New energy broadband test analysis control method and system

Abstract

The invention discloses a new energy broadband test analysis management and control method and system, and relates to the technical field of new energy detection and intelligent management and control, and the method comprises the following steps: obtaining a multi-window splicing frequency spectrum of a load switching edge, superposing phase lock consistency verification in a splicing frequency spectrum result, generating a frequency spectrum leakage suspicion chart, and carrying out the boundary constraint for energy track extraction; under the constraint of a spectrum leakage suspicion chart, an energy center track, a group delay curve and a phase derivative sequence are extracted, and a cross-frequency migration fingerprint is constructed and is used as a basic evidence for misjudgment traceability. By means of spectrum suspicion identification, feature fusion modeling, misjudgment traceability, label correction, energy reverse intervention and the like, accurate identification and closed-loop regulation and control of false operation caused by spectrum leakage are achieved, high identification accuracy, low misjudgment rate, high anti-interference performance and quick response capacity are achieved, and the operation stability and safety of a new energy field station are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of new energy detection and intelligent control technology, specifically to a method and system for broadband testing, analysis and control of new energy. Background Technology

[0002] The "New Energy Broadband Testing, Analysis and Control System" is a power monitoring and operation control platform for new energy power plants such as wind power, photovoltaic, and energy storage. It consists of a broadband measurement device and a broadband processing unit. The measurement device is responsible for high-precision real-time acquisition and event triggering recording of signals such as voltage, current, phasor, frequency, harmonics, interharmonics, and subsynchronous / broadband oscillations within a wide frequency range. The processing unit is responsible for aggregating, storing, statistically analyzing, and processing data uploaded from multiple devices. It interconnects with the WAMS master station through a communication method conforming to the power system real-time dynamic monitoring standard and the IEC61850 MMS mechanism, enabling customized uploading of raw data, pre-processed results, and diagnostic information. The system features redundant design, self-diagnostic capabilities, and strong anti-interference characteristics, enabling comprehensive monitoring and closed-loop control of frequency fluctuations, harmonic pollution, and oscillation risks during new energy grid-connected operation, thereby ensuring the safety and stability of the power grid and the efficient and reliable absorption of new energy. The existing technology has the following shortcomings: In existing technologies, grid-connected operation of renewable energy power plants typically relies on broadband measurement devices to monitor oscillation signals in the 100Hz-300Hz range in real time, in order to identify and warn of high-frequency oscillation risks. However, when the power plant is under dynamic load switching conditions (such as sudden access or disconnection of large-scale energy storage units, rapid power adjustment of wind turbine groups, and transient changes in inverter operating modes), the sampling channel is highly susceptible to transient impacts, resulting in spectral leakage. This spectral leakage causes oscillation energy that should be concentrated in the high-frequency band to migrate across frequencies in the frequency domain, manifesting as energy components falsely falling into the low-frequency band. Such false signals are easily misidentified as low-order harmonics in existing identification mechanisms, thereby interfering with the accuracy of oscillation source location and fault mechanism analysis. Furthermore, when the misjudgment result is used by the control system as a trigger condition to execute protection strategies, the system may erroneously operate under non-real fault conditions, leading to a large-scale shutdown of the entire renewable energy unit, posing a significant threat to power system security, and severely restricting the stable grid connection and effective absorption of renewable energy generation.

[0003] 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

[0004] The purpose of this invention is to provide a method and system for testing, analyzing and managing new energy broadband, so as to solve the problems in the background art mentioned above.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for testing, analyzing, and controlling new energy broadband, comprising the following steps: The multi-window spliced ​​spectrum at the load switching edge is obtained, and phase-locked consistency verification is superimposed on the spliced ​​spectrum result to generate a spectrum leakage suspicion map, which is used as a boundary constraint for energy trajectory extraction. Under the constraint of the spectrum leakage suspect map, the energy center trajectory, group delay curve and phase derivative sequence are extracted to construct cross-frequency migration fingerprint, which is used as the basic evidence for tracing the source of misjudgment; Under the constraint of cross-frequency migration fingerprint, a misjudgment tracing chain is constructed, and the cross-frequency migration fingerprint is mapped to the continuous micro-windows before and after the trigger. The contribution of low-order harmonic criteria is quantified to achieve the removal of false components. Under the constraint of misjudgment tracing chain, a robust discriminator is trained, and the characteristic consistency of energy center trajectory, group delay curve and phase derivative sequence is enhanced by self-distillation. The corrected frequency band label is output to provide a basis for linkage suppression. Under the constraints of the corrected frequency band labels, a linkage suppression command is generated, a short dwell time delay window is formed through dual threshold cross confirmation, a fixed-point disposal list is established, and input conditions are provided to dynamic control. Under the constraint of the linkage suppression command, a dynamic control sequence is generated, the time-reversal phase gating mechanism is enabled, and the programmable polarization metasurface is driven to inject a conjugate suppression envelope to construct a reverse energy channel in the leakage frequency band, thereby realizing the closed-loop release of malfunctions.

[0006] Preferably, the steps for generating the spectrum leakage suspect map are as follows: Extract the rate of change and acceleration of change of the voltage sampling sequence and the current sampling sequence, and extract the disturbance edge time window by combining the rate of change threshold and the power slope fitting method; Multiple time-scale weighted analysis windows are constructed within the edge time window, Fourier transforms are performed on each window, and frequency axis interpolation is aligned. The spliced ​​spectrum is obtained by splicing the windows using an overlapping weighted method. The phase trajectory of frequency points is extracted from the spliced ​​spectrogram, a multi-window phase evolution sequence is constructed, and the phase change rate is calculated to generate a phase consistency score map; By combining the amplitude coherence and phase consistency scores of frequency points, abnormal frequency bands in the spectrum are extracted, forming a spectrum leakage suspect map with frequency as the horizontal axis and time window index as the vertical axis, and outputting the time frequency distribution results of suspected leakage frequency bands in the spectrum.

[0007] Preferably, the cross-frequency migration fingerprint construction process is as follows: Within the frequency band covered by the suspected spectrum leakage map, the energy center trajectory of the frequency distribution in each time window is extracted to form a continuous frequency change path; Based on the energy center trajectory, the group delay curve in the frequency domain is calculated to identify the propagation trend, inflection point, and negative value segment in frequency migration. By combining the group delay curve, the phase derivative sequence of the target frequency point in the time dimension is extracted, the segment with drastic phase change is identified, and the key frequency position is marked. The energy center trajectory, group delay curve, and phase derivative sequence are normalized and fused to construct a complete cross-frequency migration fingerprint, which is used for subsequent source analysis of spectrum misjudgments.

[0008] The preferred steps for constructing a false positive traceability chain are as follows: Based on cross-frequency migration fingerprint images, the start and end frequencies of energy trajectories are extracted, and the boundary ranges of the frequency axis and time axis are determined. A continuous micro-time window sequence is constructed within the fingerprint boundary range, and the spectral amplitude, phase value and signal-to-noise ratio of frequency points are extracted to form candidate misjudgment paths; Connect the candidate frequency paths in time order to construct a misjudgment tracing chain that includes spatial overlap, phase change value and group delay jump amplitude; The contribution intensity of the misjudgment tracing chain to the low-order harmonic criterion in the micro-window before and after the disturbance is evaluated, and the low-order harmonic misjudgment contribution matrix is ​​generated. Based on the contribution matrix, frequency points with sudden energy increases after disturbance and no harmonic characteristics before disturbance are identified. Amplitude attenuation compensation and phase continuity reconstruction are performed to complete the spectrum stripping process.

[0009] The preferred steps for outputting the corrected frequency band label are as follows: Training samples were constructed based on the misjudgment tracing chain. Three types of features were extracted: energy center trajectory, group time delay curve and phase derivative sequence. A three-dimensional array structure corresponding to time window index, frequency point position and feature channel was established. Labeling was completed and a complete training dataset was formed. Based on the constructed training dataset, a discriminative model is built using a self-distillation training method. The teacher model provides intermediate feature response benchmarks, while the student model strengthens the consistent expression of the three types of features under the joint constraints of label error and feature response error, thereby improving the discriminative stability and generalization ability under perturbation conditions. The trained discrimination model is used to perform full-band reasoning on the target spectrum data, extract the frequency point classification results and corresponding feature trends, and generate corrected frequency band labels with time continuity and anti-interference ability, which are used as the basis for subsequent linkage suppression operations.

[0010] Preferably, in the self-distillation training method, the feature response of the student model is aligned layer by layer with the intermediate feature maps of the teacher model in three feature channels: energy center trajectory, group delay curve and phase derivative sequence, and the Euclidean distance is used as the optimization objective to enhance the model's discrimination consistency in the frequency perturbation boundary region.

[0011] Preferably, the steps for generating the linkage suppression instruction are as follows: Based on the corrected frequency band labels, frequency points that are continuously marked as non-true harmonic categories are extracted, a frequency dwell time map is constructed, and the dwell stability coefficient is calculated to screen out abnormal frequency points with high dwell characteristics. Based on the frequency dwell time map, a dual verification of frequency trajectory stability threshold and energy mutation threshold is introduced for the selected frequency points, and only frequency points that simultaneously meet the two threshold conditions are retained as trigger targets. Based on the center time window of the dual-threshold passing frequency point, a short dwell time window is formed by expanding forward and backward, and the disturbance amplitude, phase derivative and group delay characteristics are extracted and a fixed-point handling list is constructed. Based on the list of designated disposal points, a coordinated suppression command is generated, which clarifies the frequency location, time window range, disturbance direction and response measures, forming a coordinated response sequence that is time-ordered and resolves conflicts.

[0012] Preferably, under the constraint of the linkage suppression command, the following steps are taken to generate a dynamic control sequence, trigger time-reversal phase gating, drive the programmable polarization metasurface to inject a conjugate envelope, and construct a reverse energy channel in the leakage band: Based on the frequency points, time indices, disturbance directions, and response methods contained in the linkage suppression command, a set of control parameters including time, phase, and disturbance trends is constructed, and the phase transition path is extracted to determine the true disturbance boundary. A dynamic control sequence is generated based on the constructed set of control parameters. Using this sequence as input, a time-reversal phase-gating operation is performed to reverse the phase trajectory and construct a conjugate intervention waveform for energy reverse intervention. The conjugate interference waveform is translated into spatial control commands, which drive the programmable polarization structure to activate the polarization unit, forming a spatial beam with directionality and phase consistency, thereby realizing the injection of reverse energy channels in the spectrum and the removal of interference loops.

[0013] The new energy broadband test analysis and control system includes a spectrum suspicion generation module, a cross-frequency fingerprint construction module, a traceability chain analysis module, a robust discrimination module, a linkage suppression module, and a dynamic control module. The spectrum suspect generation module obtains the multi-window spliced ​​spectrum at the load switching edge, overlays phase-locked consistency verification on the spliced ​​spectrum result, and generates a spectrum leakage suspect map, which is used as a boundary constraint for energy trajectory extraction. The cross-frequency fingerprint construction module extracts the energy center trajectory, group delay curve and phase derivative sequence under the constraint of the spectrum leakage suspect map to construct a cross-frequency migration fingerprint, which is used as the basic evidence for tracing the source of misjudgments. The source chain analysis module constructs a misjudgment source chain under the constraint of cross-frequency migration fingerprint, maps the cross-frequency migration fingerprint to continuous micro-windows before and after the trigger, and quantifies the contribution of low-order harmonic criteria. The robust discrimination module trains a robust discriminator under the constraint of the misjudgment tracing chain. It enhances the characteristic consistency of the energy center trajectory, group delay curve and phase derivative sequence through self-distillation and outputs the corrected frequency band label to provide a basis for linkage suppression. The linkage suppression module generates linkage suppression commands under the constraints of the corrected frequency band labels, forms a short dwell time delay window through dual-threshold cross confirmation, establishes a fixed-point handling list, and provides input conditions for dynamic control. The dynamic control module generates a dynamic control sequence under the constraint of the linkage suppression command, enables the time-reversal phase gating mechanism, drives the programmable polarization metasurface to inject the conjugate suppression envelope, and constructs the reverse energy channel in the leakage frequency band.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention achieves highly sensitive capture of transient disturbances through spectrum suspicion generation and constructs a cross-frequency migration fingerprint using the fusion features of energy center trajectory, group delay curve, and phase derivative sequence, thereby improving the accurate identification of pseudo-frequency band energy components. Based on this, by constructing a misjudgment tracing chain, the origin time and starting frequency of spectrum leakage can be accurately traced, effectively separating false low-frequency signals and preventing them from being incorrectly labeled as actual fault sources in subsequent identification stages. Furthermore, the system improves the robustness of spectrum discrimination by introducing a self-distillation training mechanism and accurately corrects erroneous identification results using corrected frequency band labels, thus providing a reliable basis for downstream linkage suppression and control decisions. Finally, through time-reversal phase gating and conjugate suppression mechanisms, reverse energy intervention in the physical frequency domain is achieved, completely interrupting the triggering path of malfunctions and ensuring stable operation and uninterrupted tripping of new energy equipment under dynamic grid-connected conditions. Overall, the solution has beneficial technical effects such as high identification accuracy, low misjudgment rate, fast control response, accurate energy intervention, and strong closed-loop capability, which significantly improves the anti-disturbance capability and operational safety of new energy power plants in the face of complex disturbance scenarios. Attached Figure Description

[0015] 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.

[0016] Figure 1 This is a flowchart of the new energy broadband test, analysis and control method of the present invention.

[0017] Figure 2 This is a schematic diagram of the modules of the new energy broadband test, analysis and control system of the present invention. Detailed Implementation

[0018] 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.

[0019] This invention provides, for example Figure 1 The new energy broadband test analysis and control method shown includes the following steps: The multi-window spliced ​​spectrum at the load switching edge is obtained, and phase-locked consistency verification is superimposed on the spliced ​​spectrum result to generate a spectrum leakage suspicion map, which is used as a boundary constraint for energy trajectory extraction. To accurately identify spectrum leakage under load switching conditions in renewable energy power plants, a spectrum reconstruction and phase verification method based on edge event response is proposed to generate a spectrum leakage suspect map that can be used for energy trajectory extraction. The method includes the following steps: To address potential power surges during the operation of renewable energy power plants, such as sudden connection or disconnection of large-capacity energy storage units, drastic changes in wind turbine power output, and rapid switching of photovoltaic inverter operating modes, an edge extraction method based on sampled data surge detection is employed to identify the time region of the disturbance. Specifically, firstly, first- and second-order difference calculations are performed on high-precision synchronous sampling sequences of voltage and current to extract the corresponding current rate of change and voltage acceleration sequences. Within these derivative sequences, the location where the rate of change simultaneously exceeds two set thresholds (e.g., three times the standard deviation of the set average rate of change) is selected as the initial disturbance judgment point. Subsequently, multiple time periods are extended around this initial point, and the average power slope within each period is fitted. The abrupt boundary of the fitted curve is taken as the edge interval range of the disturbance, ultimately forming an edge time window composed of multiple sampling points. This time window does not rely on traditional fixed power threshold judgment methods but rather extracts data through dynamic characteristic slope response, enabling the identification of the entire process data of the disturbance initiation stage even in the context of multiple concurrent events, thus providing a stable starting point for subsequent frequency domain analysis.

[0020] Within the extracted edge time window, a multi-window stitching method is used to reconstruct the spectrum of the perturbation data. First, within the edge window range, several pairs of symmetrical analysis window groups are constructed, each group consisting of a short window covering 20ms and a long window covering 100ms. Each analysis window uses a weighted cosine window function to process its covered sampling segment to reduce spectral leakage caused by boundary effects, and uses integer-cycle compensation technology of the signal window function to ensure that the analyzed samples completely cover one or more fundamental frequency cycles. Discrete Fourier transform processing is performed within each window group to obtain its respective frequency domain amplitude and phase information. The spectra output from all windows are aligned to a uniform frequency resolution (e.g., one frequency point per 10Hz) through frequency axis interpolation and normalized to eliminate the influence of amplitude dimensions. In the stitching operation, a weighted coefficient fusion method based on window overlap is used. The weighting factor is set according to the center overlap ratio between each window group; for example, a weighting factor of 0.5 corresponds to a window overlap of 50%. Finally, the spectra of different time scales and resolutions are fused into a unified frequency distribution image. Compared to traditional single-window Fourier analysis with a fixed window length, this spliced ​​spectrum can simultaneously preserve the detailed resolution of high-frequency signals and the stability of low-frequency energy distribution, significantly improving the ability to present complex perturbation spectral structures.

[0021] After constructing the spliced ​​spectrogram, a frequency point reliability determination method based on phase consistency is introduced to improve the ability to distinguish between real and pseudo-spectral components. First, the multi-window phase trajectory corresponding to each frequency point in the spliced ​​spectrogram is extracted, constructing a phase evolution sequence for each frequency point in different time windows. Next, the phase consistency score for each frequency point is obtained by calculating the phase change rate between adjacent windows. Specifically, if a frequency point maintains stable phase change across multiple windows with a change rate less than a set threshold (e.g., 5 degrees / window), its phase consistency is high, and it is considered a real spectral component. If a frequency point experiences drastic phase jumps in different windows with a change rate greater than a set upper limit (e.g., 30 degrees / window), it is judged to lack phase continuity and is suspected to be a pseudo-frequency component caused by spectral leakage. Furthermore, for cases where a frequency point is missing in an intermediate window, the phase trends of the preceding and following windows are further differentially fitted. If the trend breaks significantly, the frequency point is further marked as a high-risk area. This phase consistency verification method effectively distinguishes pseudo-peaks in multi-window spliced ​​spectra that have energy but lack phase support, and solves the problem that traditional methods, which rely solely on amplitude levels, easily misidentify leaked pseudo-frequency points as real oscillations.

[0022] After completing the spectrum stitching and phase consistency verification, frequency bands with potential leakage are identified based on the amplitude continuity and phase consistency scores of the frequency points, and a spectrum leakage suspicion map is constructed. The specific construction method is as follows: First, a frequency band continuity identification threshold is set. If the phase consistency score of more than 3 out of 5 adjacent frequency points is below the threshold, and there is a local bulge in the amplitude continuity within the frequency band (i.e., the amplitude exceeds the average of the preceding and following frequency points by more than 30%), it is judged as a suspected leakage segment. Then, all segments meeting the above conditions are extracted from the frequency axis, and their starting frequency, ending frequency, maximum amplitude point, minimum phase consistency point, and other characteristic indicators are recorded to generate a two-dimensional spectrum leakage suspicion map. This map uses frequency as the horizontal axis and time window index as the vertical axis, using grayscale or thermal color coding to represent the degree of suspicion of the frequency points. In this map, continuous high-susceptibility areas are displayed as bright blocky areas, whose shape is significantly different from the actual energy trajectory distribution, typically exhibiting strong lateral diffusion and poor vertical continuity. This figure, as the final result of spectrum leakage identification, can not only be used to visually display the location of non-true frequency components, but also provide boundary condition references for subsequent steps such as energy trajectory extraction, cross-frequency migration analysis, and harmonic misjudgment removal.

[0023] Under the constraint of the spectrum leakage suspect map, the energy center trajectory, group delay curve and phase derivative sequence are extracted to construct cross-frequency migration fingerprint, which is used as the basic evidence for tracing the source of misjudgment; Based on the completed spectral leakage suspect map, in order to achieve in-depth analysis and accurate identification of suspected frequency energy migration phenomena, it is necessary to extract three types of dynamic features: energy center trajectory, group delay curve, and phase derivative sequence. These three features will be integrated to construct a complete cross-frequency migration fingerprint, providing a clear analytical basis for subsequent misjudgment tracing. The specific implementation steps are as follows: Within all frequency bands covered by the suspected spectral leakage map, energy center trajectories are extracted window-by-window for each time window's frequency axis slice. For each time window's corresponding spectral data, the power spectral density value of each frequency point within the suspected frequency range is first selected and used as a weight to perform a weighted average calculation for that frequency band, obtaining the center frequency value for that time window. To reduce interference from instantaneous abnormal peaks and improve the continuity and smoothness of the trajectory, a three-window joint mechanism is designed. The center frequency calculation result for each time window must simultaneously reference the data from the preceding and following windows, and a weighted average is taken after uniform fitting within the three windows. Subsequently, the center frequencies obtained from all time windows are arranged sequentially to form a continuous energy center trajectory curve. To enhance the visual perception of frequency migration trends, linear trend fitting and local inflection point analysis are performed on the trajectory to identify whether there is a continuous downward slope from high to low frequencies. If this behavior shows a stable trend within multiple time windows, it can be preliminarily judged as an energy migration from a high-frequency band to a low-frequency band. This trajectory extraction method is significantly different from the traditional fixed-frequency energy tracking method, and has the ability to dynamically identify the cross-frequency start and end points.

[0024] Based on the obtained energy center trajectory, the group delay variation behavior of each frequency point on the trajectory in the frequency domain is further analyzed. To achieve this analysis, high-precision phase extraction is first performed on the frequency interval near the energy trajectory in each time window. The phase increment of adjacent frequency points is obtained using phase difference calculation, and the derivative of phase with respect to frequency is calculated by combining the frequency interval, which is the group delay value. To improve the physical interpretability of the group delay data, the group delay value of each frequency point is time-series processed to form a complete group delay curve, and a moving average denoising method is used to remove high-frequency jitter components. Subsequently, second-order derivative analysis is performed on the group delay curve to identify inflection points, trend change points, and negative value troughs. These features usually characterize the possibility of feedback, back propagation, or structural resonance in the signal energy propagation path. In particular, if the group delay rapidly jumps from a positive value to a negative value in a certain frequency band, and this point happens to be located at the edge of a high-suspected frequency band in the spectrum leakage suspicion map, it can further corroborate that this frequency point is a transit area for cross-frequency energy injection. Unlike conventional methods that only observe the absolute value of group delay, this method deeply integrates the morphological changes of group delay with trajectory information, significantly enhancing the analytical power of the physical properties of the migration path.

[0025] After the energy center trajectory and group delay curve are extracted, the phase derivative sequence extraction stage begins to capture the dynamic evolution characteristics of the signal phase during spectral disturbances. In this process, for each target frequency point in the suspected spectral leakage map, its absolute phase value is recorded, and then a first-order difference calculation is performed in the time dimension to obtain the rate curve of phase change over time. To avoid error accumulation due to periodic phase folding, all phase values ​​are first unwrapped, compensating for phase points with jumps exceeding 180 degrees, ensuring the phase sequence has a continuously differentiable numerical basis. Subsequently, volatility detection is performed on the phase derivative curve to identify abrupt change points and rapid reversal points, and their corresponding time window indices and frequency positions are marked. If a frequency point experiences a sudden increase or drastic change in phase derivative across multiple consecutive time windows, and this position has a high degree of overlap with nodes in the energy center trajectory and is also within the group delay jump region, then this frequency point can be identified as a significant event point of energy migration. This method can not only extract the location of the instantaneous features of the disturbance signal, but also identify the triggering mechanism in the energy cross-frequency transfer process, thereby effectively avoiding the risk of misjudging phase distortion as a real harmonic change.

[0026] The extracted energy center trajectory, group delay curve, and phase derivative sequence are fused to construct a complete cross-frequency migration fingerprint, serving as the core data structure for misjudgment tracing analysis. In this construction process, the three features are first mapped to a unified time-frequency plane, establishing a two-dimensional feature matrix. The row dimension corresponds to the time window sequence, and the column dimension corresponds to the frequency distribution. The matrix elements consist of normalized values ​​from the three features. Subsequently, to enhance the differential representation between different features, different weights are assigned to the three features; for example, the energy center trajectory is assigned 0.5, the group delay curve 0.3, and the phase derivative sequence 0.2. These weighted superpositions are then used to synthesize a feature fusion image. In this image, the true frequency migration path typically appears as a bright trajectory sloping from high to low frequencies, accompanied by a gradual band of group delay and a phase transition edge around the trajectory. To extract this trajectory, morphological edge detection is performed on the fused image, and isolated false peaks are filtered out through path consistency checks, ultimately forming the cross-frequency migration fingerprint curve. This fingerprint curve not only reflects the starting point, path, and landing point of energy, but also embeds the group delay structure and phase evolution direction, which is a key input in subsequent source tracing judgment.

[0027] Under the constraint of cross-frequency migration fingerprint, a misjudgment tracing chain is constructed, and the cross-frequency migration fingerprint is mapped to the continuous micro-windows before and after the trigger. The contribution of low-order harmonic criteria is quantified to achieve the removal of false components. To further identify frequency components misclassified as true low-order harmonics during spectral perturbations, a method for constructing a misclassification source chain is proposed, constrained by the obtained cross-frequency migration fingerprint. Through continuous time-frequency window mapping, feature path tracing, and criterion contribution evaluation, accurate removal of non-true spectral components is achieved. The specific implementation steps are as follows: Based on the constructed cross-frequency migration fingerprint image, the start and end frequencies of the energy trajectory in the fingerprint are extracted. Combined with the start and end indices of the trajectory in the time window sequence, the boundary ranges on the frequency and time axes are clarified. Within this boundary, all frequency points coinciding with the end of the trajectory, especially those close to integer multiples of the fundamental frequency (e.g., 100 Hz, 150 Hz, 200 Hz, 250 Hz), are selected to form a set of suspected spectral leakage landing points. In each time window, the energy proportion, phase consistency score, and group delay stability of these frequency points are statistically analyzed. A fingerprint mapping path region is constructed in the two-dimensional frequency-time plane as a high-risk analysis area for the subsequent micro-window reflection region.

[0028] Within the defined fingerprint boundary, a sequence of highly overlapping micro-time windows is constructed, centered on the perturbation occurrence time window. Each time window is set to a length of 20 milliseconds, and sliding extraction is performed in 5-millisecond steps to ensure continuous coverage of the entire perturbation response process. Within each micro-time window, a Discrete Fourier Transform is performed on the original sampled data to extract the spectral amplitude, phase value, and signal-to-noise ratio (SNR) of the target frequency point. The dynamic characteristics of each frequency point during the perturbation process are recorded, such as peak occurrence time, spectral peak width expansion trend, and the rate of change of the total energy integral. The frequency domain data extracted from these micro-time windows is then matched with the target frequency path in the cross-frequency migration fingerprint to construct candidate false positive paths.

[0029] Using the matching results, overlapping frequency points in adjacent micro-windows are connected in chronological order to form a set of paths evolving from the starting frequency to the ending frequency, thus constructing a preliminary misjudgment tracing chain. In this chain, each frequency point possesses three types of characteristic indicators: spatial overlap with the fingerprint trajectory, temporal abrupt change in the phase derivative, and jump amplitude of the group delay. By setting path continuity conditions (such as the frequency point simultaneously possessing high amplitude, stable phase, and low group delay slope in at least four consecutive time windows), stable migration paths are selected. Further consistency analysis of the spectral characteristics of the path's starting point, middle section, and ending point is performed to verify whether the path might be caused by the spectral wake migration of a single high-frequency disturbance. In traditional techniques, spectral trajectories are often judged based on a single frequency point abrupt change; this method achieves multi-feature tracking of continuous frequency domain paths in the temporal evolution dimension, exhibiting a stronger ability to reconstruct the causal chain of misjudgments.

[0030] After the misjudgment tracing chain is constructed, the numerical impact of this path on the low-order harmonic criteria before and after the disturbance is evaluated. Within each micro-time window, multiple criteria are used for joint evaluation of frequency points located at integer multiples of the fundamental frequency. These criteria include whether the amplitude percentage of a single point exceeds 1.5 times the average value across the entire frequency band, whether the frequency point is misidentified as a harmonic component by the power quality assessment standard, and whether the phase continuity of the frequency point undergoes abrupt changes before and after the disturbance. These indicators for all frequency points constitute a low-order harmonic misjudgment contribution matrix, where each element represents the probability of a frequency point being misjudged as a harmonic within a micro-time window. Integrating this matrix row-wise and column-wise yields the change surface of the low-order harmonic criteria throughout the entire time window sequence. Combined with the energy trajectory of the cross-frequency migration path, a focused analysis is performed on the frequency points with the highest contribution intensity, providing a basis for the stripping decision.

[0031] Based on the analysis of the low-order harmonic misidentification contribution matrix and the source chain path, a false spectral component removal process is performed. For frequency points that experience a sudden energy increase after a disturbance but did not possess any harmonic characteristics before the disturbance, if these frequency points exhibit high phase perturbation rate, group delay instability, and energy trajectory characteristics with a high degree of matching with the cross-frequency migration endpoint, they are marked as removal targets. In each time window, the spectral amplitude of the removal target frequency points is attenuated and compensated, and continuity reconstruction is performed in the phase space to eliminate non-physical spectral structure anomalies caused by misidentification. Furthermore, the removed spectrum is re-aligned to ensure the energy conservation of the entire spectrum and the continuity of the dominant frequency.

[0032] Under the constraint of misjudgment tracing chain, a robust discriminator is trained, and the characteristic consistency of energy center trajectory, group delay curve and phase derivative sequence is enhanced by self-distillation. The corrected frequency band label is output to provide a basis for linkage suppression. After identifying the complete chain of misjudgments and pinpointing high-risk frequency paths, it is necessary to further establish a feature recognition model with high generalization capabilities to dynamically classify frequency behavior, strengthen the consistency among the three types of features, and ultimately generate corrected frequency band labels, providing an executable basis for subsequent actions. This process includes the following steps: Based on the established misjudgment tracing chain, the time window index and corresponding frequency point set contained in each cross-frequency migration path are extracted, and training samples are constructed based on these calibrated regions. The core data of the training samples comes from joint feature extraction of three dimensions: energy center trajectory, group delay curve, and phase derivative sequence. Regarding the energy center trajectory, the spectral amplitude distribution within each time window in the spectrogram is first extracted. A weighted distribution function is established with frequency as the horizontal axis, and the frequency points within the suspected frequency range are weighted and averaged to form the center frequency value representing the spectral quality of the current time window. Subsequently, based on the changes in the center frequency points of multiple adjacent time windows, the frequency drift velocity is calculated and used as an important dimension representing the energy disturbance trend. For the extraction of the group delay curve, the phase change value of adjacent frequency points within a continuous window is divided by the frequency interval to obtain the group delay value. A time series vector is further constructed, and its first and second derivatives are extracted through a sliding window to form a complete propagation delay fluctuation curve. For constructing the phase derivative sequence, phase values ​​at the same frequency points within a continuous time window are collected. Phase unwrapping is first performed to eliminate errors caused by periodic jumps. Then, the first-order difference is calculated based on the time window interval, forming a fluctuation index reflecting phase instability. Three types of features occupy fixed positions in the samples of each time window and are combined into a three-dimensional array structure. Each dimension corresponds to the time window index, frequency point position, and feature channel, ensuring that the model input has high-dimensional expressive power and physical meaning correspondence. In the training dataset, frequency points confirmed to belong to the leakage spectrum trail in the misjudgment source chain are labeled as misjudged categories, frequency points corresponding to stable harmonic signals are labeled as true harmonic categories, and other frequency points without significant disturbance behavior are labeled as background categories, thus forming a clear label structure for subsequent training.

[0033] For the constructed training dataset, a feature consistency enhancement method is used to train the model, improving its robustness in perturbation scenarios. During training, an initial classification model is first constructed, whose architecture can simultaneously receive three-dimensional time window-frequency-feature data and output the probability distribution of each frequency point belonging to a specific category. The initial training employs supervised learning, using manually labeled misclassified, harmonic, and background categories as standard outputs, optimized using a cross-entropy loss function. After the initial training, the model parameters are frozen, and a student model with the same structure but independent weights is constructed as the target model for further learning. In the next round of training, the initial model serves as the teacher model, providing a reference representation of intermediate features for the student model. Specifically, the output results of the teacher model at each layer's intermediate feature map are extracted, particularly the response values ​​in the energy center trajectory, group delay changes, and phase derivative position sensitive areas, serving as the teacher model's feature response benchmark. The student model, during training, needs to minimize not only the prediction error of the output label but also the response difference between its intermediate feature map and the teacher model; this process is accomplished by calculating the Euclidean distance between corresponding positions. This self-distillation training mechanism enables student models to learn the response structures of teacher models to three key features, strengthening the model's stable judgment ability under perturbation conditions. In particular, it significantly outperforms traditional pure label-driven methods in judging frequency points with blurred perturbation boundaries and unstable trajectory start and end points. Furthermore, slight perturbation noise is artificially added to the training samples to test the model's discriminative ability under non-ideal conditions and further improve its adaptability to time window slippage, frequency shifts, and phase flicker. This training method is completely different from traditional methods that enhance discriminative ability through multi-label classification. Its advantage lies in achieving recognition generalization ability across perturbation environments solely through model self-reinforcement, without relying on newly added labeled resources.

[0034] After the training process converges and is completed, the trained model is used to perform full-band label inference on the target spectrum data, outputting corrected frequency band classification results. During inference, the sampled data throughout the entire disturbance period is first subjected to batch feature extraction to construct a three-class joint feature data structure consistent with the training phase. This structure is then sequentially input into the model for window-by-window, frequency-by-frequency classification prediction. For each frequency point, the model outputs the probability value of being identified as a misclassified signal, a true harmonic, or background, selecting the one with the highest probability as the final classification result. Within each time window, the classification results are plotted on the frequency axis using color coding to generate a label chromatogram, visually indicating which frequency points in the spectrum belong to the true signal and which belong to misclassified frequencies caused by spectral leakage or boundary distortion. Specifically, for frequency points identified as misclassified by the model, the corresponding three-class feature change trends are further extracted, and their energy center sinking, group delay reversal, or phase derivative jump positions can be marked on the image for subsequent human review and strategy formulation. Furthermore, to eliminate classification label jumps caused by random fluctuations within short time windows, a time window label smoothing process is performed. This involves voting and fusing the classification results for the same frequency point in three adjacent time windows, retaining the most frequently occurring label as the final output. This achieves label continuity and anti-interference capability over time. The final output frequency band label results can be directly used as the frequency input source for subsequent suppression actions, serving as an important basis for determining whether to trigger suppression, response control, frequency jump delay, and other functions.

[0035] Under the constraints of the corrected frequency band labels, a linkage suppression command is generated, a short dwell time delay window is formed through dual threshold cross confirmation, a fixed-point disposal list is established, and input conditions are provided to dynamic control. After obtaining frequency band labels generated based on feature consistency enhancement and identifying suspected disturbance frequency points, it is necessary to construct a response judgment process for coordinated suppression based on this label information. This process involves timing window filtering, disturbance stability assessment, location confirmation, and behavior definition, ultimately forming coordinated instructions that can be directly used for dynamic intervention to guide subsequent refined frequency control actions. This process includes the following steps: Based on the corrected frequency band labels, frequency points continuously marked as non-true harmonic categories are extracted, and a frequency dwell time map is constructed. This map uses frequency points as the vertical axis and time windows as the horizontal axis, recording the label status of each frequency point within each time window in a two-dimensional matrix. If a frequency point is marked as "misclassified" in multiple consecutive time windows, it can be considered that the frequency point has a significant dwelling trend during the disturbance process. To improve the accuracy of screening, the duration and continuity of misclassified labels for each frequency point are statistically analyzed, a dwelling stability coefficient is calculated, and a dwelling judgment threshold is set. If a frequency point maintains a misclassified label for five or more consecutive time windows, and its frequency fluctuation range is within a set threshold (e.g., ±1Hz), then the frequency point is identified as an anomalous target with high dwelling characteristics. Based on this, the energy distribution of the dwelling frequency points is further evaluated, the spectral amplitude of each time window is extracted, and the total energy integral during the entire dwelling period is calculated. Frequency points with extremely low amplitudes that may be background noise are eliminated, and only high-energy dwelling frequency points with substantial disturbance effects are retained, providing a basis for subsequent dynamic judgment.

[0036] A dual-threshold cross-validation mechanism is introduced for the dwell frequency points to ensure that the frequency disturbance is indeed a significant behavior that can trigger a response. The first threshold is the "frequency trajectory stability threshold." Under this threshold, the trajectory of the center frequency change of the dwell frequency point in all dwell time windows is analyzed, and the drift amplitude and standard deviation of the frequency point in the time series are calculated. If the value exceeds a set range (e.g., standard deviation greater than 0.3Hz), the frequency point is considered to lack trajectory stability and is removed from the candidate set. The second threshold is the "energy mutation threshold." This threshold requires that the frequency point exhibits a significant energy mutation during the dwell period. Specifically, the energy integral value of the frequency point in the first half and the second half is calculated separately, using the midpoint of the dwell time window as the boundary, and the difference is taken. If the energy integral of the second half is significantly higher than that of the first half (e.g., an increase of more than 30%), the frequency point is considered to have an energy mutation trend and can be used as a potential trigger source for the disturbance response. Only when a certain frequency point simultaneously meets both the trajectory stability threshold and the energy mutation threshold will it be identified as a valid trigger target and enter the response preparation phase.

[0037] Within the frequency set after dual-threshold screening, a short-stay delay window is constructed for each frequency point to clarify the specific time period during which the frequency point may generate disturbance effects. The construction method is as follows: using the center time window that satisfies both thresholds as a reference, two time windows are extended forward and backward, forming a time period of five time windows, which serve as the boundaries of the short-stay delay window. Within this window, the spectral amplitude, phase derivative, and group delay values ​​of the frequency point in each time window are recorded, and their rate of change and average disturbance severity are calculated. This process aims to define the persistence and intensity of the actual disturbance behavior at the frequency point, providing a quantitative basis for subsequent control strategy development. Subsequently, all short-stay delay windows that meet the conditions are uniformly sorted, prioritizing the retention of windows with large disturbance amplitudes and high spectral jump rates, while removing windows with excessive temporal overlap or disturbance amplitudes below the trigger threshold, ensuring that the final selected delay windows possess temporal independence and disturbance significance. Finally, using frequency point location, delay window start and end time, disturbance direction (upward or downward) and interference amplitude as indexes, structured data items are formed and summarized into a targeted handling list.

[0038] Based on the designated handling list, a coordinated suppression command is generated for each frequency point to be handled, and these commands are aggregated into an execution sequence for regulating the response. Each command contains the following five fields: trigger frequency position, action start time window index, action end time window index, disturbance phase adjustment direction, and suggested response measure category. The trigger frequency position is directly derived from the previous dwelling frequency point determination result; the action time window boundary is determined by the short dwelling delay window; the disturbance direction is determined by the trend of phase derivative change (e.g., a continuously increasing phase derivative indicates an upward disturbance); and the response measure category is determined based on the disturbance energy change result, classifying it as a power absorption or power injection command. All generated coordinated suppression commands are sorted according to the start time window to form a coordinated response sequence, and conflict resolution between commands is executed. For example, if two command time windows overlap and the frequency points are too close (less than 1 Hz), they are automatically merged into a single joint response action; if the command action directions are opposite and the disturbance amplitudes are similar, the command with the greater impact is retained. The resulting linked suppression command sequence features high reliability, high time accuracy, and high response accuracy. It not only has the ability to suppress disturbances in a targeted manner, but can also serve as a standard input format in the dynamic control process, enabling seamless integration with subsequent spectrum control processes.

[0039] Under the constraint of the linkage suppression command, a dynamic control sequence is generated, the time-reversal phase gating mechanism is enabled, the programmable polarization metasurface is driven to inject the conjugate suppression envelope, and a reverse energy channel is constructed in the leakage frequency band to realize the closed-loop release of malfunctions. Upon receiving a linkage suppression command containing information such as frequency location, duration of action, and phase adjustment direction, a dynamic control sequence with timing, phase, and spatial distribution information needs to be generated based on the command content. This sequence is then used to drive the polarization structure to realize a reverse channel for spectral energy, thereby completely eliminating the malfunction path caused by spectral leakage within the frequency domain. The implementation process is as follows: Based on the frequency points, start and end time window indices, disturbance directions, and power response modes specified in the linkage suppression commands, a set of control parameters is constructed to generate a precisely matched dynamic control sequence. Each control command is converted into a set of parameter items, including the start and end time indexes, target frequency value, instantaneous phase angle characteristics, disturbance trend sign, and corresponding control direction. During the construction process, the phase trajectory of the target frequency within each time window is first discretized, and the instantaneous phase change in continuous time windows is extracted. By performing high-order differencing on this phase difference sequence, the start and end points of phase abrupt changes are captured and used to define the true boundary of the frequency disturbance. Next, based on the disturbance direction, it is determined whether the frequency point belongs to the energy rise or energy fall category, and it is designated as either injection-type or absorption-type control, respectively. For each control target point, its corresponding gating activation time period is determined, i.e., gating control is performed during its disturbance phase transition period, and its propagation direction characteristics in space are recorded to facilitate subsequent reverse excitation path matching. Finally, all this information is encoded into control sequence items as the trigger condition set driving the phase gating mechanism.

[0040] A time-reversal phase gating operation is performed using the generated dynamic control sequence to intervene in the perturbation signal in both time and frequency dimensions. The key to this process is to use the identified phase transition path as a reference, perform inverse time-series mapping on it, and execute a symmetric conjugate transformation in the frequency dimension to construct an intervention excitation path with energy cancellation capabilities. Specifically, for each frequency point's perturbation time window, the phase continuity structure of its pre-perturbation state is traced back to establish a pre-perturbation steady-state phase reference. Then, for the phase abrupt changes detected during the perturbation process, its conjugate inversion curve is constructed, that is, the phase information at the current moment is mapped backward along the time axis to the previous moment, and this process is iteratively executed until the phase stability segment before the perturbation begins is reached. During this process, a time-reversal mapping table is simultaneously established, containing the inverse phase value, frequency point position, and perturbation direction at each moment, ensuring time-matching accuracy for subsequent phase control operations. Subsequently, based on the frequency point position and perturbation direction, a synthetic intervention waveform is constructed using phase modulation. This waveform has an inverse phase structure with the original perturbation waveform in the time-frequency two-dimensional domain. The intervention waveform is embedded into the original signal path, and the energy of the original disturbance is weakened through conjugate interference. In traditional techniques, disturbance intensity can usually only be suppressed by filtering, but frequency domain closed-loop repair cannot be achieved. However, the time-reversal control and reverse phase injection mechanism in this step significantly improve the time and frequency matching accuracy of the intervention effect.

[0041] The constructed intervention waveform is injected through a programmable polarization surface to form a reverse energy channel with directional and phase control capabilities in the spectral space, thereby achieving disturbance loop release in the physical dimension. In the specific operation, firstly, based on the frequency position and phase distribution information in the intervention waveform, this information is translated into spatial control instructions for the polarization unit array, specifying the activation time, rotation angle, and polarization direction of each polarization control unit. Subsequently, each nanounit in the metasurface structure is driven electronically to achieve polarization direction and phase modulation of the incident microwave signal. Within the control activation time window, the phase vector carried by the intervention waveform achieves spatial conjugate superposition between the incident wave and the disturbance signal, generating a spatial beam with reverse intervention capabilities. This beam is symmetrical to the original disturbance signal in the spectrum, overlaps with the disturbance path in the time dimension, and generates conjugate interference through phase modulation, thereby constructing an energy reverse flow path with the leakage frequency band as its core in the energy transmission channel. This path has the ability to repair the accumulation of disturbance energy and reduce the spread of harmonic energy in real time. It can dynamically eliminate the abnormal frequency response caused by false disturbances without affecting the normal signal passage, and finally achieve closed-loop release of the malfunctioning signal source.

[0042] This invention achieves highly sensitive capture of transient disturbances through spectrum suspicion generation and constructs a cross-frequency migration fingerprint using the fusion features of energy center trajectory, group delay curve, and phase derivative sequence, thereby improving the accurate identification of pseudo-frequency band energy components. Based on this, by constructing a misjudgment tracing chain, the origin time and starting frequency of spectrum leakage can be accurately traced, effectively separating false low-frequency signals and preventing them from being incorrectly labeled as actual fault sources in subsequent identification stages. Furthermore, the system improves the robustness of spectrum discrimination by introducing a self-distillation training mechanism and accurately corrects erroneous identification results using corrected frequency band labels, thus providing a reliable basis for downstream linkage suppression and control decisions. Finally, through time-reversal phase gating and conjugate suppression mechanisms, reverse energy intervention in the physical frequency domain is achieved, completely interrupting the triggering path of malfunctions and ensuring stable operation and uninterrupted tripping of new energy equipment under dynamic grid-connected conditions. Overall, the solution has beneficial technical effects such as high identification accuracy, low misjudgment rate, fast control response, accurate energy intervention, and strong closed-loop capability, which significantly improves the anti-disturbance capability and operational safety of new energy power plants in the face of complex disturbance scenarios.

[0043] The new energy broadband test analysis and control system includes a spectrum suspicion generation module, a cross-frequency fingerprint construction module, a traceability chain analysis module, a robust discrimination module, a linkage suppression module, and a dynamic control module. The spectrum suspect generation module obtains the multi-window spliced ​​spectrum at the load switching edge, overlays phase-locked consistency verification on the spliced ​​spectrum result, and generates a spectrum leakage suspect map, which is used as a boundary constraint for energy trajectory extraction. The cross-frequency fingerprint construction module extracts the energy center trajectory, group delay curve and phase derivative sequence under the constraint of the spectrum leakage suspect map to construct a cross-frequency migration fingerprint, which is used as the basic evidence for tracing the source of misjudgments. The source chain analysis module constructs a misjudgment source chain under the constraint of cross-frequency migration fingerprint, maps the cross-frequency migration fingerprint to continuous micro-windows before and after the trigger, and quantifies the contribution of low-order harmonic criteria. The robust discrimination module trains a robust discriminator under the constraint of the misjudgment tracing chain. It enhances the characteristic consistency of the energy center trajectory, group delay curve and phase derivative sequence through self-distillation and outputs the corrected frequency band label to provide a basis for linkage suppression. The linkage suppression module generates linkage suppression commands under the constraints of the corrected frequency band labels, forms a short dwell time delay window through dual-threshold cross confirmation, establishes a fixed-point handling list, and provides input conditions for dynamic control. The dynamic control module generates a dynamic control sequence under the constraint of the linkage suppression command, enables the time-reversal phase gating mechanism, drives the programmable polarization metasurface to inject the conjugate suppression envelope, and constructs the reverse energy channel in the leakage frequency band.

[0044] The new energy broadband test analysis and control method provided in this embodiment of the invention is implemented through the aforementioned new energy broadband test analysis and control system. For details of the specific methods and processes of the new energy broadband test analysis and control system, please refer to the embodiments of the aforementioned new energy broadband test analysis and control method, which will not be repeated here.

[0045] 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 method for testing, analyzing, and controlling new energy broadband, characterized in that, Includes the following steps: The multi-window spliced ​​spectrum at the load switching edge is obtained, and phase-locked consistency verification is superimposed on the spliced ​​spectrum result to generate a spectrum leakage suspicion map, which is used as a boundary constraint for energy trajectory extraction. Under the constraint of the spectrum leakage suspect map, the energy center trajectory, group delay curve and phase derivative sequence are extracted to construct cross-frequency migration fingerprint, which is used as the basic evidence for tracing the source of misjudgment; Under the constraint of cross-frequency migration fingerprint, a misjudgment tracing chain is constructed, and the cross-frequency migration fingerprint is mapped to the continuous micro-windows before and after the trigger to quantify the contribution of low-order harmonic criteria. Under the constraint of misjudgment tracing chain, a robust discriminator is trained, and the characteristic consistency of energy center trajectory, group delay curve and phase derivative sequence is enhanced by self-distillation. The corrected frequency band label is output to provide a basis for linkage suppression. Under the constraints of the corrected frequency band labels, a linkage suppression command is generated, a short dwell time delay window is formed through dual threshold cross confirmation, a fixed-point disposal list is established, and input conditions are provided to dynamic control. Under the constraint of the linkage suppression command, a dynamic control sequence is generated, the time-reversal phase gating mechanism is enabled, and the programmable polarization metasurface is driven to inject the conjugate suppression envelope to construct the reverse energy channel in the leakage band.

2. The new energy broadband test, analysis, and control method according to claim 1, characterized in that, The steps for generating a spectrum leakage suspected map are as follows: Extract the rate of change and acceleration of change of the voltage sampling sequence and the current sampling sequence, and extract the disturbance edge time window by combining the rate of change threshold and the power slope fitting method; Multiple time-scale weighted analysis windows are constructed within the edge time window, Fourier transforms are performed on each window, and frequency axis interpolation is aligned. The spliced ​​spectrum is obtained by splicing the windows using an overlapping weighted method. The phase trajectory of frequency points is extracted from the spliced ​​spectrogram, a multi-window phase evolution sequence is constructed, and the phase change rate is calculated to generate a phase consistency score map; By combining the amplitude coherence and phase consistency scores of frequency points, abnormal frequency bands in the spectrum are extracted, forming a spectrum leakage suspect map with frequency as the horizontal axis and time window index as the vertical axis, and outputting the time frequency distribution results of suspected leakage frequency bands in the spectrum.

3. The new energy broadband test, analysis, and control method according to claim 2, characterized in that, The cross-frequency migration fingerprint construction process is as follows: Within the frequency band covered by the suspected spectrum leakage map, the energy center trajectory of the frequency distribution in each time window is extracted to form a continuous frequency change path; Based on the energy center trajectory, the group delay curve in the frequency domain is calculated to identify the propagation trend, inflection point, and negative value segment in frequency migration. By combining the group delay curve, the phase derivative sequence of the target frequency point in the time dimension is extracted, the segment with drastic phase change is identified, and the key frequency position is marked. The energy center trajectory, group delay curve, and phase derivative sequence are normalized and fused to construct a complete cross-frequency migration fingerprint, which is used for subsequent source analysis of spectrum misjudgments.

4. The new energy broadband test analysis and control method according to claim 3, characterized in that, The steps for constructing a tracing chain for misjudgments are as follows: Based on cross-frequency migration fingerprint images, the start and end frequencies of energy trajectories are extracted, and the boundary ranges of the frequency axis and time axis are determined. A continuous micro-time window sequence is constructed within the fingerprint boundary range, and the spectral amplitude, phase value and signal-to-noise ratio of frequency points are extracted to form candidate misjudgment paths; Connect the candidate frequency paths in chronological order to construct a misjudgment tracing chain; The contribution intensity of the misjudgment tracing chain to the low-order harmonic criterion in the micro-window before and after the disturbance is evaluated, and the low-order harmonic misjudgment contribution matrix is ​​generated. Based on the contribution matrix, frequency points with sudden energy increases after disturbance and no harmonic characteristics before disturbance are identified. Amplitude attenuation compensation and phase continuity reconstruction are performed to complete the spectrum stripping process.

5. The new energy broadband test analysis and control method according to claim 4, characterized in that, The corrected frequency band label output steps are as follows: Training samples were constructed based on the misjudgment tracing chain. Three types of features were extracted: energy center trajectory, group time delay curve and phase derivative sequence. A three-dimensional array structure corresponding to time window index, frequency point position and feature channel was established. Labeling was completed and a complete training dataset was formed. Based on the constructed training dataset, a discriminative model is built using a self-distillation training method. The teacher model provides intermediate feature response benchmarks, while the student model strengthens the consistent expression of the three types of features under the joint constraints of label error and feature response error. The trained discriminant model is used to perform full-band inference on the target spectrum data, extract the frequency point classification results and corresponding feature trends, and generate corrected frequency band labels, which are used as the basis for subsequent linkage suppression operations.

6. The new energy broadband test analysis and control method according to claim 5, characterized in that, In the self-distillation training method, the feature response of the student model is aligned layer by layer with the intermediate feature maps of the teacher model in three feature channels: energy center trajectory, group delay curve and phase derivative sequence. The Euclidean distance is used as the optimization objective to enhance the model's discriminative consistency in the frequency perturbation boundary region.

7. The new energy broadband test analysis and control method according to claim 5, characterized in that, The steps for generating the linkage suppression instruction are as follows: Based on the corrected frequency band labels, frequency points that are continuously marked as non-true harmonic categories are extracted, a frequency dwell time map is constructed, and the dwell stability coefficient is calculated to screen out abnormal frequency points with high dwell characteristics. Based on the frequency dwell time map, a dual verification of frequency trajectory stability threshold and energy mutation threshold is introduced for the selected frequency points, and only frequency points that simultaneously meet the two threshold conditions are retained as trigger targets. Based on the center time window of the dual-threshold passing frequency point, a short dwell time window is formed by expanding forward and backward, and the disturbance amplitude, phase derivative and group delay characteristics are extracted and a fixed-point handling list is constructed. Based on the list of designated disposal points, a coordinated suppression command is generated, which clarifies the frequency location, time window range, disturbance direction and response measures, forming a coordinated response sequence that is time-ordered and resolves conflicts.

8. The new energy broadband test analysis and control method according to claim 7, characterized in that, Under the constraint of the linkage suppression command, a dynamic control sequence is generated, which triggers time-reversal phase gating and drives the programmable polarization metasurface to inject the conjugate envelope. The steps to construct the reverse energy channel in the leakage band are as follows: Based on the frequency points, time indices, disturbance directions, and response methods contained in the linkage suppression command, a set of control parameters including time, phase, and disturbance trends is constructed, and the phase transition path is extracted to determine the true disturbance boundary. A dynamic control sequence is generated based on the constructed set of control parameters. Using this sequence as input, a time-reversal phase-gating operation is performed to reverse the phase trajectory and construct a conjugate intervention waveform for energy reverse intervention. The conjugate interference waveform is translated into spatial control commands, which drive the programmable polarization structure to activate polarization units and form a spatial beam.

9. A new energy broadband test analysis and control system, used to implement the new energy broadband test analysis and control method and system according to any one of claims 1-8, characterized in that, It includes a spectrum suspect generation module, a cross-frequency fingerprint construction module, a source chain analysis module, a robust discrimination module, a linkage suppression module, and a dynamic control module: The spectrum suspect generation module obtains the multi-window spliced ​​spectrum at the load switching edge, overlays phase-locked consistency verification on the spliced ​​spectrum result, and generates a spectrum leakage suspect map, which is used as a boundary constraint for energy trajectory extraction. The cross-frequency fingerprint construction module extracts the energy center trajectory, group delay curve and phase derivative sequence under the constraint of the spectrum leakage suspect map to construct a cross-frequency migration fingerprint, which is used as the basic evidence for tracing the source of misjudgments. The source chain analysis module constructs a misjudgment source chain under the constraint of cross-frequency migration fingerprint, maps the cross-frequency migration fingerprint to continuous micro-windows before and after the trigger, and quantifies the contribution of low-order harmonic criteria. The robust discrimination module trains a robust discriminator under the constraint of the misjudgment tracing chain. It enhances the characteristic consistency of the energy center trajectory, group delay curve and phase derivative sequence through self-distillation and outputs the corrected frequency band label to provide a basis for linkage suppression. The linkage suppression module generates linkage suppression commands under the constraints of the corrected frequency band labels, forms a short dwell time delay window through dual-threshold cross confirmation, establishes a fixed-point handling list, and provides input conditions for dynamic control. The dynamic control module generates a dynamic control sequence under the constraint of the linkage suppression command, enables the time-reversal phase gating mechanism, drives the programmable polarization metasurface to inject the conjugate suppression envelope, and constructs the reverse energy channel in the leakage frequency band.

Images