Inertial confinement fusion implosion state inversion and experimental parameter correction method and system based on multi-modal diagnostic data
Patent Information
- Application Number
- CN202611196791.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-07
- Publication Date
- 2026-10-27
AI Technical Summary
但现有技术多聚焦于单项预测、单一性能指标或设计空间搜索，尚缺少一种面向实验流程的、将多模态诊断数据、物理约束状态反演、不确定度评估和下一发实验参数修正统一起来的方法
（1）能够统一融合多源诊断数据，减小单一诊断不完备或标定误差带来的状态反演偏差；
Smart Images
Figure CN122987656A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical fields of inertial confinement fusion experimental diagnosis, radiation fluid simulation, data fusion, artificial intelligence state inversion and experimental parameter closed-loop optimization, and particularly relates to a method and system for inertial confinement fusion implosion state inversion and experimental parameter correction based on multimodal diagnostic data. Background Technology
[0002] Inertial confinement fusion experiments typically deposit energy onto a target pellet, sleeve, or load in an extremely short time using lasers, pulsed power, heavy ion beams, or other external driving methods. This causes the fuel to implode rapidly and form a high-temperature, high-density region during the stagnation phase. Key physical quantities of interest in these experiments include implosion velocity, hotspot temperature, hotspot pressure, fuel surface density, stagnation time, symmetry, mixing degree, neutron yield, and energy coupling efficiency. Many of these physical quantities cannot be directly and completely measured; they can only be indirectly inferred from multi-source information such as X-ray images, spectra, neutron diagnostics, radiation power waveforms, current and voltage waveforms, particle diagnostics, and target manufacturing parameters.
[0003] Current inertial confinement fusion experiments typically rely on radiation fluid simulations, synthetic diagnostics, and expert experience to interpret experimental results. Multiphysics simulations can provide detailed predictions of the implosion process, but these models often contain uncertainties such as material equations of state, opacity, hybrid models, laser-plasma interactions, energy coupling, and three-dimensional perturbations. Public research has indicated that the design of inertial confinement fusion experiments depends on computer simulations, which are approximations of real-world experiments and require calibration through experimental observations. Employing techniques such as transfer learning can link low-fidelity and high-fidelity simulations with experimental data, improving the predictive ability of experimental results.
[0004] On the other hand, diagnostic data itself is characterized by multimodality, asynchrony, high noise, and incompleteness. For example, X-ray images can reflect hot spot morphology and low-order modes, neutron diagnostics can reflect fusion yield, ion temperature, and areal density, spectra can reflect temperature, density, and mixing information, and driving waveforms and current / voltage waveforms can reflect energy deposition history. A single diagnostic signal often cannot uniquely determine the implosion state, and there may be timing errors, calibration errors, and spatial perspective differences between multiple diagnostics. Therefore, relying solely on artificial rules or single-mode models makes it difficult to form a stable, interpretable, and updatable post-experimental analysis workflow.
[0005] In recent years, machine learning has been used for inertial confinement fusion design optimization, simulation-experiment calibration, image diagnostic analysis, and physical state inversion. Related research shows that machine learning surrogate models can handle multiple data modalities, transfer learning can combine large-scale simulation data with limited experimental data, and physically consistent machine learning can recover physically consistent density fields from noisy radiographic images. These studies provide a technological foundation for introducing artificial intelligence into inertial confinement fusion diagnostics and experimental closed-loop systems. However, existing technologies mostly focus on single-item predictions, single performance indicators, or design space searches, lacking a method that unifies multimodal diagnostic data, physical constraint state inversion, uncertainty assessment, and next-launch experimental parameter correction for the experimental process. Summary of the Invention
[0006] The purpose of this invention is to propose a method and system for inverting the implosion state and correcting experimental parameters in inertial confinement fusion based on multimodal diagnostic data. After an inertial confinement fusion experiment, it can automatically fuse multi-source diagnostic data and simulation priors to invert the key implosion state and generate parameter correction suggestions for the next experiment under the constraints of credibility and safety boundaries. This improves the efficiency of experimental analysis, reduces the cost of manual iteration by experts, and enhances the continuous optimization capability of experimental design.
[0007] To achieve the above objectives, this invention provides a method for inverting the state of inertial confinement fusion implosion and correcting experimental parameters based on multimodal diagnostic data, comprising: The experimental configuration parameters and multimodal diagnostic data of an inertial confinement fusion experiment are obtained. The experimental configuration parameters include one or more of the following: driving parameters, target parameters, and diagnostic configuration parameters. The multimodal diagnostic data includes at least two of the following: driving waveforms, X-ray images, visible light images, spectral data, neutron diagnostic data, radiation power waveforms, current and voltage waveforms, particle diagnostic data, and target geometric parameters. The multimodal diagnostic data is time-aligned, spatially registered, intensity-calibrated, noise-estimated, and missing mode-identified to obtain a standardized diagnostic data set. A multimodal feature vector is constructed based on the standardized diagnostic data set and experimental configuration parameters; The multimodal feature vector is input into the implosion state inversion model trained with simulation data and historical experimental data to obtain the inertial confinement fusion implosion state vector and its uncertainty. The implosion state vector is input into the physical consistency verification module to calculate the diagnostic consistency index, conservation constraint residual, and state credibility index. Based on the implosion state vector, the target state vector, and the state confidence index, an experimental parameter correction and optimization model is constructed. Under the constraints of drive device capability, target preparation tolerance, diagnostic window, safety boundary and experimental resources, solve for the parameter correction amount of the next test. The diagnostic data and experimental results after the next test are written into the historical sample database to update the implosion state inversion model and the experimental parameter correction model.
[0008] Preferably, the implosion state vector includes one or more of the following: hot spot temperature, hot spot pressure, hot spot radius, fuel surface density, implosion velocity, stagnation time, radiation temperature, neutron yield, ion temperature, low-order asymmetric mode, mixing degree, shell integrity index, and energy coupling efficiency.
[0009] Preferably, the multimodal feature vector is obtained by fusing the outputs of at least two encoders selected from an image encoder, a waveform encoder, a spectral encoder, a nuclear diagnostic encoder, and a target parameter encoder.
[0010] Preferably, the implosion state inversion model includes one or more of the following: convolutional neural network, temporal neural network, Transformer model, graph neural network, variational inference model, Gaussian process model, Bayesian neural network, or physically constrained neural network.
[0011] Preferably, the implosion state inversion model adopts a multi-fidelity training method, and the training data includes one or more of the following: one-dimensional radiation fluid simulation data, two-dimensional or three-dimensional radiation fluid simulation data, synthetic diagnostic data, historical experimental diagnostic data, and posterior analysis results calibrated by experts.
[0012] Preferably, the physical consistency verification module includes at least one or more of the following constraints: energy conservation constraint, mass conservation constraint, radiation output constraint, nuclear reaction rate constraint, geometric symmetry constraint, and diagnostic mode consistency constraint.
[0013] Preferably, the experimental parameter corrections include one or more of the following: laser power waveform correction, radiation temperature waveform correction, current waveform correction, pulse width correction, target radius correction, shell thickness correction, doping ratio correction, filling pressure correction, pre-pulse parameter correction, and diagnostic time window correction.
[0014] Preferably, the objective function of the experimental parameter correction optimization model includes a state deviation term, a fusion performance term, a parameter change penalty term, an uncertainty penalty term, and a safety boundary penalty term.
[0015] Preferably, if the state credibility index is lower than a preset threshold, supplementary diagnostic suggestions, simulation supplementation suggestions, or manual review indicators are output, instead of directly outputting the experimental parameter correction amount.
[0016] Preferably, the method is applicable to laser-driven inertial confinement fusion, laser-driven inertial confinement fusion, pulsed power-driven inertial confinement fusion, Z-pinch inertial confinement fusion, heavy ion-driven inertial confinement fusion, and magnetized target inertial confinement fusion experiments.
[0017] Another aspect of the present invention provides an inertial confinement fusion implosion state inversion and experimental parameter correction system for implementing the method, including a data access module, a diagnostic preprocessing module, a multimodal feature encoding module, a state inversion module, a physical consistency verification module, a parameter correction and optimization module, an experimental sample library, and a model update module.
[0018] In another aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method.
[0019] Compared with existing methods such as post-experiment manual analysis, single-modal diagnostic inversion, or single surrogate model prediction, this invention has at least the following advantages: (1) It can unify and integrate multi-source diagnostic data, reducing the state inversion bias caused by incomplete single diagnosis or calibration error; (2) It can combine simulation priors and experimental data to improve the stability of implosion state inversion under small sample conditions; (3) It can evaluate the credibility of the inversion results through physical consistency residuals and uncertainty outputs, and avoid using low-credibility results directly for experimental correction; (4) It can convert the results of post-experimental diagnostic analysis into suggestions for the correction of the next drive parameters, target parameters and diagnostic parameters; (5) It can perform shot-to-shot model updates as experimental samples accumulate, thereby improving the prediction and correction capabilities of subsequent experiments.
[0020] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description
[0021] The above and other objects, features and advantages of the present invention will become more apparent from the more detailed description of exemplary embodiments of the invention in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same parts in the exemplary embodiments of the invention.
[0022] Figure 1 A flowchart of an inertial confinement fusion implosion state inversion and experimental parameter correction method based on multimodal diagnostic data according to an embodiment of the present invention is shown. Figure 2A schematic diagram of multimodal diagnostic data input and preprocessing is shown in an inertial confinement fusion implosion state inversion and experimental parameter correction method based on multimodal diagnostic data according to an embodiment of the present invention. Figure 3 A multimodal coding fusion and state inversion model structure diagram of an inertial confinement fusion implosion state inversion and experimental parameter correction method based on multimodal diagnostic data according to an embodiment of the present invention is shown. Figure 4 A simulation experiment multi-fidelity calibration flowchart of an inertial confinement fusion implosion state inversion and experimental parameter correction method based on multimodal diagnostic data according to an embodiment of the present invention is shown. Figure 5 The experimental parameter correction optimization closed-loop diagram of an inertial confinement fusion implosion state inversion and experimental parameter correction method based on multimodal diagnostic data according to an embodiment of the present invention is shown. Figure 6 A reliability gating and manual verification logic diagram of an inertial confinement fusion implosion state inversion and experimental parameter correction method based on multimodal diagnostic data according to an embodiment of the present invention is shown. Figure 7 A flowchart of laser-driven ICF diagnostic inversion is shown, illustrating a method for inverting the state of inertial confinement fusion implosion and correcting experimental parameters based on multimodal diagnostic data, according to an embodiment of the present invention. Figure 8 A flowchart of pulse power-driven ICF parameter correction is shown for an inertial confinement fusion implosion state inversion and experimental parameter correction method based on multimodal diagnostic data according to another embodiment of the present invention. Figure 9 The diagram illustrates the decrease in the objective function before and after parameter correction in the inertial confinement fusion implosion state inversion and experimental parameter correction method based on multimodal diagnostic data according to another embodiment of the present invention. Detailed Implementation
[0023] Preferred embodiments of the invention will now be described in more detail. While preferred embodiments of the invention are described below, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.
[0024] like Figure 1 As shown, this invention proposes a method for inverting the state of inertial confinement fusion implosion and correcting experimental parameters based on multimodal diagnostic data, including: The experimental configuration parameters and multimodal diagnostic data of an inertial confinement fusion experiment are obtained. The experimental configuration parameters include one or more of the following: driving parameters, target parameters, and diagnostic configuration parameters. The multimodal diagnostic data includes at least two of the following: driving waveforms, X-ray images, visible light images, spectral data, neutron diagnostic data, radiation power waveforms, current and voltage waveforms, particle diagnostic data, and target geometric parameters. The multimodal diagnostic data is time-aligned, spatially registered, intensity-calibrated, noise-estimated, and missing mode-identified to obtain a standardized diagnostic data set. Multimodal feature vectors are constructed based on standardized diagnostic data sets and experimental configuration parameters; By inputting the multimodal feature vector into the implosion state inversion model trained with simulation data and historical experimental data, the implosion state vector of inertial confinement fusion and its uncertainty are obtained. The implosion state vector is input into the physical consistency verification module to calculate the diagnostic consistency index, conservation constraint residual, and state credibility index. Based on the implosion state vector, the target state vector, and the state confidence index, an experimental parameter correction and optimization model is constructed. Under the constraints of drive device capability, target preparation tolerance, diagnostic window, safety boundary and experimental resources, solve for the parameter correction amount of the next test. The diagnostic data and experimental results after the next test are written into the historical sample database to update the implosion state inversion model and the experimental parameter correction model.
[0025] This invention employs a closed-loop technical approach encompassing multimodal diagnostic input, data alignment and encoding, physical constraint state inversion, reliability assessment, experimental parameter correction, and sample write-back updates. The key to this method lies not in simply using neural networks to predict a particular experimental result, but in establishing a mapping between multimodal diagnostic data and the physical state variables of inertial constraint fusion, and generating the next experimental parameter correction under physical consistency and uncertainty constraints. The details of this method are described below.
[0026] (I) Composition and Preprocessing of Multimodal Diagnostic Data
[0027] A schematic diagram of multimodal diagnostic data input and preprocessing is shown below. Figure 2 As shown. In this invention, the multimodal diagnostic data set of a single experiment is denoted as... (1) Among them, superscript ( k ) indicates the first k Conduct experiments, D xray This represents X-ray image or backlit image data. Dimg Represents visible light or self-illuminating image data. D spec Representing spectral data, D n This represents diagnostic data for neutrons or particles. D rad This represents data on radiated power or total radiated energy. D drv Represents the driving waveform data. D aux It indicates the target manufacturing parameters, environmental parameters, and other auxiliary data.
[0028] Different diagnostic modes first undergo preprocessing operator P m (·): , (2) in, m For diagnostic modal indexing, α m For strength calibration parameters, τ m R is the time zero point or time delay correction parameter. m For spatial registration parameters, σ m For noise level or confidence level parameters, This refers to the preprocessed, normalized data.
[0029] Table 1. Multimodal diagnostic data and their invertible physical information
[0030] (II) Definition of Implosion State Vector
[0031] This invention represents the implosion state corresponding to a single experiment as a state vector s. (k) : s T (3) in, T hs This refers to the hot spot temperature. P hs For hot spot pressure, R hs Where is the radius of the hot spot. ρR fuel surface density, v imp For the implosion velocity, t stag For a moment of stagnation, Y n Neutron yield, T iThe ion temperature. These are low-order asymmetric mode coefficients. M mix As an indicator of the degree of mixing, η cpl To drive the target coupling efficiency.
[0032] In practice, the state vector can be trimmed according to the type of experiment and available diagnostics; if some state variables cannot be directly labeled, posterior analysis results, synthetic diagnostic inversion results or expert confirmation results can be used as weak labels to participate in training.
[0033] (III) Multimodal coding fusion and state inversion model
[0034] Multimodal coding fusion and state inversion model, such as Figure 3 As shown. The preprocessed modal data are passed through the corresponding encoder. E m (·) Convert to modal features: (4) Among them, h m For the first m Feature vectors of a diagnostic modality θ m These are the parameters of the modal encoder. The multimodal fusion features are: z (5) in, F Θ For fusion networks or probabilistic fusion models, M For the number of modes, u (k) Let q be the experimental control parameter vector. (k) This represents the quality score vector for each modality.
[0035] The implosion state inversion model outputs state estimates and uncertainties: (6) in, G Φ Here, bs is the state inversion network, and bs is the state estimate. This is for estimating the state covariance matrix or uncertainty.
[0036] From a probabilistic perspective, state inversion can be expressed as a posterior distribution estimate: (7) in, p ( | s , u) is the diagnostic likelihood term.p (s | u) represents the state prior obtained from simulation priors or historical experimental statistics.
[0037] (iv) Physical consistency loss function
[0038] To avoid the state inversion model only fitting the apparent features of the data, this invention introduces a physical consistency loss function. The total loss function can be written as: L = L diag + λ phys L phys + λ sim L sim-exp + λ unc L unc (8) Among them, L diag To diagnose reconstruction or state supervision loss, L phys For physical constraint loss, L sim-exp For simulation-experiment calibration loss, L unc For uncertainty calibration loss, λ phys , λ sim , λ unc These are the weighting coefficients.
[0039] The diagnostic consistency loss can be written as: (9) in, H m (·) is the first m A synthetic diagnostic operator for each diagnostic modality. ω m W represents the modal weights. m This is a noise or confidence weighted matrix.
[0040] Physical constraint losses can include energy conservation and product consistency. For example, the energy residual term can be written as: R E = E in η c pl - E kin - E rad - E loss (10) in, E in For input driving energy, η cpl To improve the drive-target coupling efficiency, E kin For shell kinetic energy, E rad As radiant energy, E loss Other energy losses. The corresponding losses are: (11) in, β E , β Y , β A These are the weighting coefficients. Y n,diag The neutron yield obtained for diagnosis, These are the low-order modal coefficients obtained from image diagnosis.
[0041] Table 2 Definition of Main Formula Parameters (five)
[0043] Simulation-Experiment Multifidelity Calibration
[0044] In inertial confinement fusion experiments, high-fidelity 3D simulation is computationally expensive and has a limited number of experimental samples, while low-dimensional or reduced-order simulations can provide a large number of samples but suffer from model bias. Therefore, this invention employs a multi-fidelity calibration mechanism to progressively fuse low-fidelity simulations, high-fidelity simulations, and experimental data, such as... Figure 4 As shown.
[0045] Multifidelity calibration can be expressed as: (12) in, G LF It is a low-fidelity proxy model. R HF This refers to the residual calibration term for high-fidelity simulation compared to low-fidelity simulation. G HF For high-fidelity agent output, R EXP This is a deviation correction term between the experiment and the simulation.
[0046] (vi) Experimental parameter correction and optimization model
[0047] Experimental parameter correction and optimization closed-loop diagram as follows Figure 5 As shown. Let the control parameter of the current experiment be u. (k) The target state is The inversion state is s(k) The parameter correction Δu for the next experiment is obtained through optimization modeling: (13) Wherein, objective function Defined as: (14) Where Q is the state error weight matrix, R is the parameter change weight matrix, and Tr The trace represents the state uncertainty, and Psafe represents the safety boundary penalty term. λ u , λ Σ , λ c These are the weighting coefficients.
[0048] Optimization constraints include: u min ≤ u( k ) + Δu ≤ u max (15) |Δu| ≤ Δu max (16) (17) (18) Among them, u min with u max For the upper and lower limits of the parameter, Δu max Ctarget represents the allowable correction range for a single experiment, Cfacility represents the target preparation and material constraints, and Cfacility represents the device capability and safety constraints.
[0049] (vii) Credibility gating and manual review mechanism
[0050] To avoid directly providing parameter corrections in cases of missing diagnoses, model extrapolation, or high uncertainty, this invention employs a confidence gating system. The state confidence score is defined as follows: Γ = γ 1 S diag + γ 2 S phys + γ 3 S unc + γ 4 S domain (19) in, S diag For diagnostic consistency scoring, Sphys Scoring for physical consistency S unc Scoring the uncertainty S domain Score the coverage of the sample distribution. γ 1 to γ 4 represents the weighting coefficient.
[0051] If Γ ≥ Γ min Then, output parameter correction suggestions; if Γ < Γ min If so, supplementary diagnostic suggestions or a manual review indicator will be output. This gating mechanism is as follows: Figure 6 As shown.
[0052] Example 1: Laser-Driven Inertial Confinement Fusion Experiment
[0053] In one embodiment, the present invention is applied to laser-driven inertial confinement fusion experiments. Experimental input parameters include laser pulse power history, black cavity geometry parameters, target radius, shell thickness, doping ratio, and filling gas pressure. Diagnostic data include X-ray autoluminescence images, X-ray spectra, neutron yield, neutron spectrum, XRD radiation power waveform, and target manufacturing metrology results.
[0054] The steps of this embodiment are as follows: S1: Read the laser power waveform and target design parameters to form the experimental control vector u. (k) ; S2: Perform center localization, viewpoint registration, and low-order mode decomposition on X-ray images to extract... ; S3: Perform energy axis calibration and spectral feature extraction on the spectrum to obtain temperature and mixing correlation features; S4: Estimate the uncertainty of neutron spectrum and yield data to form nuclear diagnostic features; S5: Input the features of each modality into the fusion model and invert the model. T hs , ρR , Y n , t stag and ; S6: If the low-order modes are too large and the output is lower than the target value, optimize the model output laser pulse shaping, diagnostic time window and target parameter correction suggestions.
[0055] The flowchart of laser-indirectly driven ICF diagnostic inversion in this embodiment is as follows: Figure 7 As shown.
[0056] Example 2: Pulsed Power Driven Inertial Confinement Fusion Experiment
[0057] In another embodiment, the present invention is applied to pulsed power-driven inertial confinement fusion experiments. Experimental input parameters include current waveforms, voltage waveforms, transmission line parameters, initial load radius, load material, line mass or shell mass, pre-pulse parameters, and diagnostic configuration. Diagnostic data includes load-end current and voltage waveforms, X-ray images, radiation power waveforms, spectra, and neutron or particle diagnostic data. In this embodiment, the state inversion model primarily outputs implosion velocity, stagnation time, radiation peak time, axial asymmetric modes, mixing degree, and energy coupling efficiency. The parameter correction model can output correction suggestions for current rise time, pre-pulse intensity, initial load structure, material proportions, and diagnostic time windows. The flowchart of pulsed power-driven ICF parameter correction in this embodiment is shown below. Figure 8 As shown.
[0058] Example 3: Objective Function Descent and Model Update
[0059] In continuous experiments, the system can record the objective function value and state error for each experiment. Figure 9 This figure illustrates the results of this embodiment, showing that after applying the method of the present invention, the objective function gradually decreases and tends to stabilize as experimental samples accumulate and parameters are adjusted. This figure is only used to illustrate the closed-loop correction trend and does not limit the specific numerical range of the present invention.
[0060] Another aspect of the present invention provides an inertial confinement fusion implosion state inversion and experimental parameter correction system for implementing the method, including a data access module, a diagnostic preprocessing module, a multimodal feature encoding module, a state inversion module, a physical consistency verification module, a parameter correction and optimization module, an experimental sample library, and a model update module.
[0061] In another aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method.
[0062] In summary, the present invention solves the following technical problems: 1. Different diagnostic modalities have different temporal resolutions, spatial perspectives, and noise levels. How can they be uniformly aligned and fused? 2. The simulation data is large but has model bias, while the experimental data is small but represents the real device. How to combine the two types of data to train the state inversion model? 3. Since critical implosion states cannot be directly used as training labels to obtain a large number of results, how can physical constraints and synthetic diagnostics be introduced to improve the reliability of the inversion? 4. How to quantify the uncertainty and diagnostic consistency of the inversion results to avoid low-confidence results directly driving the next parameter adjustment; 5. How to generate parameter correction suggestions under the constraints of driving capability, target preparation tolerance, safety boundaries, and experimental resources.
[0063] The key features of this invention and their corresponding technical effects are shown in Table 3 below: Table 3 Key technical features and effects of the present invention
[0064] The main experimental parameter correction targets that can be covered by this invention are shown in Table 4.
[0065] Table 4. Main experimental parameter correction targets covered by this invention
[0066] It should also be noted that, without departing from the technical concept of this invention, the specific implementation can be substituted or extended as follows: (1) Multimodal diagnostic data can be selected from any combination of two or more diagnostic signals according to experimental conditions; (2) The state inversion model can be a deterministic neural network, a Bayesian neural network, an ensemble model, a variational autoencoder, or a Gaussian process model; (3) Physical constraints can be given by analytical formulas, synthetic diagnostic procedures, radiation fluid simulation proxy models or expert rules; (4) Parameter correction optimization can be achieved by gradient optimization, Bayesian optimization, model predictive control, genetic algorithm or semi-automatic optimization after manual verification; (5) The method can be used for offline analysis after the experiment, as well as for pre-evaluation and recommendation of experimental parameters during the experimental planning stage; (6) The sample library may include real experiments, simulation experiments, expert annotations and abnormal samples. The model can be updated periodically or online according to new samples.
[0067] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.
Claims
1. A method for inverting the state of implosion and correcting experimental parameters in inertial confinement fusion based on multimodal diagnostic data, characterized in that, include: The experimental configuration parameters and multimodal diagnostic data of an inertial confinement fusion experiment are obtained. The experimental configuration parameters include one or more of the following: driving parameters, target parameters, and diagnostic configuration parameters. The multimodal diagnostic data includes at least two of the following: driving waveforms, X-ray images, visible light images, spectral data, neutron diagnostic data, radiation power waveforms, current and voltage waveforms, particle diagnostic data, and target geometric parameters. The multimodal diagnostic data is time-aligned, spatially registered, intensity-calibrated, noise-estimated, and missing mode-identified to obtain a standardized diagnostic data set. A multimodal feature vector is constructed based on the standardized diagnostic data set and experimental configuration parameters; The multimodal feature vector is input into the implosion state inversion model trained with simulation data and historical experimental data to obtain the inertial confinement fusion implosion state vector and its uncertainty. The implosion state vector is input into the physical consistency verification module to calculate the diagnostic consistency index, conservation constraint residual, and state credibility index. Based on the implosion state vector, the target state vector, and the state confidence index, an experimental parameter correction and optimization model is constructed. Under the constraints of drive device capability, target preparation tolerance, diagnostic window, safety boundary and experimental resources, solve for the parameter correction amount of the next test. The diagnostic data and experimental results after the next test are written into the historical sample database to update the implosion state inversion model and the experimental parameter correction model.
2. The method according to claim 1, characterized in that, The implosion state vector includes one or more of the following: hot spot temperature, hot spot pressure, hot spot radius, fuel surface density, implosion velocity, stagnation time, radiation temperature, neutron yield, ion temperature, low-order asymmetric modes, mixing degree, shell integrity index, and energy coupling efficiency.
3. The method according to claim 1, characterized in that, The multimodal feature vector is obtained by fusing the outputs of at least two encoders from the image encoder, waveform encoder, spectral encoder, nuclear diagnostic encoder, and target parameter encoder.
4. The method according to claim 1, characterized in that, The implosion state inversion model includes one or more of the following: convolutional neural network, temporal neural network, Transformer model, graph neural network, variational inference model, Gaussian process model, Bayesian neural network, or physically constrained neural network.
5. The method according to claim 1, characterized in that, The implosion state inversion model adopts a multi-fidelity training method. The training data includes one or more of the following: one-dimensional radiation fluid simulation data, two-dimensional or three-dimensional radiation fluid simulation data, synthetic diagnostic data, historical experimental diagnostic data, and posterior analysis results calibrated by experts.
6. The method according to claim 1, characterized in that, The physical consistency verification module includes at least one or more of the following constraints: energy conservation constraint, mass conservation constraint, radiation output constraint, nuclear reaction rate constraint, geometric symmetry constraint, and diagnostic mode consistency constraint.
7. The method according to claim 1, characterized in that, The experimental parameter corrections include one or more of the following: laser power waveform correction, radiation temperature waveform correction, current waveform correction, pulse width correction, target radius correction, shell thickness correction, doping ratio correction, filling pressure correction, pre-pulse parameter correction, and diagnostic time window correction.
8. The method according to claim 1, characterized in that, The objective function of the experimental parameter correction and optimization model includes a state deviation term, a fusion performance term, a parameter change penalty term, an uncertainty penalty term, and a safety boundary penalty term.
9. The method according to claim 1, characterized in that, If the state credibility index is lower than the preset threshold, supplementary diagnostic suggestions, simulation supplementation suggestions, or manual review indicators will be output instead of directly outputting the experimental parameter correction amount.
10. The method according to claim 1, characterized in that, The method is applicable to laser-driven inertial confinement fusion, laser-driven inertial confinement fusion, pulsed power-driven inertial confinement fusion, Z-pinch inertial confinement fusion, heavy ion-driven inertial confinement fusion, and magnetized target inertial confinement fusion experiments.
11. A system for inertial confinement fusion implosion state inversion and experimental parameter correction for implementing the method of any one of claims 1 to 10, characterized in that, It includes a data access module, a diagnostic preprocessing module, a multimodal feature encoding module, a state inversion module, a physical consistency verification module, a parameter correction and optimization module, an experimental sample library, and a model update module.
12. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of any one of claims 1 to 10.