An error correction system and method for furnace infrared imaging temperature measurement scenarios

Abstract

This application relates to the field of metallurgical engineering technology, and in particular to an error correction system and method for infrared imaging temperature measurement in a furnace. The system's edge computing nodes are configured to: generate infrared data sequences using infrared radiation intensity data within a first time window as first data frames according to a first preset frequency; the central control server is configured to: decompose each first data frame into K eigenmode components using a variational mode decomposition algorithm; reconstruct the signals of each first data frame using the eigenmode components to obtain first radiation intensity data; convert the first radiation intensity data into preliminary temperature field data using the inverse function of Planck's blackbody radiation law; and correct the preliminary temperature field data based on dynamic compensation parameters to obtain target temperature field data and generate combustion optimization control commands for the furnace. This application utilizes variational mode decomposition and deep learning models to effectively eliminate interference such as self-radiation from optical tubes, achieving accurate compensation.

Description

Technical Field

[0001] This application relates to the field of metallurgical engineering technology, and in particular to an error correction system and method for furnace infrared imaging temperature measurement. Background Technology

[0002] In infrared thermal imaging monitoring of industrial combustion furnace temperature, existing technologies typically suffer from the following problems: First, the optical tube inserted into the furnace will be heated to a very high temperature, and the infrared radiation emitted by the tube will be superimposed on the target signal, resulting in an artificially high temperature measurement result. Existing methods lack effective compensation for this interference.

[0003] Second, the emissivity of fly ash, flue gas, and coking layer on the target surface changes constantly, resulting in uncertainty in the attenuation of the radiation transmission path. Traditional fixed parameter correction methods cannot adapt to this dynamic change, leading to large temperature measurement errors.

[0004] Third, the existing system only outputs temperature values ​​and cannot automatically identify abnormal hot zones and control combustion parameters in reverse, thus failing to form a closed-loop optimization.

[0005] Therefore, there is an urgent need for an error correction system that can dynamically compensate for path attenuation and emissivity changes, and can automatically generate combustion control commands based on the corrected temperature field. Summary of the Invention

[0006] This application provides an error correction system and method for furnace infrared imaging temperature measurement scenarios to solve the problem of poor monitoring performance of traditional monitoring systems.

[0007] In a first aspect, embodiments of this application provide an error correction system for an infrared imaging temperature measurement scenario in an industrial combustion furnace, wherein the furnace wall of the industrial combustion furnace has an observation hole; the system includes a furnace infrared temperature measurement terminal fixed at the observation hole, an edge computing node communicatively connected to the furnace infrared temperature measurement terminal, and a centralized control server communicatively connected to the edge computing node; the furnace infrared temperature measurement terminal includes at least: an optical lens tube extending into the observation hole, and an infrared thermal imager coupled to the optical path of the optical lens tube; the infrared thermal imager is used to continuously collect infrared radiation intensity data of a target area inside the furnace at a preset sampling frequency; the edge computing node is configured to: generate an infrared data sequence by taking the infrared radiation intensity data within a first time window as a first data frame according to a first preset frequency, and perform... After preprocessing, the data is uploaded to the centralized control server. The centralized control server is configured to: decompose each first data frame into K intrinsic mode components using a variational mode decomposition algorithm; reconstruct the signal of each first data frame using the intrinsic mode components to obtain the first radiation intensity data; convert the first radiation intensity data into preliminary temperature field data using the inverse function of Planck's blackbody radiation law; construct a feature vector using the first radiation intensity data and the preliminary temperature field data, input the feature vector into a pre-trained error correction strategy model to obtain dynamic compensation parameters; correct the preliminary temperature field data based on the dynamic compensation parameters to obtain the target temperature field data; and generate combustion optimization control commands for the furnace based on the target temperature field data and send them to the edge computing nodes.

[0008] In one possible implementation, the furnace infrared temperature measurement terminal further includes a thin-film thermocouple attached to the inner wall of the optical tube. The thin-film thermocouple is used to collect the real-time temperature of the optical tube at a preset sampling frequency. The edge computing node is also configured to: generate a tube temperature sequence based on the real-time tube temperature within a first time window according to a first preset frequency; wherein each first data frame in the infrared data sequence corresponds one-to-one with each temperature value in the tube temperature sequence on the timestamp; preprocess the infrared data sequence using the tube temperature sequence to obtain a preprocessed infrared data sequence; the preprocessing refers to performing tube transmittance attenuation compensation on the infrared data sequence based on the real-time tube temperature data sequence; the tube transmittance attenuation compensation is to multiply each first data frame in the infrared data sequence by the attenuation compensation coefficient related to the real-time tube temperature at the corresponding time according to a preset transmittance-temperature mapping table.

[0009] In one possible implementation, the variational mode decomposition algorithm includes: For a first data frame, set the value of the number of intrinsic mode components K, and initialize the center frequency and Lagrange multiplier operator of each intrinsic mode component; The intrinsic mode components and center frequencies corresponding to each first data frame are iteratively updated according to the following first and second formulas: First formula: ; Second formula: ; Where k is the index of the intrinsic mode component, 0 < k ≤ K, n is the index of the iteration number; ω is the angular frequency independent variable, 0 ≤ ω ≤ ω max， ω max =π f s , f s This is the preset sampling frequency for the infrared thermal imager; This represents the Fourier transform of the kth intrinsic mode component in the (n+1)th iteration. The Fourier transform of the first data frame; Let be the Fourier transform of the i-th intrinsic mode component, where i is the index of the intrinsic mode component and i ≠ k; α is the Fourier transform of the Lagrange multiplication operator; α is the penalty factor; The center frequency of the k-th intrinsic mode component; as well as, This represents the center frequency of the k-th intrinsic mode component in the (n+1)th iteration; The power spectral density corresponding to the Fourier transform of the k-th intrinsic mode component is represented. The first moment of the power spectrum; This represents the total energy of the power spectrum; When the relative change of the Fourier transform of the intrinsic mode components in two consecutive iterations is less than the preset convergence threshold, the iteration stops and the intrinsic mode components corresponding to the first data frame are output.

[0010] In one possible implementation, in the step of reconstructing the signal of each first data frame using intrinsic mode components to obtain the first radiation intensity data, the centralized control server is specifically configured to: calculate the energy spectral entropy of the corresponding intrinsic mode component for each first data frame in the infrared data sequence; mark J intrinsic mode components whose energy spectral entropy is less than or equal to a preset entropy threshold as thermal radiation dominant mode components; linearly superimpose all thermal radiation dominant mode components corresponding to the same first data frame to reconstruct the first data frame; and generate the first radiation intensity data based on each reconstructed first data frame.

[0011] In one possible implementation, the infrared radiation intensity data is a two-dimensional matrix output by the focal plane array of an infrared thermal imager. The two-dimensional matrix has H rows and W columns, and each element in the two-dimensional matrix is ​​a pixel. The value of each pixel is the spectral radiance value corresponding to that pixel. In the step of converting the first radiation intensity data into preliminary temperature field data using the inverse function of Planck's blackbody radiation law, the central control server is specifically configured to: calculate the brightness temperature of each pixel in each first data frame of the first radiation intensity data according to the following third formula. Third formula: ; in, Represents pixel ( i , j The brightness temperature is λ, where λ is the center response wavelength of the infrared thermal imager, c1 is the first radiation constant, and c2 is the second radiation constant. For pixels ( i , j The spectral radiance value of ). The calculated brightness and temperature are arranged frame by frame according to the row and column order of pixels to form a two-dimensional matrix of H rows and W columns, thus obtaining the second data frame for each frame; the second data frames are spliced ​​together in chronological order to obtain the preliminary temperature field data.

[0012] In one possible implementation, the step of constructing a feature vector using the first radiation intensity data and preliminary temperature field data, and inputting the feature vector into a pre-trained error correction strategy model to obtain dynamic compensation parameters, involves the central control server being specifically configured to: expand each first data frame in the first radiation intensity data into a one-dimensional vector, and concatenate it with the one-dimensional vector expanded from the second data frame corresponding to the first data frame to form an original feature vector; normalize the original feature vector so that the numerical range of each dimension falls within the [0,1] interval to obtain the feature vector; and input the feature vector into the pre-trained error correction strategy model to obtain dynamic compensation parameters.

[0013] In one possible implementation, the dynamic compensation parameters include an emissivity dynamic correction factor matrix and a path radiation interference compensation matrix. In the step of correcting the initial temperature field data based on the dynamic compensation parameters to obtain the target temperature field data, the centralized control server is specifically configured to: calculate the corresponding target temperature for each pixel of each second data frame according to the following fourth formula. Fourth formula: ; in, Represents pixel ( i , jThe target temperature is λ, the center response wavelength of the infrared thermal imager is λ, c1 is the first radiation constant, and c2 is the second radiation constant. For pixels ( i , j The spectral radiance value of ). This is the emissivity dynamic correction factor matrix; This is the path radiation interference compensation matrix; The calculated target temperatures are arranged frame by frame according to the row and column order of pixels to form a two-dimensional matrix of H rows and W columns, thus obtaining each target temperature frame; the target temperature frames are then stitched together in chronological order to obtain the target temperature field data.

[0014] In one possible implementation, in the step of generating combustion optimization control instructions for the furnace based on target temperature field data, the centralized control server is specifically configured to: calculate the gradient magnitude of each pixel of each target temperature frame within a first time window, and mark pixels with gradient magnitudes greater than a preset gradient threshold as edge points; use interconnected edge point regions as candidate thermal anomaly regions, and calculate the target features of each candidate thermal anomaly region; the target features include the number of pixels in the region and the average temperature of the region; determine the candidate thermal anomaly regions whose target features meet preset conditions as actual thermal anomaly regions; and generate corresponding combustion optimization control instructions based on the actual thermal anomaly regions.

[0015] In one possible implementation, in the step of generating corresponding combustion optimization control instructions based on actual thermal anomaly zones, the centralized control server is specifically configured to: calculate the average of the regional average temperature of all actual thermal anomaly zones to obtain a first average, and calculate the average of the regional area of ​​all actual thermal anomaly zones to obtain a second average; if the first average exceeds a preset temperature upper limit, generate an instruction to reduce the fuel supply rate or increase the air supply volume; if the first average does not exceed the preset temperature upper limit and the second average exceeds a preset area threshold, generate a furnace soot blowing instruction or a coking warning signal.

[0016] Secondly, embodiments of this application also provide an error correction method for an infrared imaging temperature measurement scenario in an industrial combustion furnace. The method includes: generating an infrared data sequence from infrared radiation intensity data as a first data frame according to a first preset frequency, and preprocessing it; decomposing each first data frame into K intrinsic mode components using a variational mode decomposition algorithm; reconstructing the signals of each first data frame using the intrinsic mode components to obtain first radiation intensity data; converting the first radiation intensity data into preliminary temperature field data using the inverse function of Planck's blackbody radiation law; constructing a feature vector using the first radiation intensity data and the preliminary temperature field data, and inputting the feature vector into a pre-trained error correction strategy model to obtain dynamic compensation parameters; correcting the preliminary temperature field data based on the dynamic compensation parameters to obtain target temperature field data; and generating combustion optimization control commands for the combustion furnace based on the target temperature field data.

[0017] As can be seen from the above, this application provides an error correction system and method for furnace infrared imaging temperature measurement, applied to an industrial combustion furnace. The furnace wall of the industrial combustion furnace has an observation hole. The system includes a furnace infrared temperature measurement terminal fixed at the observation hole, an edge computing node communicatively connected to the furnace infrared temperature measurement terminal, and a central control server communicatively connected to the edge computing node. The furnace infrared temperature measurement terminal includes at least: an optical lens tube extending into the observation hole, and an infrared thermal imager coupled to the optical path of the optical lens tube. The infrared thermal imager is used to continuously collect infrared radiation intensity data of the target area inside the furnace at a preset sampling frequency. The edge computing node is configured to: generate an infrared data sequence by using the infrared radiation intensity data within a first time window as a first data frame according to a first preset frequency. The data is preprocessed and then uploaded to the centralized control server. The centralized control server is configured to: decompose each first data frame into K intrinsic mode components using a variational mode decomposition algorithm; reconstruct the signal of each first data frame using the intrinsic mode components to obtain first radiation intensity data; convert the first radiation intensity data into preliminary temperature field data using the inverse function of Planck's blackbody radiation law; construct feature vectors using the first radiation intensity data and preliminary temperature field data, input the feature vectors into a pre-trained error correction strategy model to obtain dynamic compensation parameters; correct the preliminary temperature field data based on the dynamic compensation parameters to obtain target temperature field data; and generate combustion optimization control instructions for the furnace based on the target temperature field data and send them to the edge computing nodes. It can be seen that this embodiment, through edge computing and centralized control collaboration, utilizes variational mode decomposition and deep learning models to effectively eliminate interference such as optical tube self-radiation and fly ash flue gas, achieving dynamic and accurate compensation of furnace temperature, and automatically identifying thermal anomaly zones to generate combustion optimization instructions, significantly improving temperature measurement accuracy and furnace operation safety. Attached Figure Description

[0018] Figure 1This is a schematic diagram of the error correction system for furnace infrared imaging temperature measurement provided in the embodiments of this application; Figure 2 This is a flowchart illustrating the error correction method for furnace infrared imaging temperature measurement scenarios provided in this application embodiment.

[0019] Among them, 100-furnace infrared temperature measurement terminal, 101-optical lens tube, 102-infrared thermal imager, 103-thin film thermocouple, 200-edge computing node, and 300-centralized control server. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of this application.

[0021] To address issues such as interference from the mirror tube radiation, dynamic changes in emissivity and path attenuation, large temperature measurement errors, and the inability to achieve closed-loop control in infrared temperature measurement of industrial furnaces, this application provides an error correction system and method for furnace infrared imaging temperature measurement scenarios. This system and method can compensate for the mirror tube temperature in real time, perform variational mode decomposition for noise reduction, dynamically correct temperature errors using an intelligent model, and automatically identify thermal anomalies to generate combustion optimization commands. In this way, the accuracy of furnace temperature measurement is effectively improved, and closed-loop optimization of temperature measurement and combustion control is achieved.

[0022] Figure 1 This is a schematic diagram of the error correction system for furnace infrared imaging temperature measurement scenario provided in the embodiments of this application.

[0023] like Figure 1 As shown, this application embodiment provides an error correction system for furnace infrared imaging temperature measurement scenarios, applied to industrial furnaces (such as petrochemical heating furnaces, power plant boilers, pyrolysis furnaces, etc.). The furnace wall of the industrial furnace has observation holes. The system includes a furnace infrared temperature measurement terminal 100 fixed at the observation hole, an edge computing node 200 communicatively connected to the furnace infrared temperature measurement terminal 100, and a central control server 300 communicatively connected to the edge computing node 200. Typically, the edge computing node 200 is deployed in a field control cabinet, and the central control server 300 is located in a central control room. The two communicate via an Industrial Ethernet or 5G network.

[0024] The furnace infrared temperature measurement terminal 100 includes at least: an optical lens tube 101 extending into the observation aperture, an infrared thermal imager 102 coupled to the optical path of the optical lens tube, and a thin-film thermocouple 103 attached to the inner wall of the optical lens tube. The optical lens tube 101 is made of a high-temperature resistant material (e.g., sapphire or quartz glass). The infrared thermal imager 102 is used to continuously acquire infrared radiation intensity data of the target area inside the furnace at a preset sampling frequency (e.g., 50 Hz). The thin-film thermocouple 103 is used to acquire the real-time lens tube temperature of the optical lens tube 101 at the same preset sampling frequency, with a sampling accuracy of ±0.5 degrees Celsius (°C).

[0025] It should be further explained that "optical path coupling" refers to precisely aligning the outgoing optical path of the optical tube with the incident optical path of the infrared thermal imager in space, so that the infrared radiation emitted by the target inside the furnace can be transmitted through the optical tube and enter the detector focal plane of the infrared thermal imager efficiently and without obstruction, forming a clear temperature image.

[0026] It is worth noting that the infrared radiation intensity data is a two-dimensional matrix output by the focal plane array of the infrared thermal imager 102. The two-dimensional matrix has H rows and W columns, and each element in the two-dimensional matrix is ​​a pixel. The value of each pixel is the spectral radiance value corresponding to that pixel.

[0027] In this embodiment, the resolution of the infrared thermal imager (i.e., the number of pixel rows and columns H×W per frame) can be selected according to the actual furnace monitoring requirements and the model of the thermal imager. Common values ​​include: Low resolution: 160×120 or 130×240, suitable for local monitoring or cost-sensitive scenarios.

[0028] Medium resolution: 384×288 or 640×480, balancing detail and data volume.

[0029] High resolution: 640×512, 1024×768 or 1280×1024, for large field of view or high-precision temperature measurement.

[0030] In some implementations, the target area inside the furnace can refer to the surface of the furnace tube, the inner wall of the furnace wall refractory material, the flame area, the furnace outlet interface, etc., but this application does not specifically limit this.

[0031] In some implementations, the edge computing node 200 is an embedded industrial computer (e.g., an artificial intelligence (AI) computing board based on an Advanced Reduced Instruction Set Machine (ARM) architecture), which integrates a data acquisition card, a preprocessing module, and a communication module. The central control server 300 is a high-performance industrial server, deployed with a deep learning inference engine and a database.

[0032] Furthermore, the edge computing node 200 is configured to perform the following steps S101-S103.

[0033] S101: Generate an infrared data sequence by using the infrared radiation intensity data within the first time window as the first data frame according to the first preset frequency.

[0034] It is understandable that the first preset frequency is greater than the preset sampling frequency. For example, the preset sampling frequency can be from 25 Hz to 100 Hz, and the first preset frequency can be once every 1 second, corresponding to a first time window length of 1 second.

[0035] S102: Generate a tube temperature sequence based on the real-time tube temperature within the first time window according to the first preset frequency; wherein, each first data frame in the infrared data sequence corresponds one-to-one with each temperature value in the tube temperature sequence on the timestamp.

[0036] S103: After preprocessing the infrared data sequence, upload it to the centralized control server 300.

[0037] The preprocessing steps specifically include: preprocessing the infrared data sequence using the tube temperature sequence to obtain the preprocessed infrared data sequence; preprocessing refers to performing tube transmittance attenuation compensation on the infrared data sequence based on the real-time tube temperature data sequence; tube transmittance attenuation compensation is to multiply each first data frame in the infrared data sequence by the attenuation compensation coefficient related to the real-time tube temperature at the corresponding time according to the preset transmittance-temperature mapping table.

[0038] It is understandable that, since the optical tube 101 is located inside a high-temperature furnace, its own temperature rise will lead to a decrease in transmittance, and the optical tube 101 itself will also radiate infrared energy outward, which will be superimposed on the target signal. Therefore, this application compensates for this by means of the following method: The infrared transmittance of the optical lens tube 101 at different temperatures is measured in the laboratory beforehand, and a transmittance-temperature mapping table is created. For example, the transmittance is 0.92 at 500℃, 0.85 at 800℃, and 0.72 at 1200℃. This table is stored in the memory of the edge computing node 200.

[0039] For each first data frame in the infrared data sequence, the transmittance is obtained from the mapping table based on the real-time tube temperature corresponding to the same timestamp. Calculate the attenuation compensation coefficient The spectral radiance value of each pixel in the first data frame is multiplied by the attenuation compensation coefficient to obtain the compensated spectral radiance value.

[0040] Example: If the temperature of the microscope tube is 1000℃ at a certain moment, and the transmittance is found to be 0.78 from the table, then the attenuation compensation coefficient is 1 / 0.78≈1.282. Therefore, multiplying the original radiation intensity value (also known as the first data frame or infrared radiation intensity data) by 1.282 essentially eliminates the attenuation caused by the absorption of the microscope tube.

[0041] The preprocessed infrared data sequence is uploaded from edge computing node 200 to central control server 300.

[0042] The central control server 300 is configured to perform the following steps S200-S600.

[0043] S200: Using the variational mode decomposition algorithm, each first data frame is decomposed into K eigenmode components.

[0044] The specific steps of variational mode decomposition (VMD) are as follows (taking K=4 as an example). In practical applications, K can take the value of 2 or 4, and this application embodiment does not specifically limit it.

[0045] S201: For a first data frame, set the value of the number of intrinsic mode components K, and initialize the center frequency and Lagrange multiplier of each intrinsic mode component.

[0046] Where K takes the value 4. Furthermore, in the initialization step, this embodiment of the application can use values ​​from 0 to ω. max (ω) max =π fs , fs The frequency range of the infrared thermal imager 102 (preset sampling frequency) is uniformly divided into K segments, and the center frequency of each segment is taken as the initialization center frequency of the corresponding intrinsic mode component. Furthermore, in this embodiment, the Lagrange multiplication operator can be initialized to 0.

[0047] S202: Iteratively update the intrinsic mode components and center frequencies corresponding to each first data frame according to the following first and second formulas: First formula (modal component update): ; Second formula (center frequency update):

[0048] ; Where k is the index of the intrinsic mode component, 0 < k ≤ K, k is an integer, and n is the index of the iteration number; ω is the angular frequency independent variable, 0 ≤ ω ≤ ω max， ω max =π f s , f s This is the preset sampling frequency for the infrared thermal imager; This represents the Fourier transform of the kth intrinsic mode component in the (n+1)th iteration. The Fourier transform of the first data frame; Let be the Fourier transform of the i-th intrinsic mode component, where i is the index of the intrinsic mode component and i ≠ k; α is the Fourier transform of the Lagrange multiplication operator; α is the penalty factor; The center frequency of the k-th intrinsic mode component; as well as, This represents the center frequency of the k-th intrinsic mode component in the (n+1)th iteration; The power spectral density corresponding to the Fourier transform of the k-th intrinsic mode component is represented. The first moment of the power spectrum; This represents the total energy of the power spectrum.

[0049] Understandably, in the first iteration (n=0): the initial values ​​are substituted. That is, based on the initial intrinsic mode component center frequencies set in S201 and the Lagrange multiplication operator, the Fourier transform of the intrinsic mode components after the first iteration is calculated using the first formula, and then the center frequencies are updated using the second formula.

[0050] The (n+1)th iteration (n≥0): Substitute the Fourier transforms, center frequencies, and Lagrange multipliers of all intrinsic mode components obtained in the nth iteration, and calculate the result of the (n+1)th iteration using the first and second formulas.

[0051] It is understandable that the independent variable ω of angular frequency takes the value of [0, π]. f s Discrete frequency points within the interval.

[0052] S203: When the relative change of the Fourier transform of the intrinsic mode components in two adjacent iterations is less than the preset convergence threshold, stop the iteration and output the intrinsic mode components corresponding to the first data frame.

[0053] The specific convergence criterion is as follows: During the iteration process, the relative change of the Fourier transform of each eigenmode component between two adjacent iterations is calculated. The formula for calculating the relative change is as follows: ; in, The relative change is represented by k, which is the index of the intrinsic modal component, ranging from 1, 2, ... K K represents the total number of intrinsic modal components; in this implementation, K=4; n is the index of the iteration number. This represents the Fourier transform of the k-th eigenmode component in the (n+1)th iteration. This represents the Fourier transform of the k-th intrinsic mode component in the nth iteration; This represents the L2 norm (i.e., the Euclidean norm). This represents the energy difference between two consecutive Fourier transform iterations. This represents the energy of the nth iteration of the Fourier transform.

[0054] When the relative changes of the Fourier transforms of the K intrinsic mode components are less than a preset convergence threshold (e.g., 10), -6 When the iteration reaches K, it is considered to have converged and the iteration stops. At this point, K time-domain eigenmode components are output. It is understandable that the intrinsic mode component is obtained by performing an inverse Fourier transform on the output of the first formula.

[0055] In practical applications, after performing VMD decomposition on the first data frame, the following intrinsic mode components can be obtained: the first intrinsic mode component corresponds to the slowly changing temperature background of the furnace tube, the second corresponds to the pulsating component of the flame, and the third and fourth correspond to high-frequency noise (such as fly ash flicker).

[0056] S300: The signal of each first data frame is reconstructed using the intrinsic mode components to obtain the first radiation intensity data.

[0057] Specifically, step S300 may include the following steps S301-S303.

[0058] S301: For each first data frame in the infrared data sequence, calculate the energy spectral entropy of its corresponding intrinsic mode component, and mark the J intrinsic mode components whose energy spectral entropy is less than or equal to a preset entropy threshold as thermal radiation dominant mode components.

[0059] In this embodiment of the application, the step of calculating the energy spectrum entropy may be: first, the intrinsic mode components... u ( t Perform a Fourier transform to obtain U ( oh Then calculate the power spectral density. P ( oh )=| U ( oh )| 2 Furthermore, embodiments of this application can utilize power spectral density. P ( oh Normalized to a probability distribution Ultimately, probability distribution is used. p(ω) Calculate the energy spectrum entropy. The formula for calculating the energy spectrum entropy is: ; in, H Represents the energy spectrum entropy. p(ω)Represents a probability distribution. oh This represents the angular velocity independent variable.

[0060] In some implementations, the angular velocity independent variable oh The range of values ​​for is 0≤ oh ≤π f s In discrete implementation, oh Take discrete values oh m , oh m =m·2π f s / R m = 1, 2, ..., R / 2. The value of R is the number of the first data frames within the first time window.

[0061] The energy spectrum entropy reflects the degree of order of a signal: the lower the entropy, the more ordered the signal (usually corresponding to thermal radiation signals); the higher the entropy, the more random the signal (corresponding to noise or contamination signals). In this embodiment, a preset entropy threshold (e.g., 6.5, which can be calibrated experimentally) can be set, and J modal components with energy spectrum entropy less than or equal to this preset threshold can be marked as dominant thermal radiation modal components, while the remaining components (high-entropy noise components) are directly discarded.

[0062] S302: Linearly superimpose all thermal radiation dominant mode components corresponding to the same first data frame to reconstruct the first data frame.

[0063] It is worth noting that, in the embodiments of this application, linear superposition means: adding all thermal radiation dominant mode components corresponding to the same first data frame point by point at the same pixel position.

[0064] S303: Generate first radiation intensity data based on each reconstructed first data frame.

[0065] In this embodiment of the application, the first data frames are sorted according to the time sequence (or acquisition sequence) to obtain the first radiation intensity data. The first radiation intensity data has basically removed the random noise caused by fly ash and flue gas, and retained the real thermal radiation information inside the furnace.

[0066] S400: Using the inverse function of Planck's blackbody radiation law, the first radiation intensity data is converted into preliminary temperature field data.

[0067] In this embodiment of the application, the data output by the infrared thermal imager 102 is the spectral radiance value of each pixel. L ( i , jFurthermore, in the embodiments of this application, a blackbody assumption (emissivity = 1) can be adopted, and the spectral radiance value can be converted into radiance temperature (also known as radiation temperature) using the inverse function of Planck's blackbody radiation law. The specific calculation steps are as follows: S401-S403.

[0068] S401: For each pixel in each first data frame of the first radiation intensity data, calculate its brightness temperature according to the following third formula; Third formula: ; in, Represents pixel ( i , j The brightness temperature of ) is λ, the center response wavelength of the infrared thermal imager 102 is λ, c1 is the first radiation constant and c2 is the second radiation constant; For pixels ( i , j The spectral radiance value of ).

[0069] It is understandable that the third formula above is the inverse function of Planck's blackbody radiation law.

[0070] For example, suppose the spectral radiance of a certain pixel is 1000 W·m. -2 ·sr -1 The center response wavelength is 5μm, and c1 is 3.7418×10. 8 W·μm 4 ·m -2 c2 takes the value 1.4388 × 10 8 W·μm 4 K. Substituting this into the second formula, we can obtain the brightness temperature of this pixel as approximately 785 Kelvin (K), or approximately 512℃.

[0071] S402: Arrange the calculated brightness and temperature frame by frame according to the row and column order of the pixels to form a two-dimensional matrix of H rows and W columns, and then obtain the second data frame for each frame. S403: The second data frames are spliced ​​together in chronological order to obtain preliminary temperature field data.

[0072] Understandably, the above calculation is repeated for each pixel, and the brightness and temperature of all pixels are arranged in rows and columns to obtain a second data frame (i.e., a brightness and temperature frame). By stitching together all the second data frames in chronological order, the preliminary temperature field data can be obtained.

[0073] S500: Construct a feature vector using the first radiation intensity data and preliminary temperature field data, and input the feature vector into the pre-trained error correction strategy model to obtain dynamic compensation parameters.

[0074] Specifically, step S500 may include the following steps S501-S503.

[0075] S501: Expand each first data frame in the first radiation intensity data into a one-dimensional vector, and concatenate it with the one-dimensional vector expanded from the second data frame corresponding to the first data frame to form the original feature vector.

[0076] It is understood that, in the embodiments of this application, each first data frame (a two-dimensional matrix of H rows and W columns) can be expanded into a one-dimensional vector of length H×W. V rad The corresponding second data frame (a two-dimensional matrix of H rows and W columns) is also expanded into a one-dimensional vector. V temp The two vectors are concatenated to form the original feature vector. V raw =[ V rad , V temp Its length is 2×H×W.

[0077] S502: Normalize the original feature vector so that the values ​​of each dimension fall within the range of [0,1] to obtain the feature vector.

[0078] The embodiments of this application can be normalized using the max-min standardization method.

[0079] S503: Input the feature vector into the pre-trained error correction strategy model to obtain dynamic compensation parameters.

[0080] In this embodiment, the normalized feature vector can be input into a pre-trained error correction strategy model. This model is a deep neural network, employing a hybrid structure of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM), and is trained offline. The model's input is the feature vector, and its output is two two-dimensional matrices of size H×W: an emissivity dynamic correction factor matrix, where each element ranges from 0.1 to 1.0, used to correct the deviation between the actual emissivity of the target surface and the blackbody assumption; and a path radiation interference compensation matrix, with units identical to the radiance value, used to deduct interference from path radiation such as smoke and fly ash.

[0081] In some implementations, the training process of the error correction strategy model is as follows: ① Model Structure: This embodiment employs a hybrid model combining convolutional neural networks and long short-term memory networks. The convolutional neural network is used to extract spatial features (e.g., temperature distribution patterns on the furnace tube surface), while the long short-term memory network is used to capture trends in the time series. The model input is a feature vector (2×H×W in length), and the output consists of two two-dimensional matrices (emissivity dynamic correction factor matrix and path radiation interference compensation matrix).

[0082] ② Constructing a training dataset: Data sources can be derived from simulation data. Specifically, a physical model of furnace radiation transfer can be established, including parameters such as random variations in target emissivity, random disturbances in flue gas, and variations in mirror tube temperature, generating a large amount of paired data (input: radiance + assumed blackbody temperature; output: actual emissivity and path interference values). A total of 100,000 sets of simulation samples were generated. Data sources can also be derived from experimental data. Specifically, high-precision thermocouples can be installed on actual industrial furnaces as reference thermometers to measure the actual temperature at different locations on the furnace wall, while simultaneously collecting raw radiation intensity data output from an infrared thermal imager. A total of 5,000 sets of experimental data were collected under different operating conditions (load variations, fuel type variations, and different coking degrees).

[0083] After that, the data can be labeled to obtain the final dataset.

[0084] ③ Constructing the loss function: During model training, this embodiment of the application can use a multi-task loss function, the expression of which is: L=Lε+λ L ·L L .

[0085] in, L This represents the total loss value (scalar), used to guide the updating of model parameters; Yes This represents the mean square error loss of the emissivity correction factor; l L This represents the balance coefficient (hyperparameter, for example, a value of 0.5) for path interference compensation loss. L L This represents the mean square error loss of the path interference compensation coefficient.

[0086] ④ Training process: 1. Normalize the feature vectors of all training samples to the interval [0,1].

[0087] 2. Use the Adaptive Moment Estimation (Adam) optimizer with an initial learning rate of 0.001 and a batch size of 32.

[0088] 3. Divide the training data into 80% training set, 10% validation set, and 10% test set.

[0089] 4. Monitor the validation set loss during training. If the loss does not decrease for 10 consecutive epochs, stop the training prematurely.

[0090] 5. Perform 8-bit integer (INT8) quantization compression on the model so that it can be deployed to the central control server for real-time inference.

[0091] ⑤ Model Performance Verification: On the simulation test set, the average absolute error between the model's predicted emissivity and the simulated actual value was 0.03, and the average absolute error between the path interference compensation value and the simulated actual value was 12 W·m⁻²·sr⁻¹, indicating that the model has good fitting ability. In the actual furnace, through thermocouple comparison verification, the deviation between the system's final output target temperature and the thermocouple measurement value was within ±8℃, meeting the requirements of industrial applications.

[0092] S600: Corrects the initial temperature field data based on dynamic compensation parameters to obtain the target temperature field data; and generates combustion optimization control commands for the furnace based on the target temperature field data and sends them to the edge computing nodes.

[0093] Understandably, the central control server 300 corrects the radiance value using dynamic compensation parameters for each second data frame (brightness-temperature frame), and then recalculates the temperature using the Planck inverse function to obtain the true temperature (target temperature field data). This specifically includes the following steps S601-S603.

[0094] S601: For each pixel in each second data frame, calculate its corresponding target temperature according to the following fourth formula; Fourth formula: ; in, Represents pixel ( i , j The target temperature is λ, the center response wavelength of the infrared thermal imager is λ, c1 is the first radiation constant, and c2 is the second radiation constant. For pixels ( i , j The spectral radiance value of ). This is the emissivity dynamic correction factor matrix; This is the path radiation interference compensation matrix.

[0095] Understandably, the formula first subtracts the path radiance interference compensation from the original radiance (spectral radiance value), then multiplies it by the emissivity correction factor, and finally substitutes it into the Planck inverse function to calculate the true temperature.

[0096] S602: Arrange all the calculated target temperatures frame by frame according to the row and column order of the pixels to form a two-dimensional matrix of H rows and W columns, and then obtain the target temperature frame for each frame. S603: The target temperature frames are stitched together in chronological order to obtain the target temperature field data.

[0097] Furthermore, step S603 may be followed by steps S604-S607.

[0098] S604: For each target temperature frame within the first time window, calculate the gradient magnitude of each pixel and mark pixels with gradient magnitudes greater than a preset gradient threshold as edge points.

[0099] In this embodiment, the gradient magnitude specifically refers to the temperature gradient magnitude between each pixel and its neighboring pixels, which can be calculated using the central difference method. The preset gradient magnitude can be 10℃ / pixel.

[0100] For pixels ( i , j Let the target temperature of this pixel be... T ( i , j The central difference method is as follows: First, calculate the horizontal gradient: Gx ( i , j )= T ( i , j +1)- T ( i , j -1) represents a pixel ( i , j Temperature difference between adjacent pixels on the right and left.

[0101] Next, calculate the vertical gradient: Gy ( i , j )= T ( i +1, j )- T ( i -1, j ), representing a pixel ( i , j Temperature difference between adjacent pixels on the lower and upper sides.

[0102] Then the gradient magnitude of that pixel is synthesized. .

[0103] like G ( i, j If the gradient value is greater than a preset gradient threshold (e.g., 10°C / pixel), then the pixel is marked as an edge point. For boundary pixels ( i =1 or i = H or j =1 or j =W), which cannot use central difference, can be replaced by forward or backward difference, or the boundary points can be ignored and not marked.

[0104] S605: Select interconnected edge point regions as candidate thermal anomaly regions and calculate the target features of each candidate thermal anomaly region; the target features include the number of pixels in the region and the average temperature of the region.

[0105] In this embodiment, all pixels marked as edge points can be subjected to connected component analysis (using the 8-neighborhood connectivity criterion), and interconnected edge point regions can be considered as candidate thermal anomaly regions. For each candidate region, the following calculations are made: region area: the total number of pixels in the region; region average temperature: the average value of the target temperatures of all pixels in the region.

[0106] S606: The candidate thermal anomaly regions whose target features meet the preset conditions are determined as the actual thermal anomaly regions.

[0107] In this embodiment of the application, the preset conditions are, for example, that the area is ≥5 pixels (to avoid single-point noise interference) and the average temperature of the area is ≥900℃ (higher than the normal furnace tube operating temperature).

[0108] S607: Generates corresponding combustion optimization control commands based on the actual thermal anomaly zone.

[0109] Specifically, step S607 may include the following steps S6071-S6073.

[0110] S6071: Calculate the average of the regional average temperature of all actual thermal anomaly zones to obtain the first average value, and calculate the average of the regional area of ​​all actual thermal anomaly zones to obtain the second average value.

[0111] Understandably, the centralized control server 300 statistically analyzes the regional average temperature and area of ​​all actual thermal anomaly zones, calculates the arithmetic mean of each, and obtains the first and second average values.

[0112] S6072: If the first average value exceeds the preset upper limit of temperature, generate a command to reduce the fuel supply rate or increase the air supply.

[0113] The preset upper limit of temperature can be 1200℃, but this embodiment does not specifically limit it. If the first average value exceeds the preset upper limit of temperature, it indicates that the overall temperature of the furnace is too high. This embodiment can generate a command to reduce the fuel supply rate or increase the air supply.

[0114] S6073: When the first average value does not exceed the preset upper limit of temperature and the second average value exceeds the preset area threshold, generate a furnace soot blowing command or a coking warning signal.

[0115] The preset area threshold can be 100 pixels, and this application embodiment does not specifically limit it. If the first average value does not exceed the preset temperature upper limit, but the second average value exceeds the preset area threshold, it indicates that there is a large-scale thermal anomaly (such as coking) in the furnace, and a furnace soot blowing command or coking warning signal is generated.

[0116] The aforementioned command control server can distribute commands to the edge computing node 200 via the Open Platform Communications Unified Architecture (OPC UA) or the Modbus protocol to achieve closed-loop optimized control.

[0117] It should also be noted that after receiving the combustion optimization control command, the edge computing node 200 can perform corresponding operations. Specifically, when it receives the "reduce fuel supply rate command", the edge computing node 200 sends a closing signal to the fuel regulating valve of the furnace via the fieldbus (such as Modbus or Profibus) to reduce the amount of fuel per unit time; if it receives the "increase air supply command", it sends an increasing frequency signal to the blower frequency converter to increase the amount of combustion air.

[0118] When a "furnace soot blowing command" is received, the edge computing node 200 starts the pre-set soot blowing program, sequentially opening the control valves of the steam or sonic soot blower to blow away the furnace heating surface and remove surface ash and coke.

[0119] When a "coking warning signal" is received, the edge computing node 200 will display a warning pop-up window on the local human-machine interface (HMI) and issue an audible and visual alarm to remind the operator to arrange for coking removal or adjust the combustion conditions.

[0120] In addition, the edge computing node 200 can also upload the command execution results and the real-time response data of the furnace back to the central control server 300 for monitoring the closed-loop effect or further optimizing the model parameters.

[0121] It is understandable that this application primarily corrects the following three types of errors: First, the optical tube 101 itself experiences high-temperature radiation and transmittance attenuation errors: When the optical tube 101 is heated inside the furnace, it generates infrared radiation, which is superimposed on the target signal, leading to falsely high temperature readings. Simultaneously, the transmittance of the tube decreases with increasing temperature, causing signal attenuation. This application addresses this by using a thin-film thermocouple 103 attached to the inner wall of the tube to collect the tube temperature in real time. A preset transmittance-temperature mapping table is used to multiply each first data frame by an attenuation compensation coefficient, thereby eliminating the influence of the tube's own radiation and transmittance variations.

[0122] Second, path radiation interference error: Fly ash, flue gas, and water vapor in the furnace absorb and scatter the target radiation, and these high-temperature media themselves also emit radiation, which is superimposed on the path. This application outputs a path radiation interference compensation matrix through a pre-trained error correction strategy model, and subtracts this value from the original radiance in the correction formula to deduct path interference.

[0123] Third, the error due to dynamic changes in the emissivity of the target surface: Coking layers, oxide scale, etc., on the furnace tube surface alter the emissivity, which changes dynamically with time and operating conditions. Traditional methods, which assume a fixed emissivity, lead to temperature measurement deviations. This application outputs a dynamic emissivity correction factor matrix through an error correction strategy model. This factor is then multiplied by the radiance in the correction formula to adapt to real-time changes in emissivity.

[0124] Furthermore, this application further eliminates random noise interference and improves signal quality by combining variational mode decomposition with energy spectrum entropy screening. The final target temperature field data, after these multiple corrections, significantly improves temperature measurement accuracy, and combustion optimization control commands are automatically generated based on the corrected temperature field to achieve closed-loop control.

[0125] Figure 2 This is a flowchart illustrating the error correction method for furnace infrared imaging temperature measurement scenarios provided in this application embodiment.

[0126] like Figure 2 As shown in the embodiments of this application, an error correction method for furnace infrared imaging temperature measurement is also provided, applied to industrial combustion furnaces; the method includes: S701: Generate an infrared data sequence using the infrared radiation intensity data as the first data frame according to the first preset frequency, and preprocess it; S702: Using the variational mode decomposition algorithm, each first data frame is decomposed into K eigenmode components; S703: Reconstruct the signal of each first data frame using the intrinsic mode components to obtain the first radiation intensity data; S704: Using the inverse function of Planck's blackbody radiation law, the first radiation intensity data is converted into preliminary temperature field data; S705: Construct a feature vector using the first radiation intensity data and preliminary temperature field data, and input the feature vector into the pre-trained error correction strategy model to obtain dynamic compensation parameters; S706: Correct the initial temperature field data based on dynamic compensation parameters to obtain the target temperature field data; and generate combustion optimization control commands for the furnace based on the target temperature field data.

[0127] The error correction system and method for furnace infrared imaging temperature measurement provided in this application have at least the following beneficial effects: By using a thin-film thermocouple 103 to monitor the temperature of the optical tube 101 in real time and compensate for transmittance attenuation, the influence of the tube's own thermal radiation on temperature measurement is effectively eliminated. A variational mode decomposition combined with energy spectrum entropy screening method is used to adaptively extract the dominant thermal radiation signal from strong noise, significantly improving the signal-to-noise ratio. The deep learning-based error correction strategy model can dynamically output emissivity correction factors and path radiation compensation coefficients according to different operating conditions, solving the problem of poor adaptability of traditional fixed parameter correction. Utilizing the corrected high-precision temperature field, the system automatically identifies thermal anomaly zones and provides combustion optimization commands, realizing intelligent closed-loop control, which helps improve furnace thermal efficiency and reduce failure risks.

[0128] In a specific implementation, the present invention also provides a computer storage medium, wherein the computer storage medium may store a program, which, when executed, may include some or all of the steps of the error correction method for furnace infrared imaging temperature measurement scenarios provided by the present invention. The storage medium may be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0129] It is readily understood that, based on the several embodiments provided in this application, those skilled in the art can combine, split, or reorganize the embodiments of this application to obtain other embodiments, none of which exceed the protection scope of this application.

[0130] The above detailed embodiments further illustrate the purpose, technical solution, and beneficial effects of the embodiments of this application. It should be understood that the above are merely specific embodiments of the embodiments of this application and are not intended to limit the protection scope of the embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solutions of the embodiments of this application should be included within the protection scope of the embodiments of this application.

Claims

1. An error correction system for infrared imaging temperature measurement in a furnace, characterized in that, The system is applied to an industrial combustion furnace, wherein the furnace wall of the industrial combustion furnace has an observation hole; the system includes a furnace infrared temperature measurement terminal fixed at the observation hole, an edge computing node communicatively connected to the furnace infrared temperature measurement terminal, and a central control server communicatively connected to the edge computing node; The furnace infrared temperature measurement terminal includes at least: an optical lens tube extending into the observation hole, and an infrared thermal imager coupled to the optical path of the optical lens tube; the infrared thermal imager is used to continuously collect infrared radiation intensity data of the target area inside the furnace at a preset sampling frequency; The edge computing node is configured to: generate an infrared data sequence by taking the infrared radiation intensity data within a first time window as a first data frame according to a first preset frequency, and upload it to the central control server after preprocessing it; The centralized control server is configured to decompose each of the first data frames into K intrinsic mode components using a variational mode decomposition algorithm. The first data frame is reconstructed using the intrinsic mode components to obtain the first radiation intensity data. Using the inverse function of Planck's blackbody radiation law, the first radiation intensity data is converted into preliminary temperature field data; A feature vector is constructed using the first radiation intensity data and the preliminary temperature field data. The feature vector is then input into a pre-trained error correction strategy model to obtain dynamic compensation parameters. The preliminary temperature field data is corrected based on the dynamic compensation parameters to obtain target temperature field data; and combustion optimization control instructions for the furnace are generated based on the target temperature field data and sent to the edge computing node.

2. The error correction system for furnace infrared imaging temperature measurement scenario according to claim 1, characterized in that, The furnace infrared temperature measurement terminal also includes a thin-film thermocouple attached to the inner wall of the optical tube. The thin-film thermocouple is used to collect the real-time temperature of the optical tube at the preset sampling frequency. The edge computing node is further configured to: generate a tube temperature sequence based on the real-time tube temperature within a first time window according to the first preset frequency; wherein each first data frame in the infrared data sequence corresponds one-to-one with each temperature value in the tube temperature sequence on the timestamp. The infrared data sequence is preprocessed using the tube temperature sequence to obtain the preprocessed infrared data sequence. The preprocessing refers to performing tube transmittance attenuation compensation on the infrared data sequence based on the real-time tube temperature data sequence. The tube transmittance attenuation compensation is performed by multiplying each first data frame in the infrared data sequence by the attenuation compensation coefficient related to the real-time tube temperature at the corresponding time according to a preset transmittance-temperature mapping table.

3. The error correction system for furnace infrared imaging temperature measurement scenario according to claim 1, characterized in that, The variational mode decomposition algorithm includes: For a given first data frame, set the value of the number K of the intrinsic mode components, and initialize the center frequency and Lagrange multiplier operator of each intrinsic mode component; The intrinsic modal components and center frequencies corresponding to each first data frame are iteratively updated according to the following first and second formulas: First formula: ； Second formula: ； Where k is the index of the intrinsic mode component, 0 < k ≤ K, n is the index of the iteration number; ω is the angular frequency independent variable, 0 ≤ ω ≤ ω max， ω max =π f s , f s The preset sampling frequency of the infrared thermal imager; This represents the Fourier transform of the k-th intrinsic mode component in the (n+1)th iteration; The Fourier transform of the first data frame; Let i be the Fourier transform of the i-th intrinsic mode component, where i is the index of the intrinsic mode component and i ≠ k; Here, α is the Fourier transform of the Lagrange multiplication operator; α is the penalty factor. The center frequency of the k-th intrinsic mode component; as well as, This represents the center frequency of the k-th intrinsic mode component in the (n+1)th iteration; The power spectral density represents the Fourier transform of the k-th intrinsic mode component; This represents the first moment of the power spectrum; This represents the total energy of the power spectrum; When the relative change of the Fourier transform of the intrinsic mode component in two consecutive iterations is less than the preset convergence threshold, the iteration stops and the intrinsic mode component corresponding to the first data frame is output.

4. The error correction system for furnace infrared imaging temperature measurement scenario according to claim 1, characterized in that, In the step of reconstructing the signal of each of the first data frames using the intrinsic mode components to obtain the first radiation intensity data, the centralized control server is specifically configured as follows: For each first data frame in the infrared data sequence, calculate the energy spectral entropy of its corresponding intrinsic mode component, and mark J intrinsic mode components whose energy spectral entropy is less than or equal to a preset entropy threshold as thermal radiation dominant mode components. The first data frame is reconstructed by linearly superimposing all the dominant thermal radiation mode components corresponding to the same first data frame. The first radiation intensity data is generated based on each of the reconstructed first data frames.

5. The error correction system for furnace infrared imaging temperature measurement scenario according to claim 1, characterized in that, The infrared radiation intensity data is a two-dimensional matrix output by the focal plane array of the infrared thermal imager. The two-dimensional matrix has H rows and W columns, and each element in the two-dimensional matrix is ​​a pixel. The value of each pixel is the spectral radiance value corresponding to that pixel. In the step of converting the first radiation intensity data into preliminary temperature field data using the inverse function of Planck's blackbody radiation law, the centralized control server is specifically configured as follows: For each pixel in each first data frame of the first radiation intensity data, its brightness temperature is calculated according to the following third formula. Third formula: ； in, Represents pixel ( i , j The brightness temperature of the infrared thermal imager is given by λ, where λ is the center response wavelength of the infrared thermal imager, c1 is the first radiation constant, and c2 is the second radiation constant. For pixels ( i , j The spectral radiance value of ); The calculated brightness temperature is arranged frame by frame according to the row and column order of the pixels to form a two-dimensional matrix of H rows and W columns, thereby obtaining the second data frame for each frame. The preliminary temperature field data is obtained by splicing the second data frames in chronological order.

6. The error correction system for furnace infrared imaging temperature measurement scenario according to claim 5, characterized in that, In the step of constructing a feature vector using the first radiation intensity data and the preliminary temperature field data, and inputting the feature vector into a pre-trained error correction strategy model to obtain dynamic compensation parameters, the centralized control server is specifically configured as follows: Each data frame in the first radiation intensity data is expanded into a one-dimensional vector and concatenated with the one-dimensional vector of the second data frame corresponding to the first data frame to form the original feature vector. The original feature vector is normalized so that the values ​​of each dimension fall within the range of [0,1], thus obtaining the feature vector; The feature vector is input into the pre-trained error correction strategy model to obtain the dynamic compensation parameters.

7. The error correction system for furnace infrared imaging temperature measurement scenario according to claim 5 or 6, characterized in that, The dynamic compensation parameters include an emissivity dynamic correction factor matrix and a path radiation interference compensation matrix; in the step of correcting the preliminary temperature field data based on the dynamic compensation parameters to obtain the target temperature field data, the centralized control server is specifically configured as follows: For each pixel in each of the second data frames, the corresponding target temperature is calculated according to the following fourth formula; Fourth formula: ； in, Represents pixel ( i , j The target temperature is given by λ, where λ is the center response wavelength of the infrared thermal imager, c1 is the first radiation constant, and c2 is the second radiation constant. For pixels ( i , j The spectral radiance value of ); This is the emissivity dynamic correction factor matrix; The path radiation interference compensation matrix; The calculated target temperatures are arranged frame by frame according to the row and column order of the pixels to form a two-dimensional matrix of H rows and W columns, thus obtaining each frame of target temperature. The target temperature frames are stitched together in chronological order to obtain the target temperature field data.

8. The error correction system for furnace infrared imaging temperature measurement scenario according to claim 7, characterized in that, In the step of generating combustion optimization control instructions for the furnace based on the target temperature field data, the centralized control server is specifically configured as follows: For each target temperature frame within the first time window, calculate the gradient magnitude of each pixel and mark pixels with gradient magnitudes greater than a preset gradient threshold as edge points. Interconnected edge point regions are used as candidate thermal anomaly regions, and target features of each candidate thermal anomaly region are calculated; the target features include the number of pixels in the region and the average temperature of the region. The candidate thermal anomaly regions whose target features meet the preset conditions are determined as actual thermal anomaly regions; The corresponding combustion optimization control command is generated based on the actual thermal anomaly zone.

9. The error correction system for furnace infrared imaging temperature measurement scenario according to claim 8, characterized in that, In the step of generating the corresponding combustion optimization control command based on the actual thermal anomaly zone, the centralized control server is specifically configured as follows: Calculate the average of the regional average temperature of all the actual thermal anomaly zones to obtain a first average value, and calculate the average of the regional area of ​​all the actual thermal anomaly zones to obtain a second average value; If the first average value exceeds the preset temperature limit, a command to reduce the fuel supply rate or a command to increase the air supply will be generated. When the first average value does not exceed the preset temperature limit and the second average value exceeds the preset area threshold, a furnace soot blowing command or a coking warning signal is generated.

10. An error correction method for infrared imaging temperature measurement in a furnace scenario, characterized in that, Applied to industrial combustion furnaces; the method includes: According to the first preset frequency, infrared radiation intensity data is used as the first data frame to generate an infrared data sequence, and then preprocessed. Using the variational mode decomposition algorithm, each of the first data frames is decomposed into K eigenmode components; The first data frame is reconstructed using the intrinsic mode components to obtain the first radiation intensity data. Using the inverse function of Planck's blackbody radiation law, the first radiation intensity data is converted into preliminary temperature field data; A feature vector is constructed using the first radiation intensity data and the preliminary temperature field data. The feature vector is then input into a pre-trained error correction strategy model to obtain dynamic compensation parameters. The preliminary temperature field data is corrected based on the dynamic compensation parameters to obtain target temperature field data; and combustion optimization control commands for the furnace are generated based on the target temperature field data.

Images