An intelligent fault diagnosis system based on power distribution equipment
By collecting partial discharge waveforms and power frequency signals from power distribution equipment, and combining electromagnetic wave propagation models and multi-source data analysis, the problems of positioning accuracy and multi-source data fusion in power distribution equipment fault diagnosis are solved. This enables precise positioning of insulation degradation and transparency of the fault development process, providing clear maintenance guidance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INNER MONGOLIA TENGXIN SMART ELECTRONICS CO LTD
- Filing Date
- 2026-04-15
- Publication Date
- 2026-07-14
AI Technical Summary
Existing fault diagnosis technologies for power distribution equipment suffer from insufficient positioning accuracy, lack of temporal evolution analysis, poor interpretability of diagnostic results, and low fusion of multi-source data. They are difficult to accurately identify the spatial location and development process of insulation degradation, and the diagnostic conclusions are disconnected from actual maintenance.
By synchronously acquiring partial discharge waveforms and power frequency signals from power distribution equipment, extracting pulse rise time and amplitude attenuation rate, calculating pulse projection weight, and combining with electromagnetic wave propagation model for preliminary positioning, the optimal propagation path is matched using pulse three-way frequency energy ratio, reflecting signals are eliminated, a multi-source parameter anomaly time sequence chain is constructed, evolution stages are divided, dominant stress type is determined, and a fault spatiotemporal evolution path is formed.
It enables precise spatial positioning of insulation degradation areas, traceability of fault processes, and interpretability of diagnostic results, providing targeted maintenance strategies and improving the accuracy and robustness of diagnosis.
Smart Images
Figure CN122045958B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power system fault diagnosis technology, and specifically to an intelligent fault diagnosis system based on power distribution equipment. Background Technology
[0002] Power distribution equipment (such as transformers, switchgear, cable terminals, circuit breakers, etc.) is an important component of the power system, and its operational reliability is directly related to power supply security. During long-term operation, insulation degradation is one of the main causes of failures, and partial discharge is an important early symptom of insulation degradation.
[0003] Existing fault diagnosis technologies for power distribution equipment have the following main shortcomings: First, insufficient positioning accuracy. Traditional partial discharge detection often relies on a single sensor or a simple time-difference method for positioning, which is easily affected by signal reflection, attenuation, and interference from complex electromagnetic environments, making it difficult to accurately identify the spatial location of insulation degradation. Especially when the internal structure of the equipment is complex and there are multiple dielectric interfaces, the electromagnetic wave propagation path is complex, and traditional methods cannot effectively distinguish between direct waves and reflected waves.
[0004] Second, there is a lack of time-series evolution analysis. Existing diagnostic systems mostly make isolated event judgments and fail to correlate the time-series changes of various parameters such as electrical parameters, vibration, and temperature rise with the insulation degradation process. They cannot reveal the complete process from the occurrence to the evolution of the fault and are difficult to effectively predict the degradation trend.
[0005] Third, the diagnostic results are poorly interpretable. Although intelligent diagnostic methods based on black-box models can provide conclusions, they lack the ability to distinguish between fault mechanisms (such as electrical stress-dominated or mechanical stress-dominated), making it difficult to guide maintenance personnel to take targeted repair measures, resulting in a disconnect between diagnostic conclusions and actual repairs.
[0006] Fourth, the integration of multi-source data is low. Existing systems often analyze data such as partial discharge, temperature rise, and vibration independently, failing to fully utilize the inherent physical correlation between multi-source data. This results in insufficient accuracy and robustness of diagnostic conclusions, making it difficult to meet the fault diagnosis needs under complex operating conditions.
[0007] Therefore, the present invention provides an intelligent fault diagnosis system based on power distribution equipment. Summary of the Invention
[0008] The purpose of this invention is to provide an intelligent fault diagnosis system based on power distribution equipment to solve the aforementioned background problems.
[0009] The objective of this invention can be achieved through the following technical solution: A fault intelligent diagnosis system based on power distribution equipment, comprising the following modules: Feature analysis module: By synchronously acquiring the partial discharge waveform and power frequency signal of the power distribution equipment, extracting the pulse rise time and amplitude attenuation rate and calculating the pulse projection weight, determining the pulse projection point by combining the electromagnetic wave propagation model, marking the candidate areas of insulation degradation and extracting the regional features through spatial aggregation and distance determination.
[0010] Spatial positioning module: Based on the pulse of the candidate region of insulation degradation, the optimal propagation path is matched by the pulse frequency division energy ratio, and the reflected signal region is eliminated by the dispersion of amplitude attenuation rate to determine the spatial positioning region of insulation degradation.
[0011] Evolution Construction Module: Continuously collects multi-source operating parameters of power distribution equipment, constructs a parameter anomaly time sequence chain, divides the evolution stages according to the time interval and degradation cycle, determines the accelerated degradation stage and the decelerated degradation stage by parameter differential accumulation, and forms a degradation stage time sequence.
[0012] Diagnostic output module: By calculating the electrical stress contribution and mechanical stress contribution of each accelerated deterioration segment, the dominant stress type is determined, and the active deterioration area of the corresponding time period is matched. The evolution correlation is judged based on the spatial distance of the area, and the spatiotemporal evolution path of the power distribution equipment fault is formed in series.
[0013] Furthermore, the pulse projection weight is calculated by continuously capturing the time-domain waveform of the partial discharge pulse at a fixed sampling frequency using ultra-wideband high-frequency current sensors and ultra-high-frequency sensors deployed around the critical insulation parts of the power distribution equipment.
[0014] The power frequency voltage signal is obtained by a voltage transformer, and the zero-crossing moment is extracted as a unified time base reference for all sensors.
[0015] For each captured partial discharge pulse waveform, the start point and peak point are identified, the time difference between the two points is calculated and recorded as the rise time, the amplitude of the pulse is measured and compared with the reference sensitivity calibrated by the sensor at the factory to obtain the amplitude attenuation rate, and the rise time is multiplied by the amplitude attenuation rate to obtain the pulse projection weight.
[0016] Furthermore, the method for locating the pulse projection point is as follows: the three-dimensional space inside the power distribution equipment is discretized into a node grid, and each node represents a possible power source location.
[0017] For each node, the theoretical attenuation characteristics of the partial discharge pulse generated from the node are calculated through electromagnetic field simulation. The three-dimensional coordinates of each node are mapped to their corresponding theoretical attenuation values to form a model database.
[0018] For the current pulse, the pulse projection weight is compared with the theoretical attenuation value of each node position in the model point by point to calculate the deviation, and the node coordinate with the smallest deviation is selected as the pulse source projection point.
[0019] Furthermore, the process of marking candidate regions of insulation degradation and extracting region features is as follows:
[0020] A continuous time window is set to collect the pulse projection points of all pulses within the time window. A density-based clustering algorithm is used to group spatially close pulse projection points into the same cluster and calculate the aggregation center point.
[0021] A three-dimensional model of the equipment insulation structure is obtained and the geometric surfaces of each insulation boundary are predefined. The shortest three-dimensional spatial distance from the aggregation center point to the boundary of each insulation layer is calculated. Clusters with distances less than a preset boundary threshold are marked as candidate regions for insulation degradation. The average amplitude attenuation rate and average rise time of all pulses within the candidate regions for insulation degradation are recorded as regional feature identifiers.
[0022] Furthermore, the optimal propagation path matching method is as follows: perform a fast Fourier transform on each pulse in the candidate region of insulation degradation to obtain the full-band energy spectrum, and divide the frequency range into three continuous sub-bands.
[0023] For each candidate region of insulation degradation, consider the potential propagation path to each sensor, count the number of nodes passed through each potential propagation path, and query the electromagnetic wave refraction coefficient of each node.
[0024] Calculate the theoretical attenuation coefficient of the potential propagation path and correlate the theoretical attenuation coefficient with the expected subband energy percentage.
[0025] For the current pulse to the candidate region of insulation degradation, the absolute values of the element-by-element differences between the expected subband energy ratio vector and the measured subband energy ratio vector of all potential propagation paths are summed to obtain the total deviation. The path with the smallest total deviation is selected as the actual propagation path of the pulse.
[0026] Furthermore, the method for determining the spatial location area of insulation degradation is as follows: for all pulses within the candidate area of insulation degradation, the starting point of the actual propagation path is traced in reverse, and all pulses within the candidate area of insulation degradation are normalized according to their respective actual propagation path lengths to obtain the normalized amplitude attenuation rate.
[0027] Calculate the standard deviation of the normalized amplitude attenuation rate of all candidate insulation degradation regions. Insulation degradation candidate regions with standard deviations exceeding the preset discrete threshold are identified as reflected signal accumulation areas and removed from the positioning results. The remaining candidate insulation degradation regions that are not removed are used as the spatial positioning regions of insulation degradation.
[0028] Furthermore, the evolutionary stages are divided as follows: the total harmonic distortion rate of the current, the amplitude of the vibration characteristic frequency, the partial discharge repetition rate, and the temperature rise rate are continuously collected at fixed time intervals.
[0029] For each parameter, a normal operating baseline value is preset, the deviation of each collected value is calculated, and the moment when the deviation of each parameter first exceeds zero is recorded as the abnormal start time. The abnormal start times of the four parameters are arranged in ascending order of time to generate an abnormal occurrence time sequence chain.
[0030] The time interval between adjacent anomaly moments in the anomaly occurrence time sequence is calculated, and the time interval is proportionally calculated with the typical development cycle of the corresponding physical process. Adjacent parameters with a calculation result less than one are determined to belong to the same evolution stage, while those with a calculation result greater than one are determined to be divided into different evolution stages.
[0031] Furthermore, the formation process of the degradation stage time series is as follows: in each evolution stage, the difference between two adjacent acquisition values of each parameter is calculated as the time series difference value, and the time series difference values are arranged in time to form a difference series.
[0032] Five consecutive difference values are summed to obtain a five-point cumulative sum. At the same time, the number of positive signs in these five difference values is counted. If the five-point cumulative sum is greater than zero and the number of positive signs is greater than or equal to four, it is determined to be an accelerated degradation stage. If the five-point cumulative sum is less than zero and the number of positive signs is less than or equal to one, it is determined to be a deceleration degradation stage.
[0033] A degradation stage sequence is generated based on the alternating order of accelerated degradation stage and decelerated degradation stage.
[0034] Furthermore, the method for determining the dominant stress type is as follows: for each accelerated degradation segment in the degradation stage sequence, the start time and end time are extracted, and the mean values of partial discharge repetition rate, total harmonic distortion rate of current, vibration characteristic frequency amplitude, and temperature rise rate are calculated within the time interval of the accelerated degradation segment.
[0035] The contribution of electrical stress is defined as the ratio of the mean of the partial discharge repetition rate to the mean of the total harmonic distortion rate of the current, and the contribution of mechanical stress is defined as the ratio of the mean of the vibration characteristic frequency amplitude to the mean of the temperature rise rate.
[0036] Calculate the difference between the contribution of electrical stress and the contribution of mechanical stress. If the difference is greater than zero, the accelerated deterioration stage is determined to be the stage dominated by electrical stress, and if the difference is less than zero, the stage is determined to be the stage dominated by mechanical stress.
[0037] Furthermore, the formation process of the fault spatiotemporal evolution path is as follows: for each accelerated degradation segment, the average amplitude decay rate at the end of the accelerated degradation segment is compared with the average amplitude decay rate at the beginning.
[0038] If the average amplitude decay rate at the end is greater than the average amplitude decay rate at the beginning, it indicates an upward trend; if the average amplitude decay rate at the end is less than the average amplitude decay rate at the beginning, it indicates a downward trend.
[0039] In the electrical stress-dominated stage, regions with an increasing average amplitude attenuation rate are selected as active regions, while in the mechanical stress-dominated stage, regions with a decreasing average amplitude attenuation rate are selected as active regions.
[0040] The active regions at each stage are arranged in chronological order to obtain the active region sequence.
[0041] A threshold for the connection distance of the insulation structure is preset. For two adjacent active regions, the straight-line distance in three-dimensional space is calculated. Adjacent regions with a distance less than the threshold are judged to have an evolutionary association.
[0042] All active regions with evolutionary correlations are concatenated in chronological order to generate a fault spatiotemporal evolution path from the initial anomalous region to the final fault region.
[0043] The beneficial effects of this invention are as follows: high positioning accuracy and strong anti-interference ability. By constructing spatial projection weights and combining them with an electromagnetic wave propagation model, the pulse source is initially located. Then, the optimal propagation path is matched by the pulse frequency division energy ratio, and the interference of reflected signals is eliminated by the dispersion of amplitude attenuation rate. This achieves accurate spatial positioning of the insulation degradation area and effectively overcomes the positioning deviation problem of traditional methods in complex electromagnetic environments.
[0044] The fault process is traceable and the evolution path is transparent. By continuously collecting four types of parameters—electrical, vibration, partial discharge, and temperature rise—a time-series chain of parameter anomalies is constructed. Based on the time interval and physical development cycle, the evolution stages are automatically divided and acceleration or deceleration degradation segments are identified, forming a complete time-series sequence of degradation stages. On this basis, the diagnostic output module dynamically matches the stress-dominant type of each stage with the spatial location area, and generates a complete spatiotemporal evolution path from the initial anomaly to the final fault, realizing the visualization and transparency of the fault development process.
[0045] The diagnostic results are highly interpretable, not only outputting the fault location but also clearly distinguishing between the electrical stress-dominated stage and the mechanical stress-dominated stage. It also visually presents the migration path of the active area in the insulation structure, providing clear physical basis for maintenance personnel to formulate differentiated maintenance strategies (such as adjusting operating voltage, strengthening mechanical fastening, or improving heat dissipation). Attached Figure Description
[0046] The invention will now be further described with reference to the accompanying drawings.
[0047] Figure 1 This is a functional module diagram of an intelligent fault diagnosis system based on power distribution equipment according to the present invention.
[0048] Figure 2 This is the logic diagram for determining the dominant stress type in this invention. Detailed Implementation
[0049] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.
[0050] Example: Please refer to Figure 1 As shown, the intelligent fault diagnosis system based on power distribution equipment described in this invention specifically includes the following modules: Feature analysis module: By synchronously acquiring the partial discharge waveform and power frequency signal of the power distribution equipment, extracting the pulse rise time and amplitude attenuation rate and calculating the pulse projection weight, determining the pulse projection point by combining the electromagnetic wave propagation model, marking the candidate areas of insulation degradation and extracting the regional features through spatial aggregation and distance judgment.
[0051] The calculation process of the pulse projection weight includes: deploying ultra-wideband high-frequency current sensors and ultra-high frequency sensors around the key insulation parts of power distribution equipment (such as transformers, switch cabinets, cable terminals, etc.). The ultra-wideband high-frequency current sensors are used to capture high-frequency pulse current signals generated by partial discharge, and the ultra-high frequency sensors are used to receive ultra-high frequency electromagnetic waves radiated by partial discharge.
[0052] The power frequency voltage signal is obtained by a voltage transformer, and the zero-crossing moment is extracted as a unified time base reference for all sensors to ensure time alignment of multi-source data. A fixed sampling frequency (e.g., 100MHz) is set to continuously capture the time domain waveform of partial discharge pulses.
[0053] The first point to explain is that pulse parameter extraction specifically involves: for each captured partial discharge pulse waveform, identifying its starting point (the point where the signal first exceeds the background noise threshold) and peak point (the point where the waveform amplitude is at its maximum), calculating the time difference between the two points, and recording it as the rise time.
[0054] The amplitude of the pulse is measured and compared with the reference sensitivity calibrated by the sensor at the factory to obtain the amplitude attenuation rate, which reflects the degree of amplitude attenuation during the transmission of the signal from the power source to the sensor.
[0055] Secondly, it should be noted that the pulse projection weight is calculated by multiplying the rise time by the amplitude decay rate.
[0056] It is understandable that the physical meaning of pulse projection weight is: a short rise time means a steep pulse, corresponding to a shorter propagation distance or fewer reflections; a large amplitude attenuation rate means severe signal attenuation, which may be related to long-distance propagation or high-loss paths. The product of the two comprehensively characterizes the relationship between the pulse source and the sensor.
[0057] The process of locating the pulse projection point includes: First, it should be noted that an electromagnetic wave propagation attenuation model is constructed, specifically: the three-dimensional space inside the power distribution equipment is discretized into a node grid with a certain resolution (e.g., 1 cm or 2 cm), and each node represents a possible power source location.
[0058] For each node, electromagnetic field simulation (such as finite element method or finite difference time-domain method) is used to calculate the theoretical attenuation characteristics that the partial discharge pulse generated from that node should have when it propagates to each sensor. The attenuation characteristics are usually expressed as attenuation curves of different frequency components, but in the initial positioning stage, they can be simplified to a single theoretical value related to the pulse projection weight.
[0059] Secondly, it should be noted that locating the pulse projection point involves establishing a mapping relationship between the three-dimensional coordinates of each node and its corresponding theoretical attenuation value, thus forming a complete model database.
[0060] For the current pulse, the pulse projection weight is compared point by point with the corresponding value in the theoretical attenuation curve of each node position in the electromagnetic wave propagation attenuation model. The deviation is calculated, and the node coordinate with the smallest deviation is selected as the pulse source projection point.
[0061] The process of spatial aggregation and distance determination includes: setting a continuous time window and collecting the pulse projection points of all pulses within the time window.
[0062] A density-based clustering algorithm is used to group spatially close pulse projection points into the same cluster. The clustering process includes: inputting the three-dimensional coordinate set of all pulse projection points; for each pulse projection point: counting the number of other points within the neighborhood radius, where the neighborhood radius determines the maximum distance between two points that are considered to be adjacent, which is set based on the device size and positioning accuracy. For example, for a 10kV transformer, it can be set to 5cm-10cm.
[0063] Pulse projection points with a number of points greater than or equal to the minimum number of points within the neighborhood radius are marked as core points. The minimum number of points is the minimum number of points required to form a cluster, used to distinguish between real discharge sources and occasional noise. It can be set to 5-10, depending on the pulse capture frequency.
[0064] For any two core points, if they are within each other's neighborhood radius, a connection is established, and a connected graph is constructed.
[0065] The interconnected core points and all boundary points in their neighborhoods are grouped into the same cluster. The remaining unclassified points are marked as noise points and discarded directly. For each cluster, the average three-dimensional coordinates of all points are calculated to obtain the aggregation center point.
[0066] Obtain a three-dimensional model of the equipment's insulation structure, predefine the geometry of each insulation boundary, such as the winding insulation layer, the oil-paper interface, and the bushing root, and calculate the shortest three-dimensional spatial distance from the aggregation center point to the boundary of each insulation layer.
[0067] If the shortest three-dimensional spatial distance is less than the preset boundary threshold, the region corresponding to the cluster is marked as a candidate region for insulation degradation, and the amplitude attenuation rate and the average rise time of all pulses in the candidate region for insulation degradation are recorded as regional feature identifiers.
[0068] It should be noted that the purpose of marking the candidate insulation degradation region is to transform the original partial discharge pulse signal into a candidate insulation degradation region with spatial location, thus providing a basis for subsequent precise positioning.
[0069] Spatial positioning module: Based on the pulse of the candidate region of insulation degradation, the optimal propagation path is matched by the pulse frequency division energy ratio, and the reflected signal region is eliminated by the dispersion of amplitude attenuation rate to determine the spatial positioning region of insulation degradation.
[0070] The optimal propagation path matching process includes: First, it should be noted that pulse energy spectrum analysis and sub-band division are specifically: for each pulse in the candidate region of insulation degradation, a fast Fourier transform is performed to obtain the full-band energy spectrum.
[0071] The frequency range is divided into three continuous sub-bands: low frequency, mid frequency, and high frequency. The specific division process is as follows: during equipment maintenance or factory testing, partial discharge pulse samples with known propagation paths are collected through manual injection or actual measurement.
[0072] Each sample is marked with its actual propagation path, such as winding end - sensor A, insulation interface - sensor B, etc.
[0073] A fast Fourier transform is performed on each pulse sample to obtain an energy spectrum vector covering the entire frequency band. The frequency resolution is determined by the sampling time and sampling rate.
[0074] Feature selection algorithms (such as recursive feature elimination and mutual information-based feature selection) are used to evaluate the contribution of each frequency point or band to path identification.
[0075] The feature selection algorithm employs mutual information-based feature selection. The specific implementation process includes: mutual information is defined as: for each frequency point f, its energy value X... f The mutual information between X and path label Y is: I(X f ;Y)=H(Y)-H(Y|X f ), where H(Y) is the entropy of the path label, H(Y|X) f) is the conditional entropy of the path label under the condition of knowing the energy at that frequency point. The greater the mutual information, the more important that frequency point is for path recognition.
[0076] Discretize the energy values of each sample at each frequency point, that is, bin them into 10 levels. Calculate the mutual information scores between each frequency point and the path label according to the mutual information calculation formula, and integrate them into a mutual information score sequence I(f). The value of each point represents the contribution degree of that frequency to path recognition, where f = 1, 2,... represents the discretized frequency points.
[0077] Plot the mutual information scores against frequency to obtain a curve of mutual information versus frequency.
[0078] Start traversing from the second frequency point of the mutual information score sequence to the second-to-last frequency point. If the frequency point f satisfies being greater than the previous frequency point and greater than the next frequency point, then the frequency point f is a peak point, and record the indices of all peak points.
[0079] Calculate the mean and standard deviation of the entire mutual information score sequence, and define the significance threshold as the mean + k × standard deviation, where k is an empirical coefficient, usually taking k = 2 or 3, and only retain the peak points with mutual information scores greater than the significance threshold.
[0080] For each retained peak, move left from the peak to find the minimum mutual information score f left , that is, the left boundary, and stop after 3 consecutive mutual information scores are lower than the significance threshold. Similarly, expand to the right to find the maximum f right , that is, the right boundary, and finally obtain the continuous frequency bands [f left , f right corresponding to each peak.
[0081] Sort all frequency bands by the left boundary f left . If the left boundary of the next frequency band ≤ the right boundary of the current frequency band + the gap threshold (the gap threshold can be taken as half of the window width), then merge them, and the right boundary of the merged one takes the maximum of the two.
[0082] Sort the merged frequency bands by the length of the frequency band, and take the two largest frequency bands [A, B], [C, D], and B < C, as the main contribution frequency bands. Then the low-frequency band is from the lowest frequency to B, the middle-frequency band is from B to C, and the high-frequency band is from C to the highest frequency.
[0083] Secondly, it should be noted that the calculation of the propagation path theory attenuation coefficient is as follows: For each insulation degradation candidate area, consider its potential propagation paths to each sensor respectively. The potential propagation paths are determined by the internal insulation structure of the equipment and must pass through several nodes, such as the oil-paper interface, winding layer interface, insulation support bars, etc.
[0084] For each potential propagation path, the number of nodes it passes through is counted, and the electromagnetic refraction coefficient of each node is queried to reflect the energy loss of the signal when it passes through the interface.
[0085] The theoretical attenuation coefficient of a potential propagation path is calculated as the sum of the products of the electromagnetic refraction coefficient of each node and the free space attenuation caused by the path length (usually inversely proportional or exponentially related to the distance). In fact, the theoretical attenuation coefficient can be correlated with the expected sub-band energy proportion. That is, different frequency components have different attenuations in different media, so the expected energy proportion of each sub-band can be estimated from the attenuation coefficient.
[0086] The electromagnetic refractive index represents the ratio of transmitted energy to incident energy when a partial discharge signal passes through an interface of different insulating media (such as oil-paper, oil-metal, air-epoxy, etc.). It is obtained by injecting a partial discharge signal of known amplitude into a real device or simulation platform, measuring the amplitude ratio received by sensors on both sides of the interface, and statistically averaging the values as the electromagnetic refractive index of that interface. For example:
[0087] |Interface type|Incident medium|Transmission medium|Refractive index range|;
[0088] |Oil-paper interface|Transformer oil|Oil-impregnated paper|0.8-0.9|;
[0089] |Oil-Metal Interface|Transformer Oil|Steel / Copper|0.05-0.15|;
[0090] |Air-Epoxy Interface|Air|Epoxy Resin|0.7-0.8|.
[0091] Thirdly, it should be noted that the optimal propagation path matching is as follows: for the candidate insulation degradation region to which the current pulse belongs, the absolute value of the element-by-element difference between the expected subband energy ratio vector and the measured subband energy ratio vector of all possible paths to each sensor is summed to obtain the total deviation, and the path with the smallest total deviation is selected as the actual propagation path of the pulse.
[0092] The process of eliminating reflected signal regions by the dispersion of amplitude attenuation rate includes: for all pulses in the candidate region of insulation degradation, their actual propagation paths have been obtained, and the starting point of the path is traced in reverse to obtain the candidate region of insulation degradation corresponding to the pulse source.
[0093] Since pulses within the same candidate region of insulation degradation may originate from the actual discharge source or from reflected signals (i.e., the discharge source is located elsewhere, and the signal is mislocated in this region after reflection), they need to be distinguished by the degree of dispersion of amplitude attenuation rate.
[0094] All pulses within the candidate region of insulation degradation are normalized according to their actual propagation path length: the normalized amplitude attenuation rate of each pulse is calculated using the formula: normalized amplitude attenuation rate = original amplitude attenuation rate of the pulse / actual propagation path length determined after optimal propagation path matching. The purpose is to eliminate the influence of propagation path length on the original amplitude attenuation rate and make different pulses comparable.
[0095] Calculate the standard deviation of all normalized amplitude attenuation rates in the region. If the standard deviation exceeds the preset discrete threshold, it indicates that the pulse amplitude attenuation rate in the region is inconsistent and may be mixed with signals or reflected signals from different real sources. Therefore, the region is identified as a reflected signal aggregation area and removed from the positioning results. The remaining insulation degradation candidate regions that have not been removed are the final spatial positioning regions for insulation degradation.
[0096] Among them, the discrete threshold is used to determine whether the normalized amplitude attenuation rate of all pulses in the candidate region is sufficiently consistent. The basis for setting it is that the normalized amplitude attenuation rate of the pulses generated by the real discharge power source should follow a normal distribution. The fluctuation mainly comes from sensor noise, positioning error and random changes in discharge intensity. According to statistical principles, the sample standard deviation usually does not exceed 10% of the mean.
[0097] It should be noted that the function of the spatial positioning region for insulation degradation is to eliminate reflection interference from the candidate regions for insulation degradation, achieve precise positioning of the degradation location, and provide a reliable spatial reference for evolution analysis.
[0098] Evolution Construction Module: Continuously collects multi-source operating parameters of power distribution equipment, constructs a parameter anomaly time sequence chain, divides the evolution stages according to the time interval and degradation cycle, determines the accelerated degradation stage and the decelerated degradation stage by parameter differential accumulation, and forms a degradation stage time sequence.
[0099] The process of constructing the abnormal time sequence of the parameters includes: real-time acquisition of total harmonic distortion of current, amplitude of vibration characteristic frequency (amplitude of characteristic frequency related to the mechanical state of equipment), partial discharge repetition rate (number of pulses per unit time), and temperature rise rate (temperature difference between adjacent time points divided by time interval).
[0100] For each parameter, a baseline value is preset for normal operation. For each parameter value collected, the deviation is calculated, and the moment when the deviation first exceeds 0 is recorded as the abnormal start time of the parameter.
[0101] Arrange the abnormal start times of the four parameters in ascending order of time to obtain the parameter abnormality time sequence chain.
[0102] The process of dividing the evolutionary stages includes: calculating the time interval between adjacent abnormal moments, and presetting a typical development cycle for each parameter. For example, the typical cycle from the appearance of partial discharge to causing a significant temperature rise is 15 days, and the occurrence cycle of vibration abnormalities may be longer.
[0103] If the time interval between adjacent abnormal moments is less than the typical development cycle of the corresponding physical process, then the two parameter anomalies belong to the same evolutionary stage; otherwise, they are classified into different evolutionary stages.
[0104] The process of determining the accelerated degradation stage and the decelerated degradation stage includes: within each evolution stage, for each parameter, performing differential processing on its time change sequence, arranging the time difference according to time, and obtaining the differential sequence.
[0105] Five consecutive difference values are summed to obtain a five-point cumulative sum. At the same time, the number of positive signs in these five difference values is counted. If the five-point cumulative sum is greater than zero and the number of positive signs is greater than or equal to four, it is determined to be an accelerated degradation stage. If the five-point cumulative sum is less than zero and the number of positive signs is less than or equal to one, it is determined to be a deceleration degradation stage.
[0106] The formation process of the degradation stage sequence includes: arranging all accelerated degradation segments and decelerated degradation segments in chronological order to form a degradation stage sequence.
[0107] It should be noted that the purpose of constructing the time series of degradation stages is to transform discrete monitoring data into degradation development stages with time logic, thereby revealing the dynamic changes in the degradation process.
[0108] Diagnostic output module: By calculating the electrical stress contribution and mechanical stress contribution of each accelerated deterioration segment, the dominant stress type is determined, and the active deterioration area of the corresponding time period is matched. The evolution correlation is judged based on the spatial distance of the area, and the spatiotemporal evolution path of the power distribution equipment fault is formed in series.
[0109] Please see Figure 2 As shown, the process for determining the dominant stress type includes: for each accelerated degradation segment in the degradation stage sequence, extracting the start time and end time, and calculating the mean value of the partial discharge repetition rate, the mean value of the total harmonic distortion rate of the current, the mean value of the vibration characteristic frequency amplitude, and the mean value of the temperature rise rate within the time interval of the accelerated degradation segment.
[0110] The maximum and minimum values of the mean partial discharge repetition rate, the mean total harmonic distortion rate of current, the mean amplitude of vibration characteristic frequency, and the mean temperature rise rate within the time window of the accelerated deterioration stage are taken respectively. The current mean values are mapped to the interval between 0 and 1 to obtain the mean of normalized partial discharge repetition rate, the mean of normalized total harmonic distortion rate of current, the mean of normalized vibration characteristic frequency amplitude, and the mean of normalized temperature rise rate.
[0111] The contribution of electrical stress is defined as the ratio of the mean of the normalized partial discharge repetition rate to the mean of the normalized total harmonic distortion rate of the current, and the contribution of mechanical stress is defined as the ratio of the mean of the normalized vibration characteristic frequency amplitude to the mean of the normalized temperature rise rate.
[0112] Calculate the difference between the contribution of electrical stress and the contribution of mechanical stress. If the difference is greater than 0, the accelerated deterioration stage is determined to be the stage dominated by electrical stress. If the difference is less than 0, it is the stage dominated by mechanical stress.
[0113] The process of matching the active degradation area corresponding to the time period includes: for each accelerated degradation segment, querying the change direction of the average amplitude attenuation rate of each insulation degradation spatial location area within the accelerated degradation segment, that is, comparing the average amplitude attenuation rate at the end of the time period with the average amplitude attenuation rate at the beginning to determine whether it is increasing or decreasing.
[0114] If the average amplitude decay rate at the end is greater than the average amplitude decay rate at the beginning, it indicates an upward trend; if the average amplitude decay rate at the end is less than the average amplitude decay rate at the beginning, it indicates a downward trend.
[0115] The matching rule is as follows: Electrical stress-dominated stage: Select the region where the average amplitude attenuation rate shows an upward trend as the active region of this stage.
[0116] Mechanical stress-dominated stage: The region where the average amplitude attenuation rate shows a downward trend is selected as the active region of this stage.
[0117] It should be noted that the logic of active region matching is as follows: when electrical stress (such as electric field distortion) increases, partial discharge intensifies, and the signal amplitude attenuation rate tends to increase (because the discharge source is stronger, but the propagation path is the same); when mechanical stress (such as vibration and wear) increases, it may lead to poor insulation contact or gap changes, causing the signal propagation path to change, and the amplitude attenuation rate may decrease. Through this correlation, the stress type is matched with the physical change trend.
[0118] The process of determining evolutionary associations based on regional spatial distance includes: arranging active regions at each stage in chronological order to obtain an active region sequence.
[0119] A threshold for the connection distance of the insulation structure is preset, based on the geometric connectivity of the internal insulation channels of the device.
[0120] For two adjacent active regions, the straight-line distance in three-dimensional space is calculated. If the straight-line distance is less than the insulation structure connectivity threshold, then the two regions are considered to have an evolutionary association.
[0121] The formation process of the fault spatiotemporal evolution path of the power distribution equipment includes: connecting all regions with evolutionary correlation in chronological order to form a complete fault spatiotemporal evolution path from the initial abnormal region to the final fault region.
[0122] It should be noted that the role of the fault spatiotemporal evolution path is to integrate the spatial location results with the temporal evolution stages to generate a traceable and explainable entire fault development process.
[0123] The technical solution and advantages of this application are as follows: By synchronously acquiring the partial discharge waveform and power frequency signal of the power distribution equipment, the pulse rise time and amplitude attenuation rate are extracted and the pulse projection weight is calculated. The pulse projection point is determined by combining the electromagnetic wave propagation model. After spatial aggregation and distance judgment, the candidate areas of insulation degradation are marked and the regional features are extracted. Based on the pulses of the candidate areas of insulation degradation, the optimal propagation path is matched by the pulse three-way frequency energy ratio, and the reflected signal area is eliminated by the dispersion of amplitude attenuation rate to determine the spatial positioning area of insulation degradation. Multi-source operating parameters of the power distribution equipment are continuously acquired to construct the parameter anomaly time sequence chain. The evolution stages are divided according to the time interval and degradation cycle. The accelerated degradation stage and the decelerated degradation stage are determined by parameter differential accumulation to form a degradation stage time sequence. The electrical stress contribution and mechanical stress contribution of each accelerated degradation stage are calculated to determine the dominant stress type and match the active degradation area of the corresponding time period. The evolution correlation is judged according to the spatial distance of the area, and the fault spatiotemporal evolution path of the power distribution equipment is formed by connecting them. This application calculates pulse projection weights, determines pulse projection points using an electromagnetic wave propagation model, marks candidate insulation degradation areas through spatial aggregation and distance determination, matches the optimal propagation path using pulse frequency division energy ratios, eliminates reflected signal interference by amplitude attenuation rate dispersion, and determines the spatial location area of insulation degradation. It continuously collects power distribution equipment parameters, constructs a parameter anomaly time sequence chain, divides evolution stages, determines acceleration and deceleration degradation segments through parameter differential accumulation, forms a degradation stage time sequence, and forms a fault spatiotemporal evolution path by determining the dominant stress type, matching active degradation areas, and judging evolutionary correlations. Through multi-source data fusion and spatiotemporal evolution analysis, this invention achieves precise location of power distribution equipment insulation degradation and transparent tracing of the fault development process.
[0124] The embodiments of the present invention have been described in detail above, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made in accordance with the scope of the present invention should still fall within the scope of the present invention.
Claims
1. A fault intelligent diagnosis system based on power distribution equipment, characterized in that: Includes the following modules: Feature analysis module: By synchronously acquiring the partial discharge waveform and power frequency signal of the power distribution equipment, the pulse rise time and amplitude attenuation rate are extracted and the pulse projection weight is calculated. Combined with the electromagnetic wave propagation model, the pulse projection point is determined. After spatial clustering and distance judgment, candidate areas of insulation degradation are marked and regional features are extracted. The pulse projection weight is calculated as follows: By deploying ultra-wideband high-frequency current sensors and ultra-high frequency sensors around key insulation parts of power distribution equipment, the time-domain waveform of partial discharge pulses is continuously captured at a fixed sampling frequency. The power frequency voltage signal is obtained by a voltage transformer, and the zero-crossing moment is extracted as a unified time base reference for all sensors. For each captured partial discharge pulse waveform, the start point and peak point are identified, the time difference between the two points is calculated and recorded as the rise time, the amplitude of the pulse is measured and compared with the reference sensitivity calibrated by the sensor at the factory to obtain the amplitude attenuation rate, and the rise time is multiplied by the amplitude attenuation rate to obtain the pulse projection weight. The method for locating the pulse projection point is as follows: The three-dimensional space inside the power distribution equipment is discretized into a node mesh, with each node representing a possible power source location; For each node, the theoretical attenuation characteristics of the partial discharge pulse generated from the node propagating to each sensor are calculated by electromagnetic field simulation. The three-dimensional coordinates of each node are mapped to their corresponding theoretical attenuation values to form a model database. For the current pulse, the pulse projection weight is compared point-by-point with the theoretical attenuation value of each node position in the model to calculate the deviation, and the node coordinates with the smallest deviation are selected as the pulse source projection point. Spatial positioning module: Based on the pulse of the candidate region of insulation degradation, the optimal propagation path is matched by the pulse frequency division energy ratio, and the reflected signal region is eliminated by the dispersion of amplitude attenuation rate to determine the spatial positioning region of insulation degradation; The optimal propagation path matching method is as follows: A fast Fourier transform is performed on each pulse in the candidate region of insulation degradation to obtain the full-band energy spectrum, and the frequency range is divided into three continuous sub-bands; For each candidate region of insulation degradation, consider the potential propagation path to each sensor, count the number of nodes passed through each potential propagation path, and query the electromagnetic wave refraction coefficient of each node. Calculate the theoretical attenuation coefficient of the potential propagation path and correlate the theoretical attenuation coefficient with the expected sub-band energy percentage; For the current pulse to the candidate region of insulation degradation, the absolute values of the element-by-element differences between the expected sub-band energy ratio vector and the measured sub-band energy ratio vector of all potential propagation paths are summed to obtain the total deviation. The path with the smallest total deviation is selected as the actual propagation path of the pulse. Evolution construction module: continuously collects multi-source operating parameters of power distribution equipment, constructs parameter anomaly time sequence chain, divides evolution stages according to time interval and degradation cycle, determines accelerated degradation stage and decelerated degradation stage by parameter differential accumulation, and forms degradation stage time sequence; Diagnostic output module: By calculating the electrical stress contribution and mechanical stress contribution of each accelerated deterioration segment, the dominant stress type is determined, and the active deterioration area of the corresponding time period is matched. The evolution correlation is determined based on the spatial distance of the area, and the spatiotemporal evolution path of the power distribution equipment fault is formed in series. The method for determining the dominant stress type is as follows: For each accelerated degradation segment in the degradation stage sequence, the start and end times are extracted, and the mean values of partial discharge repetition rate, total harmonic distortion rate of current, vibration characteristic frequency amplitude, and temperature rise rate are calculated within the time interval of the accelerated degradation segment. The contribution of electrical stress is defined as the ratio of the mean of the partial discharge repetition rate to the mean of the total harmonic distortion rate of the current, and the contribution of mechanical stress is defined as the ratio of the mean of the vibration characteristic frequency amplitude to the mean of the temperature rise rate. Calculate the difference between the contribution of electrical stress and the contribution of mechanical stress. If the difference is greater than zero, the accelerated deterioration stage is determined to be the stage dominated by electrical stress, and if the difference is less than zero, the stage is determined to be the stage dominated by mechanical stress. The formation process of the fault spatiotemporal evolution path is as follows: For each accelerated degradation segment, compare the mean amplitude decay rate at the end of the accelerated degradation segment with the mean amplitude decay rate at the beginning of the segment. If the average amplitude decay rate at the end is greater than the average amplitude decay rate at the beginning, it is an upward trend; if the average amplitude decay rate at the end is less than the average amplitude decay rate at the beginning, it is a downward trend. In the electrical stress-dominated stage, regions with an increasing average amplitude attenuation rate are selected as active regions, while in the mechanical stress-dominated stage, regions with a decreasing average amplitude attenuation rate are selected as active regions. The active regions at each stage are arranged in chronological order to obtain the active region sequence. A threshold for the connection distance of the insulation structure is preset. For two adjacent active regions, the straight-line distance in three-dimensional space is calculated. Adjacent regions with a distance less than the threshold are judged to have an evolutionary association. All active regions with evolutionary correlations are concatenated in chronological order to generate a fault spatiotemporal evolution path from the initial anomalous region to the final fault region.
2. The intelligent fault diagnosis system based on power distribution equipment according to claim 1, characterized in that: The process of marking candidate regions of insulation degradation and extracting region features is as follows: A continuous time window is set to collect the pulse projection points of all pulses within the time window. A density-based clustering algorithm is used to group spatially close pulse projection points into the same cluster and calculate the cluster center point. A three-dimensional model of the equipment insulation structure is obtained and the geometric surfaces of each insulation boundary are predefined. The shortest three-dimensional spatial distance from the cluster center point to the boundary of each insulation layer is calculated. The regions corresponding to clusters with distances less than a preset boundary threshold are marked as candidate regions for insulation degradation. The average amplitude attenuation rate and average rise time of all pulses within the candidate regions for insulation degradation are recorded as regional feature identifiers.
3. The intelligent fault diagnosis system based on power distribution equipment according to claim 1, characterized in that: The method for determining the spatial location area of insulation degradation is as follows: For all pulses within the candidate region of insulation degradation, trace back to the starting point of the actual propagation path, and normalize all pulses within the candidate region of insulation degradation according to their respective actual propagation path lengths to obtain the normalized amplitude attenuation rate. Calculate the standard deviation of the normalized amplitude attenuation rate of all candidate insulation degradation regions. Insulation degradation candidate regions with standard deviations exceeding the preset discrete threshold are identified as reflected signal accumulation areas and removed from the positioning results. The remaining candidate insulation degradation regions that are not removed are used as the spatial positioning regions of insulation degradation.
4. The intelligent fault diagnosis system based on power distribution equipment according to claim 1, characterized in that: The evolutionary stages are divided as follows: The total harmonic distortion rate of the current, the amplitude of the vibration characteristic frequency, the partial discharge repetition rate, and the temperature rise rate are continuously collected at fixed time intervals. For each parameter, a normal operating baseline value is preset, the deviation of each collected value is calculated, and the moment when the deviation of each parameter first appears to be greater than zero is recorded as the abnormal start time. The abnormal start times of the four parameters are arranged in ascending order of time to generate an abnormal occurrence time sequence chain. The time interval between adjacent anomaly moments in the anomaly occurrence time sequence is calculated, and the time interval is proportionally calculated with the typical development cycle of the corresponding physical process. Adjacent parameters with a calculation result less than one are determined to belong to the same evolution stage, while those with a calculation result greater than one are determined to be divided into different evolution stages.
5. The intelligent fault diagnosis system based on power distribution equipment according to claim 4, characterized in that: The formation process of the degradation stage time sequence is as follows: Within each evolution stage, the difference between two consecutive acquisition values of each parameter is calculated as the temporal difference value, and the temporal difference values are arranged in time to form a difference sequence; Take five consecutive difference values and sum them up to get a five-point cumulative sum. At the same time, count the number of positive signs in these five difference values. If the five-point cumulative sum is greater than zero and the number of positive signs is greater than or equal to four, it is determined to be an accelerated deterioration stage. If the five-point cumulative sum is less than zero and the number of positive signs is less than or equal to one, it is determined to be a deceleration deterioration stage. A degradation stage sequence is generated based on the alternating order of accelerated degradation stage and decelerated degradation stage.