Ultra-wideband coherent anti-Stokes Raman scattering CO2 spectrum temperature measurement method and system

Abstract

The invention discloses an ultra-wideband coherent anti-Stokes Raman scattering CO2 spectrum temperature measurement method and system, and a bidirectional long short term memory multi-output convolutional neural network CNN-BiLSTM-Attention combined with an attention mechanism is trained based on a CO2 CARS spectrum formed by simulation. The deep learning network aims to overcome the limitation of a traditional method, the accuracy, speed and reliability of CO2 temperature estimation are improved, and an innovative solution is provided for combustion science, environment monitoring and industrial application.

Description

Technical Field

[0001] This invention relates to the field of laser combustion diagnosis, and in particular to an ultrawideband coherent anti-Stokes Raman scattering CO2 spectral temperature measurement method and system. Background Technology

[0002] Ultra-wideband coherent anti-Stokes Raman scattering (CARS) based on sub-10 femtosecond light sources has become an important tool for non-contact gas thermometry due to its high temporal resolution, strong signal coherence, and multi-gas molecule detection capabilities induced by ultra-wideband spectroscopy. This technique excites molecular vibrational modes and generates an anti-Stokes signal through the nonlinear interaction of pump light, Stokes light, and probe light. Its spectral characteristics are directly related to the molecular Boltzmann distribution, thus enabling accurate temperature prediction. For example, the literature (A. Bohlin et al., JPhys Chem Lett 6 (4), 643-649 (2015)) used ultra-wideband femtosecond pump / Stokes light and picosecond probe light to measure the CARS spectra of multiple gas molecules such as N2 and O2 in hydrocarbon flames, demonstrating the applicability of ultra-wideband femtosecond CARS in high-temperature combustion fields. Similarly, a Chinese patent application (publication number CN115452202A) proposes a high-temperature thermocouple calibration method based on CARS spectroscopy to address the challenge of accurate, real-time measurement of high-temperature combustion gases over a wide range in existing technologies. However, its target temperature-measuring molecule is primarily N2, failing to adequately extend to other gas components. Existing femtosecond CARS-based temperature measurement research largely focuses on N2 molecules, primarily due to the abundance of N2's spectral characteristic parameters and its high abundance in air, making it an ideal probe for combustion diagnostics. In contrast, CARS temperature measurement research on CO2 molecules is relatively scarce. As a key product of combustion and a significant greenhouse gas, CO2's temperature information plays an irreplaceable role in multiple fields. For example, in combustion diagnostics: in combustion devices such as engines and gas turbines, the temperature distribution of CO2 directly reflects combustion efficiency and pollutant formation mechanisms; accurate temperature measurement can provide crucial data for improving fuel utilization and reducing emissions. In chemical and energy production fields, CO2 temperature monitoring is crucial for processes such as reactor optimization and carbon capture and storage.

[0003] Although CO2 thermometry is of great significance, its CARS signals are affected by factors such as molecular structure complexity and non-resonant background interference, making signal extraction and temperature prediction difficult. Existing solutions mostly rely on traditional fitting methods (such as least squares) and genetic algorithms, which have low processing efficiency and are easily limited by local optima, making it difficult to meet the real-time temperature measurement requirements in complex environments. For example, although a recent study (Gu et al., Proc. Combust. Inst. 38(1), 1599-1606(2020)) achieved CO2 thermometry using chirped probe pulse femtosecond CARS technology, its temperature estimation requires fitting as many as 7-12 independent laser parameters of a single experimental spectrum. This method makes the temperature estimation process very time-consuming and difficult to adapt to application scenarios with high requirements for real-time performance, robustness, or automation. Currently, no research has deeply integrated deep learning with ultra-wideband femtosecond CARS technology to specifically optimize the CO2 spectrum thermometry process. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an ultra-wideband coherent anti-Stokes Raman scattering CO2 spectral temperature measurement method and system to address the shortcomings of the existing technology, thereby solving the problem of difficulty in non-contact temperature measurement under high-temperature flow field and non-flow field conditions and realizing rapid and accurate temperature measurement.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: an ultra-wideband coherent anti-Stokes Raman scattering CO2 spectroscopic temperature measurement method, comprising the following steps:

[0006] S1. Obtain the CO2 CARS training set;

[0007] S2. Use the spectra in the CO2 CARS training set as input to the CNN-LSTM-Attention model to train the CNN-LSTM-Attention model and obtain the temperature measurement model;

[0008] The specific implementation process for obtaining the CO2 CARS training and validation sets includes:

[0009] The following simulation model was established: ;in, The value represents the contribution of the CARS resonant response, and the subscript n indicates the relevant parameters and process corresponding to the nth Raman transition. This represents the difference in particle number density between the upper and lower energy levels corresponding to the nth Raman transition. Let ω be the differential scattering cross section of the nth Raman transition. n Γ represents the center angular frequency of the nth Raman transition. n This represents the attenuation coefficient of the nth Raman transition caused by molecular collisions;

[0010] The simulation model is used to generate a dataset containing simulated CO2 CARS spectra; the dataset containing simulated CO2 CARS spectra includes temperature data and probe pulse width data;

[0011] The dataset containing simulated CO2 CARS spectra is randomly divided into a training set, a validation set, and a test set.

[0012] For a thermal equilibrium system at temperature T, the particle number density of CO2 molecules at specific vibrational energy level υ and rotational energy level J is expressed as: ;

[0013] Where N is the total particle number density within the effective detection area, and Q vib and Q rot Let ω represent the vibrational partition function and the rotational partition function, respectively. e =v e / c is the molecular vibrational constant, v e For the vibration frequency, g I A represents the statistical weights of atomic spins. =1.4388 K / cm -1 h is Planck's constant, c is the speed of light, k is the force constant, and B is the velocity of light. v Represents the rotational constant of a molecule, and the number density of particles in the upper energy level corresponding to the nth Raman transition. and lower energy level particle number density Calculating the difference yields the difference in particle number density between the upper and lower energy levels. .

[0014] .

[0015] Atomic spin statistical weights g I The determination process includes:

[0016] For fundamental band transitions, when J is odd, g I = 1; when J is even, g I = 0;

[0017] For tropical transitions, for all rotational energy levels satisfying J ≥ l2, g I = 1; where l2 is the bending vibration angular momentum quantum number.

[0018] Differential scattering cross section of the nth Raman transition The calculation formula is: Where ε0 is the vacuum permittivity, It detects the number of light waves. M is the transition frequency of the nth Raman transition. n Let represent the dipole moment of the nth Raman transition.

[0019] ω n The energy difference between the upper and lower energy levels involved in Raman transitions determines the energy of a specific vibrational energy level υ and rotational energy level J. The calculation formula is: ;

[0020] Among them, T e ω represents the minimum potential energy of the electronic state. e =v e / c is the molecular vibrational constant, v e The vibration frequency, B is a higher-order anharmonic factor. v and D v These are the rotational constant and centrifugal distortion constant at the vibrational energy level υ, respectively, ω n Depend on We obtain, where, and These represent the upper and lower energy levels corresponding to the nth Raman transition, respectively. is the reduced Planck constant.

[0021] The CNN-LSTM-Attention model includes:

[0022] A convolutional neural network (CNN) is used to extract features from the input simulated CO2 CARS spectrum to obtain a spectral feature sequence.

[0023] A bidirectional long short-term memory network (BiLSTM) is used to encode the spectral feature sequence into a context vector containing global information.

[0024] The attention mechanism layer is used to obtain high-dimensional features using the context vector and output the high-dimensional features to the fully connected layer, which maps the high-dimensional features to the temperature and pulse width to be estimated.

[0025] As an inventive concept, the present invention also provides an ultrawideband coherent anti-Stokes Raman scattering CO2 spectral temperature measurement system, including a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the steps of the above method.

[0026] A computer-readable storage medium having a computer program / instructions stored thereon; the computer program / instructions, when executed by a processor, implement the steps of the above-described method.

[0027] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention employs a deep learning-based CNN-BiLSTM-Attention model to measure the temperature of ultra-wideband coherent anti-Stokes Raman scattering CO2 spectra. This method exhibits significant comprehensive advantages: First, its end-to-end design achieves a direct and accurate mapping from the raw spectrum to physical parameters, avoiding error accumulation in step-by-step estimation; second, by automatically learning local spectral details, global dependencies, and key task regions, the model achieves a temperature estimation RMSE accuracy of approximately 2.27 K on an independent test set, validating its high accuracy and robustness in application; functionally, it achieves simultaneous and high-precision prediction of the two key parameters—temperature and probe pulse width—from a single CO2 CARS spectrum, providing an efficient and reliable solution for CARS spectral diagnostics in combustion, gasification, and other fields. It solves the problem of difficult non-contact temperature measurement under high-temperature flow and non-flow field conditions, enabling rapid and accurate temperature measurement. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the training process of the CNN-LSTM-Attention model according to an embodiment of the present invention.

[0029] Figure 2 For a test set of CO2 CARS spectra at a temperature of 673 K;

[0030] Figure 3 The model outputs pulse width and temperature results for 60 test set samples. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Example 1

[0033] Embodiment 1 of this invention provides a CARS spectral thermometry method based on a CNN-BiLSTM-Attention model, used to achieve high-precision and high-efficiency estimation of key physical parameters such as temperature and detection laser pulse width from a single CARS spectrum. This embodiment mainly includes the following two parts:

[0034] 1. Generation of CARS Spectral Simulation Dataset

[0035] A simulation model was developed using MATLAB. The pump light / Stokes light was an ultra-wideband source with a wavelength of less than 10 fs, which was approximated as an ultrashort pulse with the Fourier transform limit in the model. The simulation of the ultra-wideband CARS spectrum was carried out according to formula (1):

[0036] (1)

[0037] in, This represents the difference in particle number density between the upper and lower energy levels corresponding to the nth Raman transition. Let ω be the differential scattering cross section of the nth Raman transition. n Γ represents the center angular frequency of the nth Raman transition. n This represents the attenuation coefficient for the nth Raman transition caused by molecular collisions. This invention focuses on the thermometry of CO2 CARS spectra with a probe delay <10 ps, ​​where the influence of molecular collision effects (i.e., Γ) is negligible. n ≈0), therefore the exponential decay term in formula (1) This will not be discussed further here. The simulation principle of the three key physical parameters in formula (1), excluding the collision term, will be explained below:

[0038] (1) Particle number density difference

[0039] Characterizing the number difference between the upper and lower energy levels of a Raman transition, this directly reflects the Boltzmann distribution at different vibrational-rotational energy levels and is a core parameter determining the temperature dependence of the CARS spectral intensity distribution. For a thermal equilibrium system at temperature T, the number density of CO2 molecules at specific vibrational energy level υ and rotational energy level J can be expressed as:

[0040] (2)

[0041] Where N is the total particle number density within the effective detection area, and Q vib and Q rot Let ω represent the vibrational and rotational partition functions, respectively. e =v e / c is the molecular vibrational constant, v e For the vibration frequency, g I A represents the statistical weights of atomic spins. =1.4388 K / cm -1 h is Planck's constant, c is the speed of light, k is the force constant, and B is the velocity of light. v This represents the rotational constant of the molecule.

[0042] (3)

[0043] For the CO2 molecule, the statistical weights of its atomic spins vary depending on the specific transition, which is crucial for accurately simulating CO2 CARS spectra. In this simulation, based on the molecular spectroscopic theory of CO2 (H. Finsterhölzl, Berichte der Bunsengesellschaftfürphysikalische Chemie, 86, 797-805 (1982)), the g-spin statistical weights of the major contributing transitions to CO2 are... I The rules are set as follows:

[0044] For fundamental frequency band transition (00) 0 0)→(10 0 0) 1,2 g I = 1 (when J is odd); g I = 0 (when J is even).

[0045] For tropical transition (01) 1 0)→(11 1 0) 1,2 For all rotational energy levels satisfying J ≥ l2 (where l2 is the quantum number of bending vibrational angular momentum), g I = 1.

[0046] Particle number density in this simulation model The calculation must follow the above g I rule.

[0047] (2) Raman cross section

[0048] Raman cross section The probability of Raman scattering occurring in the nth Raman transition mode is described, which determines the intrinsic signal intensity of this vibration-rotation mode in the CARS spectrum. In this simulation model, the Raman cross section of the CO2 molecule is calculated using formula (4):

[0049] (4)

[0050] Where ε0 is the vacuum permittivity; It detects the light wavenumber; M is the transition frequency of the nth Raman transition; n Let M represent the dipole moment of the nth Raman transition. For CO2, M is used in the simulation. n See the table below:

[0051] surface The CO2 transition dipole moment M used in the simulation n and its corresponding Raman translation ν f

[0052]

[0053] (3) Molecular Raman transition frequency ω n

[0054] ω n ω is the central angular frequency of the nth Raman transition, which determines the position of the spectral peak in the CARS spectrum. n The energy difference between the upper and lower energy levels involved in the Raman transition is determined by the energy level difference between them. The energy of the vibrational energy level J and the rotational energy level J at a specific υ can be expressed as shown in formula (5):

[0055] Among them, T e ω represents the minimum potential energy of the electronic state. e =v e / c is the molecular vibrational constant, v e The vibration frequency, B is a higher-order anharmonic factor. v and D v These are the rotational constant and centrifugal distortion constant at this vibrational energy level, respectively. This formula, by introducing higher-order correction terms such as centrifugal distortion, accurately describes the frequency shift caused by changes in internuclear distance during molecular rotation, ensuring the correspondence between the simulated spectral peak positions and the actual molecular energy level structure.

[0056] Based on the aforementioned physical model, a dataset containing over 60,000 simulated CO2 CARS spectra was first generated. The dataset's parameter space covers temperatures of 300-1300 K (1 K step) and probe pulse widths of 1.4-2 ps (0.01 ps step). This dataset was then randomly divided into training, validation, and test sets in a 70%:15%:15% ratio for the training and evaluation of the deep learning model.

[0057] In this embodiment, based on the CARS spectral theoretical model and the molecular spectral parameters of CO2, 60,000 simulated CO2 CARS spectra with different temperature and pulse width combinations are generated in batches within a set temperature range of 300-1300 K (step size 1 K) and a probe laser pulse width range of 1.4-2.0 ps (step size 0.01 ps). Each simulated spectrum strictly corresponds to a set of real physical parameters (temperature T, pulse width τ), which are referred to as the label of the corresponding spectral data in this embodiment. Subsequently, the simulated CARS spectral data are normalized, and preliminary spectral alignment is performed based on the maximum value of the CO2 CARS spectral peaks, and the data is divided into training, validation, and test sets.

[0058] 2. Training the CNN-LSTM-Attention model

[0059] Build as Figure 1 The CNN-BiLSTM-Attention model shown uses 70% of the training set and 15% of the validation set as input for training the network model. The spectra in the training set are fed in batches into the CNN module to automatically extract local fine features such as peak shape and width. Subsequently, the BiLSTM module is used to capture the forward and backward dependencies of the spectral sequence. The Attention mechanism module dynamically assigns weights to different spectral regions, focusing on the spectral feature regions most critical for temperature or pulse width estimation. Finally, a fully connected layer maps the learned high-dimensional features to the temperature and pulse width to be estimated.

[0060] The CO2CARS training set spectrum is input into the training module of this CNN-LSTM-Attention model for training. During training, the validation set is used to evaluate the performance. The functions of the three main modules are as follows:

[0061] (1) Convolutional Neural Networks (CNNs) are used to automatically extract local spectral features. CNNs first extract features from the input CARS spectra. Their convolutional kernels scan the wavenumber axis of the spectrum, a process that automatically learns and captures fine local patterns that are crucial for the estimation of physical parameters. For example: (a) the shape and fine structure of spectral peaks: such as the rotational fine structure feature peaks of the 2ν2 vibration band (~1388 cm⁻¹ and ~1409 cm⁻¹), their respective peak widths, asymmetries, and shoulder features; (b) the relative relationships and trends of spectral lines. Through multi-layer convolution and nonlinear activation, CNNs can abstract the original spectral data layer by layer into high-level features, that is, convert one-dimensional spectral signals into feature vectors that can characterize local spectral patterns and their sequence correlations, providing rich structured information for subsequent processing modules.

[0062] (2) A Bidirectional Long Short-Term Memory (BiLSTM) network is used to capture the global dependencies of a spectral sequence. The feature sequence extracted by the CNN is input into the BiLSTM for context information extraction. Through its unique gating mechanism, the BiLSTM can selectively memorize and fuse information, effectively capturing long-range dependencies within the sequence. Specifically, the network can scan the entire feature sequence from both positive and negative directions to understand and model the following global physical relationships: (a) analyzing the intensity ratio and peak spacing between the rotational fine structure feature peaks of the 2ν2 vibrational band, which are highly temperature-dependent features; (b) identifying spectral broadening caused by changes in the probe laser pulse width or overall spectral shift caused by systematic errors; and (c) understanding the evolution of the intensity distribution of rotational branches caused by temperature changes. Finally, the BiLSTM selectively remembers important spectral evolution patterns and forgets irrelevant details, ultimately encoding the entire spectral sequence into a context vector containing global information.

[0063] (3) Attention mechanism is used for dynamic feature selection and information weighting. The attention mechanism is the core of the decision-making of this model. It receives the global sequence dependencies extracted by BiLSTM and performs a feature importance re-evaluation and information fusion based on the task objective. Finally, it outputs a refined feature representation that is highly relevant to the regression objective, which is then used by the subsequent fully connected layers for high-precision temperature and pulse width regression estimation. Its core functions include: (a) The model automatically learns and identifies the spectral regions that are most discriminative to the current regression objective. For example, temperature estimation mainly depends on the intensity ratio of the rotational fine structure feature peaks of the 2ν2 vibration band, while pulse width estimation depends more on the overall broadening and edge morphology of the spectral lines. The attention mechanism achieves adaptive focusing on key spectral regions of different tasks by adjusting the weight distribution. (b) A task-oriented context vector is generated by weighted summation of all hidden states. This vector strengthens the contribution of important features while weakening the interference of noise or irrelevant spectral regions, thereby achieving information purification and enhancement.

[0064] The cascaded model is trained using a training set. In this embodiment, the network model is trained using a batch iterative approach. In each iteration, 64 CARS spectra from the training set are simultaneously input into the network model. The 64 predicted temperature and pulse width values ​​output by the model are compared with their corresponding labels, the loss function is calculated, and the results are used to optimize the network model weights through backpropagation. Each training iteration consists of 657 iterations to ensure that the training set is traversed. After each training iteration, the validation set spectra are input into the network model, and the output results are scored according to the labels, serving as the evaluation metric for the model's performance in that iteration. The maximum number of training iterations is set to 50. Early stopping is used to prevent overfitting until the model's loss on the validation set no longer decreases significantly, resulting in a final trained model with fixed weight parameters. During training, the root mean square error (RMSE) is used as the loss function, and the model parameters are iteratively updated using the backpropagation algorithm and the Adam optimizer. Meanwhile, the training process is monitored using an independent validation set, and early stopping is used to prevent overfitting of the model until the loss of the model on the validation set no longer decreases significantly. This yields a final trained model with fixed weight parameters, and the model performance is evaluated using a test set.

[0065] To verify the effectiveness and robustness of the trained end-to-end CNN-BiLSTM-Attention model, after the model training was completed, its temperature estimation performance was quantitatively evaluated using 15% of the reserved test set CO2 CARS spectral data that the model had not seen before.

[0066] Figure 2 A CO2 CARS spectrum of a test set at 673 K is given, with the wavenumber coverage of the input spectrum being ~1150–1550 cm⁻¹. -1 Each spectrum contains 488 data points. Sixty spectral samples are randomly selected from the test set and input into the trained CNN-BiLSTM-Attention model to obtain their temperature / pulse width predictions, which are then compared with the labels.

[0067] Figure 3The pulse width and temperature estimation results for 60 test samples are presented. Blue circles represent the model's output estimates, and red crosses represent the true label values. Test results show that the model exhibits excellent performance on unseen spectral data, with RMSEs of 2.2733 K and 4.1822 fs for estimated temperature and pulse width, respectively. In terms of efficiency, with a training set of 42,000 samples and 50 training epochs, the total training time on a desktop computer with an NVIDIA GeForce RTX 4090 GPU is ≤4 min. The single estimation time after training is <10 ms. These results directly confirm that the scheme of this invention can simultaneously, accurately, and timely estimate multiple key physical parameters from a single spectrum.

[0068] In summary, this invention proposes and verifies a deep learning-based ultrawideband CARSCO2 spectral thermometry method. The core of this method lies in constructing and training a CNN-BiLSTM-Attention cascaded neural network model, achieving end-to-end, high-precision estimation from the raw spectrum to the target physical parameters. Furthermore, this approach eliminates the reliance on manual feature engineering and complex iterative fitting algorithms, significantly improving parameter estimation efficiency and providing a promising technical solution for real-time, accurate temperature measurement in fields such as combustion diagnostics.

[0069] Example 2

[0070] Embodiment 2 of the present invention provides a system corresponding to Embodiment 1 above, including a memory, a processor, and a computer program stored in the memory; the processor executes the computer program in the memory to implement the steps of the method of Embodiment 1 above.

[0071] In some implementations, the memory may be high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device.

[0072] In other implementations, the processor can be any type of general-purpose processor, such as a central processing unit (CPU) or a digital signal processor (DSP), and there is no limitation here.

[0073] Example 3

[0074] Embodiment 3 of the present invention provides a computer-readable storage medium corresponding to Embodiment 1 above, on which a computer program / instructions are stored. When the computer program / instructions are executed by a processor, they implement the steps of the method of Embodiment 1 above.

[0075] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0076] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0077] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0078] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0079] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0080] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for measuring the temperature of CO2 using ultrawideband coherent anti-Stokes Raman scattering spectrometry, characterized in that, Includes the following steps: S1. Obtain the CO2CARS training set; S2. Use the spectra in the CO2CARS training set as input to the CNN-LSTM-Attention model to train the CNN-LSTM-Attention model and obtain the temperature measurement model; The specific implementation process for obtaining the CO2CARS training and validation sets includes: The following simulation model was established: ;in, The value represents the contribution of the CARS resonant response, and the subscript n indicates the relevant parameters and process corresponding to the nth Raman transition. This represents the difference in particle number density between the upper and lower energy levels corresponding to the nth Raman transition. Let ω be the differential scattering cross section of the nth Raman transition. n Γ represents the center angular frequency of the nth Raman transition. n This represents the attenuation coefficient of the nth Raman transition caused by molecular collisions; The simulation model is used to generate a dataset containing simulated CO2CARS spectra; the dataset containing simulated CO2CARS spectra includes temperature data and probe pulse width data; The dataset containing simulated CO2CARS spectra is randomly divided into a training set, a validation set, and a test set.

2. The ultra-wideband coherent anti-Stokes Raman scattering CO2 spectroscopic temperature measurement method according to claim 1, characterized in that, For a thermal equilibrium system at temperature T, the particle number density of CO2 molecules at specific vibrational energy level υ and rotational energy level J is expressed as: ； Where N is the total particle number density within the effective detection area, and Q vib and Q rot Let ω represent the vibrational partition function and the rotational partition function, respectively. e =v e / c is the molecular vibrational constant, v e For the vibration frequency, g I A represents the statistical weights of atomic spins. =1.4388 K / cm -1 h is Planck's constant, c is the speed of light, k is the force constant, and B is the velocity of light. v Represents the rotational constant of a molecule, and the number density of particles in the upper energy level corresponding to the nth Raman transition. and lower energy level particle number density Calculating the difference yields the difference in particle number density between the upper and lower energy levels. .

3. The ultra-wideband coherent anti-Stokes Raman scattering CO2 spectroscopic temperature measurement method according to claim 2, characterized in that, 。 4. The ultra-wideband coherent anti-Stokes Raman scattering CO2 spectroscopic temperature measurement method according to claim 2, characterized in that, Atomic spin statistical weights g I The determination process includes: For fundamental band transitions, when J is odd, g I = 1; when J is even, g I = 0; For tropical transitions, for all rotational energy levels satisfying J ≥ l2, g I = 1; where l2 is the bending vibration angular momentum quantum number.

5. The ultra-wideband coherent anti-Stokes Raman scattering CO2 spectroscopic temperature measurement method according to claim 1, characterized in that, Differential scattering cross section of the nth Raman transition The calculation formula is: Where ε0 is the vacuum permittivity, It detects the number of light waves. M is the transition frequency of the nth Raman transition. n Let represent the dipole moment of the nth Raman transition.

6. The ultra-wideband coherent anti-Stokes Raman scattering CO2 spectroscopic temperature measurement method according to claim 1, characterized in that, ω n The energy levels at a specific vibrational energy level υ and rotational energy level J are determined by the energy difference between the upper and lower energy levels involved in Raman transitions. The calculation formula is: ； Among them, T e ω represents the minimum potential energy of the electronic state. e =v e / c is the molecular vibrational constant, v e The vibration frequency, B is a higher-order anharmonic factor. v and D v These are the rotational constant and centrifugal distortion constant at the vibrational energy level υ, respectively, ω n Depend on We obtain, where, and These represent the upper and lower energy levels corresponding to the nth Raman transition, respectively. is the reduced Planck constant.

7. The ultra-wideband coherent anti-Stokes Raman scattering CO2 spectroscopic temperature measurement method according to claim 1, characterized in that, The CNN-LSTM-Attention model includes: A convolutional neural network (CNN) is used to extract features from the input simulated CO2CARS spectrum to obtain a spectral feature sequence. A bidirectional long short-term memory network (BiLSTM) is used to encode the spectral feature sequence into a context vector containing global information. The attention mechanism layer is used to obtain high-dimensional features using the context vector and output the high-dimensional features to the fully connected layer, which maps the high-dimensional features to the temperature and pulse width to be estimated.

8. A wideband coherent anti-Stokes Raman scattering CO2 spectral temperature measurement system, comprising a memory, a processor, and a computer program stored in the memory; characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instructions stored thereon; characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

Images