Distribution network fault intelligent diagnosis method based on multi-source heterogeneous data fusion

Abstract

The invention discloses an intelligent distribution network fault diagnosis method based on multi-source heterogeneous data fusion, which relates to the technical field of distribution network fault diagnosis, and comprises the following steps of: establishing a double-section buffer structure comprising a forward buffer area and a backward buffer area, acquiring electric quantity of a line in real time, and circularly writing the electric quantity into the forward buffer area; performing real-time analysis on the data in the forward buffer area, and detecting a suspicious symptom of a transient disturbance event; when a suspicious symptom is detected, freezing forward buffer data, and judging whether an event starting condition is met or not based on the data; if the event starting condition is met, starting a backward buffer area to record waves to obtain transient buffer data; performing time alignment and correction on the frozen forward buffer data and the transient buffer data to form a complete transient waveform; according to the method, the problems that the intermittent flash-over fault process cannot be completely recorded and the precursor is difficult to capture are solved by establishing a double-section transient buffer, micro-amplitude fluctuation detection and forward and backward data splicing correction method.

Description

Technical Field

[0001] This invention relates to the field of distribution network fault diagnosis technology, and more specifically, to a method for intelligent fault diagnosis of distribution networks based on multi-source heterogeneous data fusion. Background Technology

[0002] Single-phase grounding faults frequently occur in power distribution networks during operation. Existing power distribution networks typically use equipment such as feeder protection devices, line monitoring devices, fault indicators, and waveform recorders to monitor line voltage, current, and zero-sequence quantities, and trigger waveform recording or protection actions when the detected quantities reach set thresholds.

[0003] Under complex environmental conditions, some grounding faults exhibit "intermittent flashover and reconnection" characteristics, meaning that the grounding state at the fault point repeatedly appears and disappears within a very short period of time. This phenomenon is often caused by factors such as tree branches swaying and contacting the conductor, partial discharge due to insulation dampness, and instantaneous span changes caused by wind vibration. Its short duration, small fluctuation amplitude, and irregular intervals make the fault process exhibit obvious fragmented characteristics. Existing monitoring technologies have significant shortcomings in handling such intermittent faults: firstly, waveform recorders generally use a single-trigger mode, recording only limited data before and after the trigger, making it difficult to cover multiple flashover processes; secondly, devices are typically designed for continuous faults and cannot identify small, recurring abnormal fluctuations within a short period, resulting in the incomplete recording of the actual process of fault formation, intermittent faulting, and re-breakdown.

[0004] Key transient information leading to intermittent grounding faults is easily overlooked, and fault precursors cannot be captured, thus affecting engineers' judgment of the fault nature, development trend, and equipment health status, and also causing difficulties in subsequent fault location and troubleshooting. Especially in scenarios with latent faults such as tree branch contact and partial insulation deterioration of cables, the lack of complete fault waveforms may lead to misjudgment, cascading tripping, or the fault remaining undetected for a long time. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for intelligent diagnosis of distribution network faults based on multi-source heterogeneous data fusion. By establishing a dual-segment transient buffer, micro-amplitude fluctuation detection, and forward and backward data splicing and correction method, the method solves the problems in the prior art where the intermittent flashover and flashover fault process cannot be fully recorded and the precursors are difficult to capture.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This application provides a method for intelligent fault diagnosis of distribution networks based on multi-source heterogeneous data fusion. The method includes: establishing a dual-segment buffer structure containing a forward buffer and a backward buffer; collecting line electrical quantities in real time and writing them cyclically into the forward buffer; performing real-time analysis on the data in the forward buffer to detect suspicious signs of transient disturbance events; freezing the forward buffer data when a suspicious sign is detected, and determining whether the event initiation conditions are met based on the data; if the event initiation conditions are met, starting the backward buffer to record waveforms to obtain transient buffer data; and aligning and correcting the frozen forward buffer data with the transient buffer data in time to form a complete transient waveform.

[0007] In one embodiment, real-time acquisition of line electrical quantities and cyclic writing to a forward buffer includes: acquiring instantaneous sampled values ​​of the power line to form an original sampling sequence; calculating the numerical changes of adjacent sampling points in the original sampling sequence; identifying slowly changing data segments and transient abrupt data points based on the changes; compressing the identified slowly changing data segments and writing them to the forward buffer; retaining the corresponding original sampled values ​​of the identified transient abrupt data points and writing them to the forward buffer; and organizing the data written to the forward buffer in chronological order to form traceable transient cache data.

[0008] In one embodiment, the identified slowly varying data segments are compressed and written to the forward buffer, including: establishing a compression buffer unit for the sampled data identified as slowly varying; accumulating and storing the slowly varying data determined in real time into the compression buffer unit during continuous sampling; monitoring the amount of data accumulated in the compression buffer unit, and generating a compressed representation value based on the accumulated data when the batch threshold for adaptive updating is reached; and writing the compressed representation value into the forward buffer in chronological order.

[0009] In one embodiment, real-time analysis of the data in the forward buffer to detect suspicious signs of transient disturbance events includes: decompressing the compressed representation values ​​read from the forward buffer to reconstruct a reconstructed signal containing time-series information; performing wavelet transform on the reconstructed signal to decompose it into sub-band signals of different frequency scales; extracting time-domain features representing micro-amplitude fluctuations from each sub-band signal within a preset time window and combining them to form a multi-dimensional feature vector; comparing the multi-dimensional feature vector with preset judgment conditions, and if the multi-dimensional feature vector meets the preset judgment conditions, then determining that there are suspicious signs representing transient events in the current time window.

[0010] In one embodiment, the reconstructed signal is decomposed into sub-band signals of different frequency scales by wavelet transform, including: dividing the reconstructed signal into several consecutive time periods and establishing an incremental processing unit for each time period; calculating the information entropy of each candidate wavelet basis on the signal within the current time period; selecting the wavelet basis with the smallest information entropy from the candidate wavelet basis set as the optimal wavelet basis for the current time period; and using an incremental wavelet decomposition algorithm, initializing the signal using the boundary state of the previous time period stored in the incremental processing unit, and decomposing the signal of the current time period using the optimal wavelet basis to generate sub-band signals of multiple scales.

[0011] In one embodiment, when a suspicious symptom is detected, the forward buffer data is frozen, and the event triggering condition is determined based on the data. This includes: freezing the forward buffer after detecting a suspicious symptom; extracting key feature points characterizing signal waveform changes from the frozen buffer data; constructing a current waveform structure feature template based on the time distribution and amplitude relationship of the key feature points, wherein the current waveform structure feature template includes extreme point spacing pattern, energy-dominant subband switching order, and envelope morphology abrupt change direction; comparing the current waveform structure feature template with a pre-stored normal operating waveform structure template to identify structural difference features; and jointly setting the structural difference features as the flashover event determination condition.

[0012] In one embodiment, if the event triggering condition is met, the back buffer is started to record waveforms to obtain transient buffer data, including: continuously monitoring and judging the real-time sampled values ​​based on the flashover event judgment condition; when the sampled value meets the judgment condition, a trigger signal is generated and the waveform recording function of the back buffer is started; after starting, the trigger time and subsequent sampled values ​​are written into the back buffer in sequence to form transient buffer data; the sampling sequence is continuously monitored, and when the transient disturbance process is judged to be over, the writing of data to the back buffer is stopped.

[0013] In one embodiment, the frozen forward buffer data and transient buffer data are time-aligned and corrected, including: correcting the time reference deviation between the forward buffer data and the backward waveform data; identifying and processing the relative positional relationship between the forward buffer data and the backward waveform data on the time axis; and based on the processing result, splicing the two data segments in time order and performing overall correction to form a continuous and complete transient waveform.

[0014] In one embodiment, the relative positional relationship between the forward buffer data and the backward waveform recording data on the time axis is identified and processed, including: the relative positional relationship includes time overlap, time continuity and time gap; if there is time overlap, the overlapping part is fused according to the sampling rate of the two data segments; if there is time gap, the type is determined according to the gap length and corresponding data interpolation or marking processing is performed.

[0015] In one embodiment, the correction step includes: aligning the waveform data of the frozen forward and backward buffers on the time axis to form a sequence to be spliced ​​containing splicing points; extracting the time-domain features of the waveforms in the neighborhood of the splicing points and calculating the time-domain consistency deviation; extracting the frequency-domain features of the waveforms in the short-time windows before and after the splicing points and calculating the frequency-domain consistency deviation; constructing a dual-domain consistency judgment condition based on the time-domain consistency deviation and the frequency-domain consistency deviation; if the dual-domain consistency judgment condition is not met, correcting the waveforms in the neighborhood of the splicing points; and outputting the corrected sequence as the final spliced ​​transient waveform.

[0016] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: By constructing a dual-segment buffer structure consisting of a forward buffer and a backward buffer, continuous, traceable, and high-fidelity recording of the entire process of power line flashover and reconnection is achieved under high sampling rate conditions. Its advantage lies in its ability to simultaneously ensure real-time performance and integrity under limited storage resources. Compared with conventional techniques in the field that rely solely on fixed threshold triggering for waveform recording, single buffering, or post-event analysis, this solution innovatively introduces an adaptive forward buffering mechanism of "compressing gradual changes and retaining abrupt changes," a micro-amplitude transient symptom detection method based on incremental wavelet decomposition, and a flashover judgment logic centered on signal structure abrupt change fingerprints (envelope extrema, energy-dominant subband switching, waveform edge abrupt changes), enabling waveform recording triggering to accurately align with the same flashover event. At the same time, through time alignment, overlapping and gap classification processing, and a dual-domain splicing correction strategy combining time and frequency domain consistency, the problems of discontinuity in multi-buffered data splicing and pseudo-abrupt changes are solved, significantly improving the early identification capability of flashover and reconnection faults, waveform integrity, and analysis reliability, demonstrating comprehensive innovation in event precursor capture, data management efficiency, and physical consistency assurance. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the intelligent fault diagnosis method for distribution networks based on multi-source heterogeneous data fusion provided in the embodiments of this application.

[0018] Figure 2 This is a comparison chart of the sub-band root mean square amplitude and the standard root mean square amplitude provided in the embodiments of this application.

[0019] Figure 3 A comparison chart of peak value change and standard peak value change provided for embodiments of this application.

[0020] Figure 4 A comparison chart of the fluctuation frequency and the standard fluctuation frequency provided for the embodiments of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figure 1 The flowchart shown is a schematic diagram of the intelligent fault diagnosis method for distribution networks based on multi-source heterogeneous data fusion provided by the present invention, including the following steps S101 to S105. Wherein: S101, establish a dual-segment buffer structure and acquire line electrical quantities, including instantaneous sampled values ​​of voltage and current, and cyclically write the line electrical quantities into the forward buffer in time order to form traceable transient buffer data. The dual-segment buffer structure includes a forward buffer and a backward buffer.

[0023] The power distribution line monitoring device establishes a dual-segment buffer structure, including a forward buffer and a backward buffer. The forward buffer stores sampled data before triggering, while the backward buffer stores transient fault data after triggering. Both the forward and backward buffers have time-sequential storage capabilities and can be accessed by sampling time index.

[0024] In step S101, the line electrical quantities are cyclically written into the forward buffer in chronological order to form traceable transient buffer data, including steps S201 to S206, wherein: S201, based on a fixed sampling rate, acquires instantaneous sampled values ​​of line voltage and current to form the original sampling sequence; S202, Calculate the numerical changes of adjacent sampling points sequentially based on the original sampling sequence; The specific formula for calculating the change in value between adjacent sampling points is as follows:

[0025] In the formula, This represents the instantaneous sampled value of the line voltage or current at the nth sampling time. For the previous sampling time n 1. Instantaneous sampled value of line voltage or current. This represents the change at the nth sampling time relative to the previous sampling point.

[0026] S203, the change amount is compared with a preset amplitude threshold. If the change amount does not exceed the preset amplitude threshold, the change amount is marked as a slow-changing data segment with low amplitude and stable change. If the change amount exceeds the preset amplitude threshold, the change amount is marked as a transient data point with high amplitude and sudden change. S204: For slowly varying data segments, compress them and then write them to the forward buffer; S205, For data points marked as transient mutations, retain their corresponding original sampled values ​​and write them completely into the forward buffer; S206 organizes the data written to the forward buffer in chronological order to form traceable transient buffer data.

[0027] Furthermore, in step S204, the slowly varying data segment is compressed and then written to the forward buffer, including steps S301 to S305, wherein: S301, for the slowly changing data segment marked as low amplitude stable change, establish a compression buffer unit. The compression buffer unit has cyclic storage and counting functions, and initialize the state of the compression buffer unit, clearing the counter, accumulator and mark pointer to zero. S302, during continuous sampling, the slowly changing data determined in real time is accumulated and stored in the compression buffer unit; S303: Real-time acquisition of the amplitude and frequency of change of slowly varying data, and updating the batch threshold using a preset adaptive formula, where the frequency of change is the number of times the slowly varying data exceeds the amplitude threshold. The adaptive formula is specifically calculated as follows:

[0028] In the formula, To update the batch threshold, Let A be the initial batch threshold and A be the amplitude of the slowly varying data. For the frequency of change, As a monotonically decreasing function, the value of N is reduced when the amplitude or frequency of change increases, making the compression more precise and capturing abnormal changes in a timely manner.

[0029] S304, When the accumulated amount of slowly varying data reaches the update batch threshold, generate a compressed representation value; Among them, by batch accumulation and fitting of stable segments, the segment trend is preserved while reducing redundant storage, and at the same time, the trend of slight fluctuations before the flashover is traceable.

[0030] S305, writes the compressed representation value to the forward buffer in chronological order.

[0031] It should be noted that by establishing a dual-segment buffer structure, the forward buffer is used to store sampled data before the fault is triggered, and the backward buffer is used to store sampled data after the fault is triggered. Combined with differential incremental calculation, gradually changing data compression, complete storage of high-amplitude sudden changes, and dynamic batch threshold adjustment strategies, efficient data management under high sampling rates is achieved. The advantages of this scheme are: it can reduce real-time write pressure and significantly reduce redundant data storage under limited buffer capacity and hardware conditions; it can also retain the precursors and key transient characteristics of flashovers and connections, ensuring that the formed forward transient buffer data is traceable, complete, and continuous, providing a high-fidelity, traceable data foundation, while improving the real-time performance and reliability of the power distribution line monitoring device.

[0032] S102, perform real-time analysis based on the transient cache data, extract micro-amplitude transient fluctuations, and detect suspicious signs of transient disturbance events.

[0033] In this embodiment, step S102 involves real-time analysis of the transient buffer data to extract minute transient fluctuations and detect suspicious signs of transient disturbance events, including steps S401 to S404, wherein: S401, based on the compressed representation value in the transient cache data, decompresses the data, restores the trend of slight changes, and generates a reconstructed signal; S402, Perform wavelet transform on the reconstructed signal to decompose it into sub-band signals of different frequency scales; S403, based on each sub-band signal, within a preset time window, extract time-domain features representing micro-amplitude fluctuations from each sub-band signal and combine them to form a multi-dimensional feature vector to describe the micro-amplitude fluctuation state of the time window. The multi-dimensional feature vector includes root mean square amplitude, peak value change, and fluctuation frequency. S404. Based on the multidimensional feature vector, a judgment is made based on the preset judgment conditions. If the multidimensional feature vector simultaneously meets the preset judgment conditions, it is determined that there are suspicious signs of transient disturbance events in the time window. The preset judgment criteria include the sub-band standard root mean square amplitude, standard peak value change, standard fluctuation frequency, and standard duration, which are obtained through statistical analysis of data under normal conditions. If the multidimensional feature vectors simultaneously satisfy the preset judgment criteria, it can be understood that, for example... Figure 2 As shown, if the root mean square amplitude exceeds the standard root mean square amplitude, then the root mean square amplitude characteristic of this sub-band satisfies the condition; for example... Figure 3 As shown, if the peak value change exceeds the standard peak value change, then the peak value change of this sub-band meets the condition; for example... Figure 4As shown, if the fluctuation frequency exceeds the standard fluctuation frequency, then the fluctuation frequency of the sub-band meets the condition; for each of the above feature vectors, the duration of the threshold is counted within multiple consecutive sampling points or sub-time periods. If the duration exceeds the continuous standard duration, then the feature is considered to be continuously abnormal within the time window and meets the continuity condition.

[0034] Further, in step S402, the reconstructed signal is subjected to wavelet transform to decompose it into sub-band signals of different frequency scales, including S501 to S506, wherein: S501 divides the reconstructed signal into several consecutive time periods according to the time series; S502, For each time period, an incremental processing unit is established. The incremental processing unit saves the start and end times and boundary state information of the signal segment, including the wavelet coefficients or trend parameters of the end of the previous segment. S503, for the reconstructed signal of each time period, calculate the information entropy of each candidate wavelet on the signal of the current time period; The specific formula for calculating the information entropy is as follows:

[0035]

[0036] In the formula, Let M be the information entropy, and M be the total number of wavelet coefficients of the signal in the current time period. The energy percentage of the i-th wavelet coefficient. Let be the i-th wavelet coefficient of the signal, and j be the summation index.

[0037] S504, Based on information entropy, select the wavelet with the smallest entropy from the candidate wavelet basis set as the optimal wavelet basis for the current time period. The candidate wavelet basis set includes db, sym, and coif series wavelets. S505: For the selected optimal wavelet basis, the incremental wavelet decomposition algorithm is used to initialize the signal by using the boundary state of the previous time period stored in the incremental processing unit, and the optimal wavelet basis is used to decompose the signal of the current time period to generate sub-band signals of multiple scales. S506, during the decomposition process, uses the previous boundary state saved by the incremental processing unit as initialization to ensure the continuous connection of wavelet coefficients in each segment, thereby reducing mode aliasing and boundary effects.

[0038] It should be noted that by decomposing the reconstructed signal into several sub-band signals through incremental wavelet decomposition, the micro-amplitude fluctuation features at different frequency scales can be accurately extracted while maintaining signal continuity. In the current technical scenario of detecting flashover precursors in power distribution lines, this method offers the following specific advantages: it can simultaneously capture the subtle trends and high-frequency abrupt changes in slowly varying data, avoiding mode aliasing caused by improper signal segmentation or wavelet basis selection. Furthermore, incremental processing enables continuous connection and high-precision multi-scale analysis, making the identification of micro-amplitude flashover symptoms more accurate, stable, and traceable.

[0039] S103, after confirming suspicious signs, freeze the forward buffer and establish the interruption event judgment conditions based on the frozen transient buffer data.

[0040] In this embodiment, in step S103, after confirming a suspected intermittent interruption symptom, the forward buffer is frozen, and intermittent interruption event determination conditions are established based on the frozen transient buffer data, including S601 to S605, wherein: S601, after confirming a suspicious flashover symptom, immediately freeze the forward buffer. The freezing operation includes: stopping the writing of new sampled data to ensure the integrity and continuity of existing data in the buffer; locking the low-amplitude stable change compressed data and high-amplitude sudden change original sampled values ​​stored in the buffer in chronological order to form a traceable frozen transient buffer data.

[0041] S602, based on the frozen transient buffer data, extract key feature points representing changes in the signal waveform. These key feature points include local extrema of the signal envelope (referring to the local maximum and minimum points in the envelope curve after envelope extraction (e.g., through Hilbert transform, sliding extremum search, or smooth envelope algorithm), dominant trend change points of sub-band energy (referring to the time points when the dominant energy position changes in the energy trend of each sub-band (different scales, different frequencies) over time after wavelet decomposition), and abrupt shape edge points in the time series (referring to the locations where the signal waveform shows steep changes on the time axis, including spikes, rapid rising edges, rapid falling edges, and abrupt changes in inflection points, etc.), serving as the basic fingerprint for subsequent structural feature analysis. Among them, local extreme points reflect the fluctuation structure of the macroscopic energy of the entire signal over time. When the system shows signs of a flashover, a slight disturbance will cause the envelope curve to produce a sudden and dense high and low extreme points. The role of the dominant trend change point of sub-band energy is that flashover events are usually accompanied by instantaneous energy redistribution, which is manifested as a sudden increase in high-frequency energy or a sudden decrease in low-frequency energy. This jump can accurately reflect the short-term instability of the system and is a more intuitive structural indicator than conventional thresholds. Before a flashover, there are usually weak but obvious transient disturbances. The most typical feature of these disturbances is the sudden change of the waveform edge.

[0042] S603, the extracted key feature points are arranged in chronological order, and a current waveform structure feature template is constructed based on their temporal distribution and amplitude relationship. The current waveform structure feature template includes: extreme point spacing pattern, energy-dominant subband switching order, and envelope morphology abrupt change direction. Among them, the current waveform structure feature template reflects the signal morphological changes before and after freezing, and can capture the structural disturbances that are typical precursors to flashover. The extreme point spacing pattern refers to the interval distribution law of adjacent local maxima and minima in the signal envelope on the time axis, which is used to characterize the changes in signal oscillation frequency and structural fragmentation degree; the energy-dominant subband switching order refers to the order in which the energy dominance of different subbands shifts over time in the multi-scale or multi-frequency band decomposition results, which is used to reflect the dynamic migration process of the signal energy center of gravity; the direction of envelope morphological change refers to whether the trend of the signal envelope is upward enhancement or downward decay when it shows a significant steep rise or collapse in a short period of time, which is used to characterize the direction and nature of transient disturbances.

[0043] S604, based on the comparison between the current waveform structure feature template and the pre-stored normal operating waveform structure template, identify structural difference features, specifically: An abnormal increase in the density of extreme points exceeding the preset safety threshold (reflecting a sudden "fragmentation" of the signal structure); a sudden jump in the energy-dominant subband (reflecting a sudden shift of the energy focus from low frequency to high frequency or a reverse migration); a short-term collapse or sharp rise in the envelope shape (reflecting transient disturbances); the occurrence of any one of these structural abrupt changes constitutes the triggering basis for potential flashover precursors.

[0044] S605, the structural difference features are jointly set as the criteria for determining the flashover event. The criteria for determining the flashover event are that there is one of the following three conditions: a density jump at an extreme point, a switching of the energy dominant subband, or a sudden change in the envelope morphology. The flashover event is repeatable or has short-term persistence during the frozen time period.

[0045] It should be noted that the "structural mutation fingerprint," composed of three key structural points—signal envelope extrema, energy-dominant subband jumps, and waveform edge mutations—can capture flashover precursors from three independent but complementary dimensions: signal morphology, energy distribution, and transient edges. This makes the judgment conditions robust, noise-resistant, and highly sensitive. Furthermore, high-confidence early warnings can be achieved by detecting only any one type of structural mutation phenomenon, thereby significantly improving the timeliness and reliability of flashover event identification and ensuring the early detection capability of fast transient faults such as "flashover-flashover" in power systems.

[0046] S104: Based on the fault interruption event determination condition, the backward buffer is activated and formal waveform recording begins, causing the backward buffer to record transient changes during the fault duration, thus obtaining backward transient buffer data. Since the fault interruption event determination condition originates from the frozen transient buffer data, this waveform recording is precisely aligned with the same fault interruption and reconnection process.

[0047] In step S104, the backward buffer is activated according to the flashover event determination condition, and formal waveform recording begins. This causes the backward buffer to record transient changes during the fault duration, resulting in backward transient buffer data, including S701 to S708, wherein: S701 monitors the instantaneous sampled values ​​of line voltage and current point by point based on the intermittent interruption event judgment conditions; The criteria for determining the flashover event include at least one structural difference feature, namely, extreme point density jump, energy dominant subband switching, or envelope morphology abrupt change.

[0048] S702, Based on the monitoring results, determine whether the real-time sampled value meets the conditions for determining the interruption event; S703, if not satisfied, continue to keep the back buffer in standby state; If S704 is satisfied, a formal waveform recording trigger signal is generated. S705, based on the formal recording trigger signal, starts the recording function of the backward buffer; The startup operation includes initializing the write pointer, time index, and segment identifier of the backward buffer, making the backward buffer writable.

[0049] S706: After the backward buffer enters the writable state, the instantaneous sampled values ​​at the trigger time and thereafter are written into the backward buffer in the order of sampling time. S707 updates the time index range of the backward buffer in real time based on the continuously written instantaneous sampled values, forming backward transient buffer data that records transient changes during the fault. The transient changes include voltage drops, current surges, transient imbalances between phases, and phase angle shifts.

[0050] S708, when the fault duration ends, stop the write operation to the buffer according to the judgment condition that the flashover event disappears in the sampling sequence.

[0051] S105 aligns and corrects the frozen transient buffer data with the transient buffer data in time to form a complete transient waveform covering the fault precursor, the moment of flashover to the continuous flashover process, thereby capturing the entire process of flashover and flashover faults.

[0052] In step S105, the frozen transient buffer data and transient buffer data are time-aligned and corrected to form a complete transient waveform covering the fault precursor, the moment of flashover, and the continuous flashover process, including S801 to S808, wherein: S801, obtain the timestamps of the frozen transient buffer data and the backward transient buffer data, and calculate the time deviation; The specific formula for calculating the time deviation is as follows:

[0053] In the formula, Due to time deviation, The termination time for frozen transient cache data. This is the start time of the backward transient buffer data.

[0054] S802, if the time deviation exceeds the preset tolerance, time calibration is performed according to the device synchronization record or Network Time Protocol (NTP); S803, based on the time calibration results, combined with the termination time of the frozen transient buffer data and the start time of the backward transient buffer data, detects and classifies the time position relationship, which includes time overlap, time continuity and time gap. Specifically, if the end time of the frozen transient buffer data is greater than the start time of the backward transient buffer data, it is determined to be time overlap; if the end time of the frozen transient buffer data is equal to the start time of the backward transient buffer data, it is determined to be time continuity; if the end time of the frozen transient buffer data is less than the start time of the backward transient buffer data, it is determined to be time gap.

[0055] S804, based on time overlap, obtains the sampling rate of the forward and backward data segments. When the sampling rates of two data segments in the overlapping area are different and the difference exceeds a set threshold, the data segment with the largest sampling rate is selected as the main one. If the difference between the sampling rates of the two data segments does not exceed the set threshold, the overlapping area is weighted and fused according to the sampling rate ratio to generate the final sample points. The fusion method and weight are recorded in the metadata, thereby realizing the continuity of the waveform in the overlapping area, while ensuring the priority of sampling accuracy and the coherence of the time series. The specific formula for calculating the weight is as follows:

[0056]

[0057] In the formula, The weights of the frozen transient cached data, The weights for the backward transient buffer data, The sampling rate of the forward frozen data segment. This is the sampling rate for the backward data segment.

[0058] S805, based on the time gap, obtain the gap length, and determine the gap type according to the gap length, wherein the gap type includes short gaps and long gaps; The length of the gap The specific calculation formula is as follows:

[0059] S806. Gap segment processing is performed according to the gap type. For gaps with insufficient length (when the gap length is ≤ the preset gap threshold), linear / spline interpolation is used to fill the gap, and the gap is marked as "estimated filling segment" in the merged result. For long gaps (when the gap length is > the preset gap threshold), time gap areas are retained in the splicing sequence and placeholder metadata is added to mark the start and end of the gap for subsequent manual review or supplementary recording.

[0060] S807, based on the processing results of time overlap and time gap, merges the forward-frozen transient buffer data and the backward transient buffer data into a complete transient waveform in chronological order and performs splicing correction.

[0061] The complete transient waveform and its metadata are archived and indexed in the form of event packets. The event packets include the original segments of forward-frozen transient buffer data, backward transient buffer data, merging results, time correction records, and overlap / gap processing records. The event packets are uploaded to the power distribution master station or operation and maintenance platform for fault location, analysis and archiving.

[0062] Furthermore, in step S807, splicing correction is performed, including S901 to S908, wherein: S901, obtain the frozen forward transient buffer data and backward transient buffer data, arrange the two waveforms in chronological order, and form a candidate splicing sequence at the splicing point; S902: Obtain temporal features from adjacent sampling points before and after the splicing point, and calculate the temporal consistency deviation by weighted summation based on the temporal features. The time-domain consistency deviation is used to characterize the degree of abrupt change in the time domain at the splicing point. The time-domain features include the difference in voltage or current values ​​at the sampling points, the difference in the first derivative of the difference in ... S903, select multiple short-time windows before and after the splicing point, perform short-time Fourier transform on the sequence within the window, and extract frequency domain features, including the dominant frequency component, frequency band energy distribution, and phase spectrum of the window; S904, calculate the frequency domain consistency deviation based on the energy proportion and phase continuity of each frequency band in the frequency domain characteristics. The frequency domain consistency deviation describes the degree of abrupt change at the splicing point in the frequency domain. The specific formula for calculating the frequency domain consistency deviation is as follows:

[0063] In the formula, For frequency domain consistency deviation, The total number of frequency components in the analysis window can be determined by the number of decomposition layers in the STFT (Short Time Fourier Transform). This represents the energy percentage of the frequency component f in the window before the splicing point. This represents the energy percentage of the frequency component f in the window after the splicing point. This represents the phase mean of the frequency component f in the window before the splicing point. This represents the phase mean of the frequency component f in the window after the splicing point. , These are the weighting coefficients.

[0064] S905, construct dual-domain consistency conditions based on time-domain consistency deviation and frequency-domain consistency deviation, and then make a judgment; Among them, if Preset deviation threshold and If the preset deviation threshold is met, the splicing point is deemed to have good continuity and requires no correction; otherwise... >Preset deviation threshold and If the preset deviation threshold is set, it is determined to be a time-domain abrupt change, requiring amplitude or derivative correction; if Preset deviation threshold and If the preset deviation threshold is met, it is determined to be a frequency domain abrupt change, requiring frequency band energy or phase correction; if >Preset deviation threshold and If the preset deviation threshold is not met, it is considered a true transient spike and no correction is performed.

[0065] S906, based on the judgment result, corrects the time domain abrupt change and frequency domain abrupt change. For the time domain abrupt change, perform amplitude offset compensation processing on the splicing point to make the splicing point smooth in the time domain. For the frequency domain abrupt change, use the spectrum consistency filtering method to make the spectrum of the splicing point consistent with the energy trend of the adjacent window. For example, when the splicing point exhibits a sudden change in the time domain but a gradual change in the frequency domain, it indicates that the two transient waveforms are continuous in the overall spectral structure, but there is a discontinuity in the amplitude reference or instantaneous rate of change. In this case, amplitude offset compensation or first-order derivative continuity processing is performed on the splicing point: Amplitude offset compensation is achieved by calculating the average amplitude difference of several sampling points before and after the splicing point, and shifting the later or earlier waveform by the corresponding offset amount to make the signal amplitude at the splicing point continuous.

[0066] When the frequency domain consistency deviation exceeds the limit at the splicing point while the time domain change is gradual, it indicates that the signals before and after splicing are continuous in terms of instantaneous amplitude, but there are abnormal jumps in energy distribution or phase structure. In this case, spectral consistency filtering or frequency band energy redistribution method is used for correction: spectral consistency filtering constrains the spectrum near the splicing point, aligning its frequency band energy ratio and phase change trend with the statistical characteristics of the front and rear windows, thereby suppressing unreal spectral abrupt changes.

[0067] S907 outputs the corrected sequence as the final transient waveform, achieving a smooth transition between the frozen forward transient buffer data and the backward transient buffer data.

[0068] It should be noted that by uniformly processing the frozen forward transient buffer data and backward transient buffer data under various conditions such as time deviation, overlap, and gaps, and combining time-domain and frequency-domain dual-domain consistency correction, not only can complete transient waveform splicing of the precursor, the instantaneous flashover, and the continuous flashover process be achieved, but the smoothness and continuity of the waveform in amplitude, derivative, and spectrum are also guaranteed. At the same time, complete metadata is recorded, which facilitates fault analysis, location, and archiving, and improves the accuracy and reliability of capturing flashover and flashover events in the distribution network.

[0069] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0070] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0071] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0072] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0073] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0074] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent fault diagnosis of distribution networks based on multi-source heterogeneous data fusion, characterized in that, Includes the following steps: Establish a dual-segment buffer structure containing a forward buffer and a backward buffer, collect line electrical quantities in real time, and write them to the forward buffer in a loop. Real-time analysis of data in the forward buffer is performed to detect suspicious signs of transient disturbance events; When a suspicious sign is detected, the forward buffer data is frozen, and the event triggering conditions are determined based on the data. If the event triggering conditions are met, waveform recording is performed in the buffer after the event is triggered to obtain transient buffer data. The frozen forward buffer data and the transient buffer data are time-aligned and corrected to form a complete transient waveform.

2. The intelligent fault diagnosis method for distribution networks based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The real-time acquisition of line electrical quantities and their cyclic writing to the forward buffer includes: Acquire instantaneous sample values ​​of power lines to form the original sampling sequence; Calculate the numerical change between adjacent sampling points in the original sampling sequence; Based on the amount of change, slowly changing data segments and transiently abrupt data points are identified; The identified slowly varying data segments are compressed and written to the forward buffer. The identified transient abrupt change data points retain their corresponding original sample values ​​and are written to the forward buffer. The data written to the forward buffer is organized in chronological order to form a traceable transient buffer.

3. The intelligent fault diagnosis method for distribution networks based on multi-source heterogeneous data fusion according to claim 2, characterized in that, The step of compressing the identified slowly varying data segments and writing them into the forward buffer includes: A compression buffer unit is established for the sampled data that is identified as slowly varying. During continuous sampling, the slowly varying data determined in real time are accumulated and stored in the compression buffer unit; Monitor the amount of data accumulated in the compression buffer unit, and when the batch threshold for adaptive updates is reached, generate a compressed representation value based on the accumulated data; The compressed representation values ​​are written to the forward buffer in chronological order.

4. The intelligent fault diagnosis method for distribution networks based on multi-source heterogeneous data fusion according to claim 3, characterized in that, The real-time analysis of data in the forward buffer to detect suspicious signs of transient disturbance events includes: The compressed representation values ​​read from the forward buffer are decompressed to recover and generate a reconstructed signal containing time series information; The reconstructed signal is decomposed into sub-band signals of different frequency scales by wavelet transform. Within a preset time window, time-domain features representing minute fluctuations are extracted from each sub-band signal and combined to form a multi-dimensional feature vector. The multidimensional feature vector is compared with the preset judgment conditions. If the multidimensional feature vector meets the preset judgment conditions, the current time window memory is judged to be a suspicious sign of transient events.

5. The intelligent fault diagnosis method for distribution networks based on multi-source heterogeneous data fusion according to claim 4, characterized in that, The step of performing wavelet transform on the reconstructed signal to decompose it into sub-band signals of different frequency scales includes: The reconstructed signal is divided into several consecutive time periods, and an incremental processing unit is established for each time period. For the signal within the current time period, calculate the information entropy of each candidate wavelet basis on the signal; From the set of candidate wavelet bases, select the one with the minimum information entropy as the optimal wavelet base for the current time period; An incremental wavelet decomposition algorithm is adopted. The boundary state of the previous time period stored in the incremental processing unit is used for initialization, and the signal of the current time period is decomposed using the optimal wavelet basis to generate sub-band signals of multiple scales.

6. The intelligent fault diagnosis method for distribution networks based on multi-source heterogeneous data fusion according to claim 1, characterized in that, When a suspicious symptom is detected, the forward buffer data is frozen, and the event triggering conditions are determined based on this data, including: Freeze the forward buffer upon detecting suspicious signs; Extract key feature points characterizing signal waveform changes from frozen cache data; Based on the temporal distribution and amplitude relationship of key feature points, a current waveform structure feature template is constructed. The current waveform structure feature template includes the extreme point spacing pattern, the energy-dominant subband switching order, and the envelope morphology abrupt change direction. The structural differences are identified by comparing the current waveform structure feature template with the pre-stored normal operating waveform structure template. Structural difference features are combined and set as the criteria for determining the interruption event.

7. The intelligent fault diagnosis method for distribution networks based on multi-source heterogeneous data fusion according to claim 1, characterized in that, If the event triggering conditions are met, then after triggering, waveform recording is performed in the buffer to obtain transient buffer data, including: Based on the criteria for determining intermittent events, real-time sampled values ​​are continuously monitored and judged. When the sampled value meets the judgment condition, a trigger signal is generated and the waveform recording function of the backward buffer is started; After startup, the trigger time and subsequent sampled values ​​are written sequentially into the backward buffer to form transient buffer data; The sampling sequence is continuously monitored, and when the transient disturbance process is determined to be over, the writing of data to the buffer is stopped.

8. The intelligent fault diagnosis method for distribution networks based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The step of time-aligning and correcting the frozen forward buffer data and the transient buffer data includes: Correct the time reference deviation between the forward buffer data and the backward waveform recording data; Identify and process the relative positions of the forward buffer data and the backward waveform data on the time axis; Based on the processing results, the two data segments are spliced ​​together in chronological order and then subjected to overall correction to form a continuous and complete transient waveform.

9. The intelligent fault diagnosis method for distribution networks based on multi-source heterogeneous data fusion according to claim 8, characterized in that, The process of identifying and processing the relative positions of the forward buffer data and the backward waveform recording data on the time axis includes: The relative positional relationships include time overlap, time continuity, and time gaps; If there is time overlap, the overlapping part is fused according to the sampling rate of the two data segments; If a time gap exists, its type is determined based on the gap length, and corresponding data interpolation or marking processing is performed.

10. The intelligent fault diagnosis method for distribution networks based on multi-source heterogeneous data fusion according to claim 8, characterized in that, The correction steps include: Align the waveform data of the frozen forward and backward buffers on the time axis to form a sequence to be spliced, which includes splicing points; Extract the temporal features of the waveforms in the neighborhood of the splicing point and calculate the temporal consistency deviation; Frequency domain features of waveforms within short time windows before and after the splicing point are extracted, and frequency domain consistency deviation is calculated. Based on time-domain consistency deviation and frequency-domain consistency deviation, a dual-domain consistency judgment condition is constructed; If the dual-domain consistency judgment condition is not met, the waveform of the splicing point neighborhood is corrected. The corrected sequence is output as the final spliced ​​transient waveform.

Images