Digital control uninterruptible power supply parallel operation phase locking method

Abstract

The invention discloses a digital control uninterruptible power supply parallel operation phase locking method, which relates to the technical field of digital phase locking, and comprises the following steps of: in each analysis period, acquiring an output voltage waveform signal and a current frequency disturbance control parameter of each uninterruptible power supply, and respectively executing short-time Fourier transform to obtain an output voltage waveform signal and a current frequency disturbance control parameter of each uninterruptible power supply; and converting the time domain signal into a multi-dimensional dynamic spectrum matrix, extracting frequency and amplitude parameters of a disturbance signal, and constructing a corresponding uninterruptible power supply disturbance parameter set. According to the invention, through dynamic spectrum analysis and frequency domain coupling tensor modeling, accurate extraction of UPS modulation signal features and dynamic evaluation of resonance risk indexes are realized, in combination with a risk-driven frequency intervention mechanism, frequency coupling is broken, the frequency diversity and disturbance decoupling capability of the system are improved, and the reliability of the system is improved. The stability, the anti-interference performance and the intelligent response capability of the parallel system are obviously enhanced, and early warning and active suppression of lock point resonance are realized.

Description

Technical Field

[0001] This invention relates to the field of digital phase-locked loop technology, specifically to a digitally controlled phase-locked loop method for parallel operation of uninterruptible power supplies. Background Technology

[0002] Digitally controlled uninterruptible power supply (UPS) parallel operation phase-locked loop (PLL) refers to the process of precisely synchronizing the frequency and phase of the output voltage of each UPS system when multiple UPS systems are connected in parallel, using digital control technology (i.e., "phase-locking"). Since parallel operation requires each UPS output to have the same amplitude, frequency, and phase, otherwise problems such as circulating current, uneven load distribution, or system instability may occur. Therefore, it is essential to ensure a high degree of consistency between their output voltages. Digital control methods use embedded processors or digital signal processors (DSPs) to acquire the output signals of each UPS in real time. Through a phase-locked loop (PLL) or an improved digital PLL algorithm, the output phase of the UPS is dynamically adjusted to maintain consistency with the host or a common synchronization reference. This achieves precise phase synchronization between multiple UPS systems, improving system stability, fault tolerance, and power redundancy management. This method offers higher accuracy, stability, and anti-interference capabilities than traditional analog PLL methods, making it suitable for the high-reliability power supply requirements of modern intelligent power systems.

[0003] Existing technologies have the following shortcomings: In existing technologies, to improve the synchronization response sensitivity of digital phase-locked loops (DPLLs) during the parallel operation of multiple uninterruptible power supply (UPS) systems, a frequency dithering mechanism is often used. This involves introducing small-amplitude periodic or random jitter into the output frequency control signal to avoid phase-locked dead zones and enhance the ability to identify minute phase errors. However, when multiple UPS systems are operating in parallel and each is using dither modulation, if the frequency disturbance signals applied by each UPS are highly similar in frequency, amplitude, or phase parameters, multiple disturbance signals may undergo phase coupling and amplitude superposition at specific harmonic points, thereby inducing the system to fall into a latent abnormal state called the "lock-in resonance zone." In this state, although the DPLL controller surface maintains frequency lock, the system as a whole exhibits a periodic oscillation mode, causing the UPS output waveform to exhibit unstable oscillation phenomena on a micro time scale. This ultimately manifests as abnormal behaviors such as increased voltage distortion, a surge in harmonic components, and frequent system circulating current disturbances. This not only affects the power supply quality of the load but may also trigger the protection mechanism or cause the UPS to go offline. In severe cases, it will cause the entire parallel system to become unstable.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a phase-locked loop (PLL) method for parallel operation of digitally controlled uninterruptible power supplies (UPS). Through dynamic spectrum analysis and frequency domain coupling tensor modeling, it achieves accurate extraction of UPS modulation signal characteristics and dynamic assessment of resonance risk index. Combined with a risk-driven frequency intervention mechanism, it breaks frequency coupling, improves system frequency diversity and disturbance decoupling capability, significantly enhances the stability, anti-interference and intelligent response capability of the parallel system, and realizes early warning and active suppression of lock-point resonance, thereby solving the problems in the background technology mentioned above.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a digitally controlled uninterruptible power supply parallel operation phase-locked loop method, comprising the following steps:

[0007] Within each analysis cycle, the output voltage waveform signal and current frequency disturbance control parameters of each uninterruptible power supply are acquired, and short-time Fourier transforms are performed to convert the time-domain signal into a multi-dimensional dynamic spectrum matrix. At the same time, the frequency and amplitude parameters of the disturbance signal are extracted to construct the corresponding uninterruptible power supply disturbance parameter set.

[0008] In the dynamic spectrum matrix, the frequency band with structural response sensitivity is selected as the detection target frequency range. Based on the disturbance parameter set of each uninterruptible power supply, the spectral energy values ​​of its main modulation frequency component and its adjacent frequency band are extracted. The phase difference and amplitude superposition ratio between each main modulation component are calculated one by one, and a frequency domain coupling tensor is constructed based on the parameter change trajectory within the period.

[0009] By using the frequency domain coupling tensor, the energy density of each coupling frequency point in the current detection frequency band, the phase convergence between each main modulation frequency, and the aggregation of frequency intervals are mapped in the structural stability analysis space, thereby generating a set of multi-factor excitation vectors characterizing the perturbation coupling trend.

[0010] The set of multi-factor excitation vectors is normalized and loaded with a resonance risk amplification function trained based on historical operating samples and resonance precursor data to generate a lock-point resonance risk index, which is used to dynamically quantify the risk level of cooperative coupling and excitation resonance of frequency disturbance signals among multiple uninterruptible power supplies.

[0011] When the lock-point resonance risk index exceeds the preset reference threshold, an independently generated small-amplitude disturbance with random phase offset is applied to the frequency disturbance signal of each uninterruptible power supply, so that the modulation frequency of each uninterruptible power supply exhibits statistical uncorrelation, reducing the frequency coupling probability, actively disrupting the path that forms the lock-point resonance zone, and restoring the synchronous stability of parallel operation.

[0012] Preferably, the steps for acquiring the output voltage waveform signal and constructing the uninterruptible power supply disturbance parameter set in each analysis cycle are as follows:

[0013] Based on the local control processing unit of each uninterruptible power supply, the output voltage waveform signal is acquired at a fixed sampling frequency, and the frequency disturbance control parameters are extracted in the synchronous acquisition channel.

[0014] The acquired voltage waveform signal is divided into time windows of a preset length. After using a windowing function to improve the spectral leakage suppression effect, a short-time Fourier transform operation is performed on the data in each time window to obtain a three-dimensional dynamic spectrum matrix containing frequency, amplitude and time distribution.

[0015] Based on the fundamental frequency information recorded in the frequency disturbance control parameters, the corresponding frequency components are matched in the dynamic spectrum matrix, and the current modulation frequency and disturbance amplitude value of the disturbance signal are extracted by combining the instantaneous amplitude information of the corresponding frequency band. The disturbance parameter set of each power supply is constructed by combining the uninterruptible power supply identification number.

[0016] Preferably, the process of performing short-time Fourier transform on the acquired voltage waveform signal and constructing a three-dimensional dynamic spectrum matrix is ​​as follows:

[0017] The continuously acquired voltage waveform signal is divided into multiple overlapping sub-segments according to the sliding time window strategy. The length of each time window is dynamically set according to the target frequency resolution, and the overlap rate is adaptively adjusted according to the spectrum smoothness optimization criterion.

[0018] A Gaussian window function is applied to the signal sub-segment within each time window to perform windowing processing, thereby enhancing spectral accuracy and suppressing spectral leakage from interfering with adjacent frequency components, ensuring that frequency energy has accurate positioning capability in the transformed spectrum.

[0019] A fast Fourier transform operation is performed within each windowed time window to generate the frequency amplitude spectrum at the corresponding time point. The spectrum results from all time windows are then stacked and fused according to the time series to construct a complete three-dimensional dynamic spectrum matrix containing time, frequency, and amplitude information.

[0020] Preferably, the specific steps for constructing the frequency domain coupled tensor based on the parameter variation trajectory within the period are as follows:

[0021] Based on the set of sensitive frequency points defined in the structural vibration response model, the harmonic frequency band covering multiple fundamental frequency multiples is selected as the target detection interval. The main modulation frequency component and the spectral amplitude distribution curve of the adjacent frequency band are extracted from the dynamic spectrum matrix of each uninterruptible power supply.

[0022] By combining the modulation frequency index of the disturbance parameters of each uninterruptible power supply, phase alignment processing is performed on the extracted main modulation frequency components, and a spectrum curve interpolation algorithm is used to reconstruct the continuous energy transition curve between adjacent frequency points. The phase difference and amplitude superposition ratio of each uninterruptible power supply in the target frequency range are compared one by one to construct the initial feature matrix of frequency coupling.

[0023] The phase difference, amplitude superposition ratio, and their variation trajectory within a continuous analysis period are encoded in multiple dimensions and input into the tensor construction module. A frequency domain coupled tensor containing the cooperative variation mode of the disturbance source is formed through a high-order tensor structure.

[0024] Preferably, the specific steps for constructing the initial feature matrix of frequency coupling degree by performing phase alignment processing on the extracted main modulation frequency components, based on the modulation frequency index recorded in the disturbance parameters set of each uninterruptible power supply, are as follows:

[0025] Based on the modulation frequency index recorded in the disturbance parameter set of each uninterruptible power supply, the corresponding main modulation frequency component is extracted from the dynamic spectrum matrix of each power supply, and the main modulation frequency of different power supplies is phase aligned using a phase synchronization algorithm.

[0026] In the spectrum data of each uninterruptible power supply, neighboring frequency points within a fixed bandwidth range are selected with the main modulation frequency as the center, and the discrete amplitude data are reconstructed into a continuous curve using a spectrum curve interpolation algorithm to form energy transition characteristics.

[0027] Within the target frequency range, the main modulation frequency components and their reconstructed frequency bands of any two uninterruptible power supplies are compared one by one, and their phase difference and amplitude superposition ratio are calculated. All combined results are then summarized to form the initial characteristic matrix of frequency coupling.

[0028] Preferably, the specific steps for generating a set of multi-factor excitation vectors representing the perturbation coupling trend based on the frequency domain coupling tensor are as follows:

[0029] For the constructed frequency domain coupling tensor, the detection frequency band related to the harmonic resonance mechanism is selected in the structural stability analysis space. The amplitude information of the tensor in the energy channel dimension is extracted for each frequency coupling point, and the energy density index of the current frequency point in the whole spectrum is calculated.

[0030] In the phase channel dimension of the coupling tensor, the phase change trend between the main modulation frequency components is analyzed. By statistically analyzing the phase change gradient and synchronous convergence rate within its period, the phase convergence of the modulation signals between the UPS at the current moment is quantified.

[0031] The frequency distribution of the interval between the main modulation frequencies is evaluated in the frequency distribution dimension. The frequency aggregation degree is calculated by using the frequency interval dispersion function. Combined with the energy density index and the phase convergence degree, a set of excitation vectors containing multiple perturbation coupling dynamic factors is generated by a weighted fusion algorithm.

[0032] Preferably, for the multi-factor excitation vector set constructed in the previous stage, interval normalization processing is performed on the energy density index, phase convergence degree index and frequency aggregation degree index respectively;

[0033] The normalized excitation vector is input into the risk amplification function model generated by training with historical operating data and resonance precursor samples. The nonlinear mapping structure built inside the model is used to dynamically calculate the contribution of each excitation factor to the system resonance risk based on the current feature combination.

[0034] A weighted fusion operation is performed on the risk contribution of each dimension to output a lock-point resonance risk index with real-time continuous characteristics.

[0035] Preferably, when the lockpoint resonance risk index exceeds a preset reference threshold, the system enters the resonance evolution boundary state, triggering the intervention modulation mechanism. Based on the acquired lockpoint resonance risk index, a perturbation offset factor matrix D is constructed. t , Where N represents the number of uninterruptible power supply devices operating in parallel, δ i (t) represents the independent disturbance phase offset generated by the i-th uninterruptible power supply within period t, and is calculated using the following formula:

[0036]

[0037] In the formula, γ is the upper limit coefficient of disturbance intensity, and η i F is the individualized extended parameter used by the i-th uninterruptible power supply to control the frequency of disturbance phase change, and F is the lock-point resonance risk index. ref It is the reference threshold for the lock-point resonance risk index, φ i It is the initial phase offset constant of the disturbance, and 2π is the angle of the complete cycle, that is, 360° in radians;

[0038] Based on the acquired independent perturbation phase offset δ i (t), for each uninterruptible power supply, a new perturbation frequency is constructed, as shown in the following formula:

[0039]

[0040] In the formula, It is the new disturbance frequency that the i-th uninterruptible power supply finally applies within the current control cycle t. is the fundamental disturbance frequency used by the i-th uninterruptible power supply in the previous control cycle, and β is the disturbance gain amplification factor.

[0041] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0042] This invention introduces a multidimensional dynamic spectrum analysis and frequency domain coupling tensor modeling mechanism to achieve refined monitoring and feature extraction of the phase, amplitude, and frequency distribution of modulation signals among each UPS unit. Combined with a risk amplification function trained based on historical data, it dynamically calculates the lock-point resonance risk index, enabling early identification and quantitative assessment of potential system resonance states. Furthermore, by implementing statistically uncorrelated intervention disturbances on the modulation frequency, it breaks coupling links, actively repairs system frequency diversity, and decouples disturbance sources, effectively improving the stability, anti-interference capability, and power supply quality of the parallel system. This method not only enhances the system's ability to perceive frequency resonance risks but also possesses closed-loop response and real-time intervention characteristics, significantly improving the operational safety and intelligence level of the intelligent power system. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0044] Figure 1 This is a flowchart of the digital control uninterruptible power supply parallel operation phase-locked method of the present invention. Detailed Implementation

[0045] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0046] This invention provides, for example Figure 1 The digitally controlled uninterruptible power supply parallel operation phase-locked loop method shown includes the following steps:

[0047] Within each analysis cycle, the output voltage waveform signal and current frequency disturbance control parameters of each uninterruptible power supply are acquired, and short-time Fourier transforms are performed to convert the time-domain signal into a multi-dimensional dynamic spectrum matrix. At the same time, the frequency and amplitude parameters of the disturbance signal are extracted to construct the corresponding uninterruptible power supply disturbance parameter set.

[0048] The steps for acquiring the output voltage waveform signal and constructing the uninterruptible power supply disturbance parameter set in each analysis cycle are as follows:

[0049] Based on the local control processing unit of each uninterruptible power supply, the output voltage waveform signal is acquired at a fixed sampling frequency, and the frequency disturbance control parameters, including the fundamental frequency, amplitude and initial phase of the frequency disturbance signal, are extracted in the synchronous acquisition channel.

[0050] The acquired voltage waveform signal is divided into time windows of a preset length. After using a windowing function to improve the spectral leakage suppression effect, a short-time Fourier transform operation is performed on the data in each time window to obtain a three-dimensional dynamic spectrum matrix containing frequency, amplitude and time distribution.

[0051] Based on the fundamental frequency information recorded in the frequency disturbance control parameters, the corresponding frequency components are matched in the dynamic spectrum matrix, and the current modulation frequency and disturbance amplitude value of the disturbance signal are extracted by combining the instantaneous amplitude information of the corresponding frequency band. The disturbance parameter set of each power supply is constructed by combining the uninterruptible power supply identification number for subsequent frequency domain coupling analysis.

[0052] During dynamic spectrum analysis, each uninterruptible power supply (UPS) records the injected frequency disturbance control parameters within its control cycle, including the fundamental frequency (modulation frequency) as the main characteristic frequency of the disturbance signal. Based on this fundamental frequency information, the frequency component containing the fundamental frequency can be located along the frequency axis in the three-dimensional dynamic spectrum matrix corresponding to the power supply. Subsequently, the instantaneous amplitude of this frequency point is extracted in the time slice corresponding to this position as an estimate of the current modulation amplitude of the disturbance signal. If the energy intensity of this frequency point diffuses within a neighboring frequency band, a weighted average of amplitudes within a certain bandwidth can be combined to improve the stability of the amplitude estimation. Through the above frequency matching and amplitude extraction operations, combined with the unique identifier of the power supply itself (such as device address or control node ID), a set of disturbance parameters for each UPS in the current cycle can be constructed. This set typically includes fields such as modulation frequency, modulation amplitude, timestamp, and device number. The purpose of this process is to provide clear disturbance source information for subsequent frequency domain coupling analysis, enabling the system to accurately assess the frequency disturbance interaction relationship between each UPS in a multi-unit operation scenario, thereby identifying potential lock-in resonance risks.

[0053] In each analysis cycle, the output voltage waveform signals of each uninterruptible power supply (UPS) are acquired, and a short-time Fourier transform is performed in conjunction with its current frequency disturbance control parameters. The aim is to transform the original time-domain voltage signals into a multi-dimensional dynamic spectrum matrix with more observable and quantifiable analytical characteristics, providing a data foundation for subsequent identification of frequency coupling risks and resonance indicators. The dynamic spectrum matrix not only preserves the evolution of the signal on the time axis but also shows its energy distribution characteristics on the frequency axis, making it particularly suitable for capturing the spectral behavior of short-time disturbance signals. When combined with the fundamental frequency information recorded in the disturbance control parameters, the frequency components of the disturbance currently applied by each UPS can be accurately located in the spectrum matrix, and the actual modulation amplitude of the disturbance signal can be obtained by extracting the spectral amplitude, further reconstructing the frequency disturbance characteristics of each device in that cycle. By combining these disturbance frequencies, disturbance amplitudes, and UPS device numbers, the constructed disturbance parameter set not only possesses real-time performance and accuracy but also provides a structured input for subsequent frequency domain coupling analysis, enabling the system to comprehensively assess whether there are frequency synchronization risks, harmonic resonance possibilities, or coupling trends between disturbance signals among multiple UPS units. This process is the foundation of the entire lockpoint resonance risk monitoring process, and its accuracy and completeness directly determine the reliability of subsequent risk index assessment and control strategy decisions.

[0054] The process of performing short-time Fourier transform on the acquired voltage waveform signal and constructing a three-dimensional dynamic spectrum matrix is ​​as follows:

[0055] The continuously acquired voltage waveform signal is divided into multiple overlapping sub-segments according to the sliding time window strategy. The length of each time window is dynamically set according to the target frequency resolution, and the overlap rate is adaptively adjusted according to the spectrum smoothness optimization criterion.

[0056] A Gaussian window function is applied to the signal sub-segment within each time window to perform windowing processing, thereby enhancing spectral accuracy and suppressing spectral leakage from interfering with adjacent frequency components, ensuring that frequency energy has accurate positioning capability in the transformed spectrum.

[0057] Fast Fourier Transform is performed within each windowed time window to generate the frequency amplitude spectrum at the corresponding time point. The spectral results from all time windows are then stacked and fused according to the time series to construct a complete three-dimensional dynamic spectrum matrix containing time, frequency, and amplitude information, providing basic data support for subsequent frequency domain coupled tensor analysis.

[0058] A windowing function is a mathematical function applied to a finite-length signal segment before frequency domain analysis (such as Fourier transform) to control edge effects and suppress spectral leakage. Common windowing functions include the Hamming window, Hanning window, Blackman window, Gaussian window, and Kaiser window. These functions apply smaller weights at the ends of the time window, allowing the signal to attenuate smoothly at the edges, avoiding abrupt changes caused by signal truncation, and thus reducing energy diffusion in the frequency domain.

[0059] In the short-time Fourier transform, the original signal is divided into time windows of finite length. Direct transformation is equivalent to "rectangular truncation" of the signal. This abrupt interruption introduces discontinuities, causing leakage in the signal spectrum at non-true frequency locations (i.e., energy diffuses to neighboring frequencies). By using a windowing function to smoothly transition the signal at both ends of the time window to values ​​close to zero, the energy distribution at non-target frequency points in the spectrum can be significantly reduced, improving frequency resolution and spectral clarity. This makes the main frequency components more concentrated and identifiable in the spectrum, thereby enhancing the accuracy and robustness of frequency domain coupling analysis.

[0060] In the dynamic spectrum matrix, the frequency band with structural response sensitivity is selected as the detection target frequency range. Based on the disturbance parameter set of each uninterruptible power supply, the spectral energy values ​​of its main modulation frequency component and its adjacent frequency band are extracted. The phase difference and amplitude superposition ratio between each main modulation component are calculated one by one, and a frequency domain coupling tensor is constructed based on the parameter change trajectory within the period.

[0061] The specific steps for constructing a frequency-domain coupled tensor based on the trajectory of parameter changes within a period are as follows:

[0062] Based on the set of sensitive frequency points defined in the structural vibration response model, the harmonic frequency band covering multiple fundamental frequency multiples is selected as the target detection interval. The main modulation frequency component and the spectral amplitude distribution curve of the adjacent frequency band are extracted from the dynamic spectrum matrix of each uninterruptible power supply to provide target frequency domain data for subsequent feature analysis.

[0063] By combining the modulation frequency index of the disturbance parameters of each uninterruptible power supply, phase alignment processing is performed on the extracted main modulation frequency components, and a spectrum curve interpolation algorithm is used to reconstruct the continuous energy transition curve between adjacent frequency points. The phase difference and amplitude superposition ratio of each uninterruptible power supply in the target frequency range are compared one by one to construct the initial feature matrix of frequency coupling.

[0064] The phase difference, amplitude superposition ratio, and their variation trajectory within a continuous analysis period are encoded in multiple dimensions and input into the tensor construction module. A frequency domain coupling tensor containing the cooperative variation mode of the disturbance source is formed through a high-order tensor structure. This tensor reflects the harmonic interaction intensity and coupling trend of multi-source disturbances in the structure-sensitive frequency band, providing high-dimensional dynamic feature support for subsequent lockpoint resonance risk assessment.

[0065] In frequency domain analysis, the spectral amplitude distribution curve is a continuous curve describing the trend of amplitude intensity changes of a signal at different frequency points. It reflects the energy distribution of the signal's frequency components in the spectrum. In the spectrum generated by the short-time Fourier transform, the spectral amplitude distribution curve is typically plotted within a specific time window, with frequency on the horizontal axis and amplitude on the vertical axis. This curve can be used to identify the dominant frequency components, harmonic components, and energy diffusion characteristics of adjacent frequency bands. Its function is to provide a clear and continuous frequency response picture, which helps in analyzing whether the modulated signal is concentrated in a certain frequency band, whether there are multi-source coupling characteristics, and whether frequency components are extended or overlapped. Especially in frequency domain coupling analysis of multiple uninterruptible power supplies operating in parallel, the spectral amplitude distribution curve is not only used to identify the location of the dominant modulation frequency, but also to calculate the amplitude superposition ratio and analyze the energy similarity between spectra, thus providing core data for the construction of frequency coupling tensors and the identification of resonance trends.

[0066] The trajectory of change within a continuous analysis period refers to the dynamic change of key parameters such as the phase difference and amplitude superposition ratio of the modulation signals of each uninterruptible power supply (UPS) within the target frequency range over multiple adjacent analysis periods. This trajectory is obtained by repeatedly performing spectrum extraction and coupling feature calculation operations for each analysis period, recording the main modulation frequency position, phase value, and amplitude distribution of each UPS in each period, and performing time-series tracking of the phase difference and amplitude ratio between any two UPSs, thus forming a characteristic sequence that evolves with the period. The purpose of this trajectory is to capture the cooperative trend and coupling stability between multi-source disturbances, such as whether the phase difference tends to converge or whether the amplitude ratio periodically amplifies. It can serve as a key basis for judging whether the system is entering a frequency coupling critical state or a potential resonance boundary, providing dynamic time-dimensional feature input for subsequent tensor construction, and supporting the predictive identification of lockpoint resonance risks and the formulation of response strategies.

[0067] This step aims to establish a multi-dimensional frequency domain coupling model to quantify the mutual influence of frequency disturbance signals among multiple uninterruptible power supplies (UPS), providing a crucial data foundation for identifying potential harmonic coupling trends and lock-in resonance risks within the system. Specifically, by selecting structurally sensitive harmonic frequency bands in the dynamic spectrum matrix as detection targets, the model effectively focuses on frequency regions with high coupling risks to system stability, thereby improving the targeting and accuracy of the analysis. Based on the disturbance parameter set of each UPS, the model extracts its main modulation frequency component and the spectral energy values ​​of its adjacent frequency bands, enabling the reconstruction of the current disturbance behavior and energy distribution characteristics of each power supply at the frequency domain level. By comparing the main modulation frequencies of any two UPSs pairwise and calculating their phase difference and amplitude superposition ratio, the model reflects the synergy and interference potential of each frequency signal in the sensitive frequency band. Furthermore, by multi-dimensionally encoding these phase differences, amplitude ratios, and their time-varying trajectories, a frequency domain coupling tensor is constructed, which not only describes the frequency coupling strength at a certain moment but also captures its dynamic changes over multiple analysis periods. This tensor structure possesses high information compression and expressive power, providing mathematically complete and data-driven high-dimensional support for the subsequent generation of the lock-point resonance risk index and the triggering of system control strategies. It has core value in frequency domain collaborative modeling of multi-machine parallel systems.

[0068] The specific steps for constructing the initial feature matrix of frequency coupling degree by combining the modulation frequency index recorded in the disturbance parameters set of each uninterruptible power supply with the extracted main modulation frequency components for phase alignment processing are as follows:

[0069] Based on the modulation frequency index recorded in the disturbance parameter set of each uninterruptible power supply, the corresponding main modulation frequency component is extracted from the dynamic spectrum matrix of each power supply, and the main modulation frequency of different power supplies is phase aligned using a phase synchronization algorithm to ensure that subsequent comparisons are performed under a unified phase reference.

[0070] In the spectrum data of each uninterruptible power supply, neighboring frequency points within a fixed bandwidth range are selected with the main modulation frequency as the center. The discrete amplitude data are reconstructed into a continuous curve using a spectrum curve interpolation algorithm to form energy transition characteristics, thereby enhancing the analytical accuracy and comparability of the frequency amplitude distribution.

[0071] Within the target frequency range, the main modulation frequency components and their reconstructed frequency bands of any two uninterruptible power supplies are compared one by one. The phase difference and amplitude superposition ratio are calculated, and all combination results are summarized to form the initial characteristic matrix of frequency coupling, which provides the basic coupling characteristic structure for subsequent multi-cycle dynamic coupling trend modeling and tensor analysis.

[0072] Phase synchronization algorithms are used to eliminate initial phase differences in signals when comparing or fusing multiple frequency components in multi-signal analysis. The implementation process typically includes: first, extracting the phase angle value of the frequency component based on the main modulation frequency of each uninterruptible power supply; then, selecting a reference power supply as the phase reference and converting the phase values ​​of other power supplies into relative phase differences relative to this reference; finally, aligning the frequency components through phase rotation or numerical offset to ensure a unified phase reference framework for the modulation frequencies of different power supplies in frequency domain analysis. Commonly used phase synchronization algorithms include the "Hilbert transform phase demodulation method," the "phase-locked loop (PLL) algorithm," or the "short-time Fourier transform-based phase envelope calibration method." The purpose of this algorithm is to eliminate coupling errors caused by initial phase differences, ensuring that the phase difference value in the frequency coupling calculation process truly reflects the dynamic relationship between frequencies.

[0073] Spectrum curve interpolation algorithms refer to the interpolation and reconstruction of a finite number of discretely distributed frequency amplitude points in spectrum analysis, thereby forming a continuous and smooth spectral amplitude curve to more accurately represent the distribution and transition characteristics of signal energy in the frequency space. The implementation typically involves using mathematical methods such as cubic spline interpolation, B-spline fitting, Lagrange interpolation, or local polynomial fitting, based on the main modulation frequency point and several adjacent frequency sampling points, to construct a continuous curve passing through these points, filling the energy gaps between discrete frequency spectrum points. The algorithm improves the resolution and continuity of the spectrum distribution, allowing the calculation of amplitude superposition ratios to move beyond single-point comparisons and comprehensively consider energy variation trends within the frequency band, thus more accurately identifying harmonic interference effects caused by frequency coupling. This type of interpolation method is widely used in digital signal processing, communication spectrum analysis, and power system harmonic detection, and belongs to a mature existing algorithm system.

[0074] The core function of this step is to extract and quantify the mutual coupling relationship between frequency disturbance signals of multiple uninterruptible power supplies (UPS) within a specific frequency region, constructing an initial characteristic matrix of frequency coupling with physical meaning and mathematical expressive power. This provides crucial input for further modeling frequency domain resonance risk and system stability assessment. First, by using the modulation frequency index recorded in the centralized data of disturbance parameters for each UPS, the main modulation frequency component of each UPS in the dynamic spectrum matrix can be accurately located, thus avoiding analysis errors caused by spectrum ambiguity or multi-frequency interference. Subsequently, through phase alignment processing, the phases of the main frequency components of different UPSs are unified to the same reference frame, eliminating the pseudo-coupling effect caused by initial phase differences and ensuring that the physical meaning of the phase difference truly reflects the synchronization offset state between signals. Furthermore, by using a spectrum curve interpolation algorithm to continuously reconstruct the discrete amplitude data between the main modulation frequency and its neighboring frequency points, a transition curve reflecting the changes in frequency energy distribution can be formed. This allows the amplitude characteristics to not only be limited to single-point amplitude comparisons but also to integrate the energy evolution trend throughout the entire frequency band. Finally, by comparing the main modulation frequency components between different UPSs in pairs, the phase difference and amplitude superposition ratio are extracted to form a basic set of frequency coupling indicators, which are then organized into an initial feature matrix. This matrix structurally records the possible coupling paths and coupling strengths between UPSs, possesses high time-frequency resolution, and can provide direct data support and a basis for judgment for subsequent high-dimensional coupling tensor modeling, lock-point resonance risk index calculation, and disturbance intervention mechanisms. It is a key intermediate link in the entire multi-source frequency domain collaborative analysis system.

[0075] By using the frequency domain coupling tensor, the energy density of each coupling frequency point in the current detection frequency band, the phase convergence between each main modulation frequency, and the aggregation of frequency intervals are mapped in the structural stability analysis space, thereby generating a set of multi-factor excitation vectors characterizing the perturbation coupling trend.

[0076] The specific steps for generating a set of multi-factor excitation vectors representing the coupling trend of perturbations based on frequency domain coupling tensors are as follows:

[0077] For the constructed frequency domain coupling tensor, the detection frequency band related to the harmonic resonance mechanism is selected in the structural stability analysis space. The amplitude information of the tensor in the energy channel dimension is extracted for each frequency coupling point, and the energy density index of the current frequency point in the whole spectrum is calculated to characterize the degree of disturbance energy accumulation.

[0078] In the phase channel dimension of the coupling tensor, the phase change trend between the main modulation frequency components is analyzed. By statistically analyzing the phase change gradient and synchronous convergence rate within its period, the phase convergence of the modulation signals between the UPS at the current moment is quantified and used as a dynamic factor to measure resonance.

[0079] The frequency distribution of the interval between the main modulation frequencies is evaluated in the frequency distribution dimension. The frequency interval dispersion function is used to measure the degree of frequency aggregation. Combined with the energy density index and the phase convergence degree, a set of excitation vectors containing multiple perturbation coupling dynamic factors is generated by a weighted fusion algorithm. This set is used as the input of the dynamic evaluation model of the subsequent lock point resonance risk index, so as to realize the early identification of the system resonance trend and the linkage of the response mechanism.

[0080] The frequency spacing dispersion function is a mathematical function used to quantify the density of the distribution among multiple frequency points. Its core purpose is to assess whether these frequencies are clustered in a certain frequency band in the spectrum, reflecting the concentration trend of the modulated signal in the frequency domain. This function is usually implemented based on the principle of statistical distribution, such as using methods like the variance of frequency gauges, the mean nearest neighbor distance (MNND), or the frequency entropy-based aggregation metric. The implementation process is as follows: First, all main modulation frequencies are sorted in ascending order, and the difference vector between adjacent frequencies is calculated; then, variance calculation or standardized entropy calculation is performed on this difference vector to obtain a quantized value of the frequency spacing dispersion. The smaller the value, the denser the frequency distribution and the higher the degree of aggregation. This indicator can reflect whether multiple UPSs form a potential resonance region in frequency modulation and is an important basis for identifying common-mode modulation trends and system instability boundaries. When the discreteness function, along with the energy density index and the phase convergence degree, is used as input, a set of excitation vectors containing multi-dimensional features of frequency coupling trends can be formed through a weighted fusion algorithm, providing dynamic and structured input features for the resonance risk index.

[0081] This step aims to structurally extract and compress the abundant time-frequency coupling information contained in the frequency domain coupling tensor, thereby generating a high-dimensional feature vector set that reflects the coupling trend of disturbance signals between multiple uninterruptible power supplies (UPSs). This provides a calculable and traceable multi-factor input basis for the quantitative assessment of resonance risk. In a multi-UPS system, each device may inject disturbance signals with slightly different amplitude, frequency, and phase characteristics. These disturbances may form local energy superposition or phase coordination in specific frequency bands in the frequency domain, leading to frequency resonance or system instability. Therefore, by scanning the detection frequency band in the structural stability analysis space using the frequency domain coupling tensor, the energy density index of each frequency coupling point can be extracted to measure the degree of local energy accumulation at that frequency point in the entire spectrum. Simultaneously, the phase convergence between different main modulation frequencies can be analyzed, i.e., whether the signals tend towards synchronization during dynamic evolution. Furthermore, the aggregation degree of the main modulation frequencies is evaluated using the frequency interval dispersion function to determine whether the disturbance frequencies are concentrated in a narrow frequency band. When the values ​​of the above three characteristic dimensions collectively exhibit a trend of high energy, low phase difference, and frequency convergence, it can be inferred that the system may be entering a critical state of frequency resonance. Constructing these three types of indicators into a multi-factor excitation vector set through a weighted fusion strategy not only characterizes the strength and evolution trend of the current disturbance coupling but also possesses mathematical continuity and physical rationality as input variables for the subsequent lock-point resonance risk index, thus providing strong data support for early warning and active control of the system.

[0082] The set of multi-factor excitation vectors is normalized and loaded with a resonance risk amplification function trained based on historical operating samples and resonance precursor data to generate a lock-point resonance risk index, which is used to dynamically quantify the risk level of cooperative coupling and excitation resonance of frequency disturbance signals among multiple uninterruptible power supplies.

[0083] The specific steps for generating the lockpoint resonance risk index are as follows: Normalize the set of multi-factor excitation vectors and load the resonance risk amplification function trained based on historical running samples and resonance precursor data.

[0084] For the multi-factor excitation vector set constructed in the previous stage, interval normalization processing is performed on the energy density index, phase convergence degree index and frequency aggregation degree index respectively, so that all feature dimensions are uniformly mapped to the preset standard value range, eliminating the weight offset caused by different physical dimensions, and ensuring the numerical stability and feature fairness of subsequent model calculations.

[0085] This step unifies the numerical range of each activating factor, eliminates differences in physical dimensions, improves the fairness of the weights of different features in the calculation and the stability of the model, and provides clean and standardized data input for subsequent risk modeling.

[0086] The normalized excitation vector is input into the risk amplification function model trained from historical operating data and resonance precursor samples. Using the nonlinear mapping structure built inside the model, the contribution of each excitation factor to the system resonance risk is dynamically calculated based on the current feature combination. Based on the integrated decision mechanism, the risk effects of all dimensions are integrated to generate the overall risk response under the current system state.

[0087] This step evaluates the specific contribution of each excitation factor to the system's resonance risk through the trained nonlinear model, realizing the transformation from static features to dynamic risk prediction and enhancing the accuracy and sensitivity of identifying potential resonance trends.

[0088] A weighted fusion operation is performed on the risk contribution of each dimension to output a lock-point resonance risk index with real-time continuous characteristics. This index can accurately characterize the probability level of multiple uninterruptible power supplies co-coupled and stimulate structural resonance under the current frequency disturbance conditions, providing a direct quantitative judgment basis for the active triggering of system control strategies, the loading of circulating current suppression strategies, and the operation early warning mechanism.

[0089] This step integrates the risk contributions of each factor and outputs a real-time updated, continuously quantified lockpoint resonance risk value, providing a clear decision-making basis for the system's automatic regulation, early warning response, and disturbance suppression.

[0090] The "Risk Amplification Function Model Trained from Historical Operating Data and Resonance Precursor Samples" refers to a function model that reflects the nonlinear relationship between disturbance coupling characteristics and system resonance risk. This model is constructed using machine learning or data-driven modeling methods, leveraging operational monitoring data accumulated over long-term operation of multiple uninterruptible power supply (UPS) systems and samples of precursory behaviors before system resonance. The core function of this model is to map multiple disturbance characteristics (such as energy density, phase convergence, and frequency convergence) to the risk amplification level of system resonance. During the model training phase, firstly, a large amount of historical operating data is labeled and features extracted to identify the characteristic state distribution before resonance precursor events such as abnormal oscillations, harmonic increases, and voltage distortion. Then, using these precursor samples as positive classes and normal operating data as negative classes, machine learning algorithms such as neural networks, support vector machines (SVM), gradient boosting trees (GBDT), or deep regression networks are employed to construct a learnable nonlinear function model. Cross-validation and hyperparameter optimization ensure the model's generalization ability and robustness.

[0091] During operation, the risk amplification function model, serving as the core engine of the system risk calculation module, receives the normalized excitation vector at each moment and performs joint analysis on its perturbation characteristics across various dimensions. The model's internal nonlinear structure can output the strength of each factor's influence on the potential resonance trend under the current system state—its risk contribution—based on the complex interactions between features. Unlike traditional linear weighted models, the risk amplification function not only considers the independent influence of each factor but also explores their combination patterns, nonlinear synergies, and boundary effects. For example, when individual UPS systems have low phase convergence but significant frequency aggregation, the model may output a high risk because such feature combinations frequently appear in precursor samples. In this way, the model can dynamically assess whether the system is in a sensitive state near the resonance boundary, thus providing highly sensitive and discriminative data support for generating the lock-point resonance risk index, enabling the overall system to have early warning, autonomous perception, and response control capabilities. This mechanism not only improves the safety and stability of UPS systems operating in parallel but also provides a theoretical basis and practical path for the fault tolerance and self-healing of intelligent power systems.

[0092] The core function of this step is to transform complex frequency domain coupling characteristics into a risk indicator that can be monitored in real time and quantified for assessment. This dynamically reflects whether there is a cooperative coupling trend among multiple uninterruptible power supplies (UPS) during frequency disturbances, and whether this may trigger lock-in resonance. The multi-factor excitation vector set generated in the previous stage contains key feature dimensions reflecting the system state, such as frequency coupling energy density, phase convergence of modulation signals, and frequency aggregation. These features come from different sources and have different units. If directly used for risk assessment, inconsistencies in units and numerical scales will lead to misjudgments or model instability. Therefore, normalization is used to standardize all feature values ​​to the same numerical range, improving data comparability and enhancing the numerical adaptability of the model.

[0093] Subsequently, the most intelligent part of this step involves loading a risk amplification function model trained based on historical operating data and resonance precursor samples. This model establishes a nonlinear mapping relationship between excitation features and actual resonance events through machine learning algorithms. It can identify the dynamic impact of complex feature combinations on the system's resonance probability, thereby outputting the risk contribution of each feature in the current state. Finally, these contributions are weighted and fused to generate a continuous lockpoint resonance risk index. This index not only tracks whether the system is on the edge of resonance in real time but also has predictive capabilities, providing early warnings to the system control module to initiate intervention measures (such as dispersing dither frequencies and suppressing circulating channels). Therefore, this step essentially completes the transformation from feature perception to risk identification, forming the decision-making basis for the entire lockpoint resonance monitoring and prevention mechanism, and significantly improving the operational reliability and intelligence level of multi-machine parallel UPS systems.

[0094] When the lock-point resonance risk index exceeds the preset reference threshold, an independently generated small-amplitude disturbance with random phase offset is applied to the frequency disturbance signal of each uninterruptible power supply, so that the modulation frequency of each uninterruptible power supply is statistically uncorrelated, reducing the frequency coupling probability, actively destroying the path that forms the lock-point resonance zone, and restoring the synchronous stability of parallel operation.

[0095] When the lockpoint resonance risk index exceeds a preset reference threshold, the system enters the resonance evolution boundary state, triggering the intervention modulation mechanism. Based on the acquired lockpoint resonance risk index, a perturbation offset factor matrix D is constructed. t , Where N represents the number of uninterruptible power supply devices operating in parallel, δ i (t) represents the independent disturbance phase offset generated by the i-th uninterruptible power supply within period t, and is calculated using the following formula:

[0096]

[0097] In the formula, γ is the upper limit coefficient of the disturbance intensity, used to limit the maximum amplitude range of the final offset value, and its value is set between 0.01 and 1.0. η i It is the individualized extended parameter used by the i-th uninterruptible power supply to control the frequency of the disturbance phase change, through different η i Each UPS is assigned a different rate of change of disturbance frequency, giving its disturbance signal non-uniform oscillation characteristics, thereby enhancing the dispersion and asynchrony of the system disturbance spectrum. F is the lock-in resonance risk index, ranging from 0 to 1. ref It is the reference threshold for the lock-point resonance risk index, φ i It is the disturbance initial phase offset constant, which represents the unique disturbance initial phase offset value assigned to i uninterruptible power supplies during each startup or control initialization. It is usually distributed in the interval [0, 2π], where 2π is the angle of the complete cycle, i.e., 360° in radians.

[0098] The trigonometric function sin(·) in this step is used to construct a continuous, smooth, and periodically changing disturbance pattern to drive asynchronous phase shift behavior in each uninterruptible power supply (UPS) during frequency modulation. By introducing variables such as the risk index, frequency spread factor, and initial phase offset as inputs into the sine function, these linear or nonlinear factors can be mapped into a dynamic disturbance signal in radians, giving the disturbance changes periodic fluctuation characteristics and thus avoiding control instability caused by random jumps. The numerical range of the sine function itself is within [-1, 1], giving its output a natural boundary and controllability. After linear transformation (such as adding 1 and multiplying by the disturbance amplitude coefficient γ), the output can be modulated into an adjustable disturbance offset within the positive range. Therefore, the introduction of sin(·) not only gives the disturbance signal regularity and predictability but also mathematically guarantees the boundedness of the disturbance amplitude, which helps the system achieve flexible control and phase decoupling in resonance risk response.

[0099] This perturbation offset factor is designed to introduce phase randomness among frequency perturbation signals from multiple uninterruptible power supplies, enhance the statistical non-correlation between perturbation frequencies, and provide input basis for the next step of frequency modulation decoupling and reconstruction.

[0100] Based on the acquired independent perturbation phase offset δ i (t), for each uninterruptible power supply, a new perturbation frequency is constructed, as shown in the following formula:

[0101]

[0102] In the formula, The new disturbance frequency ultimately applied by the i-th uninterruptible power supply within the current control cycle t is output to the control system as a frequency modulation signal, replacing the frequency of the previous cycle. This ensures that when system risk increases, the modulation frequency automatically deviates from its original value, breaking the frequency coupling path and achieving dynamic decoupling of the disturbance frequency. is the base disturbance frequency used by the i-th uninterruptible power supply in the previous control cycle, serving as the reference starting point for disturbance frequency modulation in this cycle, ensuring that the disturbance is offset in a controlled manner above and below the base frequency. β is the disturbance gain amplification factor, used to control the influence weight of the risk index in the disturbance frequency reconstruction, enhancing the system's response strength to changes in the risk index. When the system risk index increases, the disturbance frequency offset amplitude is amplified, enabling the system to quickly escape the possible lock-point resonance region.

[0103] The above-mentioned perturbation frequency construction logic implements the risk index F and its reference threshold F ref The dynamic response modulation of the ratio between them makes the disturbance frequency generated under high-risk conditions more phase-dispersion and frequency-shift characteristics, thereby effectively weakening the coupling tendency of multiple uninterruptible power supplies in the modulation frequency. Finally, the reconstructed The frequency modulation controllers, which are respectively loaded onto each uninterruptible power supply, realize the dynamic decoupling and non-common mode distribution of the disturbance signal, destroy the frequency coupling path on which the lock point resonance region depends, thereby restoring the frequency diversity of the system and enhancing the structural stability and operational robustness of parallel operation.

[0104] This step aims to proactively intervene in the frequency disturbance modulation process of the UPS parallel system after identifying a potential risk of lock-in resonance. By injecting small-amplitude disturbances with random phase shifts, it breaks the frequency coupling path formed between multiple UPS units due to their highly consistent frequency, phase, and amplitude, thereby relieving the system from a potential "lock-in resonance zone." In traditional parallel systems, if each UPS uses similar frequency disturbance signals, the synchronization mechanism tends to unify them, and their frequency modulation components are prone to coupling and superimposing at specific harmonic frequencies, forming a hidden resonance loop. This step introduces modulation disturbances with randomness and individual differences, making the frequency characteristics of each UPS statistically uncorrelated, thereby increasing the spectral dispersion and phase dispersion between disturbance signals and effectively reducing the probability of the system forming stable resonance conditions in the frequency domain.

[0105] More importantly, this disturbance process is not a fixed-parameter intervention, but a dynamic response process driven by the lock-point resonance risk index. When the system risk increases, the disturbance amplitude automatically increases; after the risk decreases, the disturbance returns to stability, thus constructing an adaptive intervention closed-loop mechanism. This mechanism not only disrupts frequency synchronization and breaks coupling paths at the micro level, but also restores the system's frequency heterogeneity and response elasticity at the macro level, avoiding the risk of small disturbances accumulating and evolving into large-scale resonance instability, and significantly improving the stability, fault tolerance, and operational reliability of the UPS parallel system.

[0106] The aforementioned digitally controlled phase-locked loop (PLL) method for parallel operation of uninterruptible power supplies (UPS) effectively solves the lock-in resonance problem caused by excessive similarity in frequency disturbance signals during parallel operation of multiple UPS units in existing technologies. This method introduces multi-dimensional dynamic spectrum analysis and frequency domain coupling tensor modeling to achieve refined monitoring and feature extraction of the phase, amplitude, and frequency distribution of modulation signals among each UPS unit. Combined with a risk amplification function trained based on historical data, it dynamically calculates the lock-in resonance risk index, enabling early identification and quantitative assessment of potential resonance states in the system. Furthermore, by implementing statistically uncorrelated intervention disturbances on the modulation frequency, it breaks the coupling link, actively repairs the frequency diversity of the system, and decouples the disturbance sources, effectively improving the stability, anti-interference capability, and power supply quality of the parallel system. This method not only enhances the system's ability to perceive frequency resonance risks but also possesses closed-loop response and real-time intervention characteristics, significantly improving the operational safety and intelligence level of intelligent power systems.

[0107] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0108] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0109] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0110] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0111] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0112] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0113] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0115] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0116] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A digitally controlled uninterruptible power supply parallel operation phase-locked loop method, characterized in that, Includes the following steps: Within each analysis cycle, the output voltage waveform signal and current frequency disturbance control parameters of each uninterruptible power supply are acquired, and short-time Fourier transforms are performed to convert the time-domain signal into a multi-dimensional dynamic spectrum matrix. At the same time, the frequency and amplitude parameters of the disturbance signal are extracted to construct the corresponding uninterruptible power supply disturbance parameter set. In the dynamic spectrum matrix, the frequency band with structural response sensitivity is selected as the detection target frequency range. Based on the disturbance parameter set of each uninterruptible power supply, the spectral energy values ​​of its main modulation frequency component and its adjacent frequency band are extracted. The phase difference and amplitude superposition ratio between each main modulation component are calculated one by one, and a frequency domain coupling tensor is constructed based on the parameter change trajectory within the period. By using the frequency domain coupling tensor, the energy density of each coupling frequency point in the current detection frequency band, the phase convergence between each main modulation frequency, and the aggregation of frequency intervals are mapped in the structural stability analysis space, thereby generating a set of multi-factor excitation vectors characterizing the perturbation coupling trend. The set of multi-factor excitation vectors is normalized and loaded with a resonance risk amplification function trained based on historical operating samples and resonance precursor data to generate a lock-point resonance risk index, which is used to dynamically quantify the risk level of cooperative coupling and excitation resonance of frequency disturbance signals among multiple uninterruptible power supplies. When the lock-point resonance risk index exceeds the preset reference threshold, an independently generated small-amplitude disturbance with random phase offset is applied to the frequency disturbance signal of each uninterruptible power supply, so that the modulation frequency of each uninterruptible power supply exhibits statistical uncorrelation, reducing the frequency coupling probability, actively disrupting the path that forms the lock-point resonance zone, and restoring the synchronous stability of parallel operation.

2. The digitally controlled uninterruptible power supply parallel operation phase-locked method according to claim 1, characterized in that, The steps for acquiring the output voltage waveform signal and constructing the uninterruptible power supply disturbance parameter set in each analysis cycle are as follows: Based on the local control processing unit of each uninterruptible power supply, the output voltage waveform signal is acquired at a fixed sampling frequency, and the frequency disturbance control parameters are extracted in the synchronous acquisition channel. The acquired voltage waveform signal is divided into time windows of a preset length. After using a windowing function to improve the spectral leakage suppression effect, a short-time Fourier transform operation is performed on the data in each time window to obtain a three-dimensional dynamic spectrum matrix containing frequency, amplitude and time distribution. Based on the fundamental frequency information recorded in the frequency disturbance control parameters, the corresponding frequency components are matched in the dynamic spectrum matrix, and the current modulation frequency and disturbance amplitude value of the disturbance signal are extracted by combining the instantaneous amplitude information of the corresponding frequency band. The disturbance parameter set of each power supply is constructed by combining the uninterruptible power supply identification number.

3. The digitally controlled uninterruptible power supply parallel operation phase-locked method according to claim 2, characterized in that, The process of performing short-time Fourier transform on the acquired voltage waveform signal and constructing a three-dimensional dynamic spectrum matrix is ​​as follows: The continuously acquired voltage waveform signal is divided into multiple overlapping sub-segments according to the sliding time window strategy. The length of each time window is dynamically set according to the target frequency resolution, and the overlap rate is adaptively adjusted according to the spectrum smoothness optimization criterion. A Gaussian window function is applied to the signal sub-segment within each time window to perform windowing processing, thereby enhancing spectral accuracy and suppressing spectral leakage from interfering with adjacent frequency components, ensuring that frequency energy has accurate positioning capability in the transformed spectrum. A fast Fourier transform operation is performed within each windowed time window to generate the frequency amplitude spectrum at the corresponding time point. The spectrum results from all time windows are then stacked and fused according to the time series to construct a complete three-dimensional dynamic spectrum matrix containing time, frequency, and amplitude information.

4. The digitally controlled uninterruptible power supply parallel operation phase-locked method according to claim 1, characterized in that, The specific steps for constructing a frequency-domain coupled tensor based on the trajectory of parameter changes within a period are as follows: Based on the set of sensitive frequency points defined in the structural vibration response model, the harmonic frequency band covering multiple fundamental frequency multiples is selected as the target detection interval. The main modulation frequency component and the spectral amplitude distribution curve of the adjacent frequency band are extracted from the dynamic spectrum matrix of each uninterruptible power supply. By combining the modulation frequency index of the disturbance parameters of each uninterruptible power supply, phase alignment processing is performed on the extracted main modulation frequency components, and a spectrum curve interpolation algorithm is used to reconstruct the continuous energy transition curve between adjacent frequency points. The phase difference and amplitude superposition ratio of each uninterruptible power supply in the target frequency range are compared one by one to construct the initial feature matrix of frequency coupling. The phase difference, amplitude superposition ratio, and their variation trajectory within a continuous analysis period are encoded in multiple dimensions and input into the tensor construction module. A frequency domain coupled tensor containing the cooperative variation mode of the disturbance source is formed through a high-order tensor structure.

5. The digitally controlled uninterruptible power supply parallel operation phase-locked method according to claim 1, characterized in that, The specific steps for constructing the initial feature matrix of frequency coupling degree by combining the modulation frequency index recorded in the disturbance parameters set of each uninterruptible power supply with the extracted main modulation frequency components for phase alignment processing are as follows: Based on the modulation frequency index recorded in the disturbance parameter set of each uninterruptible power supply, the corresponding main modulation frequency component is extracted from the dynamic spectrum matrix of each power supply, and the main modulation frequency of different power supplies is phase aligned using a phase synchronization algorithm. In the spectrum data of each uninterruptible power supply, neighboring frequency points within a fixed bandwidth range are selected with the main modulation frequency as the center, and the discrete amplitude data are reconstructed into a continuous curve using a spectrum curve interpolation algorithm to form energy transition characteristics. Within the target frequency range, the main modulation frequency components and their reconstructed frequency bands of any two uninterruptible power supplies are compared one by one, and their phase difference and amplitude superposition ratio are calculated. All combined results are then summarized to form the initial characteristic matrix of frequency coupling.

6. The digitally controlled uninterruptible power supply parallel operation phase-locked method according to claim 1, characterized in that, The specific steps for generating a set of multi-factor excitation vectors representing the coupling trend of perturbations based on frequency domain coupling tensors are as follows: For the constructed frequency domain coupling tensor, the detection frequency band related to the harmonic resonance mechanism is selected in the structural stability analysis space. The amplitude information of the tensor in the energy channel dimension is extracted for each frequency coupling point, and the energy density index of the current frequency point in the whole spectrum is calculated. In the phase channel dimension of the coupling tensor, the phase change trend between the main modulation frequency components is analyzed. By statistically analyzing the phase change gradient and synchronous convergence rate within its period, the phase convergence of the modulation signals between the UPS at the current moment is quantified. The frequency distribution of the interval between the main modulation frequencies is evaluated in the frequency distribution dimension. The frequency aggregation degree is calculated by using the frequency interval dispersion function. Combined with the energy density index and the phase convergence degree, a set of excitation vectors containing multiple perturbation coupling dynamic factors is generated by a weighted fusion algorithm.

7. The digitally controlled uninterruptible power supply parallel operation phase-locked method according to claim 1, characterized in that, For the multi-factor excitation vector set constructed in the previous stage, interval normalization was performed on the energy density index, phase convergence degree index, and frequency convergence degree index, respectively. The normalized excitation vector is input into the risk amplification function model generated by training with historical operating data and resonance precursor samples. The nonlinear mapping structure built inside the model is used to dynamically calculate the contribution of each excitation factor to the system resonance risk based on the current feature combination. A weighted fusion operation is performed on the risk contribution of each dimension to output a lock-point resonance risk index with real-time continuous characteristics.

8. The digitally controlled uninterruptible power supply parallel operation phase-locked method according to claim 1, characterized in that, When the lockpoint resonance risk index exceeds a preset reference threshold, the system enters the resonance evolution boundary state, triggering the intervention modulation mechanism. Based on the acquired lockpoint resonance risk index, a perturbation offset factor matrix D is constructed. t , Where N represents the number of uninterruptible power supply devices operating in parallel, δ i (t) represents the independent disturbance phase offset generated by the i-th uninterruptible power supply within period t, and is calculated using the following formula: In the formula, γ is the upper limit coefficient of disturbance intensity, and η i F is the individualized extended parameter used by the i-th uninterruptible power supply to control the frequency of disturbance phase change, and F is the lock-point resonance risk index. ref It is the reference threshold for the lock-point resonance risk index, φ i It is the initial phase offset constant of the disturbance, and 2π is the angle of the complete cycle, that is, 360° in radians; Based on the acquired independent perturbation phase offset δ i (t), for each uninterruptible power supply, a new perturbation frequency is constructed, as shown in the following formula: In the formula, It is the new disturbance frequency that the i-th uninterruptible power supply finally applies within the current control cycle t. is the fundamental disturbance frequency used by the i-th uninterruptible power supply in the previous control cycle, and β is the disturbance gain amplification factor.

Images