A method for suppressing power grid harmonics in active-passive cooperation and a composite filter

Abstract

The application discloses a power grid harmonic suppression method and a composite filter, performs harmonic coupling characteristic analysis on power grid multi-source operation data, constructs a harmonic coupling characteristic spectrum and determines a passive compensation interval, introduces a mistuning polarity correction to dynamically correct the boundary, and identifies the compensation competition relationship of the active and passive in the overlapping frequency band; performs rolling harmonic prediction on the compensation division graph, extracts dominant order migration characteristics and maps them to the response time delay of each branch, generates layered compensation activation parameter groups through continuous conflict identification and activation sorting; detects the current zero-crossing point of the preferentially activated branch and generates a switching time window through time sequence misplacement allocation, and combines the predicted residual feedback and the branch adjustable margin to output a composite filter coordination control instruction. The application realizes dynamic coordination of active and passive compensation frequency band division and fine alignment of switching time sequence, and improves the control accuracy and stability of collaborative compensation under dynamic working conditions.

Description

Technical Field

[0001] This invention relates to the field of power quality control technology, and in particular to a method for active and passive coordinated power grid harmonic suppression and a composite filter. Background Technology

[0002] With the continuous increase of power electronic equipment and nonlinear loads in distribution networks, harmonic pollution is becoming increasingly prominent. Passive filters are prone to deviating from their design values ​​due to component aging and lack the ability to suppress harmonics beyond the tuning order. Active filters have wideband dynamic compensation capabilities, but their margin is insufficient under heavy load conditions due to converter capacity constraints. Combining the operation of these two types of devices is a common solution that balances cost and performance; however, existing collaborative control methods have significant shortcomings: there is a lack of competition identification when the two compensations generate reverse injection in the same frequency band; the response speeds of active and passive branches differ significantly, and there is a lack of a time-delay-based activation sequencing mechanism during collaborative switching; harmonic distribution continuously migrates with the load, and the compensation weights fail to be dynamically corrected in conjunction with prediction deviations, thus limiting control accuracy.

[0003] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the Invention

[0004] This invention discloses a method for active and passive coordinated power grid harmonic suppression and a composite filter. It aims to achieve dynamic boundary correction and frequency band competition identification of the passive compensation interval through harmonic coupling characteristic analysis, and to achieve fine coordination of active and passive activation timing by combining rolling harmonic prediction and response delay sorting. Furthermore, it generates compensation weights through zero-crossing switching window allocation and residual feedback correction, and finally outputs composite filter coordinated control commands, providing a dynamic coordinated and time-aligned harmonic control means for distribution networks containing multiple types of nonlinear loads.

[0005] The first aspect of this invention proposes a method for active and passive coordinated power grid harmonic suppression, comprising the following steps: Collect multi-source power grid operation data, and construct a harmonic coupling characteristic spectrum based on the analysis of harmonic coupling characteristics of the multi-source operation data; The passive compensation interval is determined based on the harmonic coupling characteristic spectrum. The detuning polarity characteristics of the passive branch are extracted from the multi-source operation data to generate detuning polarity correction coefficients. The detuning polarity correction coefficients are used to perform dynamic boundary correction on the passive compensation interval to form a compensation division map. Based on the compensation division map, the compensation competition interval is identified and a competition compensation identifier is generated. Rolling harmonic prediction is performed on the competitive compensation identifier and the compensation division map to generate a predicted harmonic sequence. The dominant number migration identifier is extracted from the predicted harmonic sequence. Compensation response mapping is performed on the dominant number migration identifier to obtain the filter response delay. The compensation priority is evaluated based on the filter response delay to generate a hierarchical compensation activation parameter group. The hierarchical compensation activation parameter group is analyzed for coordination constraints to determine the priority activation branch. The current zero-crossing switching window of the priority activation branch is detected. The branch adjustment margin is determined based on the current zero-crossing switching window. The compensation weight output composite filter coordination control command is generated by combining the residual feedback characteristics of the predicted harmonic sequence and the branch adjustment margin.

[0006] A second aspect of this invention provides an active-passive synergistic power grid harmonic suppression composite filter, comprising: The data acquisition module is used to collect multi-source operation data of the power grid and construct a harmonic coupling characteristic spectrum based on the analysis of harmonic coupling characteristics of the multi-source operation data. The compensation division module is used to determine the passive compensation interval based on the harmonic coupling characteristic spectrum, extract the detuning polarity characteristics of the passive branch from the multi-source operation data to generate detuning polarity correction coefficients, use the detuning polarity correction coefficients to perform dynamic boundary correction on the passive compensation interval to form a compensation division map, and identify the compensation competition interval based on the compensation division map to generate a competition compensation identifier. The prediction scheduling module is used to perform rolling harmonic prediction on the competition compensation identifier and the compensation division map to generate a prediction harmonic sequence, extract the dominant number migration identifier from the prediction harmonic sequence, perform compensation response mapping on the dominant number migration identifier to obtain the filter response delay, and evaluate the compensation priority based on the filter response delay to generate a hierarchical compensation activation parameter group. The coordination output module is used to perform coordination constraint analysis on the hierarchical compensation activation parameter group to determine the priority activation branch, detect the current zero-crossing switching window of the priority activation branch, determine the branch adjustment margin based on the current zero-crossing switching window, and generate a compensation weight output composite filter coordination control command by combining the residual feedback characteristics of the predicted harmonic sequence and the branch adjustment margin.

[0007] The beneficial effects of this invention are reflected in the following points: 1. By analyzing the cross-order correlation of harmonic amplitudes, harmonics from the same source are classified into coupled clusters. Based on the clustering results, the passive compensation interval is determined, and polarity correction is introduced based on the resonant frequency offset direction of the passive branch to adjust the interval boundary in real time. At the same time, the reverse competition relationship between active and passive harmonics in overlapping frequency bands is identified. This mechanism enables the passive compensation interval to be adaptively corrected as the component parameters drift, and issues an early warning in frequency bands where the compensation directions cancel each other out, avoiding the degradation of compensation effect caused by fixed tuning offset and frequency band competition. 2. Rolling harmonic prediction is performed based on the compensation division map and competitive compensation identifiers. The migration characteristics of the dominant harmonic order are identified and mapped to the measured response delay of each branch. The activation order is reordered by identifying the continuous conflict section of active and passive delays. This mechanism makes the arrival time of the two compensation currents tend to be aligned, reducing the control blind zone caused by the difference in response delay during the coordinated switching process. 3. By detecting the current zero-crossing point and allocating the timing misalignment of multiple branches, a switching window for each branch is generated. The upper limit of the compensation weight is constrained by the adjustable margin of the branch, and predictive residual feedback is introduced to adaptively correct the weight and output composite filter coordinated control command. This mechanism ensures that the zero-crossing switching suppresses operational overvoltage while dynamically adjusting the compensation weight towards the actual harmonic deviation direction, thereby improving the control accuracy and operational stability of active and passive collaborative compensation. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating an active-passive coordinated power grid harmonic suppression method according to the present invention.

[0009] Figure 2 This is a structural block diagram of a composite filter for power grid harmonic suppression that combines active and passive operation according to the present invention. Detailed Implementation

[0010] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0011] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0012] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0013] The technical solutions of the embodiments of this application are described below.

[0014] like Figure 1 As shown, this embodiment of the invention provides a method for active and passive coordinated power grid harmonic suppression, including the following steps S110-S140: Step S110: Collect multi-source operation data of the power grid, and construct the harmonic coupling characteristic spectrum based on the harmonic coupling characteristics analysis of the multi-source operation data.

[0015] Specifically, multi-source operational data is collected from various monitoring nodes in the power grid. The data collection targets three types of signal sources: voltage transformers, current transformers, and power metering devices. The sampling frequency of each signal source is uniformly set to 64 times the fundamental frequency to ensure that harmonic components of the 32nd order and below can be sufficiently distinguished. Insufficient sampling frequency will result in aliasing distortion, making it impossible to accurately reconstruct the amplitude and phase information of higher harmonics. The data collection locations are selected from three types of nodes: points of common coupling (PCC), main feeder branch points, and compensation device access points. Data from PCC reflects the overall harmonic level of the power grid, while data from feeder branch points reveals the propagation path of harmonics in the network. If the harmonic amplitude collected from a feeder branch point is significantly higher than that from PCC, it indicates the presence of a local harmonic injection source in that branch. Synchronization between multi-source signals is achieved through global positioning clock (GPS) timing. The timestamp accuracy of each collection node is better than 1 μs. When the synchronization error exceeds this accuracy, the data segment of the corresponding node is marked as a time-series offset. Time-series offset data is not included in the joint harmonic analysis across nodes. Multi-source operational data preprocessing includes two steps: outlier detection and baseline drift correction. Outlier detection is defined as a single point whose amplitude exceeds five times the recent average of similar signals. Points exceeding this criterion are replaced by linear interpolation of adjacent valid time intervals. Baseline drift correction is achieved by subtracting the low-frequency trend component, which is the result of low-pass filtering the original signal with a cutoff frequency of 0.1 times the fundamental frequency. The effective acquisition time of each signal source in the multi-source operational data must be no less than 10 fundamental cycles. Signal segments with insufficient effective time will have incomplete additional data labeling, and harmonic analysis conclusions corresponding to incomplete multi-source operational data will have their confidence weights reduced.

[0016] In some embodiments, the step of constructing a harmonic coupling feature spectrum based on the harmonic coupling characteristics analysis of the multi-source operating data includes: extracting the amplitude time series of the multi-source operating data according to the harmonic order to generate a harmonic amplitude time series group; performing cross-order synchronous fluctuation detection on the harmonic amplitude time series group to generate an inter-harmonic correlation distribution; generating homologous harmonic clustering labels by clustering the inter-harmonic correlation distribution according to threshold polarity; and performing joint feature mapping on the homologous harmonic clustering labels and the harmonic amplitude time series group to generate a harmonic coupling feature spectrum.

[0017] The amplitude timing sequence of multi-source operating data is extracted according to harmonic order to generate a harmonic amplitude timing sequence group. The short-time Fourier transform window length is 10 fundamental cycles. When the window length is insufficient, the frequency resolution decreases, leading to increased amplitude leakage between adjacent harmonic orders, thus compromising the independence of the amplitude timing sequence of each harmonic. The extracted harmonic order range covers 2nd to 32nd. Harmonics above 32nd are attenuated to negligible levels by the conduction impedance of most power equipment and are not included in the extraction range. The extraction results of each signal source in the multi-source operating data correspond to independent amplitude timing sequences. The amplitude timing sequences of different signal sources under the same harmonic order are aligned in the time dimension. The alignment is achieved based on the synchronization timestamp of each acquisition node. The 5th harmonic amplitude timing sequence of a node and the 5th harmonic amplitude timing sequence of adjacent nodes start and end at the same time, ensuring the timing consistency of subsequent cross-node fluctuation detection. The amplitude time series of each harmonic is arranged in ascending order of harmonic number. The set of amplitude time series of all harmonic numbers constitutes the harmonic amplitude time series group. In the harmonic amplitude time series group, the harmonic number whose amplitude is consistently lower than 0.1% of the fundamental amplitude of the multi-source operating data is marked as a low amplitude number. The amplitude time series corresponding to the low amplitude number is retained, but its weight is reduced in the subsequent correlation detection to avoid noise-dominated low amplitude time series interfering with the judgment of cross-number correlation.

[0018] Harmonic amplitude time series groups are subjected to cross-order synchronous fluctuation detection to generate inter-harmonic correlation distributions. The sliding window Pearson correlation coefficient method is employed, with a window length of 5 fundamental cycles and a step size of 1 fundamental cycle. Within each window, pairwise correlation coefficients are estimated for all harmonic order pairs in the harmonic amplitude time series group. The amplitude time series corresponding to low-amplitude orders in the harmonic amplitude time series group are assigned a 0.5-fold weighting factor when participating in correlation coefficient estimation. This weighting ensures that noise fluctuations of low-amplitude orders do not produce spurious high correlations with the true fluctuations of other orders. If the amplitude of a certain harmonic is close to the noise floor, its time series fluctuations are dominated by measurement noise; without weighting, the correlation coefficients with other orders may show misleadingly high values. The correlation coefficients of each harmonic order pair on the sliding window sequence are averaged over time. This average value serves as the representative correlation for that harmonic order pair. The representative correlations of all harmonic order pairs together constitute the inter-harmonic correlation distribution. A larger absolute value of the representative correlation indicates a higher degree of synchronization between the corresponding two harmonic fluctuations. A negative correlation indicates that the amplitudes of the two harmonics change in opposite directions in sync. This situation may occur when the active compensation device overcompensates a certain harmonic. In the inter-harmonic correlation distribution, harmonic order pairs with a representative correlation absolute value below 0.3 are marked as weakly correlated. Weakly correlated order pairs are not included in the identification of subsequent high-correlation order groups.

[0019] Harmonic cluster labels are generated by clustering harmonics of the same origin using threshold polarity based on the inter-harmonic correlation distribution. The correlation threshold is set to 0.7. Below this value, the cluster boundaries are too loose, which may misclassify harmonics driven by different excitation sources into the same group. Above this value, the boundaries are too strict, which may incorrectly split harmonics of the same origin into multiple groups due to a decrease in correlation caused by brief disturbances. 0.7 is the critical value that strikes a balance between incorrect merging and incorrect splitting. Pairs of positively correlated and negatively correlated high-value harmonics are processed separately. Pairs of positively correlated high-value harmonics are classified into the same-direction, same-origin cluster, while pairs of negatively correlated high-value harmonics are classified into the opposite-direction, same-origin cluster. The two types of clusters are each assigned a cluster polarity attribute to distinguish between homogeneous relationships with the same and opposite oscillation directions. A nonlinear load simultaneously injects 3rd and 5th harmonics, and the amplitudes of both change in the same direction with the load power, so it is classified into the positively correlated, same-direction cluster. Candidate clusters are expanded using a single-link merging rule: if frequencies 3 and 5 belong to the same candidate cluster, and frequencies 5 and 7 represent correlations with absolute values ​​exceeding 0.7, then all three are merged into the same cluster. Frequency frequencies that cannot be assigned to any cluster constitute individual single-element clusters. These single-element clusters do not share homology features with other frequencies, and their cluster polarity attribute is marked as non-polar by default. All clusters and their contained harmonic frequencies, along with their cluster polarity attributes, together constitute homology harmonic cluster labels. The more frequencies within a cluster in a homology harmonic cluster label, the wider the harmonic range influenced by the excitation source; the fewer the total number of clusters in a homology harmonic cluster label, the more concentrated the homology relationship revealed by the harmonic correlation distribution.

[0020] A joint feature mapping is performed on the harmonic clustering labels and harmonic amplitude time series groups to generate harmonic coupling feature spectra. For each cluster in the harmonic clustering labels, the amplitude time series of the corresponding order is extracted from the harmonic amplitude time series group. These are then weighted and superimposed according to the cluster polarity attribute to generate a representative time series for the cluster. In the same-direction cluster, the amplitude time series of each order are directly weighted and averaged. In the opposite-direction cluster, the amplitude time series of orders with opposite polarities are inverted before being weighted and averaged. The weighting coefficients are determined by the long-term average of the amplitude time series of each order; the larger the long-term average, the higher the weight of the order to highlight the contribution of the dominant order. The representative time series of the single-element cluster in the harmonic clustering labels directly uses the original amplitude time series of the corresponding order in the harmonic amplitude time series group, without superimposing it with other orders. The correlation between a certain isolated harmonic order and the other orders is all below a threshold, and its representative time series independently reflects the individual time-domain variation law of that order. The harmonic coupling characteristic spectrum is composed of three types of information: the representative time series of all clusters, the set of harmonic orders contained in each cluster, and the polarity attribute of the clusters. The representative time series provides a dynamic perspective in the time domain, the set of orders provides a perspective in the frequency domain grouping, and the polarity attribute provides a perspective on the coupling direction. The combination of these three types of information enables the harmonic coupling characteristic spectrum to fully describe the interrelation patterns of harmonics at different orders and their evolution characteristics over time. Clusters with large cluster sizes and high representative time series amplitudes in the harmonic coupling characteristic spectrum are given priority as references for subsequent compensation interval delineation. Clusters with representative time series amplitudes that are consistently lower than 1% of the fundamental amplitude are given a low-priority label in the harmonic coupling characteristic spectrum.

[0021] Step S120: Determine the passive compensation interval based on the harmonic coupling characteristic spectrum, extract the detuning polarity characteristics of the passive branch from the multi-source operation data to generate the detuning polarity correction coefficient, use the detuning polarity correction coefficient to perform dynamic boundary correction on the passive compensation interval to form a compensation division map, and identify the compensation competition interval based on the compensation division map to generate a competition compensation identifier.

[0022] Specifically, the passive compensation interval is determined based on the harmonic coupling characteristic spectrum. A 5% amplitude threshold is used as the criterion for passive compensation intervention. Below this threshold, the reactive power loss introduced by the passive filter exceeds its compensation benefit, making the investment in passive branches energy-inefficient. Clusters in the harmonic coupling characteristic spectrum representing time-series amplitudes consistently exceeding 5% of the fundamental amplitude are identified as strong harmonic clusters. The harmonic orders covered by strong harmonic clusters are mapped as continuous frequency bands on the frequency axis. Frequency bands of clusters with an interval of no more than two adjacent orders are merged into a single passive compensation interval. Intervals with an interval of more than two orders are retained as independent intervals to avoid forcibly merging different clusters with significant energy differences into the same compensation interval, which could lead to filter tuning point shifts. When determining the passive compensation interval, the frequency bands corresponding to the reverse homogeneous clusters in the harmonic coupling characteristic spectrum are assigned a polarity reversal attribute. The polarity reversal attribute indicates that the equivalent admittance polarity of the passive branch in this frequency band is naturally opposite to the direction of the active compensation current injection. A phase margin must be reserved for the active side when defining the interval. The boundaries of the passive compensation interval are represented by the upper and lower limits of the harmonic order. The upper limit is the highest order within the cluster plus 0.5, and the lower limit is the lowest order minus 0.5. The boundary is extended by 0.5 to cover slight drifts in the resonant frequency caused by temperature or load changes. When the drift exceeds the extended range, the corresponding passive compensation interval boundary is dynamically corrected. Clusters with low priority labeling of harmonic coupling characteristic spectra are temporarily deferred for determining the passive compensation interval. They will be re-evaluated for inclusion after a cumulative total of 10 fundamental frequency cycles of newly acquired valid data.

[0023] In some embodiments, the step of extracting the detuning polarity features of the passive branch from the multi-source operating data to generate detuning polarity correction coefficients includes: extracting the resonant frequency offset of the passive branch from the multi-source operating data to generate a frequency offset sequence; performing time-series difference analysis on the frequency offset sequence to generate a detuning rate of change distribution; identifying the offset direction based on the detuning rate of change distribution to generate a capacitive offset marker set and an inductive offset marker set; and constructing a leading-backward correction amount based on the capacitive offset marker set and the inductive offset marker set to generate detuning polarity correction coefficients.

[0024] Frequency offset sequences are generated by extracting the resonant frequency offset of passive branches from multi-source operating data. The resonant frequency of a passive branch is determined by the nominal parameters of the filter inductor and filter capacitor. The capacitor value drifts with temperature rise and aging, while the inductor value deviates from the nominal value due to current saturation. The combined effect of these two factors causes the actual resonant frequency to continuously deviate from the design tuning point. The resonant frequency offset is extracted from the spectral amplitude of the multi-source operating data. The minimum amplitude point is searched near the design tuning frequency. The difference between the frequency corresponding to the minimum point and the design tuning frequency is the resonant frequency offset at the current moment. A positive offset indicates that the resonant frequency is drifting towards higher frequencies, while a negative offset indicates that it is drifting towards lower frequencies. The offsets of each passive branch in the multi-source operating data are arranged according to the sampling time to form the offset sequence of each branch. When a certain branch is continuously under heavy load, the offset is stable and positive with a gradually increasing amplitude, indicating that the inductor value of this branch is decreasing due to current saturation, and the resonant frequency is continuously shifting upward. The frequency offset sequence is composed of the time series offsets of each branch. Branches with an absolute offset value exceeding 1% of the design tuning frequency are marked as significant offset branches, which are prioritized in the time series differential calculation. Branches with offsets that are consistently close to zero are marked as stable branches. The offset time series of stable branches are retained in the frequency offset sequence, but their differential weight is reduced to 0.3 times. In multi-source operation data, if the data of a passive branch corresponding to a certain acquisition node is missing for more than 3 fundamental frequency cycles, the corresponding time period of that branch is marked as a missing segment in the frequency offset sequence, and the missing segment is not included in the offset statistics.

[0025] A detuning rate distribution is generated by time-series differential analysis of the frequency offset sequence. The differential step size is set to a single fundamental period. If the step size is too short, sampling jitter caused by thermal noise is amplified into a spurious high rate; if the step size is too long, short-term rapid offset events are smoothed out by the mean. A single fundamental period strikes a reasonable balance between the two. The offset rate at each time step is obtained by subtracting the offset time steps of adjacent time steps in the frequency offset sequence and dividing by the sampling interval. A large absolute value of the rate indicates that the resonant frequency is shifting rapidly, while a rate close to zero indicates that the resonant frequency is in a relatively stable state. The differential results of stable branches in the frequency offset sequence are usually concentrated near zero. Occasionally, points that exceed zero reflect measurement noise interference rather than actual changes. Dead-zone filtering with ±0.02% of the tuning frequency is applied to the differential results of these branches, and the differential values ​​within the dead-zone range are forced to zero to eliminate the influence of noise. The time sequence of rate changes of all branches together constitutes the detuning rate distribution. The overall level of the absolute value of the rate in the detuning rate distribution reflects the overall dynamic characteristics of the resonant frequency of the passive branches within the current time period. If the rate of most branches increases simultaneously within a certain time period, it indicates the presence of a systematic disturbance affecting the global capacitance or inductance parameters. The rate amplitude of significantly offset branches in the detuning rate distribution is usually higher than that of stable branches. The difference in amplitude between the two types of branches can help determine the reliability of the offset polarity. The time corresponding to the missing segment in the frequency offset sequence is synchronously marked as an invalid time in the detuning rate distribution.

[0026] For example, the step of identifying the offset direction based on the detuning rate of change distribution and generating a capacitive offset marker set and a sensitive offset marker set includes: extracting the offset symbol distribution from the detuning rate of change distribution according to the segment amplitude to generate an offset symbol sequence; identifying continuously flipped segments of the offset symbol sequence to generate a polarity reversal event marker set; filtering the polarity reversal event marker set for reversal amplitude exceeding a threshold to generate a gradual switching trigger marker; and performing gradual direction calibration on the offset symbol sequence based on the gradual switching trigger marker to generate a capacitive offset marker set and a sensitive offset marker set.

[0027] The offset symbol distribution is generated by extracting the offset symbol distribution according to the amplitude of each segment of the detuning rate distribution. The ternary encoding uses +1, -1, and 0 to correspond to positive, negative, and zero offset states, respectively. This ternary structure decouples direction and amplitude information, allowing subsequent flip recognition to focus only on direction changes without being affected by amplitude magnitude. Symbols are extracted from the rate time series of each branch in the detuning rate distribution at each sampling time. Symbols are set to 0 when the absolute value of the rate is below the dead zone threshold. The dead zone threshold is set to 0.02% of the design tuning frequency of the corresponding branch. This value is consistent with the dead zone filtering threshold applied during the detuning rate distribution generation stage. Using the same threshold in both stages prevents some offset events from being filtered in one place and retained in another. The ternary symbol time series of each branch together constitute the offset symbol sequence. Multiple consecutive zero-value segments in the offset symbol sequence indicate that the resonant frequency of that branch is highly stable within the corresponding time period. For example, if the symbol for a branch is +1 for eight consecutive sampling times, the corresponding resonant frequency continuously shifts upwards. Branches in the offset symbol sequence whose zero-value segments persist for more than 20 sampling times are considered long-term stable branches. Long-term stable branches only need to detect the sudden appearance of non-zero values, rather than full-scan flip events. This differentiated processing reduces the unnecessary computational overhead for a large number of stable branches. The positions corresponding to invalid times in the detuning rate distribution are marked as invalid symbols in the offset symbol sequence. Invalid symbols are not included in the counting of consecutive flip segments, avoiding the artificial truncation of valid direction segments due to missing data.

[0028] The offset symbol sequence is used to identify consecutive flipped segments to generate a polarity reversal event marker set. A valid polarity reversal requires the opposite symbol to appear continuously for two or more sampling times after the flip. A single isolated flip followed immediately by the original symbol does not constitute a valid reversal. This duration threshold filters out instantaneous symbol jitter caused by measurement noise or switching disturbances. A flip detection is triggered when a non-zero symbol in the offset symbol sequence changes from +1 to -1 or from -1 to +1. After the flip meets the duration threshold, the flip time, the preceding and following symbol values, and the average absolute value of the preceding and following segment rates are recorded. The interval between two adjacent flip events in the offset symbol sequence reflects the switching frequency of the branch's offset direction. A short interval indicates that the branch's resonant frequency frequently oscillates between the two directions. If a branch experiences three flips within 10 sampling times, it suggests that the branch's resonant frequency fluctuates due to periodic disturbances. All flip events are summarized in chronological order to form a polarity reversal event marker set. Branches with a large number of flip events in the polarity reversal event marker set are given priority in the gradual switching screening. Branches with no flip events indicate that the offset direction is monotonically stable within the current time period. When there is a valid symbol on each side of an invalid symbol position in the offset symbol sequence, and the symbol values ​​on both sides are different, this position is considered a possible flip point and is included in the polarity reversal event marker set with a low-confidence flip mark. The amplitude threshold requirement for low-confidence marks is appropriately increased during the screening stage.

[0029] A gradual switching trigger flag is generated by filtering the polarity reversal event marker set based on reversal amplitude exceeding a threshold. The reversal amplitude is the difference between the mean absolute values ​​of the segment rates before and after the reversal in the polarity reversal event marker set. A large difference in mean indicates that the shift in offset direction is accompanied by a substantial change in offset intensity, belonging to a gradual direction shift; a small difference in mean indicates that the direction shift is only a momentary disturbance rather than a trend change. The filtering threshold is set to 30% of the mean absolute value of the rate of change of detuning of all branches in the current period. Reversal events with reversal amplitude exceeding this threshold are retained, while those below the threshold are filtered. After filtering, the remaining events in the polarity reversal event marker set all have sufficient amplitude to support direction calibration. The threshold for low-confidence reversal markers in the polarity reversal event marker set is increased to 45% when participating in the filtering. Low-confidence markers must pass a more stringent amplitude test before entering the gradual switching trigger flag stage, avoiding misjudgments introduced by data gaps from being transmitted to the direction calibration stage. The selected and retained flip events, along with their corresponding branch numbers and trigger times, constitute the gradual switching trigger markers. The trigger time serves as a segment node for the gradual direction calibration. If a branch experiences a threshold flip at the 15th sampling time, this time becomes the boundary point for the branch's offset direction calibration. When the polarity reversal event marker set is empty after filtering, the gradual switching trigger markers are also empty. An empty set indicates that there is no gradual switching in the offset direction of any branch within the current time period, and the direction calibration is uniformly processed for the entire time period.

[0030] Gradual direction calibration is performed on the offset symbol sequence based on gradual switching trigger markers to generate capacitive and inductive offset marker sets. The offset symbol sequence is divided into several time periods using the trigger time of each branch in the gradual switching trigger markers as the dividing point. Segmented statistics avoid the mixing of direction information between the two segments before and after the switch due to global counting across direction switching points, ensuring that the calibration results accurately correspond to the actual dominant offset direction within each time period. Branches with empty gradual switching trigger marker sets are not segmented into time periods. The occurrence counts of positive and negative symbols are uniformly counted for the entire offset symbol sequence. The dominant direction of the statistical results represents the overall offset tendency of that branch throughout the entire analysis period. When the positive symbol count exceeds the negative symbol count in each time period, the branch with the resonant frequency shifting upwards exhibits capacitive detuning at the tuning point, and the corresponding time period branch is included in the capacitive offset marker set. When the negative symbol count exceeds the positive symbol count, it is considered inductive detuning and is included in the inductive offset marker set. When the counts are equal, the time period is marked as having an unclear direction and is not included in any marker set. Invalid symbols in the offset symbol sequence are not included in the positive / negative count, and time periods with fewer than 3 valid symbols are also marked as having an unclear direction. If a branch has a positive dominant direction in the previous segment, it is included in the receptive offset label set; after switching, if the dominant direction becomes negative, it is included in the inductive offset label set. The same branch has one record in each label set. The receptive and inductive offset label sets together cover all time periods with a clear direction in the offset symbol sequence, and the two are complementary and do not overlap.

[0031] Based on the capacitive and inductive offset marker sets, a leading-backward correction amount is constructed to generate the detuning polarity correction coefficient. The principle of leading-backward correction is to apply inductive pre-compensation to the identified capacitive detuning direction, so that the corrected equivalent resonant frequency returns to the design tuning point. The correction amount must be injected in advance before the offset intensifies, rather than passively supplementing after the offset occurs. The offset amplitude of each branch in the capacitive offset marker set is estimated by integrating the cumulative rate of the detuning change rate distribution over the corresponding time period. The integral value reflects the total change in offset of that branch within the marked time period. The correction direction corresponding to the capacitive detuning is defined as positive correction. Similarly, the offset amplitude of each branch in the inductive offset marker set is estimated by integrating, and the correction direction is defined as negative correction. The positive and negative correction values ​​for each branch are multiplied by a normalization coefficient to obtain the corresponding detuning polarity correction coefficient. The normalization coefficient Q_j is determined by the formula Q_j=I_j / D_lim, where I_j is the cumulative integral estimate of the rate of change of offset of branch j within the marked time period (in terms of cycles), and D_lim is the maximum allowable boundary adjustment range of the passive compensation interval in the current time period (in terms of cycles). The value of Q_j is limited to between 0.3 and 1.0. If it is below 0.3, the correction amount is insufficient to offset the offset; if it is above 1.0, overcorrection will cause a reverse offset. When there is no branch marker in either the capacitive offset marker set or the inductive offset marker set, the detuning polarity correction coefficient for that branch is assigned a value of zero. The current resonant frequency of a branch is completely stable, and a correction coefficient of zero indicates that no correction needs to be applied to the compensation boundary of that branch. The detuning polarity correction coefficients are arranged according to the branch number, and each branch has a unique corresponding correction coefficient value. The larger the absolute value of the correction coefficient, the deeper the current detuning degree of the branch and the greater the adjustment range required for the compensation boundary.

[0032] A compensation division map is formed by dynamically correcting the boundaries of the passive compensation interval using detuning polarity correction coefficients. For capacitive detuning branches, the correction coefficient is positive, causing the upper boundary of the passive compensation interval to expand towards higher frequencies and the lower boundary to shrink as the resonant frequency increases. For inductive detuning branches, the correction coefficient is negative, with the opposite direction. The correction magnitude is determined by the product of the absolute value of the correction coefficient and the initial width of the boundary; a larger initial width results in a larger absolute correction magnitude, but the relative proportion remains unchanged. Passive compensation intervals with zero detuning polarity correction coefficients retain their initial values ​​without adjustment. If the resonant frequency of a certain branch is consistently stable with a zero correction coefficient, the boundaries of the corresponding passive compensation interval remain fixed throughout the analysis period. The corrected passive compensation intervals are superimposed on the active compensation allocated frequency bands. The active compensation allocated frequency bands are determined by the rated compensation bandwidth of the active filter and the priority compensation number. The frequency ranges jointly covered by the active segment and the corrected passive segment are marked with their respective compensation responsibility boundaries in the compensation division map. The compensation division diagram uses harmonic order as the horizontal axis. The active layer indicates the range of harmonic orders covered by active compensation and the upper limit of the compensation amount. The passive layer indicates the boundary range and tuning order of the passive compensation interval after correction. When the detuning polarity correction coefficient causes the boundaries of two adjacent passive compensation intervals to overlap, the overlapping section is marked as double passive coverage in the compensation division diagram. The passive compensation capability of the double-covered section is enhanced by superposition, but there is a risk of mutual interference between the two branch resonances, which needs to be carefully reviewed during the competition identification stage.

[0033] In some embodiments, the step of identifying the compensation competition interval and generating a competition compensation identifier based on the compensation division map includes: extracting active compensation allocation intervals and passive compensation allocation intervals from the compensation division map to generate a dual-path compensation interval group; performing interval overlap detection on the dual-path compensation interval group to generate an overlap distribution sequence; identifying the relationship between active and passive compensation directions within the overlap intervals based on the overlap distribution sequence to generate a reverse competition marker set; and using the reverse competition marker set to label the compensation competition intervals in the compensation division map and generate a competition compensation identifier.

[0034] The active and passive compensation allocation intervals are extracted from the compensation distribution map to generate a dual-path compensation interval group. Adjacent segments with an order interval of no more than 0.5 are merged into one interval. The merging threshold of 0.5 is less than the minimum harmonic order step size, ensuring truly continuous compensation coverage without artificial segmentation, and preventing the incorrect merging of independent intervals belonging to different tuning targets. The active and passive layers of the compensation distribution map are extracted separately. The active compensation allocation interval is determined by the order boundaries of each consecutive segment in the active layer, and the passive compensation allocation interval is determined by the order boundaries of each consecutive segment in the passive layer. The merging rules apply to both layers. The number of active compensation allocation intervals in the dual-path compensation interval group is usually less than that of passive compensation allocation intervals. The broadband compensation characteristics of active filters allow them to cover a wider frequency band with fewer intervals. A certain active compensation allocation interval spans from the 3rd to the 11th harmonics, while the corresponding passive side has multiple independently tuned intervals corresponding to the 5th, 7th, and 11th harmonics, respectively. When extracting dual passive coverage segments from the compensation allocation map to the passive compensation assignment interval, double annotations are added. Double-annotated intervals are assigned a 1.5x overlap weight in overlap detection to reflect higher competition risk. The dual-path compensation interval group fully preserves all active and passive coverage information from the compensation allocation map, without discarding any compensation responsibility attribution for any frequency segment. Frequency segments not covered by any interval in the compensation allocation map do not generate corresponding records in the dual-path compensation interval group.

[0035] Interval overlap detection is performed on dual-path compensation interval groups to generate an overlap distribution sequence. When active and passive regions simultaneously bear compensation responsibility in the same frequency band, their compensation currents are injected into the same node at the same time. The superposition effect depends on whether the current directions are consistent; when the directions are consistent, the effect is enhanced, and when the directions are opposite, they cancel each other out. Identifying overlapping intervals lays the foundation for distinguishing these two situations and defining the interval range. All active and passive intervals in the dual-path compensation interval group are intersected pairwise using a range of intersections. For combinations with non-empty intersection results, a common coverage segment is extracted. The width of the common coverage segment serves as a measure of the overlap degree of that combination. For doubly labeled passive intervals, the overlap degree measure is multiplied by 1.5 when intersecting. For example, a dual passive interval and an active interval with an original intersection width of 1 is weighted to have an equivalent overlap degree of 1.5, giving it higher priority than a single passive overlapping interval with an actual width of 1.2. The overlapping distribution sequence consists of all non-empty overlapping results arranged in ascending order of harmonic order. Each item contains two pieces of information: the range of overlapping orders and the weighted degree of overlapping. When there is no overlap between any pair of active and passive intervals in the dual-path compensation interval group, the overlapping distribution sequence is an empty set. An empty set indicates that the active and passive compensations in the current compensation division diagram are completely segmented and do not overlap on the frequency axis, and no competition direction analysis is required. If the interval with the highest weighted degree of overlapping in the overlapping distribution sequence has reverse competition, its negative impact on the compensation effect is the greatest, and it is prioritized for compensation direction relationship analysis.

[0036] Based on the overlapping distribution sequence, the relationship between the active and passive compensation directions within the overlapping interval is identified, generating a set of inverse competition markers. Passive branches tuned to the 5th order exhibit inductive admittance below the 5th order and capacitive admittance above the 5th order. The relative position of the tuning frequency and the center order of the overlapping interval directly determines the polarity of the injected current within the overlapping frequency band. This physical relationship is the fundamental basis for determining whether the active and passive compensation directions are consistent. The direction of the active compensation current corresponding to each overlapping interval in the overlapping distribution sequence is read from the active filter control command. The polarity of the equivalent admittance of the passive compensation within the overlapping frequency band is inferred from the relationship between the tuning order of the corresponding passive branch and the center order of the overlapping interval in the compensation division diagram. Overlapping intervals where the direction of the active current is opposite to the polarity of the passive admittance are identified as reverse competition intervals. The active interval number, the passive interval number, and the range of overlap counts together constitute a reverse competition marker. If the center count of an overlapping interval is slightly higher than the tuning count of the passive branch, the passive branch exhibits capacitive admittance in that frequency band, while the compensation current on the active side in that frequency band is inductively injected. Since the directions of the two are opposite, it is identified as reverse competition. All reverse competition markers are summarized to form a reverse competition marker set. The reverse competition marker set is empty when the overlapping distribution sequence is empty or when all overlapping intervals are in the same direction. Intervals in the overlapping distribution sequence with a weighted overlap degree exceeding 2 and identified as reverse competition are marked with a high priority in the reverse competition marker set.

[0037] Competitive compensation markers are generated by labeling the compensation allocation map using a reverse competition marker set. The competition risk level is divided into three tiers based on the weighted overlap: more than 2 overlaps indicate high competition risk, 1 to 2 overlaps indicate medium competition risk, and less than 1 overlap indicates low competition risk. This tier division is based on engineering experience that the tuning bandwidth of passive filters is typically approximately ±1 tuning order. Overlaps exceeding 2 overlaps cover more than the entire tuning bandwidth, resulting in the most severe damage to the compensation effect. The overlap range corresponding to each marker in the reverse competition marker set is superimposed with competition interval markers in the compensation allocation map. These competition interval markers are presented using a third type of marker, distinct from those for active and passive layers, making the competition intervals visually identifiable in the compensation allocation map. In high-competition-risk intervals, the active side removes this range from its compensation allocation range, and the passive side independently undertakes the filtering task for that frequency band. In medium-competition-risk intervals, the upper limit of the active side's compensation is reduced to 50% of the original value, with the passive side assuming the remaining share. The existing division of labor in low-competition-risk zones will remain unchanged, with only the markings retained for monitoring. The competition compensation identifier is composed of the harmonic order range of all competing zones, the competition risk level, and the active and passive zone numbers involved. When the reverse competition marker set is empty, the competition compensation identifier is also empty. An empty set indicates that in the compensation division map, active and passive compensation have the same direction in all overlapping frequency bands, and the existing division arrangement does not require intervention or adjustment. If the number of high-competition-risk entries in the competition compensation identifier is consistently high, it suggests a systemic conflict in the frequency band boundary delineation of active and passive compensation, requiring a re-examination of the initial delineation basis for the passive compensation zone.

[0038] Step S130: Perform rolling harmonic prediction on the competition compensation identifier and compensation division map to generate a predicted harmonic sequence, extract the dominant number migration identifier from the predicted harmonic sequence, perform compensation response mapping on the dominant number migration identifier to obtain the filter response delay, evaluate the compensation priority based on the filter response delay, and generate a hierarchical compensation activation parameter group.

[0039] Specifically, rolling harmonic prediction is performed on the competition compensation markers and compensation division maps to generate predicted harmonic sequences. The rolling window length is 20 fundamental cycles. If the window is too short, the harmonic evolution information captured by the model is insufficient to support the extrapolation of amplitude trends across cycles. If the window is too long, structural changes caused by load switching or compensation strategy adjustments are diluted by historical data, leading to prediction lag. A trade-off is made to ensure a balance between prediction timeliness and stability. The harmonic numbers corresponding to the high competition risk intervals in the competition compensation markers are given a weight of 1.5 times in the prediction model. This weight amplifies the contribution of historical fluctuation characteristics of the competition frequency bands to the prediction output, making the amplitude trend of the predicted harmonic sequence more accurate in terms of competition numbers. For example, if a certain number is marked as high competition risk and its amplitude has been fluctuating recently, the weighted prediction result can more sensitively reflect the evolution trend of that number. When the competition compensation identifier is an empty set, the weights of each harmonic are restored to uniform values. The upper limit of the compensation amount marked in the active layer of the compensation division map serves as the upper bound constraint of the predicted amplitude. The tuning frequency range of the passive layer in the compensation division map serves as a reference for the effective frequency band of the passive side prediction. The prediction confidence of frequencies exceeding the tuning range of the compensation division map is reduced. The prediction step size is set to a single fundamental period. In each step, the latest measured value is included in the rolling window and the oldest historical value is removed. For example, if the amplitude of a certain harmonic shows a monotonically increasing trend over several consecutive steps, the continuous increase in the amplitude time series of that frequency in the predicted harmonic sequence indicates that the compensation demand for that frequency band is expanding. The predicted harmonic sequences are grouped and arranged by harmonic number, with each number corresponding to a predicted amplitude time series, for use in subsequent time step amplitude ranking extraction and comparison of the average amplitude before and after the migration anchor point.

[0040] In some embodiments, extracting the dominant number migration identifier from the predicted harmonic sequence includes: extracting the amplitude order from the predicted harmonic sequence to generate an amplitude sorting group; comparing the dominant numbers of adjacent time steps in the amplitude sorting group to generate a number variation sequence; identifying the variation frequency density of the number variation sequence to generate a mutation-type migration marker set; and preferentially identifying the dominant number migration position based on the mutation-type migration marker set to generate a dominant number migration identifier.

[0041] The amplitude ranking group is generated by extracting the dominant and subordinate harmonics from the predicted harmonic sequence. When the predicted amplitudes of different harmonic orders within the same time step are similar or even equal, simply ranking them by amplitude size will cause random oscillations in the ranking between adjacent time steps. Therefore, the harmonic order number is introduced as a secondary sorting key. When amplitudes are equal, they are sorted from low to high harmonic order, ensuring that the ranking group for each time step is unique and reproducible. The predicted amplitudes of each harmonic in the predicted harmonic sequence are arranged from largest to smallest at each time step, with the order at the top being the dominant harmonic order for that time step. Harmonics whose amplitudes are all less than 0.5% of the fundamental amplitude at the same time step are not included in the ranking. Forcibly including low-amplitude harmonics would cause noise to occupy the high positions of the amplitude ranking group, obscuring the true dominant harmonics. If only the 3rd, 5th, and 7th harmonics exceed the threshold at a certain time step, the amplitude ranking group will only rank these three harmonics. The length of the amplitude sorting group extracted at each time step of the predicted harmonic sequence dynamically changes with the number of times the threshold is exceeded. A continuously shortening length indicates that the overall harmonic level is decreasing, while a stable length indicates that the composition of significant harmonic frequencies is relatively fixed. The amplitude sorting groups at each time step are summarized in chronological order to form a sorted time series. The sorted time series completely records the evolution of the dominant and subordinate relationships of the harmonic amplitudes of the predicted harmonic sequence at all time steps. When the first frequency of five consecutive time steps in the amplitude sorting group is the same, it is marked as a stable dominant state. The abrupt switching characteristics of frequencies occurring in the stable dominant state are more significant.

[0042] A frequency change sequence is generated by comparing the dominant frequencies of adjacent time steps within the amplitude ranking group. Events with an absolute difference of no more than two values ​​between adjacent dominant frequencies are classified as weak changes, while those with a difference of more than two values ​​are classified as strong changes. Weak changes are attributed to short-term ranking swaps of adjacent frequencies with similar amplitudes, while strong changes reflect a substantial migration of the dominant harmonic frequencies. The difference between the first and second frequencies of adjacent time steps in the amplitude ranking group is calculated, and four pieces of information are recorded at each position: the time step of change, the frequency before the change, the frequency after the change, and the event type, thus fully reconstructing the migration of the dominant harmonic at that time step. The frequency change sequence is composed of the frequency change events of each time step arranged in ascending chronological order. A large absolute difference indicates that the dominant frequency has crossed a wide range of harmonic frequencies in a short period. A jump of 8 values ​​between the dominant frequency and the 5th frequency indicates a large-scale structural shift in the power grid harmonic distribution. When adjacent time steps in the amplitude ranking group are in a stable dominant state, the difference between the corresponding positions in the frequency change sequence is always zero. If non-zero changes occur on both sides of the zero-value dense area, it forms a local abrupt change characteristic, which is a high-incidence location of abrupt migration. When the frequency variation sequence is a sequence of all zeros, it indicates that the dominant frequency of the amplitude sorting group is completely fixed throughout the entire analysis window. The mutation migration marker set will be an empty set, the compensation response mapping for the dominant frequency migration will not be executed, and each branch will be processed according to the default response parameters of the current dominant frequency.

[0043] A set of mutation-type migration markers is generated to identify the frequency density of a sequence of frequency changes. The density statistical window spans five adjacent time steps. If the window is too short, even a single, occasional change can cause the density to exceed the threshold, leading to frequent false positives. If the window is too long, multiple migrations occurring within a short period are diluted below the threshold and missed. The five-step window achieves an effective balance for engineering applications. Within each sliding window of the sequence of frequency changes, a weighted change density Z_w is calculated: Z_w = Σ(p_k × c_k) / L_w, where c_k is an indicator of whether a non-zero change occurs at the k-th time step (0 or 1), p_k is the weight coefficient of the corresponding change event (0.5 for absolute differences not exceeding 2, 1.0 for differences exceeding 2), and L_w is the window length. A Z_w value of 0.4 or higher is considered a high-frequency change segment. The migration intensity is measured by the mean of the absolute values ​​of the change differences within the window. Events with absolute differences of no more than two are weighted at 0.5 in density statistics to prevent frequent small-amplitude fluctuations from artificially inflating window density. If three out of five changes within a window have differences of only one, the effective density count is reduced after weighting, avoiding misjudging amplitude fluctuations as structural migration. The mutation-type migration marker set consists of the occurrence times and migration intensities of all high-frequency change segments. When the frequency change sequence is all zero, the mutation-type migration marker set is empty. Events with migration intensities exceeding four in the mutation-type migration marker set are marked with high-intensity migration labels. High-intensity migration labels indicate a large span of dominant harmonic orders, requiring the highest timeliness of compensation response. If the frequency change sequence continues to produce non-zero changes at the end of the current analysis window, it indicates that the migration process has not yet ended, and the mutation-type migration marker set must be updated continuously to reassess the status of high-intensity labels.

[0044] Dominant number migration identifiers are generated based on the priority labeling of the mutation-type migration marker set to determine the dominant number migration position. High-intensity migration-labeled mutation events are prioritized for migration position labeling; when intensities are equal, later events are prioritized. These two rules ensure that the most recently occurring strong migration events receive the highest processing priority. This proximity-first design reflects the physical fact that the more recent the harmonic migration, the more directly it affects the current compensation strategy. The occurrence time of each mutation event in the mutation-type migration marker set is marked as a migration anchor point on the time axis of the predicted harmonic sequence. The amplitudes of the dominant number in the predicted harmonic sequence after migration are compared for three time steps before and after the anchor point. If the average amplitude of the number after the anchor point is higher than before, it is marked as an enhanced migration, indicating an expansion of the harmonic injection compensation demand for that number; otherwise, it is marked as a decaying migration. If the average amplitude of the dominant number after a mutation event is higher for three consecutive steps than the average amplitude for the three steps before the event, it is determined to be an enhanced migration. The dominant number migration identifier is composed of the occurrence time step of each migration event, the dominant number before migration, the dominant number after migration, the migration intensity, and the migration type, comprehensively describing the temporal and directional characteristics of a dominant number migration. When the mutation-type migration marker set is empty, the dominant frequency migration marker is also empty. An empty set indicates that no mutation migration has occurred in the dominant frequencies within the current analysis window, and the compensation response mapping is not executed. When the number of enhancing migration events in the dominant frequency migration marker consistently exceeds that of decaying events, the overall trend of increasing harmonic intensity in the predicted harmonic sequence suggests that the compensation capability lags behind the harmonic growth. The migration type of high-intensity labeled events in the mutation-type migration marker set determines the adjustment direction and magnitude of the response mapping. When decaying events dominate in the dominant frequency migration marker, it indicates a trend of improvement in harmonic levels.

[0045] The filter response delay is obtained by performing compensation response mapping based on the dominant number migration identifier. The delay of the active branch originates from the superposition of the control algorithm detection delay and the inverter switching dead zone, while the delay of the passive branch is composed of the mechanical action time of the switching circuit breaker and the LC resonance establishment time. The difference between these two physical sources results in the filter response delay of the active side typically ranging from 1ms to 5ms, and the filter response delay of the passive side typically ranging from 10ms to 50ms. This order of magnitude difference is the fundamental starting point for active and passive collaborative timing design. The dominant number after migration for each migration event in the dominant number migration identifier is obtained by querying the corresponding measured response delay from the historical switching records. The measured response delay is obtained by statistically analyzing the difference between the time the compensation command is issued and the time when the compensation current reaches 90% of the rated amplitude. The average of the most recent 10 switching records is taken to eliminate the influence of occasional interference. When there are no historical switching records on the active side for a certain number of cycles, the response delay is estimated by linear interpolation of the response delay of adjacent cycles. An uncertainty label is added to the estimated value. In the dominant cycle migration identifier, the estimated delay for cycles corresponding to enhanced migrations is multiplied by a safety factor of 1.2, and the estimated delay for cycles corresponding to attenuated migrations is multiplied by a reduction factor of 0.9, which is used as the final filter response delay value. When the dominant cycle migration identifier is an empty set, the filter response delay for each branch is taken as the default response delay of the current dominant cycle. The default delay is determined based on the average of measured records over the past 20 cycles. For a certain branch, the average switching record over the past 20 cycles is 8ms, so the default filter response delay is taken as 8ms. The filter response delay is arranged by branch number and corresponding harmonic order. Branches with consistently high delay values ​​indicate mechanical aging or control delay accumulation issues. These branches are labeled with abnormal delays, which are treated with low confidence weights in subsequent layering and collaborative timing difference calculations.

[0046] In some embodiments, generating a hierarchical compensation activation parameter group based on the filter response delay assessment compensation priority includes: generating an active delay group and a passive delay group based on the branch type of the filter response delay; comparing the delay difference between the active delay group and the passive delay group to generate a coordinated timing difference; identifying segments in the coordinated timing difference where the delay difference continuously exceeds a preset time step threshold to generate a persistent conflict marker; and reordering the parameters based on the persistent conflict marker and the filter response delay to generate the hierarchical compensation activation parameter group.

[0047] The filter response delay is stratified into active and passive delay groups based on branch type. The stratification is based on the generation mechanism of the compensation current, not the branch's installation location. Branches that actively inject compensation current using power electronic converters are classified as active, while branches that passively filter harmonics using LC resonance are classified as passive. Branches with the same installation location but different mechanisms may belong to two delay groups at the same node. The delay values ​​of each branch in the filter response delay are extracted according to the above classification rules. The set of delay values ​​for active branches constitutes the active delay group, and the set of delay values ​​for passive branches constitutes the passive delay group. The overall delay value of the active delay group is lower than that of the passive delay group. For a certain harmonic, the active side response delay is 3ms, while the passive side is 18ms, a difference of 6 times. The dispersion of delay values ​​in each branch of the active delay group reflects the uniformity of the active side's response across different tuning orders. High dispersion indicates additional control overhead in active compensation at certain tuning orders. Branch delay values ​​with added uncertainty annotations are treated with low confidence weights in the active delay group. In the passive delay group, the delay values ​​of each branch typically increase with the tuning order. The resonant settling time at different tuning orders lengthens with increasing order. Branches with added abnormal delay annotations in the filter response delay are also treated with low confidence weights in the passive delay group to avoid the high delay of abnormally aged branches from skewing the cooperative timing difference.

[0048] The active and passive delay groups are compared to generate a coordinated timing difference. The direction of the difference is defined as passive delay minus active delay. A positive value is the norm for most harmonic orders. A negative difference indicates an abnormally slow response on the active side at that order. This abnormal slowness is usually caused by the active converter carrying too much compensation tasks, leading to increased delay in the control algorithm or inverter overheating and speed limiting. This signal indicates that the active side's operating point has exceeded the normal compensation capability boundary. The delay values ​​of the active and passive delay groups at the same harmonic order are subtracted, and the differences are arranged in chronological order to form the timing difference for that order. The timing differences for all orders together constitute the coordinated timing difference. The timing differences generated by the low-confidence weighted orders in the passive delay group are marked with uncertainty in the coordinated timing difference. The orders corresponding to the branches with low-confidence weights in the active delay group also inherit the uncertainty mark in the coordinated timing difference. The number of orders marked with uncertainty is increased by 10% in the conflict identification stage to avoid false alarms. When a certain frequency in the passive delay group has no corresponding branch, that frequency is not recorded in the coordinated timing difference. The same applies when a certain frequency in the active delay group has no corresponding branch. Frequency with a branch on only one side is not included in the difference calculation. Frequency with a consistently large mean difference in the coordinated timing difference indicates a significant difference in response speed between the active and passive sides for that frequency. For example, the mean coordinated timing difference for a certain frequency is as high as 25ms, indicating that the passive branch lags far behind the active side in the harmonic response for that frequency. Sufficient advance activation time must be reserved for the passive branch during coordinated switching.

[0049] For segments where the delay difference in the cooperative timing difference identification continuously exceeds a preset time step threshold, a persistent conflict marker is generated. The persistent time step threshold is set to 3 steps. When the threshold is below 3 steps, occasional sudden increases in the difference trigger the marker, resulting in a high false alarm rate. When the threshold is above 3 steps, a true persistent timing conflict requires more time steps to be accumulated before it is identified, causing a response lag. 3 steps is the minimum effective length verified in engineering practice between false alarm rate and response timeliness. The timing difference of each harmonic in the cooperative timing difference is scanned by time step. When the absolute value of the difference exceeds the amplitude threshold for 3 consecutive steps or more, it is determined to be a persistent conflict segment. The amplitude threshold is set to 1.5 times the current control cycle of the active filter. Exceeding 1.5 times the control cycle indicates that the passive branch lag can no longer wait for alignment on the active side, and a staggered switching strategy must be adopted. The persistent conflict marker consists of three pieces of information: the starting time step of the persistent conflict segment, the number of duration steps, and the average absolute value of the difference within the segment. Persistent conflict markers triggered by negative differences in the coordinated timing differences are assigned a reverse conflict attribute. The reverse conflict attribute indicates that the active side is later than the passive side and has a higher processing priority than forward conflicts. Persistent conflict markers are empty when the coordinated timing differences are an empty set or when all differences have not continuously exceeded the threshold. If the difference for a certain harmonic exceeds the amplitude threshold for four consecutive steps, a persistent conflict marker record with four consecutive steps is formed. The more consecutive steps, the more severe the timing mismatch. Conflict records triggered by uncertainty markers in the persistent conflict marker are processed with low confidence weights. Low confidence weight records do not trigger mandatory adjustments to the misalignment switching strategy and are only used as auxiliary references.

[0050] The hierarchical compensation activation parameter group is generated by reordering based on persistent conflict markers and filter response delays. The core logic of the sorting is to prioritize the activation of branches with severe conflicts and short response delays, establishing compensation current for branches with fast responses first, and staggering the activation of branches with slow responses. The staggered timing amount is determined by the difference in filter response delays. This design ensures that the active and passive compensation currents are aligned in timing when they reach the compensation node. The active and passive delay values ​​for each conflict record in the persistent conflict markers are extracted from the filter response delays. Conflicts with larger delay differences are assigned larger activation timing intervals, which are equal to the absolute value of the delay difference multiplied by a safety margin coefficient of 1.2. The activation order of reverse conflict events in the persistent conflict markers is adjusted so that the passive side is activated before the active side, and the activation order of forward conflict events is activated so that the active side is activated before the passive side. The activation intervals for both types of conflicts are determined by multiplying the difference in filter response delays by a coefficient of 1.2. In the hierarchical compensation activation parameter group, each harmonic order corresponds to one activation parameter record. The activation priority is determined by the number of conflict duration steps for the corresponding order in the persistent conflict flag. The more duration steps, the higher the priority. For example, an order with 6 duration steps has a higher activation priority than an order with only 3 duration steps. When the persistent conflict flag is an empty set, the activation time offset of all records in the hierarchical compensation activation parameter group is zero. Zero offset indicates simultaneous activation of active and passive components. The hierarchical compensation activation parameter group record corresponding to the uncertainty flag branch in the filter response delay is marked with a low-confidence flag. The low-confidence record must be confirmed twice during the activation execution phase before the switching command can be issued.

[0051] Step S140: Perform coordination constraint analysis on the hierarchical compensation activation parameter group to determine the priority activation branch, detect the current zero-crossing switching window of the priority activation branch, determine the branch adjustment margin based on the current zero-crossing switching window, and generate compensation weight output composite filter coordination control command by combining the residual feedback characteristics of the predicted harmonic sequence and the branch adjustment margin.

[0052] Specifically, the priority activation branches are determined by coordinating constraint analysis of the hierarchical compensation activation parameter group. Coordination constraints are derived from two types of boundary conditions: the maximum allowable injection current of the compensation node and the superimposed current when adjacent branches are activated simultaneously does not exceed the rated capacity of the circuit breaker. Satisfying only one type of constraint is insufficient to guarantee activation safety; only activation combinations that satisfy both types of constraints constitute an effective coordination scheme. The priority of each activation record in the hierarchical compensation activation parameter group is determined by the number of continuous conflict steps. Records with higher priority enter constraint verification first. Records that pass verification are included in the priority activation branches, while branches that fail verification are postponed to the next switching window to rejoin the queue. Constraint verification adopts a line-by-line accumulation method, superimposing the expected injection current of each branch from highest to lowest priority. Accumulation stops when the accumulated value reaches 90% of the upper limit of the allowable injection current of the compensation node. The set of accumulated branches constitutes the priority activation branches for the current switching window. Branches that are later in the queue after accumulation stops rejoin constraint verification in the next switching window and are not permanently excluded due to not being selected in the current window. In the hierarchical compensation activation parameter group, the expected injected current of the corresponding branch marked with low confidence is multiplied by a safety factor of 1.3 during constraint verification to prevent low-confidence branches from triggering overload protection due to excessively high actual current exceeding capacity constraints. The ratio of active to passive branches in the priority activation branch set must be no less than 1:3. If the ratio is unbalanced, low-confidence passive branches are removed from low to high priority until the ratio is restored. Removed branches are postponed to participate in verification again in the next switching window to ensure that the active side maintains sufficient fast response capability. If the proportion of passive branches still exceeds 80% after removal, it is necessary to check whether the active side is abnormally offline. The hierarchical compensation activation parameter group participates in constraint verification independently in each switching window. The size of the priority activation branch set is adjusted in real time according to the dynamic changes of grid harmonic load. The verification results of each round are independent and do not inherit the branch selection status of the previous round.

[0053] In some embodiments, detecting the current zero-crossing switching window of the priority activation branch includes: acquiring the compensation current waveform of the priority activation branch to generate a current waveform sequence; performing zero-crossing phase scanning on the current waveform sequence to generate a phase alignment time sequence; performing zero-crossing overlap detection on the phase alignment time sequence to generate a phase zero-crossing overlap mark; and calibrating the effective switching range based on the phase zero-crossing overlap mark to generate a current zero-crossing switching window.

[0054] The compensation current waveforms of the priority activation branches are collected to generate a current waveform sequence. The sampling resolution must ensure that at least 128 sample points are collected within a single fundamental cycle. If the resolution is lower than this, the zero-crossing position positioning error exceeds 0.8% of the fundamental cycle, causing the uncertainty of the zero-crossing time in phase scanning to exceed the switching control accuracy requirements. The compensation current waveforms of each branch in the priority activation branches are output in real time by current sensors installed at the branch outlets. The sensor bandwidth must cover the 50th harmonic frequency to ensure that higher harmonic components do not affect the accuracy of the waveform's zero-crossing position due to sensor cutoff distortion. The current waveform sequence is arranged in stripes according to the branch number. Each record corresponds to the sampling sequence of one branch within 10 consecutive fundamental cycles. The 10 fundamental cycles ensure that the number of zero-crossings in each record is not less than 20. 20 zero-crossings is the minimum effective sample size for phase statistical analysis. The compensation current waveform of passive branches in the priority activation branch contains multiple harmonic superposition components, and the zero-crossing density of the waveform is higher than that of the pure fundamental wave. When a passive branch is tuned to the 7th harmonic, the 7th harmonic component dominates in its compensation current, and the number of zero-crossings within the fundamental cycle is approximately 14, rather than the 2 of the pure fundamental wave. The difference in zero-crossing density must be distinguished in the phase scan. Branches with peak amplitudes lower than 5% of the rated compensation current in the current waveform sequence are marked as low-amplitude branches. Low-amplitude branches have a higher proportion of waveform noise, and the zero-crossing identification error is relatively large. The low-amplitude label is used to relax the overlap judgment threshold of the branch in the overlap detection stage. If a branch in the current waveform sequence has no effective samples for three consecutive fundamental cycles due to signal interruption within the acquisition window, the corresponding entry is marked as acquisition abnormal. Branches with acquisition abnormalities do not participate in the phase scan, and their switching time is postponed until normal acquisition is restored and re-determined.

[0055] A phase-aligned time sequence is generated by performing zero-crossing phase scanning on the current waveform sequence. Linear interpolation is used to accurately locate the zero-crossing time near the zero value by extrapolating the slope of adjacent sample points. Compared to directly taking the nearest sample point, interpolation compresses the zero-crossing location error from the order of magnitude of the sampling interval to 1 / 10 of the sampling interval, resulting in a positioning accuracy better than 0.06 ms at a sampling resolution of 128 points per cycle. The sampling sequence of each branch in the current waveform sequence is scanned sample by sample. Zero-crossing detection is triggered when the product of adjacent sample points is negative. Within this negative value range, the precise zero-crossing time is estimated using linear interpolation, and the estimation result is recorded as a zero-crossing time sample. The zero-crossing time samples of the low-amplitude labeled branches in the current waveform sequence inherit the low-amplitude label. Low-amplitude samples are retained in the phase-aligned time sequence, but the judgment threshold for participating in overlap detection is relaxed to 1.5 times the standard value. The difference between two adjacent zero-crossing moments reflects the instantaneous half-cycle of the compensation current of that branch. If the half-cycle deviates from the fundamental half-cycle by more than 10%, it indicates that the branch has inter-harmonic modulation leading to uneven zero-crossing point distribution. Unevenly distributed zero-crossing points must be converted to the equivalent fundamental phase before participating in multi-branch phase comparison. The conversion method involves normalizing the measured zero-crossing time interval to the fundamental half-cycle and then re-interpolating evenly. The phase alignment time sequence is arranged in bars according to branch number. Each record contains all zero-crossing moments of that branch within the acquisition window. One active branch recorded 21 zero-crossing moments, while another passive branch recorded 143. The density difference reflects the fundamental difference in the waveform composition of the compensation current of the two types of branches. Branches with abnormal acquisition in the current waveform sequence are not included in the phase scan, and there are no corresponding entries in the phase alignment time sequence.

[0056] Zero-crossing overlap detection is performed on the phase alignment time sequence to generate phase zero-crossing overlap markers. An overlap is determined when the absolute value of the zero-crossing time difference between two branches is less than 1% of the fundamental period. At power frequency, this threshold corresponds to approximately 0.2 ms. Within this time window, the mechanical feasibility of two branches simultaneously completing the switching action is low, and it is considered a substantial simultaneous switching conflict requiring misalignment processing. The zero-crossing times of each branch in the phase alignment time sequence are paired and their time differences are calculated. Pairs with an absolute time difference value lower than the overlap determination threshold are recorded as overlap events. The overlap event records the branch numbers involved and the corresponding zero-crossing time that caused the conflict. The overlap determination threshold for low-amplitude marker entries is extended to 2% of the fundamental period. After this extension, low-amplitude branches are more likely to trigger overlap markers, causing them to be assigned to non-overlapping times during candidate time selection to avoid switching conflicts. Before pairing, the time differences of branches with unevenly distributed zero-crossing points in the phase alignment time sequence are converted according to the equivalent fundamental phase. This conversion eliminates the interference of uneven zero-crossing point density caused by harmonic modulation on the overlap determination. For example, the zero-crossing time interval of a certain passive branch was only 0.1 ms before conversion, but after conversion, the equivalent interval was restored to about half a period of the fundamental wave, and the pairing result can better reflect the true phase proximity. The phase zero-crossing point overlap mark is composed of the sum of all overlap events. A zero number of overlap events indicates that the zero-crossing times of all branches in the phase alignment time sequence are staggered. When the phase zero-crossing point overlap mark is an empty set, the most recent zero-crossing time of each branch is directly used as the effective switching time. Branches with abnormal acquisition in the phase alignment time sequence have no corresponding entries, do not participate in pairing, and have no associated records in the phase zero-crossing point overlap mark.

[0057] For example, the step of generating a current zero-crossing switching window based on the phase zero-crossing overlap marker to calibrate the effective switching range includes: extracting the effective switching candidate times of each active branch based on the phase zero-crossing overlap marker to generate a candidate time group; performing multi-branch time overlap detection on the candidate time group to generate a simultaneous switching conflict marker; identifying conflicting branch pairs on the simultaneous switching conflict marker to generate a timing misalignment priority arrangement; and performing misaligned timing allocation on the candidate time group according to the timing misalignment priority arrangement to generate a current zero-crossing switching window.

[0058] Based on the phase zero-crossing overlap marker, effective switching candidate times for each active branch are extracted to generate candidate time groups. Three candidate times are selected for each branch to achieve an effective balance between the flexibility of misalignment allocation and the computational burden of conflict detection. Too few candidates result in insufficient misalignment scheduling options, forcing some branches to be postponed to the next cycle; too many candidates lead to a quadratic increase in the number of pairing detections, resulting in unnecessary computational overhead. Zero-crossing times of each branch that are not marked as conflicting in the phase zero-crossing overlap marker are prioritized for inclusion in the candidate time group, arranged in ascending order of their distance from the current time. If there are fewer than three candidates, the zero-crossing times corresponding to adjacent fundamental cycles are used to supplement them to three. Zero-crossing times marked as conflicting by the phase zero-crossing overlap marker are not included in the candidates. If the three most recent zero-crossing times of a branch are all marked as conflicting, then the three non-conflicting zero-crossing times are selected from the 4th, 5th, and 6th zero-crossing times. In the candidate time group, the three candidate times for each branch are arranged in ascending order of time value, with the earliest time being the first choice and the latter two times as alternatives. Branches with abnormal acquisition in the phase alignment time sequence have no candidate entries in the corresponding candidate time group and cannot participate in the switching allocation within the current fundamental cycle. If a low-amplitude labeled branch has fewer than three candidate times, it will participate in collision detection with the existing number of candidates as the upper limit. The candidate time group is summarized by branch number, and all active branches have corresponding entries in the candidate time group. The switching of branches with empty candidate entries is postponed to the next fundamental cycle.

[0059] The candidate time group undergoes multi-branch time overlap detection to generate simultaneous switching conflict markers. When the number of active branches exceeds 10, the number of detection pairs generated by pairwise pairing exceeds 45. A scan-line algorithm based on time sorting is adopted to reduce the detection time complexity from quadratic to linear. The scan-line algorithm mixes and sorts all candidate times and then sequentially scans adjacent times to see if they fall into the overlap judgment threshold, effectively avoiding redundant pairing of long-distance time pairs. All candidate times of each branch in the candidate time group are mixed into the scan. The overlap judgment threshold continues to use the standard of 1% of the fundamental period. The overlap judgment threshold of candidate times for low-amplitude labeled branches is extended to 2% of the fundamental period. The extension processing is consistent with the threshold setting in the zero-crossing overlap detection stage. The same processing rules are used in the two stages to avoid low-amplitude branches being mistakenly relaxed in one place and mistakenly tightened in another. Simultaneous switching conflict markers are composed of all conflict pairs detected during the scanning process. Each conflict record contains two branch numbers and the candidate time value that caused the conflict. The more times the same branch appears in the simultaneous switching conflict markers, the more clustered the candidate time of that branch is with the overall candidate time group. If all three candidate times of a branch conflict with different branches, and the occurrence count is 3, it indicates that the branch should be prioritized for misalignment processing. When the simultaneous switching conflict markers are an empty set, the preferred times of all branches in the candidate time group do not conflict. The current zero-crossing switching time window is directly generated based on the preferred times of each branch without entering the misalignment allocation process. Branches with empty candidate entries in the candidate time group correspond to no associated records in the simultaneous switching conflict markers.

[0060] For simultaneously switched conflict markers, conflicting branch pairs are identified and prioritized based on timing misalignment. Branches with higher priority retain their preferred time, while lower priority branches relinquish their preferred time and move to the alternative time slot. Priority is inherited from the activation priority attributes carried by each branch in the priority active branch. This design ensures consistency between the conflict concession order and the urgency of compensation, rather than simply mechanically sorting by branch number. Conflicting branch pairs in simultaneously switched conflict markers are sorted in descending order of the difference in activation priorities, with branches having larger differences receiving priority. Branch pairs with the same priority are sorted in ascending order by branch number as a secondary sorting criterion. Low-amplitude marker branches automatically drop to the last position when priorities are the same. Since low-amplitude branches have higher timing uncertainty, prioritizing their time relinquishes reduces the propagation of switching errors to other branches. For each additional occurrence of a branch in the simultaneous switching conflict markers, its concession responsibility weight in the temporal misalignment priority ranking increases by 0.1. A branch appearing three times has a concession weight of 0.3, higher than the 0.1 weight of a branch appearing only once. Branches with higher concession weights bear more responsibility for time adjustment in misalignment allocation. The temporal misalignment priority ranking is composed of the results of sorting all conflicting branch pairs according to the above rules. The more conflict records in the simultaneous switching conflict markers, the longer the temporal misalignment priority ranking, and branch pairs ranked higher receive misalignment allocation resources first. When the simultaneous switching conflict markers are an empty set, the temporal misalignment priority ranking is also an empty set; the misalignment temporal allocation step is not executed, and the preferred time of each branch in the candidate time group is directly used as the switching time.

[0061] Based on the timing misalignment priority ranking, candidate time groups are allocated using a staggered timing sequence to generate current zero-crossing switching windows. The allocation process is carried out pairwise according to the timing misalignment priority ranking. After each pair is processed, the occupancy status of the candidate time group is updated immediately. Time slots already occupied by higher-priority branches are considered unavailable in subsequent pairings, ensuring global consistency of the allocation results rather than just local optima. When processing each pair of branches in the timing misalignment priority ranking, the higher-priority branch retains its current preferred time slot, while the lower-priority branch searches sequentially for the first unoccupied alternative time slot in the candidate time group. Once found, the alternative time slot is marked as occupied and recorded as the staggered allocation time slot for that branch. Branches that have no available time slots after searching through all alternative time slots are marked as having no available time slots. If all three candidate time slots of a low-priority branch are occupied by higher-priority branches, the branch cannot be switched in the current fundamental cycle and is postponed to the next fundamental cycle to regenerate a candidate time group for allocation. After each branch completes the staggered allocation, the current zero-crossing switching time window is generated with the allocation result time as the center and the width determined according to the consistency error of the corresponding circuit breaker. The current zero-crossing switching time window covers the time range within which the switching command can be issued for the branch. For branches with no available time, the current zero-crossing switching time window is marked with a delayed switching label. When the timing stagger priority is arranged as an empty set, the preferred time of each branch in the candidate time group is directly used as the center time of the current zero-crossing switching time window.

[0062] The adjustable margin of a branch is determined based on the current zero-crossing switching time window. The adjustable margin reflects the space within which each branch can further expand its compensation under the current zero-crossing switching time window constraint. A large margin indicates that the branch's compensation capacity still has reserves that can be utilized, while a small margin indicates that further increases near the operating boundary pose an overload risk. A wider current zero-crossing switching time window allows the circuit breaker to complete switching within a larger time deviation range, indirectly allowing for greater amplitude adjustment flexibility in the compensation current waveform. The adjustable margin of the active branch is also constrained by the DC bus voltage of the inverter. When the bus voltage is low, the modulation depth of the inverter is close to saturation. Even if the window is wide when the current is switched at zero crossing, the injected current amplitude cannot be increased. The margin of this type of branch is reduced according to the degree of low bus voltage. The reduction factor γ_s is determined by the formula γ_s=(U_s-U_low) / (U_ref-U_low), where U_s is the current DC bus voltage, U_low is the minimum allowable bus voltage of the inverter, and U_ref is the rated bus voltage. The value of γ_s is cut off between 0 and 1. When it is lower than 0.2, the adjustable margin of the branch is directly assigned to zero. The adjustable margin of a passive branch is determined by its switching status. When in operation, the margin of a passive branch is the difference between its rated reactive capacity and the reactive output at the current operating point. When out of operation, the margin is its full rated reactive capacity. For example, a passive branch in operation with a rated reactive capacity of 100 kvar and a current output of 80 kvar has an adjustable margin of 20 kvar. For branches with a delayed switching mark added to the current zero-crossing switching window, the adjustable margin is set to zero in the current cycle. Delayed switching branches cannot participate in compensation scheduling in this cycle. When all branch margins are zero, the composite filter coordination control command remains unchanged from the previous cycle.

[0063] Based on the residual feedback characteristics and the adjustable margin of the branches, compensation weights are generated to output composite filter coordinated control commands. The residual feedback characteristics are extracted from the difference between the predicted harmonic sequence and the measured amplitude at the current time step of the multi-source operating data. Harmonic orders with large absolute differences indicate that the compensation demand deviates from the predicted direction. Directly using the predicted weights will lead to a systematic mismatch in the compensation amount for that order, necessitating residual correction. A positive residual indicates that the actual harmonics are higher than the predicted value, and the compensation demand is underestimated; a negative residual indicates that the actual harmonics are lower than the predicted value, and the compensation demand is overestimated. For example, if the residual for a certain order is consistently positive, it suggests that the harmonic injection source in that frequency band is increasing while the prediction model has not yet caught up, and the corresponding weight must be increased to compensate for the underestimated compensation demand. The corrected compensation weight G_i for each order i is determined by the formula G_i = G_{f,i} + μ × ε_i, where G_{f,i} is the baseline value of the prediction weight for the corresponding order of the predicted harmonic sequence, ε_i is the ratio of the residual to the fundamental amplitude for that order, and μ is the correction coefficient, set to 0.5. A value of 0.5 for μ is used to avoid drastic weight jumps that could cause compensation current oscillations when the single-step residual is too large. The upper bound H_i is determined by the formula H_i = R_i / ΣR, where R_i is the adjustable margin of branch i, and ΣR is the sum of the margins of all branches. When G_i exceeds H_i, it is truncated to H_i to ensure that the command does not exceed the physical carrying capacity of the branch. If the absolute value of the predicted harmonic sequence residual is consistently lower than 5% of the predicted amplitude for all orders, residual correction is not triggered, and the weight is directly determined by the proportion of the predicted harmonic sequence amplitude. The compensation weights after residual correction and margin constraints for each iteration are summarized into a full-evolution compensation weight vector, which is mapped to the compensation current command amplitude and switching status command respectively according to active and passive modes. The two types of commands are combined to form a composite filter coordination control command. The timing of the issuance is based on the staggered arrangement of the current zero-crossing switching window to ensure that each branch receives the command within the effective time window. The composite filter coordination control command for branches with zero adjustable margin maintains the command value of the previous cycle unchanged until the margin is restored.

[0064] To implement the above-described method embodiments, a combined active and passive power grid harmonic suppression method is proposed to achieve the corresponding functions and technical effects. See also... Figure 2 , Figure 2 This diagram illustrates a structural block diagram of an active-passive coordinated power grid harmonic suppression composite filter 200 provided in an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The active-passive coordinated power grid harmonic suppression composite filter 200 provided in this embodiment includes: Data acquisition module 201 is used to acquire multi-source operation data of the power grid and construct a harmonic coupling characteristic spectrum based on the multi-source operation data to analyze harmonic coupling characteristics. The compensation division module 202 is used to determine the passive compensation interval based on the harmonic coupling characteristic spectrum, extract the detuning polarity characteristics of the passive branch from the multi-source operation data to generate detuning polarity correction coefficients, use the detuning polarity correction coefficients to perform dynamic boundary correction on the passive compensation interval to form a compensation division map, and identify the compensation competition interval based on the compensation division map to generate a competition compensation identifier. Prediction scheduling module 203 is used to perform rolling harmonic prediction on the competition compensation identifier and the compensation division map to generate a prediction harmonic sequence, extract the dominant number migration identifier from the prediction harmonic sequence, perform compensation response mapping on the dominant number migration identifier to obtain the filter response delay, and evaluate the compensation priority based on the filter response delay to generate a hierarchical compensation activation parameter group. The coordination output module 204 is used to perform coordination constraint analysis on the hierarchical compensation activation parameter group to determine the priority activation branch, detect the current zero-crossing switching window of the priority activation branch, determine the branch adjustment margin based on the current zero-crossing switching window, and generate a compensation weight output composite filter coordination control command by combining the residual feedback characteristics of the predicted harmonic sequence and the branch adjustment margin.

[0065] The aforementioned active-passive coordinated power grid harmonic suppression composite filter 200 can implement an active-passive coordinated power grid harmonic suppression method according to the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining contents of this application embodiment can be referred to the contents of the above method embodiments, and will not be repeated in this embodiment.

[0066] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0067] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A method for active and passive coordinated power grid harmonic suppression, characterized in that, include: Collect multi-source power grid operation data, and construct a harmonic coupling characteristic spectrum based on the analysis of harmonic coupling characteristics of the multi-source operation data; The passive compensation interval is determined based on the harmonic coupling characteristic spectrum. The detuning polarity characteristics of the passive branch are extracted from the multi-source operation data to generate detuning polarity correction coefficients. The detuning polarity correction coefficients are used to perform dynamic boundary correction on the passive compensation interval to form a compensation division map. Based on the compensation division map, the compensation competition interval is identified and a competition compensation identifier is generated. Rolling harmonic prediction is performed on the competitive compensation identifier and the compensation division map to generate a predicted harmonic sequence. The dominant number migration identifier is extracted from the predicted harmonic sequence. Compensation response mapping is performed on the dominant number migration identifier to obtain the filter response delay. The compensation priority is evaluated based on the filter response delay to generate a hierarchical compensation activation parameter group. The hierarchical compensation activation parameter group is analyzed for coordination constraints to determine the priority activation branch. The current zero-crossing switching window of the priority activation branch is detected. The branch adjustment margin is determined based on the current zero-crossing switching window. The compensation weight output composite filter coordination control command is generated by combining the residual feedback characteristics of the predicted harmonic sequence and the branch adjustment margin.

2. The method according to claim 1, characterized in that, The process of constructing a harmonic coupling characteristic spectrum based on the analysis of harmonic coupling characteristics using the multi-source operating data includes: Harmonic amplitude timing groups are generated by extracting the amplitude timing sequence of the multi-source operating data according to the harmonic order. The harmonic amplitude time series is subjected to cross-order synchronous fluctuation detection to generate the inter-harmonic correlation distribution. Harmonic clustering labels are generated by clustering the inter-harmonic correlation distribution according to threshold polarity. A joint feature mapping is performed on the homogeneous harmonic clustering labels and the harmonic amplitude time series groups to generate a harmonic coupling feature spectrum.

3. The method according to claim 1, characterized in that, The step of extracting the detuning polarity features of passive branches from the multi-source operating data to generate detuning polarity correction coefficients includes: The passive branch resonant frequency offset is extracted from the multi-source operating data to generate a frequency offset sequence; The frequency offset sequence is subjected to time-series difference analysis to generate a detuning rate distribution; Based on the distribution of the detuning rate of change, the offset direction is identified, and a capacitive offset marker set and a perceptual offset marker set are generated; Based on the capacitive offset marker set and the inductive offset marker set, a leading-back correction amount is constructed to generate a detuning polarity correction coefficient.

4. The method according to claim 1, characterized in that, The step of identifying the competition interval based on the compensation division map and generating a competition compensation identifier includes: The active compensation allocation interval and the passive compensation allocation interval are extracted from the compensation division map to generate a dual-path compensation interval group; The dual-path compensation interval group is subjected to interval overlap detection to generate an overlap distribution sequence; Based on the overlapping distribution sequence, the relationship between active and passive compensation directions within the overlapping interval is identified to generate an inverse competition tag set; The compensation competition interval is marked by the reverse competition tag set to generate a competition compensation identifier in the compensation division map.

5. The method according to claim 1, characterized in that, The step of extracting the dominant order migration identifier from the predicted harmonic sequence includes: The amplitudes of the predicted harmonic sequences are extracted and sorted by primary and secondary values ​​to generate amplitude sorting groups; The amplitude sorting group is compared with the dominant frequency of adjacent time steps to generate a frequency variation sequence; A set of mutation-type migration markers is generated by identifying the frequency density of the aforementioned sequence of changes. Based on the mutant migration marker set, the dominant migration position is preferentially identified to generate a dominant migration identifier.

6. The method according to claim 1, characterized in that, The step of generating a hierarchical compensation activation parameter group based on the filter response delay evaluation compensation priority includes: The filter response delay is hierarchically generated into active delay groups and passive delay groups according to branch type; The delay difference between the active delay group and the passive delay group is compared to generate a coordinated timing difference. For the collaborative timing difference identification segment where the delay difference continuously exceeds the preset time step threshold, a continuous conflict marker is generated; The hierarchical compensation activation parameter group is generated by reordering based on the persistent conflict markers and the filter response delay.

7. The method according to claim 1, characterized in that, The detection window for the zero-crossing switching of the current in the priority activated branch includes: The compensation current waveform of the preferentially activated branch is collected to generate a current waveform sequence; The current waveform sequence is subjected to zero-crossing phase scanning to generate a phase-aligned time sequence; The phase alignment time sequence is subjected to zero-crossing overlap detection to generate phase zero-crossing overlap markers; The effective switching range is calibrated based on the phase zero-crossing overlap mark, and a current zero-crossing switching window is generated.

8. The method according to claim 3, characterized in that, The process of identifying the offset direction based on the detuning rate of change distribution and generating a capacitive offset marker set and a perceptual offset marker set includes: The offset symbol distribution is extracted from the detuning rate of change distribution according to the segment amplitude to generate an offset symbol sequence; The offset symbol sequence is used to identify continuously flipped segments to generate a polarity reversal event marker set; The polarity reversal event marker set is filtered for reversal amplitude exceeding a threshold to generate a gradual switching trigger marker; Based on the gradual switching trigger flag, the offset symbol sequence is subjected to gradual direction calibration to generate a capacitive offset flag set and an inductive offset flag set.

9. The method according to claim 7, characterized in that, The step of generating the current zero-crossing switching window based on the phase zero-crossing overlap mark to calibrate the effective switching range includes: Based on the phase zero-crossing overlap marker, the effective switching candidate times of each active branch are extracted to generate a candidate time group; The candidate time group is subjected to multi-branch time overlap detection to generate simultaneous switching conflict markers. The simultaneous switching conflict markers are used to identify conflict branch pairs and generate a time-sequence misalignment priority sort. Based on the timing misalignment priority arrangement, the candidate time group is allocated with misaligned timing to generate the current zero-crossing switching window.

10. A composite filter for power grid harmonic suppression using a combination of active and passive power grid harmonic suppression, employing the active-passive collaborative power grid harmonic suppression method described in any one of claims 1-9, characterized in that... include: The data acquisition module is used to collect multi-source operation data of the power grid and construct a harmonic coupling characteristic spectrum based on the analysis of harmonic coupling characteristics of the multi-source operation data. The compensation division module is used to determine the passive compensation interval based on the harmonic coupling characteristic spectrum, extract the detuning polarity characteristics of the passive branch from the multi-source operation data to generate detuning polarity correction coefficients, use the detuning polarity correction coefficients to perform dynamic boundary correction on the passive compensation interval to form a compensation division map, and identify the compensation competition interval based on the compensation division map to generate a competition compensation identifier. The prediction scheduling module is used to perform rolling harmonic prediction on the competition compensation identifier and the compensation division map to generate a prediction harmonic sequence, extract the dominant number migration identifier from the prediction harmonic sequence, perform compensation response mapping on the dominant number migration identifier to obtain the filter response delay, and evaluate the compensation priority based on the filter response delay to generate a hierarchical compensation activation parameter group. The coordination output module is used to perform coordination constraint analysis on the hierarchical compensation activation parameter group to determine the priority activation branch, detect the current zero-crossing switching window of the priority activation branch, determine the branch adjustment margin based on the current zero-crossing switching window, and generate a compensation weight output composite filter coordination control command by combining the residual feedback characteristics of the predicted harmonic sequence and the branch adjustment margin.

Images