of images of the flame inside the furnace of the target boiler to obtain a new comprehensive stability score, and judging the combustion state based on the new comprehensive stability score; if the combustion state is stable, recording that the executed combustion parameter optimization and adjustment strategy is effective, and adjusting the weight coefficients corresponding to the stability scores of different dimensions according to the stability scores of different dimensions, and updating them to the expert knowledge base.

[0131] Specifically, after executing the combustion parameter optimization and adjustment strategy, the next set of consecutive multi-frame images is acquired to perform a new round of comprehensive stability score calculation. If the new round of comprehensive stability score indicates that the combustion state is stable and there is no obvious drift phenomenon in the root region, then the executed combustion parameter optimization and adjustment strategy is effective. At this time, the weight coefficients corresponding to the dimensions whose stability scores have reached 1 (indicating that the threshold of 1 for the corresponding dimension is stable) are reduced, while the weight coefficients corresponding to the dimensions that have not reached 1 are increased, so that the determination of the comprehensive stability score can focus more on the unstable dimensions. The weight coefficients corresponding to the stability scores of each dimension in the diagnostic rules in the expert knowledge base are updated according to the adjusted weight coefficients for use in the next round of judgment. In the subsequent combustion stability monitoring process, the weight coefficients are dynamically adjusted according to the combustion stability, the execution of the combustion parameter optimization and adjustment strategy, etc., to achieve dynamic closed-loop feedback.

[0132] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed in this application; however, this does not mean that other units are absent in this embodiment.

[0133] Example 4: Another embodiment of this application relates to an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the image recognition-based boiler combustion stability judgment and optimization adjustment method in the above embodiments.

[0134] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0135] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0136] Example 5: Another embodiment of this application relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the above-described embodiment of the image recognition-based boiler combustion stability judgment and optimization adjustment method.

[0137] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0138] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.

Claims

1. A method for judging and optimizing boiler combustion stability based on image recognition, characterized in that, include: Acquire consecutive multi-frame images of the flame inside the furnace of the target boiler; The flame region in each frame image is divided into root region, center region, edge region and top region. Multidimensional features of the flame region in each frame image are extracted, deep features of each frame image are extracted, and fused with the multidimensional features to obtain a fused feature vector. The multidimensional features include morphological features, color features, intensity features and motion features. Based on the fused feature vectors of each frame image, different dimension stability scores of the flame region are obtained. A comprehensive stability score is calculated based on the different dimension stability scores, and the combustion state is determined based on the comprehensive stability score. When the combustion state is unstable, the cause of the instability is determined and a corresponding combustion parameter optimization and adjustment strategy is issued.

2. The method for judging and optimizing boiler combustion stability based on image recognition according to claim 1, characterized in that, Based on the geometric features of the flame, the flame region in each frame image is divided into a root region, a central region, an edge region, and a top region. The root region includes the flame region extending from the burner outlet into the furnace within a first distance range. The central region includes the flame region with the highest brightness and the largest area. The edge region includes a region that extends outward from the outer contour of the flame by a preset number of pixels and intersects with the background; the top region is the region extending within a second distance below the highest point of the flame.

3. The method for judging and optimizing boiler combustion stability based on image recognition according to claim 2, characterized in that, Using a pre-trained convolutional neural network model based on spatial attention mechanism, deep features of each frame of image are extracted and fused with the multidimensional features to obtain a fused feature vector; The convolutional neural network model based on spatial attention mechanism includes: The feature extraction layer is used to extract deep features from each frame of the image to obtain a deep feature map; A spatial attention module, connected to the feature extraction layer, is used to generate a spatial attention feature map based on the deep feature map. The feature fusion layer, connected to the spatial attention module, concatenates and fuses the multidimensional features with the spatial attention feature map to obtain the fused feature vector of each frame image; The output layer, connected to the feature fusion layer, is used to output the fused feature vector of each frame of the image; The convolutional neural network model based on the spatial attention mechanism is trained by assigning relatively high weights to the root region in the spatial attention module.

4. The method for judging and optimizing boiler combustion stability based on image recognition according to claim 2, characterized in that, The morphological features include the area of ​​the central region and / or the centroid position of the root region; the color features include the RGB three-channel mean and / or color temperature; the intensity features include grayscale value distribution; and the motion features include optical flow vectors. Based on the fused feature vectors of each frame image, stability scores for different dimensions of the flame region are obtained, including: The morphological stability score is determined based on the rate of change of morphological features in consecutive multi-frame images. The color stability score is determined based on the fluctuation range of the RGB three-channel mean and / or color temperature in consecutive frames of images. Intensity stability score is determined by calculating the temporal stability of brightness in multiple consecutive frames of images based on gray value distribution; The motion stability score is determined based on the uniformity of the distribution of optical flow motion vectors.

5. The method for judging and optimizing boiler combustion stability based on image recognition according to claim 1, characterized in that, A comprehensive stability score is calculated based on stability scores from different dimensions. The combustion state is then determined based on the comprehensive stability score, including: Calculate the overall stability score using stability scores from different dimensions and their corresponding weighting coefficients; The current combustion state is determined based on the pre-defined correspondence between the comprehensive stability score range and the combustion state.

6. The method for judging and optimizing boiler combustion stability based on image recognition according to claim 5, characterized in that, When the combustion state is unstable, the cause of the instability is determined and a corresponding combustion parameter optimization and adjustment strategy is issued, including: Using an expert knowledge base, the causes of instability and the corresponding combustion parameter optimization and adjustment strategies are determined based on stability scores across different dimensions, and the combustion parameter optimization and adjustment strategies are then issued. The expert knowledge base includes: Diagnostic rules include the weight coefficients corresponding to the stability scores of each dimension, as well as the mapping relationship between the stability scores of different dimensions and the causes of instability; Adjustment rules, including the mapping relationship between the causes of instability and the optimization adjustment strategy for combustion parameters; Constraint rules include the optimal adjustment range and safety boundaries for combustion parameters.

7. The method for judging and optimizing boiler combustion stability based on image recognition according to claim 6, characterized in that, Also includes: After implementing the combustion parameter optimization and adjustment strategy, the system continues to acquire multiple consecutive frames of flame images inside the furnace of the target boiler to obtain a new comprehensive stability score, and then judges the combustion state based on the new comprehensive stability score. If the combustion state is stable, the effective combustion parameter optimization and adjustment strategy is recorded, and the weight coefficients corresponding to the stability scores of each dimension are adjusted according to the stability scores of different dimensions and updated to the expert knowledge base.

8. A boiler combustion stability judgment and optimization adjustment device based on image recognition, characterized in that, include: The image acquisition module is used to acquire multiple consecutive frames of images of the flame inside the furnace of the target boiler. The image processing module is used to divide the flame region in each frame image into the root region, the center region, the edge region and the top region, extract the multidimensional features of the flame region in each frame image, extract the deep features of each frame image, and fuse them with the multidimensional features to obtain a fused feature vector. The multidimensional features include morphological features, color features, intensity features and motion features. The stability judgment module is used to obtain the stability scores of the flame region in different dimensions based on the fused feature vector of each frame image, calculate the comprehensive stability score based on the stability scores in different dimensions, and judge the combustion state based on the comprehensive stability score. The optimization and adjustment module is used to determine the cause of instability and issue corresponding combustion parameter optimization and adjustment strategies when the combustion state is unstable.

9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the image recognition-based boiler combustion stability judgment and optimization adjustment method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the image recognition-based method for judging and optimizing boiler combustion stability as described in any one of claims 1 to 7.

Images