Output control method and system for a pulsed power supply

Abstract

The present application relates to a kind of output control method and system of pulse power, comprising the following steps, pulse power output voltage waveform is collected, and output voltage sample sequence is obtained;Wavelet transform decomposition is carried out to it, and voltage waveform characteristic coefficient is extracted;Adaptive threshold segmentation is carried out based on characteristic coefficient, and pulse edge feature point is identified;Waveform compensation parameter sequence is generated by dynamic compensation calculation;PWM control signal is modulated using the sequence, and the optimized PWM waveform sequence is obtained;Finally, the parameter of driving circuit is adjusted based on the optimized PWM sequence, and the stable control of output voltage waveform is realized, the technical problem that traditional pulse power is difficult to deal with the output fluctuation caused by load mutation, temperature drift and device aging factor, leading to waveform distortion is solved, in which fixed parameter PWM control strategy is used.

Description

Technical Field

[0001] This invention relates to the field of pulse power supply technology, and in particular to an output control method and system for a pulse power supply. Background Technology

[0002] With the rapid development of modern industrial manufacturing, materials processing, medical equipment, and new energy fields, pulse power supplies, as critical energy supply devices, directly impact process quality and system efficiency through the accuracy and stability of their output waveforms. Especially in applications such as precision welding, electrochemical processing, and pulsed laser driving, pulse power supplies are required to output voltage waveforms with steep rise times, stable pulse widths, and low distortion rates. However, traditional pulse power supplies often employ fixed-parameter PWM control strategies, which struggle to cope with output fluctuations caused by factors such as sudden load changes, temperature drift, and device aging. This leads to waveform distortion and blurred pulse edges, severely affecting the consistency and reliability of the process. Summary of the Invention

[0003] The main objective of this invention is to provide an output control method for a pulse power supply, which solves the technical problem that traditional pulse power supplies, which mostly use PWM control strategies with fixed parameters, are unable to cope with output fluctuations caused by load changes, temperature drift, and device aging, resulting in waveform distortion.

[0004] To achieve the above objectives, the present invention provides an output control method for a pulse power supply, comprising the following steps:

[0005] The voltage waveform output by the pulse power supply is acquired to obtain the output voltage sampling sequence;

[0006] The output voltage sampling sequence is decomposed by wavelet transform to obtain the voltage waveform characteristic coefficients;

[0007] Based on the voltage waveform characteristic coefficients, adaptive threshold segmentation is performed on the voltage waveform to obtain pulse edge feature points;

[0008] The waveform compensation parameter sequence is obtained by dynamically compensating the pulse edge feature points using a fuzzy neural network.

[0009] The optimized PWM waveform sequence is obtained by modulating the PWM control signal of the pulse power supply based on the waveform compensation parameter sequence.

[0010] Based on the optimized PWM waveform sequence, the driving circuit parameters of the pulse power supply are controlled to obtain a stable output voltage waveform.

[0011] Furthermore, the acquisition of the voltage waveform output by the pulse power supply to obtain the output voltage sampling sequence includes:

[0012] The voltage waveform output by the pulse power supply is sampled by a differential probe in high-voltage isolation to obtain the original voltage sampling signal. The original voltage sampling signal is then subjected to bandpass filtering to obtain a filtered voltage signal sequence.

[0013] Time base compensation is performed on the filtered voltage signal sequence to obtain the output voltage sampling sequence.

[0014] Furthermore, the step of performing wavelet transform decomposition on the output voltage sampling sequence to obtain voltage waveform characteristic coefficients includes:

[0015] Wavelet basis function mapping is performed on the output voltage sampling sequence to obtain the initial scaling coefficient matrix, and the energy density in the initial scaling coefficient matrix is ​​calculated to obtain the energy distribution sequence of each frequency band.

[0016] Based on the energy distribution sequences of each frequency band, wavelet coefficients are reconstructed to obtain a frequency band decomposition coefficient group, and singular value decomposition is performed on the frequency band decomposition coefficient group to obtain a feature space mapping matrix.

[0017] The feature space mapping matrix is ​​subjected to a nonlinear projection transformation to obtain a set of projection coefficients, and feature weight fusion is performed based on the set of projection coefficients to obtain the voltage waveform feature coefficients.

[0018] Furthermore, the step of adaptively thresholding the voltage waveform based on the voltage waveform characteristic coefficients to obtain pulse edge feature points includes:

[0019] Multi-scale gradient analysis is performed on the characteristic coefficients of the voltage waveform to obtain the characteristic gradient matrix, and the extreme points in the characteristic gradient matrix are detected to obtain the gradient jump sequence.

[0020] Envelope extraction is performed on the gradient jump sequence to obtain an envelope curve feature group, and the dynamic threshold of the envelope curve feature group is calculated to obtain a threshold distribution curve;

[0021] Based on the threshold distribution curve, the voltage waveform is segmented using local variance weighting to obtain candidate feature points, and density clustering analysis is performed on the candidate feature points to obtain feature point clusters.

[0022] Edge localization calibration is performed on the clusters of feature points to obtain an edge localization sequence, and temporal features are extracted from the edge localization sequence to obtain pulse edge feature points.

[0023] Furthermore, the step of performing local variance-weighted segmentation on the voltage waveform based on the threshold distribution curve to obtain candidate feature points includes:

[0024] The threshold distribution curve is subjected to a second-order differential transformation to obtain the threshold curvature change, and the voltage waveform is subjected to a sliding window variance calculation to obtain a local variance distribution sequence.

[0025] The local variance distribution sequence is weighted and fused by the threshold curvature change to obtain a weighted variance feature matrix, and peak detection analysis is performed based on the weighted variance feature matrix to obtain the variance peak position sequence.

[0026] The voltage waveform segmentation boundary is determined based on the variance peak position sequence, and the voltage waveform segmentation boundary is obtained by marking and locating feature points on the voltage waveform segmentation boundary to obtain candidate feature points.

[0027] Furthermore, the step of dynamically compensating the pulse edge feature points using a fuzzy neural network to obtain a waveform compensation parameter sequence includes:

[0028] The pulse edge feature points are mapped to fuzzy membership using a fuzzy neural network to obtain a fuzzy quantization sequence of feature points. Neuron activation calculation is then performed based on the fuzzy quantization sequence of feature points to obtain node response features.

[0029] Multi-level propagation operation is performed on the node response features to obtain a compensation parameter estimation sequence, and feedback correction is performed on the compensation parameter estimation sequence to obtain compensation error correction;

[0030] Based on the compensation error correction, parameter optimization and fusion are performed to obtain the compensation parameter update sequence, and the compensation parameter update sequence is then subjected to nonlinear mapping transformation to obtain the waveform compensation parameter sequence.

[0031] Furthermore, the step of modulating the PWM control signal of the pulse power supply based on the waveform compensation parameter sequence to obtain the optimized PWM waveform sequence includes:

[0032] Modulation depth mapping is performed on the waveform compensation parameter sequence to obtain a modulation parameter vector group, and carrier synchronization analysis is performed based on the modulation parameter vector group to obtain carrier modulation features;

[0033] The carrier modulation features are subjected to multiple modulation transformations to obtain a pulse modulation sequence, and the pulse modulation sequence is subjected to edge compensation processing to obtain a compensated modulation waveform.

[0034] Based on the compensated modulation waveform, waveform shaping and reconstruction are performed to obtain a reconstructed waveform sequence. Then, based on the reconstructed waveform sequence, the PWM control signal of the pulse power supply is modulated to obtain an optimized PWM waveform sequence.

[0035] The present invention also provides an output control system for a pulse power supply, comprising:

[0036] The acquisition module is used to acquire the voltage waveform output by the pulse power supply to obtain the output voltage sampling sequence;

[0037] The decomposition module is used to perform wavelet transform decomposition on the output voltage sampling sequence to obtain voltage waveform characteristic coefficients;

[0038] The segmentation module is used to perform adaptive threshold segmentation of the voltage waveform based on the characteristic coefficients of the voltage waveform to obtain pulse edge feature points;

[0039] The calculation module is used to perform dynamic compensation calculations on the pulse edge feature points through a fuzzy neural network to obtain a waveform compensation parameter sequence;

[0040] The modulation module is used to modulate the PWM control signal of the pulse power supply based on the waveform compensation parameter sequence to obtain an optimized PWM waveform sequence.

[0041] The control module is used to control the driving circuit parameters of the pulse power supply based on the optimized PWM waveform sequence in order to obtain a stable output voltage waveform.

[0042] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0043] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0044] This invention provides an output control method for a pulse power supply, comprising the following steps: acquiring the voltage waveform output by the pulse power supply to obtain an output voltage sampling sequence; performing wavelet transform decomposition on the output voltage sampling sequence to obtain voltage waveform characteristic coefficients; performing adaptive threshold segmentation on the voltage waveform based on the voltage waveform characteristic coefficients to obtain pulse edge feature points; performing dynamic compensation calculation on the pulse edge feature points to obtain a waveform compensation parameter sequence; modulating the PWM control signal of the pulse power supply based on the waveform compensation parameter sequence to obtain an optimized PWM waveform sequence; and controlling the driving circuit parameters of the pulse power supply based on the optimized PWM waveform sequence to obtain a stable output voltage waveform. This method solves the technical problem that traditional pulse power supplies often use fixed-parameter PWM control strategies, which are difficult to cope with output fluctuations caused by load changes, temperature drift, and device aging, leading to waveform distortion. It realizes the use of the waveform compensation parameter sequence to modulate the PWM control signal, achieving closed-loop optimization from waveform feature perception to control parameter adjustment, and enabling the generated optimized PWM waveform sequence to more accurately track the ideal pulse shape. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the steps of a pulse power supply output control method in one embodiment of the present invention;

[0046] Figure 2 This is a structural block diagram of the output control system of a pulse power supply in one embodiment of the present invention;

[0047] Figure 3 This is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0048] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0050] like Figure 1 As shown, Figure 1 This invention provides an output control method for a pulse power supply, comprising the following steps:

[0051] Step S1: Acquire the voltage waveform output by the pulse power supply to obtain the output voltage sampling sequence.

[0052] Specifically, the voltage waveform output by the pulse power supply is acquired to obtain the output voltage sampling sequence. This step is the foundation of the entire output control method, aiming to acquire the actual output voltage signal of the pulse power supply in real time for subsequent precise characteristic analysis and control adjustment. In the specific implementation, a high-precision voltage sensor or voltage divider circuit is connected to the output terminal of the pulse power supply to convert the high-voltage or high-frequency pulse voltage signal into a low-level analog signal suitable for processing by the acquisition system. A high-speed analog-to-digital converter (ADC) periodically samples this analog signal at a fixed sampling frequency, thereby forming a discrete digital sequence, i.e., the output voltage sampling sequence. This sampling sequence must ensure sufficient sampling rate and resolution to accurately reproduce the key morphological features of the pulse waveform, such as the rising edge, falling edge, and pulse width, avoiding waveform distortion or feature loss due to undersampling. For example, in the application scenario of precision resistance welding machine, the pulse power supply needs to output high-energy pulses with a width of milliseconds and a rise time of less than 10 microseconds. At this time, it is necessary to use an ADC with a sampling rate of at least tens of megabits to continuously acquire the output voltage, so as to ensure that the obtained output voltage sampling sequence can accurately reflect the actual output state of each pulse, providing a reliable data basis for subsequent wavelet transform decomposition and edge feature extraction, thereby ensuring the accuracy and stability of the entire control system.

[0053] Step S2: Perform wavelet transform decomposition on the output voltage sampling sequence to obtain voltage waveform characteristic coefficients.

[0054] Specifically, the output voltage sampling sequence is decomposed using wavelet transform to obtain voltage waveform characteristic coefficients. This step is a crucial process for in-depth time-frequency analysis of the output voltage sampling sequence after voltage signal acquisition. Since the voltage waveform output by a pulsed power supply typically exhibits rapidly changing transient characteristics, such as steep rising and falling edges, traditional Fourier transforms struggle to provide good time and frequency resolution simultaneously. Wavelet transforms, however, can effectively capture local features of the signal at different time scales through multi-scale decomposition. In practice, the aforementioned acquired output voltage sampling sequence is used as the input signal. Appropriate wavelet basis functions (such as Daubechies wavelets or Morlet wavelets) and decomposition levels are selected to perform discrete or continuous wavelet transforms on the signal. This unfolds the original voltage waveform across multiple frequency sub-bands, separating high-frequency detail coefficients reflecting waveform abrupt changes, edges, and distortions, as well as low-frequency approximation coefficients characterizing the overall trend. These coefficients collectively constitute the voltage waveform characteristic coefficients, which not only preserve the time-series information of the original waveform but also enhance sensitivity to pulse edges and transient disturbances. For example, in the application of precision resistance welding machines, when the output voltage experiences a slight delay or oscillation in the rising edge due to poor load contact, wavelet transform can significantly amplify these abnormal features in the high-frequency subband, enabling subsequent processing to accurately identify the problem and providing solid data support for achieving high-precision adaptive control.

[0055] Step S3: Based on the voltage waveform characteristic coefficients, perform adaptive threshold segmentation on the voltage waveform to obtain pulse edge feature points.

[0056] Specifically, the voltage waveform is adaptively thresholded based on its characteristic coefficients to obtain pulse edge feature points. This step is a core processing step after wavelet transform decomposition, using the extracted voltage waveform characteristic coefficients to accurately identify key transition positions in the pulse waveform. In practice, firstly, the distribution characteristics of the high-frequency detail coefficients obtained from the wavelet transform are analyzed to determine their amplitude variation trends on the time axis, as these coefficients reflect abrupt changes in the voltage waveform, such as the start and end times of rising and falling edges. Based on this, an adaptive threshold algorithm (such as a dynamic threshold based on statistical standard deviation or energy entropy) is used to segment these coefficients. This threshold is not fixed but automatically adjusted according to the local characteristics of the current output voltage sampling sequence, effectively avoiding misjudgments caused by noise interference or amplitude fluctuations. When a coefficient exceeds the dynamically generated threshold, a significant waveform change is determined to exist near that moment, thereby locating pulse edge feature points, including key positions such as the start of the rising edge, peak point, and end of the falling edge. For example, in the application of precision resistance welding machines, if the pulse rise edge becomes slower due to electrode wear, the adaptive threshold segmentation can automatically reduce the threshold sensitivity according to the real-time voltage waveform characteristic coefficient, and still accurately capture the actual edge start time, ensuring the accuracy of subsequent compensation control, thereby ensuring the consistency of welding energy and process stability.

[0057] Step S4: Dynamic compensation calculation is performed on the pulse edge feature points using a fuzzy neural network to obtain a waveform compensation parameter sequence.

[0058] Specifically, a waveform compensation parameter sequence is obtained by dynamically compensating the pulse edge feature points using a fuzzy neural network. This step is crucial for real-time prediction and correction of waveform deviations based on accurate extraction of pulse edge feature points and the use of intelligent algorithms. In practice, the obtained pulse edge feature points, such as the rising edge start time, pulse width duration, and falling edge end point, are used as input variables and fed into a pre-trained fuzzy neural network model. This model integrates the reasoning ability of fuzzy logic systems for nonlinear relationships with the self-learning and adaptive capabilities of neural networks. It can determine the degree of deviation between the actual edge and the ideal waveform based on historical data and current trends, and output the corresponding compensation amount. These compensation amounts exist as a continuous numerical sequence, constituting the waveform compensation parameter sequence, which guides subsequent PWM signal adjustments. Since the fuzzy rules can set initial parameters based on expert experience, and the neural network can continuously optimize weights through online learning, the entire system possesses strong robustness and adaptability, capable of handling uncertainties such as load changes and temperature drift. For example, in the application of precision resistance welding machines, when electrode aging causes the rise time of each pulse to gradually lengthen, the fuzzy neural network can dynamically calculate the compensation parameters that need to be triggered earlier for the PWM signal based on the continuously monitored trend of pulse edge feature points. This generates a drive command to turn on earlier, effectively offsetting the impact of hardware delay and ensuring that the output voltage waveform remains stable and accurate.

[0059] Step S5: Modulate the PWM control signal of the pulse power supply based on the waveform compensation parameter sequence to obtain the optimized PWM waveform sequence.

[0060] Specifically, the PWM control signal of the pulse power supply is modulated based on the waveform compensation parameter sequence to obtain an optimized PWM waveform sequence. This step is a crucial conversion process that transforms the compensation information obtained from intelligent calculation into actual control commands. In practice, the waveform compensation parameter sequence output through the fuzzy neural network, such as pulse start time offset, pulse width adjustment coefficient, or duty cycle correction value, is first input into the PWM signal generation module to dynamically modulate the original PWM control signal of the pulse power supply. This modulation is not a simple linear superposition, but rather a fine-tuning of key parameters such as the PWM trigger time, high-level duration, or dead time based on the compensation parameters, thereby generating an optimized PWM waveform sequence that matches the current output voltage characteristics. Because this process achieves closed-loop feedback from waveform error perception to control signal correction, it can effectively eliminate output distortion caused by power device delays, drive circuit nonlinearity, or load fluctuations. For example, in the application of precision resistance welding machines, when the pulse rising edge is detected to be lagging due to IGBT switching delay, the waveform compensation parameter sequence will include the compensation amount for early conduction. Based on this, the control system will issue a PWM drive signal in advance, so that the actual output voltage waveform can be accurately aligned with the set timing, thereby ensuring that the energy output of each welding is highly consistent, significantly improving process repeatability and product quality stability.

[0061] Step S6: Based on the optimized PWM waveform sequence, control the driving circuit parameters of the pulse power supply to obtain a stable output voltage waveform.

[0062] Specifically, the optimized PWM waveform sequence is used to control the driving circuit parameters of the pulse power supply to obtain a stable output voltage waveform. This step is the execution link for the entire control method to achieve closed-loop regulation and ultimately ensure output quality. In specific implementation, the optimized PWM waveform sequence generated by modulation is input into the driving circuit of the pulse power supply as a control signal for switching devices (such as MOSFETs or IGBTs). The operating state of the driving circuit is dynamically adjusted by precisely regulating the turn-on and turn-off timing of these power devices. Since the optimized PWM waveform sequence has incorporated the correction information obtained from the original output voltage sampling sequence through wavelet transform, adaptive segmentation, and fuzzy neural network compensation, it has higher timing accuracy and dynamic response capability, and can effectively suppress waveform distortion caused by circuit parasitic parameters, temperature drift, or load abrupt changes. Driving circuit parameters such as gate drive current, dead time, and drive voltage amplitude can also be adaptively configured according to the frequency and duty cycle changes of the PWM waveform, thereby further improving the stability and efficiency of the switching process. For example, in the application of precision resistance welding machines, when the welding current needs to rise to thousands of amperes in a very short time, the optimized PWM waveform sequence can ensure that the drive circuit accurately controls the rapid turn-on and turn-off of the IGBT, avoiding rise edge delay or energy fluctuation caused by insufficient drive or overlapping conduction, and finally achieving a high degree of consistency and stability of the output voltage waveform of each pulse, meeting the stringent requirements of high-precision welding processes for power supply dynamic performance.

[0063] In a specific embodiment, the step of acquiring the voltage waveform output by the pulse power supply to obtain the output voltage sampling sequence includes:

[0064] The voltage waveform output by the pulse power supply is sampled by a differential probe in high-voltage isolation to obtain the original voltage sampling signal. The original voltage sampling signal is then subjected to bandpass filtering to obtain a filtered voltage signal sequence.

[0065] Time base compensation is performed on the filtered voltage signal sequence to obtain the output voltage sampling sequence.

[0066] Specifically, the voltage waveform output by the pulse power supply is acquired to obtain an output voltage sampling sequence. This process involves not only signal acquisition but also precise assurance of signal quality and time consistency. Specifically, it includes high-voltage isolation sampling of the pulse power supply output voltage waveform using a differential probe to obtain the raw voltage sampling signal. This raw voltage sampling signal is then bandpass filtered to obtain a filtered voltage signal sequence. Time base compensation is then applied to the filtered voltage signal sequence to finally obtain the output voltage sampling sequence used for subsequent analysis. In practical operation, since pulse power supplies typically operate in complex electromagnetic environments with high voltage, high current, and rapid switching, directly acquiring the output voltage signal is highly susceptible to common-mode interference, ground loop noise, and the risk of high-voltage breakdown. Therefore, a differential probe with high common-mode rejection ratio and electrical isolation capability must be used for high-voltage isolation sampling. The differential probe, through its internal high-impedance voltage divider network and isolation amplifier, proportionally attenuates and converts the voltage waveform at the high-voltage output terminal into a low-voltage differential signal. This ensures operational safety while accurately reproducing the dynamic changes of the original voltage, forming the raw voltage sampling signal. For example, in applications involving precision resistance welding machines, the output voltage may rapidly change between tens and hundreds of volts with a rise time of less than 10 microseconds. Using ordinary single-ended probes or non-isolated acquisition methods could introduce severe noise, potentially damaging the acquisition equipment or distorting the data. Differential probes, on the other hand, can effectively suppress high-frequency common-mode interference in the welding circuit, ensuring the integrity and accuracy of the original voltage sampling signal. Subsequently, due to the inevitable introduction of high-frequency noise (such as switching noise and electromagnetic radiation interference) and low-frequency drift (such as temperature drift and power fluctuations) during actual transmission and acquisition, the original voltage sampling signal needs to be bandpass filtered to retain the main frequency band components of the pulse signal while suppressing out-of-band interference. The design of a bandpass filter requires setting the passband range based on the typical frequency characteristics of the pulse waveform. For example, for millisecond-level pulses, the main energy is concentrated between several hundred hertz and tens of kilohertz. Therefore, the lower cutoff frequency of the bandpass filter can be set to 100 Hz and the upper cutoff frequency to 50 kHz, effectively filtering out power frequency interference (50 / 60 Hz) and high-frequency switching noise (>100 kHz), resulting in a cleaner filtered voltage signal sequence. This step significantly improves the signal-to-noise ratio, providing a high-quality data foundation for subsequent wavelet transform and edge detection. However, even if the signal amplitude is effectively recovered, there may still be deviations on the time axis. Especially during multi-channel synchronous acquisition or long-distance signal transmission, factors such as analog front-end circuitry, filter group delays, and ADC sampling clock asynchrony can cause a slight shift in the time axis of the filtered voltage signal sequence. Although this delay is short, it is sufficient to affect the positioning accuracy of pulse edge feature points in high-precision control.Therefore, time base compensation is necessary for the filtered voltage signal sequence. This involves correcting the timestamps of the sampled data by calibrating known delays or introducing synchronization reference signals (such as synchronization trigger pulses or coded time stamps) to ensure that each sampling point precisely corresponds to its actual physical occurrence time. This results in an output voltage sampling sequence that accurately reflects the pulse timing characteristics. In precision resistance welding machines, without time base compensation, even a delay of only a few microseconds can lead to misjudging the rising edge start point, affecting the compensation calculation of the fuzzy neural network and the accuracy of PWM modulation, ultimately causing welding energy fluctuations. Therefore, only through the synergistic effect of differential probe high-voltage isolation sampling, bandpass filtering, and time base compensation can the obtained output voltage sampling sequence be ensured to have both high fidelity and accurate time consistency, providing reliable data support for subsequent stages of the entire pulse power supply output control method.

[0067] In a specific embodiment, the step of performing wavelet transform decomposition on the output voltage sampling sequence to obtain voltage waveform characteristic coefficients includes:

[0068] Wavelet basis function mapping is performed on the output voltage sampling sequence to obtain the initial scaling coefficient matrix, and the energy density in the initial scaling coefficient matrix is ​​calculated to obtain the energy distribution sequence of each frequency band.

[0069] Based on the energy distribution sequences of each frequency band, wavelet coefficients are reconstructed to obtain a frequency band decomposition coefficient group, and singular value decomposition is performed on the frequency band decomposition coefficient group to obtain a feature space mapping matrix.

[0070] The feature space mapping matrix is ​​subjected to a nonlinear projection transformation to obtain a set of projection coefficients, and feature weight fusion is performed based on the set of projection coefficients to obtain the voltage waveform feature coefficients.

[0071] Specifically, the output voltage sampling sequence is decomposed using wavelet transform to obtain voltage waveform characteristic coefficients. This process not only involves traditional wavelet multi-scale analysis but also introduces a deep feature extraction mechanism based on energy distribution, matrix reconstruction, and nonlinear mapping to achieve a refined characterization of the intrinsic structure of the pulse voltage waveform. In practice, the output voltage sampling sequence obtained after time base compensation is first mapped using wavelet basis functions. That is, a suitable wavelet basis (such as db4, sym6, orthogonal wavelets with tight support and good regularity) is selected to perform a discrete wavelet transform (DWT) on the discrete-time sequence, thereby projecting the original signal into wavelet spaces at different scales and shifts, generating an initial scale coefficient matrix containing approximation coefficients and detail coefficients at each level. This matrix not only reflects the distribution of the signal in different frequency subbands (such as low-frequency trends, mid-frequency pulse main body, high-frequency edges and noise) but also preserves its localization characteristics on the time axis, providing a multi-dimensional data foundation for subsequent analysis. Based on this, the energy density in the initial scaling coefficient matrix is ​​further calculated by squaring and summing the wavelet coefficients at each scaling level to obtain the energy distribution sequence for each frequency band. This sequence can intuitively show which frequency ranges the pulse energy is mainly concentrated in. For example, in the application scenario of precision resistance welding machines, the main energy of an ideal pulse is usually concentrated in the mid-to-high frequency range, while poor electrode contact or load changes can lead to abnormal increases in high-frequency energy or aggravated low-frequency drift. Therefore, the energy distribution sequence for each frequency band becomes an important indicator for judging the health status of the waveform. Subsequently, based on the energy distribution sequence for each frequency band as a weight guide, the wavelet coefficients in the initial scaling coefficient matrix are reconstructed by frequency band weighting, highlighting the high-frequency components that are sensitive to the pulse edge and suppressing the noise-dominated invalid components, thereby obtaining a more physically meaningful frequency band decomposition coefficient group. Next, in order to explore the potential structural features within this coefficient group, singular value decomposition (SVD) is performed on it, decomposing the original high-dimensional coefficient data into left singular vectors, singular values, and right singular vectors, and then constructing a feature space mapping matrix. This matrix is ​​essentially a projection representation of the wavelet coefficients in the principal component direction, which can effectively compress redundant information and retain the most discriminative waveform feature patterns. Subsequently, to further enhance the nonlinear separability and expressive power of the features, a nonlinear projection transformation is performed on the feature space mapping matrix. For example, a kernel function (such as the RBF kernel) or a deep activation function (such as the Sigmoid or Swish) is used to map the matrix elements, thereby transforming linearly inseparable waveform differences into separable patterns in a high-dimensional nonlinear space, resulting in a set of projection coefficients. Finally, based on this set of projection coefficients and weighting factors set by prior knowledge or training data, a feature weight fusion operation is performed. This involves weighted summation or attention mechanism fusion of coefficients in different projection directions, highlighting feature dimensions sensitive to pulse rising edges, falling edges, and plateau stability. Ultimately, a set of highly condensed and highly discriminative voltage waveform feature coefficients is generated.These coefficients not only contain the amplitude, frequency, and timing information of the original waveform, but also integrate its deep features in multi-scale energy distribution, principal component structure, and nonlinear space, providing a more robust and accurate input basis for subsequent adaptive threshold segmentation. For example, during the operation of the same precision resistance welding machine, when the number of welding operations increases and the electrodes age, the output voltage waveform may exhibit subtle distortions such as rising edge delay and plateau fluctuations. Traditional single-scale wavelet coefficients are difficult to accurately capture such complex anomalies. However, through the aforementioned composite processing flow that includes energy distribution-guided reconstruction, singular value decomposition, and nonlinear projection fusion, the obtained voltage waveform characteristic coefficients can significantly amplify these early degradation features, enabling the system to activate the compensation mechanism before the distortion affects the welding quality, thereby ensuring process consistency and equipment reliability.

[0072] In a specific embodiment, the step of performing adaptive threshold segmentation on the voltage waveform based on the voltage waveform feature coefficients to obtain pulse edge feature points includes:

[0073] Multi-scale gradient analysis is performed on the characteristic coefficients of the voltage waveform to obtain the characteristic gradient matrix, and the extreme points in the characteristic gradient matrix are detected to obtain the gradient jump sequence.

[0074] Envelope extraction is performed on the gradient jump sequence to obtain an envelope curve feature group, and the dynamic threshold of the envelope curve feature group is calculated to obtain a threshold distribution curve;

[0075] Based on the threshold distribution curve, the voltage waveform is segmented using local variance weighting to obtain candidate feature points, and density clustering analysis is performed on the candidate feature points to obtain feature point clusters.

[0076] Edge localization calibration is performed on the clusters of feature points to obtain an edge localization sequence, and temporal features are extracted from the edge localization sequence to obtain pulse edge feature points.

[0077] Specifically, adaptive threshold segmentation of the voltage waveform is performed based on the voltage waveform characteristic coefficients to obtain pulse edge feature points. This process is a key step in achieving high-precision and interference-resistant pulse edge recognition after acquiring the voltage waveform characteristic coefficients containing multi-scale energy distribution and nonlinear structure information. In practice, firstly, multi-scale gradient analysis is performed on the voltage waveform characteristic coefficients, i.e., the first or second-order differential gradient is calculated at different decomposition scales to capture the rate of change of the waveform within a local time window, thus forming a characteristic gradient matrix reflecting the intensity and direction of signal abrupt changes. This matrix can effectively enhance the gradient response in key regions such as the rising and falling edges of the pulse, while suppressing false edges caused by flat regions or noise. Subsequently, extreme points are detected in the characteristic gradient matrix, i.e., the locations where the gradient amplitude reaches a local maximum or minimum are found. These locations often correspond to moments when the voltage waveform undergoes significant jumps, thus obtaining a gradient jump sequence as a preliminary candidate set of potential edges. To further improve detection stability and avoid misjudgments due to transient interference, the system extracts the envelope of the gradient jump sequence and uses Hilbert transform or moving maximum method to fit its overall trend, generating an envelope curve feature set. This set of data reflects the overall distribution and dynamic range of the gradient jumps. Based on this, the statistical characteristics of the envelope curve feature set, such as mean, standard deviation, or quantiles, are calculated. A dynamic threshold that adjusts with signal strength is adaptively determined in conjunction with real-time operating conditions, forming a threshold distribution curve that changes with waveform energy fluctuations. This overcomes the problem of poor adaptability of a fixed threshold under different loads or amplitudes. Next, based on the threshold distribution curve, the original voltage waveform is segmented using local variance weighting. That is, the local variance of the signal is calculated near each time point and used as a weighting factor to enhance the segmentation sensitivity in regions of drastic waveform changes, while reducing the probability of false triggering in stable regions. This identifies a series of candidate feature points that meet the conditions. To further eliminate isolated noise points or false edges, the system performs density clustering analysis on the candidate feature points, for example, using the DBSCAN algorithm. Clusters are formed based on the temporal distance and neighborhood density between points, grouping points concentrated near the true pulse edges into the same cluster, while discrete noise points are identified as noise and eliminated, resulting in more representative feature point clusters. Subsequently, for each cluster, edge localization calibration is performed by fitting the temporal distribution center or weighted average position of its internal points, eliminating localization errors caused by sampling discreteness or cluster offset, ultimately generating an edge localization sequence accurate to the microsecond level. Finally, key temporal features are extracted from this sequence, such as the start time of the first significant cluster as the rising edge start time, the end time of the last cluster as the falling edge end time, and the time interval between clusters as pulse width information, forming a complete set of pulse edge feature points.For example, in the application of precision resistance welding machines, when the contact resistance increases due to oxidation of the electrode surface, the rising edge of the actual output voltage will experience a slight delay and oscillation. Traditional fixed threshold methods may misjudge the edge position due to the decrease in amplitude. However, this solution can accurately capture the initial jump through multi-scale gradient analysis. Combined with envelope extraction and dynamic threshold generation, it can still accurately set the segmentation boundary. Furthermore, it can eliminate multiple pseudo-jump points caused by oscillation through density clustering. Finally, after edge positioning calibration, it accurately restores the true pulse start and end times, ensuring that the input data for subsequent fuzzy neural network compensation calculations is highly reliable, thereby ensuring the consistency of welding energy and the stability of process quality.

[0078] In a specific embodiment, the step of performing local variance-weighted segmentation on the voltage waveform based on the threshold distribution curve to obtain candidate feature points includes:

[0079] The threshold distribution curve is divided into intervals to obtain a threshold partition sequence, and the threshold partition sequence is dynamically adjusted to obtain an adaptive threshold group.

[0080] Based on the adaptive threshold group, the local variance of the voltage waveform is calculated to obtain the waveform variance matrix, and weight coefficients are assigned to the waveform variance matrix to obtain weighted segmentation parameters;

[0081] The voltage waveform is segmented based on the weighted segmentation parameters to obtain candidate feature points.

[0082] Specifically, based on the threshold distribution curve, the voltage waveform is segmented using local variance weighting to obtain candidate feature points. This process is a key step in improving the accuracy of pulse edge recognition by further improving the dynamic threshold distribution curve through partitioning modeling and weighting mechanisms. In practice, the threshold distribution curve is first divided into intervals, that is, the entire continuous threshold curve is divided into multiple continuous sub-intervals according to time or amplitude characteristics, forming a threshold partitioning sequence. Each partition corresponds to a different operating stage in the voltage waveform, such as the pre-pulse steady state, rising edge transition region, plateau maintenance region, and falling edge region, thereby achieving segmented modeling of complex waveform structures. Subsequently, the threshold partitioning sequence is dynamically adjusted, that is, the threshold interval is adaptively scaled according to the signal fluctuation characteristics (such as standard deviation, range, or energy density) within each partition, making the threshold more sensitive in high-change regions and more robust in low-change regions. Finally, an adaptive threshold set matching the local operating conditions is generated. This set of parameters not only reflects the global trend but also adapts to local distortions. Based on this, local variance calculation is performed on the voltage waveform using the adaptive threshold group. Specifically, within each threshold partition, the variance of the voltage signal within a short time window is calculated using a sliding window approach, forming a waveform variance matrix that reflects the severity of waveform fluctuations. This matrix highlights the high variance characteristics of edge regions and suppresses the low-frequency drift effect of stable segments. Next, weighting coefficients are assigned to the waveform variance matrix. Different weights are assigned to the corresponding variance data based on the adaptive threshold sensitivity of each partition. For example, higher weights are given to edge-sensitive areas, while weights are appropriately reduced in noise-prone areas, thus constructing a weighted segmentation parameter that prioritizes key areas. Subsequently, the voltage waveform is segmented using this weighted segmentation parameter. Within each threshold partition, the weighted variance threshold is used as a criterion to identify time points where significant signal changes occur and mark them as potential waveform transition locations. Finally, the results from each partition are integrated to obtain a preliminary set of candidate feature points. For example, in the application scenario of precision resistance welding machines, when the rising edge slows down and is accompanied by small oscillations due to electrode oxidation during the welding process, traditional single threshold segmentation is prone to misjudging the oscillation peak as multiple edge points. However, this solution divides the threshold distribution curve into intervals, allowing for more refined threshold partitioning in the rising edge region. Combined with dynamic range adjustment, it enhances the detection sensitivity of this region. Furthermore, the local variance weighting mechanism highlights the continuous change characteristics of the true edge and suppresses the pseudo-jumps caused by instantaneous oscillations. This accurately extracts the unique rising edge start point as a candidate feature point, providing high-quality input for subsequent density clustering and edge calibration, ensuring the robustness of the control system and process consistency.

[0083] In a specific embodiment, the step of performing local variance-weighted segmentation on the voltage waveform based on the threshold distribution curve to obtain candidate feature points includes:

[0084] The threshold distribution curve is subjected to a second-order differential transformation to obtain the threshold curvature change, and the voltage waveform is subjected to a sliding window variance calculation to obtain a local variance distribution sequence.

[0085] The local variance distribution sequence is weighted and fused by the threshold curvature change to obtain a weighted variance feature matrix, and peak detection analysis is performed based on the weighted variance feature matrix to obtain the variance peak position sequence.

[0086] The voltage waveform segmentation boundary is determined based on the variance peak position sequence, and the voltage waveform segmentation boundary is obtained by marking and locating feature points on the voltage waveform segmentation boundary to obtain candidate feature points.

[0087] Specifically, the voltage waveform is segmented using local variance weighting based on the threshold distribution curve to obtain candidate feature points. This process is a key step in further integrating local statistical characteristics of the signal under dynamic threshold guidance to improve edge recognition accuracy. In practice, firstly, a second-order differential transformation is performed on the threshold distribution curve calculated from the envelope curve feature set, i.e., its second-order difference at continuous time points is calculated, thereby obtaining the threshold curvature change reflecting the abrupt change in the threshold change rate. This matrix can identify the locations where significant inflection points occur in the threshold itself. These locations often correspond to key regions in the original voltage waveform that transition from a stable segment to the start or end of a pulse, thus containing important prior information about waveform structure changes. Simultaneously, a sliding window variance calculation is performed on the original voltage waveform. That is, a fixed-length time window (e.g., 50 sampling points) is set around each sampling point, and the variance of the voltage value within this window is calculated. As the window moves along the time axis, a continuous local variance distribution sequence is generated. This sequence effectively characterizes the intensity of voltage waveform fluctuations at different times: near the rising or falling edge of the pulse, voltage changes drastically, and the local variance increases significantly; while in the steady-state high or low level region, the variance tends to flatten. Subsequently, the local variance distribution sequence is weighted and fused by the threshold curvature change, that is, the weight coefficients in the threshold curvature change are multiplied point by point by the values ​​at the corresponding positions in the local variance distribution sequence, thereby constructing a weighted variance feature matrix. The significance of this weighting mechanism is that when the threshold itself changes drastically (i.e., large curvature), it indicates that the current region is in a critical area of ​​waveform dynamic characteristic transformation. At this time, the contribution of the variance feature in this region should be enhanced, making it more prioritized in subsequent segmentation, thus achieving "dual sensitivity"—paying attention to the volatility of the signal itself, and intelligently enhancing it in combination with the changing trend of the dynamic threshold. On this basis, peak detection analysis is performed on the weighted variance feature matrix, and the local maximum points are identified by the sliding extreme value search or threshold comparison method. These peak points usually correspond to the most drastic changes in the voltage waveform, thus forming a variance peak position sequence. Next, the segmentation boundaries of the voltage waveform are determined based on the variance peak position sequence. This involves mapping each variance peak point back to the original voltage time axis, serving as a candidate segmentation point for potential pulse start, peak, or end times, forming a set of voltage waveform segmentation boundaries accurate to the sampling point level. Finally, feature point labeling and localization are performed on these segmentation boundaries. This involves combining time-domain logic rules such as the direction of voltage amplitude change (rising or falling), duration, and interval between adjacent peaks to determine whether each boundary point belongs to the start of a rising edge, the edge of a top plateau, or the end of a falling edge, and assigning corresponding semantic labels. Ultimately, a set of candidate feature points with timestamps and type identifiers is output.For example, in the application of precision resistance welding machines, when loose contact in the welding circuit causes voltage waveform surges or ringing, traditional single threshold segmentation can easily misjudge these high-frequency oscillations as multiple independent pulse edges. However, this solution introduces a local variance weighting mechanism, which can suppress the weight of variance response in the ringing region when there are high-frequency fluctuations but the overall trend has not fundamentally changed. At the same time, it relies on the change in threshold curvature to identify the true starting inflection point of the main pulse, thereby accurately locking the unique and effective rising edge position, avoiding missegmentation, and ensuring that the candidate feature points input in the subsequent density clustering and edge calibration stages have high authenticity and low noise interference, providing a solid guarantee for the stability and accuracy of the entire control system.

[0088] In a specific embodiment, the step of dynamically compensating the pulse edge feature points using a fuzzy neural network to obtain a waveform compensation parameter sequence includes:

[0089] The pulse edge feature points are mapped to fuzzy membership using a fuzzy neural network to obtain a fuzzy quantization sequence of feature points. Neuron activation calculation is then performed based on the fuzzy quantization sequence of feature points to obtain node response features.

[0090] Multi-level propagation operation is performed on the node response features to obtain a compensation parameter estimation sequence, and feedback correction is performed on the compensation parameter estimation sequence to obtain compensation error correction;

[0091] Based on the compensation error correction, parameter optimization and fusion are performed to obtain the compensation parameter update sequence, and the compensation parameter update sequence is then subjected to nonlinear mapping transformation to obtain the waveform compensation parameter sequence.

[0092] Specifically, a fuzzy neural network is used to dynamically compensate the pulse edge feature points to obtain a waveform compensation parameter sequence. This process is the core intelligent decision-making step that transforms the previously extracted precise edge information into executable controllable correction quantities. In specific implementation, the pulse edge feature points obtained after edge localization and calibration, such as the rising edge start time, peak arrival time, falling edge end point, and pulse width duration, are first input into the input layer of the fuzzy neural network for fuzzy membership mapping. This mapping process, based on a preset fuzzy rule base (such as linguistic variables like "small rising edge delay" and "significant pulse width offset"), converts precise numerical feature points into fuzzy sets with semantic expressive capabilities, thereby generating a fuzzy quantization sequence of feature points. This sequence not only retains the numerical characteristics of the original data but also introduces tolerance to uncertainties and nonlinear relationships, enabling the system to maintain stable inference even when faced with small fluctuations or measurement errors. Next, based on the fuzzy quantization sequence of the feature points, the neurons within the network perform weighted summation of the input signal according to their connection weights and apply activation functions (such as Sigmoid or ReLU) to complete the calculation of the node response features. These response features reflect the network's initial perception and internal representation of the current waveform deviation pattern. Subsequently, these node response features undergo multi-layer propagation operations within the network, that is, through forward propagation between hidden layers, information is continuously abstracted and integrated, gradually approximating the ideal compensation relationship model, and finally outputting a preliminary compensation parameter estimation sequence. This sequence may contain control correction suggestions such as the advance trigger amount of the PWM signal, the pulse width extension amount, or the duty cycle adjustment coefficient. However, due to potential undertraining or operating condition deviations in the initial model, the estimated value still contains errors. Therefore, a feedback mechanism is needed to correct it: the deviation between the actual output and the desired waveform is used as a feedback signal and compared with the compensation parameter estimation sequence to calculate the compensation error correction amount. This correction amount can be used to adjust the network weights online or directly superimposed on the original estimate, thereby improving the compensation accuracy and robustness. Building upon this, the system further optimizes and fuses parameters based on the compensation error correction. This involves employing weighted averaging, adaptive filtering, or gradient descent strategies to organically combine historical compensation experience with the current correction amount, generating a smoother and more stable compensation parameter update sequence. This avoids control jitter caused by excessive single-error errors. Finally, to adapt to the nonlinear response characteristics of the pulse power supply drive circuit, the compensation parameter update sequence undergoes a nonlinear mapping transformation. For example, through sigmoid functions or piecewise linearization, it is converted into a waveform compensation parameter sequence that matches the switching characteristics of the power devices. This ensures that the final output compensation command effectively corrects waveform distortion without causing oscillations or overshoot due to excessive adjustment.For example, in the application of precision resistance welding machines, when the electrodes gradually heat up during continuous welding, causing the contact resistance to decrease, and resulting in an accelerated rise edge and increased energy accumulation for each pulse, the fuzzy neural network can identify the trend of "earlier rise edge" through fuzzy quantization. After multi-layer propagation and feedback correction, it generates compensation parameters for a moderately delayed PWM signal and converts them into specific timing offsets through nonlinear mapping. This dynamically balances the changes caused by thermal effects, ensuring that the voltage waveform of each weld is highly consistent, maintaining process stability and the reliability of joint quality.

[0093] In a specific embodiment, obtaining an optimized PWM waveform sequence by modulating the PWM control signal of the pulse power supply based on the waveform compensation parameter sequence includes:

[0094] Modulation depth mapping is performed on the waveform compensation parameter sequence to obtain a modulation parameter vector group, and carrier synchronization analysis is performed based on the modulation parameter vector group to obtain carrier modulation features;

[0095] The carrier modulation features are subjected to multiple modulation transformations to obtain a pulse modulation sequence, and the pulse modulation sequence is subjected to edge compensation processing to obtain a compensated modulation waveform.

[0096] Based on the compensated modulation waveform, waveform shaping and reconstruction are performed to obtain a reconstructed waveform sequence. Then, based on the reconstructed waveform sequence, the PWM control signal of the pulse power supply is modulated to obtain an optimized PWM waveform sequence.

[0097] Specifically, the PWM control signal of the pulse power supply is modulated based on the waveform compensation parameter sequence to obtain an optimized PWM waveform sequence. This process is a key conversion step in transforming the abstract compensation instructions output by the intelligent computing layer into a practically executable high-precision drive signal. In practice, the waveform compensation parameter sequence output by the fuzzy neural network is first subjected to modulation depth mapping. This involves classifying and normalizing compensation parameters such as time offset, pulse width correction coefficient, and duty cycle increment according to their physical meaning and control dimensions, constructing a multi-dimensional modulation parameter vector group. This vector group not only contains amplitude adjustment information but also integrates timing adjustment and dynamic response characteristics, providing structured input for subsequent modulation. Based on this, the system performs carrier synchronization analysis based on the modulation parameter vector group. This involves combining the pulse power supply's current operating frequency, carrier ratio, and phase information to time-align and phase-match the modulation parameters with the periodic characteristics of a triangular or sawtooth carrier, extracting carrier modulation features that reflect the compensation intention. This ensures that the modulation process is strictly synchronized with the power supply system's switching cycle on the time axis, avoiding control delays or harmonic distortion caused by phase mismatch. Subsequently, the carrier modulation characteristics are input into a multi-modulation conversion module. This module can employ a combination of various modulation strategies, such as SPWM (Sinusoidal Pulse Width Modulation), SVPWM (Space Vector Modulation), or Specific Harmonic Elimination Modulation (SHEPWM), to adaptively select the optimal modulation path based on the current load state and output target, thereby generating a preliminary pulse modulation sequence. This sequence already possesses pulse width and distribution characteristics that reflect the compensation intention. However, due to the non-ideal characteristics of power devices (such as IGBTs or MOSFETs), such as turn-on delay and turn-off tailing, the original modulation sequence may still cause edge distortion in actual driving. Therefore, edge compensation processing is required for the pulse modulation sequence. That is, based on the switching time parameters provided in the device datasheet or the online identified dynamic delay model, the rising and falling edges of each pulse are advanced or delayed to generate a compensated modulation waveform to offset the timing errors caused by the transistor drive circuit. Next, waveform shaping and reconstruction are performed based on the compensated modulated waveform. Through digital filtering, dead-time compensation, and minimum pulse width clamping, narrow pulses, glitches, or oscillating components are further eliminated to ensure that the reconstructed waveform sequence meets high reliability requirements in terms of both electrical safety and signal integrity. Finally, the PWM control signal of the modulated pulse power supply is based on the reconstructed waveform sequence. This digital waveform sequence is downloaded to a PWM generator (such as a hardware comparator unit in a DSP or FPGA) to drive it to generate an optimized PWM waveform sequence consistent with the ideal compensation target. This sequence not only has superior time accuracy compared to traditional control methods but also dynamically adapts to changes in operating conditions, achieving precise tracking and stable control of the output voltage waveform.For example, in the application scenario of precision resistance welding machines, when the thickness of the welding material changes and causes fluctuations in load impedance, the waveform compensation parameter sequence will include compensation instructions to extend the pulse width and increase the peak current. After modulation depth mapping and carrier synchronization analysis, the system generates a corresponding high duty cycle modulation signal, and ensures that the IGBT can still conduct quickly under high current through edge compensation processing. Finally, the optimized PWM waveform sequence output after waveform shaping and reconstruction can accurately control the energy output, ensure the consistency of the penetration depth and joint strength of each weld, and significantly improve the process stability and product quality consistency of the automated welding production line.

[0098] The output control method of the pulse power supply in the embodiments of the present invention has been described above. The output control system of the pulse power supply in the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 2 One embodiment of the pulse power supply output control system of the present invention includes:

[0099] The acquisition module 21 is used to acquire the voltage waveform output by the pulse power supply to obtain the output voltage sampling sequence;

[0100] Decomposition module 22 is used to perform wavelet transform decomposition on the output voltage sampling sequence to obtain voltage waveform characteristic coefficients;

[0101] The segmentation module 23 is used to perform adaptive threshold segmentation on the voltage waveform based on the voltage waveform feature coefficients to obtain pulse edge feature points;

[0102] The calculation module 24 is used to perform dynamic compensation calculation on the pulse edge feature points through a fuzzy neural network to obtain a waveform compensation parameter sequence;

[0103] Modulation module 25 is used to modulate the PWM control signal of the pulse power supply based on the waveform compensation parameter sequence to obtain an optimized PWM waveform sequence;

[0104] The control module 26 is used to control the driving circuit parameters of the pulse power supply based on the optimized PWM waveform sequence in order to obtain a stable output voltage waveform.

[0105] In this embodiment, the specific implementation of each unit in the above system embodiment is described in the above method embodiment, and will not be repeated here.

[0106] Reference Figure 3 This invention also provides a computer device whose internal structure can be as follows: Figure 3As shown, the computer device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the data corresponding to this embodiment. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0107] Those skilled in the art will understand that Figure 3 The structures shown are merely block diagrams of some structures related to the present invention and do not constitute a limitation on the computer devices on which the present invention is applied.

[0108] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0109] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the present invention and embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0110] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0111] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for output control of a pulse power supply, characterized in that, Includes the following steps: The voltage waveform output by the pulse power supply is acquired to obtain the output voltage sampling sequence; The output voltage sampling sequence is decomposed by wavelet transform to obtain the voltage waveform characteristic coefficients; Based on the voltage waveform characteristic coefficients, adaptive threshold segmentation is performed on the voltage waveform to obtain pulse edge feature points; The waveform compensation parameter sequence is obtained by dynamically compensating the pulse edge feature points using a fuzzy neural network. The optimized PWM waveform sequence is obtained by modulating the PWM control signal of the pulse power supply based on the waveform compensation parameter sequence. Based on the optimized PWM waveform sequence, the driving circuit parameters of the pulse power supply are controlled to obtain a stable output voltage waveform; The wavelet transform decomposition of the output voltage sampling sequence to obtain voltage waveform characteristic coefficients includes: Wavelet basis function mapping is performed on the output voltage sampling sequence to obtain the initial scaling coefficient matrix, and the energy density in the initial scaling coefficient matrix is ​​calculated to obtain the energy distribution sequence of each frequency band. Based on the energy distribution sequences of each frequency band, wavelet coefficients are reconstructed to obtain a frequency band decomposition coefficient group, and singular value decomposition is performed on the frequency band decomposition coefficient group to obtain a feature space mapping matrix. The feature space mapping matrix is ​​subjected to a nonlinear projection transformation to obtain a set of projection coefficients, and feature weight fusion is performed based on the set of projection coefficients to obtain voltage waveform feature coefficients. The step of adaptive thresholding the voltage waveform based on the voltage waveform feature coefficients to obtain pulse edge feature points includes: Multi-scale gradient analysis is performed on the characteristic coefficients of the voltage waveform to obtain the characteristic gradient matrix, and the extreme points in the characteristic gradient matrix are detected to obtain the gradient jump sequence. Envelope extraction is performed on the gradient jump sequence to obtain an envelope curve feature group, and the dynamic threshold of the envelope curve feature group is calculated to obtain a threshold distribution curve; Based on the threshold distribution curve, the voltage waveform is segmented using local variance weighting to obtain candidate feature points, and density clustering analysis is performed on the candidate feature points to obtain feature point clusters. Edge localization calibration is performed on the clusters of the feature points to obtain an edge localization sequence, and the temporal features in the edge localization sequence are extracted to obtain pulse edge feature points; The step of dynamically compensating the pulse edge feature points using a fuzzy neural network to obtain a waveform compensation parameter sequence includes: The pulse edge feature points are mapped to fuzzy membership using a fuzzy neural network to obtain a fuzzy quantization sequence of feature points. Neuron activation calculation is then performed based on the fuzzy quantization sequence of feature points to obtain node response features. Multi-level propagation operation is performed on the node response features to obtain a compensation parameter estimation sequence, and feedback correction is performed on the compensation parameter estimation sequence to obtain compensation error correction; Based on the compensation error correction, parameter optimization and fusion are performed to obtain the compensation parameter update sequence, and the compensation parameter update sequence is then subjected to nonlinear mapping transformation to obtain the waveform compensation parameter sequence.

2. The output control method for a pulse power supply according to claim 1, characterized in that, The acquisition of the voltage waveform output by the pulse power supply to obtain the output voltage sampling sequence includes: The voltage waveform output by the pulse power supply is sampled by a differential probe in high-voltage isolation to obtain the original voltage sampling signal. The original voltage sampling signal is then subjected to bandpass filtering to obtain a filtered voltage signal sequence. Time base compensation is performed on the filtered voltage signal sequence to obtain the output voltage sampling sequence.

3. The output control method for a pulse power supply according to claim 1, characterized in that, The step of performing local variance-weighted segmentation on the voltage waveform based on the threshold distribution curve to obtain candidate feature points includes: The threshold distribution curve is subjected to a second-order differential transformation to obtain the threshold curvature change, and the voltage waveform is subjected to a sliding window variance calculation to obtain a local variance distribution sequence. The local variance distribution sequence is weighted and fused by the threshold curvature change to obtain a weighted variance feature matrix, and peak detection analysis is performed based on the weighted variance feature matrix to obtain the variance peak position sequence. The voltage waveform segmentation boundary is determined based on the variance peak position sequence, and the voltage waveform segmentation boundary is obtained by marking and locating feature points on the voltage waveform segmentation boundary to obtain candidate feature points.

4. The output control method for a pulse power supply according to claim 1, characterized in that, The PWM control signal of the modulated pulse power supply based on the waveform compensation parameter sequence is used to obtain an optimized PWM waveform sequence, including: Modulation depth mapping is performed on the waveform compensation parameter sequence to obtain a modulation parameter vector group, and carrier synchronization analysis is performed based on the modulation parameter vector group to obtain carrier modulation features; The carrier modulation features are subjected to multiple modulation transformations to obtain a pulse modulation sequence, and the pulse modulation sequence is subjected to edge compensation processing to obtain a compensated modulation waveform. Based on the compensated modulation waveform, waveform shaping and reconstruction are performed to obtain a reconstructed waveform sequence. Then, based on the reconstructed waveform sequence, the PWM control signal of the pulse power supply is modulated to obtain an optimized PWM waveform sequence.

5. An output control system for a pulse power supply, characterized in that, include: The acquisition module is used to acquire the voltage waveform output by the pulse power supply to obtain the output voltage sampling sequence; The decomposition module is used to perform wavelet transform decomposition on the output voltage sampling sequence to obtain voltage waveform characteristic coefficients; The segmentation module is used to perform adaptive threshold segmentation of the voltage waveform based on the characteristic coefficients of the voltage waveform to obtain pulse edge feature points; The calculation module is used to perform dynamic compensation calculations on the pulse edge feature points through a fuzzy neural network to obtain a waveform compensation parameter sequence; The modulation module is used to modulate the PWM control signal of the pulse power supply based on the waveform compensation parameter sequence to obtain an optimized PWM waveform sequence. The control module is used to control the driving circuit parameters of the pulse power supply based on the optimized PWM waveform sequence in order to obtain a stable output voltage waveform. The wavelet transform decomposition of the output voltage sampling sequence to obtain voltage waveform characteristic coefficients includes: Wavelet basis function mapping is performed on the output voltage sampling sequence to obtain the initial scaling coefficient matrix, and the energy density in the initial scaling coefficient matrix is ​​calculated to obtain the energy distribution sequence of each frequency band. Based on the energy distribution sequences of each frequency band, wavelet coefficients are reconstructed to obtain a frequency band decomposition coefficient group, and singular value decomposition is performed on the frequency band decomposition coefficient group to obtain a feature space mapping matrix. The feature space mapping matrix is ​​subjected to a nonlinear projection transformation to obtain a set of projection coefficients, and feature weight fusion is performed based on the set of projection coefficients to obtain voltage waveform feature coefficients. The step of adaptive thresholding the voltage waveform based on the voltage waveform feature coefficients to obtain pulse edge feature points includes: Multi-scale gradient analysis is performed on the characteristic coefficients of the voltage waveform to obtain the characteristic gradient matrix, and the extreme points in the characteristic gradient matrix are detected to obtain the gradient jump sequence. Envelope extraction is performed on the gradient jump sequence to obtain an envelope curve feature group, and the dynamic threshold of the envelope curve feature group is calculated to obtain a threshold distribution curve; Based on the threshold distribution curve, the voltage waveform is segmented using local variance weighting to obtain candidate feature points, and density clustering analysis is performed on the candidate feature points to obtain feature point clusters. Edge localization calibration is performed on the clusters of the feature points to obtain an edge localization sequence, and the temporal features in the edge localization sequence are extracted to obtain pulse edge feature points; The step of dynamically compensating the pulse edge feature points using a fuzzy neural network to obtain a waveform compensation parameter sequence includes: The pulse edge feature points are mapped to fuzzy membership using a fuzzy neural network to obtain a fuzzy quantization sequence of feature points. Neuron activation calculation is then performed based on the fuzzy quantization sequence of feature points to obtain node response features. Multi-level propagation operation is performed on the node response features to obtain a compensation parameter estimation sequence, and feedback correction is performed on the compensation parameter estimation sequence to obtain compensation error correction; Based on the compensation error correction, parameter optimization and fusion are performed to obtain the compensation parameter update sequence, and the compensation parameter update sequence is then subjected to nonlinear mapping transformation to obtain the waveform compensation parameter sequence.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

7. 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 steps of the method according to any one of claims 1 to 4.

Images