A torch combustion monitoring and optimization control system based on multispectral infrared and visible light imaging

Abstract

The application discloses a torch combustion monitoring and optimization control system based on multispectral infrared and visible light imaging, comprising a multispectral imaging module for synchronously collecting infrared images and visible light images of a torch flame and performing spatial registration to obtain multispectral radiation data of the flame; a physical model calculation module for calculating a first combustion state parameter according to the multispectral radiation data based on a Planck radiation formula and a gray body radiation model; an artificial intelligence calculation module for calculating a second combustion state parameter according to the multispectral radiation data by using a pre-trained multiscale visual Transformer network; a data fusion and correction module for fusing and correcting the first combustion state parameter and the second combustion state parameter by using a Kalman filtering algorithm to output a final combustion state parameter; and a closed-loop control module for dynamically adjusting operation parameters of the torch according to the final combustion state parameter to optimize combustion efficiency.

Description

Technical Field

[0001] This invention belongs to the field of industrial combustion monitoring and control technology, and particularly relates to a flare combustion monitoring and optimization control system based on multispectral infrared and visible light imaging. Background Technology

[0002] Flare systems are crucial equipment in industries such as chemical, petroleum, and natural gas, used for treating process waste gases and venting gases in emergencies. Their core function is to convert harmful gases into relatively harmless carbon dioxide and water vapor through combustion, thereby reducing harm to the environment and human health. However, the combustion efficiency of a flare directly affects the environmental impact of the emitted gases. Low combustion efficiency leads to the emission of large amounts of unburned hydrocarbons, carbon monoxide, particulate matter, and other pollutants, severely polluting the atmosphere. Furthermore, fluctuations in combustion efficiency can affect the operational stability of the flare equipment and even cause safety accidents.

[0003] Flare systems are widely used in the chemical industry. Due to the unique design and operation of flares, determining their Disposal Removal Efficiency (DRE) is extremely challenging. The total amount of pollutants emitted by flares has long been a subject of debate, but a definitive answer remains elusive.

[0004] In 2010, the Texas Commission on Environmental Quality (TCEQ) commissioned the University of Texas at Austin (UT) to conduct a comprehensive study on flare combustion efficiency (CE) and destruction removal efficiency (DRE). This study employed two complementary remote sensing monitoring systems: a Hyper-Cam infrared hyperspectral imager from Telops and a passive / active Fourier transform infrared spectrometer (PFTIR / AFTIR) from IMACC. The study showed that the IMACC system's CE values ​​differed from those of the traditional "sampling measurement method" by approximately 2%-2.5%, with data availability reaching 99%-100%; while the Telops system showed an average difference of 19.9%, with data availability of only 39%. Furthermore, neither system could provide real-time monitoring data, and both were expensive to acquire (especially the Telops equipment). Therefore, a solution for real-time monitoring of flare efficiency and providing feedback to operators for adjustments to operating conditions is currently lacking.

[0005] Currently, the measurement of flare combustion efficiency mainly relies on methods such as sampling analysis, infrared spectral imaging, Fourier transform infrared spectroscopy, video imaging radiometers, and automatic thermocouple disassembly and assembly systems. While these methods can measure combustion efficiency to some extent, they generally suffer from problems such as insufficient measurement accuracy, poor real-time performance, high cost, limited applicability, and lack of closed-loop control capabilities.

[0006] For example, while sampling analysis is accurate, it suffers from drawbacks such as response lag, complex operation, and lack of real-time feedback. Infrared spectral imaging technology, while offering advantages like non-contact and remote measurement, suffers from expensive equipment, low data availability, and large measurement errors. Fourier transform infrared spectroscopy, despite its high measurement accuracy, also suffers from expensive equipment, lack of real-time feedback, and complex installation. While video imaging radiometers offer some real-time capability, their measurement accuracy is still significantly affected by the combustion state. Automatic thermocouple disassembly and assembly systems, while solving the problem of traditional thermocouple maintenance requiring downtime, are contact-based measurements, resulting in short equipment lifespan, high maintenance costs, and limited measurement points.

[0007] Therefore, there is an urgent need for a torch combustion monitoring and optimization control system that can achieve high-precision, real-time measurement and has closed-loop control capabilities to solve the problems existing in the current technology. Summary of the Invention

[0008] One of the objectives of this application is to provide a torch combustion monitoring and optimization control system based on multispectral infrared and visible light imaging, which can achieve high-precision, real-time measurement of key parameters such as torch combustion temperature, combustion efficiency, and gas flow rate.

[0009] The purpose of this application is to provide a flare combustion monitoring and optimization control system based on multispectral infrared and visible light imaging, which can dynamically adjust the flare operating parameters based on measurement results, thereby optimizing combustion efficiency, reducing harmful gas emissions, and improving environmental friendliness and industrial safety.

[0010] The purpose of this application is to provide a torch combustion monitoring and optimization control system based on multispectral infrared and visible light imaging. By constructing a dual-link collaborative mechanism of physical model and artificial intelligence model, and combining multispectral image fusion and Kalman filtering algorithm, high-precision, real-time monitoring and closed-loop optimization control of torch combustion status can be achieved.

[0011] The purpose of this application is to provide a flare combustion monitoring and optimization control system based on multispectral infrared and visible light imaging. The physical model calculation module, based on Planck's radiation formula and the gray body radiation model, combines multispectral image data to calculate flame temperature and combustion efficiency. For measurement results from multiple spectral channels, a weighted Gaussian formula is used for correction to improve calculation accuracy. The combustion efficiency calculation unit calculates combustion efficiency based on flame temperature and gas composition.

[0012] The purpose of this application is to provide a torch combustion monitoring and optimization control system based on multispectral infrared and visible light imaging. The data fusion and correction module uses a Kalman filter algorithm to fuse and correct the outputs of the physical model and the artificial intelligence model, reducing single-measurement errors and improving overall measurement accuracy. The error analysis and correction unit analyzes and corrects measurement errors, further improving measurement accuracy.

[0013] The purpose of this application is to provide a torch combustion monitoring and optimization control system based on multispectral infrared and visible light imaging. It adopts multispectral image fusion technology and acquires multispectral image information of the flame by registering an uncooled infrared camera with a megapixel zoom visible light lens. It can maintain high image quality and measurement accuracy under different combustion conditions. Based on Planck's radiation formula and gray body radiation model, the measurement results of multiple spectral channels are corrected by a weighted Gaussian formula to improve the calculation accuracy.

[0014] The purpose of this application is to provide a torch combustion monitoring and optimization control system based on multispectral infrared and visible light imaging. It proposes a multi-scale visual Transformer network based on the Transformer architecture to extract features from multispectral images and predict combustion parameters, correct measurement errors in the multispectral system, and improve measurement accuracy and generalization ability.

[0015] This application provides a flare combustion monitoring and optimization control system based on multispectral infrared and visible light imaging, characterized in that it includes:

[0016] The multispectral imaging module is used to simultaneously acquire infrared and visible light images of the torch flame and perform spatial registration to obtain multispectral radiation data of the flame.

[0017] The physical model calculation module is used to calculate the first combustion state parameters based on the Planck radiation formula and the gray body radiation model, according to the multispectral radiation data.

[0018] The artificial intelligence computing module uses a pre-trained multi-scale visual Transformer network to calculate the second combustion state parameter based on the multispectral radiation data.

[0019] The data fusion and correction module is used to fuse and correct the first combustion state parameters and the second combustion state parameters using a Kalman filter algorithm to output the final combustion state parameters; and

[0020] The closed-loop control module is used to dynamically adjust the torch's operating parameters to optimize combustion efficiency based on the final combustion state parameters.

[0021] This application provides a torch combustion monitoring and optimization control system based on multispectral infrared and visible light imaging, including a multispectral imaging module, a physical model calculation module, an artificial intelligence calculation module, a data fusion and correction module, a closed-loop control module, and a data storage and visualization module.

[0022] In one possible implementation, the multispectral imaging module includes an uncooled infrared camera, a visible light camera, and a multispectral registration unit. The uncooled infrared camera operates in wavelengths covering 3-5 micrometers and 8-12 micrometers, and is used to acquire infrared radiation images of the flame; the visible light camera is used to acquire visible light images of the flame; and the multispectral registration unit is used to spatially register the infrared radiation images and the visible light images to generate an aligned array of flame positions.

[0023] In one possible implementation, the torch combustion monitoring and optimization control system includes a temperature calculation module that integrates a multi-wavelength feature fusion model based on a Transformer architecture. This multi-wavelength feature fusion model includes an encoding layer, a multi-head attention mechanism layer, and a decoding layer. The encoding layer encodes the radiation intensity data of the red, green, and blue light bands and the infrared band into feature vectors. The multi-head attention mechanism layer achieves cross-spectral feature correlation through self-attention weight allocation. The decoding layer outputs self-corrected flame temperature distribution data. The Transformer model of the multi-wavelength feature fusion model is trained end-to-end, receiving the original spectral data as input and directly generating the corrected temperature distribution matrix as output.

[0024] In one possible implementation, the physical model calculation module includes a weighted Gaussian correction unit, which establishes a weight matrix based on the response sensitivity coefficients of each spectral channel and uses the Levenberg-Marquardt algorithm to perform nonlinear optimization on the first combustion state parameters calculated based on multi-band, so that the system measurement error is controlled within ±2%.

[0025] On the other hand, a method for optimizing flare combustion control is provided. This method includes the following steps:

[0026] Step 1: Acquire flame radiation data using a multispectral imaging system;

[0027] Step 2: Calculate the initial temperature distribution based on Planck's formula and the gray body model;

[0028] Step 3: Multispectral data fusion and correction are performed using the weighted Gaussian algorithm and the Levenberg-Marquardt algorithm;

[0029] Step 4: Calculate combustion efficiency using the Transformer model;

[0030] Step 5: Input the corrected temperature data into the combustion optimization control module;

[0031] Step 6: Generate control commands based on the physical lookup table and flow meter data;

[0032] Step 7: Adjust the combustion parameters via the actuator and return to the monitoring step.

[0033] Thirdly, this application provides an electronic device including a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the flare combustion optimization control method as described in the second aspect. The electronic device can be integrated into the local control cabinet of the flare device or deployed in a server cluster of a remote monitoring center.

[0034] Compared with the prior art, this application has at least one or more of the following advantages and positive effects:

[0035] 1. High-precision measurement: By using infrared and visible light image fusion technology, combined with physical models and artificial intelligence algorithms, the measurement accuracy of flame temperature and combustion efficiency is significantly improved;

[0036] 2. Real-time performance and dynamic response: The system has real-time image acquisition and processing capabilities, and can quickly respond to changes in combustion status to achieve dynamic closed-loop control;

[0037] 3. Closed-loop optimization control: Based on combustion efficiency feedback, the flare operating parameters are dynamically adjusted to improve combustion efficiency and reduce pollutant emissions;

[0038] 4. High system robustness: The dual-link model fusion mechanism and Kalman filter algorithm are adopted to effectively improve the stability and reliability of the system under complex working conditions;

[0039] 5. Environmentally friendly and energy-saving: By optimizing combustion control, emissions of unburned hydrocarbons and harmful gases are reduced, and energy utilization efficiency is improved;

[0040] 6. Remote monitoring and data management: Supports remote data access and visualization, facilitating industrial site management and maintenance. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings of the embodiments will be briefly introduced below. Obviously, the drawings described below only involve some embodiments of this application and are not intended to limit this application.

[0042] Figure 1 This is a schematic diagram of the basic architecture of a torch combustion monitoring and optimization control system based on multispectral infrared and visible light imaging according to some embodiments of this application.

[0043] Figure 2 This is a schematic diagram of a dual-link architecture system in which the physical model and the artificial intelligence model work in parallel, according to some embodiments of this application.

[0044] Figure 3A This is a schematic diagram of an optimized system integration scheme for a torch combustion monitoring and optimization control system according to some embodiments of this application.

[0045] Figure 3B This is a schematic diagram illustrating the operation of an improved PID controller according to some embodiments of this application.

[0046] Figure 3C This is a schematic diagram of system integration and dual-stream fusion sensing according to some embodiments of this application.

[0047] Figure 4 This is a schematic diagram of the workflow of a torch combustion monitoring and optimization control system according to some embodiments of this application.

[0048] Figure 5 This is a schematic diagram of a torch combustion monitoring and optimization control method based on multispectral infrared and visible light imaging, according to some embodiments of this application.

[0049] In the diagram: 10. Multispectral Imaging Module; 101. Uncooled Infrared Camera; 102. Visible Light Camera; 103. Multispectral Registration Unit; 104. Dynamic Spectrum Selection Module; 111. Visual Embedding Module; 112. Infrared Embedding Module; 113. Multi-head Attention Mechanism; 114. Layer Normalization Module; 115. Feedforward Network Module; 116. Random Deactivation Module; 117. Fusion Output Module; 118. Loss Function Module; 140. Image Acquisition and Transmission Module; 1041. Tunable Filter Array; 1042. Principal Component Analysis Unit; 20. Physical Model Calculation Module; 201. Planck Formula Calculation Unit; 202. Weighted Gaussian Correction Unit; 203. Combustion Efficiency Calculation Unit; 30. Closed-Loop Control Module; 301. Physical Lookup Table; 302. Interpolation Calculation Unit; 303. Difference Calculation and Basic Control Unit; 304. Actuator Interface; 305. Quality... Flow meter; 306, Combustion additive control valve; 307, Improved PID controller; 307a, Feedforward compensation unit; 307b, Dynamic weight adjustment unit; 307c, Multivariable coupled control interface; 307d, Emission constraint optimization unit; 307e, Hybrid control strategy generation unit; 310, Multi-scale visual Transformer network; 3041, Flow control interface; 40, Auxiliary function module; 401, Environmental compensation unit; 402, Data storage unit; 403, Ambient temperature and humidity sensor; 404, Atmospheric transmittance calculation unit; 50, Artificial intelligence calculation module; 60, Data fusion and correction module; 601, Physical link calibration subunit; 602, Artificial intelligence link subunit; 603, Adaptive parameter adjustment subunit; 604, Dual-link output fusion subunit; 605, Online calibration triggering subunit; 80, Data storage and visualization module. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings showing multiple embodiments according to this application. It should be understood that the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments described in this application without creative effort will fall within the scope of protection of this application.

[0051] In this application, the term "implementation" means that a specific feature, structure, or characteristic described in connection with an implementation may be included in at least one implementation of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same implementation, nor is it a separate or alternative implementation mutually exclusive with other implementations. It will be explicitly and implicitly understood by those skilled in the art that the implementations described in this application can be combined with other implementations.

[0052] As mentioned above, it should be emphasized that when the term "comprising / including" is used in this specification, it is used to explicitly indicate the presence of the stated feature, integer, step, or component, but does not exclude the presence or addition of one or more other features, integers, steps, components, or groups of features, integers, steps, or components. As used in this application, the singular forms "a," "an," and "the" also include the plural forms, unless the context clearly indicates otherwise.

[0053] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used in the description of this application is for the purpose of describing particular embodiments only and is not intended to limit this application; the terms "comprising," "including," "having," "containing," "comprise," etc., in the description, claims, and accompanying drawings of this application are open-ended terms. Therefore, "comprising," "including," or "having" refers to, for example, a method or apparatus having one or more steps or elements, but is not limited to having only these one or more elements. The terms "first," "second," etc., in the description, claims, or accompanying drawings of this application are used to distinguish different objects, not to describe a specific order or hierarchy. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0054] Concept Definition

[0055] To facilitate understanding, some concepts involved in the embodiments of the present invention will be explained below.

[0056] Multispectral infrared and visible light imaging system: This refers to the core device that uses an uncooled infrared camera and a visible light zoom camera to work together to acquire flame radiation data and calculate temperature distribution. The system uses multispectral registration technology to achieve spatial alignment of infrared and visible light images, forming a flame position array.

[0057] Weighted Gaussian Correction Algorithm: This algorithm addresses the spectral response discrepancies in multispectral vector measurements by introducing a weight matrix to fuse the measurement results from each band. Specifically, it establishes the response sensitivity coefficient matrix for each spectral channel and uses the Levenberg-Marquardt algorithm to perform nonlinear optimization on the multispectral temperature calculation results, ensuring that the system measurement error reaches the same level as VISR equipment, preferably within ±2%.

[0058] Physical lookup table: This is a database that stores the mapping relationship between fuel type, combustion conditions, exhaust gas composition, and optimal combustion parameters. When the current operating parameters are detected to match a known parameter combination in the physical lookup table, the corresponding optimal control parameters are automatically invoked to achieve closed-loop feedback control.

[0059] To better understand the present invention, the system architecture and workflow of embodiments of the present invention will be described below with reference to the accompanying drawings.

[0060] Example 1: Flare Combustion Monitoring and Optimization Control System

[0061] like Figure 1 As shown, this embodiment provides a flare combustion monitoring and optimization control system based on multispectral infrared and visible light imaging, enabling real-time monitoring and basic control of the flare combustion status. This flare combustion monitoring and optimization control system includes a multispectral imaging module 10, a physical model calculation module 20, a closed-loop control module 30, and an auxiliary function module 40.

[0062] The multispectral imaging module 10 includes an uncooled infrared camera 101 and a visible light camera 102. The visible light camera 102 preferably has a lens focal length range of 25-100mm and an operating wavelength range of 400-700nm. In flame monitoring, it primarily acquires visible light radiation information in the 0.4-0.7μm range for collecting visible light images of the flame. In some optional embodiments, the uncooled infrared camera 101 operates in wavelengths covering 3-5 micrometers and 8-12 micrometers to acquire infrared radiation images of the flame. The multispectral imaging module 10 also includes a multispectral registration unit 103, which combines the image data acquired by the uncooled infrared camera 101 and the visible light camera 102 to generate a flame position array.

[0063] In this embodiment, an uncooled infrared camera 101 and a visible light camera 102 simultaneously acquire flame radiation data, and spatial alignment is performed by a multispectral registration unit 103 to generate a flame position array. This scheme establishes a flame radiation characteristic database through multispectral position registration, providing accurate multispectral radiation data for subsequent temperature calculations and combustion efficiency analysis. The physical model calculation module 20 includes a Planck formula calculation unit 201, a weighted Gaussian correction unit 202, and a combustion efficiency calculation unit 203, wherein the combustion efficiency calculation unit 203 is preferably a Transformer model. Based on Planck's formula and the gray body radiation model, the physical model calculation module 20 receives radiation intensity data from the red, green, and blue light bands and the infrared band, and obtains the flame temperature distribution matrix through multi-wavelength combination calculations.

[0064] refer to Figure 3A As shown, in some optional embodiments, the Planck formula calculation unit 201 is based on the multispectral image data collected by the multispectral imaging module 10. The flame temperature calculation model is based on Planck's radiation formula and the gray body radiation model. Based on Planck's radiation law and the gray body radiation model, the flame temperature is calculated through the relationship between single-wave radiation energy and parameters such as emissivity, wavelength, and temperature. The formula is as follows:

[0065]

[0066] in, It is a single-wave radiation energy. For emission rate, For wavelength, For temperature, Let be Planck's constant. At the speed of light, is the Boltzmann constant.

[0067] For high-temperature flames, the calculation can be simplified using the Wien approximation formula:

[0068]

[0069] The flame temperature calculation model is based on Planck's radiation formula. For high-temperature flames, Wien's approximation can be used to simplify the calculation. For gray-body radiation sources like flames, the actual radiation intensity is equal to the product of the spectral emissivity and the blackbody radiance.

[0070] Furthermore, to address the issue of spectral response differences in multispectral vector measurements, the weighted Gaussian correction unit 202 establishes a weighting matrix based on the response sensitivity coefficients of each spectral channel:

[0071]

[0072] The weight matrix includes weights for the infrared long-wave band, infrared mid-wave band, visible red band, and visible green band, with typical values ​​of 0.4, 0.3, 0.2, and 0.1, respectively.

[0073] in The signal-to-noise ratio of each band is dynamically adjusted (e.g., infrared band weighting). Visible light band weight The Levenberg-Marquardt algorithm is used to perform nonlinear optimization on the multispectral temperature calculation results, so that the system measurement error reaches the same level as the VISR equipment, for example, the optimal range is within ±2%.

[0074] In one optional embodiment, for monitoring the dynamic combustion process of the torch to be monitored, the weighted Gaussian correction unit 202 uses a Kalman filter to perform state estimation on multiple time-series measurement data to achieve real-time updates of the torch's flame temperature field.

[0075] The combustion efficiency calculation unit 203 calculates the combustion efficiency based on flame temperature distribution and combustion theory. The unit obtains the ratio of the difference between the current average temperature and the ambient temperature to the difference between the theoretical combustion temperature and the ambient temperature, calculated using the following formula:

[0076]

[0077] in, T represents combustion efficiency. avg T represents the average flame temperature obtained through multispectral measurements. env For ambient temperature, T theoretical The theoretical combustion temperature is determined based on fuel type and combustion conditions. A combustion efficiency (CE) ≥ 95% is considered a high-efficiency combustion state, while a CE < 85% is considered an inefficient combustion state. Furthermore, the combustion efficiency calculation unit 203 integrates a multi-wavelength feature fusion model based on a Transformer architecture. This multi-wavelength feature fusion model includes an encoding layer, a multi-head attention mechanism layer, and a decoding layer. The encoding layer encodes the radiation intensity data of the red, green, and blue light bands and the infrared band into feature vectors. The multi-head attention mechanism layer achieves cross-spectral feature correlation through self-attention weight allocation. The decoding layer outputs self-corrected flame temperature distribution data. The Transformer model of this multi-wavelength feature fusion model receives the original spectral data as input and directly generates the corrected temperature distribution matrix as output.

[0078] Furthermore, the closed-loop control module 30 includes a physical lookup table 301, an interpolation calculation unit 302, a difference calculation and basic control unit 303, and an actuator interface 304. The closed-loop control module 30 is further connected to a mass flow meter 305 and a combustion additive control valve 306. The preferred measurement accuracy of the mass flow meter 305 is ±1%FS, and its output signal is used to adjust the combustion additive flow rate via a PID (proportional-integral-derivative) controller.

[0079] The physical lookup table 301 of the closed-loop control module 30 stores the mapping relationship between fuel type, combustion conditions, exhaust gas composition, and optimal combustion parameters. The table structure includes fields such as fuel type, combustion efficiency range, optimal air-fuel ratio, fuel flow coefficient, and combustion air flow rate, covering optimal control parameters for different fuel types (natural gas, liquefied petroleum gas, refinery gas, etc.) under different combustion efficiency ranges. When the current operating condition parameters match a known parameter combination in the physical lookup table 301, the closed-loop control module 30 automatically calls the corresponding optimal control parameters to achieve closed-loop feedback control. For example, in dynamic combustion control, the closed-loop control module 30 calls the physical lookup table 301 to match the current fuel type (e.g., natural gas) with combustion conditions (e.g., lean / rich oxygen) to generate optimal control parameters, such as combustion air flow rate. .

[0080] In an optional embodiment, the closed-loop control module 30 is further configured with an interpolation calculation unit 302. When the current operating condition parameters do not match a completely corresponding parameter combination in the physical lookup table 301, the interpolation calculation unit 302 uses a bilinear interpolation algorithm to perform linear interpolation calculation based on the two nearest neighbor known parameter combinations to generate control parameters adapted to the current operating condition. For example, if the current operating condition does not match the parameters in the physical lookup table 301, the interpolation calculation unit 302 performs linear interpolation calculation based on the two nearest neighbor known parameters, such as... , Perform bilinear interpolation to generate adaptation parameters, such as For extreme operating conditions that exceed the known parameter range, a conservative control strategy using the nearest neighbor parameter combination is adopted.

[0081] The difference calculation and basic control unit 303 receives the target control parameters generated by the interpolation calculation unit 302, calculates the deviation from the current actual parameters, and generates a control output using a proportional adjustment algorithm, with a proportional gain. = 1.2. When the deviation exceeds ±10%, this unit limits the rate of change of the control output to no more than 5% / s to ensure system stability.

[0082] The actuator interface 304 is the output interface of the closed-loop control module 30. It connects the mass flow meter 305 and the combustion additive control valve 306, converts control commands into standard signal outputs, and realizes precise regulation of fuel flow and combustion air flow.

[0083] The auxiliary function module 40 further includes an environmental compensation unit 401 and a data storage unit 402. The environmental compensation unit 401 connects to an ambient temperature and humidity sensor 403 and an atmospheric transmittance calculation unit 404. The atmospheric transmittance calculation unit 404, based on the MODTRAN radiative transfer model, calculates the atmospheric attenuation coefficient in real time and feeds it back to the physical model calculation module 20, achieving dynamic compensation of the measurement results by environmental factors. The data storage unit 402 uses a database to store historical data, including timestamps, temperature distribution, combustion efficiency, control parameters, and other information, providing data support for system operation analysis and fault diagnosis.

[0084] Example 2: Dual-link Enhancement Detection System

[0085] Based on the infrastructure of Embodiment 1, this embodiment further expands the system functionality. For example... Figure 2 As shown, this embodiment further introduces an artificial intelligence computing link, forming a dual-link architecture where the physical model and the artificial intelligence model work in parallel, significantly improving measurement accuracy and system robustness. Based on Embodiment 1, this enhanced monitoring system further includes an artificial intelligence computing module 50, a data fusion and correction module 60, and expands the functionality of the multispectral imaging module 10.

[0086] The multispectral imaging module 10 further includes a dynamic spectral selection module 104, which comprises a tunable filter array 1041 and a principal component analysis unit 1042. The tunable filter array 1041 is preferably an acousto-optic tunable filter (AOTF) with a wavelength range of 3-8 μm and a spectral resolution of less than or equal to 10 nm. The principal component analysis unit 1042 optimizes the spectral band combination in real time, dynamically selecting the optimal spectral band combination based on the current fuel type (e.g., natural gas / liquefied petroleum gas / heavy hydrocarbons) and combustion state (e.g., complete / incomplete combustion). In some optional embodiments, for example, 4.3 μm is suitable for monitoring the CO2 peak, 3.4 μm is suitable for monitoring the CH group absorption band, and 1.6 μm is suitable for monitoring the H2O absorption band.

[0087] Specifically, the principal component analysis unit 1042 calculates the contribution of each principal component using a principal component analysis algorithm. The first and second principal components are linear combinations of each wavelength. Optimal bands are selected based on different combustion states: 4.3 μm and 1.6 μm bands are selected for complete combustion, while 3.4 μm and 2.0 μm bands are added for incomplete combustion. Compared to a fixed band design, this dynamic spectral selection module 104 optimizes the band configuration in real time through online spectral analysis, reducing smoke interference and improving the signal-to-noise ratio.

[0088] The artificial intelligence computing module 50 employs a multi-scale visual Transformer network, which performs comprehensive processing of multispectral radiation data and joint estimation of combustion efficiency and gaseous substance concentration to obtain a second combustion state parameter. This multi-scale visual Transformer network includes a multi-scale feature extraction module, a pooling attention mechanism, local and global attention mechanisms, and a feedforward neural network.

[0089] This pooling attention mechanism reduces resolution by pooling the query tensor, key tensor, and value tensor, and reduces computational complexity by pooling the query, key, and value tensors. Specifically, it reduces resolution by pooling the query tensor Q and reduces computational complexity by pooling the key-value tensor (K,V). The calculation formula is as follows:

[0090] (5)

[0091] Here, Q, K, and V represent the query, key, and value tensors, respectively, and d is the feature dimension. This pooling attention mechanism significantly reduces computational complexity while maintaining feature expressiveness by pooling Q, K, and V.

[0092] The local and global attention mechanisms include: local attention that performs self-attention computation within an n×n window, global attention that achieves global feature extraction through a sparse uniform grid, and axial attention that handles long-range dependencies between channels, where n is a natural number ≥ 4.

[0093] Furthermore, the artificial intelligence computing module 50 integrates a dynamic axial attention module, which achieves a dynamic balance between spatial resolution and computational efficiency through learnable grid partitioning parameters. Specifically, it introduces a grid partitioning parameter α based on window attention. When α=0.5, the feature map is divided into 2×2 grids and attention weights are calculated independently.

[0094] The data fusion and correction module 60 receives output data from the physical model calculation module 20 and the artificial intelligence calculation module 50, respectively, and uses an adaptive algorithm to calibrate the two calculation links in real time. The data fusion and correction module 60 includes a physical link calibration subunit 601, an artificial intelligence link subunit 602, an adaptive parameter adjustment subunit 603, a dual-link output fusion subunit 604, and an online calibration triggering subunit 605.

[0095] The physical link calibration subunit 601 is designed using a modified Wien radiation formula. This modified Wien radiation formula adds an emissivity temperature dependence correction term and an environmental radiation compensation term to the standard Wien approximation formula. The calculation formula is as follows:

[0096] (6)

[0097] in This is the corrected spectral radiance. This is a temperature-dependent emissivity correction function. For the environmental radiation compensation term, h is Planck's constant, c is the speed of light, λ is the wavelength, k is Boltzmann's constant, and T is the absolute temperature. Furthermore, this physical link calibration subunit 601, based on the output of the first-layer physical model and combined with historical deviation data, calculates the systematic deviation of the physical model using a moving average algorithm to achieve the calibration coefficient K. phy Dynamically updated, the calculation formula is as follows

[0098] (7)

[0099] Where K phy(t) K is the calibration coefficient at the current time t. phy(t-1) The calibration coefficient from the previous moment. The learning rate is T, which ranges from 0.01 to 0.05. ref For reference temperature value, T phy The temperature value output by the physical model, (T) ref - T phy The temperature deviation is used to calibrate the physical model output of the first layer in real time based on the above formula. Furthermore, the physical link calibration subunit 601 employs an improved Kalman filter to fuse the calibrated physical model output with other sensor data. Its weighting coefficients are dynamically adjusted according to combustion stability. In some optional embodiments, the weighting scheme prioritizes the artificial intelligence model output when the combustion efficiency CE ≥ 95% and prioritizes the physical model output when CE < 85%, ensuring measurement reliability.

[0100] The AI ​​link subunit 602 employs the xViT++ spatiotemporal attention network mechanism to monitor the stability of the AI ​​model's output. For example, when the standard deviation of 10 consecutive sampling points exceeds 20K, a calibration procedure is initiated, and the bias parameters of the last layer of the network are fine-tuned through the backpropagation algorithm. The AI ​​link subunit 602 integrates a dynamic axial attention module, which achieves a dynamic balance between spatial resolution and computational efficiency through learnable grid partitioning parameters. Specifically, it introduces a grid partitioning parameter α based on window attention. In some optional embodiments, when α=0.5, the feature map is divided into ≥2×2 grids and attention weights are calculated independently.

[0101] The adaptive parameter adjustment subunit 603, as the core parameter management unit of the data fusion and correction module 60, has its input terminals connected to the feedback output terminals of the physical link calibration subunit 601 and the artificial intelligence link subunit 602, respectively, receiving combustion condition monitoring data from the multispectral imaging module 10. The adaptive parameter adjustment subunit 603 dynamically adjusts the key parameters of each calibration subunit based on the magnitude of combustion efficiency changes, flame temperature fluctuation rate, and differences in dual-link calibration accuracy. It dynamically adjusts calibration parameters according to changes in combustion conditions; when the combustion efficiency change exceeds 5%, the applicability of the calibration coefficients is reassessed, and parameter reset is triggered if necessary.

[0102] During parallel processing of the two links, the dual-link output fusion subunit 604 uses a weighted average algorithm to fuse the outputs calibrated by the physical link calibration subunit 601 and the artificial intelligence link subunit 602. The weighting coefficients are dynamically allocated based on the historical accuracy of each link. The weights are dynamically adjusted according to the combustion efficiency (CE): when CE ≥ 95%, w = 0.3, emphasizing the artificial intelligence model; when CE < 85%, w = 0.7, emphasizing the physical model.

[0103] In some optional embodiments, the online calibration triggering subunit 605 is equipped with a laser-induced fluorescence (LIF) calibration module, whose triggering conditions include: temperature fluctuations exceeding a threshold (e.g., exceeding ±50K / second), or dust concentrations exceeding a preset threshold (e.g., 500mg / m³). When the temperature fluctuation exceeds the threshold, the laser-induced fluorescence (LIF) calibration module is automatically activated to perform instantaneous spectral calibration.

[0104] To further explain, the dual-link output fusion subunit 604 of the data fusion and correction module 60 is used to fuse and correct the first combustion state parameters output by the physical model calculation module 20 and the second combustion state parameters output by the artificial intelligence calculation module 50 using a Kalman filter algorithm, so as to output the final combustion state parameters. The data fusion and correction module 60 achieves dual-model data fusion through the Kalman filter unit of the dual-link output fusion subunit 604, specifically including:

[0105] A state-space model is established, and the first combustion state parameters output by the physical model calculation module 20 are used as the predicted values ​​of the system state. The second combustion state parameter output by the artificial intelligence calculation module 50 is used as the observation value z. k ;

[0106] The observed noise covariance matrix R is dynamically adjusted based on the combustion efficiency CE: when CE ≥ 95%, the value of R is reduced to improve the Kalman gain K. k For observations biased towards artificial intelligence models, when CE < 85%, increasing the R value increases the Kalman gain K. k Predictions are biased towards physical modeling.

[0107] The Kalman filter fusion unit uses the Kalman filter algorithm to fuse and correct the outputs of the physical model calculation module 20 and the artificial intelligence calculation module 50, thereby reducing single measurement errors and improving overall measurement accuracy.

[0108] The Kalman filter algorithm is used to fuse and update measurement results of a multispectral system at different time points.

[0109] Kalman gain is calculated based on the observation matrix H, the observation noise covariance matrix R, and the current state covariance matrix P. k|k-1 Calculate the optimal Kalman gain:

[0110] (8),

[0111] Its state update equation is as follows:

[0112]

[0113] in, Estimate the state at the current moment. The current observation value. For Kalman gain, For the observation matrix, To observe the noise covariance matrix, This is the state estimation covariance matrix.

[0114] This Kalman filter algorithm enables the fusion and updating of measurement results from the multispectral imaging module 10 at different time points, reducing single measurement errors and improving overall measurement accuracy. This method can control single measurement errors to within one percent and overall errors to below one percent, achieving levels approaching or exceeding those of VISR equipment.

[0115] When the output deviation between the physical model and the artificial intelligence model exceeds a preset threshold, the data fusion and correction module 60 triggers the online calibration trigger subunit 605 to start the laser-induced fluorescence calibration program.

[0116] The artificial intelligence (AI) calculation module 50 and the physical model calculation module 20 form a dual-link verification mechanism. The AI ​​calculation module 50 employs an AI multi-physics factor link based on the Transformer architecture, offering high accuracy; the physical model calculation module 20 uses a physical derivation link based on the radiative transfer model, exhibiting strong generalization capabilities. Through cross-verification of the two links, the system can automatically select the output result that best meets industrial requirements. When the combustion efficiency is greater than or equal to 95%, the system prioritizes the AI ​​model output; when the combustion efficiency is less than 85%, the system prioritizes the physical model output, ensuring measurement reliability.

[0117] Through the above technical solution, this embodiment achieves a significant performance improvement compared to embodiment one. The accuracy of temperature measurement, combustion efficiency measurement, and system response time can be improved, and the environmental adaptability is significantly enhanced. It can maintain stable performance in complex environments such as rainy and foggy weather and strong background light interference.

[0118] Example 3: Complete System Integration and Dual-Stream Fusion Sensing

[0119] Based on Embodiments 1 and 2, this embodiment further provides an optimized system integration solution. For example... Figure 3A As shown, the multispectral infrared and visible light imaging torch combustion monitoring and optimization control system further includes a multispectral imaging module 10, a physical model calculation module 20, an artificial intelligence calculation module 50, a data fusion and correction module 60, a closed-loop control module 30, and a data storage and visualization module 80.

[0120] The multispectral imaging module 10 includes an uncooled infrared camera 101, a visible light camera 102, a multispectral registration unit 103, and an image acquisition and transmission module 140. The uncooled infrared camera 101 preferably operates in the 3-5 micrometer and 8-12 micrometer wavelength range and is used to acquire infrared radiation images of flames. The visible light camera 102 uses a zoom lens with over one megapixel resolution, operates in the visible light band, and is used to acquire visible light images of flames. The lens focal length is preferably 25-100mm, and the operating wavelength is 400-700nm. In flame monitoring, it mainly acquires visible light radiation information in the 0.4-0.7μm range for acquiring visible light images of flames. The multispectral registration unit 103 is used to spatially register the infrared and visible light images to ensure consistent image coordinates. The image acquisition and transmission module 140 is used for image data acquisition, compression, and transmission, supporting remote monitoring.

[0121] Furthermore, the physical model calculation module 20 and the artificial intelligence calculation module 50 are connected in parallel to the output of the multispectral imaging module 10, forming a dual-link calculation architecture. The physical model calculation module 20 calculates the first combustion state parameter based on Planck's radiation formula and the gray body radiation model, and the artificial intelligence calculation module 50 outputs the second combustion state parameter through the multi-scale visual Transformer network 310. The temperature distribution matrix of the physical model calculation module 20 is used as the supervised learning input feature of the artificial intelligence calculation module 50.

[0122] The input of the data fusion and correction module 60 is connected to the output of the physical model calculation module 20 and the artificial intelligence calculation module 50, respectively.

[0123] Further reference Figure 3B The closed-loop control module 30 further includes an improved PID controller 307, whose control objective function is:

[0124]

[0125] in For temperature standard deviation, This represents the average temperature.

[0126] The control objective function further includes an emission concentration penalty term:

[0127]

[0128] Where γ is the penalty coefficient. The current nitrogen oxide concentration, These are emission limits.

[0129] The input of the improved PID controller 307 is connected to the output of the data fusion and correction module 60. The output of the improved PID controller 307 is connected to the mass flow meter 305 and the combustion additive control valve 306 through the flow control interface 3041. The improved PID controller 307 further includes a feedforward compensation unit 307a, a dynamic weight adjustment unit 307b, a multivariable coupling control interface 307c, an emission constraint optimization unit 307d, and a hybrid control strategy generation unit 307e. Each unit interacts with the other through an internal data bus.

[0130] The input of the feedforward compensation unit 307a is connected to the prediction output of the physical lookup table 301. Based on the matching result of fuel type and combustion conditions, a pre-adjustment control quantity is generated: the higher the calorific value of the fuel, the larger the pre-adjustment quantity; the higher the ambient humidity (such as rainy days), the lower the pre-adjustment quantity is, in order to compensate for the influence of humidity on combustion efficiency.

[0131] The dynamic weight adjustment unit 307b is connected to the output of the data fusion and correction module 60 to extract the combustion efficiency CE parameter, which is used to dynamically adjust the proportional gain coefficient according to the fluctuation range of combustion efficiency. The dynamic weight adjustment unit 307b uses the combustion efficiency feedback value CE as the weight adjustment factor: when the combustion efficiency exceeds 95%, the proportional gain coefficient Kp is reduced to 80% of the reference value to avoid over-adjustment; when the flame temperature fluctuates drastically (e.g., changes exceeding 150K per second), the integration time T_i is shortened to 50% of the reference value to improve the response speed.

[0132] The multivariable coupling control interface 307c has its output terminals connected to the mass flow meter 305 and the combustion additive control valve 306, respectively, and uses a decoupling matrix D to synchronously regulate the fuel flow and combustion air flow.

[0133] The emission constraint optimization unit 307d is used to trigger a step response of the combustion additive control valve 306 when the corrected emission concentration exceeds the standard value, thereby increasing the excess air coefficient and executing the following control logic.

[0134] The emission concentration can be indirectly derived by the multispectral imaging module 10 and the physical model calculation module 20, or the multispectral imaging module 10 can be instantaneously calibrated by the laser-induced fluorescence (LIF) calibration module of the online calibration trigger subunit 605 to obtain the original emission concentration. The corrected concentration can then be calculated by combining the environmental humidity correction factor and the atmospheric attenuation coefficient.

[0135] When the corrected emission concentration exceeds the standard value, the step response of the combustion additive control valve 306 is triggered, causing the excess air coefficient to increase to the range of 1.15-1.25.

[0136] The control objective function is generated by combining the stability index of flame temperature distribution with the emission concentration penalty term:

[0137]

[0138] Where J represents the control objective function value, indicating the overall system performance index; α represents the stability weighting coefficient, used to adjust the importance of temperature stability in the objective function, with a typical value of 0.3-0.7. The stability index is based on the standard deviation of flame temperature. With average temperature Calculations are used to normalize the degree of temperature fluctuation. A higher value indicates greater combustion fluctuation; β represents the emission concentration weighting coefficient, used to adjust the importance of emission control in the objective function, with a typical value of 0.2-0.5. This represents the sum of squares of the ratios of various pollutant concentrations, used to comprehensively assess emission compliance; γ represents the emission penalty coefficient, the intensity of the penalty when emissions exceed standards, typically ranging from 1.0 to 3.0. The emission penalty is based on the current nitrogen oxide concentration. With emission limits The gap is dynamically adjusted when Approaching or exceeding At that time, the penalty coefficient γ is segmented to enhance the control strength; the optimal control command under the constraint conditions is solved by optimization algorithm to ensure that the fuel flow rate change does not exceed the safety threshold.

[0139] The hybrid control strategy generation unit 307e sets a control mode switching threshold: when the predicted combustion efficiency output by the improved PID controller 307 is lower than 85% or the control command mutation rate exceeds the threshold, such as 10% / s, it automatically switches to conservative control mode. At this time, the fuel flow rate is limited to 90%-95% of the matching parameters in the physical lookup table 301, the combustion air flow rate is dynamically adjusted according to the safe air-fuel ratio, and when the emission concentration exceeds the standard value, the fuel flow rate is further limited to 85%-90% of the matching value.

[0140] Through the above technical solution, the improved PID controller 307 achieves synergistic optimization of combustion stability and emission constraints, and can simultaneously ensure operational safety under extreme conditions through conservative control mode.

[0141] refer to Figure 3A The input of the data storage and visualization module 80 is connected to the output of the multispectral imaging module 10, the physical model calculation module 20, the artificial intelligence calculation module 50, and the data fusion and correction module 60, respectively. Its data storage unit stores historical data and combustion efficiency trend charts, and its visualization and alarm unit generates a flame temperature field thermogram and triggers a combustion abnormality alarm signal.

[0142] Further revealing the closed-loop optimization process, the closed-loop control module 30 establishes a two-way data channel with the environmental compensation unit 401, dynamically corrects the control parameters based on the ambient temperature and humidity and atmospheric attenuation coefficient output by the atmospheric transmittance calculation unit 404, and the calibration output terminal of the online calibration trigger subunit 605 is connected to the weighted Gaussian correction unit 202 of the physical model calculation module 20 and the model training and optimization unit of the artificial intelligence calculation module 50.

[0143] Based on the above method, the torch combustion monitoring and optimization control system with multispectral infrared and visible light imaging forms a closed-loop optimization process. That is, the registration image data collected by the multispectral imaging module 10 is processed in parallel by the physical model calculation module 20 and the artificial intelligence calculation module 50. The final combustion parameters output by the data fusion and correction module 60 drive the closed-loop control module 30 to adjust the combustion parameters. The adjusted environmental parameters are fed back to the multispectral imaging module 10 to form a closed-loop iteration. At the same time, when the temperature fluctuation rate exceeds ±50K / s or the smoke concentration exceeds 500mg / m³, the laser-induced fluorescence calibration procedure is triggered.

[0144] To address the limitations of existing technologies in processing single-modal data and their inability to maintain robustness under complex lighting or occlusion conditions, the multispectral imaging module 10 further includes a Transformer-based dual-stream fusion sensing system, the network topology of which is as follows: Figure 3C The system includes the following functional modules:

[0145] The visual embedding module 111 (VE) is used to convert the input visible light image into a first feature vector sequence; the infrared embedding module 112 (IE) is used to convert the input infrared image into a second feature vector sequence. By mapping the visible light image and the infrared image acquired by the visible light camera 102 and the uncooled infrared camera 101 to a "common semantic space" with the same dimension and the same sequence length, modal differences are eliminated, their unique information is preserved, and an aligned token sequence is provided for subsequent multi-head attention fusion. Multi-head attention (MHA) mechanism 113 is applied to the first feature vector sequence and the second feature vector sequence respectively to capture cross-pixel / cross-channel dependencies; Layer normalization (LN) module 114 is set before and after each MHA and each feedforward network to stabilize training; Feed-forward network (FFN) module 115 is used to perform nonlinear transformation on the LN output; Dropout (DO) module 116 is set after multi-head attention mechanism 113 (MHA) and feedforward network module 115 (FFN) to prevent overfitting; Output fusion module 117 is used to generate the target task result based on the fused feature vector sequence; Loss function module 118 (LF) is used to calculate the error between the predicted result and the true value during the training phase and update the network parameters through backpropagation via gradient backpropagation.

[0146] The weighted Gaussian correction unit 202 performs weighted Gaussian correction on the measurement results of multiple spectral channels to improve calculation accuracy. This unit uses the same weight matrix calculation method as the weighted Gaussian correction unit 202 in Example 1, as shown in formula (3).

[0147] Specifically, in this embodiment, the artificial intelligence computing module 50 employs a multi-scale visual Transformer network 310 for feature extraction and combustion efficiency calculation. This network is designed based on a MultiScale-ViTs architecture, achieving hierarchical representation of multi-scale features by progressively expanding the channel width D and reducing the sequence length L. This design overcomes the accuracy loss caused by simplification assumptions in traditional multispectral radiation physics formulas (such as assuming a constant emissivity and ignoring environmental radiation interference).

[0148] The multi-scale visual Transformer network 310 includes the following architectural variants:

[0149] Group42: The input is divided into two parts, including an RGB wavelength channel and an infrared channel;

[0150] Group43: The input includes two RGB wavelength channels and one infrared channel;

[0151] Group42s: contains one RGB wavelength channel, one infrared channel, and randomly samples one RGB channel using dropout;

[0152] Group43s: Contains two RGB wavelength channels, one infrared channel, and randomly samples one RGB channel using dropout;

[0153] Group44: Contains all RGB and infrared wavelength channels. This embodiment preferably uses the Group43s architecture, which maintains high accuracy while offering good computational efficiency.

[0154] The MultiScale-ViTs architecture in this embodiment adopts a parallel / serial hybrid connection method, specifically including: three different types of pooling attention, organized through parallel and serial skip connections; three serial FFN modules to realize the step-by-step transformation of features; and the mixed use of Group31, Group32, and Group33 modules.

[0155] In practical applications, this network achieves high-precision temperature measurement through the following steps: calibrating the measurement data at a specific time using MultiScale-ViTs; obtaining intermediate results through optimal fitting; updating the state using Kalman filtering; and outputting the final Transformer link prediction value.

[0156] In this embodiment, the artificial intelligence calculation module 50 and the physical model calculation module 20 constitute a dual-link verification mechanism. The artificial intelligence calculation module 50 employs an artificial intelligence multi-physics factor link, based on the aforementioned Transformer architecture, and possesses high accuracy; the physical model calculation module 20 employs a physical derivation link, based on the radiative transfer model, and possesses strong generalization ability. Through cross-verification of the two links, the multispectral infrared and visible light imaging torch combustion monitoring and optimization control system can automatically select the output result that best meets industrial needs. When CE is ≥95%, the system prioritizes the output of the artificial intelligence model; when CE is <85%, the system prioritizes the output of the physical model to ensure measurement reliability.

[0157] The above technical solution achieves breakthroughs in temperature measurement accuracy and control performance by constructing a two-way optimization closed loop of "physical constraint AI + AI-enhanced physics" while ensuring industrial reliability, and provides a reliable temperature field input for the subsequent closed-loop control module 30.

[0158] System Workflow

[0159] like Figure 4 As shown, the workflow of this system is further revealed, including:

[0160] Provide at least one multispectral imaging module 10 to acquire information in the visible and infrared bands.

[0161] The process of entering the physical model calculation module 20 includes:

[0162] Step S11: The Planck formula calculation unit 201 receives the visible light and infrared band radiation intensity data collected by the multispectral imaging module 10 and calculates the initial temperature distribution based on formula (1) and formula (2).

[0163] Step S12: The multi-band results are fused based on the weighted Gaussian correction unit 202, and the weight matrix W is dynamically adjusted according to the signal-to-noise ratio of each band.

[0164] Step S13: The combustion efficiency calculation unit 203 based on the Transformer architecture extracts cross-spectral features through a self-attention mechanism, outputs the corrected temperature distribution matrix, and outputs the first combustion state parameters;

[0165] In some optional embodiments, the temperature distribution matrix error after correction in step S3 is ≤ ±50K.

[0166] Parallel execution of the artificial intelligence computing module 50 processes, including:

[0167] Step S21: The multi-scale visual Transformer network receives the same multispectral radiometric data and performs initial feature extraction through the multi-scale feature extraction module;

[0168] Step S22: The pooling attention mechanism reduces resolution by pooling the query tensor Q, key tensor K, and value tensor V, and calculates it according to the formula Attention(Q,K,V) = softmax(QK^T / √d)V;

[0169] Step S23: The local and global attention mechanism performs self-attention calculation within an n×n window, while simultaneously extracting global features through a sparse uniform grid and outputting the second combustion state parameters;

[0170] The data fusion and correction module 60 process includes:

[0171] Step S31: Establish a state-space model and use the first combustion state parameters output by the physical model calculation module 20 as the system state prediction values. The second combustion state parameter output by the artificial intelligence calculation module 50 is used as the observation value z. k ;

[0172] Step S32: Dynamically adjust the observed noise covariance matrix R based on the combustion efficiency CE. When CE ≥ 95%, reduce the value of R to increase the Kalman gain K. k For observations biased towards artificial intelligence models, when CE < 85%, increasing the R value will increase the Kalman gain K. k Predictions are biased towards physical modeling.

[0173] Step S33: Update the equation through state update Output the final combustion state parameters after fusion;

[0174] When the deviation between the physical model and the artificial intelligence model exceeds a preset threshold, an online calibration process is executed.

[0175] Step S41: When the online calibration trigger subunit 605 detects that the temperature fluctuation exceeds the threshold, such as ±50K / s, or the dust concentration exceeds the threshold, such as 500mg / m³, the laser-induced fluorescence calibration module is automatically started.

[0176] Step S42: The laser-induced fluorescence calibration module performs instantaneous spectral calibration on the multispectral imaging module 10 and updates the response sensitivity coefficients of each spectral channel;

[0177] Step S43: Update the weight matrix of the weighted Gaussian correction unit 202 based on the calibration results to ensure that the measurement accuracy is maintained within ±2%;

[0178] Subsequently, the workflow of the closed-loop control module 30 includes:

[0179] Step S51: Based on the physical lookup table 301, match the current fuel type and combustion conditions to generate optimal control parameters;

[0180] Step S52: The interpolation calculation unit 302 performs bilinear interpolation based on the nearest neighbor parameters to generate adaptation parameters;

[0181] Step S53: The improved PID controller 307 performs multivariable coupled control based on the final combustion state parameters, and the feedforward compensation unit 307a generates a pre-adjustment control quantity based on the matching result of fuel type and combustion conditions;

[0182] The improved dynamic adjustment process of the PID controller 307 includes:

[0183] Step S54: The dynamic weight adjustment unit 307b uses the combustion efficiency feedback value CE as the weight adjustment factor. When CE≥95%, the proportional gain coefficient Kp is reduced to 80% of the reference value. When CE<85%, the integral time Ti is shortened to 50% of the reference value.

[0184] Step S55: The emission constraint optimization unit 307d monitors the corrected emission concentration. When it exceeds the standard value, it triggers the step response of the combustion additive control valve 306, increasing the excess air coefficient to the range of 1.15-1.25.

[0185] Step S56: The multivariable coupling control interface 307c uses a decoupling matrix D to synchronously adjust the fuel flow rate and the combustion air flow rate. Under normal operating conditions, the coefficients are d11=1.0, d12=-0.15×theoretical air-fuel ratio, d21=0.08×theoretical air-fuel ratio, and d22=1.0.

[0186] Ultimately, the execution and feedback process includes:

[0187] Step S61: The actuator interface 304 sends control commands to the mass flow meter 305 and the combustion additive control valve 306 to adjust the flow rate of the combustion additive and realize closed-loop feedback control;

[0188] Step S62: The adjusted combustion parameters are fed back to the multispectral imaging module 10 through the environmental compensation unit 401, forming a complete closed-loop iteration;

[0189] Step S63: The data storage and visualization module 80 records the system operation data in real time, generates a flame temperature field thermogram, and triggers an alarm signal when an abnormal combustion is detected;

[0190] Step S64: When the temperature fluctuation rate exceeds the threshold, trigger the laser-induced fluorescence calibration procedure.

[0191] Through the above steps, high-precision, real-time monitoring and closed-loop optimization control of the combustion status of the object awaiting detection in the flare are achieved. This complete workflow establishes a closed-loop optimization mechanism of "acquisition-computation-fusion-control-feedback," ensuring continuous optimization of the combustion efficiency of combustibles within the flare.

[0192] Furthermore, refer to Figure 5 In some optional embodiments of this application, a method for torch combustion monitoring and optimized control based on multispectral infrared and visible light imaging is provided, the method comprising the following steps:

[0193] Step 1: Acquire flame radiation data using a multispectral imaging system;

[0194] Step 2: Calculate the initial temperature distribution based on Planck's formula and the gray body model;

[0195] Step 3: Multispectral data fusion and correction are performed using the weighted Gaussian algorithm and the Levenberg-Marquardt algorithm;

[0196] Step 4: Calculate combustion efficiency using the Transformer model;

[0197] Step 5: Input the corrected temperature data into the combustion optimization control module;

[0198] Step 6: Generate control commands based on the physical lookup table and flow meter data;

[0199] Step 7: Adjust the combustion parameters via the actuator and return to the monitoring step.

[0200] Furthermore, in step four, “Calculating combustion efficiency using the Transformer model”, this embodiment uses a multi-scale visual Transformer network 310 as the specific implementation method. This network receives the fusion correction results of the weighted Gaussian algorithm and the Levenberg-Marquardt algorithm in step three as input.

[0201] The multi-scale visual Transformer network 310 can adopt one of the following architectural variants:

[0202] Group42 architecture: Configured with one RGB wavelength channel and one infrared channel input branch, suitable for scenarios with limited computing resources;

[0203] Group43 architecture: configured with two RGB wavelength channels and one infrared channel input branch, suitable for scenarios requiring high precision;

[0204] Group42s architecture: Adds dropout random sampling mechanism to Group42 to improve model robustness;

[0205] Group43s architecture: Based on Group43, a dropout random sampling mechanism is added, which maintains high accuracy while having good computational efficiency;

[0206] Group44 architecture: Configured with all RGB and infrared wavelength channels, suitable for scenarios where computational resource requirements are not critical. Any of the above architecture variants are structurally identical to the multispectral data fusion results from step three.

[0207] The combustion efficiency calculation process includes the following key calculation modules:

[0208] Pooling attention mechanism reduces computational complexity while maintaining feature representation ability; and

[0209] The local and global attention mechanisms further include: local window attention, used to capture spatial local features within an n×n window, where n is a natural number ≥ 4; global grid attention, used to extract long-range dependencies; and axial attention, used to handle inter-channel correlations.

[0210] The reconstructed output employs a dual-link verification mechanism. The reconstructed temperature field is compared with the initial temperature distribution calculated using Planck's formula in step two. When the combustion efficiency (CE) ≥ 95%, the system prioritizes the output of the artificial intelligence model; when CE < 85%, the system prioritizes the output of the physical model. This ensures that the output meets the input requirements of the combustion optimization control module in step five.

[0211] The reconstructed temperature field data is transmitted to step five via a standardized interface to provide accurate temperature field input for subsequent combustion optimization control.

[0212] In summary, this scheme employs multispectral image fusion technology, acquiring multispectral image information of the flame through registration of an uncooled infrared camera and a visible light camera. This technology maintains high image quality and measurement accuracy under different combustion states. Based on Planck's radiation formula and the gray body radiation model, combined with multispectral image data, the flame temperature is calculated. For the measurement results of multiple spectral channels, a weighted Gaussian formula is used for correction to improve calculation accuracy.

[0213] Through the above technical solution, the present invention achieves the following technical effects:

[0214] 1. Cost reduction: By replacing traditional high-cost equipment with a multispectral imaging system, the hardware cost of the flare monitoring system is reduced by more than 40%;

[0215] 2. Improved accuracy: The algorithm architecture combining the Transformer model and Kalman filtering is adopted to control the flame temperature measurement error within ±50K, achieving the same accuracy level as VISR equipment.

[0216] 3. Comprehensive coverage: Through the collaborative work of physical lookup tables and interpolation algorithms, it can cover the control requirements of 100% normal combustion conditions and 98% abnormal combustion conditions.

[0217] 4. Wide applicability: The system supports real-time monitoring and optimized control of flare devices for various fuel types (natural gas, refinery gas, syngas, etc.) and different combustion conditions (oxygen-deficient / oxygen-enriched combustion), meeting increasingly stringent environmental regulations.

[0218] The basic principles, main features, and advantages of this application have been described above. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely the principles of this application. Various changes and modifications can be made to this application without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claims. The scope of protection claimed by this application is defined by the appended claims and their equivalents.

Claims

1. A torch combustion monitoring and optimization control system based on multispectral infrared and visible light imaging, characterized in that, include: The multispectral imaging module is used to simultaneously acquire infrared and visible light images of the torch flame and perform spatial registration to obtain multispectral radiation data of the flame. The physical model calculation module is used to calculate the first combustion state parameters based on the Planck radiation formula and the gray body radiation model, according to the multispectral radiation data. The artificial intelligence computing module uses a pre-trained multi-scale visual Transformer network to calculate the second combustion state parameter based on the multispectral radiation data. The data fusion and correction module is used to fuse and correct the first combustion state parameters and the second combustion state parameters using a Kalman filter algorithm to output the final combustion state parameters; and A closed-loop control module is used to dynamically adjust the torch's operating parameters to optimize combustion efficiency based on the final combustion state parameters. The data fusion and correction module achieves dual-model data fusion through a Kalman filter unit, specifically including: A state-space model is established, and the first combustion state parameter output by the physical model calculation module is used as the system state prediction value, and the second combustion state parameter output by the artificial intelligence calculation module is used as the observation value. The observation noise covariance matrix R is dynamically adjusted based on the combustion efficiency CE: when CE≥95%, the value of R is reduced to bias the Kalman gain toward the observation of the artificial intelligence model; when CE<85%, the value of R is increased to bias the Kalman gain toward the prediction of the physical model.

2. The torch combustion monitoring and optimization control system according to claim 1, characterized in that, The multispectral imaging module includes: An uncooled infrared camera, with an operating wavelength covering 3-5 micrometers and 8-12 micrometers, is used to acquire infrared radiation images of the flame. A visible light camera is used to acquire visible light images of the flame; and A multispectral registration unit is used to spatially register the infrared radiation image with the visible light image to generate an aligned array of flame positions.

3. The torch combustion monitoring and optimization control system according to claim 2, characterized in that, The multispectral imaging module further includes a dynamic spectral selection module, which includes a tunable filter array and a principal component analysis unit. The principal component analysis unit dynamically selects the optimal combination of spectral bands based on the current fuel type and combustion state to guide the tunable filter array in imaging.

4. The torch combustion monitoring and optimization control system according to claim 3, characterized in that, The system also includes an environmental compensation module, which is connected to an environmental temperature and humidity sensor and calculates the atmospheric attenuation coefficient in real time based on the MODTRAN radiative transfer model to dynamically compensate the data collected by the multispectral imaging module.

5. The torch combustion monitoring and optimization control system according to claim 1, characterized in that, The physical model calculation module includes a weighted Gaussian correction unit. The weighted Gaussian correction unit establishes a weight matrix based on the response sensitivity coefficients of each spectral channel and uses the Levenberg-Marquardt algorithm to perform nonlinear optimization on the first combustion state parameters calculated based on multi-band.

6. The flare combustion monitoring and optimization control system according to claim 1, characterized in that, The multi-scale visual Transformer network includes a multi-scale feature extraction module, a pooling attention mechanism, and local and global attention mechanisms, wherein: The pooling attention mechanism reduces resolution by pooling query tensors, key tensors, and value tensors, and reduces computational complexity by performing pooling operations on query, key, and value tensors. The local and global attention mechanisms include local attention, which performs self-attention computation within an n×n window, and global attention, which extracts global features through a sparse uniform grid, where n is a natural number ≥ 4.

7. The flare combustion monitoring and optimization control system according to claim 1, characterized in that, The closed-loop control module includes: A physical lookup table stores the mapping relationship between fuel type, combustion conditions and optimal combustion parameters; The interpolation calculation unit, when detecting that the current operating condition parameters do not match in the physical lookup table, employs a bilinear interpolation algorithm to perform linear interpolation calculations based on the nearest neighbor known parameter combinations in the physical lookup table, in order to generate control parameters adapted to the current operating condition; and An improved PID controller is used to generate control commands based on the final combustion state parameters.

8. The flare combustion monitoring and optimization control system according to claim 7, characterized in that, The improved PID controller includes: The feedforward compensation unit is used to generate a pre-adjustment control quantity based on the matching result of fuel type and combustion conditions; The dynamic weighting adjustment unit is used to dynamically adjust the proportional gain coefficient according to the fluctuation range of combustion efficiency; and The multivariable coupling control interface has its output connected to a mass flow meter and a combustion additive control valve, respectively, to synchronously regulate the fuel flow and combustion air flow.

9. The flare combustion monitoring and optimization control system according to claim 1, characterized in that, The system forms a closed-loop optimization process: The registered image data acquired by the multispectral imaging module is processed in parallel by the physical model calculation module and the artificial intelligence calculation module; The final combustion parameters output by the data fusion and correction module drive the closed-loop control module to adjust the combustion parameters; and the adjusted environmental parameters are fed back to the multispectral imaging module to form a closed-loop iteration.

Images