Voltage transformer real-time monitoring system based on broadband and edge calculation

Abstract

The invention relates to the technical field of voltage transformers, in particular to a voltage transformer real-time monitoring system based on broadband and edge calculation, which comprises the steps of collecting multiple paths of voltage transformer data, extracting peak values and zero cross points, comparing jump rising edge time differences, unifying time sequences, extracting extreme values and slope identification disturbance sections, and monitoring the voltage transformers in real time. Measuring amplitude continuous change to judge the offset trend, analyzing trend direction classification waveform fragments, and combining the three indexes to generate an operation state label. According to the method, a unified time sequence is constructed by comparing cross-channel jump rising edge time differences, multi-channel jump process alignment is achieved, the synchronous anomaly recognition capability is improved, the number of extreme values and the slope direction are extracted to recognize a disturbance dense section, continuous anomaly is judged by combining statistics of overrun, and the fluctuation trend is judged by tracking the deviation trend between extreme values. And based on the three-index combination classification operation state, the anomaly identification precision is improved.

Description

Technical Field

[0001] This invention relates to the field of voltage transformer technology, and in particular to a real-time monitoring system for voltage transformers based on wideband and edge computing. Background Technology

[0002] Voltage transformers encompass voltage measurement, monitoring, and protection technologies in power systems. A voltage transformer is a device used to provide voltage signals in a power system, primarily measuring high voltages in power lines and converting them into signals suitable for low-voltage electrical equipment. Voltage transformer technology involves multiple aspects, including equipment design, structural optimization, accuracy improvement, and application methods. With the increasing demands for automation and intelligence in power systems, real-time monitoring technology for voltage transformers has gradually developed, particularly by combining broadband signal processing and edge computing technologies to improve the response speed and reliability of power monitoring systems and meet the requirements for higher precision and faster data processing.

[0003] The voltage transformer real-time monitoring system based on broadband and edge computing refers to a voltage transformer monitoring system that integrates broadband signal processing technology and edge computing technology. Addressing the real-time monitoring needs of voltage transformers, it employs broadband technology to achieve high-precision acquisition of voltage signals, and combines this with edge computing technology for local data processing and analysis, thereby improving the monitoring response speed and real-time performance. Through the combination of these technologies, efficient monitoring and management of the voltage transformer status is achieved, avoiding the latency problems of traditional centralized data processing and ensuring the accuracy and stability of voltage signals in the power system.

[0004] In existing technologies, voltage transformer sampling data does not uniformly process the timing relationship of transition responses between multiple channels, which can easily cause feature misalignment between channels and affect the accuracy of subsequent waveform analysis. Disturbance identification is based solely on local feature extraction within a period, lacking analysis of disturbance trend aggregation behavior, making it difficult to identify dense disturbance sections. The processing of amplitude changes mainly relies on the amplitude itself, lacking structural judgment of continuous offset trends, affecting the complete extraction of trend sections. The single-dimensional indicator in the operational status classification leads to ambiguous status judgments, resulting in misjudgments and response lags. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a real-time monitoring system for voltage transformers based on wideband and edge computing.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a real-time monitoring system for voltage transformers based on broadband and edge computing, the system comprising, The jump synchronization module collects multi-channel voltage transformer and transient signal data, compares time differences of each channel jump edge, performs delay insertion, complement and rearrangement of the waveforms, and generates a set of time-aligned waveforms; The disturbance marking module extracts extreme points and slope change characteristics from the aligned waveform set, identifies unidirectional dense distribution super-period disturbance segments, marks them as abnormal sections, and forms a disturbance section sequence; The fluctuation extraction module measures amplitude changes based on the disturbance section sequence, calculates the direction and amplitude of adjacent extreme points, and if the direction is consistent and the amplitude continuously exceeds the reference interval, it is marked as a traceable deviation, and a continuous deviation information group is generated; The trend classification module determines the trend consistency according to the continuous deviation information group, classifies segments with consistent direction and similar change speed as rising, falling or mixed trend, and forms a trend classification data set.

[0007] The state judgment module extracts disturbance frequency, deviation density and duration from the trend classification data set, comprehensively judges the voltage transformer running state, classifies it as stable, fluctuation or abnormal, and generates a state classification result.

[0008] As a further scheme of the application, the time-aligned waveform set includes jump delay difference characteristics, time rearrangement structure, synchronization correction identifier, and channel alignment marker. The jump disturbance section sequence specifically includes disturbance dense segment number, abnormal distribution pattern, statistical mutation characteristics, and period comparison result. The continuous deviation section information group includes deviation direction attribute, continuous deviation amplitude, section fluctuation characteristics, and reference interval deviation amount. The trend classification structure data set includes trend type identifier, change direction classification, duration parameter, and speed consistency index. The voltage transformer running state classification result includes state judgment label, disturbance statistical value, fluctuation density, and deviation time span.

[0009] As a further scheme of the application, the jump synchronization module includes a jump detection submodule, a delay calculation submodule, and a time rearrangement submodule: The jump detection submodule collects multi-channel voltage transformer sampling data and corresponding transient jump signals, extracts waveform segments before and after each channel jump based on the peak points and zero-crossing points of the power frequency main waveform in each channel signal, selects rising edge morphology mutation time points, and generates a channel jump candidate time point sequence; The delay calculation submodule calls the channel jump candidate time point sequence, compares the rising edge time points of the same jump event in each channel, judges according to the time point difference and the set jump response delay reference value, selects the channel combination with early or late relationship, and obtains a cross-channel jump delay difference list; The time rearrangement submodule inserts a fixed amplitude delay mark to the channel waveform advanced at the jump time point according to the cross-channel jump delay difference list, inserts a fixed amplitude delay mark to the channel waveform delayed, performs data segment supplementing according to the adjacent data change trend of the delayed channel waveform, and rearranges all channels according to a unified time axis to generate a time sequence alignment waveform set.

[0010] As a further scheme of the present application, the disturbance marking module comprises a disturbance extraction submodule, a threshold judgment submodule, and a sequence construction submodule: The disturbance extraction submodule extracts the extreme value point number of each channel waveform in a continuous sampling period and the slope direction change of adjacent data points based on the time sequence alignment waveform set, constructs an extreme value count sequence and a slope direction change sequence arranged according to the sampling period, and generates a periodic disturbance morphology parameter set; The threshold judgment submodule calls the periodic disturbance morphology parameter set, takes the number of disturbances in a plurality of continuous periods as a reference, accumulates the extreme value number and the total number of direction changes in each period, calculates a disturbance density value, compares the density value with a disturbance number statistical threshold, and obtains a disturbance density over-limit interval group; The sequence construction submodule detects the continuity of the slope direction change trend in the interval according to the disturbance density over-limit interval group, counts the time span and the number of disturbances in the same direction disturbance interval, divides each continuous disturbance interval according to the time sequence number, and generates a jump disturbance interval sequence.

[0011] As a further scheme of the present application, the fluctuation extraction module comprises an amplitude calculation submodule, a trend discrimination submodule, and an interval screening submodule: The amplitude calculation submodule continuously processes each piece of data in the jump disturbance interval sequence, collects the amplitude change amount and the sampling time interval of adjacent two extreme value points in each paragraph, calculates the amplitude change rate between each two extreme value points and collects the amplitude change rate into a sequence to generate a continuous amplitude change rate sequence; The trend discrimination submodule calls the continuous amplitude change rate sequence, judges the change rate direction in each paragraph, counts whether the adjacent change rate signs are consistent, and takes the continuous three groups of positive or negative changes as the consistent reference of the shift direction to identify the change segments meeting the conditions and obtain a unidirectional shift trend interval group; The interval screening submodule calculates the difference between the maximum value and the minimum value in each change interval according to the unidirectional shift trend interval group, judges whether the difference continuously exceeds the upper limit of the set reference amplitude interval, integrates the fragments meeting the conditions according to the channel number and the time sequence to generate a continuous shift interval information group.

[0012] As a further scheme of the present application, the trend classification module comprises a trend judgment submodule, a speed evaluation submodule, and a type division submodule: The trend determination submodule screens out continuous segments with consistent offset directions according to the offset direction and duration of each segment in the set of continuous offset segment information, detects whether the duration of the continuous segment is greater than a set duration threshold, filters out segments that do not meet the condition, and generates a same-direction trend segment set; The speed evaluation submodule calls the same-direction trend segment set, calculates the amplitude change amount per unit time in each offset segment as a change speed value, calculates the change speed difference between adjacent segments, marks the segments with a difference value lower than a change speed judgment threshold, and obtains a trend speed consistency segment set; The type division submodule judges the change direction of the segment to be a positive value, a negative value or an alternating change direction according to the trend speed consistency segment set, respectively marks the segment as an upward type, a downward type or a mixed type, arranges and numbers each trend type segment in chronological order, and generates a trend classification structure data set.

[0013] As a further scheme of the application, the state determination module includes an index extraction submodule, a segment comparison submodule and a state recognition submodule. The index extraction submodule calls the upward type, downward type and mixed type trend segments in the trend classification structure data set, collects the disturbance frequency, amplitude change amount of each sampling point and offset duration in each trend segment, respectively counts the numerical range corresponding to each type of index, and generates a trend segment fluctuation index set. The segment comparison submodule performs weighted calculation on the disturbance frequency and amplitude change density in each trend segment according to the trend segment fluctuation index set, obtains a fluctuation density value, extracts the corresponding offset duration, forms a combined feature sequence with the three indexes, and obtains a state feature combination parameter group. The state recognition submodule performs interval judgment on each group of data with respect to the state feature combination parameter group, a set stable state threshold, a fluctuation state threshold and an abnormal state threshold, classifies the waveform segment as stable, fluctuation or abnormal according to the index falling into the range, adds a corresponding label, and generates a voltage transformer operation state classification result.

[0014] Compared with the prior art, the application has the advantages and positive effects that: In the application, by comparing the time difference of the jump rising edge of the cross-channel waveform, a unified time sequence is constructed, the alignment of multi-channel data in the jump response process can be realized, and the identification ability of the synchronization anomaly is enhanced. Based on the extreme value quantity and the slope direction, the disturbance distribution characteristics are extracted, the disturbance dense trend is judged combined with the statistical overrun, and the continuous abnormal segment is effectively located. The trend direction is judged according to the direction consistency and the amplitude continuity, and the stable classification of the fluctuation segment is supported. Through the combination of the disturbance frequency, the amplitude fluctuation density and the offset time length, the clear classification of the running state is realized, and the identification accuracy of the abnormal evolution is improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The method flowchart of the application is shown in the figure. Figure 2 The acquisition flowchart of the jump synchronization module of the application is shown in the figure. Figure 3 The acquisition flowchart of the disturbance marking module of the application is shown in the figure. Figure 4 The acquisition flowchart of the fluctuation extraction module of the application is shown in the figure. Figure 5 The acquisition flowchart of the trend classification module of the application is shown in the figure. Figure 6 The acquisition flowchart of the state judgment module of the application is shown in the figure. DETAILED DESCRIPTION

[0016] The application will be further described in detail in combination with the drawings. These drawings are all simplified schematic diagrams, and only show the basic structure of the application in a schematic manner, and thus only show the structure related to the application.

[0017] Reference Figures 1 to 6 A voltage transformer real-time monitoring system based on wide frequency band and edge calculation, comprising: A jump synchronization module collects multi-channel voltage transformer sampling data and transient jump signals, extracts the peak points and zero-crossing points of each channel in the power frequency main waveform, compares the time difference of the rising edge of the jump waveform in each channel, judges whether there is a jump delay difference across channels, if there is a difference, inserts a delay identifier at the corresponding time point for the advanced waveform, performs data segmenting and uniformly rearranges the time sequence for the lagging waveform, and generates a time sequence alignment waveform set; A disturbance marking module extracts the extreme value point quantity and the slope change direction in the continuous sampling period based on the time sequence alignment waveform set, identifies the segment in which the waveform disturbance frequency exceeds the single period statistical value in a limited time period, judges whether the disturbance in these segments presents a one-way dense distribution trend, if the dense disturbance continuous condition is met, records the segment as an abnormal marking segment, and numbers it in time sequence, and generates a jump disturbance segment sequence; The wave fluctuation extraction module measures the continuity of the voltage amplitude variation of each paragraph in the jump disturbance section sequence, calculates the change direction and change amplitude between the two consecutive extreme points in each paragraph, judges whether there is a trend of continuous upward or downward deviation of the amplitude, and if the deviation direction is consistent and the fluctuation range continuously exceeds the reference interval, the paragraph is marked as a traceable deviation behavior section, and a continuous deviation section information group is generated; The trend classification module judges whether the change of each waveform section maintains the same direction trend according to the deviation direction and duration in the continuous deviation section information group. If the deviation direction is consistent and the change speed difference is less than the set judgment range in the continuous section, the section is classified as a linear trend section. Further, all trend sections are classified and sorted, marked as rising type, falling type or mixed type trend, and a trend classification structure data set is generated. The state judgment module calls the waveform paragraphs involved in each type of trend in the trend classification structure data set, calculates the corresponding disturbance times, amplitude fluctuation density and offset continuous time span, judges whether the current voltage transformer is in a stable state, a fluctuation state or an abnormal state according to the combination of the three fluctuation indicators, and generates corresponding running state classification labels to generate a voltage transformer running state classification result.

[0018] The time alignment waveform set includes jump delay difference characteristics, time rearrangement structure, synchronization correction identifier and channel alignment marker. The jump disturbance section sequence specifically includes disturbance intensive section number, abnormal distribution mode, statistical mutation characteristics and period comparison result. The continuous deviation section information group includes deviation direction attribute, continuous deviation amplitude, section fluctuation characteristics and reference interval deviation amount. The trend classification structure data set includes trend type identifier, change direction classification, duration parameter and speed consistency index. The voltage transformer running state classification result includes state judgment label, disturbance statistical value, fluctuation density and offset time span.

[0019] As shown in Figure 2 , the jump synchronization module includes a jump detection submodule, a delay calculation submodule and a time rearrangement submodule: The jump detection submodule collects multi-channel voltage transformer sampling data and corresponding transient jump signals, extracts waveform segments before and after each channel jump as jump candidate intervals based on the peak points and zero-crossing points of the power frequency main waveform in each channel signal, selects rising edge mutation time points, and generates a channel jump candidate time point sequence. The multi-channel voltage transformer sampling data and corresponding transient jump signals collect data sequences of three-channel voltage transformers at a sampling rate of kHz, and read the data sequences at The transient jump signal emitted by the protection device at milliseconds (ms) is used as the center. Waveform segments near the peak point or zero-crossing point of the main waveform of each channel signal are extracted as candidate jump intervals. Then, the time points where the rising edge shape changes abruptly are screened out. Specifically, the instantaneous rate of change sequence is obtained by calculating the first-order forward difference value of the waveform data in the candidate interval of each channel, and a sudden change detection threshold is set. This threshold is based on the average instantaneous rate of change within the stable power frequency cycle before the jump. with standard deviation ,pass Calculated, for example, channel A's for V / sample, for V / sample, then its threshold for V / sample iterates through the instantaneous rate of change sequence, identifying the first time point exceeding the threshold as a candidate time point, such as channel A in... The difference in ms is V / sample, greater than V / sample, then ms is the candidate time point for channel A. Similarly, the candidate time point for channel B is obtained as follows. ms, channel C is ms, ultimately generating a candidate time point sequence for channel transitions.

[0020] The delay calculation submodule calls the candidate time point sequence of channel transitions, compares the rising edge time points of the same transition event in each channel pairwise, judges based on the difference between time points and the set transition response delay benchmark value, filters out channel combinations with an early or late relationship, and obtains a list of cross-channel transition delay differences. Channel transition candidate time point sequence, calling the candidate time point sequence generated by the previous submodule { ms, ms, The time points of each channel are compared pairwise, and their absolute differences are calculated to obtain the difference between channels A and B. ms, the difference between A and C is ms, the difference between B and C is ms, depending on the device Allowable sampling rate at kHz The error at each sampling point is used to calculate and set the jump response delay baseline value. for The time difference between each channel combination is compared with this baseline value, and channels with a difference greater than ms are selected. The channel combination of ms, i.e. the combination of channel A and B, the combination of channel B and C, is determined to have an advance or lag relationship, and finally, the specific difference of the channel combination having the delay relationship is obtained, and the channel A is calculated to lead the channel B ms, the channel C leads the channel B ms, thereby obtaining a cross-channel jump delay difference list.

[0021] The time rearrangement submodule inserts a fixed amplitude delay identifier into the channel waveform of the jump time point in advance according to the cross-channel jump delay difference list, performs data segment filling on the lag channel waveform according to the adjacent data change trend, and reorders all channels according to a unified time axis to generate a timing alignment waveform set.

[0022] The cross-channel jump delay difference list is used to select the time point of the channel B having the latest jump in all channels ms as a global alignment reference, the channel waveform of the jump time point is processed, the channel A is advanced by ms (corresponding to sampling points), the channel C is advanced by ms (corresponding to sampling points), and therefore, a delay identifier of fixed amplitude is inserted after the ms time point of the original waveform of the channel A, a delay identifier of fixed amplitude is inserted after the ms time point of the original waveform of the channel C, and for the lag channel, interpolation segment filling is performed according to the change trend of adjacent data, since the channel B is the reference in this example, no segment filling operation is required, after the above processing, the waveform data of all channels is reordered according to a unified time axis, so that the data sequence lengths of all channels are equal, and a timing alignment waveform set is generated.

[0023] Further, wavelet transform or empirical mode decomposition and other multi-scale signal processing methods can be used for jump detection, and dynamic time warping is combined to match the cross-channel waveform shape, and an adaptive threshold mechanism is introduced. The adaptability of jump detection to noise and complex working conditions is enhanced, and false judgment is avoided.

[0024] The cross-channel jump synchronization with higher precision and robustness is realized, and the reliability of subsequent analysis is improved.

[0025] As shown in Figure 3 , the disturbance marking module includes a disturbance extraction submodule, a threshold judgment submodule, and a sequence construction submodule: The disturbance extraction submodule extracts the number of extreme points of each channel waveform in a continuous sampling period and the slope direction change of adjacent data points based on the time-aligned waveform set, constructs an extreme value count sequence and a slope direction change sequence arranged by sampling period, and generates a periodic disturbance morphology parameter set; The time-aligned waveform set is based on the time-aligned waveforms generated in the previous step Divide the data into one power frequency period, and then extract the number of extreme points of each channel waveform in each period and the number of slope direction changes of adjacent data points. Specifically, the extreme points are counted by judging the amplitude relationship of the sampling points and their adjacent points, and the number of direction changes is counted by calculating the sign change of adjacent slopes. When the system is running stably, the number of extreme points in one power frequency period is , and the number of slope direction changes is In the 6th period of the jump, the number of extreme points of channel B increases to , and the number of slope direction changes increases to In the 7th period, the two values are and respectively. The two parameters calculated in all periods are constructed into an extreme value count sequence and a slope direction change sequence arranged by period, and a periodic disturbance morphology parameter set is generated.

[0026] The threshold judgment submodule calls the periodic disturbance morphology parameter set, takes the number of disturbances in consecutive multiple periods as a reference, accumulates the number of extreme values and the total number of direction changes in each period, calculates the disturbance density value, and compares the density value with the disturbance number statistical threshold to obtain the disturbance density overrun interval group. The periodic disturbance morphology parameter set calls the parameter set generated in the previous step and sets the disturbance number statistical threshold This threshold is based on the analysis of a large number of stable period data under normal working conditions, and is set to the upper limit of the total number of stable period disturbances (the sum of the number of extreme points and the number of direction changes) , which is set to Subsequently, the number of extreme values and the total number of direction changes in each period are accumulated to calculate the disturbance density value of the period. For example, the disturbance density value of the stable period is , the density value of the 6th period of the jump is , and the density value of the 7th period is Compare the density values calculated in each period with the threshold , and select the periods with density values greater than , i.e. the 6th and 7th periods. Combine these two periods to obtain the disturbance density overrun interval group.

[0027] The sequence construction submodule detects the continuity of the slope direction change trend in the section according to the disturbance density over-limit interval group, counts the time span and the number of disturbances in the same disturbance interval, divides each continuous disturbance interval in time sequence number, and generates a jump disturbance section sequence.

[0028] The disturbance density over-limit interval group, according to the obtained over-limit interval group {[6th period, 7th period]}, detects the continuity of the slope direction change trend of the waveform in the section. Since the 6th period and the 7th period are continuous in time, they are divided into a continuous disturbance interval. Then, the time span and the total number of disturbances in the same disturbance interval are counted. The interval contains two power frequency periods, the time span is ms, and the total number of disturbances is the sum of the number of slope direction changes in the 6th period and the 7th period, that is, times, finally, the continuous disturbance interval is numbered in time sequence, and it is marked as disturbance section 1, its attributes are start time ms, end time ms, span ms, number of disturbances times, and a jump disturbance section sequence is generated.

[0029] Further, short-time Fourier transform or spectral energy analysis method can be used to capture the disturbance characteristics, and Bayesian discriminant model is introduced for probabilistic judgment, and disturbance knowledge graph based on historical data is constructed. Reduce the misjudgment caused by single threshold, and enhance the recognition ability of different disturbance types. Realize more intelligent and more adaptive labeling of disturbance events, support long-term evolution of online learning optimization.

[0030] As shown in Figure 4 , the fluctuation extraction module includes an amplitude calculation submodule, a trend discrimination submodule, and an interval screening submodule: The amplitude calculation submodule continuously processes each segment of data in the jump disturbance section sequence, collects the amplitude change and sampling time interval of adjacent two extreme points in each paragraph, calculates the amplitude change rate between each two extreme points and summarizes it into a sequence to generate a continuous amplitude change rate sequence. The jump disturbance section sequence processes the data of the disturbance section 1 (starting at ms) determined in the previous step, first collects all adjacent extreme points of the channel B waveform in the section, for example, the first extreme point P1 ( ms, V) and the second extreme point P2 ( ms, V), the amplitude change between the two points is calculated as V, the time interval is ms, and the amplitude change rate is calculated as V / ms, this method is used to sequentially process all adjacent extreme point pairs such as P2 and P3, P3 and P4 within this segment, and calculate a series of amplitude change rate values, for example V / ms, V / ms, V / ms, etc., are summarized in chronological order to generate a continuous amplitude change rate sequence.

[0031] The trend judgment submodule calls the continuous amplitude change rate sequence to judge the direction of the change rate in each segment, counts whether the positive and negative signs of adjacent change rates are consistent, and uses three consecutive sets of positive or negative changes as the benchmark for consistent offset direction to identify the change segments that meet the conditions, thus obtaining the unidirectional offset trend segment group. A continuous rate of change sequence of amplitude, which calls the rate of change sequence of amplitude generated by the previous submodule, i.e. [ , , , , , , V / ms, for each value in the sequence, the direction, i.e., the sign, is judged to obtain the sign sequence [negative, positive, negative, positive, negative, positive, negative]. A benchmark with consistent offset direction is set, that is, three consecutive sets of positive or negative changes. Then, the sign sequence is traversed to determine whether there are three consecutive signs that are all positive or all negative. After checking the entire sign sequence, no segment satisfying three consecutive sets of changes in the same direction is found. Therefore, no change segment that meets the condition is identified, resulting in an empty unidirectional offset trend segment group.

[0032] The interval filtering submodule calculates the difference between the maximum and minimum amplitude values ​​in each interval based on the unidirectional offset trend segment group. It then determines whether the difference continuously exceeds the upper limit of the set reference amplitude interval and integrates the segments that meet the conditions according to the channel number and time order to generate a continuous offset segment information group.

[0033] For the unidirectional offset trend segment group, since the segment group obtained in the previous steps is empty, this step does not perform calculations. For complete explanation, it is assumed that a segment group starting from [a certain point] was obtained in another example. ms, stop at If the unidirectional offset trend segment is defined in milliseconds, the waveform data within that segment is processed to find the maximum amplitude. V and minimum value V, calculate the difference between the two. V, setting an upper limit for a reference amplitude range. for V, this value is based on the withstand capability of this type of voltage transformer. The difference is determined by the impulse test data at twice the rated voltage, and the calculated difference is used. V and Comparison, V greater than V. The segment is determined to meet the conditions. Finally, the channel number, start and end time and amplitude difference of the segment are integrated to generate a continuous offset segment information group.

[0034] Furthermore, nonlinear dynamic indicators can be used to measure wave complexity, combined with multi-channel principal component analysis to extract common-mode perturbation features, and convolutional neural networks can be used to automatically identify wave patterns. This comprehensively reveals wave characteristics and overcomes the limitations of linear analysis methods. It can more accurately distinguish complex wave patterns and improve the early identification capability of potential anomalies.

[0035] like Figure 5 As shown, the trend classification module includes a trend determination submodule, a speed assessment submodule, and a type classification submodule: The trend determination submodule filters out continuous segments with the same offset direction based on the offset direction and duration of each segment in the continuous offset segment information group, and checks whether the duration of the continuous segments is greater than the set duration threshold. It then filters out segments that do not meet the conditions and generates a set of segments with the same trend. The continuous offset segment information group is the result of filtering waveform disturbances from multiple channels in the previous steps. It contains multiple segments with their own offset direction and duration. The processing unit first filters out segments with the same offset direction and are temporally continuous for merging. For example, two adjacent segments in channel A, both with positive offsets, are merged into one segment with a duration of [duration missing]. A continuous segment of milliseconds is defined, and then the duration of this continuous segment is checked to see if it exceeds a set duration threshold. This threshold is based on the definition in the power grid regulations that a voltage sag event must last for three power frequency cycles. Hz system set to ms, because ms greater than The duration is ms, therefore this segment is retained, while the other segments have insufficient duration after merging. The segments with a value of ms are then filtered out, and finally all segments that meet the conditions are aggregated to generate a set of segments with the same trend.

[0036] The velocity assessment submodule calls the set of trend segments in the same direction, calculates the change in amplitude per unit time within each offset segment as the velocity value, calculates the difference in velocity between adjacent segments, marks segments whose difference is lower than the velocity judgment threshold, and obtains the set of trend velocity consistent segments. Call the set of segments with the same trend, and filter out those starting from ms and lasts a same-direction trend section of channel A of the ms, a change rate value of the section is calculated, specifically, a starting amplitude value of the section is extracted from the original waveform data V and a terminal amplitude value V, a change amount of the amplitude is calculated as V, a change rate value per unit time is V / ms, and a section immediately adjacent to the section with the change rate value of V / ms is processed, a change rate difference value between the two adjacent sections is calculated as V / ms, and the difference value is compared with a change rate judgment threshold The threshold is set to V / ms, since the calculated difference value V / ms is lower than the threshold, the processing unit labels both the two sections, and obtains a trend speed consistency section set.

[0037] A type division sub-module, according to the trend speed consistency section set, judges a change direction of the section as a positive value, a negative value or a change direction alternation, and respectively labels as an upward type, a downward type or a mixed type, arranges and numbers each trend type section in chronological order, and generates a trend classification structure data set.

[0038] According to the labeled section set obtained from the trend speed consistency section set, a whole change direction of the section set is judged, since the offset directions of all sections in the section set are positive, the section set is labeled as an upward type as a whole, if the offset directions of the sections are negative, the section set is labeled as a downward type, if the directions are alternated, the section set is labeled as a mixed type, then, the upward type section set is arranged and numbered, a starting time of a first section contained in the section set is taken as a total starting time, a terminal time of a last section is taken as a total terminal time, and the whole section set is numbered as trend 1 and is labeled with an “upward type” label, if there are other section sets meeting the conditions subsequently, the section sets are numbered as trend 2, trend 3 and are labeled with corresponding type labels, finally, all these structured information is summarized, and a trend classification structure data set is generated. ms as a total starting time, and a terminal time of a last section as a total terminal time, and the whole section set is numbered as trend 1 and is labeled with an “upward type” label, if there are other section sets meeting the conditions subsequently, the section sets are numbered as trend 2, trend 3 and are labeled with corresponding type labels, finally, all these structured information is summarized, and a trend classification structure data set is generated.

[0039] Further, a clustering algorithm can be used to automatically divide the trend patterns, and morphological labels such as “step type” and “oscillation type” are introduced, and an online learning model is used to correct the classification boundary. The automation and flexibility of the trend classification are improved, and the limitation of fixed rules is eliminated. More detailed and adaptive trend classification structures are obtained, and clearer explanations are provided for the abnormal analysis.

[0040] For example Figure 6 ​As shown, the state determination module includes an indicator extraction submodule, a section comparison submodule, and a state recognition submodule: The indicator extraction submodule calls the rising, falling, and mixed trend sections in the trend classification structure dataset, collects the disturbance frequency, amplitude change quantity of each sampling point, and offset duration in each trend section, respectively counts the numerical range corresponding to each type of indicator, and generates a trend section fluctuation indicator set; The trend classification structure dataset calls the dataset containing the rising, falling, and mixed trend sections generated in the previous step, collects the disturbance frequency, amplitude change quantity of each sampling point, and offset duration in each trend section. Specifically, for the rising trend section numbered trend 1 (channel A, start ms, end ms), the processing unit traces back the original record and counts a total disturbance frequency of times, an amplitude change quantity of times, and an offset duration of ms in this time period. Using the same method, the disturbance frequency of the falling trend section numbered trend 2 (channel B, start ms, end ms) is times, the amplitude change quantity is times, the offset duration is ms, and the disturbance frequency of the mixed trend section numbered trend 3 (channel C, start ms, end ms) is times, the amplitude change quantity is times, and the offset duration is ms. All collected and counted indicator values are summarized to generate a trend section fluctuation indicator set.

[0041] The section comparison submodule performs weighted calculation on the disturbance frequency and amplitude change intensity in each trend section according to the trend section fluctuation indicator set to obtain a fluctuation intensity value, extracts the corresponding offset duration length, and forms a combined feature sequence with the three indicators to obtain a state feature combination parameter group. The trend section fluctuation indicator set performs weighted calculation on the disturbance frequency and amplitude change intensity in each trend section according to the indicators generated in the previous step. This calculation calls two preset weight coefficients, and which are determined based on the contribution degree determined by historical fault data analysis. Taking trend 1 as an example, the disturbance frequency intensity is times / ms, the amplitude change intensity is times / ms, and the fluctuation intensity value is and extract the corresponding offset duration length ms, the combination feature is ( , ) , the same calculation is performed on trend 2, the fluctuation intensity value is , the duration is ms, the combination feature is ( , ) , the calculation is performed on trend 3, the fluctuation intensity value is , the duration is ms, the combination feature is ( , ) , all the calculated combination features are summarized to obtain the state feature combination parameter group.

[0042] The state recognition submodule is used for interval judgment of each group of data and the set stable state threshold, fluctuation state threshold and abnormal state threshold according to the state feature combination parameter group, and the waveform segment is classified as stable, fluctuation or abnormal according to the index falling range, and the corresponding label is added, and the voltage transformer running state classification result is generated.

[0043] The state feature combination parameter group is used for interval judgment of each group of data and the set state threshold according to the parameter group obtained in the foregoing step, that is, trend 1 ( , ) , trend 2 ( , ) , and trend 3 ( , ) , the stable state threshold is set to be less than and the duration is less than ms, the abnormal state threshold is set to be greater than or the duration is greater than ms, and the rest is the fluctuation state, and the specific judgment process is that, for the parameters of trend 1 ( , ) , the fluctuation intensity value is greater than , and the duration ms is greater than ms, the abnormal state condition is met, so the waveform segment is classified as abnormal, for the parameters of trend 2 ( , ) , the indexes all fall into the fluctuation state interval, so the classification is fluctuation, for the parameters of trend 3 ( , ) , the same also falls into the fluctuation state interval, and the classification is fluctuation, finally, the classification result is associated with the corresponding trend information and the label is added, and the voltage transformer running state classification result is generated.

[0044] Further, a multi-index fusion health index method can be used to comprehensively integrate the disturbance density, the duration and the trend speed into a continuous state score, and a machine learning classifier is combined to improve the accuracy of the determination, and a state confidence is output. Avoid ambiguous judgments caused by hard thresholds and improve diagnostic transparency. A more reliable and interpretable operating state recognition is realized to provide a diagnostic reference with confidence for operation and maintenance personnel.

[0045] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A wideband and edge computing based real-time monitoring system for voltage transformers, characterized by, The system comprises: The jump synchronization module collects multi-channel voltage transformer and transient signal data, compares the time difference of each channel jump edge, performs delay insertion and completion rearrangement of the waveform, and generates a set of time-aligned waveforms; The disturbance marking module is used to extract extreme points and slope change characteristics from the aligned waveform set, identify unidirectional dense distribution super-period disturbance segments, and mark them as abnormal sections to form a disturbance section sequence; The fluctuation extraction module is used to measure the amplitude change based on the disturbance section sequence, calculate the direction and amplitude of adjacent extreme points, and if the direction is consistent and the amplitude continuously exceeds the reference interval, it is marked as a traceable deviation, and a continuous deviation information group is generated; The trend classification module is used to determine the trend consistency according to the continuous deviation information group, classify the segments with consistent direction and similar change speed into rising, falling or mixed trend, and form a trend classification data set; The state judgment module is used to extract disturbance frequency, deviation density and continuous time length based on the trend classification data set, comprehensively judge the voltage transformer running state, classify it as stable, fluctuation or abnormal, and generate a state classification result.

2. The wideband and edge computing based voltage transformer real-time monitoring system according to claim 1, characterized in that: The time-aligned waveform set includes jump delay difference characteristics, time rearrangement structure, synchronization correction identifier, and channel alignment marker. The jump disturbance section sequence specifically includes disturbance dense segment number, abnormal distribution mode, statistical mutation characteristics, and period comparison result. The continuous deviation section information group includes deviation direction attribute, continuous deviation amplitude, section fluctuation characteristics, and reference interval deviation amount. The trend classification structure data set includes trend type identifier, change direction classification, continuous time length parameter, and speed consistency index. The voltage transformer running state classification result includes state judgment label, disturbance statistical value, fluctuation density, and deviation time span.

3. The wideband and edge computing based voltage transformer real-time monitoring system according to claim 1, wherein, The jump synchronization module comprises: Collecting multi-channel voltage transformer sampling data and corresponding transient jump signals, extracting the waveform segments before and after each channel jump based on the peak points and zero-crossing points of the power frequency main waveform in each channel signal as jump candidate intervals, selecting the rising edge mutation time points, and generating a channel jump candidate time point sequence; Calling the channel jump candidate time point sequence, comparing the rising edge time points of the same jump event in each channel, judging according to the time point difference and the set jump response delay reference value, selecting the channel combinations with advance or lag relationship, and obtaining a cross-channel jump delay difference list; According to the cross-channel jump delay difference list, inserting a fixed amplitude delay identifier into the waveform of the channel with advanced jump time point, supplementing the data segment of the lag channel waveform according to its adjacent data change trend, and reordering all channels according to the unified time axis to generate a time-aligned waveform set.

4. The wideband and edge computing based voltage transformer real-time monitoring system of claim 1, wherein, The disturbance marking module comprises: Based on the time-aligned waveform set, the extreme point number and the slope direction change of adjacent data points of each channel waveform in the continuous sampling period are extracted, the extreme value count sequence and the slope direction change sequence arranged according to the sampling period are constructed, and the period disturbance form parameter set is generated. Call the periodic disturbance form parameter set, with reference to the number of disturbances in a plurality of consecutive periods, accumulate the number of extreme values and the total number of direction changes in each period, calculate the disturbance density value, and compare the density value with the disturbance number statistical threshold to obtain the disturbance density over-limit interval group; According to the disturbance density over-limit interval group, the continuity of the slope direction change trend in the section is detected, the time span and the number of disturbances in the same direction disturbance interval are counted, and each continuous disturbance interval is divided according to the time sequence number to generate a jump disturbance section sequence.

5. The wideband and edge computing based real-time monitoring system for voltage transformers as claimed in claim 1, wherein, The fluctuation extraction module comprises: According to the continuous amplitude change rate sequence generated by the continuous amplitude change rate sequence in each section of data, the direction of the change rate in each paragraph is judged, whether the signs of adjacent change rates are consistent is counted, and the change section meeting the condition is marked as the offset direction consistent reference with continuous three groups of positive or negative changes, to obtain the unidirectional offset trend section group; According to the unidirectional offset trend section group, the difference between the maximum value and the minimum value in each change interval is calculated, it is judged whether the difference continuously exceeds the upper limit of the set reference amplitude interval, and the segments meeting the condition are integrated according to the channel number and time sequence to generate a continuous offset section information group. The trend classification module comprises:

6. The wideband and edge computing based voltage transformer real-time monitoring system according to claim 1, wherein, According to the offset direction and the duration of each section in the continuous offset section information group, the continuous sections with consistent offset direction are screened out, and it is detected whether the duration of the continuous section is greater than the set duration threshold, and the segments not meeting the condition are screened out to generate a same direction trend section set; Call the same direction trend section set, calculate the amplitude change amount per unit time in each offset section as the change speed value, calculate the change speed difference between adjacent sections, and mark the sections with a difference value lower than the change speed judgment threshold to obtain a trend speed consistency section set; According to the trend speed consistency section set, it is judged whether the section change direction is positive, negative or the change direction is alternating, and is respectively marked as the rising type, the falling type or the mixed type, each trend type section is arranged and numbered according to the time sequence, and a trend classification structure data set is generated. The state determination module comprises:

7. The wideband and edge computing based voltage transformer real-time monitoring system according to claim 1, wherein, Call the rising type, falling type and mixed type trend sections in the trend classification structure data set, collect the number of disturbances, the amplitude change amount of each sampling point and the offset duration in each trend section, respectively count the numerical range corresponding to each type of index, and generate a trend section fluctuation index set; According to the trend section fluctuation index set, the number of disturbances and the amplitude change density in each trend are weighted calculated to obtain the fluctuation density value, and the corresponding offset duration length is extracted, and the three indexes are combined to form a combined feature sequence to obtain a state feature combination parameter group; ​ For the state feature combination parameter group, interval judgment is performed on each group of data with the set steady state threshold, fluctuation state threshold and abnormal state threshold, the waveform segment is classified as steady, fluctuation or abnormal according to the index falling into the range, and corresponding labels are attached, to generate the voltage transformer operation state classification result.

Images