A method, medium, and equipment for detecting NOx emissions from combustion using an image-based approach.

Abstract

This invention proposes an image-based method, medium, and device for detecting NOx emissions from combustion, comprising: acquiring flame images of a furnace burner and preprocessing them using a bilateral filtering noise reduction algorithm; establishing an empirical formula to extract statistical features from the flame images; establishing a convolutional autoencoder to extract deep features from the flame images; using principal component analysis to reduce the dimensionality of the flame images and obtain principal component features; concatenating the statistical features, deep features, and principal component features into a comprehensive feature; establishing a fully connected autoencoder to extract high-level features from the comprehensive feature; inputting the extracted high-level features into a long short-term memory network to extract temporal features; introducing a self-attention mechanism to process the temporal features; and finally predicting and outputting the combustion NOx concentration. This invention can more accurately describe the combustion state of the flame, improve the ability to capture data change trends, enhance the generalization and adaptability of the model, and achieve accurate monitoring and prediction of combustion NOx emissions.

Description

Technical Field

[0001] This invention belongs to the field of combustion emission monitoring technology, specifically relating to an image-based method, medium, and device for detecting NOx emissions from combustion. Background Technology

[0002] Nitrogen oxides (NOx) are common pollutants in combustion processes, seriously threatening the ecological environment and human health. Timely and accurate access to NOx emission information from combustion provides a crucial data foundation for developing combustion optimization strategies and implementing flue gas treatment solutions. Traditional NOx emission monitoring primarily relies on gas analyzers, but due to high maintenance costs and frequent offline calibrations, these instruments often struggle to achieve economical and continuous measurements. In contrast, soft measurement methods are favored due to their ease of maintenance and low operating costs. Practice has shown that data-driven models can predict NOx emission concentrations using only combustion process data without requiring knowledge of the combustion mechanism, offering advantages such as high computational efficiency and strong generalization performance. However, NOx formation during combustion is influenced by various factors, making the accurate establishment of a nonlinear mapping between flame images and NOx emission concentrations a significant challenge. Therefore, an efficient NOx emission detection method is urgently needed to improve existing detection schemes and achieve high-precision monitoring of NOx emissions from combustion. Summary of the Invention

[0003] This invention addresses the shortcomings of existing technologies by providing an image-based method, medium, and device for detecting NOx emissions from combustion.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A method for detecting NOx emissions from combustion using an image-based approach, characterized by comprising the following steps:

[0006] Images of the furnace burner flames were acquired and preprocessed using a bilateral filtering noise reduction algorithm;

[0007] Establish empirical formulas to extract statistical features from flame images;

[0008] Convolutional autoencoder CAE is established to extract deep features from flame images;

[0009] Principal component analysis was used to reduce the dimensionality of the flame images to obtain principal component features;

[0010] Statistical features, deep features, and principal component features are combined to form comprehensive features;

[0011] Establish a fully connected autoencoder FAE to extract high-level features of the comprehensive features;

[0012] The extracted high-level features are input into a Long Short-Term Memory (LSTM) network to extract temporal features;

[0013] A self-attention mechanism is introduced to handle time features, and the output combustion NOx concentration is ultimately predicted.

[0014] To optimize the above technical solution, the specific measures also include:

[0015] Furthermore, the bilateral filtering noise reduction algorithm is expressed as follows:

[0016] ;

[0017] In the formula, Indicates the position of the filtered flame image. pixel values, Indicates the original flame image at location pixel values, Indicates A local neighborhood centered on the center Represents the normalization coefficient. The spatial Gaussian kernel parameters are initialized to constants and are expressed through... Dynamic adjustment, among which Indicates the width and height of the image. Indicates the reference resolution. ,in Indicates the intensity Gaussian kernel parameter. Indicates position The local variance, This represents the adjustment factor.

[0018] Furthermore, the statistical features include mean gray level, flame brightness, effective flame area, gray level standard deviation, flame skewness, and flame kurtosis.

[0019] Furthermore, the convolutional autoencoder CAE consists of an encoder and a decoder, and the specific implementation process is as follows:

[0020] Flame image The encoder e1 processes the features, specifically including: first, feature extraction is performed by Q1 convolutional filters C1 with c×c windows and q strides; then, feature activation is performed by the ReLU activation function; finally, feature dimensionality reduction is performed by a max-pooling layer P1 with o×o windows and l strides, outputting convolutional features with dimensions a1×b1×c1. ;

[0021] The convolutional feature h1 is processed sequentially by encoders e2, e3, and e4. The data processing procedures for encoders e2, e3, and e4 are the same as those for encoder e1, and the final output is a deep feature with dimensions a4×b4×c4. ;

[0022] Deep features The data is processed by decoder d1, specifically including: first, data dimensionality enhancement is performed by an upsampling layer U1 with a window size of g×g; then, feature extraction is performed by Q5 convolutional filters with a window size of c×c and a stride of q; finally, feature activation is performed by the ReLU activation function, outputting convolutional features with dimensions a3×b3×c3. ;

[0023] Convolutional features The image is processed sequentially by encoders d2, d3, and d4. The data processing procedures for encoders d2, d3, and d4 are the same as those for decoder d1, and the final output is a reconstructed image with dimensions O×P×3. .

[0024] Furthermore, the convolutional autoencoder CAE employs the following loss function during training:

[0025] ;

[0026] ;

[0027] ;

[0028] ;

[0029] In the formula, Represents the loss function. This represents the mean square error. Indicates peak signal-to-noise ratio. Indicates edge preservation loss, Indicates the number of training images. and They represent the first Zhang inputs the flame image and the reconstructed image. This represents the maximum number of gray levels in the image. This represents the gradient operator. Indicates the image index position. Indicates the location pixel values, This represents the square of the Euclidean norm.

[0030] Furthermore, the fully connected self-encoding FAE consists of an input layer, a hidden layer, and an output layer, and the specific implementation process is as follows:

[0031] Dimensional comprehensive features Mapped to via a fully connected encoder Dimensional high-level features The mapping process is represented as ,in , Represents the weight matrix. Indicates the bias parameter. Represents the Sigmoid function;

[0032] Dimensional high-level features Reconstructed by a fully connected decoder Dimensional Reconstruction Features The reconstruction process is represented as ,in This represents the weight matrix connecting the hidden layer and the output layer. This represents the bias parameter.

[0033] Furthermore, the Long Short-Term Memory (LSTM) network consists of an input gate i, a forget gate f, and an output gate o, and the specific calculation process is as follows:

[0034] ;

[0035] ;

[0036] ;

[0037] ;

[0038] ;

[0039] ;

[0040] In the formula, This represents the high-level features at time t. This represents the temporal characteristics at time t. This represents the temporal characteristics at time t-1. , and Let represent the activation vectors of the input gate, forget gate, and output gate, respectively. Represents a hyperbolic sine function. Indicates the candidate storage state. Indicates the internal state; , , , and , , , Both represent weight matrices. , , , Indicates bias. This represents the dot product of matrices.

[0041] Furthermore, the introduction of a self-attention mechanism to process time features and ultimately predict the output combustion NOx concentration is as follows:

[0042] ;

[0043] ;

[0044] ;

[0045] In the formula, the attention score function Indicates in The correlation between the output vector of the Long Short-Term Memory (LSTM) network and the NOx concentration from combustion. and All represent weighting coefficients. Indicates the bias term; This represents the probability distribution of the attention mechanism output to the hidden layer of the Long Short-Term Memory (LSTM) network; express The output of the attention layer at each moment is the final predicted NOx concentration from combustion.

[0046] Accordingly, the present invention provides a computer-readable storage medium storing a computer program, characterized in that the computer program causes a computer to execute the image-based combustion NOx emission detection method as described above.

[0047] Accordingly, the present invention provides an electronic device, characterized in that it includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the image-based combustion NOx emission detection method as described above.

[0048] The beneficial effects of this invention are as follows: The convolutional autoencoder CAE and fully connected autoencoder FAE established in this invention can extract deep and high-level features from flame images, thereby more accurately describing the combustion state of the flame; the long short-term memory network LSTM and self-attention mechanism introduced in this invention significantly enhance the processing capability of time series data and greatly improve the ability to capture data change trends; this invention achieves accurate monitoring and prediction of NOx emissions from combustion by integrating a variety of improved image analysis and machine learning techniques; at the same time, this invention also effectively improves the generalization ability and adaptability to complex and changing environments by combining static and dynamic features. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the structure of the convolutional autoencoder CAE of this invention.

[0050] Figure 2 This is a schematic diagram of the structure of the fully connected self-encoding FAE of the present invention.

[0051] Figure 3 This is a schematic diagram of the Long Short-Term Memory (LSTM) network structure of the present invention.

[0052] Figure 4 This is a schematic diagram of the self-attention mechanism of the present invention.

[0053] Figure 5 These are furnace flame images under eight combustion conditions according to the present invention.

[0054] Figure 6 This is the predicted NOx concentration from combustion according to the present invention. Detailed Implementation

[0055] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0056] In one embodiment, the present invention proposes an image-based method for detecting NOx emissions from combustion, which specifically includes the following steps.

[0057] S1: Acquire images of the furnace burner flame and preprocess them using a bilateral filtering noise reduction algorithm.

[0058] In this embodiment, the bilateral filtering algorithm is expressed as:

[0059] ;

[0060] In the formula, Indicates the position after filtering (like Location, Indicates the row where the pixel is located. The pixel value (representing the column in which the pixel is located). Indicates the original image at position pixel values, Indicates A local neighborhood centered on the center This represents the normalization coefficient. The spatial Gaussian kernel parameters are initialized to constants and are expressed through... Dynamic adjustment, among which Indicates the width and height of the image. Indicates the reference resolution; The intensity Gaussian kernel parameter, adjusted based on local variance, is defined as follows: ,in This represents the original intensity Gaussian kernel parameters (which are set to a fixed constant by default). Indicates position The local variance, This represents the adjustment factor (usually set to 0.1).

[0061] In this embodiment, 8000 images of the furnace flame were acquired, representing eight different combustion conditions, such as... Figure 5 As shown, a bilateral filtering noise reduction algorithm is used to preprocess it.

[0062] S2: Establish six empirical formulas to extract statistical features from flame images. .

[0063] In this embodiment, the six statistical features are gray mean, flame brightness, effective flame area, gray standard deviation, flame skewness, and flame kurtosis, and the calculation formulas are shown in Table 1.

[0064] Table 1. Empirical formulas for six statistical characteristics

[0065]

[0066] S3: Establish a convolutional autoencoder (CAE) to extract deep features from flame images. .

[0067] In this embodiment, the structure of the convolutional autoencoder CAE is as follows: Figure 1 As shown, it consists of an encoder and a decoder, and the specific implementation process is as follows:

[0068] Flame image The encoder e1 processes the features, specifically including: first, feature extraction is performed by Q1 convolutional filters C1 with c×c windows and q strides; then, feature activation is performed by the ReLU activation function; finally, feature dimensionality reduction is performed by a max-pooling layer P1 with o×o windows and l strides, outputting convolutional features with dimensions a1×b1×c1. ;

[0069] The convolutional feature h1 is processed sequentially by encoders e2, e3, and e4. The data processing process is similar to that of encoder e1, and the final output is a deep feature with dimensions a4×b4×c4. ;

[0070] Deep features The data is processed by decoder d1, specifically including: first, data dimensionality enhancement is performed by an upsampling layer U1 with a window size of g×g; then, feature extraction is performed by Q5 convolutional filters with a window size of c×c and a stride of q; finally, feature activation is performed by the ReLU activation function, outputting convolutional features with dimensions a3×b3×c3. ;

[0071] Convolutional features The data is processed sequentially by encoders d2, d3, and d4, with the data processing procedure similar to that of decoder d1. The final output is a reconstructed image with dimensions O×P×3. .

[0072] In this embodiment, the loss function used during the training of the convolutional autoencoder CAE is:

[0073] ;

[0074] Mean square error Peak signal-to-noise ratio Edge preservation loss In the formula Indicates the number of training images. and They represent the first Zhang input image and reconstructed image, This indicates the maximum number of gray levels in the image (usually set to 255). This represents the gradient operator. Indicates the image index position. Indicates the location pixel values, This represents the square of the Euclidean norm.

[0075] S4: Principal component analysis is used to reduce the dimensionality of the flame image to obtain principal component features. .

[0076] S5: Statistical characteristics Deep features Principal component characteristics By splicing together, a comprehensive feature is formed. .

[0077] S6: Establish a fully connected autoencoder (FAE) and extract comprehensive features. Advanced features .

[0078] In this embodiment, the structure of the fully connected self-encoding FAE is as follows: Figure 2 As shown, it consists of an input layer, a hidden layer, and an output layer. The specific implementation process is as follows:

[0079] Dimensional comprehensive features Mapped to via a fully connected encoder Dimensional high-level features The mapping process is represented as ,in , Represents the weight matrix. Indicates the bias parameter. ;

[0080] Dimensional high-level features Reconstructed by a fully connected decoder Dimensional Reconstruction Features The reconstruction process is represented as ,in This represents the weight matrix connecting the hidden layer and the output layer. This represents the bias parameter.

[0081] S7: Extracted high-level features Input is fed into a Long Short-Term Memory (LSTM) network to extract temporal features. .

[0082] In this embodiment, the structure of the Long Short-Term Memory (LSTM) network is as follows: Figure 3 As shown, it consists of input gate i, forget gate f, and output gate o. The specific calculation process is as follows:

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] In the formula, This represents the high-level features at time t. This represents the temporal characteristics at time t. This represents the temporal characteristics at time t-1. , and Let represent the activation vectors of the input gate, forget gate, and output gate, respectively. Represents a hyperbolic sine function. Indicates the candidate storage state. Indicates the internal state; and Both represent weight matrices. Indicates bias. This represents the dot product of matrices.

[0090] S8: Introducing a self-attention mechanism to handle temporal features The final output is the NOx concentration from combustion.

[0091] In step S8, the structure of the self-attention mechanism is as follows: Figure 4As shown, the calculation process for predicting NOx concentration during combustion is as follows:

[0092] ;

[0093] ;

[0094] ;

[0095] In the formula, the attention score function Indicates in The correlation between the output vector of the Long Short-Term Memory (LSTM) network and the NOx concentration from combustion. and All represent weighting coefficients. Indicates the bias term; This represents the probability distribution of the attention mechanism output to the hidden layer of the Long Short-Term Memory network; express The output of the attention layer at any given time is the final predicted NOx concentration from combustion.

[0096] S9: Validate the accuracy of NOx concentration predictions using a test set, employing mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination. Evaluate the effectiveness of the practice.

[0097] In this embodiment, , , ,in Indicates the number of test samples. Indicates the actual value. This represents the predicted value.

[0098] In this embodiment, the predicted NOx concentration from combustion is as follows: Figure 6 As shown in the figure, the prediction results of the image-based combustion NOx emission detection method proposed in this embodiment are completely consistent with the actual observed values, thus achieving accurate monitoring and prediction of combustion NOx emissions.

[0099] In another embodiment, the present invention provides a computer-readable storage medium storing a computer program that causes a computer to perform the image-based combustion NOx emission detection method as described in Embodiment 1.

[0100] In another embodiment, the present invention provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the image-based combustion NOx emission detection method as described in Embodiment 1.

[0101] In the embodiments disclosed in this application, a computer storage medium may be a tangible medium that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CDROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0102] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0103] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method of image-based combustion NOx emission detection, characterized by, It comprises the following steps: Collecting a furnace burner flame image and pre-processing it by using a bilateral filter denoising algorithm; Establishing an empirical formula to extract statistical features of the flame image; the statistical features include mean gray value, flame brightness, flame effective area, gray standard deviation, flame skewness and flame kurtosis; Establishing a convolutional autoencoder (CAE) to extract deep features of the flame image; Using principal component analysis to reduce the dimension of the flame image to obtain principal component features; Splicing the statistical features, deep features and principal component features to form comprehensive features; Establishing a fully connected autoencoder (FAE) to extract high-level features of the comprehensive features; Inputting the extracted high-level features into a long short-term memory network (LSTM) to extract time features; Introducing a self-attention mechanism to process the time features to finally predict the output combustion NOx concentration.

2. A method of image-based combustion NOx emission sensing according to claim 1, characterized in that: The bilateral filter denoising algorithm is represented as: ； wherein denotes the pixel value of the filtered flame image at position , denotes the pixel value of the original flame image at position , denotes a local neighborhood centered at , denotes a normalization coefficient, denotes a spatial Gaussian kernel parameter, initially set to a constant value and dynamically adjusted by wherein denotes the width and height of the image, denotes a reference resolution, wherein denotes an intensity Gaussian kernel parameter, denotes a local variance at position , denotes an adjustment coefficient.

3. A method of image-based combustion NOx emission sensing according to claim 2, wherein: The convolutional autoencoder (CAE) is composed of an encoder and a decoder, and the specific implementation process is as follows: Flame image The encoder e1 processes, specifically including: first, feature extraction by a convolutional filter C1 of Q1 c x c window size, q step length, then feature activation by a ReLU activation function, and finally, feature dimension reduction by a max-pooling layer P1 of o x o window size, l step length, and the output dimension is a1 x b1 x c1 convolutional features ; The convolution feature h1 is processed by the encoders e2, e3, and e4 in sequence, the data processing processes of the encoders e2, e3, and e4 are the same as that of the encoder e1, and finally the deep feature with the output dimension of a4×b4×c4 is output ; Deep features After decoder d1 processing, specifically including: first dimensionality increasing by up-sampling layer U1 with g x g window size, then feature extraction by convolution filter C5 with q step size and c x c window size, finally feature activation by ReLU activation function, outputting convolution features with dimension a3 x b3 x c3 ; Convolutional features The reconstructed image with the final output dimension of OXP3 is outputted after sequentially processing the encoders d2, d3, d4, the data processing processes of which are the same as the decoder d1 .

4. A method of image-based combustion NOx emission sensing according to claim 3, wherein: The convolutional autoencoder (CAE) uses the following loss function in the training process: ； ； ； ； In the formula, Represents the loss function. This represents the mean square error. Indicates peak signal-to-noise ratio. This indicates edge preservation loss. Indicates the number of training images. and They represent the first Zhang inputs the flame image and the reconstructed image. This represents the maximum number of gray levels in the image. This represents the gradient operator. ( () indicates the image index position. Indicates the location pixel values, This represents the square of the Euclidean norm.

5. A method of image-based combustion NOx emission sensing according to claim 4, characterized in that: The fully connected autoencoder (FAE) is composed of an input layer, a hidden layer and an output layer, and the specific implementation process is as follows: comprehensive features mapped by a fully connected encoder high-level features , the mapping process is represented as wherein , represents a weight matrix, represents a bias parameter, represents a Sigmoid function; higher dimensional features reconstructed by a fully connected decoder reconstructed features , the reconstruction process is represented as where W represents a weight matrix connecting the hidden layer to the output layer, b represents a bias parameter.

6. A method of image-based combustion NOx emission sensing according to claim 5, characterized in that: The long short-term memory network (LSTM) is composed of an input gate i, a forget gate f and an output gate o, and the specific calculation process is as follows: ； ； ； ； ； ； wherein, denotes the high-level feature at time t, denotes the time feature at time t, denotes the time feature at time t-1, and denote the activation vectors of the input gate, the forget gate and the output gate, respectively, denotes the hyperbolic sine function, denotes the candidate memory state, denotes the internal state; and all denote weight matrices, denotes a bias, denotes the dot product of matrices.​​​​​​​​​​ 7. A method of image-based combustion NOx emission sensing according to claim 6, characterized in that: The long short-term memory network (LSTM) is composed of an input gate i, a forget gate f and an output gate o, and the specific calculation process is as follows: ； ； ； wherein, the attention score function represents the correlation between the output vector of the long short-term memory network LSTM at the time and the concentration of combustion NOx, and represent the weight coefficients, represents the bias term; represents the probability distribution of the attention mechanism output to the long short-term memory network LSTM hidden layer; represents the output of the attention layer at the time, which is the final predicted concentration of combustion NOx.

8. A computer readable storage medium storing a computer program, characterized in that, The long short-term memory network (LSTM) is composed of an input gate i, a forget gate f and an output gate o, and the specific calculation process is as follows:

9. An electronic device, comprising: The computer program enables the computer to perform the image-based combustion NOx emission detection method according to any one of claims 1-7. It comprises: A memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the image-based combustion NOx emission detection method according to any one of claims 1-7.

Images