High-temperature pure gas-phase gas dual-band neural network radiation temperature measurement method and device

Abstract

The application provides a high-temperature pure gas-phase fuel gas double-waveband neural network radiation temperature measurement method and device, which is suitable for temperature measurement of propellant products being high-temperature fuel gas, and aims to solve the problems of traditional double-color temperature measurement, such as dependence on Planck iteration, necessity of given prior fuel gas components, large CO2 / H2O radiation coupling error of 2.7 mu m band, and poor real-time performance of dynamic fuel gas. The application can complete fuel gas temperature inversion only by relying on two groups of characteristic waveband radiation brightness; based on an SNB narrowband gas radiation transmission model and a HITRAN spectral database, a batch of pure gas-phase simulation data sets are constructed, a multilayer perception machine (MLP) is used as a network backbone, and the simulation data sets are used for offline training. In the reasoning stage, only the corrected double-waveband radiation brightness is input, and the fuel gas thermodynamic temperature is directly output, without performing Planck spectral integration and radiation transmission iteration solving, and without relying on any prior conditions of gas-phase components, so that the high-temperature pure gas-phase fuel gas temperature measurement demand can be met.

Description

Technical Field

[0001] This invention relates to the field of high-temperature gas temperature measurement technology, specifically to a high-temperature pure gas phase gas dual-band neural network radiation temperature measurement method and device. Background Technology

[0002] The core gaseous components of propellant combustion products are CO2 and H2O. Existing pure gas-phase two-color radiometric temperature measurement algorithms have the following drawbacks: (1) The observation equation is underdetermined and depends on the prior knowledge of the components: There are only two sets of radiance observations in the bands, and the gas temperature T and water vapor optical thickness exist. Carbon dioxide optical thickness Three unknowns: Traditional iterative methods must predetermine component conditions such as propellant ratio and gas partial pressure; the gas composition fluctuates dynamically during operation, and the iteration is prone to divergence when there is no stable prior. (2) The nonlinear solution of the spectral band coupling is complicated: the CO2 and H2O absorption bands in the 2.7μm~2.95μm bands are highly overlapping. The traditional WSGG and SNB gray gas models fit the coupled emissivity in segments, and the temperature measurement error is large in the high temperature and high optical thickness saturation range. (3) Planck iteration has poor real-time performance: Traditional inversion requires multiple spectral integrations and narrowband emissivity calculations in each iteration, which takes a long time to compute; (4) Existing gas temperature measurement neural network algorithms do not distinguish between the MLP basic model and the PINN physical constraint model. The two architectures are separate, and it is impossible to balance lightweight real-time inference and high-temperature boundary condition inversion accuracy. Moreover, most schemes still require input of component and pressure auxiliary parameters, and cannot achieve independent inversion without prior knowledge. Summary of the Invention

[0003] To address the problems of lack of prior knowledge regarding high-temperature pure CO2-H2O high-temperature gas components, radiation coupling of overlapping dual-band spectra, high computational delay of traditional iterative algorithms, and large errors in the high-temperature saturation range, this invention provides a dual-band neural network radiation temperature measurement method and device for high-temperature pure gas phase gas.

[0004] To achieve the above objectives, the present invention provides the following technical solution: On the one hand, a high-temperature pure gas phase combustion dual-band neural network radiation thermometry method is provided, including the following steps: Considering only the gaseous components CO2 and H2O, a sample space was constructed based on the SNB narrowband gas radiative transfer model and the HITRAN high-precision molecular spectral line database. Variables included gas combustion temperature, water vapor optical thickness, and carbon dioxide optical thickness. For each set of variables, the corresponding simulated brightness L was calculated using forward modeling in the first band (2.7 μm–2.95 μm) and the second band (4.2 μm–4.45 μm). 1,sim and L 2,simSimultaneously, temperature labels corresponding to each group of variables are generated to construct a simulation dataset; A temperature prediction model with a 2-input, 1-output multilayer perceptron (MLP) as the network backbone was constructed. The temperature prediction model was trained using a simulation dataset to obtain the trained temperature prediction model. The measured integrated radiance of high-temperature gas gas in the first band (2.7μm~2.95μm) and the second band (4.2μm~4.45μm) was obtained. The background radiation interference was subtracted and the influence of optical path attenuation was eliminated to obtain the pure gas phase intrinsic radiance corresponding to the first and second bands. The inherent radiance of the pure gas phase corresponding to the first and second bands is input into the trained temperature prediction model, and the thermodynamic temperature of the high-temperature pure gas phase gas is output through network forward propagation.

[0005] Preferably, when training the temperature prediction model using a simulation dataset, the network training is completed by constructing the mean square error supervision loss using only temperature labels.

[0006] Preferably, when training the temperature prediction model using the simulation dataset, the PINN physical information network training method is adopted. On the basis of constructing the mean square error supervision loss with temperature labels, a pure gas phase radiation transfer equation constraint loss is added to construct the total loss function, which forces the network prediction output to satisfy the physical laws of CO2-H2O gas phase radiation.

[0007] On the other hand, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described high-temperature pure gas phase combustion dual-band neural network radiation temperature measurement method.

[0008] On the other hand, the present invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described high-temperature pure gas phase combustion dual-band neural network radiation temperature measurement method.

[0009] On the other hand, the present invention provides a computer program product stored on a computer-readable storage medium and including computer instructions that, when executed by a processor, cause an electronic device to implement the steps of the above-described high-temperature pure gas phase combustion dual-band neural network radiation temperature measurement method.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a high-temperature pure gas-phase combustion gas dual-band neural network radiation thermometry method suitable for measuring the temperature of propellant products that are high-temperature combustion gases. It solves the problems of traditional two-color thermometry, which relies on Planck iteration, requires prior knowledge of combustion gas components, suffers from large coupling errors in the 2.7μm spectral band CO2 / H2O radiation, and has poor real-time performance for dynamic combustion gases. This invention can complete the combustion gas temperature inversion using only two sets of characteristic band radiation brightness; offline training introduces physical constraints on gas-phase radiation to improve accuracy; the inference is lightweight and iterative-free, and the entire process does not depend on any prior conditions for gas-phase components.

[0011] Specifically, this invention selects a first band of 2.7μm–2.95μm and a second band of 4.2μm–4.45μm. The first band of 2.7μm–2.95μm represents the overlapping gas-phase absorption band of CO2 and H2O, while the second band of 4.2μm–4.45μm represents the single gas-phase characteristic band of pure CO2. Two sets of integrated radiance values ​​from the first band of 2.7μm–2.95μm and the second band of 4.2μm–4.45μm are selected as inputs. A pure gas-phase simulation dataset is constructed in batches based on the SNB narrowband gas radiative transfer model and the HITRAN spectral database. A multilayer perceptron (MLP) is used as the backbone, providing two training modes: one is a lightweight MLP with only label supervision, and the other is a Physical Information Neural Network (PINN) with constraints from the gas-phase radiative transfer equation superimposed on label supervision. During the inference and prediction phase, only the corrected dual-band radiance is input, and the gas thermodynamic temperature is directly output. Planck spectral integration and non-radiative transport iterative solutions are not performed throughout the process, and prior input of components such as CO2 partial pressure, H2O mole fraction, etc., is unnecessary. PINN suppresses prediction drift in the high-temperature and high-optical-thickness saturation range through gas-phase radiation physical regularization. Its lightweight MLP has extremely low computational cost, meeting the requirements for high-temperature gas temperature measurement. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a high-temperature pure gas phase combustion dual-band neural network radiation temperature measurement method in one embodiment; Figure 2 This is a block diagram of an electronic device in one embodiment. Detailed Implementation

[0014] The technical solution of the present invention will now be clearly and completely described through specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0015] This invention provides a dual-band neural network radiation thermometry method for high-temperature pure gas phase combustion gas, which is suitable for the temperature measurement of propellant products that are high-temperature combustion gases. It solves the problems of traditional two-color thermometry relying on Planck iteration, requiring prior knowledge of combustion gas components, large radiative coupling error of CO2 / H2O in the 2.7μm spectral band, and poor real-time performance of dynamic combustion gases.

[0016] In one embodiment, reference is made to Figure 1 A method for measuring the radiation temperature of high-temperature pure gas phase combustion gas using a dual-band neural network is provided, comprising the following steps: Considering only the gaseous components CO2 and H2O, a sample space was constructed based on the SNB narrowband gas radiative transfer model and the HITRAN high-precision molecular spectral line database. Variables included gas combustion temperature, water vapor optical thickness, and carbon dioxide optical thickness. For each set of variables, the corresponding simulated brightness L was calculated using forward modeling in the first band (2.7 μm–2.95 μm) and the second band (4.2 μm–4.45 μm). 1,sim and L 2,sim Simultaneously, a temperature label T corresponding to each group of variables is generated. label Construct a simulation dataset; A temperature prediction model with a 2-input, 1-output multilayer perceptron (MLP) as the network backbone was constructed. The temperature prediction model was trained using a simulation dataset to obtain the trained temperature prediction model. The measured integrated radiance of high-temperature gas gas in the first band (2.7μm~2.95μm) and the second band (4.2μm~4.45μm) was obtained. The background radiation interference was subtracted and the influence of optical path attenuation was eliminated to obtain the pure gas phase intrinsic radiance corresponding to the first and second bands. The inherent radiance of the pure gas phase corresponding to the first and second bands is input into the trained temperature prediction model, and the thermodynamic temperature of the high-temperature pure gas phase gas is output through network forward propagation.

[0017] The first band, from 2.7 μm to 2.95 μm, is a strong absorption band shared by both CO2 and H2O, and its brightness simultaneously carries three-dimensional coupled information about temperature, water vapor, and carbon dioxide. The second band, from 4.2 μm to 4.45 μm, exhibits only characteristic CO2 gas phase absorption; the absorption cross-section of H2O in this band is approximately zero, and its brightness carries coupled information about temperature and CO2 optical thickness. These two brightness sets constitute a complete observation vector, with each set of observations uniquely corresponding to a set of gas phase conditions.

[0018] This invention is specifically designed for CO2-H2O gaseous media and is suitable for temperature measurement scenarios where the propellant products are pure combustion gases. The temperature prediction model of this invention only requires the input of two pre-processed sets of band radiance values; no other gaseous component parameters such as propellant ratio, water vapor partial pressure, CO2 mole fraction, or gas layer optical path length need to be input. During the training phase, the network model autonomously learns all nonlinear mapping relationships between temperature, water vapor optical thickness, carbon dioxide optical thickness, and dual-band radiance values. This invention is completely free of prior component constraints, adapts to dynamic fluctuations in combustion gas composition under varying operating conditions, and eliminates the need for pre-calibration of combustion product components, significantly reducing the workload of data processing.

[0019] In one embodiment, the sampling range of the sample space is: gas temperature T∈400K~1400K; carbon dioxide optical thickness. ∈0.05~3.5; water vapor optical thickness ∈0.1~6.0. Sampling methods can include uniform sampling, Latin hypercube sampling, or densified sampling in high-temperature and high-optical-thickness regions. The absorption coefficients of each narrowband are calculated based on HITRAN high-precision molecular spectral line parameters and the SNB narrowband gas radiative transfer model, further yielding the equivalent emissivity of the CO2-H2O mixture in the i-th band. .

[0020] When constructing the simulation data model, only the gas phase radiation model was used throughout. Full-condition samples were generated in batches using the HITRAN high-precision molecular spectral line database combined with the SNB narrowband gas radiative transfer model, covering the operating temperature range of 400K~1400K. To further enhance the adaptability to real-world testing, Gaussian noise from the infrared detector, calibration errors, and optical path transmittance perturbations were superimposed on the simulated brightness to create enhanced samples [L]. 1,sim ,L 2,sim ,T label To expand the simulation dataset and improve the algorithm's generalization ability under real-world dynamic gas-phase environments, the simulation dataset can be divided into training, validation, and test sets. The test set should retain samples in the high-temperature saturation range to evaluate the temperature inversion stability under extreme conditions.

[0021] This invention uses a simulation dataset to train the neural network offline, employing a 2-input, 1-output multilayer perceptron (MLP) as the network backbone. In one embodiment, the network topology of the temperature prediction model is as follows: 2 neurons in the input layer → 64 neurons in the first hidden layer with ReLU activation → 128 neurons in the second hidden layer with ReLU activation → 64 neurons in the third hidden layer with ReLU activation → 1 neuron in the output layer. The hidden layers use ReLU or SiLU activation functions, and the output layer provides the thermodynamic temperature of the high-temperature pure gas phase combustion gas.

[0022] For the temperature prediction model, this invention provides two different training modes to choose from: In one implementation, when training the temperature prediction model using a simulation dataset, the network training is completed by constructing a mean squared error supervised loss using only temperature labels. This supervised learning relies solely on simulated temperature labels, without physical equation constraints, resulting in fast training speed and small model weight size, thus meeting the requirements for high-frequency, real-time time-series temperature measurement.

[0023] In another implementation, when training the temperature prediction model using a simulation dataset, a PINN (Physical Information Network) training method is employed. This method constructs a total loss function by adding a constraint loss based on the pure gas-phase radiative transfer equation, in addition to the mean square error supervised loss built using temperature labels. This forces the network's prediction output to satisfy the physical laws of CO2-H2O gas-phase radiation. Regularizing the network parameters using the physical laws of CO2-H2O radiation corrects prediction biases under high-temperature and high-concentration saturation conditions, significantly improving the inversion accuracy under extreme high-temperature conditions.

[0024] Specifically, in one embodiment, a total loss function is constructed by adding a pure gas-phase radiative transfer equation constraint loss to the mean square error monitoring loss built using temperature labels. for: ; in, Predicting temperature for the network Temperature labels obtained from simulation Mean square error monitoring loss; The weighting parameters for the constraint loss in the pure gas phase radiative transport equation are: (0,1); This represents the constraint loss for the pure gas phase radiative transport equation.

[0025] Among them, the constraint loss of the pure gas phase radiative transport equation The construction method is as follows: Predicting temperature via network Substituting into the SNB gas phase radiation forward model, calculate the theoretical radiance L of the first band. 1,phy Theoretical radiance L in the second band 2,phy and the corresponding simulated brightness L 1,sim and L 2,sim Constructing residuals: ; in and These are the brightness normalization scaling factors for the first and second bands, respectively.

[0026] Monotonicity constraint term The construction method is as follows: ; in, Predict temperature for the network. For the first j Radiance input in each band, Corresponding to the radiance input of the first band, 2.7μm~2.95μm, Corresponding to the second band of radiance input of 4.2um~4.45um, As a monotonic factor, take .

[0027] In one embodiment, a constraint loss based on the pure gas-phase radiative transfer equation is added to the mean square error monitoring loss constructed using temperature labels. Additionally, a monotonicity constraint term is included to construct the total loss function. for: ; in, Predicting temperature for the network Temperature labels obtained from simulation Mean square error monitoring loss; The constraint loss is for the pure gas phase radiative transport equation. The monotonicity constraint term is used to suppress prediction drift in the high optical thickness saturation region. The monotonicity constraint term is constrained within a fixed optical thickness neighborhood, and the network prediction brightness maintains a non-negative response as the temperature increases. and These are the weight parameters for the constraint loss and monotonicity constraint terms in the pure gas phase radiative transport equation, respectively. and Constraint loss of pure gas phase radiative transport equation The construction method is the same as above, and will not be repeated here.

[0028] During the actual measurement, the integrated radiance of high-temperature gas was simultaneously acquired using a calibrated dual-band infrared detection system in the first band (2.7μm–2.95μm, covering both strong absorption bands of CO2 and H2O) and the second band (4.2μm–4.45μm, primarily characterizing CO2 gas phase radiation). Background radiation interference was subtracted, and the effects of optical path attenuation were eliminated to obtain the pure gas phase intrinsic radiance corresponding to the first and second bands. Response calibration was performed using a blackbody furnace before and after acquisition to obtain the detector response coefficient, dark current, and optical window transmittance.

[0029] For the i-th band, the measured integrated radiance of the i-th band can be expressed as: ; in , indicating the band number, Corresponding to the first band, Corresponding to the second band; For the first Measured integrated radiance in the band; This is a combined coefficient of optical transmittance and detector response. For the first The inherent radiance of the pure gas phase corresponding to the band; For the CO2-H2O mixed gas in the first i Equivalent emissivity of the band; For the first Band-band wall blackbody radiation brightness; For the first Dark field offset in the band, For the first Measurement noise in the bands. Using the above formulas, after dark field subtraction, transmittance correction, and wall background separation, the pure gas phase intrinsic radiance corresponding to the first and second bands can be obtained.

[0030] To reduce the impact of dimensional differences on network training, after obtaining the pure gas phase intrinsic radiance corresponding to the first and second bands, a logarithmic normalization preprocessing is performed on the pure gas phase intrinsic radiance corresponding to the first and second bands to obtain the input vector. ,in To prevent taking the smallest positive value of the logarithm for zero brightness, the above preprocessing method is not limited to logarithmic normalization. Mean-variance normalization or max-min normalization can also be used. The normalization parameters are obtained from the training set statistics and saved along with the model weights.

[0031] In one embodiment, the pure gas phase intrinsic radiance corresponding to the first and second bands is input into the trained temperature prediction model. After the high-temperature pure gas phase thermodynamic temperature is output through network forward propagation, a confidence level determination step is also included: if the pure gas phase intrinsic radiance corresponding to the first and second bands exceeds the radiance envelope of the simulation dataset, or the dual-band radiance ratio exceeds a preset threshold, or the network integrated output variance exceeds a preset threshold, then the confidence level of the temperature measurement result output by the current temperature prediction model is determined to be low, and a low confidence level marker is output to indicate possible wall obstruction, residual condensed particles, window contamination, or detector saturation abnormality.

[0032] Traditional two-color gas thermometry typically requires pre-defined CO2 / H2O mole fractions, partial pressures, or optical path lengths, followed by iterative solutions to the temperature using Planck's integral and the radiative transfer equation. This invention offline incorporates composition and optical thickness variations into the simulation sample space and neural network weights, allowing for direct online temperature output using only two gas phase brightness observations. This avoids underdetermined solutions and iterative divergence problems caused by missing composition priors.

[0033] This invention can be used in scenarios such as rapid assessment of the temperature field of high-temperature pure gas-phase combustion products. This method does not rely on new hardware structures and can be used in conjunction with existing dual-band infrared testing systems. It features low implementation cost, strong real-time performance, and the ability to process experimental time-series data in batches.

[0034] In another embodiment, a high-temperature pure gas phase combustion dual-band neural network radiation temperature measurement device is provided, comprising: The simulation dataset construction module is used to construct a sample space based on the SNB narrowband gas radiative transfer model and the HITRAN high-precision molecular spectral line database, considering only gas phase components CO2 and H2O. Variables include gas combustion temperature, water vapor optical thickness, and carbon dioxide optical thickness. For each set of variables, the corresponding simulated brightness L is calculated using forward modeling in the first band (2.7μm–2.95μm) and the second band (4.2μm–4.45μm). 1,sim and L 2,sim Simultaneously, temperature labels corresponding to each group of variables are generated to construct a simulation dataset; The neural network model training module is used to construct a temperature prediction model with a multilayer perceptron (MLP) as the network backbone, and to train the temperature prediction model using a simulation dataset to obtain the trained temperature prediction model. The dual-band radiation acquisition and preprocessing module is used to obtain the measured integrated radiance of high-temperature gas in the first band (2.7μm~2.95μm) and the second band (4.2μm~4.45μm), and to subtract background radiation interference and eliminate the influence of optical path attenuation, thereby obtaining the pure gas phase inherent radiance corresponding to the first and second bands. The temperature inference output module is used to input the pure gas phase intrinsic radiance corresponding to the first and second bands into the trained temperature prediction model, and output the high-temperature pure gas phase gas thermodynamic temperature through network forward propagation.

[0035] Figure 2 The figure shows a block diagram of an electronic device according to an embodiment, such as Figure 2As shown, the electronic device includes one or more processors and a memory. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the high-temperature pure gas phase combustion dual-band neural network radiation temperature measurement method provided in any of the above embodiments. The processor can be a central processing unit (CPU) or other processing unit with data processing capabilities and / or instruction execution capabilities, and can control other components in the electronic device to perform desired functions. The memory can include one or more computer program products, which can include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory can include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory can include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0036] In one example, the electronic device may also include input devices and output devices, which are interconnected via a bus system and / or other forms of connection mechanism (not shown).

[0037] Of course, for the sake of simplicity, Figure 2 Only some of the components of the electronic device relevant to this application are shown in this illustration; components such as buses, input / output interfaces, etc., are omitted. In addition, the electronic device may include any other suitable components depending on the specific application.

[0038] Embodiments of the present invention may also be computer-readable storage media storing a computer program thereon. When executed by a processor, the computer program implements the steps of the high-temperature pure gas phase combustion dual-band neural network radiation thermometry method provided in any of the above embodiments. The computer-readable storage medium may be any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may include, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0039] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not restrict the application from being implemented using the specific details described above.

[0040] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0041] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0042] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application should not be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0043] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A high-temperature pure gas phase combustion dual-band neural network radiation thermometry method, characterized in that, Includes the following steps: Considering only the gas phase components CO2 and H2O, a sample space is constructed based on the SNB narrowband gas radiative transfer model and the HITRAN high-precision molecular spectral line database, with variables including gas temperature, water vapor optical thickness, and carbon dioxide optical thickness. For each set of variables, the simulated brightness L was calculated using forward modeling in the first band (2.7μm–2.95μm) and the second band (4.2μm–4.45μm). 1,sim and L 2,sim Simultaneously, temperature labels corresponding to each group of variables are generated to construct a simulation dataset; A temperature prediction model with a 2-input, 1-output multilayer perceptron (MLP) as the network backbone was constructed. The temperature prediction model was trained using a simulation dataset to obtain the trained temperature prediction model. The measured integrated radiance of high-temperature gas gas in the first band (2.7μm~2.95μm) and the second band (4.2μm~4.45μm) was obtained. The background radiation interference was subtracted and the influence of optical path attenuation was eliminated to obtain the pure gas phase intrinsic radiance corresponding to the first and second bands. The inherent radiance of the pure gas phase corresponding to the first and second bands is input into the trained temperature prediction model, and the thermodynamic temperature of the high-temperature pure gas phase gas is output through network forward propagation.

2. The high-temperature pure gas phase combustion dual-band neural network radiation thermometry method according to claim 1, characterized in that, The sampling range of the sample space is: gas temperature T∈400K~1400K; carbon dioxide optical thickness. ∈0.05~3.5; water vapor optical thickness ∈0.1~6.

0.

3. The high-temperature pure gas phase combustion dual-band neural network radiation thermometry method according to claim 1 or 2, characterized in that, When training the temperature prediction model using a simulation dataset, the network training is completed by constructing the mean square error supervision loss using only the temperature labels.

4. The high-temperature pure gas phase combustion dual-band neural network radiation thermometry method according to claim 1 or 2, characterized in that, When training the temperature prediction model using the simulation dataset, the PINN physical information network training method is adopted. Based on the mean square error supervision loss constructed with temperature labels, a pure gas phase radiation transfer equation constraint loss is added to construct the total loss function, which forces the network prediction output to satisfy the physical laws of CO2-H2O gas phase radiation.

5. The high-temperature pure gas phase combustion dual-band neural network radiation thermometry method according to claim 4, characterized in that, Total loss function for: in, Predicting temperature for the network Temperature labels obtained from simulation Mean square error monitoring loss; The weighting parameters for the constraint loss in the pure gas phase radiative transport equation; This represents the constraint loss for the pure gas phase radiative transport equation.

6. The high-temperature pure gas phase combustion dual-band neural network radiation thermometry method according to claim 5, characterized in that, Constraint loss of pure gas phase radiative transport equation The construction method is as follows: Predicting temperature via network Substituting into the SNB gas phase radiation forward model, calculate the theoretical radiance L of the first band. 1,phy Theoretical radiance L in the second band 2,phy and the corresponding simulated brightness L 1,sim and L 2,sim Constructing residuals: in and These are the brightness normalization scaling factors for the first and second bands, respectively.

7. The high-temperature pure gas phase combustion dual-band neural network radiation thermometry method according to claim 4, characterized in that, Total loss function for: in, Predicting temperature for the network Temperature labels obtained from simulation Mean square error monitoring loss; The constraint loss is for the pure gas phase radiative transport equation. The monotonicity constraint term is used to suppress prediction drift in the high optical thickness saturation region. The monotonicity constraint term is constrained within a fixed optical thickness neighborhood, and the network prediction brightness maintains a non-negative response as the temperature increases. and These are the weight parameters for the constraint loss and monotonicity constraint terms in the pure gas phase radiative transport equation, respectively. and ; Constraint loss of pure gas phase radiative transport equation The construction method is as follows: Predicting temperature via network Substituting into the SNB gas phase radiation forward model, calculate the theoretical radiance L of the first band. 1,phy Theoretical radiance L in the second band 2,phy and the corresponding simulated brightness L 1,sim and L 2,sim Constructing residuals: in and These are the brightness normalization scaling factors for the first and second bands, respectively; Monotonicity constraint term The construction method is as follows: in, Predict temperature for the network. For the first Radiance input in each band, Corresponding to the radiance input of the first band, 2.7μm~2.95μm, Corresponding to the second band of radiance input of 4.2um~4.45um, As a monotonic factor, take .

8. The high-temperature pure gas phase combustion dual-band neural network radiation thermometry method according to claim 1, 2, 5, 6, or 7, is characterized in that, The pure gas phase intrinsic radiance corresponding to the first and second bands can be obtained using the following formula: in , indicating the band number, Corresponding to the first band, Corresponding to the second band; For the first Measured integrated radiance in the band; This is a combined coefficient of optical transmittance and detector response. For the first The inherent radiance of the pure gas phase corresponding to the band; For the CO2-H2O mixed gas in the first i Equivalent emissivity of the band; For the first Band-band wall blackbody radiation brightness; For the first Dark field offset in the band, For the first Measurement noise in the band.

9. The high-temperature pure gas phase combustion dual-band neural network radiation thermometry method according to claim 8, characterized in that, The pure gas phase intrinsic radiance corresponding to the first and second bands is input into the trained temperature prediction model. After the high-temperature pure gas phase thermodynamic temperature is output through network forward propagation, a confidence judgment step is also included: if the pure gas phase intrinsic radiance corresponding to the first and second bands exceeds the radiance envelope of the simulation dataset, or the ratio of the two bands exceeds a preset threshold, or the variance of the network integrated output exceeds a preset threshold, then the temperature measurement result output by the current temperature prediction model is determined to have low confidence, and a low confidence mark is output to indicate that there may be wall obstruction, residual condensed particles, window contamination, or abnormal detector saturation.

10. A high-temperature pure gas phase combustion dual-band neural network radiation temperature measurement device, characterized in that, include: The simulation dataset construction module is used to construct a sample space based on the SNB narrowband gas radiative transfer model and the HITRAN high-precision molecular spectral line database, considering only gas phase components CO2 and H2O. The variables include gas temperature, water vapor optical thickness, and carbon dioxide optical thickness. For each set of variables, the simulated brightness L1,sim and L2,sim were calculated by forward modeling in the first band (2.7μm~2.95μm) and the second band (4.2μm~4.45μm), respectively. At the same time, temperature labels corresponding to each set of variables were generated to construct a simulation dataset. The neural network model training module is used to construct a temperature prediction model with a multilayer perceptron (MLP) as the network backbone, and to train the temperature prediction model using a simulation dataset to obtain the trained temperature prediction model. The dual-band radiation acquisition and preprocessing module is used to acquire the measured integrated radiance of high-temperature gas in the first band (2.7μm–2.95μm) and the second band (4.2μm–4.45μm), subtract background radiation interference, and eliminate the effect of optical path attenuation to obtain the pure gas phase inherent radiance corresponding to the first and second bands. The temperature inference output module is used to input the pure gas phase intrinsic radiance corresponding to the first and second bands into the trained temperature prediction model, and output the high-temperature pure gas phase gas thermodynamic temperature through network forward propagation.

Images