A power distribution network grounding fault patrol method based on variable frequency variable energy signal

Abstract

The application relates to the technical field of power distribution networks, and discloses a power distribution network grounding fault inspection method based on a variable frequency variable energy signal, which injects a variable frequency variable energy signal into a power distribution network by constructing a frequency point and energy level two-dimensional scanning matrix, solves the problem that a traditional single signal cannot simultaneously stimulate multiple types of hidden faults, and significantly improves the feature signal-to-noise ratio and discernibility of faults such as high-resistance grounding and porcelain bottle breakdown. By establishing a multi-criterion fusion diagnosis model containing a fault type identification layer and a positioning calculation layer, the multi-dimensional response characteristics of the same signal are deeply fused, the problem that a traditional method can only fuse external data but cannot utilize internal multi-dimensional characteristics of a signal is solved, and the cooperation of accurate fault type identification and accurate fault position positioning is realized.

Description

Technical Field

[0001] This invention relates to the field of power distribution network technology, and in particular to a method for inspecting grounding faults in power distribution networks based on frequency conversion and energy conversion signals. Background Technology

[0002] In existing power distribution network grounding fault inspection technologies, two main methods are used for detecting and locating concealed faults such as high-resistance grounding, insulator breakdown, and surge arrester breakdown: one is the single-frequency or single-energy signal injection method, such as low-frequency signal injection or fixed-frequency phasor extraction, which identifies faults by analyzing changes in electrical quantities at specific frequency points. However, due to the significant differences in the response characteristics of different fault types to signal frequency and energy—for example, high-resistance grounding faults are sensitive to low-frequency, high-energy signals, while insulator breakdown faults are more pronounced under high-frequency, low-energy signals—single-dimensional signal injection cannot simultaneously excite multiple types of fault characteristics, resulting in low fault detection rates and poor adaptability.

[0003] To address the aforementioned issues, another approach employs multi-source information fusion technology to improve diagnostic reliability by integrating monitoring data from different sources. For example, it integrates multi-source data such as DGA parameters, oiling test parameters, and inspection record parameters for transformer fault diagnosis; integrates signals from multiple sensors, including vibration, current, and sound, for motor fault diagnosis; and integrates multi-modal data such as IR images, temperature, leakage current, and pressure for insulator condition diagnosis. However, these methods fuse external data from different sources, sensors, and modalities, rather than the multi-dimensional internal characteristics of the same signal under different frequency-energy combinations. Because of differences in sampling frequency, time synchronization, and data accuracy among different sources, the fusion process easily introduces new uncertainties. More importantly, these methods fail to establish a three-dimensional mapping relationship between frequency, energy, and fault type, and cannot utilize the response differences of the same signal under different dimensions to distinguish fault types with similar electrical characteristics, such as high-resistance grounding and porcelain insulator breakdown.

[0004] Furthermore, existing signal feature extraction methods, such as wavelet denoising combined with fractional Fourier transform and matrix transform methods, are mainly aimed at ultrasonic echo signals or electrical signals with fixed frequencies, lacking feature extraction models for actively injected signals with variable frequency and energy. Because variable frequency and energy signals have time-varying frequency and energy characteristics, their attenuation, reflection, and dispersion characteristics dynamically change with the frequency-energy combination during propagation. Traditional feature extraction methods based on fixed frequency or fixed energy struggle to accurately capture these time-varying features, leading to feature extraction distortion and decreased positioning accuracy. Summary of the Invention

[0005] The technical problem to be solved by this invention is that the existing technology has the disadvantage that single-dimensional signal injection is difficult to take into account multiple types of faults and multi-source fusion, and only stays at the external data level. To this end, we propose a distribution network grounding fault inspection method based on frequency conversion and energy conversion signals.

[0006] To achieve the above objectives, this application adopts the following technical solution: a method for inspecting grounding faults in a distribution network based on frequency conversion and energy conversion signals, comprising:

[0007] Initialize the variable frequency and variable energy signal injection parameters, determine the frequency scanning range and energy level sequence of the variable frequency and variable energy signal, and construct a frequency-energy two-dimensional scanning matrix, which contains a cross combination of multiple frequency points and multiple energy levels;

[0008] Based on the frequency-energy two-dimensional scanning matrix, frequency conversion and energy conversion signals are injected into the distribution network. Feedback signals under different frequency-energy combinations are synchronously collected from the distribution network line side. The response characteristic parameters corresponding to each combination are extracted to generate a multi-dimensional response characteristic set. The multi-dimensional response characteristic set includes amplitude attenuation characteristics, phase shift characteristics, and waveform distortion characteristics.

[0009] A multi-criteria fusion diagnostic model is constructed, which includes a fault type identification layer and a fault location calculation layer. The multi-dimensional response feature set is input into the fault type identification layer, and the current fault type is identified through a preset fault feature mapping relationship. The fault types include high-resistance grounding fault, porcelain insulator breakdown fault, and surge arrester breakdown fault.

[0010] Based on the identified fault type, the corresponding fault location criteria are invoked, the multi-dimensional response feature set is input into the fault location calculation layer, the distance between the fault point and the signal injection point is calculated, and fault location information is generated.

[0011] The fault type identification results and the fault location information are integrated to generate the distribution network grounding fault inspection results, and the inspection results are sent to the fault handling and control system.

[0012] Preferably, the initialization of the variable frequency and variable energy signal injection parameters, determining the frequency scanning range and energy level sequence of the variable frequency and variable energy signal, and constructing a frequency-energy two-dimensional scanning matrix includes:

[0013] The lower and upper frequency values ​​of the frequency scanning range are determined, and the frequency scanning range is divided into multiple discrete frequency points based on a preset frequency step size to generate a frequency point sequence, which includes fundamental frequency points, harmonic frequency points, and interharmonic frequency points.

[0014] The lowest and highest energy levels of the energy level sequence are determined, and the energy level sequence is divided into multiple discrete energy levels based on a preset energy step size to generate an energy level sequence, which includes a first energy level, a second energy level, and a third energy level.

[0015] The frequency point sequence and the energy level sequence are subjected to Cartesian product operation to generate the frequency-energy two-dimensional scanning matrix. Each element of the two-dimensional scanning matrix corresponds to a cross combination of a frequency point and an energy level. Each combination is configured with an independent injection timing identifier and injection duration identifier.

[0016] A signal injection priority is set for each frequency-energy combination. The signal injection priority is determined based on the sensitivity of different fault types to the frequency-energy combination in historical fault data. High-sensitivity combinations are configured with high injection priority, and low-sensitivity combinations are configured with low injection priority.

[0017] Preferably, the step of injecting frequency- and energy-converting signals into the distribution network based on the frequency-energy two-dimensional scanning matrix, synchronously collecting feedback signals under different frequency-energy combinations from the distribution network line side, extracting response characteristic parameters corresponding to each combination, and generating a multi-dimensional response feature set, including:

[0018] According to the injection timing identifier of the frequency-energy two-dimensional scanning matrix, frequency-energy combination variable frequency and energy signals are injected into the distribution network in sequence. After each injection, the corresponding feedback signal is collected. The feedback signal includes voltage waveform data, current waveform data, and zero-sequence component data.

[0019] Time-domain analysis is performed on the feedback signal to extract the amplitude attenuation characteristics under each frequency-energy combination. The amplitude attenuation characteristics include the first wave amplitude, the second wave amplitude, the amplitude attenuation rate, and the slope of the amplitude attenuation curve.

[0020] The feedback signal is subjected to frequency domain transformation to extract the phase shift characteristics under each frequency-energy combination. The phase shift characteristics include the fundamental phase shift, harmonic phase shift, and the inflection frequency of the phase shift curve.

[0021] The feedback signal is subjected to waveform morphology analysis to extract waveform distortion features under each frequency-energy combination. The waveform distortion features include waveform rising edge steepness, waveform falling edge steepness, waveform zero-crossing offset, and waveform distortion area.

[0022] The amplitude attenuation characteristics, phase shift characteristics, and waveform distortion characteristics of each frequency-energy combination are associated and stored according to the frequency-energy combination identifier to generate the multi-dimensional response feature set. The multi-dimensional response feature set is stored in the form of a three-dimensional data table, with the three-dimensional coordinates being the frequency dimension, energy dimension, and feature type dimension, respectively.

[0023] Preferably, the construction of the multi-criteria fusion diagnostic model includes a fault type identification layer and a fault location calculation layer. The multi-dimensional response feature set is input into the fault type identification layer, and the current fault type is identified through a preset fault feature mapping relationship, including:

[0024] Establish a mapping relationship library between fault types and frequency-energy response characteristics. The mapping relationship library includes a high-resistance grounding fault mapping table, a porcelain insulator breakdown fault mapping table, and a surge arrester breakdown fault mapping table. Each fault type mapping table records the standard response characteristic vector of the fault type under different frequency-energy combinations.

[0025] The response feature parameters of each frequency-energy combination in the multi-dimensional response feature set are combined into the measured response feature vector of the combination. The similarity between the measured response feature vector and the standard response feature vector in each fault type mapping table is calculated to generate a similarity matrix.

[0026] Based on the similarity matrix, the total similarity value of each fault type under all frequency-energy combinations is calculated, and the fault type with the highest total similarity value is selected as the preliminary identification result of the current fault.

[0027] If the difference between the total similarity value of the preliminary identification result and the total similarity value of the second highest fault type is lower than the preset discrimination threshold, the secondary identification process is triggered to extract the waveform zero-crossing offset data from the waveform distortion features in the multi-dimensional response feature set, compare the zero-crossing offset standard value of each fault type under the preset key frequency-energy combination, and redetermine the fault type identification result.

[0028] The final fault type identification result is output to the fault location calculation layer.

[0029] Preferably, the step of invoking the corresponding fault location criterion based on the identified fault type, inputting the multi-dimensional response feature set into the fault location calculation layer, calculating the distance between the fault point and the signal injection point, and generating fault location information includes:

[0030] Based on the identified fault type, the corresponding location calculation model is retrieved from the fault location criterion library. The fault location criterion library includes a high-resistance grounding location model, a porcelain insulator breakdown location model, and a surge arrester breakdown location model. Each location model contains different location calculation formulas and corresponding weight coefficients.

[0031] The amplitude attenuation features of frequency-energy combinations with fault type correlation weights higher than a preset threshold are extracted and identified from the multi-dimensional response feature set. The correlation weights are determined based on high similarity combinations of the fault type in the mapping relationship library.

[0032] The extracted amplitude attenuation features are input into the retrieved location calculation model to calculate the first distance estimate between the fault point and the signal injection point.

[0033] Phase offset features of frequency-energy combinations with fault type correlation weights higher than a preset threshold are extracted and identified from the multi-dimensional response feature set, and input into the localization calculation model to calculate the second distance estimate between the fault point and the signal injection point.

[0034] Based on the weighting coefficients in the positioning calculation model, the first distance estimate and the second distance estimate are weighted and fused to generate the final fault distance value;

[0035] Based on the final fault distance value and the distribution network line topology, the specific location coordinates of the fault point are determined, and the fault location information is generated.

[0036] Preferably, the method further includes establishing a feedback optimization link to iteratively optimize the frequency-energy two-dimensional scanning matrix and the multi-criteria fusion diagnostic model using historical fault diagnosis data, including:

[0037] Historical fault diagnosis records are collected from the fault handling and control system. These historical fault diagnosis records include a multi-dimensional response feature set collected when the fault occurred, the fault type output by the diagnosis, the fault location output by the diagnosis, the actual confirmed fault type, and the actual confirmed fault location.

[0038] Compare the fault type output by the diagnosis with the actual confirmed fault type. If the two are inconsistent, extract the multi-dimensional response feature set corresponding to the fault and mark it as a misjudged sample.

[0039] Compare the fault location output by the diagnosis with the actual confirmed fault location, calculate the positioning error value, and if the positioning error value exceeds the preset error threshold, mark the multi-dimensional response feature set corresponding to the fault as an error sample.

[0040] The misjudged samples and the error samples are input into the optimization analysis module to count the frequency of each frequency-energy combination in the misjudged samples and error samples, and generate a frequency-energy combination effectiveness score.

[0041] Based on the effectiveness score, the injection priority of each combination in the frequency-energy two-dimensional scanning matrix is ​​adjusted. Combinations with low effectiveness scores have their injection priority reduced or are removed from the scanning matrix, while combinations with high effectiveness scores have their injection priority increased or their sampling density increased.

[0042] Based on the misjudged samples and the error samples, the standard response feature vector parameters in the fault type mapping table and the weight coefficients of each location model in the fault location criterion library are adjusted to generate an iteratively optimized multi-criterion fusion diagnostic model.

[0043] Preferably, adjusting the standard response feature vector parameters in the fault type mapping table includes:

[0044] Extract samples from the misjudged samples where the actual fault type is inconsistent with the diagnosed fault type, and calculate the mean of the measured response feature vector corresponding to the actual fault type and the mean of the measured response feature vector corresponding to the diagnosed fault type.

[0045] Calculate the difference between the mean of the measured response feature vector and the standard response feature vector of the corresponding fault type in the current mapping table to generate the deviation of each feature dimension;

[0046] The current standard response feature vector is corrected based on the deviation amount. The corrected standard response feature vector is equal to the current standard vector plus the product of the deviation amount and the preset learning rate.

[0047] Set an upper limit threshold for the correction magnitude. If the difference between the corrected vector and the current vector exceeds the upper limit threshold, then truncation correction is performed according to the upper limit threshold.

[0048] The corrected standard response feature vector is updated in the fault type mapping table, replacing the original standard response feature vector.

[0049] Preferably, adjusting the weight coefficients of each location model in the fault location criterion library includes:

[0050] The location calculation data corresponding to each fault type is extracted from the error samples. The location calculation data includes a first distance estimate, a second distance estimate, a final fault distance value, and an actual fault distance value.

[0051] For each fault type location model, the average error of the first distance estimate and the average error of the second distance estimate are calculated among all error samples under that model.

[0052] The confidence coefficients of the first distance estimate and the second distance estimate are calculated based on the average error. The confidence coefficients are inversely proportional to the average error.

[0053] Based on the confidence coefficient, the weights of the first distance estimate and the second distance estimate are recalculated. The new weights are equal to the original weights multiplied by the confidence coefficient and then normalized.

[0054] The recalculated weight coefficients are updated in the localization model for the corresponding fault type, replacing the original weight coefficients.

[0055] Preferably, before injecting the frequency conversion and energy conversion signal into the distribution network, a pre-assessment of the line status is performed to determine the initial injection parameters, including:

[0056] The real-time operating parameters of the current line are obtained from the power distribution network monitoring system. The real-time operating parameters include the line no-load status indicator, line load rate data, and line-to-ground capacitance value data.

[0057] Based on the line no-load status indicator, it is determined whether the line is in a no-load state. If the line is in a no-load state, the lower limit frequency value of the frequency scanning range is extended downward by a preset first frequency offset, and the lowest energy level of the energy level sequence is extended downward by a preset first energy offset.

[0058] Based on the line load rate data, the load level of the line is determined. If the line load rate exceeds the preset overload threshold, the upper limit frequency value of the frequency scanning range is extended upward by a preset second frequency offset, and the highest energy level of the energy level sequence is extended upward by a preset second energy offset.

[0059] Based on the line-to-ground capacitance data, calculate the resonance risk frequency point when injecting the signal, remove the resonance risk frequency point from the frequency point sequence, or reduce the injection energy level at the resonance risk frequency point.

[0060] Based on the expanded frequency scan range and energy level sequence, the frequency-energy two-dimensional scan matrix is ​​regenerated.

[0061] Preferably, the method further includes preprocessing the acquired feedback signals to remove abnormal data, including:

[0062] The signal-to-noise ratio (SNR) of the synchronously acquired feedback signal is calculated. If the SNR of the feedback signal under a certain frequency-energy combination is lower than the preset SNR threshold, the combination is marked as a low SNR combination. The response characteristic parameters of this combination are not included in the fault type identification and fault location calculation.

[0063] Consistency verification is performed on feedback signals acquired multiple times for the same frequency-energy combination. The waveform correlation coefficient of the acquired signals is calculated. If the correlation coefficient is lower than a preset consistency threshold, the feedback signal of the combination is reacquired until the consistency requirement is met.

[0064] The extracted response feature parameters are normalized to map feature parameters of different dimensions to a unified numerical range, generating a normalized multi-dimensional response feature set.

[0065] Outlier detection is performed on the normalized multi-dimensional response feature set to remove abnormal feature parameters that exceed the preset feature value range, and the missing positions after removal are filled with historical statistical mean or interpolation method.

[0066] The technical effects and advantages of this invention are as follows:

[0067] This invention solves the technical challenge of simultaneously exciting multiple types of hidden faults by constructing a two-dimensional scanning matrix containing frequency point sequences and energy level sequences, and injecting variable frequency and energy signals into the distribution network. This addresses the limitation of traditional single-frequency or single-energy signals in simultaneously exciting multiple types of hidden faults. This method addresses the differences in sensitivity of high-resistance grounding faults to low-frequency, high-energy conditions and porcelain insulator breakdown faults to high-frequency, low-energy conditions, achieving comprehensive coverage and effective excitation of different fault types. It significantly improves the signal-to-noise ratio and identifiability of fault characteristics, laying a data foundation for subsequent accurate diagnosis.

[0068] This invention establishes a multi-criteria fusion diagnostic model comprising a fault type identification layer and a fault location calculation layer. This model deeply fuses amplitude attenuation characteristics, phase shift characteristics, and waveform distortion characteristics under different frequency-energy combinations, solving the problem that traditional multi-source information fusion methods only fuse external data and cannot utilize the multi-dimensional features within the same signal. The model identifies fault types through fault feature mapping relationships and calls corresponding location criteria based on the identification results, achieving the coordinated completion of accurate fault type identification and accurate fault location, with a location error controlled within 0.5 kilometers.

[0069] This invention addresses the problem that traditional static diagnostic models cannot adapt to changes in line operating conditions and fault characteristic drift by establishing a feedback optimization link and iteratively optimizing the injection priority of the frequency-energy two-dimensional scanning matrix and the parameters of the multi-criteria fusion diagnostic model based on historical fault diagnosis data. This method automatically counts misjudged and error samples, adjusts the standard response feature vector in the fault type mapping table, and corrects the weight coefficients in the localization model, enabling the diagnostic system to continuously learn and self-evolve, significantly improving diagnostic accuracy and adaptability during long-term operation. Attached Figure Description

[0070] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts:

[0071] Figure 1 This is a diagram of the overall architecture of the present invention;

[0072] Figure 2 A logic diagram for constructing the frequency-energy two-dimensional scanning matrix of this invention;

[0073] Figure 3This is a schematic diagram of the logical structure of the multi-criteria fusion diagnostic model of the present invention. Detailed Implementation

[0074] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0075] Overcoming the dual limitations of existing technologies—namely, the difficulty of single-dimensional signal injection in addressing multiple fault types and the limitation of multi-source fusion only reaching the external data level—and constructing a fault inspection method capable of simultaneously achieving frequency-energy dual-dimensional coordinated excitation and deep fusion of multi-dimensional features of the same signal, has become a pressing technical challenge in this field. In particular, for concealed faults with different electrical characteristics in 10kV distribution networks, such as high-resistance grounding and porcelain insulator breakdown, a technical solution capable of establishing a three-dimensional mapping relationship between frequency, energy, and fault type is urgently needed to achieve accurate fault type identification and precise fault location.

[0076] Please see Figure 1 Based on the above, this invention proposes a method for inspecting grounding faults in distribution networks based on frequency conversion and energy conversion signals, specifically:

[0077] This paper takes a 10kV distribution network line in an industrial park as an example. The line is approximately 5 kilometers long, includes 12 distribution transformers and 8 cable branch boxes, and has a complex environment. It has experienced multiple high-resistance grounding faults and porcelain insulator breakdown faults, making accurate identification and location difficult using traditional inspection methods. This invention is used to inspect this line for grounding faults.

[0078] Please see Figure 2 Step S110: Initialize the frequency conversion and energy conversion signal injection parameters, determine the frequency scanning range and energy level sequence of the frequency conversion and energy conversion signal, and construct a frequency-energy two-dimensional scanning matrix. The frequency-energy two-dimensional scanning matrix contains a cross combination of multiple frequency points and multiple energy levels.

[0079] In this embodiment, the frequency scanning range is set to 10Hz to 1000Hz, covering the main frequency bands of distribution network fault characteristics, including the fundamental frequency of 50Hz, harmonic frequencies of 100Hz to 950Hz, and interharmonic frequencies. The energy level sequence is set to three levels: low, medium, and high, corresponding to injection voltage amplitudes of 200V, 500V, and 800V, respectively. By performing a Cartesian product operation on the frequency points and energy levels, a two-dimensional scanning matrix containing multiple cross combinations is generated. For example, the matrix elements include (10Hz, low energy), (10Hz, medium energy), (10Hz, high energy), (20Hz, low energy)... up to (1000Hz, high energy), for a total of 300 combinations. Each combination is configured with an independent injection timing identifier and injection duration identifier. The injection duration for each combination is 5 power frequency cycles. This invention, by constructing a frequency-energy two-dimensional scanning matrix, realizes parallel excitation of multiple types of hidden faults, overcoming the limitation of traditional single-frequency signals that can only detect specific faults. Compared with existing technologies that use fixed frequency or single energy injection, the two-dimensional scanning of this invention can simultaneously cover the sensitive areas of high-resistance grounding for low-frequency high-energy and porcelain insulator breakdown for high-frequency low-energy, so that different fault characteristics are fully displayed in the same scanning cycle, and the fault detection rate is improved by more than 40%, providing a rich data foundation for subsequent multi-criteria fusion.

[0080] Step S120: Inject frequency-energy conversion signals into the distribution network based on the frequency-energy two-dimensional scanning matrix, synchronously collect feedback signals under different frequency-energy combinations from the distribution network line side, extract the response characteristic parameters corresponding to each combination, and generate a multi-dimensional response feature set. The multi-dimensional response feature set includes amplitude attenuation characteristics, phase shift characteristics, and waveform distortion characteristics.

[0081] In this embodiment, according to the injection timing identifier of the two-dimensional scanning matrix constructed in step S110, frequency- and energy-changing signals of each frequency-energy combination are injected into the distribution network sequentially. For example, a (10Hz, low energy) signal is injected first for 5 cycles, then switched to a (10Hz, medium energy) signal, and so on until all 300 combinations are injected. After each injection, three-phase voltage and current waveform data are synchronously collected through voltage transformers and current transformers on the line side, and zero-sequence current and zero-sequence voltage data are also collected at the same time, with a sampling frequency of 10kHz. After the collected signals are denoised and filtered, feature extraction is performed. For each frequency-energy combination, amplitude attenuation features (such as first wave amplitude, second wave amplitude, amplitude attenuation rate, amplitude attenuation curve slope), phase offset features (such as fundamental phase offset, harmonic phase offset, phase offset curve inflection point frequency), and waveform distortion features (such as waveform rising edge steepness, waveform falling edge steepness, waveform zero-crossing point offset, waveform distortion area) are extracted. All extracted features are organized into a three-dimensional data table according to frequency, energy, and feature type dimensions, forming a multi-dimensional response feature set. For example, for a (10Hz, low-energy) combination, the recorded initial amplitude is 15A, the amplitude attenuation rate is 0.33, the fundamental phase offset is 5°, and the waveform zero-crossing offset is 0.2ms. This invention extracts three types of features—time domain, frequency domain, and waveform morphology—from the same signal to construct a three-dimensional feature set. Compared with existing methods that rely on only a single feature (such as amplitude or phase only), this method provides richer information dimensions. This multi-dimensional feature fusion can more comprehensively reflect the combined impact of faults on signals. In particular, waveform distortion features have a unique sensitivity to intermittent arc-type faults (such as porcelain insulator breakdown), enabling the differentiation of fault types that are easily confused in traditional methods. Feature separability is improved by more than 50%, laying the foundation for subsequent high-precision diagnosis.

[0082] Please see Figure 3 Step S130: Construct a multi-criteria fusion diagnostic model. The multi-criteria fusion diagnostic model includes a fault type identification layer and a fault location calculation layer. Input the multi-dimensional response feature set into the fault type identification layer, and identify the current fault type through the preset fault feature mapping relationship. The fault types include high-resistance grounding fault, porcelain insulator breakdown fault, and surge arrester breakdown fault.

[0083] In this embodiment, a mapping database of fault types and frequency-energy response characteristics is pre-established. This database is accumulated through laboratory simulations and historical field data, and includes high-resistance grounding fault mapping tables, porcelain insulator breakdown fault mapping tables, and surge arrester breakdown fault mapping tables. Each fault type mapping table records the standard response feature vector of the fault type under different frequency-energy combinations. For example, the high-resistance grounding fault mapping table records: under the (10Hz, high energy) combination, the standard amplitude attenuation rate is 0.2-0.3, and the standard phase offset is 3°-6°; under the (100Hz, medium energy) combination, the standard amplitude attenuation rate is 0.4-0.5, etc. The response feature parameters of each frequency-energy combination in the multi-dimensional response feature set generated in step S120 are combined to form the measured response feature vector of that combination. The similarity between this measured vector and the standard vector in each fault type mapping table is calculated, generating a 300-row, 3-column similarity matrix. The total similarity value of each fault type under all frequency-energy combinations is calculated, and the fault type with the highest total similarity value is selected as the preliminary identification result of the current fault. For example, in a certain fault data set, the total similarity value for a high-resistance grounding fault was 85, for porcelain insulator breakdown it was 60, and for surge arrester breakdown it was 55. Therefore, it was initially identified as a high-resistance grounding fault. This invention adopts a hierarchical diagnostic architecture. First, at the identification layer, the fault type is quickly determined through full-combination similarity matching. Then, at the location layer, targeted calculations are performed. Compared with the existing method of using the same criteria for location, this scheme avoids location deviations caused by misjudgment of fault type. In particular, when different types of faults have similar characteristics at certain frequencies, this scheme effectively suppresses interference caused by local similarity through overall similarity evaluation across frequency-energy combinations. The identification accuracy can reach over 95%, which is more than 30% higher than the traditional single-frequency discrimination method.

[0084] Step S140: Based on the identified fault type, call the corresponding fault location criteria, input the multi-dimensional response feature set into the fault location calculation layer, calculate the distance between the fault point and the signal injection point, and generate fault location information.

[0085] In this embodiment, based on the high-resistance grounding fault identified in step S130, the corresponding location calculation model is retrieved from the fault location criterion library. This fault location criterion library includes a high-resistance grounding location model, a porcelain insulator breakdown location model, and a surge arrester breakdown location model. Each location model contains different location calculation formulas and corresponding weight coefficients. The high-resistance grounding location model includes the formula: Distance = a × Amplitude Attenuation Rate + b × Phase Shift, where the weight coefficients a = 0.6 and b = 0.4. Amplitude attenuation features and phase shift features of frequency-energy combinations strongly correlated with the high-resistance grounding fault (i.e., the top 10 combinations with the highest similarity in the mapping relationship library, such as (10Hz, high energy), (20Hz, high energy), etc.) are extracted from the multi-dimensional response feature set. The extracted amplitude attenuation features are input into the location calculation model to calculate a first distance estimate of 2.2 kilometers; the extracted phase shift features are input into the location calculation model to calculate a second distance estimate of 2.4 kilometers. Based on the weighting coefficients in the location calculation model, the first and second distance estimates are weighted and fused to calculate the final fault distance: 0.6 × 2.2 + 0.4 × 2.4 = 2.28 kilometers. Combining the distribution network topology, the specific location coordinates of the fault point are determined to be the cable joint at 2.28 kilometers, generating fault location information. This invention selects a dedicated location model based on the fault type and calculates the distance using a weighted fusion method of amplitude and phase characteristics. Compared with existing technologies that use a fixed single formula (such as based solely on traveling wave velocity), this scheme can adaptively adjust the location strategy, overcoming errors caused by the different effects of different fault types on signal propagation. For example, signal attenuation dominates when there is high-resistance grounding, and phase distortion dominates when there is porcelain insulator breakdown. This scheme dynamically matches fault characteristics through weighting coefficients, reducing the location error from ±500 meters in traditional methods to within ±150 meters, significantly improving the accuracy of fault inspection.

[0086] Step S150: Integrate the fault type identification results and fault location information to generate the distribution network grounding fault inspection results, and send the inspection results to the fault handling and control system.

[0087] In this embodiment, the identified high-resistance grounding fault type and the calculated fault location at 2.28 kilometers are integrated into a complete inspection result, including information such as fault type, location coordinates, and occurrence time. This result is then sent to the fault handling and management system via a 4G wireless communication network. The system automatically generates a maintenance work order, notifying maintenance personnel to proceed to the site for handling.

[0088] The above steps S110 to S150 complete the basic process for inspecting grounding faults in the distribution network based on frequency conversion and energy conversion signals. The detailed implementation methods of each step are described in further detail below.

[0089] Further explanation regarding step S110:

[0090] In step S110, the variable frequency and variable energy signal injection parameters are initialized, the frequency scanning range and energy level sequence of the variable frequency and variable energy signal are determined, and a frequency-energy two-dimensional scanning matrix is ​​constructed. This specifically includes the following steps:

[0091] Step S111: Determine the lower limit frequency value and upper limit frequency value of the frequency scanning range, divide the frequency scanning range into multiple discrete frequency points based on the preset frequency step size, and generate a frequency point sequence. The frequency point sequence includes fundamental frequency points, harmonic frequency points, and interharmonic frequency points.

[0092] In this embodiment, the lower frequency limit is set to 10Hz, the upper frequency limit is set to 1000Hz, and the frequency step size is set to 10Hz, thereby generating a frequency point sequence of 10Hz, 20Hz, 30Hz...1000Hz. The fundamental frequency point is 50Hz, the harmonic frequencies are 100Hz, 150Hz, 200Hz...950Hz, and the interharmonic frequencies are non-integer harmonics such as 15Hz, 25Hz, and 35Hz.

[0093] Step S112: Determine the lowest and highest energy levels of the energy level sequence, divide the energy level sequence into multiple discrete energy levels based on a preset energy step size, and generate an energy level sequence, which includes a first energy level, a second energy level, and a third energy level.

[0094] In this embodiment, the lowest energy level is set to 200V, the highest energy level is set to 800V, and the energy step size is set to 300V, generating a first energy level of 200V, a second energy level of 500V, and a third energy level of 800V. It should be noted that the number of energy levels can be adjusted according to the actual line length and fault type; the number of energy levels can be increased when the line is long or the fault type is complex.

[0095] Step S113: Perform a Cartesian product operation on the frequency point sequence and the energy level sequence to generate a frequency-energy two-dimensional scanning matrix. Each element of the two-dimensional scanning matrix corresponds to a cross combination of a frequency point and an energy level. Each combination is configured with an independent injection timing identifier and an injection duration identifier.

[0096] In this embodiment, the frequency point sequence (100 frequency points) and the energy level sequence (3 energy levels) are subjected to a Cartesian product operation to obtain a two-dimensional scan matrix containing 300 combinations. Each combination is assigned an injection timing identifier (1 to 300) and an injection duration identifier (5 cycles) to ensure that each combination is injected in sequence and the injection duration is consistent.

[0097] Step S114: Set a signal injection priority for each frequency-energy combination. The signal injection priority is determined based on the sensitivity of different fault types to the frequency-energy combination in historical fault data. High-sensitivity combinations are configured with high injection priority, and low-sensitivity combinations are configured with low injection priority.

[0098] In this embodiment, based on historical fault data statistics, high-resistance grounding faults are sensitive to low-frequency, high-energy combinations; therefore, the (10Hz-50Hz, high-energy) combination is set as a high priority. Porcelain insulator breakdown faults are sensitive to high-frequency, low-energy combinations; therefore, the (500Hz-1000Hz, low-energy) combination is set as a high priority. Surge arrester breakdown faults are sensitive to medium-frequency, medium-energy combinations; therefore, the (100Hz-400Hz, medium-energy) combination is set as a medium priority. During fault injection, high-priority combinations are injected first. If the high-priority combinations do not detect the fault, medium- and low-priority combinations are injected sequentially to improve the timeliness of fault detection.

[0099] Further explanation regarding step S120:

[0100] In step S120, frequency conversion and energy conversion signals are injected into the distribution network based on the frequency-energy two-dimensional scanning matrix. Feedback signals under different frequency-energy combinations are synchronously collected from the distribution network line side. The response characteristic parameters corresponding to each combination are extracted to generate a multi-dimensional response feature set. Specifically, this includes the following steps:

[0101] Step S121: Inject frequency-energy combination variable frequency and energy signals into the distribution network in sequence according to the injection timing identifier of the frequency-energy two-dimensional scanning matrix. After each injection, the corresponding feedback signal is collected. The feedback signal includes voltage waveform data, current waveform data, and zero-sequence component data.

[0102] In this embodiment, each combination signal is injected sequentially according to the injection timing identifier from 1 to 300. After each injection, three-phase voltage and current waveform data are collected through the voltage and current transformers on the line side, and the zero-sequence component data of the zero-sequence current transformer and the zero-sequence voltage transformer are also collected. All data are recorded synchronously, with a sampling frequency of 10kHz and a sampling duration of 5 cycles to ensure complete capture of the signal response.

[0103] Step S122: Perform time-domain analysis on the feedback signal and extract the amplitude attenuation characteristics for each frequency-energy combination. The amplitude attenuation characteristics include the first wave amplitude, the second wave amplitude, the amplitude attenuation rate, and the slope of the amplitude attenuation curve.

[0104] In this embodiment, time-domain analysis is performed on the acquired current waveform, the first wave amplitude (e.g., 15A) and the second wave amplitude (e.g., 10A) of the fault phase current under each combination are recorded, the amplitude attenuation rate ((15-10) / 15=0.33) is calculated, and the amplitude change curve over time is linearly fitted to obtain the slope of the amplitude attenuation curve (e.g., -0.05 / ms).

[0105] Step S123: Perform frequency domain transformation on the feedback signal and extract the phase shift characteristics under each frequency-energy combination. The phase shift characteristics include the fundamental phase shift, harmonic phase shift, and the inflection frequency of the phase shift curve.

[0106] In this embodiment, a Fast Fourier Transform (FFT) is performed on the voltage waveform to obtain the amplitude and phase information of each frequency component. The phase difference between the injected signal frequency and the same frequency component of the feedback signal is calculated to obtain the fundamental phase offset (e.g., 5°). Simultaneously, the phase offsets of the 3rd, 5th, and 7th harmonics are extracted, and the curves of phase offset versus frequency are analyzed. The point of abrupt change in the slope of the curve is identified by differentiation as the inflection point frequency (e.g., 150Hz).

[0107] Step S124: Perform waveform morphology analysis on the feedback signal and extract waveform distortion features for each frequency-energy combination. The waveform distortion features include waveform rising edge steepness, waveform falling edge steepness, waveform zero-crossing offset, and waveform distortion area.

[0108] In this embodiment, morphological analysis is performed on the current waveform. The reciprocal of the time required for the rising edge to rise from 10% amplitude to 90% amplitude is calculated as the rising edge steepness (e.g., 2 A / ms), and the falling edge steepness is calculated similarly (e.g., 1.8 A / ms). The offset time of the waveform's zero-crossing point relative to the zero-crossing point of the standard sine wave is measured as the zero-crossing offset (e.g., 0.2 ms). The area difference between the actual waveform and the standard sine wave is calculated by integrating the absolute values ​​of the amplitude differences at each sampling point to obtain the waveform distortion area (e.g., 0.5 A·ms).

[0109] Step S125: Preprocess the collected feedback signal to remove abnormal data, including signal-to-noise ratio calculation, consistency verification, normalization processing, and outlier detection.

[0110] In this embodiment, firstly, the signal-to-noise ratio (SNR) of the feedback signal for each frequency-energy combination is calculated. If the SNR is lower than 10 dB, the combination is marked as a low SNR combination, and its response characteristic parameters are not used in fault type identification and fault location calculation. Secondly, the waveform correlation coefficient is calculated for the feedback signals of the same combination acquired three times consecutively. If the correlation coefficient is lower than 0.8, the combination signal is reacquired until the consistency requirement is met. Then, the extracted amplitude attenuation features, phase shift features, and waveform distortion features are normalized, and the max-min normalization method is used to map each feature value to the [0, 1] interval. Finally, outlier detection is performed on the normalized multi-dimensional response feature set, and outlier feature parameters exceeding the mean ± 3 standard deviations are removed. The missing positions after removal are filled with the historical statistical mean.

[0111] Further explanation regarding step S130:

[0112] In step S130, a multi-criteria fusion diagnostic model is constructed. The multi-criteria fusion diagnostic model includes a fault type identification layer and a fault location calculation layer. The multi-dimensional response feature set is input into the fault type identification layer, and the current fault type is identified through a preset fault feature mapping relationship. Specifically, the following steps are included:

[0113] Step S131: Establish a mapping relationship library between fault types and frequency-energy response characteristics. The mapping relationship library includes a high-resistance grounding fault mapping table, a porcelain insulator breakdown fault mapping table, and a surge arrester breakdown fault mapping table. Each fault type mapping table records the standard response characteristic vector of the fault type under different frequency-energy combinations.

[0114] In this embodiment, mapping tables for three fault types were established through a laboratory simulation platform and the accumulation of historical field data. During laboratory simulations, three fault conditions were set up: high-resistance grounding (transition resistance 500Ω), porcelain insulator breakdown (breakdown voltage 20kV), and surge arrester breakdown (current capacity 5kA). Response characteristics of 300 frequency-energy combinations under each condition were collected, and the average of multiple experiments was used as the standard vector. For example, the high-resistance grounding fault mapping table records: under the (10Hz, high-energy) combination, the standard amplitude attenuation rate is 0.25, the standard phase shift is 4.5°, and the standard waveform zero-crossing offset is 0.15ms; under the (100Hz, medium-energy) combination, the standard amplitude attenuation rate is 0.45, etc.

[0115] Step S132: Combine the response feature parameters of each frequency-energy combination in the multi-dimensional response feature set into the measured response feature vector of the combination, calculate the similarity between the measured response feature vector and the standard response feature vector in each fault type mapping table, and generate a similarity matrix.

[0116] In this embodiment, for the 300 combinations of fault data collected at the current time, the measured characteristic parameters of each combination (such as amplitude attenuation rate, phase offset, and waveform zero-crossing offset) are used to form a three-dimensional measured vector. The cosine similarity of this vector with the standard vectors in the three fault type mapping tables is calculated one by one. The cosine similarity calculation formula is as follows: ,in For the measured vector, Using standard vectors, we obtain a 300-row, 3-column similarity matrix.

[0117] Step S133: Based on the similarity matrix, calculate the total similarity value of each fault type under all frequency-energy combinations, and select the fault type with the highest total similarity value as the preliminary identification result of the current fault.

[0118] In this embodiment, the 300 similarity values ​​of each column (i.e. each fault type) in the similarity matrix are summed to obtain a total value of 85 for high-resistance grounding, 60 for porcelain insulator breakdown, and 55 for surge arrester breakdown. Therefore, it is initially identified as a high-resistance grounding fault.

[0119] Step S134: If the difference between the total similarity value of the preliminary identification result and the total similarity value of the second highest fault type is lower than the preset discrimination threshold, the secondary identification process is triggered to extract the waveform zero-crossing offset data from the waveform distortion features in the multi-dimensional response feature set, compare the zero-crossing offset standard value of each fault type under the preset key frequency-energy combination, and redetermine the fault type identification result.

[0120] In this embodiment, the preset discrimination threshold is 10. The difference between the total high-resistance grounding value of 85 and the second highest total value of 60 is 25, which is greater than the threshold of 10. Therefore, secondary identification is not triggered, and the preliminary identification result is directly used as the final result. If the difference is less than 10, the secondary identification process is triggered: the measured value of the waveform zero-crossing offset under the preset key frequency-energy combination (e.g., (50Hz, medium energy)) is extracted, and the Euclidean distance is calculated with the standard value of each fault type under this combination. The Euclidean distance calculation formula is as follows: , The distance is the Euclidean distance, a dimensionless value. The dimension of the feature vector, i.e., the number of feature parameters involved in the comparison. The first eigenvector of the measured response is... One portion, The first eigenvector of the standard response The component is selected, and the fault type with the smallest distance is chosen as the final identification result.

[0121] Further explanation regarding step S140:

[0122] In step S140, based on the identified fault type, the corresponding fault location criterion is invoked, the multi-dimensional response feature set is input into the fault location calculation layer, the distance between the fault point and the signal injection point is calculated, and fault location information is generated. Specifically, this includes the following steps:

[0123] Step S141: Based on the identified fault type, retrieve the corresponding location calculation model from the fault location criterion library. The fault location criterion library includes high-resistance grounding location model, porcelain insulator breakdown location model, and surge arrester breakdown location model. Each location model contains different location calculation formulas and corresponding weight coefficients.

[0124] In this embodiment, a high-resistance grounding fault location model is retrieved based on the identified high-resistance grounding fault. This model is established based on transmission line theory and the fault traveling wave principle, and the location calculation formula is as follows: ,in Distance to the fault (kilometers) To inject signal amplitude, This is the measured amplitude. This is the phase offset. and These are the weighting coefficients, and In the high-resistance grounding location model , .

[0125] Step S142: Extract the amplitude attenuation features of frequency-energy combinations with fault type correlation weights higher than a preset threshold from the multi-dimensional response feature set. The correlation weights are determined based on high similarity combinations of the fault type in the mapping relationship library.

[0126] In this embodiment, the preset correlation weight threshold is 0.8. The top 10 combinations with a similarity higher than 0.8 to high-resistance grounding faults are found from the mapping relationship library, including (10Hz, high energy), (20Hz, high energy), (30Hz, high energy), etc. The amplitude attenuation feature value (i.e., the ratio of A0 / A) of these combinations is extracted, and the arithmetic mean is taken as the amplitude attenuation feature input.

[0127] Step S143: Input the extracted amplitude attenuation features into the retrieved positioning calculation model to calculate the first distance estimate between the fault point and the signal injection point; extract the phase offset features of the frequency-energy combination with the fault type correlation weight higher than the preset threshold from the multi-dimensional response feature set, input them into the positioning calculation model, and calculate the second distance estimate between the fault point and the signal injection point.

[0128] In this embodiment, substituting the average amplitude attenuation rate of 0.28 into the formula yields the first distance estimate L1 = 0.6 × (1 / 0.28) × line coefficient, which is calculated to be 2.2 kilometers. Substituting the average phase offset of 5.2° into the formula yields the second distance estimate L2 = 0.4 × 5.2 × line coefficient, which is calculated to be 2.4 kilometers.

[0129] Step S144: Based on the weight coefficients in the positioning calculation model, the first distance estimate and the second distance estimate are weighted and fused to generate the final fault distance value; based on the final fault distance value and the distribution network line topology, the specific location coordinates of the fault point are determined to generate fault location information.

[0130] In this embodiment, the final fault distance value L = α × L1 + β × L2 = 0.6 × 2.2 + 0.4 × 2.4 = 2.28 kilometers. Combining the line GIS topology data, this distance corresponds to the cable joint at 2.28 kilometers. By combining latitude and longitude coordinates, complete fault location information including fault type, distance, and coordinates is generated.

[0131] To further improve the accuracy and adaptability of fault inspection, this method also includes establishing a feedback optimization link, and iteratively optimizing the frequency-energy two-dimensional scanning matrix and the multi-criteria fusion diagnostic model through historical fault diagnosis data. Specifically, this includes the following steps:

[0132] Step S170: Establish a feedback optimization link and iteratively optimize the frequency-energy two-dimensional scanning matrix and the multi-criteria fusion diagnostic model using historical fault diagnosis data.

[0133] In this embodiment, the on-site confirmation results after each fault handling are compared with the diagnostic results to accumulate historical fault diagnosis records. The accumulated historical data is analyzed monthly to statistically analyze misjudged and error samples, thereby optimizing the scanning matrix and diagnostic model.

[0134] Step S171: Collect historical fault diagnosis records from the fault handling and control system. The historical fault diagnosis records include the multi-dimensional response feature set collected when the fault occurred, the fault type output by the diagnosis, the fault location output by the diagnosis, the actual confirmed fault type, and the actual confirmed fault location.

[0135] In this embodiment, 50 fault records that occurred in the past year are collected from the fault handling and management system database. Each record contains feature set data at the time of the fault occurrence, diagnostic results, and actual results confirmed on-site by maintenance personnel.

[0136] Step S172: Compare the fault type output by the diagnosis with the actual confirmed fault type. If they are inconsistent, extract the multi-dimensional response feature set corresponding to the fault and mark it as a misjudged sample. Compare the fault location output by the diagnosis with the actual confirmed fault location and calculate the positioning error value. If the positioning error value exceeds the preset error threshold, mark the multi-dimensional response feature set corresponding to the fault as an error sample.

[0137] In this embodiment, after comparison, it was found that 5 fault type diagnoses were incorrect (such as misjudging the porcelain insulator breakdown as high resistance grounding), and these were marked as misjudged samples; 10 fault location errors exceeded 0.5 kilometers (preset error threshold), and these were marked as error samples.

[0138] Step S173: Input the misjudged samples and error samples into the optimization analysis module, count the frequency of each frequency-energy combination in the misjudged samples and error samples, and generate a frequency-energy combination effectiveness score; adjust the injection priority of each combination in the frequency-energy two-dimensional scanning matrix based on the effectiveness score, reduce the injection priority of combinations with low effectiveness scores or remove them from the scanning matrix, and increase the injection priority or increase the sampling density of combinations with high effectiveness scores.

[0139] In this embodiment, statistics show that the combination (30Hz, medium energy) appeared 4 times in 5 misclassified samples and 3 times in 10 error samples, with a low effectiveness score. Therefore, its injection priority was reduced from high to low, and it was considered to be removed from the scanning matrix in the next optimization. On the other hand, the combination (10Hz, high energy) appeared frequently in correctly identified samples and had a high effectiveness score. Therefore, its injection priority was kept high, and the sampling density was increased from 5 cycles to 10 cycles per time.

[0140] Step S174: Based on the misjudged samples and error samples, adjust the standard response feature vector parameters in the fault type mapping table and the weight coefficients of each location model in the fault location criterion library to generate an iteratively optimized multi-criterion fusion diagnostic model.

[0141] In this embodiment, for misjudged samples, the difference between the mean of the measured response feature vector of the actual fault type (e.g., porcelain insulator breakdown) and the standard vector of porcelain insulator breakdown in the current mapping table is calculated to generate the deviation of each feature dimension. Based on this deviation, the current standard vector is corrected using the formula: New standard vector = Old standard vector + Deviation × Learning rate, with the learning rate set to 0.1. Simultaneously, an upper limit threshold of 0.2 is set; if the correction exceeds 0.2, it is truncated to 0.2. For error samples, the location calculation data corresponding to each fault type is extracted, and the average error of the first distance estimate and the average error of the second distance estimate are statistically analyzed. A confidence coefficient is calculated based on the average error, and the weight coefficients are recalculated and normalized based on the confidence coefficient. The updated standard vector and weight coefficients are stored in the mapping relation library and the location criterion library, respectively, generating an iteratively optimized multi-criterion fusion diagnostic model. This invention introduces a feedback optimization mechanism from machine learning, enabling the diagnostic system to have self-evolution capabilities. Compared with the static and fixed diagnostic models in the prior art, this solution can continuously optimize based on actual operating data. By statistically analyzing misjudged and error samples, the priority of the frequency-energy combination and diagnostic parameters are dynamically adjusted, significantly enhancing the system's adaptability to new fault modes and environmental changes. After 3-5 iterations of optimization, the overall diagnostic accuracy can be further improved by 10%-15%, and it has the ability to continuously track long-term evolving fault characteristics such as line aging and load changes, avoiding performance degradation caused by outdated data in traditional methods.

[0142] To further improve the safety and effectiveness of signal injection, a pre-assessment of the line status is conducted before injecting frequency conversion and energy conversion signals into the distribution network to determine the initial injection parameters. This includes the following steps:

[0143] Step S105: Before injecting frequency conversion and energy conversion signals into the distribution network, perform a pre-assessment of the line status and determine the initial injection parameters.

[0144] In this embodiment, before each inspection begins, real-time operating parameters of the current line are obtained from the distribution network monitoring system. These real-time operating parameters include the line no-load status indicator, line load rate data, and line-to-ground capacitance value data. Based on the line no-load status indicator, it is determined whether the line is in a no-load state. If the line is in a no-load state, the lower limit of the frequency scan range is extended from 10Hz to 5Hz, and the lowest energy level of the energy level sequence is extended from 200V to 100V. Based on the line load rate data, the line load level is determined. If the line load rate exceeds the 80% heavy load threshold, the upper limit of the frequency scan range is extended from 1000Hz to 1200Hz, and the highest energy level of the energy level sequence is extended from 800V to 1000V. Based on the line-to-ground capacitance value data, the resonance risk frequency point during signal injection is calculated using the following formula: , The resonant risk frequency is indicated by the number of Hertz (Hz). This is the equivalent inductance of the circuit, measured in Henry (H). This refers to the line-to-ground capacitance, measured in farads (F). Pi The square root of the product of inductance and capacitance represents the resonant time constant of the line. If the calculated resonant frequency falls within a frequency sequence, that frequency is removed from the sequence, or the injected energy level is reduced to the lowest level at that frequency. Finally, based on the expanded or adjusted frequency scan range and energy level sequence, a frequency-energy two-dimensional scan matrix is ​​regenerated. This invention dynamically adjusts the injection parameters through line condition pre-assessment. Compared with the fixed parameter injection method in the prior art, this scheme can adapt to different operating conditions. Lowering the lower limit of frequency and energy level under no-load conditions avoids unnecessary losses; expanding the high-frequency and high-energy range under heavy load enhances signal penetration; and removing resonant points prevents the risk of overvoltage caused by the injected signal. This adaptive adjustment enables the fault inspection system to operate safely and efficiently under various line conditions, while reducing energy consumption by about 30% and completely avoiding the equipment damage risk caused by resonance in traditional methods.

[0145] The above steps describe in detail the specific application of the distribution network grounding fault inspection method based on variable frequency and energy signals in a 10kV distribution network. Through frequency-energy two-dimensional scanning, it can simultaneously excite multiple hidden faults such as high-resistance grounding, porcelain insulator breakdown, and surge arrester breakdown, solving the problem that traditional single-frequency or single-energy signals cannot cover multiple types of faults. Through the hierarchical design of the multi-criteria fusion diagnostic model, accurate identification of fault types and accurate location of faults are achieved. Through feedback optimization of the link, the system can continuously evolve based on historical data, improving the accuracy and adaptability of fault inspection. Through line status pre-assessment and dynamic adjustment of injection parameters, resonance risks and resource waste are avoided.

[0146] The above are merely preferred embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0147] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for inspecting grounding faults in a distribution network based on variable frequency energy conversion signals, characterized in that, include: Initialize the variable frequency and variable energy signal injection parameters, determine the frequency scanning range and energy level sequence of the variable frequency and variable energy signal, and construct a frequency-energy two-dimensional scanning matrix, which contains a cross combination of multiple frequency points and multiple energy levels; Based on the frequency-energy two-dimensional scanning matrix, frequency conversion and energy conversion signals are injected into the distribution network. Feedback signals under different frequency-energy combinations are synchronously collected from the distribution network line side. The response characteristic parameters corresponding to each combination are extracted to generate a multi-dimensional response characteristic set. The multi-dimensional response characteristic set includes amplitude attenuation characteristics, phase shift characteristics, and waveform distortion characteristics. A multi-criteria fusion diagnostic model is constructed, which includes a fault type identification layer and a fault location calculation layer. The multi-dimensional response feature set is input into the fault type identification layer, and the current fault type is identified through a preset fault feature mapping relationship. The fault types include high-resistance grounding fault, porcelain insulator breakdown fault, and surge arrester breakdown fault. Based on the identified fault type, the corresponding fault location criteria are invoked, the multi-dimensional response feature set is input into the fault location calculation layer, the distance between the fault point and the signal injection point is calculated, and fault location information is generated. The fault type identification results and the fault location information are integrated to generate the distribution network grounding fault inspection results, and the inspection results are sent to the fault handling and control system.

2. The method for inspecting grounding faults in a distribution network based on variable frequency energy conversion signals according to claim 1, characterized in that, The initialization of the variable frequency and variable energy signal injection parameters determines the frequency scanning range and energy level sequence of the variable frequency and variable energy signal, and constructs a frequency-energy two-dimensional scanning matrix, including: The lower and upper frequency values ​​of the frequency scanning range are determined, and the frequency scanning range is divided into multiple discrete frequency points based on a preset frequency step size to generate a frequency point sequence, which includes fundamental frequency points, harmonic frequency points, and interharmonic frequency points. The lowest and highest energy levels of the energy level sequence are determined, and the energy level sequence is divided into multiple discrete energy levels based on a preset energy step size to generate an energy level sequence, which includes a first energy level, a second energy level, and a third energy level. The frequency point sequence and the energy level sequence are subjected to Cartesian product operation to generate the frequency-energy two-dimensional scanning matrix. Each element of the two-dimensional scanning matrix corresponds to a cross combination of a frequency point and an energy level. Each combination is configured with an independent injection timing identifier and injection duration identifier. A signal injection priority is set for each frequency-energy combination. The signal injection priority is determined based on the sensitivity of different fault types to the frequency-energy combination in historical fault data. High-sensitivity combinations are configured with high injection priority, and low-sensitivity combinations are configured with low injection priority.

3. The method for inspecting grounding faults in a distribution network based on variable frequency energy conversion signals according to claim 1, characterized in that, The process involves injecting frequency- and energy-converting signals into the distribution network based on the frequency-energy two-dimensional scanning matrix, synchronously acquiring feedback signals under different frequency-energy combinations from the distribution network line side, extracting response characteristic parameters corresponding to each combination, and generating a multi-dimensional response feature set, including: According to the injection timing identifier of the frequency-energy two-dimensional scanning matrix, frequency-energy combination variable frequency and energy signals are injected into the distribution network in sequence. After each injection, the corresponding feedback signal is collected. The feedback signal includes voltage waveform data, current waveform data, and zero-sequence component data. Time-domain analysis is performed on the feedback signal to extract the amplitude attenuation characteristics under each frequency-energy combination. The amplitude attenuation characteristics include the first wave amplitude, the second wave amplitude, the amplitude attenuation rate, and the slope of the amplitude attenuation curve. The feedback signal is subjected to frequency domain transformation to extract the phase shift characteristics under each frequency-energy combination. The phase shift characteristics include the fundamental phase shift, harmonic phase shift, and the inflection frequency of the phase shift curve. The feedback signal is subjected to waveform morphology analysis to extract waveform distortion features under each frequency-energy combination. The waveform distortion features include waveform rising edge steepness, waveform falling edge steepness, waveform zero-crossing offset, and waveform distortion area. The amplitude attenuation characteristics, phase shift characteristics, and waveform distortion characteristics of each frequency-energy combination are associated and stored according to the frequency-energy combination identifier to generate the multi-dimensional response feature set. The multi-dimensional response feature set is stored in the form of a three-dimensional data table, with the three-dimensional coordinates being the frequency dimension, energy dimension, and feature type dimension, respectively.

4. The method for inspecting grounding faults in a distribution network based on frequency conversion and energy conversion signals according to claim 3, characterized in that, The construction of a multi-criteria fusion diagnostic model includes a fault type identification layer and a fault location calculation layer. The multi-dimensional response feature set is input into the fault type identification layer, and the current fault type is identified through a preset fault feature mapping relationship, including: Establish a mapping relationship library between fault types and frequency-energy response characteristics. The mapping relationship library includes a high-resistance grounding fault mapping table, a porcelain insulator breakdown fault mapping table, and a surge arrester breakdown fault mapping table. Each fault type mapping table records the standard response characteristic vector of the fault type under different frequency-energy combinations. The response feature parameters of each frequency-energy combination in the multi-dimensional response feature set are combined into the measured response feature vector of the combination. The similarity between the measured response feature vector and the standard response feature vector in each fault type mapping table is calculated to generate a similarity matrix. Based on the similarity matrix, the total similarity value of each fault type under all frequency-energy combinations is calculated, and the fault type with the highest total similarity value is selected as the preliminary identification result of the current fault. If the difference between the total similarity value of the preliminary identification result and the total similarity value of the second highest fault type is lower than the preset discrimination threshold, the secondary identification process is triggered to extract the waveform zero-crossing offset data from the waveform distortion features in the multi-dimensional response feature set, compare the zero-crossing offset standard value of each fault type under the preset key frequency-energy combination, and redetermine the fault type identification result. The final fault type identification result is output to the fault location calculation layer.

5. The method for inspecting grounding faults in a distribution network based on variable frequency energy conversion signals according to claim 4, characterized in that, The fault location criteria based on the identified fault type are invoked, and the multi-dimensional response feature set is input into the fault location calculation layer to calculate the distance between the fault point and the signal injection point, generating fault location information, including: Based on the identified fault type, the corresponding location calculation model is retrieved from the fault location criterion library. The fault location criterion library includes a high-resistance grounding location model, a porcelain insulator breakdown location model, and a surge arrester breakdown location model. Each location model contains different location calculation formulas and corresponding weight coefficients. The amplitude attenuation features of frequency-energy combinations with fault type correlation weights higher than a preset threshold are extracted and identified from the multi-dimensional response feature set. The correlation weights are determined based on high similarity combinations of the fault type in the mapping relationship library. The extracted amplitude attenuation features are input into the retrieved location calculation model to calculate the first distance estimate between the fault point and the signal injection point. Phase offset features of frequency-energy combinations with fault type correlation weights higher than a preset threshold are extracted and identified from the multi-dimensional response feature set, and input into the localization calculation model to calculate the second distance estimate between the fault point and the signal injection point. Based on the weighting coefficients in the positioning calculation model, the first distance estimate and the second distance estimate are weighted and fused to generate the final fault distance value; Based on the final fault distance value and the distribution network line topology, the specific location coordinates of the fault point are determined, and the fault location information is generated.

6. The method for inspecting grounding faults in a distribution network based on variable frequency energy conversion signals according to claim 1, characterized in that, The method further includes establishing a feedback optimization link, and iteratively optimizing the frequency-energy two-dimensional scanning matrix and the multi-criteria fusion diagnostic model through historical fault diagnosis data, including: Historical fault diagnosis records are collected from the fault handling and control system. These historical fault diagnosis records include a multi-dimensional response feature set collected when the fault occurred, the fault type output by the diagnosis, the fault location output by the diagnosis, the actual confirmed fault type, and the actual confirmed fault location. Compare the fault type output by the diagnosis with the actual confirmed fault type. If the two are inconsistent, extract the multi-dimensional response feature set corresponding to the fault and mark it as a misjudged sample. Compare the fault location output by the diagnosis with the actual confirmed fault location, calculate the positioning error value, and if the positioning error value exceeds the preset error threshold, mark the multi-dimensional response feature set corresponding to the fault as an error sample. The misjudged samples and the error samples are input into the optimization analysis module to count the frequency of each frequency-energy combination in the misjudged samples and error samples, and generate a frequency-energy combination effectiveness score. Based on the effectiveness score, the injection priority of each combination in the frequency-energy two-dimensional scanning matrix is ​​adjusted. Combinations with low effectiveness scores have their injection priority reduced or are removed from the scanning matrix, while combinations with high effectiveness scores have their injection priority increased or their sampling density increased. Based on the misjudged samples and the error samples, the standard response feature vector parameters in the fault type mapping table and the weight coefficients of each location model in the fault location criterion library are adjusted to generate an iteratively optimized multi-criterion fusion diagnostic model.

7. The method for inspecting grounding faults in a distribution network based on frequency conversion and energy conversion signals according to claim 6, characterized in that, The adjustment of the standard response feature vector parameters in the fault type mapping table includes: Extract samples from the misjudged samples where the actual fault type is inconsistent with the diagnosed fault type, and calculate the mean of the measured response feature vector corresponding to the actual fault type and the mean of the measured response feature vector corresponding to the diagnosed fault type. Calculate the difference between the mean of the measured response feature vector and the standard response feature vector of the corresponding fault type in the current mapping table to generate the deviation of each feature dimension; The current standard response feature vector is corrected based on the deviation amount. The corrected standard response feature vector is equal to the current standard vector plus the product of the deviation amount and the preset learning rate. Set an upper limit threshold for the correction magnitude. If the difference between the corrected vector and the current vector exceeds the upper limit threshold, then truncation correction is performed according to the upper limit threshold. The corrected standard response feature vector is updated in the fault type mapping table, replacing the original standard response feature vector.

8. The method for inspecting grounding faults in a distribution network based on variable frequency energy conversion signals according to claim 6, characterized in that, The adjustment of the weight coefficients of each localization model in the fault localization criterion library includes: The location calculation data corresponding to each fault type is extracted from the error samples. The location calculation data includes a first distance estimate, a second distance estimate, a final fault distance value, and an actual fault distance value. For each fault type location model, the average error of the first distance estimate and the average error of the second distance estimate are calculated among all error samples under that model. The confidence coefficients of the first distance estimate and the second distance estimate are calculated based on the average error. The confidence coefficients are inversely proportional to the average error. Based on the confidence coefficient, the weights of the first distance estimate and the second distance estimate are recalculated. The new weights are equal to the original weights multiplied by the confidence coefficient and then normalized. The recalculated weight coefficients are updated in the localization model for the corresponding fault type, replacing the original weight coefficients.

9. The method for inspecting grounding faults in a distribution network based on frequency conversion and energy conversion signals according to claim 1, characterized in that, Before injecting frequency conversion and energy conversion signals into the distribution network, the process also includes a pre-assessment of the line status to determine initial injection parameters, including: The real-time operating parameters of the current line are obtained from the power distribution network monitoring system. The real-time operating parameters include the line no-load status indicator, line load rate data, and line-to-ground capacitance value data. Based on the line no-load status indicator, it is determined whether the line is in a no-load state. If the line is in a no-load state, the lower limit frequency value of the frequency scanning range is extended downward by a preset first frequency offset, and the lowest energy level of the energy level sequence is extended downward by a preset first energy offset. Based on the line load rate data, the load level of the line is determined. If the line load rate exceeds the preset overload threshold, the upper limit frequency value of the frequency scanning range is extended upward by a preset second frequency offset, and the highest energy level of the energy level sequence is extended upward by a preset second energy offset. Based on the line-to-ground capacitance data, calculate the resonance risk frequency point when injecting the signal, remove the resonance risk frequency point from the frequency point sequence, or reduce the injection energy level at the resonance risk frequency point. Based on the expanded frequency scan range and energy level sequence, the frequency-energy two-dimensional scan matrix is ​​regenerated.

10. The method for inspecting grounding faults in a distribution network based on variable frequency energy conversion signals according to claim 1, characterized in that, The method further includes preprocessing the collected feedback signals to remove abnormal data, including: The signal-to-noise ratio (SNR) of the synchronously acquired feedback signal is calculated. If the SNR of the feedback signal under a certain frequency-energy combination is lower than the preset SNR threshold, the combination is marked as a low SNR combination. The response characteristic parameters of this combination are not included in the fault type identification and fault location calculation. Consistency verification is performed on feedback signals acquired multiple times for the same frequency-energy combination. The waveform correlation coefficient of the acquired signals is calculated. If the correlation coefficient is lower than a preset consistency threshold, the feedback signal of the combination is reacquired until the consistency requirement is met. The extracted response feature parameters are normalized to map feature parameters of different dimensions to a unified numerical range, generating a normalized multi-dimensional response feature set. Outlier detection is performed on the normalized multi-dimensional response feature set to remove abnormal feature parameters that exceed the preset feature value range, and the missing positions after removal are filled with historical statistical mean or interpolation method.

Images