A method and apparatus for deconvolution of pulsed x-ray energy spectra

Abstract

A deconvolution method and system for pulse X-ray energy spectrum measurement are applied to pulse X-rays generated by an ultra-short and ultra-strong laser device, the method comprising: combining a predetermined number of experimental data and simulation data into a first data set, selecting part of the data from the first data set as a training data set of a neural network model; training the neural network model by using the training data set; calculating a predicted energy spectrum by using a diagnostic physical parameter in the ultra-short and ultra-strong laser device and the trained neural network model according to the real-time collected diagnostic physical parameter; and obtaining a pulse X-ray energy spectrum by a pre-set deconvolution algorithm according to the predicted energy spectrum.

Description

Technical Field

[0001] This application relates to the field of energy spectrum interpretation, and more particularly to a method and apparatus for interpreting pulsed X-ray energy spectrum measurements. Background Technology

[0002] Pulsed X-ray energy spectrum diagnostics is an important component of plasma diagnostics, condition monitoring, and radiation protection for ultrashort, ultra-intense laser devices. When an ultrashort, ultra-intense laser interacts with a target material, the pulsed laser ionizes the atoms on the target surface in an extremely short time, forming a high-temperature, high-density plasma. The main pulsed laser couples with the plasma, transferring energy to electrons through different acceleration mechanisms. The interaction between the electrons and the target generates pulsed X-rays.

[0003] Pulsed X-rays have the advantages of short pulse duration (~ps) and large dose (~10). 12 keV -1 ·sr -1 Due to their wide energy spectrum (~ tens of MeV), commonly used detectors are limited by dead time and accumulation effect and cannot be applied.

[0004] Therefore, different energy spectrum measurement methods have been developed for pulsed X-rays of different energies. Low-energy (eV to keV) X-rays are often measured using crystal diffraction, while high-energy (several MeV to tens of MeV) X-rays are measured using Compton scattering or activation methods. For scenarios such as laser inertial confinement fusion, monitoring the operational status of laser devices, and X-ray imaging, researchers are more concerned with measuring medium-energy (keV to several MeV) X-rays, which are mainly measured using absorption methods. Some technologies struggle to meet the requirements for real-time and accurate measurement of pulsed X-ray energy spectra within ultrashort, ultra-intense laser devices. Therefore, there is an urgent need for a real-time spectral interpretation method for pulsed X-rays in ultrashort, ultra-intense laser devices. Summary of the Invention

[0005] This application provides a method and apparatus for interpreting the spectrum of pulsed X-ray energy spectrum measurement. By using a neural network model and a spectrum interpretation algorithm, the accurate distribution of pulsed X-rays inside and outside the target chamber can be obtained in real time. This solves the problem of dependence on the accuracy of the initial spectrum in traditional iterative methods, as well as the problem of not being able to interpret the spectrum of pulsed X-rays in real time.

[0006] This application provides a spectral interpretation method for pulsed X-ray energy spectrum measurement, applied to pulsed X-rays generated by an ultrashort, ultra-intense laser device. The method includes:

[0007] A predetermined amount of experimental data and simulation data are combined into a first dataset, and a portion of the data is selected from the first dataset as the training dataset for the neural network model.

[0008] The neural network model is trained using this training dataset;

[0009] Based on the diagnostic physical parameters collected in real time from the ultra-short and ultra-intense laser device, the predicted energy spectrum is calculated using the diagnostic physical parameters and the trained neural network model.

[0010] The pulsed X-ray energy spectrum is obtained from the predicted energy spectrum using a pre-set spectral decomposition algorithm.

[0011] In one exemplary embodiment, the simulation data is obtained through the following steps:

[0012] The electron distribution information of superthermal electrons in the plasma on the target surface at different times was determined by plasma numerical simulation method. The electron distribution information at different times was fitted and calculated to determine the final characteristic temperature.

[0013] The electron distribution information at the time corresponding to the determined final characteristic temperature is used as the electron source term, and the electron source term is input into the preset Monte Carlo calculation model to calculate the pulse X-ray energy spectrum at that time. The pulse X-ray energy spectrum calculated at that time is used as the pulse X-ray energy spectrum corresponding to the simulation conditions.

[0014] The calculated pulsed X-ray energy spectrum was used as simulation data.

[0015] In one exemplary embodiment, before determining the electron distribution information of superthermal electrons at different times within the plasma on the target surface using plasma numerical simulation methods, the method further includes:

[0016] The empirical characteristic temperature is determined based on the ultrashort and ultra-intense laser power density in the experimental data.

[0017] The initial characteristic temperature is obtained by multiplying the empirical characteristic temperature by a predetermined scaling parameter.

[0018] The second characteristic temperature is obtained by adding a pre-set increment to the initial characteristic temperature;

[0019] The i-th characteristic temperature is obtained by adding the pre-set increment to the (i-1)-th characteristic temperature; where i is an integer greater than or equal to 3.

[0020] The maximum value of the i-th characteristic temperature does not exceed the cutoff characteristic temperature; the cutoff characteristic temperature is κ times the empirical characteristic temperature, where κ is 3, 4 or 5.

[0021] In one exemplary embodiment, the step of fitting and calculating the electron distribution information at different times to determine the final characteristic temperature includes:

[0022] Obtain electron distribution information at different times from the start to the end of the laser pulse, and determine the electron distribution information corresponding to multiple characteristic temperatures between the initial characteristic temperature and the cutoff characteristic temperature;

[0023] The obtained electron distribution information data points were fitted using Boltzmann, Maxwell, and relativistic Maxwell distribution fitting algorithms, respectively.

[0024] Calculate the goodness-of-fit value for each fitted electronic spectrum data curve;

[0025] The curve generated by the fitting algorithm corresponding to the minimum value among the goodness-of-fit values ​​is selected as the optimal fitted electronic spectrum data curve.

[0026] Determine the optimal fitted electron spectrum data curve and corresponding characteristic temperature at different times, select the characteristic temperature with the largest absolute value, and take the characteristic temperature with the largest absolute value as the final characteristic temperature.

[0027] In one exemplary embodiment, after training the neural network model using the training dataset, the method further includes:

[0028] The remaining data from the first dataset will be used as the test dataset for the neural network model;

[0029] And the evaluation parameters are used to judge the neural network model;

[0030] When the evaluation parameters of the neural network model are less than the preset evaluation parameter threshold, the neural network model is determined to be a well-trained neural network model.

[0031] The evaluation parameters are as follows:

[0032]

[0033] In the formula, λ1, λ2, and λ3 are constants defined in the region (0, 1), set in a ratio of 10:1:1; M is the experimental value measured by the absorption detector, a vector with m values; A is the response probability of the detector system, an m×n matrix, where any element a ij The probability that the energy of a pulsed X-ray with energy i is measured by the j-th detection unit of the detector is represented; N is the true pulsed X-ray energy spectrum distribution, which is a vector with n values; ||M-AN||2 represents the L2 norm of the vector; ndims() represents the dimension of the vector.

[0034] In one exemplary embodiment, the diagnostic physical parameters include at least one of the following: laser energy, pulse width, focusing size, energy concentration, incident angle, laser polarization mode, target thickness and mass number, and pre-plasma density distribution.

[0035] In one exemplary embodiment, after obtaining the pulsed X-ray energy spectrum based on the predicted energy spectrum using a pre-set spectral resolution algorithm, the method further includes:

[0036] The pulsed X-ray energy spectrum obtained by spectrum decomposition is compared with the predicted energy spectrum obtained by neural network calculation to make a judgment;

[0037] When the difference between the pulsed X-ray energy spectrum obtained by spectral analysis and the predicted energy spectrum is greater than a preset value, it is then determined whether the number of data groups with a difference greater than the preset value reaches a specific percentage of the total data.

[0038] When a certain percentage of the total data is reached, update the data in that data group to the first dataset.

[0039] The neural network model is retrained based on the updated first dataset.

[0040] In one exemplary embodiment, the pulsed X-ray energy spectrum obtained by a pre-set spectral resolution algorithm is displayed.

[0041] In one exemplary embodiment, after determining whether the number of data groups greater than a preset value reaches a specific percentage of the total data, the method further includes:

[0042] If the number of data groups exceeding a preset value does not reach a specific percentage of the total data, the data in that data group will be stored.

[0043] This application also provides a spectroscopic interpretation device for pulsed X-ray energy spectroscopy measurement, the device comprising: a memory and a processor; characterized in that, the memory is used to store a program for spectroscopic interpretation of pulsed X-ray energy spectroscopy measurement, and the processor is used to read and execute the program for spectroscopic interpretation of pulsed X-ray energy spectroscopy measurement, and execute the spectroscopic interpretation method for pulsed X-ray energy spectroscopy measurement as described in any of the above embodiments.

[0044] Compared with related technologies, this application provides a method and apparatus for interpreting the energy spectrum of pulsed X-rays, applicable to pulsed X-rays generated by an ultra-short, ultra-intense laser device. The method includes: combining a predetermined amount of experimental data and simulated data into a first dataset; selecting a portion of data from the first dataset as a training dataset for a neural network model; training the neural network model using the training dataset; calculating a predicted energy spectrum based on real-time acquired diagnostic physical parameters from the ultra-short, ultra-intense laser device and the trained neural network model; and obtaining the pulsed X-ray energy spectrum using a pre-set interpretation algorithm based on the predicted energy spectrum. This application utilizes a neural network model to obtain the accurate distribution of pulsed X-rays inside and outside the target chamber in real time, solving the problem of dependence on the accuracy of the initial spectrum in traditional iterative methods and the inability to interpret the pulsed X-ray spectrum in real time.

[0045] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. Other advantages of this application can be realized and obtained by means of the solutions described in the description and the accompanying drawings. Attached Figure Description

[0046] The accompanying drawings are used to provide an understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0047] Figure 1 This is a flowchart of the spectral interpretation method for pulsed X-ray energy spectroscopy measurement according to an embodiment of this application;

[0048] Figure 2 This is a schematic diagram of the spectral interpretation device for pulsed X-ray energy spectroscopy measurement according to an embodiment of this application;

[0049] Figure 3 This is a schematic diagram of the spectral interpretation system for pulsed X-ray energy spectroscopy measurement according to an embodiment of this application;

[0050] Figure 4 This is a flowchart illustrating an exemplary embodiment of a method for interpreting the spectrum of pulsed X-rays in a real-time energy spectrum measurement using an updated neural network combining Monte Carlo and PIC methods. Detailed Implementation

[0051] This application describes several embodiments, but these descriptions are exemplary and not restrictive, and it will be apparent to those skilled in the art that many more embodiments and implementations are possible within the scope of the embodiments described herein. Although many possible combinations of features are shown in the drawings and discussed in the detailed description, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with, or may replace, any feature or element of any other embodiment.

[0052] This application includes and contemplates combinations of features and elements known to those skilled in the art. The embodiments, features, and elements disclosed in this application may also be combined with any conventional features or elements to form a unique inventive scheme as defined by the claims. Any feature or element of any embodiment may also be combined with features or elements from other inventive schemes to form another unique inventive scheme as defined by the claims. Therefore, it should be understood that any feature shown and / or discussed in this application may be implemented individually or in any suitable combination. Therefore, the embodiments are not limited except by the limitations imposed by the appended claims and their equivalents. Furthermore, various modifications and changes may be made within the scope of the appended claims.

[0053] Furthermore, in describing representative embodiments, the specification may have presented methods and / or processes as a specific sequence of steps. However, the method or process should not be limited to the specific order of steps described herein, to the extent that it does not depend on such a specific order. As will be understood by those skilled in the art, other sequences of steps are also possible. Therefore, the specific order of steps set forth in the specification should not be construed as a limitation of the claims. Moreover, the claims concerning the method and / or process should not be limited to the steps performed in the written order, and those skilled in the art will readily understand that these orders can be varied and still remain within the spirit and scope of the embodiments of this application.

[0054] Pulsed X-rays have the advantages of short pulse duration (~ps) and large dose (~10). 12 keV -1 〃sr -1 The wide energy spectrum range (~ tens of MeV) of pulsed X-rays makes commonly used detectors unsuitable due to limitations in dead time and stacking effects. Different energy spectrum measurement methods have been developed for pulsed X-rays of different energies. Low-energy (eV~keV) X-rays are often measured using crystal diffraction, while high-energy (several MeV~tens of MeV) X-rays are measured using Compton scattering or activation methods. For applications such as laser inertial confinement fusion, monitoring the operational status of laser devices, and X-ray imaging, researchers are generally more interested in measuring medium-energy (keV~several MeV) X-rays using absorption methods.

[0055] Some technologies employ absorption-based spectrometers for medium-energy (keV to several MeV) X-rays. These spectrometers offer advantages such as small size, simple structure, and the ability to measure absolute X-ray fluence. Most ultra-short, ultra-intense laser facilities both domestically and internationally are equipped with absorption-based pulsed X-ray spectrometers. By selecting appropriate spectral interpretation methods, these absorption-based pulsed X-ray spectrometers can be used to measure the spatial distribution of medium-energy pulsed X-ray spectra both inside and outside the target chamber of ultra-short, ultra-intense laser facilities.

[0056] In pulsed X-ray spectrometers based on absorption methods, the most widely used spectral interpretation methods are the least squares method and the maximum entropy method. While these two methods are relatively mature, their slow interpretation speed is a major limitation. With the development of artificial intelligence, deep learning and other technologies have been applied to nuclear measurement, providing rapid prediction results. However, the accuracy of these predictions largely depends on the dimensionality and accuracy of the dataset. None of the above methods can meet the requirements for real-time and accurate measurement of pulsed X-ray energy spectra in ultrashort, ultra-intense laser devices. Based on these problems, the reasons are analyzed as follows:

[0057] I. With the continuous development and upgrading of laser devices, the target firing frequency is also rapidly increasing, especially in low power density (10⁻⁶) laser applications. 14 W / cm 2 Under these conditions, the firing frequency has reached 1MHz, and some ultra-intense and ultra-short laser experimental devices have achieved 10PW beamlines at high power densities (10 PW). 19 W / cm 2 With a firing interval of only 1 minute and a 1PW beam repetition rate as high as 10Hz, this places higher speed requirements on online pulsed X-ray energy spectrum measurement.

[0058] Both the maximum entropy method and the least squares method are essentially iterative methods. If the initial spectrum is input based on personal experience or a uniformly distributed spectrum is used as the input parameter of the spectrum solving program, the time spent searching the parameter space will be increased, and it may even be difficult to reach the pre-set convergence threshold, increase the number of iterations, or fail to give an ideal convergence result within a given number of iterations.

[0059] Second, considering the inherent characteristics of ultrashort and ultra-intense laser devices, numerous factors influence the final pulsed X-ray energy spectrum distribution. For example, unavoidable energy or time fluctuations exist in current experiments, altering the plasma state and ultimately affecting the X-ray spectrum distribution. In spectral analysis methods, taking the widely used SAND-II algorithm as an example, some research results indicate a direct relationship between the spectral analysis result and the accuracy of the input spectrum. Manually inputting the energy spectrum based on experience will reduce the accuracy of the analysis to some extent. Reasonably predicting the energy spectrum based on physical experimental conditions and narrowing the solution space will not only significantly shorten the analysis time but also simultaneously meet the requirements for accurate measurement.

[0060] In experiments involving ultrashort, ultra-intense lasers interacting with targets, parameter tuning is essential. To obtain particle beams with ideal divergence, energy, charge, and quasi-monoenergetic properties, or to achieve a certain acceleration mechanism, it is necessary to continuously adjust a series of parameters, including laser energy, pre-pulse laser energy, pulse width, incident angle, polarization direction, focused peak power, target physical state (solid, gaseous, or a combination of both), target mass number, and shape. Neural networks can provide reasonable predictions based on a large amount of reliable training data, but they perform poorly when predictions are not included in the training set. Clearly, neither experimental nor simulated data provides a sufficient training set for neural networks. Therefore, in predicting pulsed X-ray spectra in experiments involving ultrashort, ultra-intense lasers interacting with targets, traditional neural network models can only provide reasonable assumptions within the training set, and their universality needs further optimization. Based on these issues, the applicant, based on experimental conditions such as laser power density and target material, classifies and processes experimental data to improve the accuracy of neural network predictions, providing a reasonable energy spectrum distribution and improving the accuracy of the input spectrum.

[0061] Based on the above analysis, a method and system for real-time energy spectrum interpretation of pulsed X-rays in ultra-short and ultra-intense laser devices are proposed.

[0062] This disclosure provides a method for interpreting the spectrum of pulsed X-ray energy spectroscopy measurements, such as... Figure 1 As shown, the method includes steps S100-S130.

[0063] S100. Combine a predetermined number of experimental data and simulation data into a first dataset, and select a portion of the data from the first dataset as the training dataset for the neural network model;

[0064] S110. Train the neural network model using the training dataset;

[0065] S120. Based on the diagnostic physical parameters collected in real time from the ultra-short and ultra-intense laser device, the predicted energy spectrum is calculated using the diagnostic physical parameters and the trained neural network model.

[0066] S130. The pulsed X-ray energy spectrum is obtained by using a pre-set spectral resolution algorithm based on the predicted energy spectrum.

[0067] In this embodiment, the method can be applied to pulsed X-rays generated by an ultrashort, ultra-intense laser device. An ultrashort, ultra-intense laser device refers to a laser with a power density of 10... 18 W / cm 2 um 2 Above the horizontal level, devices with laser pulse lengths ranging from fs to ps.

[0068] In one exemplary embodiment, experimental data from an ultrashort, ultra-intense laser device is classified. This classification can be based on laser power density and target material properties, with a neural network model trained for each category. For example, the experimental data can be divided into four categories: the first category contains data with a power density of 10... 18 ~10 20 W / cm 2 um 2 Horizontal laser and low-Z target experiments, the second type is with power density at 10 18 ~10 20 W / cm 2 um 2 Horizontal laser and high-Z target experiments, the third type is power density greater than 10 20 W / cm 2 um 2 Horizontal laser and low-Z target experiments, the fourth category is power density greater than 10 20 W / cm 2 um 2 Horizontal laser and high Z target experiment.

[0069] In one exemplary embodiment, for each type of data, simulated data is calculated using the PIC method. The simulated data is obtained through the following steps: determining the electron distribution information of hyperthermic electrons within the target surface plasma at different times using the PIC plasma method; fitting the electron distribution information at different times to determine the final characteristic temperature; using the electron distribution information corresponding to the determined final characteristic temperature as an electron source term, and inputting this electron source term into a pre-set Monte Carlo calculation model to calculate the pulsed X-ray energy spectrum under the simulated conditions; and using the calculated pulsed X-ray energy spectrum as the simulated data. Specifically, the hyperthermic electron information includes the spatial position of the hyperthermic electrons within the target surface plasma, the angle between the hyperthermic electrons and the positive directions of the X, Y, and Z axes of the Cartesian coordinate system, the cosine value of the angle, the energy and momentum of the hyperthermic electrons, the laser-to-hyperthermic electron energy conversion efficiency, and weighting information, etc. In this embodiment, the process of inputting the electron source term into a preset Monte Carlo calculation model to calculate the pulsed X-ray energy spectrum at that moment is as follows: Using Monte Carlo software, a full model of the geometry and materials of the target chamber and target material of the ultrashort, ultra-intense laser device is performed. The phase space file is used as the radiation source term to calculate the pulsed X-ray energy spectrum at typical locations inside and outside the target chamber. The flux (MeV) of the pulsed X-ray spectrum at 350 keV within the target chamber is calculated. -1 Normalization was performed, and the external X-ray flux (MeV) was calculated based on a pulsed X-ray at 600 keV. -1 After normalization, the pulsed X-ray energy spectra corresponding to the experimental conditions of interaction between four types of ultrashort and ultra-intense lasers and targets were finally obtained.

[0070] In one exemplary embodiment, before determining the electron distribution information of superthermal electrons at different times within the target surface plasma using the PIC method, the method further includes: a first step of determining an empirical characteristic temperature based on the ultrashort, ultra-intense laser power density in the experimental data. In this embodiment, the empirical characteristic temperature T is given based on the ultrashort, ultra-intense laser power density. wilks :

[0071]

[0072] In the formula, I represents the laser power density, W / cm². 2 λ is the laser wavelength, in μm;

[0073] The second step is to multiply the empirical characteristic temperature by a predetermined scaling parameter to obtain the initial characteristic temperature; for example: T wilks The empirical characteristic temperature is multiplied by a predetermined scaling parameter to obtain the initial characteristic temperature; that is, the initial characteristic temperature T is T0. wilks A certain proportion γ, i.e., t = γT wilks γ can be 5%, 10%, or 15%.

[0074] The third step is to add a pre-set increment to the initial characteristic temperature to obtain the second characteristic temperature.

[0075] The maximum value of the i-th characteristic temperature does not exceed the cutoff characteristic temperature; the cutoff characteristic temperature is κ times the empirical characteristic temperature, where κ is 3, 4, or 5. In this step, the calculation process is as follows: the i-th characteristic temperature is obtained by adding the preset increment to the (i-1)-th characteristic temperature; where i is an integer greater than or equal to 3. For example: Step 1. Add the preset increment ΔT to the initial characteristic temperature T to obtain the second characteristic temperature, and further determine whether the second characteristic temperature reaches the cutoff characteristic temperature; if not, proceed to Step 2; Step 2. Add the preset increment ΔT to the second characteristic temperature to obtain the third characteristic temperature; and determine whether the third characteristic temperature reaches the cutoff characteristic temperature; if not, proceed to Step 3; ... Step i-1. Add the preset increment ΔT to the (i-1)-th characteristic temperature to obtain the i-th characteristic temperature, and further determine whether the i-th characteristic temperature reaches the cutoff characteristic temperature; if it reaches the cutoff characteristic temperature, the calculation ends. In this embodiment, the cutoff characteristic temperature is a multiple of the empirical characteristic temperature (κ), which is an upper limit of the increment, i.e., the maximum value. In other words, the minimum value of T is 5%T, and the maximum is κT. The final selected characteristic problem is a temperature value located within the interval (5%T, κT). For example: T wilks The empirical characteristic temperature is multiplied by a predetermined scaling parameter to obtain the initial characteristic temperature; that is, the initial characteristic temperature T is T0. wilks A certain proportion γ, i.e., T = γT wilksγ can be 5%, 10%, or 15%. An increment ΔT is assigned to the characteristic temperature T, i.e., t = T + ΔT, and the magnitude of the increment can be T. wilks A certain proportion γ, i.e., ΔT = γT wilks γ can be 5%, 10%, or 15%, and step three is repeated until T reaches T0. wilks A certain multiple κ, where κ can be 3, 4 or 5, and T is the cutoff characteristic temperature.

[0076] In an exemplary embodiment, the step of fitting and calculating the electron distribution information at different times to determine the final characteristic temperature includes: acquiring electron distribution information corresponding to multiple characteristic temperatures between the initial characteristic temperature and the cutoff characteristic temperature; fitting the acquired electron distribution information data point set using Boltzmann, Maxwell, and relativistic Maxwell distribution fitting algorithms respectively; calculating the goodness-of-fit coefficient value for each fitted electron energy spectrum data curve; selecting the curve formed by the fitting algorithm corresponding to the minimum goodness-of-fit coefficient value as the optimal fitted electron energy spectrum data curve; determining the characteristic temperature at different times in the optimal fitted electron energy spectrum data curve, determining the characteristic temperature with the largest absolute value, and using the characteristic temperature with the largest absolute value as the final characteristic temperature. In this embodiment, the above process may include the following steps:

[0077] Step 1. For a certain moment and a characteristic temperature T, fit the hyperthermal electron spectrum using the Boltzmann, Maxwell and relativistic Maxwell distributions respectively;

[0078] N(E) fit 1 = Aexp -E / T T---Boltzmann fitting formula

[0079] N(E) fit 2 = BE 1 / 2 exp -E / T T 3 / 2 ---Maxwell's Fitting Formula

[0080] N(E) fit 3 = CE 2 exp -E / T T 3 ----Maxwell's Relativity Formula

[0081] In the three fitting formulas above, N(E) fit 1. N(E) fit 2. N(E) fit 3 represent the fitted hyperthermal electron energy spectrum distributions, in MeV. -1 ;

[0082] E is the kinetic energy of the superheated electron, MeV;

[0083] A, B, and C are constants, determined based on the laser energy and the laser-to-superthermal electron energy conversion efficiency.

[0084] T is the characteristic temperature of the superthermal electrons, in MeV;

[0085] Step 2. Calculate and save the goodness of fit of the above three formulas according to the following definitions:

[0086]

[0087] In the formula, N(E) pic The electronic energy spectrum data are obtained by the PIC method;

[0088] N(E) fit These are the three fitted distributions obtained through step 1 above.

[0089] Step 3. Compare within the range of the initial characteristic temperature and the cutoff characteristic temperature. The size, take The characteristic temperature corresponding to the minimum case is used as the characteristic temperature at this moment.

[0090] Step 4. Repeat steps 1-3 within the laser pulse wavelength or the laser-target interaction time, compare the characteristic temperatures at different times, select the characteristic temperature with the largest absolute value, and use the electronic information at the time corresponding to that characteristic temperature as the source term information.

[0091] In step S100, a predetermined amount of experimental data and simulated data are combined into a first dataset, and a portion of the data is selected from the first dataset as the training dataset for the neural network model. For example, the first dataset is composed of the calculated simulated data and the experimental data of the ultrashort, ultra-intense laser device, and then classified into four categories: the first category consists of data with a power density of 10... 18 ~10 20 W / cm 2 um 2 Data on the interaction between horizontal and low Z targets, the second category being power densities at 10 18 ~10 20 W / cm 2 um 2 Data on horizontal and high Z-target interactions, the third category being power densities greater than 10. 20 W / cm 2 um 2 Data on horizontal and low-Z targets, the fourth category is power density greater than 10. 20 W / cm 2 um 2The data on the horizontal and high Z-target effects are used to select a predetermined proportion α from each of the above four types of data as the training set, and the remaining 1-α is used as the test set; where α can be 70%, 80%, or 90%.

[0092] In step S110, the selected neural network model is trained using the training dataset selected in step S100. After training, the trained neural network model is evaluated using evaluation parameters. When the evaluation parameters of the neural network model are less than a preset evaluation parameter threshold, the neural network model is determined to be a well-trained neural network model. The evaluation parameters are:

[0093]

[0094] In the formula, λ1, λ2, and λ3 are constants defined in the region (0, 1), set in a ratio of 10:1:1; M is the experimental value measured by the absorption detector, a vector with m values; A is the response probability of the detector system, an m×n matrix, where any element a ij The probability that the energy of a pulsed X-ray with energy i is measured by the j-th detection unit of the detector is represented; N is the true pulsed X-ray energy spectrum distribution, which is a vector with n values; ||M-AN||2 represents the L2 norm of the vector; ndims() represents the dimension of the vector.

[0095] S120. Based on the diagnostic physical parameters collected in real time from the ultra-short and ultra-intense laser device, the predicted energy spectrum is calculated using the diagnostic physical parameters and the trained neural network model; the diagnostic physical parameters include at least one of the following: laser energy, pulse width, incident angle, polarization direction, target thickness and mass number, and pre-plasma density distribution.

[0096] In an exemplary embodiment, after obtaining the pulsed X-ray energy spectrum using a pre-set spectral resolution algorithm based on the predicted energy spectrum in step S130, the spectrally resolved pulsed X-ray energy spectrum is compared with the predicted energy spectrum calculated by the neural network. When the difference between the spectrally resolved pulsed X-ray energy spectrum and the predicted energy spectrum is greater than a preset value, it is then determined whether the number of data groups with a difference greater than the preset value reaches a specific percentage of the total data. When the specific percentage of the total data is reached, the data in that data group is updated to the first dataset. Based on the updated first dataset, the neural network model is retrained. For example: first, the difference between the predicted energy spectrum and the pulsed X-ray energy spectrum obtained by the pre-set spectral resolution algorithm is calculated. When the difference is greater than χ... 2 , χ 2The percentage can be 100%, 150%, or 200%. The laser parameters and X-ray spectrum information used for neural network training are then stored and tagged. Furthermore, if the number of data groups exceeding a preset value does not reach a specific percentage of the total data, the stored data is added to the neural network database to update the neural network model. For example, if the amount of difference data does not reach β (which can be 20%, 25%, or 30%) of the total data, the stored data is added to the neural network database to update the neural network model. The difference parameter χ between the predicted energy spectrum and the pulsed X-ray energy spectrum obtained through a pre-set spectral resolution algorithm is used. 2 The calculation formula is:

[0097]

[0098] In the formula f ANN (E, N) represents the pulsed X-ray energy spectrum distribution predicted by the neural network;

[0099] f unfold( E, N) represent the experimentally measured pulsed X-ray energy spectrum distribution; where the experimentally measured pulsed X-ray energy spectrum is the parameter given by the spectrum decomposition algorithm, which serves as the final experimental measurement result, i.e., the measured pulsed X-ray energy spectrum.

[0100] ||f ANN (E, N)||1 represents f ANN The 1-norm of the (E, N) vector.

[0101] In one exemplary embodiment, the pulsed X-ray energy spectrum obtained by spectrum deconvolution is shown.

[0102] In one exemplary embodiment, when the number of data groups exceeding a preset value does not reach a specific percentage of the total data, the data in that data group is stored.

[0103] This application also provides a spectroscopic interpretation device for pulsed X-ray energy spectroscopy measurement, such as... Figure 2 As shown, the device includes a memory 210 and a processor 220; the memory is used to store a program for interpreting the spectrum of pulsed X-ray energy spectroscopy, and the processor is used to read and execute the program for interpreting the spectrum of pulsed X-ray energy spectroscopy, and execute the method for interpreting the spectrum of pulsed X-ray energy spectroscopy as described in any of the above embodiments.

[0104] This application also provides a spectral interpretation system for pulsed X-ray energy spectroscopy measurements, such as... Figure 3 As shown, the system includes: a data preparation module 300, a neural network training module 310, a spectrum interpretation module 320, and a database and neural network update module 330.

[0105] Data Preparation Module 300: First, experimental data from typical ultrashort, ultra-intense laser devices are categorized into four types based on laser power density and target material properties. Further, the PIC method is used to simulate the electron distribution information of superthermal electrons within the target surface plasma at different times under the interaction of the ultrashort, ultra-intense laser and the target material. Characteristic electron temperatures are obtained through fitting methods, and the electron distribution information corresponding to these characteristic temperatures is used as the input source term for Monte Carlo software, ultimately obtaining the pulsed X-ray energy spectrum inside and outside the target chamber; the obtained pulsed X-ray energy spectrum is used as simulation data. Finally, a predetermined amount of experimental and simulation data is combined into a first dataset, and a portion of the data from this first dataset is selected as the training dataset for the neural network model.

[0106] Neural Network Training Module 310: After creating training and testing sets using experimental data from the ultra-short, ultra-intense laser device experiment obtained by data preparation module 300 and simulated data of pulsed X-ray energy spectra inside and outside the target chamber obtained through simulation algorithms, a neural network model is selected for training. Based on the trained neural network model and the simultaneously received real-time diagnostic experimental parameters of the ultra-short, ultra-intense laser device, the predicted energy spectra of pulsed X-rays at typical locations inside and outside the target chamber are predicted, and the predicted energy spectra are sent to the spectrum interpretation module.

[0107] The spectrum decomposition module 320: On the one hand, it receives the predicted energy spectrum transmitted by the neural network system and uses it as the initial energy spectrum input program. It performs iterative spectrum decomposition based on the iterative algorithm and the initial input energy spectrum. On the other hand, it measures the actual experimental energy spectrum. Finally, it updates the database and the neural network module by inputting the neural network predicted energy spectrum and the actual experimental energy spectrum.

[0108] The database and neural network module 330 are updated on the one hand to receive the neural network predicted energy spectrum and the actual experimental measured energy spectrum transmitted by the spectral decomposition system and compare the two; on the other hand, they perform statistical analysis on cases where the predicted energy spectrum and the measured energy spectrum differ significantly and determine whether the neural network model needs to be updated.

[0109] The following example illustrates the process of applying an updated neural network combining Monte Carlo and PIC methods to the spectral interpretation method for real-time energy spectral measurements of pulsed X-rays from ultrashort, ultra-intense laser devices:

[0110] 400. Data Preparation

[0111] Experimental data from ultrashort, ultra-intense laser devices are categorized into four types based on laser power density and target properties: The first type includes data with power densities between 10 and 10^6 GHz. 18 ~10 20 W / cm 2 um 2 Horizontal laser and low-Z target experiments, the second type is with power density at 1018 ~10 20 W / cm 2 um 2 Horizontal laser and high Z-target experiments, the third type is power density greater than 10 20 W / cm 2 um 2 Horizontal laser and low-Z target experiments, the fourth type is power density greater than 10 20 W / cm 2 um 2 Horizontal laser and high Z-target experiments;

[0112] 410. Calculate simulation data

[0113] For the above four types of experimental data, the distribution information of superthermal electrons in the target surface plasma at different times under the experimental conditions of ultrashort and ultra-intense laser-target interaction in step 400 is calculated based on the PIC method. Among them, the superthermal electron information is obtained by simulating using the PIC method and a phase space file is created. The superthermal electron information specifically includes the spatial position of the superthermal electrons in the target surface plasma, the angle between the superthermal electrons and the positive directions of the X, Y, and Z coordinate axes of the spatial Cartesian rectangular coordinate system and the cosine value of the angle, the energy and momentum of the superthermal electrons, the laser-to-superthermal electron energy conversion efficiency, and weight information, etc.

[0114] 420. Calculate the electronic source term.

[0115] The electron energy spectrum at different times was fitted using Boltzmann, Maxwell, and relativistic Maxwell distributions to obtain the electron characteristic temperature. The electron information corresponding to the time with the maximum characteristic temperature was selected as the electron source term. The process for determining the electron source term is as follows:

[0116] 4201. Based on the power density of ultrashort, ultra-intense lasers, given the empirical characteristic temperature T... wilks :

[0117]

[0118] In the formula, I represents the laser power density, W / cm². 2 ;

[0119] λ is the laser wavelength, in μm;

[0120] 4202. Assign a value to the initial characteristic temperature T, and take T as the value. wilks A certain proportion γ, i.e., t = γT wilks γ can be 5%, 10%, or 15%;

[0121] 4203. For a certain moment and a characteristic temperature T, fit the hyperthermal electron spectrum using the following Boltzmann, Maxwell and relativistic Maxwell distributions;

[0122] N(E) fit =Aexp -E / T T

[0123] N(E) fit =BE 1 / 2 exp -E / T T 3 / 2

[0124] N(E) fit =CE 2 exp -E / T T 3

[0125] In the formula, N(E) fit To fit the ultrathermal electron energy spectrum distribution, MeV -1 ;

[0126] E is the kinetic energy of the superheated electron, MeV;

[0127] A, B, and C are constants, determined based on the laser energy and the laser-to-superthermal electron energy conversion efficiency.

[0128] T is the characteristic temperature of the superthermal electrons, in MeV;

[0129] 4204. Calculate and save the goodness of fit of the three distributions according to the following definitions:

[0130]

[0131] In the formula, N(E) pic The electronic energy spectrum data are obtained by the PIC method;

[0132] N(E) fit The fitted distribution obtained through step 4203;

[0133] 4205. Assign an increment ΔT to the characteristic temperature T, i.e., T = T + ΔT, where the magnitude of the increment can be T. wilks A certain proportion γ, i.e., ΔT = γT wilks γ can be 5%, 10%, or 15%, and the incremental step of assigning an increment ΔT to the characteristic temperature T is repeated until t reaches T. wilks A certain multiple κ, where κ can be 3, 4 or 5, and T is the cutoff characteristic temperature;

[0134] 4206. Within the range of the initial characteristic temperature and the cutoff characteristic temperature T, compare... The size, take The characteristic temperature T corresponding to the minimum case is determined and used as the characteristic temperature at this moment.

[0135] 4207. During the laser pulse wavelength or the laser-target interaction time, repeat steps 4201-4206, compare the characteristic temperatures at different times, select the characteristic temperature T with the largest absolute value, and use the electronic information at the time corresponding to this characteristic temperature as the source term information.

[0136] 430. Musical Interpretation

[0137] Monte Carlo software was used to perform a full modeling of the geometry and materials of the target chamber and target material of the ultrashort, ultra-intense laser device. The phase space file was used as the radiation source term to calculate the pulsed X-ray energy spectrum at typical locations inside and outside the target chamber. The flux (MeV) of the pulsed X-ray spectrum at 350 keV within the target chamber was calculated. -1 Normalization was performed, and the external X-ray flux (MeV) was calculated based on a pulsed X-ray at 600 keV. -1 After normalization, the pulsed X-ray energy spectra corresponding to the four types of experimental conditions for the interaction between ultrashort and ultra-intense lasers and targets were finally obtained.

[0138] 440. Neural Network Training

[0139] 4401. The first dataset, composed of the pulsed X-ray energy spectrum results obtained in step 430 and the experimental data from the ultrashort, ultra-intense laser device, is classified. The classification includes: the first category consists of data with a power density of 10... 18 ~10 20 W / cm 2 um 2 Data on the interaction between horizontal and low Z targets, and the second type is power density at 10 18 ~10 20 W / cm 2 um 2 Horizontal and high Z-target interaction data, the third category is power density greater than 10 20 W / cm 2 um 2 Horizontal and low Z-target interaction data and the fourth type with power density greater than 10 20 W / cm 2 um 2 Data on the interaction between horizontal and high Z-targets. From each of the four data categories mentioned above, a certain proportion α is selected as the training set, and the remaining 1-α is used as the test set; where α can be 70%, 80%, or 90%.

[0140] 4402. The trained neural network is evaluated, and if the evaluation result is less than a set value, the neural network is considered to have met the requirements; where the evaluation parameter "goodness" for the artificial neural network is:

[0141]

[0142] In the formula, λ1, λ2, and λ3 are constants whose domain is (0, 1), and are given in a ratio of 10:1:1;

[0143] M is the experimental value measured by the absorption detector, a vector with m values; A is the response probability of the detector system, an m×n matrix, where any element a... ij The probability that the energy of a pulsed X-ray with energy i is measured by the j-th detection unit of the detector is represented; N is the true pulsed X-ray energy spectrum distribution, which is a vector with n values; ||M-AN||2 represents the L2 norm of the vector; ndims() represents the dimension of the vector.

[0144] 450. Musical interpretation

[0145] Based on the neural network model and the diagnostic experimental parameters in the ultrashort and ultra-intense laser and target experiment, the predicted initial energy spectrum is calculated. The predicted initial energy spectrum is then used as an input parameter to the spectrum decomposition program to obtain the actual experimental measurement results.

[0146] 460. Update the database and neural network

[0147] 4601. The difference between the predicted initial energy and the actual experimentally measured energy spectrum is given by the following formula. When the difference is greater than χ², 2 , χ 2 If the value can be 100%, 150%, or 200%, then the laser parameters and X-ray spectrum information from this experiment will be stored and tagged.

[0148] χ 2 The calculation formula is:

[0149]

[0150] In the formula f ANN (E, N) represents the pulsed X-ray energy spectrum distribution given by the neural network;

[0151] f unfold (E, N) represents the experimentally measured pulsed X-ray energy spectrum distribution;

[0152] ||f ANN (E, N)||1 represents the 1 norm of the vector.

[0153] 4602. If the amount of differential data reaches β, which is the proportion of the total amount of existing experimental data, the stored data is added to the neural network database to update the neural network model; where β can be 20%, 25% or 30%.

[0154] In this embodiment, the updated neural network combining Monte Carlo and PIC methods is applied to the real-time energy spectrum measurement of pulsed X-rays in ultra-short, ultra-intense laser devices for real-time energy spectrum interpretation. This method is suitable for real-time energy spectrum interpretation of pulsed X-rays in ultra-short, ultra-intense laser devices, and compared to conventional absorption spectrometers, it considers the real-time performance of energy spectrum interpretation and the rationality of the energy spectrum input to the interpretation program. The specific implementation effects are as follows:

[0155] I. The PIC (Particle-In-Cell) plasma numerical simulation method can be used to simulate the interaction process between high-power lasers and plasmas. By simulating the motion of charged particles in self-consistent and applied electromagnetic fields, it studies the collective effects of plasmas and ultimately obtains X-ray energy spectra based on experimental conditions. In high-power-density target firing experiments, strong nonlinearity and coupling phenomena exist, which to some extent limit researchers' understanding of the physical mechanisms. The X-ray spectrum obtained by the PIC method significantly reduces the gap with the true spectral distribution, and is more reasonable and accurate than the initial X-ray spectrum guessed by human experience. It solves the problem of dependence on the accuracy of the initial spectrum in traditional iterative methods, further improving the reliability of the final output results.

[0156] Second, under different power densities, the interaction between ultrashort, ultra-intense lasers and plasma corresponds to different physical acceleration mechanisms and laser-thermionic conversion efficiencies. Furthermore, different mass numbers of the target material correspond to different bremsstrahlung ratios, leading to different characteristic temperatures and characteristic X-rays in the pulsed X-ray spectrum. Classifying the experimental data types allows for obtaining pulsed X-rays with a distribution closer to the true distribution, shortening the time required for spectral interpretation, and improving interpretation accuracy.

[0157] Third, since laser power density and target material directly affect the distribution of hyperthermic electrons on the target surface, including their quantity and energy, collecting hyperthermic electron information and further fitting and analyzing it can yield the electron distribution corresponding to the maximum characteristic temperature, thus accurately obtaining the pulsed X-ray distribution. Furthermore, because the electron information corresponds to the maximum characteristic temperature, it can provide a radiation safety assessment reference for personnel around the laser device, satisfying the conservative principle in radiation protection design.

[0158] Fourth, obtaining the pulsed X-ray energy spectrum by coupling the PIC method with the MC Monte Carlo method takes a long time. By using a neural network method to train on a dataset with a certain degree of completeness, with laser parameters as input and X-ray spectrum as output, the time used for solution space search of the spectral solution algorithm is greatly reduced, and the spectral solution speed is improved.

[0159] V. Neural networks are widely used due to their good performance in nonlinear problems, but the reliability and rationality of their prediction results depend heavily on the completeness and accuracy of the training data. Considering the differences between the X-ray spectral interpretation requirements of ultrashort and ultra-intense laser interaction scenarios and other application scenarios, the training database and neural network model in this method are further updated according to certain conditions. That is, experimental data and simulation data are combined to further optimize the neural network and make full use of the generalization ability of the neural network to meet the actual engineering needs.

[0160] Through the above five aspects, this spectral interpretation method can meet the needs of real-time measurement of pulsed X-ray energy spectra in ultrashort and ultra-intense laser devices.

[0161] The following example illustrates the process of applying an updated neural network combining Monte Carlo and PIC methods to the spectral interpretation method for real-time energy spectrum measurement of pulsed X-rays in ultrashort, ultra-intense laser devices, such as... Figure 4 As shown:

[0162] 1. The parameters obtained under laser experimental conditions are classified and the distribution information of superthermal electrons in the target surface plasma at different times is simulated using PIC software to simulate the interaction between ultrashort and ultraintense lasers and target materials.

[0163] 2. The distribution information of plasma within the target surface at different times was obtained using the PIC method. The maximum characteristic electron temperature was obtained through a fitting method, and the electron information corresponding to this characteristic temperature was used to create a phase space file. Using the phase space as the simulation source term, the pulsed X-ray energy spectrum was calculated using the Monte Carlo method.

[0164] 3. The data in the laser experimental parameter database and the pulsed X-ray spectrum data are combined into a dataset; the data in the dataset are classified according to laser power density and target material properties, with a part of the data used as the training set and a part used as the test set;

[0165] 4. Select a neural network for training and verify whether the trained neural network model achieves the target to determine the neural network model;

[0166] 5. To achieve the goal, determine (update) the neural network model;

[0167] 6. Data from the laser device is collected using diagnostic equipment (the collected data includes diagnostic physical parameters from the diagnostic device).

[0168] 7. This involves integrating the data into the neural network computation model;

[0169] 8. The predicted pulsed X-ray energy spectrum is given by using an updated artificial neural network model combined with experimental parameter input;

[0170] 9 is the spectrum decomposition iteration program that uses the predicted pulsed X-ray energy spectrum as the initial energy spectrum input;

[0171] 10 outputs the pulsed X-ray energy spectrum through a spectrum decomposition iteration program;

[0172] 11 displays the pulsed X-ray energy spectrum obtained from the spectrum deconstruction;

[0173] 12 compares the output of the spectrum interpretation program with the predicted pulse X-ray energy spectrum given by the neural network;

[0174] 13 compares the output of the spectrum decomposition program with the pulsed X-ray energy spectrum predicted by the neural network, and makes the next judgment;

[0175] 14 is to store data when the number of data groups with significant differences does not reach a specific percentage of the total data;

[0176] 15 indicates that if the number of data groups with significant differences exceeds a certain percentage of the total data, the data is added to the training and testing datasets, and the neural network model is updated and validated again.

[0177] In this embodiment, steps 1 and 2 constitute the data processing part, steps 3, 4, and 5 constitute the model update part, and steps 6-15 constitute the spectral output part. This embodiment has the following technical effects:

[0178] By establishing a physical model of the interaction between laser and target material, a Monte Carlo transport model, and a neural network model between laser target firing experimental parameters and pulsed X-ray energy spectrum, the accurate distribution of pulsed X-rays inside and outside the target chamber can be obtained in real time. This solves the problem of dependence on the accuracy of the initial spectrum in traditional iterative methods and the inability to perform real-time spectrum analysis of pulsed X-rays.

[0179] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

Claims

1. A deconvolution method of a pulse X-ray energy spectrum measurement applied to a pulse X-ray generated by an ultra-short and ultra-strong laser device, characterized in that, The method comprises: combining a predetermined number of experimental data and simulation data into a first data set, selecting part of the data from the first data set as a training data set of a neural network model; training the neural network model using the training data set; calculating a predicted energy spectrum using a diagnostic physical parameter and the trained neural network model according to a real-time collected diagnostic physical parameter in an ultra-short and ultra-strong laser device, wherein the diagnostic physical parameter comprises: laser energy, pulse width, focusing size, energy concentration, incident angle, laser polarization mode, target material thickness and mass number, and plasma density distribution; obtaining a pulse X-ray energy spectrum through a pre-set deconvolution algorithm according to the predicted energy spectrum.

2. The deconvolution method of a pulse X-ray energy spectrum measurement according to claim 1, characterized in that, The simulation data is obtained through the following steps: determining the electron distribution information of the superhot electrons in the target surface plasma at different times by using a plasma numerical simulation method, and calculating the electron distribution information at different times to determine the final characteristic temperature; inputting the determined electron distribution information at the corresponding time of the final characteristic temperature into a pre-set Monte Carlo calculation model as an electron source term, and calculating the pulse X-ray energy spectrum at the time, wherein the calculated pulse X-ray energy spectrum at the time is taken as the pulse X-ray energy spectrum corresponding to the simulation condition; taking the calculated pulse X-ray energy spectrum as the simulation data.

3. The deconvolution method of a pulse X-ray energy spectrum measurement according to claim 2, characterized in that, Before the step of determining the electron distribution information of the superhot electrons in the target surface plasma at different times by using the plasma numerical simulation method, the method further comprises: determining an empirical characteristic temperature according to the ultra-short and ultra-strong laser power density in the experimental data; multiplying the empirical characteristic temperature by a predetermined proportion parameter to obtain an initial characteristic temperature; adding a pre-set increment to the initial characteristic temperature to obtain a second characteristic temperature; the i-th characteristic temperature is obtained by adding the pre-set increment to the (i-1)-th characteristic temperature, wherein i is an integer greater than or equal to 3; the maximum value of the ith characteristic temperature does not exceed a cutoff characteristic temperature; the cutoff characteristic temperature is the empirical characteristic temperature plus times the cutoff characteristic temperature, wherein, .

4. The deconvolution method for measuring the pulse X-ray energy spectrum according to claim 3, wherein the step of calculating the electron distribution information at different times to determine the final characteristic temperature comprises: obtaining the electron distribution information at different times from the beginning of the laser pulse to the end of the laser pulse, and determining the electron distribution information corresponding to a plurality of characteristic temperatures between the initial characteristic temperature and the cutoff characteristic temperature; fitting the obtained electron distribution information data points using Boltzmann, Maxwell and relativistic Maxwell distribution fitting algorithms respectively; calculating the fitting degree of each fitting electron energy spectrum data curve respectively; selecting the curve formed by the fitting algorithm corresponding to the minimum value in the fitting degree as the optimal fitting electron energy spectrum data curve; determining the corresponding optimal fitting electron energy spectrum data curve at different times and the corresponding characteristic temperature, selecting the characteristic temperature with the largest absolute value as the final characteristic temperature.

5. The deconvolution method of a pulse X-ray energy spectrum measurement according to claim 1, wherein, After the step of training the neural network model using the training data set, the method further comprises: taking the remaining part of the data in the first data set as a test data set of the neural network model; and using evaluation parameters to judge the neural network model; When the evaluation parameter of the neural network model is less than a preset evaluation parameter threshold, the neural network model is determined as a trained neural network model; The evaluation parameter is: wherein, , , is a constant with domain in (0, 1) and is set in the ratio of 10:1:1; M is the measured value of the absorption method detector, which is a vector with values; A is the response probability of the detector system, which is a matrix, any element in the matrix represents the probability of the pulse X-ray energy of being measured by the first detection unit of the detector; N is the real pulse X-ray energy spectrum distribution, which is a vector with values; ||M-AN||2represents the two-norm of the vector; ndims() represents the vector dimension.

6. The deconvolution method of a pulse X-ray energy spectrum measurement according to claim 1, wherein, After obtaining the pulse X-ray energy spectrum through the preset deconvolution algorithm according to the predicted energy spectrum, the method further comprises: Comparing the pulse X-ray energy spectrum obtained by deconvolution with the predicted energy spectrum calculated by the neural network to determine whether the difference between the two is greater than a preset value; When the difference between the pulse X-ray energy spectrum obtained by deconvolution and the predicted energy spectrum is greater than the preset value, it is determined whether the number of data groups greater than the preset value reaches a certain percentage of the total data; When the certain percentage of the total data is reached, the data in the data group is updated to the first data set; According to the updated first data set, the neural network model is retrained.

7. The deconvolution method of a pulse X-ray energy spectrum measurement according to claim 1, wherein, After obtaining the pulse X-ray energy spectrum through the preset deconvolution algorithm according to the predicted energy spectrum, the method further comprises: The pulse X-ray energy spectrum obtained by the preset deconvolution algorithm is displayed.

8. The deconvolution method of a pulse X-ray energy spectrum measurement according to claim 6, wherein, After determining whether the number of data groups greater than the preset value reaches a certain percentage of the total data, the method further comprises: When it is determined that the number of data groups greater than the preset value does not reach a certain percentage of the total data, the data in the data group is stored.

9. A deconvolution device for a pulsed X-ray energy spectrum measurement, characterized by The device comprises a memory and a processor; the memory is used to save the deconvolution program for pulse X-ray energy spectrum measurement, and the processor is used to read and execute the deconvolution program for pulse X-ray energy spectrum measurement, and execute the pulse X-ray energy spectrum measurement deconvolution method of any one of claims 1-8.

Images