A switch trigger timing self-adaptive correction method for multi-channel LTD pulse power device

Abstract

The application discloses a switch trigger timing self-adaptive correction method for a multi-channel LTD pulse power device, which comprises the following steps: acquiring current state parameters and historical discharge waveform parameters of multi-channel LTD pulse power branches; using a switch breakdown delay prediction model trained by historical discharge data and / or circuit simulation data to predict switch breakdown delays of the branches in the next discharge; combining trigger time setting values, branch propagation delays, transmission path delays and the like to predict times when current waves of the branches reach a load end; solving trigger compensation amounts of the branches, and delivering the compensation amounts to a multi-channel trigger controller. After the discharge is completed, the system collects measured waveforms and updates the prediction model, forming a shot-to-shot closed-loop correction. The application can reduce dispersion degrees of multi-channel switch on times and load end arrival times, improve load end current wave superposition quality, peak current and waveform repeatability.

Description

Technical Field

[0001] This invention belongs to the field of pulse power device control, and particularly relates to an adaptive correction method for the switching trigger timing of multi-channel LTD pulse power devices. Background Technology

[0002] LTD pulsed power devices typically consist of multiple energy storage branches connected in parallel to form a single-stage module, and then multiple modules are connected in series or parallel to form a higher voltage or higher current output. A typical branch includes an energy storage capacitor, a switch, an equivalent branch inductance, an equivalent branch resistance, and a trigger input. Multiple branches need to be synchronously turned on within a very short time to form a high-quality superimposed waveform at the bus structure, transmission lines, and load.

[0003] In practical applications, switch breakdown delay is affected by factors such as charging voltage, trigger voltage, gas pressure, ambient temperature, trigger link delay, electrode erosion, residual charge, insulation condition, and cumulative discharge count. Manufacturing errors and operational variations exist between different branches, and even within the same branch, slow drift and random jitter occur with repeated discharges. This results in inconsistent arrival times of current waveforms at the load end for each branch, thereby reducing the peak current at the load end, widening the rise time, reducing energy coupling efficiency, and adversely affecting the drive symmetry of Z-pinch loads or other high-power loads.

[0004] Traditional methods often employ fixed trigger delay tables, manual empirical tuning, or periodic offline calibration. These methods have the following drawbacks: First, they struggle to adapt to the slow drift caused by changes in switch states due to variations in discharge frequency, gas pressure, and electrode states; second, the large number of parameters in multi-branch, multi-module systems leads to low efficiency in manual tuning; third, the control target often remains at the trigger signal moment, rather than the actual arrival time of the load-side current wave; fourth, they lack a safety constraint mechanism that integrates trigger voltage, gas pressure, insulation margin, and branch current upper limits into the solution process; and fifth, a large amount of historical waveform data is not fully utilized.

[0005] Therefore, an AI adaptive correction method is needed that integrates with specific pulse power hardware, waveform diagnostics, and trigger execution systems to automatically predict delays and generate compensations based on branch states and historical waveforms before each discharge, and to continuously update the model after discharge. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to accurately predict the switching breakdown delay and the time when the branch current wave arrives at the load end in the next discharge in a multi-channel LTD pulse power device based on the current state and historical waveform information of each branch, and to automatically generate the trigger compensation amount of each branch under safety constraints, thereby improving the synchronization, peak current and repeatability of the load end current wave.

[0007] To achieve the above objectives, the present invention provides an adaptive correction method for the switching trigger timing of a multi-channel LTD pulse power device, comprising: Obtain the current state parameters and historical discharge waveform parameters of multiple pulse power branches; Construct state feature vectors for each branch based on the current state parameters and historical discharge waveform parameters; The state feature vector is input into the switch breakdown delay prediction model trained with historical discharge data and / or circuit simulation data to obtain the predicted value of the switch breakdown delay for each branch. Based on the predicted value of the switch breakdown delay, the set value of the trigger time, the branch propagation delay and the transmission path delay of each branch, the arrival time of the current wave to the load end of each branch is predicted. With the goal of minimizing the arrival time dispersion of each branch, and constrained by the trigger timing adjustment range, trigger voltage range, gas pressure range, insulation margin, and upper limit of branch current, the trigger timing compensation of each branch is solved. The triggering time of each branch in the next discharge is corrected according to the triggering timing compensation amount; After the discharge is completed, the measured waveform is acquired, and the switch breakdown delay prediction model and / or arrival time prediction model are updated according to the deviation between the measured waveform and the prediction result.

[0008] Preferably, the current status parameters include at least one of the following: charging voltage, trigger voltage, gas switch pressure, ambient temperature, branch residual voltage, trigger link delay, cumulative discharge count, and maintenance status indicator.

[0009] Preferably, the historical discharge waveform parameters include at least one of the following: branch current waveform, branch voltage waveform, trigger signal waveform, load end current waveform, load end voltage waveform, switch breakdown delay, branch current rising edge, waveform half width at half maximum (FWHM), branch current peak value, and energy integral value.

[0010] Preferably, the switch breakdown delay prediction model includes a fully connected neural network, a temporal neural network, a Transformer model, a gradient boosting tree model, a Gaussian process model, a physically constrained neural network, or a combination of the above models.

[0011] Preferably, the arrival time prediction includes a delay correction term caused by changes in transmission line length, transmission line impedance, branch inductance, branch resistance, bus structure, and load impedance.

[0012] Preferably, the objective function for solving the trigger timing compensation includes an arrival time dispersion term, a branch current imbalance term, a trigger compensation amplitude penalty term, a prediction uncertainty penalty term, and a safety margin penalty term.

[0013] Preferably, the trigger compensation amount includes a trigger timing compensation amount, and further includes at least one of a trigger voltage compensation amount, a gas pressure compensation amount, or a pre-trigger parameter compensation amount.

[0014] Preferably, the deviation between the measured waveform and the predicted result includes at least one of the following: breakdown delay error, load arrival time error, load peak current error, branch peak current imbalance error, and waveform similarity error.

[0015] Preferably, the circuit simulation data includes MNA simulation data, TLCODE simulation data, SPICE simulation data, electromagnetic field simulation data, or a combination of the above simulation data.

[0016] Preferably, the method is implemented in layers at the branch level, module level, and system level, wherein the branch level is used for single switch delay prediction, the module level is used for branch synchronization within a module, and the system level is used for load-side waveform consistency control.

[0017] Preferably, the method is applicable to LTD devices, Marx devices, pulse forming line devices, high-current electron beam sources, Z-pinch drive devices, high-power X-ray source drive devices, and other multi-channel high-power pulse generating devices.

[0018] Another aspect of the present invention provides a trigger timing adaptive correction system for implementing the aforementioned method, comprising a data acquisition module, a waveform feature extraction module, a switch breakdown delay prediction module, a load arrival time prediction module, a safety constraint optimization module, a trigger control module, and a model update module.

[0019] The beneficial effects of this invention include: (1) It can predict the switch breakdown delay based on the current state of the branch and historical waveforms, thereby reducing the impact of random jitter and slow drift; (2) The control target is shifted from synchronization at the trigger time to synchronization at the arrival time of the load current wave, which directly improves the waveform quality at the load end; (3) Optimize safety constraints to avoid overcompensation and reduce the risks of trigger failure, current overrun and insulation failure; (4) The model can be updated using the measured waveform after discharge, thereby improving the adaptability under repeated discharge conditions; (5) It can be extended to branch level, module level and whole machine level control, and is suitable for complex multi-module pulse power devices.

[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 components in the exemplary embodiments of the invention.

[0022] Figure 1 shows a schematic diagram of the principle of the multi-channel LTD pulse power device and AI trigger control system according to an embodiment of the present invention; Figure 2 shows the overall workflow of the adaptive correction method for switching trigger timing of a multi-channel LTD pulse power device according to an embodiment of the present invention; Figure 3 illustrates the data acquisition, features, model, and execution architecture of an adaptive correction method for switching trigger timing of a multi-channel LTD pulse power device according to an embodiment of the present invention. Figure 4 shows the internal structure of the AI ​​prediction and optimization model for the adaptive correction method of switching trigger timing of a multi-channel LTD pulse power device according to an embodiment of the present invention; Figure 5 shows a schematic diagram of the arrival time dispersion definition of the switching trigger timing adaptive correction method for a multi-channel LTD pulse power device according to an embodiment of the present invention; Figure 6 shows a conceptual diagram of the current wave synchronization effect before and after AI correction in an adaptive correction method for switching trigger timing of a multi-channel LTD pulse power device according to an embodiment of the present invention; Figure 7 shows a schematic diagram of the arrival time deviation improvement of the switching trigger timing adaptive correction method for a multi-channel LTD pulse power device according to Embodiment 1 of the present invention; Figure 8 illustrates a hierarchical control architecture diagram at the branch level, module level, and system level for an adaptive correction method for switching trigger timing of a multi-channel LTD pulse power device according to an embodiment of the present invention. Figure 9 shows a schematic diagram of the improved stability of repetitive discharge peak current according to the adaptive correction method for switching trigger timing of a multi-channel LTD pulse power device according to Embodiment 2 of the present invention; Figure 10 shows a joint compensation flowchart of the adaptive correction method for switching trigger timing of a multi-channel LTD pulse power device according to Embodiment 2 of the present invention; Figure 11 illustrates the safe feasible region and optimal compensation amount of the adaptive correction method for switching trigger timing of a multi-channel LTD pulse power device according to an 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] This invention employs a closed-loop control architecture consisting of state perception, delay prediction, arrival time prediction, safety constraint optimization, trigger execution, waveform feedback, and model update. Its key lies not in the abstract use of machine learning models, but in embedding AI models into the specific engineering process of trigger control for multi-branch LTD devices, enabling the model output to become executable trigger timing compensation, trigger voltage compensation, or gas pressure compensation.

[0025] Figure 1 This is a schematic diagram of the multi-channel LTD pulse power device and AI trigger control system to which this invention applies. Multiple branches are connected in parallel to output to the transmission path and the load. Branch sensors and load sensors collect waveforms and feed them back to the controller. The AI ​​trigger controller outputs trigger correction values ​​for each branch based on a prediction model and an optimization model. The technical details of this invention are explained in detail below.

[0026] For ease of description, Table 1 below lists the meanings of the main parameters appearing in this specification. Unless otherwise stated, all parameters appearing in this specification are defined in Table 1.

[0027] Table 1 Parameter Definition Table

[0028] (I) Overall Process

[0029] The overall flow of the adaptive correction method for switching trigger timing of a multi-channel LTD pulse power device of the present invention is as follows: Figure 2 As shown. Before discharge, the system collects the voltage, air pressure, temperature, residual voltage and discharge count of each branch, and reads the historical waveform database; after feature extraction, it inputs the data into the AI ​​prediction model; the prediction model outputs the next breakdown delay, load arrival time and uncertainty; the optimizer solves for the trigger compensation amount under safety constraints; the trigger controller executes the compensation; after discharge, the system rewrites the measured waveform into the database and updates the model.

[0030] (II) System Architecture

[0031] The system architecture of this invention is as follows: Figure 3 As shown. The data acquisition layer is responsible for acquiring high-speed waveforms and state variables; the feature and database layer is responsible for sample organization, feature extraction, and abnormal sample labeling; the model service layer is responsible for breakdown delay prediction, arrival time prediction, and uncertainty assessment; the optimization control layer is responsible for compensation solutions under safety constraints; the execution layer is responsible for distributing the compensation amount to the multi-channel delay generator or trigger controller; and the device layer completes the discharge and feeds back the measured waveform to the data acquisition layer.

[0032] (III) Input Variables and Feature Construction

[0033] No. i Road branch road in the k The state feature vector before discharge can be represented as:

[0034] in, V res,i Residual voltage I , V and L These represent feature extraction operators for branch current, branch voltage, and load current waveforms, respectively.

[0035] Table 2 Typical input variables of the triggering timing adaptive correction model

[0036] For branch current waveforms, the arrival time can be defined using the threshold method: (2) The relative delay of the waveform can also be defined using the cross-correlation method: (3) (iv) AI Prediction Model Structure AI prediction model structure as follows Figure 4 As shown. The model input consists of branch states and waveform characteristics. After passing through the feature encoder, it outputs the breakdown delay prediction value on one hand and the prediction uncertainty on the other. The breakdown delay prediction result is combined with the propagation delay model to form the load-end arrival time prediction, which is finally input into the safety constraint optimizer.

[0037] No. i The prediction of the switch breakdown delay for a branch circuit can be written as: (4) If the model also outputs uncertainty, then: (5) In a preferred embodiment, a hybrid model combining the physical baseline and AI residuals is employed: (6) Where Δ t phys,i Δ is the baseline prediction value given by voltage, pressure, trigger margin, or empirical breakdown law. t AI,i This is the residual correction term learned by the model based on historical waveforms and branch states.

[0038] The model training objective function can be written as: (7) (v) Load arrival time prediction Switch breakdown synchronization does not necessarily mean load-side synchronization. i The time it takes for the current wave in the branch circuit to reach the load terminal can be expressed as: (8) in To delay the spread, This is the equivalent time delay correction term caused by branch impedance differences, bus structure, and load conditions. If the equivalent propagation length from the branch to the load is... l i The speed of transmission is v i Then it is acceptable: (9) in, δt TL,i This reflects the impact of termination, impedance mismatch, and reflection on arrival time extraction.

[0039] Define average arrival time: (10) The time dispersion of arrival is: (11) The load arrival time dispersion is defined as follows: Figure 5 As shown in the schematic diagram.

[0040] (vi) Theoretical basis for synchronicity optimization

[0041] Assume that the response of each branch at the load end can be approximated by the same reference waveform. s ( t Time shift: (12) The total load current is: (13) like A i ≈ A and τ i = ¯ τ + δ i And satisfy P i δi = 0, then Taylor expands to: (14) This equation shows that after first-order offset cancellation, the waveform peak and rising edge degradation are mainly affected by... Control. Therefore, reducing the arrival time dispersion at the load end can directly improve the quality of coherent superposition of current waves.

[0042] Figure 6 A conceptual diagram illustrating the synchronization effect before and after the correction is provided. This diagram, created using a professional drawing library, is used to express the core mechanism of this invention: through AI prediction and triggering compensation, the branch current wave at the load end changes from dispersed arrival to concentrated arrival.

[0043] (vii) Safety Constraint Optimization Model

[0044] Let the predicted arrival time after triggering compensation be: (15) The compensated average arrival time is: (16) The optimization objective of this invention can be written as: (17) in, J I This is a penalty term for branch current imbalance. J σ For the uncertainty penalty term of prediction, J safe This is a penalty item for safety margin.

[0045] Typical constraints include:

[0046] In implementations that only perform timing compensation, the following can be adopted: clip (twenty three) In the joint compensation implementation method δt i , δV i and δp i The solution can be obtained through quadratic programming, model predictive control, Bayesian optimization, or reinforcement learning strategies.

[0047] (viii) Model update and outlier handling

[0048] After discharge, the system extracts the measured breakdown delay and load arrival time. The prediction error is defined as: (twenty four) Online updates can be written as: (25) Where 0 < ρ ≤ 1 is the historical sample attenuation factor. Abnormal samples that fail to trigger, lack diagnosis, or have severe waveform distortion can be written into the abnormal sample library, and a decision on whether to include them in the training set will be made after manual review.

[0049] (ix) Implementation of Layered Control

[0050] For multi-module devices, the following can be used: Figure 8 The diagram illustrates hierarchical control at the branch level, module level, and system level. The branch-level model predicts the breakdown delay of a single switch; the module-level model coordinates the synchronization of branches within a module; and the system-level model, aiming at load-side waveform consistency, collaboratively corrects the trigger timing of multiple modules.

[0051] Example 1: Timing Adaptive Correction of a 40-Branch Single-Stage LTD Module

[0052] In this embodiment, the LTD module includes 40 parallel branches, each configured with an independent trigger channel and a branch current measurement channel. The system adopts the most recent M The branch breakdown delay prediction model is trained using discharge data, and the load arrival time is extracted using a combination of threshold and cross-correlation methods. This embodiment only corrects the triggering time and does not link the triggering voltage or gas pressure.

[0053] The specific steps are as follows: S1: Before discharging, read 40 channels of charging voltage, trigger voltage, air pressure, temperature, residual voltage, and cumulative number of discharges; S2: Extract the most recent data from the historical database M The branch current, branch voltage, and load waveform are generated; S3: Extract breakdown delay, branch peak current, waveform rise time, cross-correlation delay, and load arrival time; S4: Calculate each branch and ; S5: In The compensation amount is calculated under the constraints. S6: Send the compensation amount to the multi-channel delay generator and perform discharge; S7: After discharge, the actual waveform is collected, the synchronous improvement index is calculated, and the model is updated.

[0054] Figure 9The results are illustrative. It can be seen that before correction, the arrival time deviation of some branches was relatively large, while after correction, the deviation converged significantly. The values ​​in this figure are used to illustrate the effect of the method and do not limit the specific scope of application of this invention.

[0055] Example 2: Joint Compensation Correction under Multi-Module Repetitive Discharge Conditions

[0056] In this embodiment, the device consists of multiple LTD modules. Unlike Embodiment 1, this embodiment employs a combined correction strategy of trigger timing compensation, trigger voltage compensation, and gas pressure compensation to adapt to the slow drift of branch states under repeated discharge conditions. The system not only predicts the breakdown delay but also outputs the uncertainty. When the average predicted delay of a certain branch is too large and the uncertainty is high, the trigger voltage or gas pressure can be appropriately adjusted within a safe range, rather than relying solely on timing advancement.

[0057] The joint compensation process is as follows Figure 10 As shown.

[0058] The joint compensation objective function can be written as: (26) Its constraints include the upper limit of trigger voltage, the upper and lower limits of gas pressure, the maximum timing adjustment range, the upper limit of branch peak current, and the lower limit of insulation margin.

[0059] Figure 11 The illustration shows that after adopting AI adaptive joint correction in the repetitive discharge sequence, the normalized load peak current is higher and the fluctuation is smaller, indicating that the present invention can adapt to the switching state drift caused by repetitive discharge.

[0060] Example 3: Deployment method combining simulation pre-training and experimental fine-tuning

[0061] When the experimental sample size is small or the new device has not yet accumulated enough discharge data, a combination of simulation pre-training and experimental fine-tuning can be used. First, a dataset covering different branch inductances, resistances, switching delays, transmission line parameters, and load states is generated using MNA, TLCODE, or SPICE models; then, an initial prediction model is trained; after the device is put into operation, it is fine-tuned using real experimental data.

[0062] The advantages of this embodiment are: first, it enables the formation of a basic model before experimentation, reducing the number of initial manual parameter adjustments; second, the simulation data can cover boundary conditions that are not suitable for frequent testing in real devices; and third, fine-tuning of experimental data can correct deviations between simulation and real devices. This embodiment is particularly suitable for multi-level LTD devices, gradually expanding from single-module debugging to full-system operation.

[0063] In summary, the differences between this invention and conventional trigger compensation methods are summarized in the following table: Table 3. Differences between the present invention and conventional trigger compensation methods

[0064] Without departing from the concept of the invention, the invention can be made the following substitutions or extensions: (1) The switch type may be a gas switch, vacuum switch, magnetic switch, photoconductive switch, semiconductor switch or a combination thereof; (2) The prediction model can be trained with experimental data or pre-trained with simulation data and then fine-tuned with experimental data; (3) Synchronization indicators may include threshold arrival time, peak time, waveform centroid, cross-correlation peak time, or a combination thereof; (4) The compensation object may be only the trigger timing, or it may include the trigger voltage, gas pressure, pre-ionization parameters and pre-trigger pulse parameters; (5) This method can be used for Z-pinch drive, inertial confinement fusion related pulse power experimental device, high power X-ray source, high current electron beam source, Marx generator, pulse forming line system and other multi-channel high power pulse generation system, as well as Marx generator, pulse forming line, multi-channel laser triggered high voltage switch array, high current electron beam source and multi-channel X-ray source drive system.

[0065] 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 adaptive correction of switching trigger timing for multi-channel LTD pulse power devices, characterized in that, include: Obtain the current state parameters and historical discharge waveform parameters of multiple pulse power branches; Construct state feature vectors for each branch based on the current state parameters and historical discharge waveform parameters; The state feature vector is input into the switch breakdown delay prediction model trained with historical discharge data and / or circuit simulation data to obtain the predicted value of the switch breakdown delay for each branch. Based on the predicted breakdown delay of each branch, the set trigger time, the branch propagation delay, and the transmission path delay, the arrival time of the current wave in each branch to the load end is predicted. With the goal of minimizing the arrival time dispersion of each branch, and constrained by the trigger timing adjustment range, trigger voltage range, gas pressure range, insulation margin, and upper limit of branch current, the trigger timing compensation of each branch is solved. The triggering time of each branch in the next discharge is corrected according to the triggering timing compensation amount; After the discharge is completed, the measured waveform is acquired, and the switch breakdown delay prediction model and / or arrival time prediction model are updated according to the deviation between the measured waveform and the prediction result.

2. The method according to claim 1, characterized in that, The current status parameters include at least one of the following: charging voltage, trigger voltage, gas switch pressure, ambient temperature, branch residual voltage, trigger link delay, cumulative discharge count, and maintenance status indicator.

3. The method according to claim 1, characterized in that, The historical discharge waveform parameters include at least one of the following: branch current waveform, branch voltage waveform, trigger signal waveform, load end current waveform, load end voltage waveform, switch breakdown delay, branch current rising edge, waveform half width at half maximum (WHM), branch current peak value, and energy integral value.

4. The method according to claim 1, characterized in that, The switch breakdown delay prediction model includes a fully connected neural network, a temporal neural network, a Transformer model, a gradient boosting tree model, a Gaussian process model, a physically constrained neural network, or a combination of the above models.

5. The method according to claim 1, characterized in that, The arrival time prediction includes a delay correction term caused by changes in transmission line length, transmission line impedance, branch inductance, branch resistance, bus structure, and load impedance.

6. The method according to claim 1, characterized in that, The objective function for solving the timing compensation quantity includes the arrival time dispersion term, the branch current imbalance term, the trigger compensation magnitude penalty term, the prediction uncertainty penalty term, and the safety margin penalty term.

7. The method according to claim 1, characterized in that, The trigger compensation amount includes a trigger timing compensation amount, and further includes at least one of a trigger voltage compensation amount, a gas pressure compensation amount, or a pre-trigger parameter compensation amount.

8. The method according to claim 1, characterized in that, The deviation between the measured waveform and the predicted result includes at least one of the following: breakdown delay error, load arrival time error, load peak current error, branch peak current imbalance error, and waveform similarity error.

9. The method according to claim 1, characterized in that, The circuit simulation data includes MNA simulation data, TLCODE simulation data, SPICE simulation data, electromagnetic field simulation data, or a combination of the above simulation data.

10. The method according to claim 1, characterized in that, The method is implemented in layers at the branch level, module level, and whole machine level. The branch level is used for single switch delay prediction, the module level is used for branch synchronization within a module, and the whole machine level is used for load-side waveform consistency control.

11. The method according to claim 1, characterized in that, The method is applicable to LTD devices, Marx devices, pulse forming line devices, high-current electron beam sources, Z-pinch drive devices, high-power X-ray source drive devices, and other multi-channel high-power pulse generators.

12. A trigger timing adaptive correction system for implementing the method of any one of claims 1 to 11, characterized in that, It includes a data acquisition module, a waveform feature extraction module, a switch breakdown delay prediction module, a load arrival time prediction module, a safety constraint optimization module, a trigger control module, and a model update module.

Images