Method and system for inverting composition and concentration of complex high-temperature flue gas in boiler

Abstract

The present invention provides a method and system for inverting the composition and concentration of complex high-temperature boiler flue gas, belonging to the field of thermal engineering technology. The method includes: obtaining the absorption spectrum data of the flue gas of the boiler to be measured; inputting the absorption spectrum data of the flue gas of the boiler to be measured into a multi-task deep learning model to obtain the composition and concentration parameters of the boiler flue gas; wherein, the multi-task deep learning model is established based on the absorption spectrum data of complex boiler flue gas at different temperatures and different concentrations. A deep learning algorithm is constructed based on the absorption spectrum data of complex boiler flue gas at different temperatures and different concentrations, and the absorption spectrum data of the actual measured boiler flue gas is quickly and accurately inverted through the constructed deep learning algorithm to obtain the main components and concentration parameters of the boiler flue gas, which can effectively cover the range of the composition and concentration of complex high-temperature boiler flue gas and reduce redundant data.

Description

Technical Field

[0001] The present invention relates to the field of thermal engineering technologies, and particularly to a method for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectroscopy, and a system for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectroscopy. Background Art

[0002] The components of boiler flue gas mainly include CO 2 , H 2 O, CO, O 2 , NO, NO 2 , SO 2 , H 2 S, etc. Among them, the corrosion of H 2 S gas is the main cause of high-temperature corrosion of the water wall, and the concentrations of CO and O 2 gases are related to the high-temperature corrosion of the water wall. The measurement of H 2 S, CO, O 2 and other gases has important reference value for studying the corrosion mechanism of the boiler water wall and predicting the high-temperature corrosion of the boiler water wall. In addition, gases such as NO, NO 2 , SO 2 are the main pollutants discharged from the boiler, and gases such as CO 2 , H 2 O are important data for evaluating the operation of the boiler. Therefore, the measurement of boiler flue gas components is one of the important contents of boiler thermal engineering measurement.

[0003] At present, the components of boiler flue gas are mainly analyzed by a flue gas analyzer composed of different gas electrochemical sensors to obtain the composition and concentration of various flue gases. Due to the cross-interference of various flue gas components by the electrochemical method, the measurement accuracy is low. The absorption spectroscopy technology utilizes a certain quantitative relationship between the concentration of the gas to be measured and the attenuation degree of the light intensity in a specific wavelength band. Through this mathematical relationship, the gas to be measured can be conveniently determined and the concentration value of the gas to be measured can be monitored. It has the advantages of fast response speed, on-line measurement, convenient maintenance, real-time monitoring, strong anti-cross-interference ability, etc., and the measurement accuracy can be effectively improved by combining with a multi-reflection absorption cell. However, the absorption spectroscopy technology currently used for flue gas analysis usually analyzes a single gas and does not consider the influence of the coexistence of complex multiple gases on the measurement of the concentration of a single gas. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a method and system for inverting the composition and concentration of complex high-temperature boiler flue gas. A multi-task deep learning model is established through the absorption spectra of complex boiler flue gas at different temperatures and different concentrations, and the absorption spectrum data of the measured boiler flue gas is quickly and accurately inverted through a deep learning algorithm to obtain the main components and concentration parameters of various boiler flue gases.

[0005] To achieve the above object, a first aspect of the present invention provides a method for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectroscopy. The complex high-temperature boiler flue gas mainly includes CO 2 , H 2 O, CO, O 2 , NO, NO 2 , SO 2 , H 2 S, etc. The method includes:

[0006] Obtain the absorption spectroscopy data of the boiler flue gas to be measured;

[0007] Input the absorption spectroscopy data of the boiler flue gas to be measured into a multi-task deep learning model to obtain the composition and concentration parameters of the boiler flue gas;

[0008] Among them, the multi-task deep learning model is established based on the absorption spectroscopy data of complex boiler flue gas at different temperatures and different concentrations.

[0009] Optionally, the absorption spectroscopy data of complex boiler flue gas at different temperatures and different concentrations are obtained through the following steps:

[0010] Taking the temperature, main gas components and concentration parameters of the complex high-temperature boiler flue gas as independent variables, calculate the optical depth OD of complex boiler flue gas at different temperatures and different concentrations according to the absorption spectroscopy principle;

[0011] Draw the absorption spectroscopy data of complex boiler flue gas at the same temperature and the same concentration according to the optical depth OD of different bands of complex boiler flue gas at the same temperature and the same concentration;

[0012] Change the temperature, main gas components and concentration parameters of the complex high-temperature boiler flue gas, and draw the absorption spectroscopy data of complex boiler flue gas at different temperatures and different concentrations.

[0013] Optionally, store the absorption spectroscopy data of complex boiler flue gas at different temperatures and different concentrations in an absorption spectroscopy library.

[0014] Optionally, the absorption spectroscopy principle is:

[0015] For a single gas, when irradiated with light of different wavelengths, the intensity of the incident light and the transmitted light conforms to the following relationship:

[0016] I(λ) = I 0 (λ)exp[-σ(λ)CL] (1)

[0017] In formula (1), I(λ) is the transmitted light intensity of the light with wavelength λ emitted by the light source passing through the gas to be measured, I 0$I_0(\lambda)$ is the initial light intensity of the light with wavelength $\lambda$ emitted by the light source, $L$ is the optical path of the gas, $C$ is the concentration of the gas to be measured, and $\sigma(\lambda)$ is the absorption cross-section of the gas to be measured, which is related to the wavelength $\lambda$;

[0018] According to the principle that the light attenuation of the complex high-temperature flue gas in the boiler includes the extinction caused by Rayleigh scattering and Mie scattering, formula (1) is changed to:

[0019]

[0020] In formula (2), $\varepsilon$ R is the Rayleigh scattering extinction coefficient, and $\varepsilon$ M is the Mie scattering extinction coefficient;

[0021] Since the light emitted by the light source is absorbed by various gas components in the complex high-temperature flue gas of the boiler, considering the influence of other gas molecules and particles, formula (2) is expressed as:

[0022]

[0023] In formula (3), $\sigma$ i and $C$ i respectively represent the absorption cross-section and concentration of the $i$-th gas that absorbs light;

[0024] According to the "slow change" and "fast change" principles of gas absorption, the absorption cross-section $\sigma(\lambda)$ of the gas to be measured is expressed as:

[0025] $\sigma(\lambda)=\sigma$ s $(\lambda)+\sigma$ f $(\lambda)\ (4)$

[0026] In formula (4), $\sigma$ s $(\lambda)$ is the slow-varying absorption cross-section of gas absorption, and $\sigma$ f $(\lambda)$ is the fast-varying absorption cross-section of gas absorption;

[0027] According to the "slow change" and "fast change" principles in gas absorption, broadband absorption and narrowband absorption are obtained:

[0028]

[0029] The optical thickness OD at different wavelengths $\lambda$ is defined as:

[0030]

[0031] In formula (6), is the broadband absorption, that is, the broadband absorption caused by broadband absorption and scattering in gas absorption; is the narrowband absorption;

[0032] Considering the influence of temperature on the optical attenuation of complex high-temperature flue gas in boilers, the optical thickness OD is corrected for temperature according to Equation (7):

[0033]

[0034] The accurate optical thickness of complex high-temperature flue gas in boilers can be calculated through the above method.

[0035] Optionally, the multi-task deep learning model is established through the following steps:

[0036] Using the spectral two-dimensional image conversion algorithm, convert the absorption spectral data in the absorption spectral library into spectral images to obtain a two-dimensional spectral information matrix;

[0037] Construct a multi-task deep learning model based on the two-dimensional spectral information matrix.

[0038] Optionally, the step of using the spectral two-dimensional image conversion algorithm to convert the absorption spectral data in the absorption spectral library into spectral images to obtain a two-dimensional spectral information matrix includes:

[0039] Process the absorption spectral data according to Equation (8):

[0040] S = XX T (8)

[0041] In Equation (8), S is the two-dimensional spectral information matrix, and X is the spectral data column vector. A typical two-dimensional spectral information matrix can be obtained:

[0042]

[0043] In Equation (9), a i is one-dimensional spectral data.

[0044] Optionally, the step of constructing a multi-task deep learning model based on the two-dimensional spectral information matrix includes:

[0045] Construct a common layer including multiple convolutional kernels and a max pooling layer;

[0046] Construct convolutional kernel branches for different tasks to obtain an initial multi-task deep learning model including a common layer and convolutional kernel branches for different tasks;

[0047] Input the two-dimensional spectral information matrix into the initial multi-task deep learning model, perform multiple iterative trainings, confirm the principal component factors and optimal weights of the initial multi-task deep learning model to obtain a multi-task deep learning model;

[0048] Among them, each convolutional kernel branch corresponds to different temperatures, different main gas components, and different concentration parameters.

[0049] Optionally, the construction includes a common layer of multiple convolutional kernels and a max pooling layer, including:

[0050] Construct using the convolutional neural network of Equation (10):

[0051]

[0052] In Equation (10), f(m) is the typical two-dimensional spectral information matrix constructed by Equation (9), n is the length of the signal f(n), and S(n) is the convolution result sequence, with a length of len(f(n)) + len(g(n)) - 1.

[0053] The second aspect of the present invention provides a system for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectra, the system including:

[0054] A data acquisition module for acquiring absorption spectral data of the boiler flue gas to be measured;

[0055] An inversion module for inputting the absorption spectral data of the boiler flue gas to be measured into a multi-task deep learning model to obtain the composition and concentration parameters of the boiler flue gas;

[0056] Wherein, the multi-task deep learning model is established based on the absorption spectral data of complex boiler flue gas at different temperatures and different concentrations.

[0057] On the other hand, the present invention provides a machine-readable storage medium, on which instructions are stored, and these instructions are used to cause a machine to execute the method for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectra.

[0058] Through the above technical solution, a method for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectra is provided. The complex high-temperature boiler flue gas in this method mainly includes CO 2 , H 2 O, CO, O 2 , NO, NO 2 , SO 2 , H 2 S, etc. This method obtains the absorption spectral data of complex boiler flue gas at different temperatures and different concentrations, constructs a multi-task deep learning algorithm based on the absorption spectral data, and quickly and accurately inverts the measured absorption spectral data of the boiler flue gas through the multi-task deep learning algorithm to obtain the main components and concentration parameters of the boiler flue gas, which can effectively cover the composition and concentration range of complex high-temperature boiler flue gas and reduce redundant data.

[0059] Based on a multi-task deep learning algorithm, the measured absorption spectrum data of boiler flue gas can be quickly and accurately inverted, enabling the synchronous acquisition of various main components and concentration parameters of the boiler flue gas. It can effectively eliminate the mutual interference of various gas components and the influence of particles, etc., and improve the inversion accuracy.

[0060] In the present invention, through a spectral two-dimensional image conversion algorithm, the absorption spectrum data of complex boiler flue gas is converted into a spectrogram, thereby establishing a two-dimensional spectral information matrix. A multi-task network is constructed by a multi-task convolutional network form construction method, and multiple iterative trainings are carried out to find the main component factors of the network and their optimal weights. Thus, the measured absorption spectrum data of boiler flue gas is inverted by the obtained main component factors and their optimal weights, effectively improving the iterative efficiency and reducing the inversion calculation time.

[0061] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific embodiment part. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The drawings are used to provide a further understanding of the embodiments of the present invention, and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation to the embodiments of the present invention. In the drawings:

[0063] Figure 1 is a flow chart of a method for inverting the components and concentrations of complex high-temperature boiler flue gas based on absorption spectrum provided by an embodiment of the present invention;

[0064] Figure 2 is a schematic diagram of a multi-task network constructed by a multi-task convolutional network form construction method provided by an embodiment of the present invention;

[0065] Figure 3 is a block diagram of a system for inverting the components and concentrations of complex high-temperature boiler flue gas based on absorption spectrum provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] The following will describe in detail the specific embodiments of the present invention with reference to the drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0067] Figure 1 is a flow chart of a method for inverting the components and concentrations of complex high-temperature boiler flue gas based on absorption spectrum provided by an embodiment of the present invention. As Figure 1 shown, the method includes:

[0068] Step 1: Obtain the absorption spectrum data of the boiler flue gas to be measured; in this embodiment, the absorption spectrum data of the boiler flue gas to be measured is obtained by measuring the boiler flue gas to be measured according to the absorption spectrum technology.

[0069] Step 2: Input the absorption spectrum data of the boiler flue gas to be measured into the multi-task deep learning model to obtain the composition and concentration parameters of the boiler flue gas;

[0070] Among them, the multi-task deep learning model is established based on the absorption spectrum data of complex boiler flue gas at different temperatures and different concentrations.

[0071] In this embodiment, the complex high-temperature boiler flue gas mainly includes CO 2 , H 2 O, CO, O 2 , NO, NO 2 , SO 2 , H 2 S, etc.

[0072] In this embodiment, the absorption spectrum data of complex boiler flue gas at different temperatures and different concentrations are obtained through the following steps:

[0073] 1) Taking the temperature of the complex high-temperature boiler flue gas, the main gas components and concentration parameters as independent variables, calculate the optical depth OD of complex boiler flue gas at different temperatures and different concentrations according to the absorption spectrum principle.

[0074] 2) Draw the absorption spectrum data of complex boiler flue gas at the same temperature and the same concentration according to the optical depth OD of different bands of complex boiler flue gas at the same temperature and the same concentration. The absorption spectrum is drawn with the optical depth OD as the vertical axis and the band as the horizontal axis. The drawing process is the splicing process of the optical depth OD of different bands of complex boiler flue gas at the same temperature and the same concentration on the horizontal axis.

[0075] 3) Change the temperature of the complex high-temperature boiler flue gas, the main gas components and concentration parameters, and draw the absorption spectrum data of complex boiler flue gas at different temperatures and different concentrations.

[0076] In some embodiments, the absorption spectrum data of complex boiler flue gas at different temperatures and different concentrations are stored in the absorption spectrum library.

[0077] In this application, the principle of the absorption spectrum technology lies in that there is a certain quantitative relationship between the concentration of the gas to be measured and the attenuation degree of the light intensity in a specific band. Through this mathematical relationship, the gas to be measured can be conveniently determined and the concentration value of the gas to be measured can be monitored. In this embodiment, the absorption spectrum principle is:

[0078] For a single gas, under the irradiation of light with different wavelengths, the intensity of the incident light and the transmitted light conforms to the following relationship:

[0079] I(λ) = I 0 (λ)exp[-σ(λ)CL] (1)

[0080] In formula (1), I(λ) is the transmitted light intensity of the light with wavelength λ emitted by the light source after passing through the gas to be measured, and I 0 (λ) is the initial light intensity of the light with wavelength λ emitted by the light source, L is the optical path of the gas, C is the concentration of the gas to be measured, and σ(λ) is the absorption cross-section of the gas to be measured, which is related to the wavelength λ;

[0081] According to the principle that the light attenuation of the complex high-temperature flue gas in the boiler includes extinction caused by Rayleigh scattering and Mie scattering, formula (1) becomes:

[0082]

[0083] In formula (2), ε R is the Rayleigh scattering extinction coefficient, and ε M is the Mie scattering extinction coefficient;

[0084] According to the fact that the complex high-temperature flue gas in the boiler is a mixture of multiple gas components, and multiple gas components absorb the light emitted by the light source, and there are also the influences of other gas molecules and particles at the same time, formula (2) is expressed as:

[0085]

[0086] In formula (3), σ i and C i respectively represent the absorption cross-section and concentration of the i-th gas that absorbs light;

[0087] According to the fact that the gas absorption contains two parts of "slow change" and "fast change", the absorption cross-section σ(λ) of the gas to be measured is expressed as:

[0088] σ(λ) = σ s (λ) + σ f (λ) (4)

[0089] In formula (4), σ s (λ) is the slow-varying absorption cross-section of gas absorption, and σ f (λ) is the fast-varying absorption cross-section of gas absorption;

[0090] According to the "slow change" and "fast change" principles in gas absorption, broadband absorption and narrowband absorption can be obtained:

[0091]

[0092] Define the optical thickness OD at different wavelengths λ as:

[0093]

[0094] In formula (6), is broadband absorption, that is, broadband absorption caused by broadband absorption and scattering in gas absorption, which can be removed from the optical thickness through analysis; is narrowband absorption, that is, caused by gas absorption;

[0095] Considering that the optical attenuation of the complex high-temperature flue gas in the boiler is affected by temperature, the optical thickness OD is corrected for temperature according to Equation (7):

[0096]

[0097] Through the above method, the accurate optical thickness of the complex high-temperature flue gas in the boiler can be calculated.

[0098] In this embodiment, the deep learning model is established through the following steps:

[0099] 1) Adopt the spectral two-dimensional image conversion algorithm to convert the absorption spectral data in the absorption spectral library into a spectrogram, and obtain a two-dimensional spectral information matrix, including:

[0100] Process the absorption spectral data according to the following formula:

[0101] S = XX T (8)

[0102] In Equation (8), S is the two-dimensional spectral information matrix, and X is the spectral data column vector, obtaining a typical two-dimensional spectral information matrix:

[0103]

[0104] In Equation (9), a i is one-dimensional spectral data.

[0105] 2) Construct a multi-task deep learning model according to the two-dimensional spectral information matrix, including:

[0106] 2-1) Construct a common layer including multiple convolutional kernels and a max pooling layer;

[0107] 2-2) Construct convolutional kernel branches for different tasks, obtaining an initial multi-task deep learning model including a common layer and convolutional kernel branches for different tasks;

[0108] 2-3) Input the two-dimensional spectral information matrix into the initial multi-task deep learning model, perform multiple iterative trainings, confirm the principal component factors and optimal weights of the initial multi-task deep learning model, and obtain the target multi-task deep learning model;

[0109] Among them, each convolutional kernel branch corresponds to different temperatures, different main gas components, and different concentration parameters.

[0110] In this embodiment, the construction of the common layer including multiple convolutional kernels and a max pooling layer includes:

[0111] The construction is carried out using the convolutional neural network of Equation (10):

[0112]

[0113] In Equation (10), f(m) is the typical two-dimensional spectral information matrix constructed by Equation (9), n is the length of the signal f(n), and S(n) is the convolution result sequence with a length of len(f(n)) + len(g(n)) - 1.

[0114] By the above steps, multiple shared convolutional layers are architected, a common layer including a series of convolutional kernels and a max pooling layer is set. After extracting the spectral map data in the common layer, the information is directed to multiple different branches, and each branch corresponds to temperature, main gas components, and concentration parameters, thereby forming a multi-task network.

[0115] In this embodiment, the constructed multi-task network is as Figure 2 shown. This multi-task network includes convolutional kernel A, pooling layer A, convolutional kernel B, pooling layer B, convolutional kernel C, pooling layer C, convolutional kernel D as the common layer, and different branches. The absorption spectral data of the boiler flue gas to be measured is first input into convolutional kernel A, and after being processed by convolutional kernel A, pooling layer A, convolutional kernel B, pooling layer B, convolutional kernel C, pooling layer C, and convolutional kernel D in sequence, the data is transmitted to different branches. Figure 2 is another embodiment of the multi-task network constructed in this application, in which at least 8 typical branches are shown. For example, the spectral map branches of H 2 S with different concentrations and temperatures. According to this branch, the concentration of H 2 S can be inversely obtained; another example is the spectral map branches of SO 2 with different concentrations and temperatures. According to this branch, the concentration of SO 2 can be inversely obtained; another example is the spectral map branches of NO with different concentrations and temperatures. According to this branch, the concentration of NO can be inversely obtained; another example is the spectral map branches of NO 2 with different concentrations and temperatures. According to this branch, the concentration of NO 2 can be inversely obtained; for example, the spectral map branches of H 2 O with different concentrations and temperatures. According to this branch, the concentration of H 2 O can be inversely obtained; for example, the spectral map branches of O 2 with different concentrations and temperatures. According to this branch, the concentration of O 2 can be inversely obtained; for example, the spectral map branches of CO with different concentrations and temperatures. According to this branch, the concentration of CO can be inversely obtained; for example, the spectral map branches of CO 2 with different concentrations and temperatures. According to this branch, the concentration of CO2 Concentration.

[0116] In the above embodiments, only the composition and concentration of the complex flue gas in the boiler are measured. The inversion of the concentration of any other gas or gases can also adopt the above inversion method.

[0117] Figure 3 It is a block diagram of a system for inverting the composition and concentration of complex high-temperature flue gas in a boiler based on absorption spectroscopy provided by an embodiment of the present invention. As Figure 3 shown, the system includes:

[0118] A data acquisition module for acquiring absorption spectroscopy data of the flue gas of the boiler to be measured;

[0119] An inversion module for inputting the absorption spectroscopy data of the flue gas of the boiler to be measured into a multi-task deep learning model to obtain the composition and concentration parameters of the flue gas of the boiler;

[0120] Among them, the multi-task deep learning model is established based on the absorption spectroscopy data of complex boiler flue gas at different temperatures and different concentrations.

[0121] The embodiment of the present invention also provides a machine-readable storage medium, on which instructions are stored, and the instructions are used to make a machine execute the method for inverting the composition and concentration of complex high-temperature flue gas in a boiler based on absorption spectroscopy.

[0122] Those skilled in the art can understand that all or part of the steps in the methods of the above embodiments can be completed by instructing relevant hardware through a program. The program is stored in a storage medium, including several instructions for making a single-chip microcomputer, a chip or a processor execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disc that can store program codes.

[0123] The above has described in detail the optional embodiments of the present invention in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all belong to the protection scope of the embodiments of the present invention. In addition, it should be noted that in the above specific embodiments, the various specific technical features described can be combined in any appropriate manner without conflict. To avoid unnecessary repetition, the embodiments of the present invention will not separately describe various possible combination methods.

[0124] In addition, any combination can be made among various different embodiments of the present invention, as long as it does not violate the idea of the embodiments of the present invention, and it should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A method for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectra, characterized in that, the method includes: Obtaining the absorption spectrum data of the boiler flue gas to be measured; Inputting the absorption spectrum data of the boiler flue gas to be measured into a multi-task deep learning model to obtain the composition and concentration parameters of the boiler flue gas; wherein, the multi-task deep learning model is established based on the absorption spectrum data of complex boiler flue gas at different temperatures and different concentrations; The absorption spectrum data of complex boiler flue gas at different temperatures and different concentrations are obtained through the following steps: Taking the temperature of the complex high-temperature flue gas in the boiler, the main gas components and concentration parameters as independent variables, the optical thickness of the complex boiler flue gas at different temperatures and different concentrations is calculated according to the principle of absorption spectroscopy ; According to the optical thickness of complex boiler flue gas at the same temperature and the same concentration in different wavebands The absorption spectral data of complex boiler flue gas at the same temperature and the same concentration are obtained by plotting; Changing the temperature, main gas components and concentration parameters of the complex high-temperature boiler flue gas, and drawing the absorption spectrum data of the complex boiler flue gas at different temperatures and different concentrations; The multi-task deep learning model is established through the following steps: Using a spectral two-dimensional image conversion algorithm to convert the absorption spectrum data in the absorption spectrum library into a spectrogram to obtain a two-dimensional spectral information matrix; Constructing a multi-task deep learning model based on the two-dimensional spectral information matrix, including: Constructing a common layer including multiple convolutional kernels and max pooling layers; Constructing convolutional kernel branches for different tasks to obtain an initial multi-task deep learning model including a common layer and convolutional kernel branches for different tasks; Inputting the two-dimensional spectral information matrix into the initial multi-task deep learning model, performing multiple iterative trainings, and confirming the principal component factors and optimal weights of the initial multi-task deep learning model to obtain the multi-task deep learning model; wherein, each convolutional kernel branch corresponds to different temperatures, different main gas components and different concentration parameters.

2. The method for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectra according to claim 1, characterized in that, Storing the absorption spectrum data of the complex boiler flue gas at different temperatures and different concentrations in an absorption spectrum library.

3. The method for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectra according to claim 1, characterized in that, The absorption spectrum principle is: For a single gas, when irradiated with light of different wavelengths, the intensities of the incident light and the transmitted light conform to the following relationship: In the formula , is the transmitted light intensity after the light with a wavelength of emitted by the light source passes through the gas to be measured, is the initial light intensity of the light with a wavelength of emitted by the light source, is the optical path of the gas, is the concentration of the gas to be measured, is the absorption cross-section of the gas to be measured, which is related to the wavelength . According to the principle that the light attenuation of the complex high-temperature flue gas in the boiler is caused by extinction including Rayleigh scattering and Mie scattering, the formula becomes: In the formula , is the Rayleigh scattering extinction coefficient, is the Mie scattering extinction coefficient; According to the fact that various gas components in the complex high-temperature flue gas of the boiler all absorb the light emitted by the light source, considering the influence of other gas molecules and particles, the formula is expressed as: Formula wherein and respectively represent the absorption cross-section and concentration of the th gas for light absorption; According to the "slow change" and "fast change" principles in gas absorption, the absorption cross-section of the gas to be measured is expressed as: In the formula , is the slow-varying absorption cross section of gas absorption, and the fast-varying absorption cross section of gas absorption; According to the "slow change" and "fast change" principles in gas absorption, broadband absorption and narrowband absorption are obtained: Define different wavelengths The optical thickness below is defined as: In the formula , is broadband absorption; is narrowband absorption; Considering the influence of temperature on the optical attenuation of complex high-temperature flue gas in boilers, the optical thickness According to the formula Perform temperature correction: 。 4. The method for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectra according to claim 1, characterized in that, The step of using a spectral two-dimensional image conversion algorithm to convert the absorption spectrum data in the absorption spectrum library into a spectrogram to obtain a two-dimensional spectral information matrix includes: Process the absorption spectral data according to the formula as follows: In the formula , is a two-dimensional spectral information matrix, is a spectral data column vector, and a typical two-dimensional spectral information matrix is obtained: In the formula , is one-dimensional spectral data.

5. The method for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectra according to claim 4, characterized in that, The step of constructing a common layer including multiple convolutional kernels and max pooling layers includes: Constructed using the convolutional neural network of the formula : In the formula , is the typical two-dimensional spectral information matrix constructed for formula (9), is the signal length, is the convolution result sequence, with a length of .

6. A system for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectra, characterized in that, the system includes: A data acquisition module for obtaining the absorption spectrum data of the boiler flue gas to be measured; An inversion module for inputting the absorption spectrum data of the boiler flue gas to be measured into a multi-task deep learning model to obtain the composition and concentration parameters of the boiler flue gas; wherein, the multi-task deep learning model is established based on the absorption spectrum data of complex boiler flue gas at different temperatures and different concentrations; The absorption spectral data of complex boiler flue gas at different temperatures and different concentrations are obtained through the following steps: Taking the complex high-temperature flue gas temperature, main gas components and concentration parameters of the boiler as independent variables, calculate the optical thickness of complex boiler flue gas at different temperatures and different concentrations according to the absorption spectrum principle ; According to the optical thickness of complex boiler flue gas at the same temperature and the same concentration in different wavebands Absorption spectral data of complex boiler flue gas at the same temperature and the same concentration are obtained by plotting; Change the temperature, main gas components and concentration parameters of the complex high-temperature boiler flue gas, and plot the absorption spectral data of the complex boiler flue gas at different temperatures and different concentrations; A multi-task deep learning model is established through the following steps: Adopt a spectral two-dimensional image conversion algorithm to convert the absorption spectral data in the absorption spectral library into a spectrogram to obtain a two-dimensional spectral information matrix; Construct a multi-task deep learning model according to the two-dimensional spectral information matrix, including: Construct a common layer including multiple convolutional kernels and max-pooling layers; Construct convolutional kernel branches for different tasks to obtain an initial multi-task deep learning model including a common layer and convolutional kernel branches for different tasks; Input the two-dimensional spectral information matrix into the initial multi-task deep learning model, perform multiple iterative trainings, confirm the principal component factors and optimal weights of the initial multi-task deep learning model, and obtain the multi-task deep learning model; Among them, each convolutional kernel branch corresponds to different temperatures, different main gas components and different concentration parameters.

7. A machine-readable storage medium, on which instructions are stored, and the instructions are used to make the machine execute the method for inverting the composition and concentration of complex high-temperature boiler flue gas based on absorption spectrum according to any one of claims 1-5.

Images