Plasma light source multi-mode adaptive control system and method

Abstract

The invention relates to a plasma light source multi-mode adaptive control system and method, and the system comprises a multi-source data collection and fusion layer which is used for obtaining multi-source data related to the plasma state and system operation; the digital twinning and artificial intelligence model layer is used for constructing and updating a virtual model of a plasma dynamic process based on the multi-source data and outputting a control decision through an artificial intelligence model; the intelligent decision and cooperative control layer is used for generating a cooperative optimization control instruction for the LHDP laser heating subsystem and the main power supply subsystem based on the control decision; and the execution mechanism layer is used for executing the collaborative optimization control instruction. The method has the advantages that energy is supplemented continuously through the LHDP technology, the service life of the high-temperature and high-density plasma is remarkably prolonged, the total radiation amount of 2-50 nm is increased, and the power of 2-50 nm is increased by 2-3 times.

Description

Technical Field

[0001] This invention belongs to the field of plasma technology, and particularly relates to a multimodal adaptive control system and method for a plasma source. Background Technology

[0002] Plasma technology, as a highly efficient method for generating radiation in the 2-50nm wavelength range, typically uses gaseous media such as xenon as the radiation source. It generates plasma through the induction of a pulsed high current in a pre-filled gas, resulting in magnetic compression and thus radiating light in the 2-50nm range. However, traditional plasma light sources still face bottlenecks in terms of energy conversion efficiency and output power. 1. Short radiation lifetime: The plasma expands and cools rapidly after compression, resulting in a short radiation time of 2-50nm and limited single-pulse energy; 2. Discontinuous energy injection: Relying solely on electrical pulse energy injection, lacking a continuous heating mechanism, the plasma temperature and density drop rapidly; 3. Limited control dimensions: Existing systems mainly rely on electrical parameter adjustments and lack the ability to intervene in the dynamic state of plasma in multiple dimensions; 4. Lack of adaptive heating mechanism: Although some studies have proposed laser-assisted heating, most of them involve open-loop fixed parameter injection, which cannot dynamically match and continuously reinforce according to the real-time plasma state.

[0003] In recent years, laser continuous heating (LHDP) technology has been proposed to prolong the high-temperature, high-density state of plasma. However, its control still relies on human experience and has failed to form an integrated intelligent closed loop of perception, decision-making, and execution. Therefore, developing an adaptive control system that can perceive the plasma state in real time, intelligently decide on the LHDP laser continuous supply parameters, and coordinate with the main pulse power supply for optimization has become the key to achieving breakthroughs in the performance of 2-50nm light sources. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a multimodal adaptive control system and method for plasma light sources.

[0005] Firstly, a multimodal adaptive control system for a plasma source is provided, comprising: A multi-source data acquisition and fusion layer is used to acquire multi-source data related to plasma state and system operation; The digital twin and artificial intelligence model layer is connected to the multi-source data acquisition and fusion layer. It is used to construct and update a virtual model of the plasma dynamic process based on the multi-source data, and output control decisions through the artificial intelligence model. The intelligent decision-making and collaborative control layer, connected to the digital twin and artificial intelligence model layer, is used to generate collaborative optimization control commands for the LHDP laser heating subsystem and the main power supply subsystem based on the control decisions. The execution layer, connected to the intelligent decision-making and collaborative control layer, is used to execute the collaborative optimization control instructions.

[0006] Preferably, the data integrated by the multi-source data acquisition and fusion layer includes: theoretical simulation data, real-time sensor data, and user-defined target data.

[0007] Preferably, the real-time sensing data includes: electrical parameters, plasma parameters, optical parameters, and environmental parameters.

[0008] Preferably, the digital twin and artificial intelligence model layer includes a deep reinforcement learning model, which is configured as follows: The plasma state characteristics extracted from the multi-source data, the current system control parameters, and the user target are used as state inputs; Optimized values ​​of target parameters for the output laser heating subsystem and the main pulse power supply subsystem; The reward function of the deep reinforcement learning model is constructed based on the degree of conformity between the actual output performance of the light source and the user's objective.

[0009] Preferably, the intelligent decision-making and collaborative control layer employs a model predictive control algorithm, which solves for the optimal control sequence over multiple future working cycles based on the control decisions output by the deep reinforcement learning model.

[0010] Preferably, the laser heating subsystem includes a pulsed laser and a beam directional focusing device. The instructions generated by the intelligent decision-making and collaborative control layer are used to dynamically adjust at least one of the following parameters of the laser heating subsystem: LHDP laser energy adjustment amount, pulse width, pulse timing, and three-dimensional adjustment amount of focusing position.

[0011] Secondly, a multimodal adaptive control method for a plasma source is provided, executed by any of the systems described in the first aspect, comprising: S1. Acquire and fuse multi-source data related to plasma state and system operation; S2. Based on the multi-source data, analyze and make decisions through an artificial intelligence model to generate collaborative control decisions for the LHDP laser heating subsystem and the main power supply subsystem. S3. Based on the aforementioned collaborative control decision, generate and execute collaborative optimization control commands for the laser heating subsystem and the main pulse power supply subsystem to perform dynamic energy matching and replenishment during plasma evolution. S4. Update the artificial intelligence model and control instructions based on the system feedback after execution.

[0012] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the method described in the second aspect.

[0013] Fourthly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements the method as described in the second aspect.

[0014] The beneficial effects of this invention are: 1. This invention uses LHDP technology to continuously replenish energy, significantly extending the lifetime of high-temperature, high-density plasma, increasing the total radiation in the 2-50nm range, and boosting the power in the 2-50nm range by 2-3 times.

[0015] 2. This invention enables dynamic and precise matching, specifically through real-time AI-controlled laser parameters, ensuring that energy is always injected into the spatiotemporal region most in need of heating, maximizing energy utilization efficiency. Furthermore, users only need to set advanced objectives, and the system automatically completes the complex collaborative optimization of the LHDP laser and main power supply.

[0016] 3. This invention compensates for performance drift caused by equipment aging and gas composition fluctuations in real time through an online learning mechanism, thus exhibiting strong adaptability and robustness. Furthermore, all successful "LHDP + main pulse" collaborative strategies of this invention are structured and stored, forming a transferable and optimizable expert system. Attached Figure Description

[0017] Figure 1 A schematic diagram of the architecture of a multimodal adaptive control system for a plasma source provided by the present invention; Figure 2 This is a schematic diagram of the AI ​​training and decision-making process provided by the present invention; Figure 3 This is a schematic diagram of the LHDP technology and plasma dynamic matching provided by the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to embodiments. The description of the embodiments below is only for the purpose of helping to understand the present invention. It should be noted that those skilled in the art can make several modifications to the present invention without departing from the principle of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0019] Example 1: To address the problems of low output power, short radiation lifetime, lack of adaptive laser continuous heating mechanism, and inaccurate plasma state control in existing plasma light sources (2-50nm), Embodiment 1 of this application provides a multi-modal adaptive control system for plasma light sources. Its core lies in constructing a plasma dynamic digital twin that integrates multi-source data, and realizing the collaborative intelligent control of the LHDP laser heating subsystem and the main power supply subsystem through deep reinforcement learning.

[0020] include: The multi-source data acquisition and fusion layer is used to acquire multi-source data related to plasma state and system operation.

[0021] The data integrated by the multi-source data acquisition and fusion layer includes: theoretical simulation data, real-time sensor data, and user-defined target data.

[0022] The theoretical simulation data includes ionization states, radiation spectra, and absorption coefficients of gaseous media such as xenon (Xe) and nitrogen (N2) at different electron temperatures and densities, provided by atomic physics data (FLYCHK, HULLAC). Additionally, it includes plasma compression dynamics, energy dissipation paths, and laser absorption characteristics from radiation magnetohydrodynamic simulation data (CESZAR, HELIOS).

[0023] Real-time sensing data includes: electrical parameters, plasma parameters, optical parameters, and environmental parameters.

[0024] Specifically, real-time sensing data is acquired through a real-time sensing array. This array is used for electrical diagnostics, plasma diagnostics, optical diagnostics, and LHDP laser diagnostics.

[0025] The electrical diagnostics utilizes Rogowski coils, Bdot probes, and high-voltage differential probes to acquire current I(t), dI / dt, and voltage V(t). Plasma diagnostics uses Langmuir probes to obtain electron temperature Te and density ne, and a grating spectrometer to acquire characteristic spectral line intensities such as Xe XI and Xe X, used for plasma state inversion. Optical diagnostics uses photodiodes to monitor integrated power in the 2-50nm band; a high-speed CCD to acquire the spatiotemporal evolution of the laser spot and the centroid trajectory; and a high-speed visible / near-infrared CCD to monitor the macroscopic morphology of the plasma column. LHDP laser diagnostics uses an energy meter to monitor laser pulse energy in real time; and a four-quadrant detector and wavefront sensor to monitor beam direction, focusing position, and spot morphology. In addition, the real-time sensing array includes environmental sensors such as mass flow controllers, vacuum gauges, and optical element temperature sensors.

[0026] Furthermore, the multi-source data acquisition and fusion layer also includes a user target definition module, which supports "high power mode," "high brightness mode," "long-term stable operation mode," and user-defined multi-target weight combinations. For example, 30W output for 2-50nm; 50W / mm^2-sr brightness for 2-50nm; 1% standard deviation for long-term stability; and a user-defined 20W + 1550W / mm^2-sr + 1.5%.

[0027] The digital twin and artificial intelligence model layer is connected to the multi-source data acquisition and fusion layer. It is used to construct and update a virtual model of the plasma dynamic process based on the multi-source data, and output control decisions through the artificial intelligence model.

[0028] The digital twin and artificial intelligence model layer constructs a virtual mapping of plasma dynamic processes and enables intelligent decision-making. This layer includes a deep reinforcement learning model, which is configured as follows: The plasma state characteristics extracted from the multi-source data, the current system control parameters, and the user target are used as state inputs; Optimized values ​​of target parameters for the output laser heating subsystem and the main pulse power supply subsystem; The reward function of the deep reinforcement learning model is constructed based on the degree of conformity between the actual output performance of the light source and the user's objective.

[0029] It should be noted that the deep reinforcement learning model, as the core decision engine of the subsequent intelligent decision-making and collaborative control layer, is responsible for generating the optimized target parameter values ​​for the laser heating subsystem and the main pulse power supply subsystem. The intelligent decision-making and collaborative control layer then transforms these target values ​​into specific collaborative action commands.

[0030] For example, the state inputs of the deep reinforcement learning model include: current control parameters (LHDP laser energy, pulse width, focus position, repetition rate, main power supply voltage, frequency, pre-ionization intensity), real-time plasma characteristics, and the user target vector. The action outputs of the deep reinforcement learning model include: LHDP laser energy adjustment (800 mJ–2 J), three-dimensional focus position adjustment (via piezoelectric ceramic and fast reflector), pulse timing (delay relative to the main pulse), pulse width; main power supply frequency adjustment (10 Hz–3 kHz), and voltage fine-tuning. The reward function of the deep reinforcement learning model is dynamically calculated based on the total output energy from 2–50 nm, instantaneous peak power, brightness, spatial stability, and target compliance.

[0031] Furthermore, the training data for the deep reinforcement learning model integrates atomic physics simulation data of gaseous media, magnetohydrodynamic simulation data, historical experimental data, and online incremental data, and adopts an update strategy that combines offline pre-training with online incremental learning.

[0032] In addition, the digital twin and artificial intelligence model layer also includes a feature engineering unit, which is used to: extract the "inductance drop" time, current rise rate, and net current integral from the current waveform; extract the spot contraction rate, expansion rate, and morphological symmetry from the CCD image sequence; and calculate key parameters of laser absorption: based on plasma density Ne and temperature Te, calculate the position of the laser critical density surface and the inverse bremsstrahlung absorption coefficient in real time.

[0033] Furthermore, the digital twin and artificial intelligence model layer also incorporates the pre-ionization circuit state parameters.

[0034] Furthermore, the digital twin and artificial intelligence model layer can also include a model training engine for model training. The data sources for the model training engine include: fused first-principles simulation data, a library of historical experimental "recipes," and incremental data generated during online execution. The model training process employs a three-stage strategy: offline pre-training (simulation + historical data) → online transfer learning (new equipment) → continuous incremental fine-tuning (running data). The model update mechanism is triggered by accumulating 10^4 impulses or detecting performance drift exceeding a threshold.

[0035] The intelligent decision-making and collaborative control layer, connected to the digital twin and artificial intelligence model layer, is used to generate collaborative optimization control commands for the LHDP laser heating subsystem and the main power supply subsystem based on the control decisions.

[0036] The intelligent decision-making and collaborative control layer employs a model predictive control algorithm, which, based on the control decisions output by the deep reinforcement learning model, solves for the optimal control sequence over multiple future working cycles.

[0037] The laser heating subsystem includes a pulsed laser and a beam-directing focusing device, used to dynamically adjust the laser energy, focusing position, and pulse timing during the plasma expansion and cooling stages to achieve continuous heating. The instructions generated by the intelligent decision-making and collaborative control layer are used to dynamically adjust at least one of the following parameters of the laser heating subsystem: LHDP laser energy adjustment, pulse width, pulse timing, and three-dimensional adjustment of the focusing position.

[0038] The execution layer, connected to the intelligent decision-making and collaborative control layer, is used to execute the collaborative optimization control instructions.

[0039] The system analyzes plasma spectra and spatial images in real time to invert its temperature, density, and location, and then dynamically adjusts the focus of the LHDP laser to ensure that it is always aligned with the optimal absorption and heating region.

[0040] Example 2: Based on Example 1, Example 2 of this application provides a more specific multimodal adaptive control system for a plasma source, including: The multi-source data acquisition and fusion layer is used to acquire multi-source data related to plasma state and system operation.

[0041] The digital twin and artificial intelligence model layer is connected to the multi-source data acquisition and fusion layer. It is used to construct and update a virtual model of the plasma dynamic process based on the multi-source data, and output control decisions through the artificial intelligence model.

[0042] The intelligent decision-making and collaborative control layer, connected to the digital twin and artificial intelligence model layer, is used to generate collaborative optimization control commands for the LHDP laser heating subsystem and the main power supply subsystem based on the control decisions.

[0043] The intelligent decision-making and collaborative control layer specifically includes: A multi-objective optimization decision maker is used to: solve for the optimal control sequence over several future pulse cycles based on the output of an AI model and using the model predictive control (MPC) algorithm, while satisfying safety constraints.

[0044] LHDP - Power Co-controller, including: LHDP laser subsystem controller and main power subsystem controller.

[0045] The LHDP laser subsystem controller is used to control Nd:YAG (1064 nm) or CO2 (10.6 μm) lasers. The single pulse energy is adjustable (800 mJ–2 J), the repetition frequency is synchronized with the main pulse and can be independently fine-tuned (10 Hz–3 kHz), and the beam is dynamically controlled by a piezoelectric ceramic fast mirror and an adaptive optics system to achieve real-time adjustment of the focusing position and spot shape.

[0046] The main power supply subsystem controller is used to control the main pulse power supply, with a single pulse energy of 1–5 J and a repetition frequency of 10 Hz–3 kHz, precisely synchronized with the LHDP laser triggering timing.

[0047] Furthermore, the intelligent decision-making and collaborative control layer may also include: The anomaly detection and self-recovery unit is used to monitor the residual between sensor data and digital twin predictions in real time. When a sudden change in laser absorptivity, rapid plasma position drift, or a sharp drop in 2-50nm signal is detected, a safety protocol is immediately triggered, and an AI model is called to perform root cause analysis and generate a recovery control sequence (such as cleaning pulse or parameter rollback).

[0048] The execution layer, connected to the intelligent decision-making and collaborative control layer, is used to execute the collaborative optimization control instructions.

[0049] The actuator layer includes an LHDP laser subsystem for continuous energy supply to the plasma. Specifically, the actuator layer includes: a high-power LHDP pulsed laser; a high-speed beam orientation and focusing system (including piezoelectric ceramic fast mirrors and optical mirrors); a pulsed power module; a pre-ionization and gas delivery system; and a synchronization triggering and timing control unit.

[0050] It should be noted that the parts in this embodiment that are the same as or similar to those in Embodiment 1 can be referred to each other, and will not be repeated in this application.

[0051] Example 3: Based on Example 2, Example 3 of this application provides a multimodal adaptive control method for a plasma source, including: S1. Acquire and fuse multi-source data related to plasma state and system operation.

[0052] S2. Based on the multi-source data, analyze and make decisions through an artificial intelligence model to generate collaborative control decisions for the LHDP laser heating subsystem and the main power supply subsystem.

[0053] Specifically, the pre-trained AI model is loaded and a data link is established; the user sets the output target of 2-50nm; pre-ionization is initiated to generate initial electron seeds in the gas medium; the main pulse is triggered to collect plasma compression and radiation process data in real time; the AI ​​model predicts the plasma evolution trajectory based on real-time conditions (such as temperature, density, and spatial distribution) and calculates the optimal LHDP laser continuous heating parameters.

[0054] S3. Based on the aforementioned collaborative control decision, generate and execute collaborative optimization control commands for the laser heating subsystem and the main pulse power supply subsystem to perform dynamic energy matching and replenishment during plasma evolution.

[0055] Specifically, by controlling the energy injection of the LHDP laser during the initial or plateau phase of plasma expansion, and adjusting the energy, focusing position, and pulse width, the plasma is maintained in an optimal radiation state (e.g., Te≈20 eV, Ne≈1×10). 19 cm -3 Based on feedback signals such as 2-50nm power and spot shape, the parameters of the LHDP laser and the main power supply are dynamically adjusted to achieve closed-loop control.

[0056] It should be noted that the LHDP laser pulse is injected at the initial stage of expansion after the plasma compression peak. Dynamic matching with the plasma cluster is achieved by adjusting the delay between the laser pulse and the main pulse, the laser energy, and the three-dimensional focusing coordinates.

[0057] S4. Update the artificial intelligence model and control instructions based on the system feedback after execution.

[0058] Specifically, it continuously collects operational data to drive the AI ​​model to learn and update online, so as to adaptively compensate for the system's time-varying characteristics and individual differences; and enters the protection and self-recovery process in case of anomalies.

[0059] The following two more specific examples illustrate the method provided in this application.

[0060] In one example, this application will describe EUV power multiplication based on xenon media.

[0061] In this example, the initial state is as follows: operating in normal mode, using xenon (Xe) as the working gas, with an average power of 15 W from 2-50 nm. Then, the LHDP system is activated: the AI ​​model, based on real-time spectral (Xe and N2 line intensities) and current characteristics, determines that the plasma enters a rapid cooling phase approximately 30 ns after the compression peak. Intelligent decision-making and execution are then performed: the decision-maker controls the LHDP laser (1064 nm) to inject the first heating pulse (energy 1.2 J) 105 ns after the main pulse, with the focusing position dynamically biased based on the plasma mass center retrieved from the CCD. Simultaneously, the main pulse repetition frequency is increased from 1.5 kHz to 2.4 kHz to match the heating rhythm. The above method achieves the following results: after adjustment, the 2-50 nm radiation duration is extended from ~40 ns to ~180 ns, the single pulse energy is increased by 2.8 times, and the average power stabilizes at 42 W. The system automatically records this "high-frequency main pulse + delayed double-peak LHDP heating" strategy to the knowledge base for rapid startup under similar gas conditions and power targets.

[0062] In another example, this application will describe the achievement of a target plasma state by dynamically adjusting the laser incident angle.

[0063] In this example, the initial state and target settings are as follows: the plasma source uses pure xenon (Xe, N2) as the working medium and operates under standard parameters: main pulse voltage 400 V, frequency 1.8 kHz, and static pressure 90 mTorr. At this time, the average power output of the system from 2 to 50 nm is 22 W. The user sets the target plasma core parameters through the interactive interface as follows: electron temperature Te = 20 eV, electron density ne = 1 × 10⁻⁶ eV. 19 cm -3 The system was designed to maximize brightness in the 2-50 nm band while maintaining this state. Following this, state monitoring and feature extraction were performed: the real-time data acquisition layer operated synchronously after system startup. The Langmuir probe array measured the current electron temperature in the plasma core region to be approximately 17.5 eV, and the electron density to be approximately 7.5 × 10⁻⁶ eV. 18 cm -3The CCD image sequence shows that the light spot exhibits asymmetric expansion after the compression peak, with a periodic jitter of about 15 μm in the horizontal direction at the centroid. The intensity ratio of Xe (λ=9.4 nm) and XeXI (λ=11.2 nm) spectral lines acquired by the spectrometer further confirms that the temperature is lower than the target value. The feature engineering unit calculates in real time that the current laser absorption rate is estimated to be 65%, and there is a spatial offset of about 200 μm between the laser focus position and the plasma density peak region. Then, AI intelligent decision-making and beam angle adjustment are performed: After receiving the above multi-dimensional features, the AI ​​model performs millisecond-level deduction in combination with the target state. The model determines that in order to achieve the target state, the operation of adjusting the laser incident angle must be performed simultaneously: through the piezoelectric ceramic fast mirror system, the focus of the LHDP laser (10.6 μm) is adjusted in three-dimensional space: horizontal offset +0.25 mm, vertical offset -0.1 mm, and axial (optical path direction) fine adjustment +0.05 mm, so that the laser focus accurately tracks and falls into the density that is about to reach 1×10 19 cm -3 The plasma region was then targeted. Specifically, the laser energy and pulse width were matched, increasing the laser pulse energy from 1.0 J to 1.4 J and adjusting the pulse width from 10 ns to 30 ns to increase energy deposition depth and heating time. Furthermore, the main pulse parameters were coordinated: the main pulse frequency was increased from 1.8 kHz to 2.1 kHz to match the new heating rhythm, and the charging voltage was fine-tuned to 395 V to stabilize the compression dynamics. This decision was based on the physical mapping of the digital twin: precisely injecting laser energy near the density critical point allows for the most efficient heating of electrons through the inverse bremsstrahlung absorption mechanism; simultaneously, a slightly higher main pulse frequency maintains the plasma in a relatively hot state for the next compression, forming a virtuous cycle. The electron temperature and electron density of the electrodeless Z-Pinch plasma were precisely controlled to target values ​​(20 eV, 1 × 10⁻⁶). 19 cm -3 This allows for the maximization of radiation efficiency in the 2-50nm range.

[0064] It should be noted that the method provided in this embodiment is the corresponding method of the system provided in Embodiment 2. Therefore, the parts in this embodiment that are the same as or similar to those in Embodiment 2 can be referred to each other, and will not be described again in this application. Example 4: Based on Example 3, Example 4 of this application provides another multimodal adaptive control method for a plasma source, including: Step S1: System Synchronization and Mode Initialization: Establish a communication link, perform time synchronization calibration, and set the collaborative working mode; Step S2: Subsystem Independent Parameter Preset: Send instructions to the laser heating subsystem and the main pulse power supply subsystem respectively to set their independent initial operating parameters; prioritize the laser because the adjustment response is fast enough, fine-tune the laser output power in increments of 1W, and then adjust the time delay in increments of 10ns.

[0065] Step S3: Load cooperative safety boundary: Send a command to the safety interlock module to set the maximum allowable values ​​and derating factor for power, delay time, output voltage, and frequency, and activate cooperative protection; Step S4: Perform timing coordination fine-tuning: Send timing fine-tuning command to adjust the delay of the laser pulse relative to the main pulse, and verify whether the pulse width overlap reaches the preset threshold; Step S5: Perform energy coupling optimization: Based on the completion of timing fine-tuning, send an energy coupling optimization command. The collaborative tuning engine iteratively adjusts the parameters in a closed-loop manner to bring the actual coupled energy to the target energy range. Step S6: Steady-state operation and monitoring: Enter the steady-state operation loop, continuously monitor key parameters, and trigger safety interlocks or adjust the process in case of anomalies.

[0066] It should be noted that the parts in this embodiment that are the same as or similar to those in Embodiment 3 can be referred to each other, and will not be repeated in this application.

Claims

1. A multimodal adaptive control system for a plasma light source, characterized in that, include: A multi-source data acquisition and fusion layer is used to acquire multi-source data related to plasma state and system operation; The digital twin and artificial intelligence model layer is connected to the multi-source data acquisition and fusion layer. It is used to construct and update a virtual model of the plasma dynamic process based on the multi-source data, and output control decisions through the artificial intelligence model. The intelligent decision-making and collaborative control layer, connected to the digital twin and artificial intelligence model layer, is used to generate collaborative optimization control commands for the LHDP laser heating subsystem and the main power supply subsystem based on the control decisions. The execution layer, connected to the intelligent decision-making and collaborative control layer, is used to execute the collaborative optimization control instructions.

2. The plasma source multimodal adaptive control system according to claim 1, characterized in that, The data integrated by the multi-source data acquisition and fusion layer includes: theoretical simulation data, real-time sensor data, and user-defined target data.

3. The plasma source multimodal adaptive control system according to claim 2, characterized in that, The real-time sensing data includes electrical parameters, plasma parameters, optical parameters, and environmental parameters.

4. The plasma light source multimodal adaptive control system according to claim 3, characterized in that, The digital twin and artificial intelligence model layer includes a deep reinforcement learning model, which is configured as follows: The plasma state characteristics extracted from the multi-source data, the current system control parameters, and the user target are used as state inputs; Optimized values ​​of target parameters for the output laser heating subsystem and the main pulse power supply subsystem; The reward function of the deep reinforcement learning model is constructed based on the degree of conformity between the actual output performance of the light source and the user's objective.

5. The plasma source multimodal adaptive control system according to claim 4, characterized in that, The intelligent decision-making and collaborative control layer employs a model predictive control algorithm, which, based on the control decisions output by the deep reinforcement learning model, solves for the optimal control sequence over multiple future working cycles.

6. The plasma source multimodal adaptive control system according to claim 5, characterized in that, The laser heating subsystem includes a pulsed laser and a beam directional focusing device. The instructions generated by the intelligent decision-making and collaborative control layer are used to dynamically adjust at least one of the following parameters of the laser heating subsystem: LHDP laser energy adjustment amount, pulse width, pulse timing, and three-dimensional adjustment amount of focusing position.

7. A multimodal adaptive control method for a plasma source, characterized in that, Performed by the system according to any one of claims 1 to 6, comprising: S1. Acquire and fuse multi-source data related to plasma state and system operation; S2. Based on the multi-source data, analyze and make decisions through an artificial intelligence model to generate collaborative control decisions for the LHDP laser heating subsystem and the main power supply subsystem. S3. Based on the aforementioned collaborative control decision, generate and execute collaborative optimization control commands for the laser heating subsystem and the main pulse power supply subsystem to perform dynamic energy matching and replenishment during plasma evolution. S4. Update the artificial intelligence model and control instructions based on the system feedback after execution.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in claim 7.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in claim 7.

Images