Automatic monitoring system and method for flue gas emission of stationary pollution source

Abstract

The invention relates to the field of flue gas monitoring, and particularly discloses an automatic monitoring system and method for flue gas emission of a stationary pollution source, and the system employs an artificial intelligence technology based on the deep learning field to carry out the feature extraction and coding of a multispectral image of air at an input end of a sampling pipe and a multispectral image of air at an output end of the sampling pipe. And a classification result of whether the sampling tube has an abnormal condition is obtained. Thus, by intelligently judging whether the sampling pipe is loosened or not and other abnormal conditions, the difficulty of manually checking the sampling pipe is reduced, and the situation that CEMS data is false due to the abnormal sampling pipe can be effectively avoided.

Description

Technical Field

[0001] The present application relates to the field of flue gas monitoring, and in particular to a system and method for automatically monitoring flue gas emissions from fixed pollution sources. Background Art

[0002] Stationary pollution sources are facilities that emit or release harmful substances, negatively impacting the environment. They are categorized as organized emission sources (such as chimneys) and unorganized emission sources (such as open-air facilities). The exhaust gases they emit contain both solid soot and dust, as well as a variety of harmful substances in gaseous and aerosol forms. Major types of stationary pollutants include soot, carbon dioxide, sulfur dioxide, carbon monoxide, nitrogen oxides, fluoride, hydrogen sulfide, chlorine, hydrogen chloride, sulfuric acid and aerosols, phosphorus oxides, phenols, benzene, gasoline, mercury, lead and its compounds, arsine, cadmium, chromium and its compounds, etc. Stationary pollution refers to pollution of the atmosphere. Stationary atmospheric pollution sources are fixed sources of atmospheric pollutants, including boilers and kilns in factories, enterprises, institutions, and catering services, as well as exhaust pipes used by residents for daily life. Boilers are one of the main emission devices of stationary atmospheric pollution sources. With the development of human production and the prosperity of life, especially the development of modern industry, agriculture, and energy, the harm caused to the environment by pollutants emitted by stationary atmospheric pollution sources has become increasingly apparent. In practical applications, a continuous monitoring system (CEMS) for flue gas emissions from fixed pollution sources is often used to detect the content of atmospheric pollutants. CEMS is a real-time monitoring system used to measure the concentration of pollutants emitted from chimneys or other emission sources. It uses various sensors, sampling pipes and analyzers to continuously measure specific pollutants and analyze the concentration of particulate matter, flue gas temperature, flue gas pressure, flue gas flow rate, flue gas humidity, flue gas oxygen content, SO2 concentration, and NOx concentration in flue gas samples. However, due to economic benefits, some companies will falsify CEMS data. One typical case is to loosen the sampling pipe, resulting in the actual monitored gas being diluted gas, deviating from the true value. However, due to the harsh environment in which the sampling pipe is located, it is difficult to conduct manual inspection alone.

[0003] Therefore, an optimized automatic monitoring system for flue gas emissions from fixed pollution sources is expected. Summary of the Invention

[0004] In order to solve the above-mentioned technical problems, the present application is proposed. The embodiment of the present application provides an automatic monitoring system and method for flue gas emissions from fixed pollution sources, which uses artificial intelligence technology based on the field of deep learning to extract and encode features from the multispectral image of the air at the input end of the sampling tube and the multispectral image of the air at the output end of the sampling tube to obtain a classification result of whether there is an abnormality in the sampling tube. In this way, by intelligently judging whether the sampling tube is loose or other abnormal conditions, the difficulty of manually checking the sampling tube is reduced, and the occurrence of CEMS data falsification caused by abnormalities in the sampling tube can be effectively avoided.

[0005] According to one aspect of the present application, there is provided an automatic monitoring system for flue gas emissions from a stationary pollution source, comprising:

[0006] A sampling tube gas data acquisition module is used to obtain a multispectral image of the air at the input end of the sampling tube and a multispectral image of the air at the output end of the sampling tube;

[0007] An air analysis module at the input end of the sampling tube is used to analyze a multispectral image of the air at the input end of the sampling tube to obtain a multi-scale input gas correlation feature map;

[0008] An air analysis module at the output end of the sampling tube is used to analyze a multispectral image of the air at the output end of the sampling tube to obtain a multi-scale output gas correlation feature map;

[0009] The sampling tube abnormality judgment module is used to comprehensively process the multi-scale input gas correlation characteristic map and the multi-scale output gas correlation characteristic map to obtain a result of whether the sampling tube has an abnormality.

[0010] According to another aspect of the present application, a method for automatically monitoring flue gas emissions from a stationary pollution source is provided, comprising:

[0011] Acquire a multispectral image of the air at the input end of the sampling tube and a multispectral image of the air at the output end of the sampling tube;

[0012] Analyzing a multispectral image of the air at the input end of the sampling tube to obtain a multi-scale input gas correlation feature map;

[0013] Analyzing a multispectral image of the air at the output end of the sampling tube to obtain a multi-scale output gas correlation feature map;

[0014] The multi-scale input gas correlation characteristic diagram and the multi-scale output gas correlation characteristic diagram are comprehensively processed to obtain a result of whether there is any abnormality in the sampling tube.

[0015] In summary, the automatic monitoring system and method for flue gas emissions from fixed pollution sources provided by this application uses artificial intelligence technology based on deep learning to extract and encode features from multispectral images of the air at the input end of the sampling tube and the multispectral images of the air at the output end of the sampling tube, thereby obtaining a classification result indicating whether the sampling tube has an abnormality. In this way, by intelligently determining whether the sampling tube is loose or has other abnormalities, the difficulty of manually troubleshooting the sampling tube is reduced, and the occurrence of CEMS data falsification caused by sampling tube anomalies can be effectively avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without inventive effort. In the drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 4 is a block diagram of an automatic monitoring system for flue gas emissions from stationary pollution sources according to an embodiment of the present application.

[0018] Figure 2 This is a block diagram of an air analysis module at the input end of a sampling tube in an automatic monitoring system for flue gas emissions from a stationary pollution source according to an embodiment of the present application.

[0019] Figure 3 4 is a block diagram of a multispectral image spatial feature enhancement unit in an automatic monitoring system for flue gas emissions from stationary pollution sources according to an embodiment of the present application.

[0020] Figure 4 Flowchart of a method for automatically monitoring flue gas emissions from stationary pollution sources according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] Below, the exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings, clearly and completely describing the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0022] Figure 1 FIG is a block diagram of an automatic monitoring system for flue gas emissions from fixed pollution sources according to an embodiment of the present application. Figure 1As shown, according to an embodiment of the present application, the automatic monitoring system 100 for flue gas emissions from fixed pollution sources includes: a sampling tube gas data acquisition module 110, which is used to obtain a multispectral image of the air at the input end of the sampling tube and a multispectral image of the air at the output end of the sampling tube; a sampling tube input end air analysis module 120, which is used to analyze the multispectral image of the air at the input end of the sampling tube to obtain a multi-scale input gas correlation feature map; a sampling tube output end air analysis module 130, which is used to analyze the multispectral image of the air at the output end of the sampling tube to obtain a multi-scale output gas correlation feature map; and a sampling tube abnormality judgment module 140, which is used to comprehensively process the multi-scale input gas correlation feature map and the multi-scale output gas correlation feature map to obtain a result of whether there is an abnormality in the sampling tube.

[0023] In the above-mentioned automatic monitoring system 100 for flue gas emissions from fixed pollution sources, the sampling tube gas data acquisition module 110 is used to obtain a multispectral image of the air at the input end of the sampling tube and a multispectral image of the air at the output end of the sampling tube. As mentioned in the above-mentioned background technology, due to economic benefits, some companies will falsify CEMS data. One typical case is that by loosening the sampling pipeline, the gas actually monitored is diluted gas, which deviates from the true value. However, since the environment in which the sampling tube is located is relatively harsh, it is difficult to simply check it manually. Therefore, an optimized automatic monitoring system for flue gas emissions from fixed pollution sources is expected.

[0024] To address these technical issues, we proposed an automated monitoring system for flue gas emissions from fixed pollution sources. This system uses deep learning-based artificial intelligence to extract and encode features from multispectral images of the air at the input and output ends of sampling tubes, thereby classifying whether the tubes are abnormal. This intelligent system reduces the difficulty of manually troubleshooting sampling tubes, effectively preventing CEMS data falsification caused by tube anomalies.

[0025] Specifically, the system first acquires multispectral images of the air at the sampling tube's input and output ends. This allows for comprehensive monitoring and analysis of potential anomalies during flue gas emissions, such as pollutant leaks, pipe blockages, or other issues. By comparing the multispectral images of the input and output ends, the system can detect changes and anomalies within the sampling tube, enabling timely repairs or adjustments to ensure accurate and reliable emission data.

[0026] Specifically, the methods for obtaining multispectral images include: 1. Multispectral camera: Using a camera device specially designed for multispectral image acquisition, it can simultaneously capture spectral information in multiple wavelength ranges. This camera can be installed near the sampling tube to obtain multispectral images of the air at the input and output ends. 2. Spectrometer: Another way is to use a spectrometer to obtain multispectral images. The spectrometer can analyze the wavelength and intensity of light to generate a multispectral image. 3. Sensor: In some cases, a multispectral sensor can be directly installed on the sampling tube to collect multispectral image data in real time. The obtained multispectral image needs to undergo appropriate processing and analysis, including noise reduction, feature extraction and other steps, so that the system can accurately identify and analyze the information in the image. By combining these methods, the system can obtain multispectral images of the air at the input and output ends of the sampling tube, providing necessary data support for subsequent analysis and monitoring.

[0027] In the above-mentioned automatic monitoring system 100 for flue gas emissions from stationary pollution sources, the sampling tube input end air analysis module 120 is used to analyze the multispectral image of the air at the sampling tube input end to obtain a multi-scale input gas correlation feature map.

[0028] Figure 2 FIG. 1 is a block diagram of an air analysis module at the input end of a sampling tube in an automatic monitoring system for flue gas emissions from a fixed pollution source according to an embodiment of the present application. Figure 2 As shown, the sampling tube input end air analysis module 120 includes: an air multispectral image denoising unit 121, used to perform denoising processing on the multispectral image of the air at the sampling tube input end to obtain a generated sampling tube input end multispectral cubemap; a multispectral image spatial feature enhancement unit 122, used to pass the generated sampling tube input end multispectral cubemap through a multispectral spatial feature extraction module to obtain a spatially enhanced input end multispectral feature map; an input end feature dual-stream detection unit 123, used to pass the spatially enhanced input end multispectral feature map through an input end feature dual-stream detection module including a first input end multispectral feature extractor and a second input end multispectral feature extractor to obtain a first-scale input multispectral feature map and a second-scale input multispectral feature map; and a multi-scale input gas correlation unit 124, used to fuse the first-scale input multispectral feature map and the second-scale input multispectral feature map to obtain a multi-scale input gas correlation feature map.

[0029] More specifically, to improve data quality and accuracy, the multispectral image of the air at the sampling tube input is processed through an automatic codec-based image denoiser to generate a multispectral cubemap of the sampling tube input. Considering that in real-world environments, multispectral images may be affected by various interferences and noise, such as lighting variations, equipment problems, or environmental disturbances, denoising can reduce the impact of these interferences on image quality and subsequent analysis, improving data reliability and accuracy. Furthermore, denoising helps improve image clarity and detail, making the image easier to analyze and understand. Generating the multispectral cubemap of the sampling tube input allows for better extraction of feature information, providing a better foundation for subsequent processing and analysis. The denoised image makes it easier to extract meaningful features, facilitating subsequent feature extraction and analysis. This improves the system's ability to understand and monitor air conditions at the input, enabling more accurate detection of potential anomalies. Furthermore, denoising improves the efficiency and accuracy of the system's image data processing, thereby enhancing the performance and reliability of the entire automatic monitoring system. This helps the system better cope with complex monitoring environments and various challenges.

[0030] Specifically, in the embodiment of the present application, the air multispectral image denoising unit 121 is used to: pass the multispectral image of the air at the input end of the sampling tube through an image denoiser based on an automatic codec to generate a multispectral cubemap at the input end of the sampling tube.

[0031] Specifically, in an embodiment of the present application, the air multispectral image denoising unit 121 includes: an encoding subunit, configured to input the multispectral image of the air at the input end of the sampling tube into an encoder of the image denoiser, wherein the encoder uses a convolution layer to perform explicit spatial encoding on the multispectral image of the air at the input end of the sampling tube to obtain multispectral image features; and a decoding subunit, configured to input the multispectral image features into a decoder of the image denoiser, wherein the decoder uses a deconvolution layer to perform deconvolution processing on the multispectral image features to obtain the generated multispectral cube map at the input end of the sampling tube.

[0032] Specifically, in an embodiment of the present application, the multispectral spatial feature extraction module is a convolutional neural network model using a spatial attention mechanism.

[0033] More specifically, the generated multispectral cubemap at the input of the sampling tube is processed by a multispectral spatial feature extraction module to obtain a spatially enhanced input multispectral feature map. The spatial attention mechanism helps the model better understand the spatial relationships and importance between different regions in the image. By introducing spatial attention, the model can focus on the features of different regions in the image, thereby better capturing the spatial information of the input multispectral image. Furthermore, the spatial attention mechanism helps the model focus on important spatial locations and structures in the image when learning feature representations, thereby improving the representational power and discriminability of features. Through spatial enhancement, the resulting multispectral feature map better reflects the characteristics and changes of the input air, providing richer information for subsequent analysis and decision-making. Furthermore, the spatial attention mechanism helps the model automatically learn which regions of the image are most important for the task, thereby reducing the impact of redundant information and improving the compactness and effectiveness of features. Convolutional neural network models that incorporate the spatial attention mechanism are generally better able to adapt to complex spatial features and structures, thereby improving model performance and generalization. This helps enhance the system's understanding and analysis of the input multispectral image, further improving the accuracy of monitoring and recognition.

[0034] Figure 3 FIG. 1 is a block diagram of a multispectral image spatial feature enhancement unit in an automatic monitoring system for flue gas emissions from stationary pollution sources according to an embodiment of the present application. Figure 3 As shown, the multispectral image spatial feature enhancement unit 122 includes: a deep convolution encoding subunit 11, which is used to use the convolution encoding part of the multispectral spatial feature extraction module to perform deep convolution encoding on the multispectral cube map at the input end of the generated sampling tube to obtain an initial convolution feature map; a spatial attention subunit 12, which is used to input the initial convolution feature map into the spatial attention part of the multispectral spatial feature extraction module to obtain a spatial attention map; an activation subunit 13, which is used to pass the spatial attention map through a Softmax activation function to obtain a spatial attention feature map; and a spatial attention application subunit 14, which is used to calculate the position point multiplication of the spatial attention feature map and the initial convolution feature map to obtain a multispectral feature map at the spatial enhancement input end.

[0035] More specifically, the spatially enhanced input multispectral feature map is processed by an input feature dual-stream detection module comprising a first input multispectral feature extractor and a second input multispectral feature extractor to obtain a first-scale input multispectral feature map and a second-scale input multispectral feature map. In the technical solution of the present application, the first input multispectral feature extractor and the second input multispectral feature extractor are first and second convolutional neural networks of different scales, and the input feature dual-stream detection module is a dual-stream detection network model. Using a dual-stream detection network model, feature information can be extracted simultaneously from different scales. The first and second convolutional neural networks can process the input feature maps separately, extracting feature information at different scales, thereby capturing richer and more diverse feature representations. At the same time, features at different scales may have different importance for different tasks and scenarios. By processing feature maps of different scales in the dual-stream detection network, the model can better adapt to feature requirements at different scales, improving the model's generalization and adaptability. Multi-scale feature extraction and fusion help improve the detection network's perception and recognition capabilities of targets. By combining feature information at different scales, we can more comprehensively understand the target information in the input multispectral feature map and improve the accuracy and robustness of detection.

[0036] Specifically, in the embodiment of the present application, the input-end feature dual-stream detection module is a dual-stream detection network model.

[0037] Specifically, in an embodiment of the present application, the first input-end multispectral feature extractor and the second input-end multispectral feature extractor are convolutional neural network models with different scales.

[0038] Specifically, in an embodiment of the present application, the input-end feature dual-stream detection unit 123 includes a first-scale input-end multispectral feature extraction unit, which is used to use the first input-end multispectral feature extractor with the first scale to perform the following on the input data in the forward pass of the layer: performing three-dimensional convolution processing, mean pooling processing and nonlinear activation processing on the input data to obtain a first feature map; and a second-scale input-end multispectral feature extraction unit, which is used to use the second input-end multispectral feature extractor with the second scale to perform the following on the input data in the forward pass of the layer: performing convolution processing, mean pooling processing and nonlinear activation processing on the input data to obtain a second feature map.

[0039] More specifically, the first-scale input multispectral feature map and the second-scale input multispectral feature map are fused to generate a multiscale input gas correlation feature map. Fusion of feature maps at different scales enables the model to obtain richer and more diverse feature representations. Multiscale feature fusion helps the system more comprehensively capture information at different scales, improving feature expression and discrimination. Combining feature information at different scales enhances feature robustness and stability. Multiscale feature fusion effectively reduces the impact of missing features or noise at a single scale on system performance, improving system stability and generalization. Furthermore, the multiscale input gas correlation feature map better reflects the gas-related features and correlation information in the input multispectral image. This helps improve the system's detection and recognition capabilities for gas targets, further enhancing system performance and accuracy. Fusion of multiscale feature maps comprehensively utilizes information at different scales, enabling the system to more comprehensively understand the gas-related features in the input multispectral image. This helps improve the system's understanding and analysis of gas targets, providing richer information for subsequent decision-making and application.

[0040] Specifically, in the embodiment of the present application, the multi-scale input gas correlation unit 124 is configured to fuse the first-scale input multi-spectral feature map and the second-scale input multi-spectral feature map using the following multi-scale correlation formula to obtain a multi-scale input gas correlation feature map; wherein the multi-scale correlation formula is:

[0041] F c =αF1+βF2

[0042] Among them, F c is the multi-scale input gas correlation feature map, F1 is the first-scale input multi-spectral feature map, F2 is the second-scale input multi-spectral feature map, "+" represents the addition of the elements at corresponding positions of the first-scale input multi-spectral feature map and the second-scale input multi-spectral feature map, α and β are weighting parameters used to control the balance between the first-scale input multi-spectral feature map and the second-scale input multi-spectral feature map in the multi-scale input gas correlation feature map.

[0043] In the aforementioned automatic monitoring system 100 for flue gas emissions from stationary pollution sources, the sampling tube output air analysis module 130 is configured to analyze a multispectral image of the air at the sampling tube output to obtain a multi-scale output gas correlation characteristic map. By analyzing the multispectral image of the air at the sampling tube output, the spectral characteristics of the gas at different wavelengths can be obtained. Specifically, the multispectral image of the air at the sampling tube output is processed in a manner similar to the processing steps for the multispectral image of the air at the sampling tube input.

[0044] In the above-mentioned automatic monitoring system 100 for flue gas emissions from fixed pollution sources, the sampling pipe abnormality judgment module 140 is used to comprehensively process the multi-scale input gas correlation characteristic diagram and the multi-scale output gas correlation characteristic diagram to obtain a result of whether there is an abnormality in the sampling pipe.

[0045] Specifically, in an embodiment of the present application, the sampling pipe abnormality judgment module 140 includes: a multi-scale gas difference unit, used to calculate the difference between the multi-scale input gas correlation characteristic map and the multi-scale output gas correlation characteristic map to obtain a multi-scale gas differential characteristic map; a characteristic map expansion unit, used to expand the multi-scale gas differential characteristic map into a multi-scale gas differential characteristic vector; a characteristic structure correction unit, used to perform characteristic fine-grained internal structure correction on the multi-scale gas differential characteristic vector based on eigenvalue decomposition to obtain a corrected multi-scale gas differential characteristic vector; and a sampling pipe abnormality judgment unit, used to pass the corrected multi-scale gas differential characteristic vector through a classifier to obtain a classification result, and the classification result is used to indicate whether there is an abnormality in the sampling pipe.

[0046] More specifically, the difference between the multi-scale input gas correlation feature map and the multi-scale output gas correlation feature map is calculated to obtain a multi-scale gas differential feature map. By calculating the difference between the multi-scale input and output gas correlation feature maps, the variation information of the gas at different scales can be highlighted. This helps to capture the differences between gas concentration, distribution, and other characteristics, and further extract the key features of gas variation. The multi-scale gas differential feature map contains the variation information between the input and output gas correlation feature maps, and can provide richer and more diverse feature representations. This helps to enhance the system's ability to represent the characteristics of gas targets and improve the system's detection and recognition performance.

[0047] Specifically, in an embodiment of the present application, the multi-scale gas difference unit is configured to calculate the difference between the multi-scale input gas correlation characteristic map and the multi-scale output gas correlation characteristic map using the following multi-scale gas difference formula to obtain a multi-scale gas difference characteristic map; wherein the multi-scale gas difference formula is:

[0048]

[0049] Among them, F a represents the multi-scale input gas correlation feature map, Indicates difference by position, F b represents the multi-scale output gas correlation feature map, and F n represents the multi-scale gas differential characteristic map.

[0050] More specifically, the multi-scale gas differential feature map is expanded into a multi-scale gas differential feature vector. This converts high-dimensional feature representations into low-dimensional feature vectors, thereby reducing data dimensionality. This helps reduce computational complexity, improves computational efficiency, and better adapts to certain algorithms.

[0051] In particular, in the technical solution of the present application, when using multi-scale gas differential eigenvectors to determine whether a fixed pollution source sampling tube is abnormal, two major problems are faced: internal feature information redundancy and noise interference, and effective representation of the complex internal structure of the high-dimensional feature vector. First, the original multi-scale gas differential eigenvector may have information redundancy, which increases the complexity of the model, increases the risk of overfitting, and reduces the generalization ability of the model. At the same time, noise interference will also affect the accuracy of the feature vector, resulting in deviations in the judgment results. Secondly, there are complex linear and nonlinear correlation structures between the dimensions within the multi-scale gas differential eigenvector, and these structures contain patterns that are crucial for judging the abnormality of the sampling tube. In order to solve this technical problem, in the technical solution of the present application, the multi-scale gas differential eigenvector is subjected to feature fine-grained internal structure correction based on eigenvalue decomposition to obtain a corrected multi-scale gas differential eigenvector.

[0052] Specifically, in the embodiment of the present application, the characteristic structure correction unit is used to calculate the global fine-grained autocorrelation topology matrix of the multi-scale gas differential characteristic vector, which is expressed as:

[0053]

[0054] M=D1⊙D2

[0055] v i ,v j ∈V

[0056] Where V represents the multi-scale gas differential eigenvector, v i and v j They represent the i-th and j-th eigenvalues of the multi-scale gas differential eigenvector, w1, w2, w3 and w4 represent different weight hyperparameters, ⊙ represents matrix dot product, D1 represents the forward weight matrix of the multi-scale gas differential feature, and D2 represents the reverse weight matrix of the multi-scale gas differential feature. Represents the value of the (i, j)th position of the multi-scale gas differential feature forward weighting matrix, represents the value of the (i, j)th position of the inverse weighted matrix of the multi-scale gas differential characteristics, and M represents the global fine-grained autocorrelation topology matrix.

[0057] Specifically, by calculating a global, fine-grained autocorrelation topology matrix, the fine-grained interactions and structural adjacency patterns of all dimensional features in the multiscale gas differential feature vector are explicitly captured. This quantifies the linear correlations or statistical dependencies within the components of the multiscale gas differential feature, providing an explicit mathematical analysis target for in-depth analysis of structural differences within the features and for determining sampling tube anomalies. The resulting global, fine-grained autocorrelation topology matrix provides a more discriminative structural feature basis for subsequent comprehensive processing of input and output features and for determining whether sampling tube anomalies exist, thereby improving the accuracy and reliability of anomaly judgments.

[0058] Specifically, the characteristic structure correction unit is further used to: perform eigenvalue decomposition on the global fine-grained autocorrelation topological matrix to obtain a set of multi-scale gas differential characteristic fine-grained characteristic component encoding vectors, which is expressed as follows:

[0059]

[0060] Where Λ represents a diagonal matrix, λ1 and λ m denote the first and mth eigenvalues of the diagonal matrix, respectively, (·) T represents the transpose of the vector, U represents the set of encoding vectors of fine-grained characteristic components of multi-scale gas differential characteristics, x1, x2 and x m They respectively represent the first, second and mth multi-scale gas differential feature fine-grained feature component encoding vectors in the set of multi-scale gas differential feature fine-grained feature component encoding vectors.

[0061] Specifically, by performing eigenvalue decomposition on the global fine-grained autocorrelation topological matrix, we obtain a set of encoding vectors for the fine-grained characteristic components of the multi-scale gas differential signature. This allows us to construct a new orthogonal basis guided by the internal structure of the data, representing the different independent modes of variation within the signature, effectively decoupling the complex internal structure of the signature. Specifically, the complex and interrelated characteristic structure is decomposed into a series of independent basic structural patterns ranked by importance, making the previously entangled characteristic information clear and organized, facilitating subsequent targeted adjustment and analysis of each independent structural aspect.

[0062] Specifically, the characteristic structure correction unit is further configured to: perform information compression on each multi-scale gas differential characteristic fine-grained characteristic component encoding vector in the set of multi-scale gas differential characteristic fine-grained characteristic component encoding vectors to obtain a set of multi-scale gas differential fine-grained characteristic component compressed encoding vectors, which is expressed as follows:

[0063]

[0064] Among them, x irepresents the i-th multi-scale gas differential feature fine-grained feature component encoding vector in the set of multi-scale gas differential feature fine-grained feature component encoding vectors, ||·|| represents the Euclidean norm, y i represents the compressed encoding vector of the i-th multi-scale gas differential fine-grained feature component.

[0065] Specifically, by compressing the information encoded in the fine-grained characteristic components of multi-scale gas differential signatures, higher-order dependencies beyond second-order statistics are captured, enhancing discriminative capabilities. Each decoupled structural component undergoes refined, independent nonlinear modulation, introducing model complexity and nonlinear processing capabilities. This breaks the purely linear framework, enhances the useful information of the fine-grained characteristic components of multi-scale gas differential signatures, suppresses noise, and makes the resulting compressed encoding vectors of these fine-grained characteristic components more representative and discriminative, thereby improving the accuracy and reliability of determining abnormalities in sampling tubes.

[0066] Specifically, the characteristic structure correction unit is further used to calculate the internal structure significant adjustment factor of each multi-scale gas difference fine granularity characteristic component compressed code vector in the set of multi-scale gas difference fine granularity characteristic component compressed code vectors to obtain a set of internal structure significant adjustment factors, which is expressed as follows:

[0067]

[0068] Among them, α and β represent different weight hyperparameters, ||·||1 represents the first norm, ||·||2 represents the second norm, L represents the length of the compressed encoding vector of the fine-grained characteristic component of multi-scale gas difference, a i represents y i The corresponding internal structure significant adjustment factor.

[0069] That is, a scalar value is calculated for each compressed coding vector of the multi-scale gas differential fine-grained characteristic component to quantify the value of the structural information it carries, and the current value of the compressed coding vector of each multi-scale gas differential fine-grained characteristic component is re-evaluated to provide a dynamic importance basis based on the modulated state for subsequent weighted fusion.

[0070] Specifically, the characteristic structure correction unit is further configured to normalize the set of internal structure significant adjustment factors to obtain a set of internal structure significant adjustment weight factors, which is expressed as follows:

[0071] w i =Softmax(a i )

[0072] Among them, Softmax(·) represents the normalization function, w i Indicates ai The corresponding internal structure significantly adjusts the weighting factors.

[0073] That is, by normalizing the set of significant internal structure adjustment factors, we generate a set of significant internal structure adjustment weight factors suitable for feature fusion. This controls the sharpness of the weight distribution, ensures that the weight distribution is reasonable and conforms to the law of probability distribution, and avoids extreme situations. This makes the subsequent fusion process more stable and controllable, and avoids judgment errors caused by numerical instability or extreme weight distribution.

[0074] Specifically, the characteristic structure correction unit is further configured to: based on the set of weight factors significantly adjusted by the internal structure, perform fine-grained integration on the set of multi-scale gas difference fine-grained characteristic component compression coding vectors to obtain a corrected multi-scale gas difference characteristic vector, which is expressed as:

[0075]

[0076] Where V' represents the corrected multi-scale gas difference eigenvector.

[0077] That is, by utilizing the internal structure to significantly adjust the weight factor, the compressed coding vector of the fine-grained feature component of the multi-scale gas difference is fine-grainedly integrated, and the corrected multi-scale gas differential feature vector is obtained, which concentrates the key structural information of the original feature and improves the characterization capability, providing better features for the subsequent accurate judgment of sampling tube anomalies.

[0078] More specifically, the corrected multi-scale gas differential feature vector is passed through a classifier to obtain a classification result, and then these classification results are used to indicate whether there is an abnormality in the sampling tube. By inputting the corrected multi-scale gas differential feature vector into the classifier for classification, abnormality detection of the gas output from the sampling tube can be achieved. The classifier can identify patterns and features that are different from normal conditions, thereby promptly detecting abnormalities in the sampling tube. By combining the corrected multi-scale gas differential feature vector with a pre-trained classifier, it is possible to automatically determine whether the gas output from the sampling tube is in a normal state, reducing manual intervention and monitoring costs. The classification results can provide real-time feedback, so that appropriate measures can be taken in a timely manner when an abnormality occurs in the sampling tube. Timely anomaly detection and response helps avoid potential safety risks and problems and ensure the normal operation and use of the system.

[0079] Specifically, in an embodiment of the present application, the sampling tube abnormality judgment unit is configured to: use the classifier to process the corrected multi-scale gas difference feature vector using the following classification formula to obtain the classification result; wherein the classification formula is:

[0080] O=softmax{(Wc ,B c )|Project(F)}

[0081] Among them, Project(F) represents the projection of the corrected multi-scale gas difference eigenvector into a vector, W c is the weight matrix, B c represents a bias vector, softmax represents a normalized exponential function, and O represents the classification result.

[0082] In summary, the automatic monitoring system for flue gas emissions from fixed pollution sources according to the embodiments of the present application has been explained. It uses artificial intelligence technology based on deep learning to extract and encode features from multispectral images of the air at the input and output ends of the sampling tubes to obtain a classification result indicating whether the sampling tubes are experiencing abnormalities. This intelligent determination of abnormalities such as loose sampling tubes reduces the difficulty of manually troubleshooting the sampling tubes, effectively preventing the falsification of CEMS data due to sampling tube anomalies.

[0083] As described above, the automatic monitoring system 100 for flue gas emissions from stationary pollution sources according to the embodiment of the present application can be implemented in various terminal devices, such as a server for automatic monitoring of flue gas emissions from stationary pollution sources. In one example, the automatic monitoring system 100 for flue gas emissions from stationary pollution sources according to the embodiment of the present application can be integrated into a terminal device as a software module and / or a hardware module. For example, the automatic monitoring system 100 for flue gas emissions from stationary pollution sources can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the automatic monitoring system 100 for flue gas emissions from stationary pollution sources can also be one of the many hardware modules of the terminal device.

[0084] Alternatively, in another example, the automatic monitoring system 100 for flue gas emissions from fixed pollution sources and the terminal device may also be separate devices, and the automatic monitoring system 100 for flue gas emissions from fixed pollution sources may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0085] Based on the same inventive concept, an embodiment of the present application also provides a method for automatically monitoring flue gas emissions from a fixed pollution source, which can be used to implement the system described in the above embodiment, as described in the following embodiment.

[0086] Figure 4 Flowchart of the method for automatically monitoring flue gas emissions from fixed pollution sources according to an embodiment of the present application. Figure 4As shown, the automatic monitoring method for flue gas emissions from fixed pollution sources according to the embodiment of the present application includes the steps of: S110, obtaining a multispectral image of the air at the input end of the sampling tube and a multispectral image of the air at the output end of the sampling tube; S120, analyzing the multispectral image of the air at the input end of the sampling tube to obtain a multi-scale input gas correlation feature map; S130, analyzing the multispectral image of the air at the output end of the sampling tube to obtain a multi-scale output gas correlation feature map; and S140, comprehensively processing the multi-scale input gas correlation feature map and the multi-scale output gas correlation feature map to obtain a result of whether there is any abnormality in the sampling tube.

[0087] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. Here, for the automatic monitoring method for flue gas emissions from fixed pollution sources disclosed in the embodiment, those skilled in the art will understand that the specific operations of each step in the automatic monitoring method for flue gas emissions from fixed pollution sources have been referred to above. Figures 1 to 3 It has been introduced in detail in the description of the automatic monitoring system for flue gas emissions from fixed pollution sources, so the description is relatively simple. For relevant details, please refer to the description of the automatic monitoring system for flue gas emissions from fixed pollution sources, and therefore, its repeated description will be omitted.

[0088] In summary, the method for automatically monitoring flue gas emissions from fixed pollution sources according to the embodiments of the present application has been explained. It uses artificial intelligence technology based on deep learning to extract and encode features from multispectral images of the air at the input and output ends of the sampling tubes to obtain a classification result indicating whether the sampling tubes are experiencing abnormalities. This intelligent determination of abnormalities such as loose sampling tubes reduces the difficulty of manually troubleshooting the sampling tubes, effectively preventing the falsification of CEMS data due to sampling tube anomalies.

[0089] In summary, after reading this detailed disclosure, those skilled in the art will appreciate that the foregoing detailed disclosure may be presented by way of example only and may not be limiting. Although not explicitly stated herein, those skilled in the art will understand that this application is intended to encompass various reasonable changes, improvements, and modifications to the embodiments. Such changes, improvements, and modifications are intended to be proposed by this application and are within the spirit and scope of the exemplary embodiments of this application.

[0090] In addition, certain terms in this application have been used to describe embodiments of the present application. For example, "one embodiment," "an embodiment," and / or "some embodiments" mean that a particular feature, structure, or characteristic described in conjunction with that embodiment may be included in at least one embodiment of the present application. Therefore, it is emphasized and should be understood that two or more references to "an embodiment," "one embodiment," or "an alternative embodiment" in various parts of this specification do not necessarily refer to the same embodiment. Furthermore, particular features, structures, or characteristics may be appropriately combined in one or more embodiments of the present application.

[0091] Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to encompass non-exclusive inclusion, such that an article or device comprising a list of elements includes not only those elements but also other elements not explicitly listed or inherent to such article or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the article or device comprising the aforementioned elements.

[0092] It should be understood that in the foregoing description of the embodiments of this application, in order to facilitate understanding of a feature and to simplify this application, this application combines various features into a single embodiment, figure, or description thereof. However, this does not mean that the combination of these features is required. When reading this application, it is entirely possible for those skilled in the art to extract some of the features and understand them as separate embodiments. In other words, the embodiments of this application can also be understood as the integration of multiple secondary embodiments. This also applies when the content of each secondary embodiment is less than all the features of a single aforementioned disclosed embodiment.

[0093] Finally, it should be understood that the embodiments of the application disclosed herein are illustrations of the principles of the embodiments of the present application. Other modified embodiments are also within the scope of the present application. Therefore, the embodiments disclosed in the present application are merely examples and not limitations. Those skilled in the art can adopt alternative configurations to implement the application in the present application based on the embodiments in the present application.

[0094] Therefore, the embodiments of the present application are not limited to the precise embodiments described in the application.

Claims

1. An automatic monitoring system for flue gas emissions from fixed pollution sources, characterized in that: include: A sampling tube gas data acquisition module is used to obtain a multispectral image of the air at the input end of the sampling tube and a multispectral image of the air at the output end of the sampling tube; An air analysis module at the input end of the sampling tube is used to analyze a multispectral image of the air at the input end of the sampling tube to obtain a multi-scale input gas correlation feature map; An air analysis module at the output end of the sampling tube is used to analyze a multispectral image of the air at the output end of the sampling tube to obtain a multi-scale output gas correlation feature map; The sampling tube abnormality judgment module is used to comprehensively process the multi-scale input gas correlation characteristic map and the multi-scale output gas correlation characteristic map to obtain a result of whether the sampling tube has an abnormality.

2. The automatic monitoring system for flue gas emissions from stationary pollution sources according to claim 1 is characterized in that: The air analysis module at the input end of the sampling tube includes: an air multispectral image noise reduction unit, configured to perform noise reduction processing on the multispectral image of the air at the input end of the sampling tube to generate a multispectral cube map at the input end of the sampling tube; A multispectral image spatial feature enhancement unit, configured to pass the generated sampling tube input end multispectral cube map through a multispectral spatial feature extraction module to obtain a spatial enhancement input end multispectral feature map; an input-end feature dual-stream detection unit, configured to pass the spatially enhanced input-end multispectral feature map through an input-end feature dual-stream detection module comprising a first input-end multispectral feature extractor and a second input-end multispectral feature extractor to obtain a first-scale input multispectral feature map and a second-scale input multispectral feature map; The multi-scale input gas correlation unit is configured to fuse the first-scale input multi-spectral feature map and the second-scale input multi-spectral feature map to obtain a multi-scale input gas correlation feature map.

3. The automatic monitoring system for flue gas emissions from stationary pollution sources according to claim 2 is characterized in that: The air multispectral image noise reduction unit is used to: The multispectral image of the air at the input end of the sampling tube is passed through an image denoiser based on an automatic codec to generate a multispectral cube map of the input end of the sampling tube.

4. The automatic monitoring system for flue gas emissions from stationary pollution sources according to claim 3 is characterized in that: The multispectral spatial feature extraction module is a convolutional neural network model using a spatial attention mechanism.

5. The automatic monitoring system for flue gas emissions from stationary pollution sources according to claim 4 is characterized in that: The input-end feature dual-stream detection module is a dual-stream detection network model.

6. The automatic monitoring system for flue gas emissions from stationary pollution sources according to claim 5 is characterized in that: The multi-scale input gas correlation unit is used to: The first-scale input multispectral feature map and the second-scale input multispectral feature map are fused using the following multiscale correlation formula to obtain a multiscale input gas correlation feature map; Wherein, the multi-scale association formula is: F c =αF1+βF2 Among them, F c is the multi-scale input gas correlation feature map, F1 is the first-scale input multi-spectral feature map, F2 is the second-scale input multi-spectral feature map, "+" represents the addition of elements at corresponding positions of the first-scale input multi-spectral feature map and the second-scale input multi-spectral feature map, α and β are weighting parameters used to control the balance between the first-scale input multi-spectral feature map and the second-scale input multi-spectral feature map in the multi-scale input gas correlation feature map.

7. The automatic monitoring system for flue gas emissions from stationary pollution sources according to claim 6 is characterized in that: The sampling tube abnormality judgment module includes: a multi-scale gas difference unit, configured to calculate a difference between the multi-scale input gas correlation characteristic map and the multi-scale output gas correlation characteristic map to obtain a multi-scale gas difference characteristic map; A feature map expansion unit, configured to expand the multi-scale gas difference feature map into a multi-scale gas difference feature vector; a characteristic structure correction unit, configured to perform characteristic fine-grained internal structure correction on the multi-scale gas differential characteristic vector based on eigenvalue decomposition to obtain a corrected multi-scale gas differential characteristic vector; The sampling pipe abnormality judgment unit is used to pass the corrected multi-scale gas differential feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether there is an abnormality in the sampling pipe.

8. The automatic monitoring system for flue gas emissions from stationary pollution sources according to claim 7 is characterized in that: The characteristic structure correction unit is used to: Calculating a global fine-grained autocorrelation topology matrix of the multi-scale gas differential eigenvector; Performing eigenvalue decomposition on the global fine-grained autocorrelation topological matrix to obtain a set of multi-scale gas differential feature fine-grained feature component encoding vectors; performing information compression on each multi-scale gas differential feature fine-grained feature component encoding vector in the set of multi-scale gas differential feature fine-grained feature component encoding vectors to obtain a set of multi-scale gas differential feature fine-grained feature component compressed encoding vectors; Calculating an internal structure significant adjustment factor of each multi-scale gas difference fine-grained characteristic component compression encoding vector in the set of multi-scale gas difference fine-grained characteristic component compression encoding vectors to obtain a set of internal structure significant adjustment factors; Normalizing the set of internal structure significant adjustment factors to obtain a set of internal structure significant adjustment weight factors; Based on the set of weight factors significantly adjusted by the internal structure, the set of multi-scale gas difference fine-grained feature component compression encoding vectors is fine-grainedly integrated to obtain a corrected multi-scale gas difference feature vector.

9. A method for automatically monitoring flue gas emissions from fixed pollution sources, characterized in that: include: Acquire a multispectral image of the air at the input end of the sampling tube and a multispectral image of the air at the output end of the sampling tube; Analyzing a multispectral image of the air at the input end of the sampling tube to obtain a multi-scale input gas correlation feature map; Analyzing a multispectral image of the air at the output end of the sampling tube to obtain a multi-scale output gas correlation feature map; The multi-scale input gas correlation characteristic diagram and the multi-scale output gas correlation characteristic diagram are comprehensively processed to obtain a result of whether there is any abnormality in the sampling tube.

Images