Automatic anchor point searching and processing method for grid-connected test data of photovoltaic inverter

Abstract

The invention discloses an automatic anchor point searching and processing method for grid-connected test data of a photovoltaic inverter, and belongs to the technical field of automatic test of a power system. According to the method, a three-phase voltage and current signal output by a power grid simulator and an inverter power instruction signal are aligned through a high-precision time synchronization device; wavelet transform multi-scale noise reduction and moving average filtering combined preprocessing is adopted to improve the signal-to-noise ratio; identifying a voltage zero crossing point candidate set, a drop starting point candidate set and a recovery termination point candidate set based on a self-adaptive dynamic threshold value; transient energy characteristic verification is introduced for a voltage drop starting point; effective anchor points are confirmed through time window association of power instruction step changes. According to the method, the problems of low efficiency of manual key event point identification, misjudgment caused by noise interference, grid event and inverter response time sequence correlation missing and the like are solved, and the automation degree of test data analysis, anchor point positioning precision and control response time sequence analysis reliability are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of power system automation testing technology, and in particular to a method for automatically finding and processing grid-connected test data of photovoltaic inverters. Background Technology

[0002] Grid-connected testing of photovoltaic inverters is a core step in verifying whether the equipment complies with grid specifications, requiring precise analysis of the inverter's dynamic response characteristics during grid faults. During the test, a grid simulator generates transient events such as voltage dips / recoveries, while simultaneously recording the three-phase voltage and current waveforms and power command signals issued by the inverter controller. The accuracy of the location of key event anchor points in this data (such as the onset of voltage dips, the recovery stabilization time, and zero-crossing points) directly affects the evaluation results of the inverter's low-voltage ride-through capability, response speed, and other performance characteristics.

[0003] Currently, the industry primarily relies on manual analysis of oscilloscope waveforms to determine anchor points: operators must visually identify abrupt changes in voltage waveforms, manually mark timestamps, and compare them with the power command change timing. This method has significant drawbacks: First, manual identification is inefficient, requiring several hours to process data for a single test, making it difficult to meet the needs of batch testing; second, in the presence of noise interference or complex transient processes (such as oscillation recovery), different personnel may have subjective differences in their judgments of the same event, leading to reduced repeatability of test conclusions; more importantly, there is a lack of quantitative means to establish the causal relationship between grid events and inverter control responses, for example, it is impossible to automatically verify whether a voltage drop truly triggered a power command jump.

[0004] Existing automated solutions attempt to identify voltage abrupt changes using fixed threshold methods, but face technical challenges. Background harmonics and random noise from the power grid can easily lead to misjudgments, especially in scenarios with shallow voltage dips. The gradual variation characteristics of power command signals (such as ramp regulation) make it difficult for traditional differential methods to accurately detect step changes, causing event correlation failures. Furthermore, oscillations during the voltage recovery phase often lead to misjudgments of termination points, while zero-crossing point identification is significantly affected by frequency fluctuations. These problems severely limit the accuracy of test data analysis and its engineering guidance value. Therefore, there is an urgent need to develop an anchor point location method with strong noise resistance and the ability to automatically correlate power grid events with control responses to improve testing efficiency and result reliability. Summary of the Invention

[0005] This invention overcomes the problems of low efficiency in manual identification of key event points, misjudgment caused by noise interference, and lack of correlation between grid events and inverter response timing in the prior art, and significantly improves the automation level of test data analysis, anchor point positioning accuracy and control response timing analysis reliability.

[0006] To achieve the above objectives, the present invention adopts the following solution: An automated method for finding and processing grid-connected test data of photovoltaic inverters includes the following steps: Acquire grid-connected test data of photovoltaic inverters, including three-phase voltage signals, three-phase current signals output by the grid simulator, and power command signals issued by the inverter controller. Then, align the three-phase voltage signals, three-phase current signals, and power command signals with timestamps using a high-precision time synchronization device. The three-phase voltage signal and the three-phase current signal are preprocessed. The preprocessing includes performing multi-scale denoising based on wavelet transform and moving average filtering in sequence, and outputting the denoised voltage signal and current signal. Based on the denoised voltage signal and the preset anchor point type, perform adaptive dynamic threshold anchor point identification. The anchor point type includes voltage zero crossing point, voltage drop start point and voltage recovery termination point, and output an initial anchor point candidate set. For the candidate points of voltage drop initiation point in the initial anchor point candidate point set, perform transient energy feature verification: calculate the transient energy value of the voltage signal within the power frequency cycle before and after the candidate point of voltage drop initiation point. The transient energy value is defined as the integral of the square of the voltage deviation, where the voltage deviation is the difference between the real-time voltage and the root mean square value of the voltage before the drop. When the calculated transient energy value is greater than the preset energy threshold, the candidate point is confirmed as an effective voltage drop initiation point. The timestamp of the power command signal is associated with the timestamp of the effective voltage drop start point. If the power command signal has a step change within a set time window before the effective voltage drop start point occurs, the candidate point is marked as the final effective anchor point; otherwise, the candidate point is marked as an invalid anchor point and removed from the anchor point candidate point set. Output the timestamps and signal characteristic values ​​of the final valid anchor points corresponding to all anchor point types in the anchor point candidate set.

[0007] Preferably, the method for determining the preset energy threshold corresponding to the transient energy value includes: Obtain voltage signal waveform data for consecutive power frequency cycles before the candidate point of voltage drop initiation, and extract this voltage signal waveform data as the reference voltage segment; The root mean square value of the reference voltage is calculated for the entire cycle of the reference voltage range to obtain the root mean square value of the reference voltage. A dynamic energy threshold calculation model is established based on the root mean square value of the reference voltage. The dynamic energy threshold calculation model satisfies the following relationship: the dynamic energy threshold is equal to the square of the root mean square value of the reference voltage multiplied by the scaling factor. The proportionality coefficient is obtained by calibration through historical test datasets. The calibration method is as follows: select the actual voltage drop start point that has been manually verified in the historical test data, calculate the actual transient energy value corresponding to each actual voltage drop start point and the square of the root mean square value of the corresponding reference voltage, fit the linear relationship between the actual transient energy value and the square of the root mean square value of the reference voltage using the least squares method, and the slope of the fitted line is the proportionality coefficient. The dynamic energy threshold calculated in real time is used as the preset energy threshold for the transient energy characteristic verification process.

[0008] As a preferred option, when the number of actual voltage drop initiation points in the historical test dataset does not reach the set number, the following method is used for scaling factor physical model compensation: Establish a theoretical model for the transient energy of voltage sag: E theory V is the theoretical energy value. base The reference voltage is the root mean square value, d is the real-time voltage drop depth, and k is the base value of the proportional coefficient. base Obtained through electromagnetic transient simulation data of typical power grid topologies; V dip The root mean square value of the voltage at the candidate point where the voltage drop begins; Calculate the theoretical energy value and use it as the preset energy threshold; confirm the new effective voltage drop initiation point through the voltage drop transient energy theoretical model; The actual transient energy value corresponding to the newly confirmed effective voltage drop initiation point and the square of the root mean square value of the reference voltage are added to the historical test dataset. When the number of effective voltage drop initiation points reaches the set number, the process switches to obtaining the scaling factor through calibration using the historical test dataset.

[0009] As a preferred method, the determination of a step change in the power command signal includes the following steps: Acquire the data sequence of the power command signal within a set time window, perform piecewise linear fitting on the data sequence, divide the data sequence into continuous sub-segments, and ensure that the sum of squared residuals of the linear fitting of data points in each sub-segment is less than a set residual threshold. Calculate the absolute value of the slope difference between adjacent segments. When the absolute value of the slope difference exceeds the set slope change threshold, mark the connection point of the adjacent segments as a potential step point. Extract power command data within a set time period before and after the potential step point, calculate the steady-state mean of the segment before and after the potential step point; calculate the absolute value of the difference between the two steady-state means, and when the absolute value of the difference is greater than the set percentage of the full-scale power command value and greater than the set minimum change amount, confirm the potential step point as a valid step change point; record the timestamp of the valid step change point and use it for the correlation verification process.

[0010] As a preferred approach, the residual threshold is dynamically adjusted during the piecewise linear fitting process. Specific methods include: Calculate the absolute value of the average rate of change of the power command signal within a set time window; When the absolute value of the average rate of change is less than the set rate threshold, the set residual threshold will be adjusted to the baseline residual threshold. When the absolute value of the average rate of change is greater than or equal to the set rate threshold, the set residual threshold is adjusted according to the following relationship: Adjust the set residual threshold to the ratio of the baseline residual threshold multiplied by the absolute value of the average rate of change to the set rate threshold; Piecewise linear fitting is performed using the adjusted residual threshold; the baseline residual threshold is determined by the fitting residual statistics of the steady-state power command segment in historical test data.

[0011] Preferably, the following method is used to verify the transient oscillation stability for the voltage recovery termination point candidate points in the initial anchor point candidate point set: Extract the noise-reduced voltage signal waveform for multiple consecutive power frequency cycles after the candidate point of voltage recovery termination. Detect the local maximum and local minimum sequence of the voltage signal waveform, calculate the amplitude difference between adjacent maximum and minimum points, and generate an amplitude difference sequence. Calculate the attenuation rate of the amplitude difference sequence. The attenuation rate is the ratio of the amplitude difference of the next set to the amplitude difference of the previous set. If three consecutive sets of attenuation rates are greater than the set attenuation threshold, the voltage signal is determined to be in a stable recovery state. Only when the candidate point of voltage recovery termination point simultaneously meets the adaptive dynamic threshold anchor point identification condition and the stable recovery state determination condition is the candidate point confirmed as an effective voltage recovery termination point.

[0012] Preferably, when there is no situation where three consecutive sets of attenuation rate determinations are all greater than the set attenuation threshold, a segmented stability weighted evaluation is performed, including the following steps: The amplitude difference sequence of continuous power frequency cycles after the candidate point of voltage recovery termination is divided into a first subsequence and a second subsequence. The average attenuation rate of the first subsequence is calculated. The attenuation rate of the front section δ i Let m be the amplitude difference between adjacent maxima and minima in the i-th group, and m be the length of the first subsequence; calculate the average decay rate of the second subsequence. The attenuation rate in the latter part δ j is the magnitude difference between adjacent maxima and minima in the j-th group, and p is the length of the subsequent subsequence; Calculate the overall stability coefficient using the weighted evaluation formula. , where w front For the front-end weights, wrear For the weight of the later stage; If the overall stability coefficient is greater than the set attenuation threshold, the voltage signal is determined to be in a stable recovery state, and this determination result is used as the final determination result for the three consecutive sets of attenuation rates.

[0013] Preferably, for the candidate points with zero voltage crossings in the initial anchor point candidate set, the following method is used for multi-cycle phase consistency verification: Extract the denoised voltage signal from multiple consecutive power frequency cycles before the candidate voltage zero-crossing point; for each power frequency cycle, detect the voltage zero-crossing point and generate a historical zero-crossing point timestamp sequence. Calculate the difference between adjacent timestamps in the historical zero-crossing timestamp sequence to obtain the actual periodic value sequence, and calculate the absolute deviation value sequence between the actual periodic value sequence and the standard power frequency cycle. If the maximum value of the absolute deviation value sequence does not exceed the set phase tolerance threshold, the current voltage zero-crossing candidate point is confirmed as an effective voltage zero-crossing point; otherwise, the median of the historical zero-crossing timestamp sequence is used to correct the timestamp of the current candidate point.

[0014] As a preferred method, the standard power frequency cycle is dynamically generated using the following method: Obtain the actual periodic value sequence of multiple consecutive power frequency cycles before the candidate point of voltage zero crossing, and perform a moving average processing on the obtained actual periodic value sequence, with the moving window width being half the number of cycles. The median of the sequence after moving average processing is used as the dynamic reference period. When calculating the absolute deviation value sequence, the dynamic reference period is used instead of the fixed standard power frequency period. If the maximum value of the absolute deviation value sequence is less than the set phase tolerance threshold, the median of the historical zero-crossing timestamp sequence is still used to correct the current candidate point timestamp. If the deviation between the dynamic reference period and the rated power frequency period exceeds the set frequency tolerance threshold, a frequency abnormality alarm signal will be sent to the test platform.

[0015] As a preferred method, the specific methods for outputting the timestamps and signal characteristic values ​​of the final valid anchor points corresponding to all anchor point types include: Generate a visualization chart containing timestamp-aligned denoised voltage signal waveforms, power command signal waveforms, and final effective anchor point location markers; Data labels are associated with the final valid anchor point location marker. The data labels include the anchor point type name, the precise timestamp, and the root mean square voltage value at the corresponding time. A manual position correction interface is set up. This interface responds to the user's drag operation on the final effective anchor point position mark in the visualization chart, obtains the new timestamp coordinates after dragging, and calculates the difference between the new timestamp coordinates and the original timestamp as the manual offset. When the absolute value of the manual offset does not exceed the predetermined tolerance window, the corrected anchor point data is generated based on the new timestamp coordinates and stored independently. The signal feature values ​​of the corrected anchor point data are updated through interpolation calculation.

[0016] The present invention includes at least the following beneficial effects: (1) By using a closed-loop process of time synchronization, signal denoising, adaptive anchor point identification, transient energy verification and power command association, manual visual inspection is replaced, reducing the single test analysis time to the minute level and greatly improving test efficiency; (2) By combining wavelet denoising, dynamic energy threshold, oscillation attenuation rate determination and multi-cycle phase verification, noise and harmonic interference are effectively suppressed, and the anchor point identification accuracy is maintained even in shallow drop and high distortion scenarios; (3) By using the time window association mechanism of power command step change, the causal sequence of grid events and inverter control actions is automatically established, providing millisecond-level precision data support for response delay analysis; (4) Dynamic threshold compensation, residual adaptive adjustment, segmented stability assessment and frequency tracking form a collaborative mechanism to adapt to different grid impedance characteristics, power change rate and frequency fluctuation scenarios; (5) Visual charts integrate the original waveform and anchor point markings, combined with tolerance manual correction interface and data interpolation update, while ensuring the automation of the main algorithm, an expert intervention channel is provided for edge cases, reducing the data waste rate. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the principle of the anchor point finding and processing method provided by the present invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0019] like Figure 1 As shown, the automated anchor point finding and processing method for photovoltaic inverter grid-connected test data provided by the present invention includes the following steps: Acquire grid-connected test data of photovoltaic inverters, including three-phase voltage signals, three-phase current signals output by the grid simulator, and power command signals issued by the inverter controller. Then, align the three-phase voltage signals, three-phase current signals, and power command signals with timestamps using a high-precision time synchronization device. The three-phase voltage signal and the three-phase current signal are preprocessed. The preprocessing includes performing multi-scale denoising based on wavelet transform and moving average filtering in sequence, and outputting the denoised voltage signal and current signal. Based on the denoised voltage signal and the preset anchor point type, perform adaptive dynamic threshold anchor point identification. The anchor point type includes voltage zero crossing point, voltage drop start point and voltage recovery termination point, and output an initial anchor point candidate set. For the candidate points of voltage drop initiation point in the initial anchor point candidate point set, perform transient energy feature verification: calculate the transient energy value of the voltage signal within the power frequency cycle before and after the candidate point of voltage drop initiation point. The transient energy value is defined as the integral of the square of the voltage deviation, where the voltage deviation is the difference between the real-time voltage and the root mean square value of the voltage before the drop. When the calculated transient energy value is greater than the preset energy threshold, the candidate point is confirmed as an effective voltage drop initiation point. The timestamp of the power command signal is associated with the timestamp of the effective voltage drop start point. If the power command signal has a step change within a set time window before the effective voltage drop start point occurs, the candidate point is marked as the final effective anchor point; otherwise, the candidate point is marked as an invalid anchor point and removed from the anchor point candidate point set. Output the timestamps and signal characteristic values ​​of the final valid anchor points corresponding to all anchor point types in the anchor point candidate set.

[0020] The automated anchor point finding and processing method for photovoltaic inverter grid-connected test data provided in this application accurately and automatically locates key time nodes (anchor points) in grid events through a series of coordinated technical steps. The entire process begins with the acquisition and time synchronization of test data. Specifically, the method first needs to collect the raw data stream generated by the photovoltaic inverter during grid-connected testing. This data includes the instantaneous three-phase voltage signals output by the grid simulator (reflecting the grid state), the instantaneous three-phase current signals (reflecting the current injected into the grid by the inverter), and the power command signals issued by the inverter's own controller (usually the target power value set by the inverter according to the operating strategy or response requirements). Considering that these signals may come from different measurement devices or system modules and have slight time offsets, this method uses a high-precision time synchronization device (such as a synchronization clock based on GPS or the IEEE 1588 protocol) to accurately timestamp each sampling point of each signal, ensuring strict alignment of different physical quantities on the time axis and providing a time reference for accurately capturing the moment of event occurrence.

[0021] After acquiring the time-synchronized data, the next step is to preprocess the original voltage and current signals to improve the signal-to-noise ratio. The power grid testing environment is complex, and the original signals often contain various high-frequency noises, harmonic interferences, or measurement noise, which can severely affect the accuracy of subsequent key point identification. A combined filtering strategy is adopted: first, noise reduction is performed using the multi-scale analysis capabilities of wavelet transform. Wavelet transform can adaptively select thresholds to process each wavelet coefficient based on the different performance of noise and signal at different scales (frequency), effectively separating and suppressing high-frequency noise while better preserving the signal's abrupt change characteristics (crucial for capturing voltage drop or recovery transition edges). After wavelet denoising, the signal is further processed by moving average filtering. Moving average smooths out residual random fluctuations and minor distortions within the power frequency cycle by averaging the data over a specified window (the window size can be selected as an integer fraction of the power frequency cycle, such as 1 / 4 cycle), resulting in smoother and cleaner denoised voltage and current signals. Preprocessing significantly improves signal quality, creating favorable conditions for subsequent anchor point detection.

[0022] Based on the preprocessed high-quality voltage signal, the process enters the initial anchor point identification stage using adaptive dynamic thresholding. Here, "anchor point" specifically refers to critical time points in power grid event analysis, including: the moment the voltage waveform crosses zero (voltage zero-crossing, crucial for phase and frequency analysis), the instant the grid voltage begins to drop suddenly (voltage drop initiation point, marking the start of a fault or disturbance), and the end point where the grid voltage recovers to a relatively stable state (voltage recovery termination point, marking the end of the disturbance and restoration of stable operation). Identifying these points cannot rely on fixed thresholds because the grid voltage itself may fluctuate or have background harmonics. Therefore, this method employs an "adaptive dynamic thresholding" strategy, analyzing the local statistical characteristics of the voltage signal (e.g., calculating the signal mean, standard deviation, or extreme values ​​within a sliding window) to dynamically generate a detection threshold that adjusts according to the signal's own variations. By comparing the real-time voltage signal with this dynamic threshold (e.g., detecting whether the voltage is below a dynamic threshold calculated based on recent voltage levels), candidate points that meet specific conditions can be initially screened (e.g., a point where the voltage changes from above the threshold to below the threshold may be a candidate point for the start of the drop), forming an initial set of anchor point candidate points, thus achieving the initial location of key event points.

[0023] The initially identified candidate point set may contain false positives (such as spurious jumps caused by noise). Therefore, it is necessary to verify the transient energy characteristics of the candidate points for the voltage drop initiation point, which is a crucial step to ensure accuracy. At the instant of the onset of a real voltage drop event, the voltage waveform usually changes drastically, accompanied by a significant release of transient energy. To quantify this transient characteristic, this method calculates the "transient energy value" of the voltage signal within a time window of a specific number of power frequency cycles (e.g., half a cycle to one cycle) before and after the candidate point. This energy value is defined as the integral of the square of the "voltage deviation" over time within this window. Here, the "voltage deviation" refers to the real-time instantaneous voltage value (or effective value) at each sampling moment minus a reference value. This reference value is usually the root mean square (RMS) value of the voltage within several power frequency cycles before the voltage drop occurs, representing the steady-state voltage level before the drop. Therefore, the voltage deviation reflects the degree of instantaneous deviation from the steady state before the drop. Integrating the square of this deviation gives the result representing the total energy magnitude of the voltage deviation from the steady state within this time window. A true voltage drop initiation point is necessarily accompanied by a significant jump in this transient energy value. This method sets a preset energy threshold (the specific method for determining this threshold can vary, such as based on historical data statistical analysis, theoretical model estimation, or dynamic calculation model). Only when the transient energy value corresponding to the calculated candidate point is greater than this preset threshold is the candidate point considered a valid voltage drop initiation point. This verification step effectively distinguishes the initiation point of real power grid events from signal noise or small fluctuations.

[0024] After confirming the valid voltage dip initiation point, the correlation of the power command signal needs to be verified. The behavior of the photovoltaic inverter during grid faults is directly affected by its control strategy, which is typically reflected in the power command signal (e.g., upon detecting a voltage dip, the control strategy may instruct the inverter to reduce active or reactive power output). To determine whether the voltage dip event actually triggered the inverter control action, this method performs a correlation analysis between the timestamp of the power command signal and the timestamp of the newly confirmed valid voltage dip initiation point. Specifically, a time window is set (e.g., 100 to 200 milliseconds before the dip initiation point), and the power command signal data sequence within this time window is checked for a significant "step change" (i.e., a rapid and significant change in the command value). There are various methods to determine step changes (e.g., piecewise linear fitting to find the slope abrupt change point, or calculating that the difference or standard deviation of the data within the time window exceeds a threshold, etc.). If a valid step change in the power command signal is detected within the set time window preceding the voltage dip initiation point, it indicates that the voltage dip event likely did indeed trigger a control response from the inverter. In this case, the candidate point is marked as the final, valid anchor point. Conversely, if no significant change in the power command is detected within the set time window, the voltage dip point is considered not to have been effectively responded to by the inverter (potentially due to islanding, test failure, or measurement issues). The candidate point is then marked as invalid and removed from the candidate set. Correlation verification greatly enhances the engineering significance of anchor point identification, linking simple grid event point identification with the actual control behavior of the inverter.

[0025] Finally, after the rigorous identification, verification, and correlation steps described above, the method outputs information on the final valid anchor points for all anchor point types (voltage zero-crossing point, voltage sag initiation point, and voltage recovery termination point). The output information includes not only the precise timestamp of each valid anchor point (derived from the initial high-precision time synchronization) but also the signal characteristic values ​​corresponding to that anchor point location (e.g., the root mean square voltage value at the voltage sag initiation point, sag depth, power command step change, etc.). This structured anchor point information and its characteristic values ​​provide a precise time reference and quantitative basis for subsequent in-depth analysis of key indicators such as the low voltage ride-through (LVRT) performance, dynamic response characteristics, and protection action time of photovoltaic inverters.

[0026] Compared to traditional methods for analyzing grid-connected photovoltaic inverter test data, this technical solution offers a significant improvement in technical performance. Traditional methods typically rely heavily on manual experience to identify and label key anchor points (such as drop start, recovery end, and zero-crossing points) on waveform diagrams, a tedious, time-consuming, and highly subjective process. This solution automates the entire anchor point finding process through clearly defined steps (preprocessing, adaptive identification, transient energy verification, and command association), freeing analysts from tedious visual inspections, resulting in a qualitative leap in data processing efficiency and a substantial reduction in the analysis cycle. Manual identification is susceptible to fatigue, experience differences, and visual errors, especially in noisy and complex waveform test data, where the accuracy of key point location is difficult to guarantee and consistency is poor. This solution introduces a multi-level verification mechanism: wavelet + moving average preprocessing effectively suppresses noise; adaptive dynamic thresholds adapt to voltage fluctuations; key transient energy feature verification uses physical properties (energy mutations) to distinguish real events from noise; and power command association provides secondary confirmation from a control behavior perspective. This multi-feature, multi-step collaborative verification significantly improves the accuracy, reliability, and robustness to different test conditions (noise levels, disturbance types) of anchor point identification. Traditional methods often provide fragmented or imprecise time correlation analysis between grid events (voltage dips / recoveries) and inverter control actions (power command changes). This scheme, through high-precision timestamp alignment and active search and time window correlation of power command step changes, accurately quantifies the time difference from the occurrence of a grid event to the start of inverter response (power command change), providing direct and objective data support for evaluating the response speed of the inverter control system. This is crucial for a deeper understanding of inverter dynamic performance. The automated process eliminates subjective differences introduced by manual operation, ensuring that the same dataset processed by this method always yields highly consistent anchor point identification results, improving the comparability of test results and the authority of reports. The adaptive threshold, transient energy calculation, and power command correlation mechanisms designed in this scheme enable it to better handle non-ideal situations that may be encountered in actual testing, such as slight voltage drops under voltage fluctuation backgrounds, waveforms containing harmonics, and diverse command variations caused by complex inverter control strategies, thus expanding the applicability of the method. Through systematic step design and comprehensive application of key technologies, this method effectively solves the problem of automated and high-precision identification of key anchor points in the analysis of photovoltaic inverter grid-connected test data, providing strong technical support for the objective, efficient, and in-depth evaluation of inverter performance.

[0027] In another technical solution, the method for determining the preset energy threshold corresponding to the transient energy value includes: Obtain voltage signal waveform data for consecutive power frequency cycles before the candidate starting point of voltage drop, and extract the voltage signal waveform data as the reference voltage segment; The root mean square value of the reference voltage is calculated for the entire cycle of the reference voltage range to obtain the root mean square value of the reference voltage. A dynamic energy threshold calculation model is established based on the root mean square value of the reference voltage. The dynamic energy threshold calculation model satisfies the following relationship: the dynamic energy threshold is equal to the square of the root mean square value of the reference voltage multiplied by the scaling factor. The proportionality coefficient is obtained by calibration through historical test datasets. The calibration method is as follows: select the actual voltage drop start point that has been manually verified in the historical test data, calculate the actual transient energy value corresponding to each actual voltage drop start point and the square of the root mean square value of the corresponding reference voltage, fit the linear relationship between the actual transient energy value and the square of the root mean square value of the reference voltage using the least squares method, and the slope of the fitted line is the proportionality coefficient. The dynamic energy threshold calculated in real time is used as the preset energy threshold for the transient energy characteristic verification process.

[0028] This solution primarily addresses the dynamic generation of preset energy thresholds in the transient energy verification process. First, the system needs to acquire raw voltage waveform data for consecutive power frequency cycles (typically 2-4 cycles) before the candidate voltage drop initiation point, defining this as the reference voltage segment. This data represents the steady-state operation of the power grid before the disturbance, and its importance lies in providing a reference benchmark for subsequent threshold calculations. By performing full-cycle root mean square (RMS) value calculation on the reference voltage segment, the reference voltage RMS value V is obtained. base This value quantifies the stability level of the grid voltage before the voltage drop. Based on this, a dynamic energy threshold calculation model was established, the core relationship of which can be expressed as follows: the dynamic energy threshold is proportional to the square of the root mean square value of the reference voltage, and the proportionality coefficient (e.g., within the range of 0.15-0.25) is obtained through calibration using historical data. This design stems from the physical law that the transient energy of the grid is positively correlated with the square of the steady-state voltage level, thus ensuring that the threshold can adapt to test scenarios of different voltage levels.

[0029] The calibration of the proportional gain depends on historical test datasets. It requires manual selection of rigorously verified actual voltage drop initiation points from the historical data (at least 50 sets are recommended). For each initiation point, two key values ​​are calculated: first, the actual transient energy value within the power frequency cycle before and after the candidate point (i.e., the integral of the square of the voltage deviation); and second, the corresponding root mean square value of the reference voltage, V. base The square of the value is then used. Subsequently, these data points are linearly fitted using the least squares method, with the actual transient energy value as the vertical axis and the root mean square value of the reference voltage V as the horizontal axis. base The square of the result is used as the horizontal axis, and the slope of the fitted line is the proportionality coefficient. This process is completed in the offline calibration stage, and the fitting results can be stored as a configuration file for real-time system calls. If historical data is insufficient, a theoretical model compensation strategy can be temporarily used, but the calibration method is still the first choice because it directly reflects the statistical characteristics of the actual test environment.

[0030] During the online detection phase, the system calculates the current root mean square value V of the reference voltage in real time.base By substituting the calibrated scaling factor, a dynamic energy threshold is instantly generated. This threshold serves as the criterion for transient energy verification: only when the measured transient energy value of a candidate point exceeds this threshold is it confirmed as the effective voltage drop initiation point. The dynamic threshold overcomes the misjudgment problem of fixed thresholds in low-voltage (e.g., 220V) and high-voltage (e.g., 380V) testing scenarios, significantly reducing false positives caused by noise interference. The scaling factor, calibrated based on historical data, makes the threshold more closely match the actual power grid characteristics, significantly improving the detection rate of weak voltage drop events (e.g., 10%-20% depth) compared to traditional empirical methods. Continuous expansion of the calibration dataset automatically optimizes the scaling factor without requiring readjustment of algorithm parameters.

[0031] When the number of actual voltage drop start points in the historical test dataset does not reach the set number, the following method is used for scaling factor physical model compensation: Establish a theoretical model for the transient energy of voltage sag: E theory V represents the theoretical energy value, i.e., the preset energy threshold. base The reference voltage is the root mean square value, d is the real-time voltage drop depth, and k is the base value of the proportional coefficient. base Obtained through electromagnetic transient simulation data of typical power grid topologies; Calculate the real-time voltage drop depth V dip The root mean square value of the voltage at the candidate point where the voltage drop begins; Calculate and use the theoretical energy value as the preset energy threshold; confirm the new effective voltage drop initiation point through the voltage drop transient energy theoretical model; The actual transient energy value corresponding to the newly confirmed effective voltage drop initiation point and the square of the root mean square value of the reference voltage are added to the historical test dataset. When the number of effective voltage drop initiation points reaches the set number, the process switches to obtaining the scaling factor through calibration using the historical test dataset.

[0032] When the number of actual voltage sag initiation points in the historical dataset is insufficient (e.g., less than 20 sets), the system automatically switches to theoretical model compensation mode. The core of this mode is constructing a theoretical model of voltage sag transient energy: theoretical energy value E. theory It consists of a product of three parts: the root mean square value of the reference voltage V base The square of , the square of the real-time voltage drop depth d, and the base value of the proportional coefficient k base The drop depth d is calculated in real time (V). base -V dip ) / V base , obtain (V dip, where is the root mean square voltage value at the candidate point, its physical meaning is to quantify the severity of the voltage drop (e.g., 20% depth corresponds to d=0.2). The base value of the scaling factor k. base The inherent laws of energy release under different network structures are reflected by electromagnetic transient simulation of typical power grid topologies (such as IEEE 9-node or 14-node systems).

[0033] The system first bases its V on the candidate points base and V dip Calculate the real-time drop depth d, and then combine it with the pre-stored k. base (For example, in the range of 0.05-0.15) Generate the theoretical energy value E theory This is directly used as the preset threshold for transient energy verification. Simultaneously, for each newly confirmed effective voltage drop initiation point, its corresponding actual transient energy value is correlated with V... base The squared value is automatically added to the historical dataset. When the dataset reaches a set size (e.g., 50 sets), the system automatically switches back to the calibration mode mentioned earlier, replacing the theoretical model with a statistical model. This design achieves a smooth transition during the cold start phase: initially relying on physical laws to ensure basic accuracy, and later gradually approaching the real scenario through data accumulation. This method solves the problem of threshold generation in scenarios with scarce historical data, avoiding algorithm failure due to insufficient calibration data. The theoretical model covers typical power grid topologies (e.g., radial / ring networks), ensuring reasonable accuracy in unknown test environments, especially more stable for deep drop (>40%) event identification. The automatic collection of new data and seamless switching of threshold calculation modes enable the system to have continuous optimization capabilities, and the accuracy approaches that of the data-driven model after long-term operation. Only a small number of typical topology simulations are needed to determine V. base This avoids the need for high-cost full-domain simulation. A dual-guarantee mechanism for adaptive energy thresholds is constructed: a dynamic calibration mode utilizes historical data to reflect the statistical patterns of the real power grid, improving accuracy in common test scenarios; a theoretical compensation mode, based on a physical model, covers data-scarce scenarios, ensuring system availability throughout its entire lifecycle. Threshold setting is freed from reliance on manual experience, adapting to different voltage levels, voltage dip depths, and power grid topologies; the false alarm rate in the transient energy verification process is significantly reduced, especially for shallow voltage dips and high-noise scenarios; and through the co-evolution of data and models, long-term maintenance costs are significantly reduced.

[0034] In another technical solution, the method for determining whether the power command signal has a step change includes the following steps: Acquire the data sequence of the power command signal within a set time window, perform piecewise linear fitting on the data sequence, divide the data sequence into continuous sub-segments, and ensure that the sum of squared residuals of the linear fitting of data points in each sub-segment is less than a set residual threshold. Calculate the absolute value of the slope difference between adjacent segments. When the absolute value of the slope difference exceeds the set slope change threshold, mark the connection point of the adjacent segments as a potential step point. Extract power command data within a set time period before and after the potential step point, calculate the steady-state mean of the segment before and after the potential step point; calculate the absolute value of the difference between the two steady-state means, and when the absolute value of the difference is greater than the set percentage of the full-scale power command value and greater than the set minimum change amount, confirm the potential step point as a valid step change point; record the timestamp of the valid step change point and use it for the correlation verification process.

[0035] An anti-interference quantization method is designed to determine step changes in power command signals. The system first extracts a power command signal data sequence centered on a candidate point where the voltage drop inflection point is located, extending forward by a predetermined time window (e.g., 100-300 milliseconds). This time window covers the possible response delay range of the inverter. Traditional differential methods are prone to misjudgment when faced with noise or gradual fluctuations in the signal. Therefore, this scheme employs a piecewise linear fitting technique, dividing the continuous data sequence into several segments. Iterative calculations ensure that the sum of squared residuals between the data points in each segment and the fitted line is less than a preset residual threshold (e.g., the sum of squares corresponding to 0.5%-2% of the full-scale value). Real step changes disrupt the linear continuity of the data, while noise fluctuations can maintain local linearity. By controlling the fitting residuals within each segment, it ensures that each segment is an approximately linear steady-state or gradually changing process, thereby pushing nonlinear jump points towards the segment boundaries.

[0036] After data segmentation, the system calculates the absolute value of the slope difference between the fitted lines of adjacent segments. If this value exceeds a set slope change threshold (e.g., corresponding to an angle change of 10°-30°), the connection point between the two segments is marked as a potential step point. To further eliminate transient interference, the system extracts data for a fixed duration (e.g., 20-50 milliseconds) before and after the potential point, and calculates the steady-state mean of the segment before the step point (excluding data near the jump point) and the steady-state mean of the segment after the step point. A valid step is confirmed only when the absolute value of the difference between the two means simultaneously meets two conditions: first, the difference is greater than a set percentage of the full-scale power command (e.g., 5%-15%); second, the difference is greater than the minimum change (e.g., 1%-3% of the rated power). This dual threshold design avoids misjudgment of minor fluctuations while ensuring the detection of command changes that are significant in engineering terms.

[0037] This scheme employs a piecewise fitting strategy to effectively filter out high-frequency noise and slowly varying interference. Compared to the traditional derivative method, the false alarm rate is significantly reduced in test environments with high noise levels (e.g., signal-to-noise ratio <20dB). The dual threshold mechanism considers both relative magnitude and absolute change, adapting to both the micro-change commands of small-power inverters (e.g., 10kW level) and the drastic jumps of high-power systems (MW level). By locating the segment connection point rather than the entire jump interval, the step timestamp determination accuracy can reach the millisecond level, providing a reliable basis for response delay analysis.

[0038] In the piecewise linear fitting process, the set residual threshold is dynamically adjusted. Specific methods include: Calculate the absolute value of the average rate of change of the power command signal within a set time window; When the absolute value of the average rate of change is less than the set rate threshold, the set residual threshold will be adjusted to the baseline residual threshold. When the absolute value of the average rate of change is greater than or equal to the set rate threshold, the set residual threshold is adjusted according to the following relationship: Adjust the set residual threshold to the ratio of the baseline residual threshold multiplied by the absolute value of the average rate of change to the set rate threshold; Piecewise linear fitting is performed using the adjusted residual threshold; the baseline residual threshold is determined by the fitting residual statistics of the steady-state power command segment in historical test data.

[0039] The residual threshold for piecewise linear fitting is not fixed because a fixed residual threshold might lead to overly fine segmentation during periods of rapid power command change (such as ramp tests), generating numerous spurious step points. To avoid this problem, a real-time adaptive mechanism for the residual threshold is introduced. When the overall power command changes smoothly, a smaller residual threshold should be used to capture subtle jumps; when the command is in a rapid adjustment period, the threshold should be relaxed to avoid mistakenly segmenting continuous changes into multiple segments. The system first calculates the absolute value of the average rate of change of the power command within a set time window (unit: % / millisecond) and uses a preset rate threshold (e.g., 0.1%-0.3% / millisecond) as the operating condition boundary. The baseline residual threshold is determined by analyzing the fitting residual statistics (e.g., 90th percentile) of steady-state power command segments (duration > 1 second) in historical data, ensuring coverage of the vast majority of steady-state fluctuation scenarios.

[0040] The system monitors the average rate of change of power commands in real time. If the rate is lower than a preset rate threshold, it determines that the system is in a quasi-steady-state condition and directly uses the baseline residual threshold to perform piecewise fitting. If the rate is higher than or equal to the preset threshold, the residual threshold is scaled up proportionally: adjusted threshold = baseline value × (real-time average rate / preset rate threshold). For example, when the real-time rate is twice the preset value, the residual threshold is simultaneously multiplied by two. This linear scaling relationship ensures that the granularity of the segment division matches the command change rate. During periods of rapid change, a coarser fit is allowed (reducing the number of segments), while during periods of slow change, a finer fit is pursued (capturing small jumps). The adjusted threshold is immediately applied to the piecewise linear fitting of the current time window, and the entire process is completed in milliseconds.

[0041] This solution employs a dynamic threshold to enable the step detection algorithm to be compatible with both steady-state testing (such as fixed power point operation) and dynamic testing (such as power ramp-up and ramp-down), avoiding the high false alarm rate (tens of times higher) of traditional methods in variable-speed scenarios. It automatically reduces the number of segments during rapid changes, lowering computational complexity and improving real-time processing efficiency, especially for long time windows (>500ms) or high sampling rates (>10kHz). The benchmark threshold is derived from historical data statistics, and the rate threshold setting has a clear physical meaning (typical change within the power frequency cycle), eliminating the need to readjust parameters for different inverter models. An intelligent closed-loop system for power command step detection is constructed, providing a core method for high-precision step positioning. Accuracy is ensured through piecewise fitting and dual verification; dynamic environment adaptability is provided, and the residual threshold is optimized in real time to solve the false alarm problem in variable-speed scenarios. Even under complex test conditions (noise superimposed with slowly / rapidly changing commands), it maintains an extremely low rate of missed detections and false alarms; it provides reliable technical support for inverter control response timing analysis, meeting the stringent millisecond-level time accuracy requirements in grid connection standards; it significantly reduces the algorithm's reliance on prior knowledge of the test scenario, and improves the generalization ability of the automated test system.

[0042] In another technical solution, the transient oscillation stability is verified using the following method for the voltage recovery termination point candidate points in the initial anchor point candidate point set: Extract the noise-reduced voltage signal waveform for multiple consecutive power frequency cycles after the candidate point of voltage recovery termination. Detect the local maximum and local minimum sequence of the voltage signal waveform, calculate the amplitude difference between adjacent maximum and minimum points, and generate an amplitude difference sequence. Calculate the attenuation rate of the amplitude difference sequence. The attenuation rate is the ratio of the amplitude difference of the next set to the amplitude difference of the previous set. If three consecutive sets of attenuation rate values ​​are all greater than the set attenuation threshold, the voltage signal is determined to be in a stable recovery state. Only when the candidate point of voltage recovery termination point simultaneously meets the adaptive dynamic threshold anchor point identification condition and the stable recovery state determination condition is the candidate point confirmed as an effective voltage recovery termination point.

[0043] This solution proposes a transient oscillation stability verification method for candidate voltage recovery termination points. After grid fault clearance, the voltage recovery process is often accompanied by damped oscillations (such as LC resonance or controller overshoot). Prematurely determining the termination point will lead to distortion in subsequent analysis. To quantify the degree of oscillation attenuation, the system extracts the denoised voltage waveform for several consecutive power frequency cycles (usually 3-5 cycles) after the candidate point. By real-time detection of the local maximum and minimum point sequences of the voltage signal (e.g., using differential sign change or curvature extremum method for localization), the amplitude difference (i.e., peak-to-peak change) between adjacent maxima and minima is calculated, forming an amplitude difference sequence. This sequence is essentially a discrete sampling of the oscillation envelope, and its attenuation trend directly reflects the system's energy dissipation rate.

[0044] Based on the amplitude difference sequence, the system calculates the ratio of the amplitude differences between adjacent groups, defined as the attenuation rate (the amplitude difference of the later group divided by the previous group). For example, the attenuation rates of the sequence δ1, δ2, δ3, and δ4 are δ2 / δ1, δ3 / δ2, and δ4 / δ3, respectively. An attenuation threshold (recommended range 0.7-0.9) is set as a stability criterion: when three consecutive attenuation rates are all greater than this threshold (i.e., δ2 / δ1 > threshold, δ3 / δ2 > threshold, δ4 / δ3 > threshold), the oscillation is determined to be in a continuously decaying state. This condition implies that the envelope exhibits a monotonically decreasing trend, and the attenuation rate is controllable (non-divergent or constant-amplitude oscillation). Only when a candidate point simultaneously satisfies both the initial adaptive dynamic threshold identification condition and this attenuation stability condition is it confirmed as an effective voltage recovery termination point.

[0045] This scheme avoids misjudging transient extreme points during oscillation as termination points, and significantly improves the identification accuracy, especially for underdamped systems (such as scenarios with a high proportion of new energy access). The decay rate sequence is not sensitive to local waveform distortions (such as harmonic glitches) and is more robust than methods based on derivatives or the spectrum. The short time window design of 3-5 periods meets the real-time requirements and is suitable for online detection systems with high sampling rates (10kHz level).

[0046] When there is no situation where three consecutive sets of attenuation rate determinations are all greater than the set attenuation threshold, a segmented stability weighted evaluation is performed, including the following steps: The amplitude difference sequence of continuous power frequency cycles after the candidate point of voltage recovery termination is divided into a first subsequence and a second subsequence. The average attenuation rate of the first subsequence is calculated. The attenuation rate of the front section δ i Let m be the amplitude difference between adjacent maxima and minima in the i-th group, and m be the length of the first subsequence; calculate the average decay rate of the second subsequence. The attenuation rate in the latter part δ j is the magnitude difference between adjacent maxima and minima in the j-th group, and p is the length of the subsequent subsequence; Calculate the overall stability coefficient using the weighted evaluation formula. , where w front For the front-end weights, w rear For the weight of the later stage; If the overall stability coefficient is greater than the set attenuation threshold, the voltage signal is determined to be in a stable recovery state, and this determination result is used as the final determination result for the three consecutive sets of attenuation rates.

[0047] In some cases, when the previously set continuous decay rate judgment condition is not met (such as the presence of non-monotonic decay or noise interference), the system adopts a segmented stability weighted evaluation strategy. First, the amplitude difference sequence of continuous power frequency cycles after the candidate point (suggested length 6-10 groups) is divided into a first subsequence (close to the recovery start, with violent oscillations) and a second subsequence (closer to the current time point, approaching steady state). The division point can be selected as the midpoint of the sequence or adaptively determined based on energy mutation detection. For example, for 8 groups of sequences, the first 4 groups are the first segment (m=4), and the last 4 groups are the second segment (p=4).

[0048] The mean decay rate (arithmetic mean of the ratios of adjacent amplitude differences) is calculated for the first subsequence, reflecting the decay intensity in the initial stage of oscillation. The mean decay rate is also calculated for the second subsequence, characterizing the recent decay trend. A comprehensive stability coefficient is calculated based on the weighting principle (suggested weights for the first subsequence: 0.3-0.4; 0.6-0.7 for the second subsequence): Comprehensive coefficient = Mean of first subsequence × Weight of first subsequence + Mean of second subsequence × Weight of second subsequence. If this coefficient is greater than the set decay threshold, the system is still considered to be in a stable recovery state. In the weighting design, the state of the second subsequence is more decisive for stability, but the violent oscillations in the first subsequence need to be appropriately tolerated.

[0049] This scheme effectively handles non-ideal attenuation waveforms (such as bi-peak oscillations and noise interference interrupting monotonicity), significantly improving coverage of actual power grid waveforms compared to strict continuous conditions. By differentiating the weights of the preceding and following segments, it better reflects the essential characteristic of "gradual convergence of oscillation amplitude" in physical systems. Abnormal attenuation rates in a single group (such as due to measurement interference) do not affect the overall judgment, reducing stringent requirements for data quality. A dual-track mechanism for determining the voltage recovery termination point is constructed. On the one hand, the strict continuous attenuation condition is applicable to ideal damping scenarios, ensuring high accuracy; on the other hand, the segmented weighting strategy covers complex oscillation conditions, providing a degradation solution. It can reliably determine the recovery endpoint under various power grid impedance characteristics (strong damping / weak damping), avoiding the systematic bias of traditional fixed-time-window methods. It significantly reduces the oscillation misjudgment rate caused by controller interaction in new energy power plant testing and provides a technical basis for grid-connected stability assessment during inverter fault recovery (such as LVRT recovery time).

[0050] In another technical solution, for the candidate voltage zero-crossing points in the initial anchor point candidate point set, the following method is used for multi-cycle phase consistency verification: Extract the denoised voltage signal from multiple consecutive power frequency cycles before the candidate voltage zero-crossing point; for each power frequency cycle, detect the voltage zero-crossing point and generate a historical zero-crossing point timestamp sequence. Calculate the difference between adjacent timestamps in the historical zero-crossing timestamp sequence to obtain the actual periodic value sequence, and calculate the absolute deviation value sequence between the actual periodic value sequence and the standard power frequency cycle. If the maximum value of the absolute deviation value sequence does not exceed the set phase tolerance threshold, the current voltage zero-crossing candidate point is confirmed as an effective voltage zero-crossing point; otherwise, the median of the historical zero-crossing timestamp sequence is used to correct the timestamp of the current candidate point.

[0051] Power grid frequency fluctuations and harmonic pollution can cause phase shifts in traditional single-cycle zero-crossing detection. This solution employs a multi-cycle phase consistency verification mechanism: for initially identified voltage zero-crossing candidate points, the system extracts denoised voltage signals from several consecutive preceding power frequency cycles (typically 4-8 cycles). Zero-crossings are detected independently for each cycle (e.g., through zero-crossing or curve fitting), generating a historical zero-crossing timestamp sequence. This sequence implicitly contains the actual trajectory of power grid frequency changes; its physical meaning is that if the power grid operates stably, the interval between adjacent zero-crossings should be close to the theoretical power frequency cycle (20ms@50Hz).

[0052] The system calculates the difference between adjacent timestamps in the historical sequence to obtain the actual period value sequence. Each actual period is compared with the standard power frequency period (e.g., 20ms for 50Hz) to generate an absolute deviation value sequence (microsecond level). A phase tolerance threshold is set (recommended 100-300 microseconds, corresponding to 0.9°-2.7° electrical angle) as the maximum allowable deviation. If all deviation values ​​are below the threshold, the current candidate point is confirmed as a valid zero-crossing point. Otherwise, it indicates the existence of cumulative phase error. In this case, the median position of the historical timestamp sequence is used to correct the timestamp of the current candidate point (e.g., taking the median of the first 6 zero-crossing timestamps as the new timestamp). This correction essentially uses statistical stability to resist single-cycle disturbances, eliminates zero-crossing drift caused by small frequency fluctuations in the power grid (within ±0.5Hz), and improves phase accuracy to sub-millisecond level; multi-cycle statistics avoid harmonic zero-crossing distortion (e.g., false zero-crossing caused by the 3rd harmonic); and provides a highly reliable time reference for subsequent voltage / current phase difference analysis, meeting the requirements of power factor testing.

[0053] The standard power frequency cycle is dynamically generated using the following method: Obtain the actual periodic value sequence of multiple consecutive power frequency cycles before the candidate point of voltage zero crossing, and perform a moving average processing on the obtained actual periodic value sequence, with the moving window width being half the number of cycles. The median of the sequence after moving average processing is used as the dynamic reference period. When calculating the absolute deviation value sequence, the dynamic reference period is used instead of the fixed standard power frequency period. If the maximum value of the absolute deviation value sequence is less than the set phase tolerance threshold, the median of the historical zero-crossing timestamp sequence is still used to correct the current candidate point timestamp. If the deviation between the dynamic reference period and the rated power frequency period exceeds the set frequency tolerance threshold, a frequency abnormality alarm signal will be sent to the test platform.

[0054] Fixed standard power frequency cycles (e.g., 50Hz) may be insufficiently applicable in power grids with high penetration of new energy sources. This solution proposes a dynamic reference cycle generation technique. The system acquires a sequence of actual cycle values ​​for several consecutive cycles (6-10 recommended) preceding a candidate point, and smooths out random fluctuations through moving average processing (window width half the number of cycles). For example, for 8 cycles of data, with a window width of 4, the average values ​​of cycles 1-4, 2-5, ..., 5-8 are calculated sequentially. The median of these moving averages is taken as the dynamic reference cycle (e.g., 20.05ms), which reflects the current average frequency of the power grid in real time.

[0055] In phase consistency verification, the system uses a dynamic reference period instead of a fixed standard period to calculate the absolute deviation. If the deviation still exceeds the limit, the median of historical zero-crossing points is used to correct the timestamp. Simultaneously, the deviation between the dynamic reference period and the rated value (e.g., 20ms) is monitored: when it exceeds the frequency tolerance threshold (recommended ±0.15-0.25Hz), a frequency anomaly alarm is sent to the test platform. For example, an alarm is triggered when the dynamic reference period is 21ms (corresponding to 47.6Hz) and the threshold is set to ±0.2Hz. This design deeply couples anchor point correction with grid condition monitoring, supporting frequency fluctuation scenarios of 45-55Hz and covering special operating conditions such as islanded testing; the frequency alarm function provides testers with early warnings of grid anomalies (e.g., primary frequency regulation failure); the dynamic reference compensates for system-level frequency drift, further reducing phase measurement errors. A dual-protection system for voltage zero-crossing point identification is constructed to solve the single-cycle disturbance problem, ensuring local accuracy through multi-cycle statistics; and to address system-level frequency drift, global adaptability is achieved through the dynamic reference. It can provide highly reliable zero-crossing markers under various power grid environments (stable / weak / islanded), laying the foundation for grid-connected synchronization performance testing; it upgrades simple anchor point identification into a power grid state perception tool, outputting key indicators of frequency stability; and it significantly reduces the data invalidation rate caused by background frequency fluctuations in new energy grid-connected testing.

[0056] In another technical solution, the specific method for outputting the timestamps and signal characteristic values ​​of the final valid anchor points corresponding to all anchor point types includes: Generate a visualization chart containing timestamp-aligned denoised voltage signal waveforms, power command signal waveforms, and final effective anchor point location markers; Data labels are associated with the final valid anchor point location marker. The data labels include the anchor point type name, the precise timestamp, and the root mean square voltage value at the corresponding time. A manual position correction interface is set up. This interface responds to the user's drag operation on the final effective anchor point position mark in the visualization chart, obtains the new timestamp coordinates after dragging, and calculates the difference between the new timestamp coordinates and the original timestamp as the manual offset. When the absolute value of the manual offset does not exceed the predetermined tolerance window, the corrected anchor point data is generated based on the new timestamp coordinates and stored independently. The signal feature values ​​of the corrected anchor point data are updated through interpolation calculation.

[0057] This solution features an efficient output format and interaction mechanism for anchor point data. The system first generates a multi-signal coaxial visualization chart, displaying the noise-reduced voltage waveform and power command signal waveform, with strictly aligned timestamps, superimposed on the same time axis using different colors / line styles. The chart also precisely marks the positions of various final effective anchor points (e.g., a red triangle marks the start of a voltage drop, and a green circle marks the end of a voltage recovery). This visualization design allows testers to intuitively understand the timing relationship between grid events and inverter responses. At each anchor point, the system associates dynamic data labels: when the cursor hovers over or clicks the label, a label box automatically pops up displaying key information: anchor point type name (e.g., "Voltage DipStart"), precise timestamp (formatted as "hour:minute:second.millisecond"), and the corresponding root mean square voltage value (e.g., "215.3V"). The label data directly originates from the algorithm processing results, creating a seamless connection between machine interpretation and human understanding.

[0058] To handle extremely complex waveform scenarios (such as multiple superimposed disturbances), the system includes a manual position correction interface. Users can drag anchor point markers in the visualization chart (fine-tuning the time position horizontally). The background system captures the new timestamp coordinates after dragging in real time and calculates the difference between them and the original algorithm marker position (manual offset). The system presets a tolerance window (recommended within ±5ms) as the allowable correction range. If the offset is within this window (e.g., +3ms), corrected anchor point data is generated based on the new coordinates and stored independently (while retaining the original data for comparison). Simultaneously, the signal characteristic values ​​of the anchor point are automatically updated: through linear interpolation or spline interpolation techniques, the instantaneous voltage value, power command value, etc., corresponding to the new timestamp are extracted from the original signal data, and derived parameters such as the root mean square voltage value are recalculated. This design respects expert experience while preventing irrational offsets through tolerance limits.

[0059] The visualization charts integrate multi-dimensional information (original waveform + anchor point + feature value), reducing the time required for traditional manual correlation analysis (which typically takes hours) to minutes, significantly improving the efficiency of batch test report generation. The tolerance-based manual correction interface balances algorithm accuracy with expert experience, resolving ambiguities in complex edge cases (such as scenarios where voltage oscillations and command jumps coincide), and reducing the rework rate of data post-processing. The correction records (original anchor point + manual offset + corrected data) are stored independently, meeting the requirements of the laboratory quality system for data change auditing. Interpolation calculations ensure that the manually corrected feature values ​​strictly match the new timestamp, avoiding errors from secondary manual calculations.

[0060] It should be noted that although the steps are described in a specific order above, this does not mean that they must be performed in that order. In fact, some of these steps can be executed concurrently, or even in a different order, as long as the required functionality is achieved. The number of devices and processing scale described herein are for simplification of the invention; applications, modifications, and variations of this invention will be readily apparent to those skilled in the art.

[0061] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. An automated method for finding and processing grid-connected test data of photovoltaic inverters, characterized in that, Includes the following steps: Acquire grid-connected test data of photovoltaic inverters, including three-phase voltage signals, three-phase current signals output by the grid simulator, and power command signals issued by the inverter controller. Then, align the three-phase voltage signals, three-phase current signals, and power command signals with timestamps using a high-precision time synchronization device. The three-phase voltage signal and the three-phase current signal are preprocessed. The preprocessing includes performing multi-scale denoising based on wavelet transform and moving average filtering in sequence, and outputting the denoised voltage signal and current signal. Based on the denoised voltage signal and the preset anchor point type, perform adaptive dynamic threshold anchor point identification. The anchor point type includes voltage zero crossing point, voltage drop start point and voltage recovery termination point, and output an initial anchor point candidate set. For the candidate points of voltage drop initiation point in the initial anchor point candidate point set, perform transient energy feature verification: calculate the transient energy value of the voltage signal within the power frequency cycle before and after the candidate point of voltage drop initiation point. The transient energy value is defined as the integral of the square of the voltage deviation, where the voltage deviation is the difference between the real-time voltage and the root mean square value of the voltage before the drop. When the calculated transient energy value is greater than the preset energy threshold, the candidate point is confirmed as an effective voltage drop initiation point. The timestamp of the power command signal is associated with the timestamp of the effective voltage drop start point. If the power command signal has a step change within a set time window before the effective voltage drop start point occurs, the candidate point is marked as the final effective anchor point; otherwise, the candidate point is marked as an invalid anchor point and removed from the anchor point candidate point set. Output the timestamps and signal characteristic values ​​of the final valid anchor points corresponding to all anchor point types in the anchor point candidate set.

2. The method for automated anchor point finding and processing of photovoltaic inverter grid-connected test data according to claim 1, characterized in that, The methods for determining the preset energy threshold corresponding to the transient energy value include: Obtain voltage signal waveform data for consecutive power frequency cycles before the candidate starting point of voltage drop, and extract the voltage signal waveform data as the reference voltage segment; The root mean square value of the reference voltage is calculated for the entire cycle of the reference voltage range to obtain the root mean square value of the reference voltage. A dynamic energy threshold calculation model is established based on the root mean square value of the reference voltage. The dynamic energy threshold calculation model satisfies the following relationship: the dynamic energy threshold is equal to the square of the root mean square value of the reference voltage multiplied by the scaling factor. The proportionality coefficient is obtained by calibration through historical test datasets. The calibration method is as follows: select the actual voltage drop start point that has been manually verified in the historical test data, calculate the actual transient energy value corresponding to each actual voltage drop start point and the square of the root mean square value of the corresponding reference voltage, fit the linear relationship between the actual transient energy value and the square of the root mean square value of the reference voltage using the least squares method, and the slope of the fitted line is the proportionality coefficient. The dynamic energy threshold calculated in real time is used as the preset energy threshold for the transient energy characteristic verification process.

3. The method for automated anchor point finding and processing of photovoltaic inverter grid-connected test data according to claim 2, characterized in that, When the number of actual voltage drop start points in the historical test dataset does not reach the set number, the following method is used for scaling factor physical model compensation: Establish a theoretical model for the transient energy of voltage sag: E theory V is the theoretical energy value. base The reference voltage is the root mean square value, d is the real-time voltage drop depth, and k is the base value of the proportional coefficient. base Obtained through electromagnetic transient simulation data of typical power grid topologies; V dip The root mean square value of the voltage at the candidate point where the voltage drop begins; Calculate the theoretical energy value and use it as the preset energy threshold; confirm the new effective voltage drop initiation point through the voltage drop transient energy theoretical model; The actual transient energy value corresponding to the newly confirmed effective voltage drop initiation point and the square of the root mean square value of the reference voltage are added to the historical test dataset. When the number of effective voltage drop initiation points reaches the set number, the process switches to obtaining the scaling factor through calibration using the historical test dataset.

4. The method for automated anchor point finding and processing of photovoltaic inverter grid-connected test data according to claim 1, characterized in that, The method for determining whether the power command signal has a step change includes the following steps: Acquire the data sequence of the power command signal within a set time window, perform piecewise linear fitting on the data sequence, divide the data sequence into continuous sub-segments, and ensure that the sum of squared residuals of the linear fitting of data points in each sub-segment is less than a set residual threshold. Calculate the absolute value of the slope difference between adjacent segments. When the absolute value of the slope difference exceeds the set slope change threshold, mark the connection point of the adjacent segments as a potential step point. Extract power command data within a set time period before and after the potential step point, calculate the steady-state mean of the segment before and after the potential step point; calculate the absolute value of the difference between the two steady-state means, and when the absolute value of the difference is greater than the set percentage of the full-scale power command value and greater than the set minimum change amount, confirm the potential step point as a valid step change point; record the timestamp of the valid step change point and use it for the correlation verification process.

5. The method for automated anchor point finding and processing of photovoltaic inverter grid-connected test data according to claim 4, characterized in that, In the piecewise linear fitting process, the set residual threshold is dynamically adjusted. Specific methods include: Calculate the absolute value of the average rate of change of the power command signal within a set time window; When the absolute value of the average rate of change is less than the set rate threshold, the set residual threshold will be adjusted to the baseline residual threshold. When the absolute value of the average rate of change is greater than or equal to the set rate threshold, the set residual threshold is adjusted according to the following relationship: Adjust the set residual threshold to the ratio of the baseline residual threshold multiplied by the absolute value of the average rate of change to the set rate threshold; Piecewise linear fitting is performed using the adjusted residual threshold; the baseline residual threshold is determined by the fitting residual statistics of the steady-state power command segment in historical test data.

6. The method for automated anchor point finding and processing of photovoltaic inverter grid-connected test data according to claim 1, characterized in that, For the candidate voltage recovery termination points in the initial anchor point candidate set, the following method is used to verify the transient oscillation stability: Extract the noise-reduced voltage signal waveform for multiple consecutive power frequency cycles after the candidate point of voltage recovery termination. Detect the local maximum and local minimum sequence of the voltage signal waveform, calculate the amplitude difference between adjacent maximum and minimum points, and generate an amplitude difference sequence. Calculate the attenuation rate of the amplitude difference sequence, which is the ratio of the amplitude difference of the next set to the amplitude difference of the previous set. If three consecutive attenuation rates are greater than the set attenuation threshold, the voltage signal is determined to be in a stable recovery state. Only when the candidate point for voltage recovery termination simultaneously meets the adaptive dynamic threshold anchor point identification condition and the stable recovery state determination condition is the candidate point confirmed as an effective voltage recovery termination point.

7. The method for automated anchor point finding and processing of photovoltaic inverter grid-connected test data according to claim 6, characterized in that, When there is no situation where three consecutive sets of attenuation rate determinations are all greater than the set attenuation threshold, a segmented stability weighted evaluation is performed, including the following steps: The amplitude difference sequence of continuous power frequency cycles after the candidate point of voltage recovery termination is divided into a first subsequence and a second subsequence. The average attenuation rate of the first subsequence is calculated. The attenuation rate of the front section δ i Let m be the amplitude difference between adjacent maxima and minima in the i-th group, and m be the length of the first subsequence; calculate the average decay rate of the second subsequence. The attenuation rate in the latter part δ j is the magnitude difference between adjacent maxima and minima in the j-th group, and p is the length of the subsequent subsequence; Calculate the overall stability coefficient using the weighted evaluation formula. , where w front For the front-end weights, w rear For the weight of the later stage; If the overall stability coefficient is greater than the set attenuation threshold, the voltage signal is determined to be in a stable recovery state, and this determination result is used as the final determination result for the three consecutive sets of attenuation rates.

8. The method for automated anchor point finding and processing of photovoltaic inverter grid-connected test data according to claim 1, characterized in that, For the candidate zero-crossing voltage points in the initial anchor point candidate set, the following method is used for multi-cycle phase consistency verification: Extract the denoised voltage signal from multiple consecutive power frequency cycles before the candidate voltage zero-crossing point; for each power frequency cycle, detect the voltage zero-crossing point and generate a historical zero-crossing point timestamp sequence. Calculate the difference between adjacent timestamps in the historical zero-crossing timestamp sequence to obtain the actual periodic value sequence, and calculate the absolute deviation value sequence between the actual periodic value sequence and the standard power frequency cycle. If the maximum value of the absolute deviation value sequence does not exceed the set phase tolerance threshold, the current voltage zero-crossing candidate point is confirmed as an effective voltage zero-crossing point; otherwise, the median of the historical zero-crossing timestamp sequence is used to correct the timestamp of the current candidate point.

9. The method for automated anchor point finding and processing of photovoltaic inverter grid-connected test data according to claim 8, characterized in that, The standard power frequency cycle is dynamically generated using the following method: Obtain the actual periodic value sequence of multiple consecutive power frequency cycles before the candidate point of voltage zero crossing, and perform a moving average processing on the obtained actual periodic value sequence, with the moving window width being half the number of cycles. The median of the sequence after moving average processing is used as the dynamic reference period. When calculating the absolute deviation value sequence, the dynamic reference period is used instead of the fixed standard power frequency period. If the maximum value of the absolute deviation value sequence is less than the set phase tolerance threshold, the median of the historical zero-crossing timestamp sequence is still used to correct the current candidate point timestamp. If the deviation between the dynamic reference period and the rated power frequency period exceeds the set frequency tolerance threshold, a frequency abnormality alarm signal will be sent to the test platform.

10. The method for automated anchor point finding and processing of photovoltaic inverter grid-connected test data according to claim 1, characterized in that, The specific methods for outputting the timestamps and signal characteristic values ​​of the final valid anchor points corresponding to all anchor point types include: Generate a visualization chart containing timestamp-aligned denoised voltage signal waveforms, power command signal waveforms, and final effective anchor point location markers; Data labels are associated with the final valid anchor point location marker. The data labels include the anchor point type name, the precise timestamp, and the root mean square voltage value at the corresponding time. A manual position correction interface is set up. This interface responds to the user's drag operation on the final effective anchor point position mark in the visualization chart, obtains the new timestamp coordinates after dragging, and calculates the difference between the new timestamp coordinates and the original timestamp as the manual offset. When the absolute value of the manual offset does not exceed the predetermined tolerance window, the corrected anchor point data is generated based on the new timestamp coordinates and stored independently. The signal feature values ​​of the corrected anchor point data are updated through interpolation calculation.

Images