Distribution line fault intelligent diagnosis method based on multi-source information fusion

Abstract

The invention discloses a distribution line fault intelligent diagnosis method based on multi-source information fusion, and belongs to the technical field of power system fault detection and diagnosis, and the method comprises the steps: obtaining distribution line data and meteorological data; superposing the distribution line data with the meteorological data to generate a space-time fusion feature matrix based on the wire space displacement; dynamically adjusting a preset initial threshold value by using the space-time fusion feature matrix; and performing multi-round verification on the abnormal fluctuation of the electrical parameters of the line according to the fault judgment threshold, generating a fault positioning instruction, performing real-time matching with the topological structure of the line, correcting the priority queue of the fault section based on the switching action feedback data of the feeder terminal, and triggering the fault removal action matched with the priority. According to the method, spatial-temporal feature fusion of current, voltage, weather and spatial displacement data and a dynamic threshold adjustment mechanism are adopted, the fault type can be accurately identified, the fault section can be positioned in real time, and the diagnosis precision and the response speed in a complex environment are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system fault detection and diagnosis, and in particular to a distribution line fault intelligent diagnosis method based on multi-source information fusion. Background Art

[0002] Traditional distribution line fault diagnosis methods mainly rely on single electrical signal thresholds such as current and voltage, which have the problem of insufficient data dimensions. When the line is disturbed by changes in ambient temperature and humidity or mechanical deformation, fixed thresholds are prone to misjudgment. For example, the fluctuation of conductor resistance due to temperature may lead to false overcurrent alarms, and strong winds may cause conductor swings to be mistakenly identified as short circuit faults.

[0003] In the prior art, the abnormality is usually handled by comparing regular inspections with preset rule bases. For example, offline fault recording data is used to match typical waveforms, or the status of fault indicators is manually verified. Such methods need to rely on historical experience data, and the response delay is large, which cannot adapt to the needs of rapid analysis of dynamic changes in line topology and transient faults.

[0004] Traditional solutions have obvious defects when facing multi-factor coupled faults. Hidden defects such as mechanical deformation and insulation aging are difficult to identify in time through a single electrical parameter, and sudden changes in meteorological conditions may mask early fault characteristics. Manual inspections are inefficient and cannot cover real-time monitoring, resulting in a long troubleshooting cycle and affecting power supply reliability. Summary of the invention

[0005] To solve the above problems, the present invention provides an intelligent diagnosis method for distribution line faults based on multi-source information fusion. It adopts the spatiotemporal feature fusion and dynamic threshold adjustment mechanism of current, voltage, meteorological and spatial displacement data, which can accurately identify the fault type and locate the fault section in real time, significantly improving the diagnosis accuracy and response speed in complex environments.

[0006] The above objectives can be achieved through the following solutions: The invention discloses an intelligent fault diagnosis method for distribution lines based on multi-source information fusion, comprising obtaining current signal, voltage signal and temperature signal of distribution lines, and synchronously collecting humidity gradient change rate in meteorological data and conductor spatial displacement in geographic coordinate data to generate a multi-dimensional heterogeneous data set; performing entropy value calculation on the time domain distortion characteristics of the current signal and the voltage signal, and superimposing the humidity gradient change rate to generate a dynamic entropy change parameter, and performing spatial morphology calibration on the dynamic entropy change parameter based on the conductor spatial displacement to generate a spatiotemporal fusion feature matrix; dynamically adjusting a preset initial threshold value by using the spatiotemporal fusion feature matrix; wherein, the adjustment The process includes calculating the conductor resistance change rate according to the real-time temperature data in the meteorological data, and performing weighted compensation on the resistance change rate through a sliding window algorithm to generate an adaptive fault discrimination threshold; performing multiple rounds of verification on the abnormal fluctuation of the electrical parameters of the line according to the fault discrimination threshold, and generating a fault location instruction with a confidence weight when the fluctuation amplitude of three consecutive sampling cycles exceeds the fault discrimination threshold and the proportion of high-frequency harmonics reaches a preset threshold; matching the fault location instruction with the line topology in real time, and correcting the fault section priority queue based on the switch action feedback data of the feeder terminal to trigger a fault removal action matching the priority.

[0007] Optionally, generating the spatiotemporal fusion feature matrix includes: extracting the time domain waveform distortion characteristics of the current signal and the third harmonic mutation rate of the voltage signal, and calculating the composite distortion entropy value of the two; superimposing the logarithmic attenuation coefficient of the humidity gradient change rate on the composite distortion entropy value to generate a dynamic entropy change parameter; performing morphological corrosion processing on the dynamic entropy change parameter based on the fluctuation trajectory of the spatial displacement of the conductor, and outputting the spatiotemporal fusion feature matrix.

[0008] Optionally, generating an adaptive fault discrimination threshold includes: obtaining historical mean data of the conductor resistance change rate, and fitting a resistance change curve based on real-time temperature data; combining the spatiotemporal fusion feature matrix, using a dynamic Bayesian network to identify abnormal segments of the resistance change curve, and generating a compensation weight factor; performing a convolution operation on the compensation weight factor and the initial threshold to generate a fault discrimination threshold that includes a dynamic coupling relationship between environmental parameters.

[0009] Optionally, the multiple rounds of verification of abnormal fluctuations in electrical parameters of the line according to the fault judgment threshold include: when the first fluctuation amplitude exceeds the fault judgment threshold, collecting hydrogen gas integral score data of the distribution transformer oil chromatogram; if the hydrogen gas integral score exceeds a preset fractional threshold and the high-frequency harmonic proportion reaches a threshold threshold, triggering a partial discharge prediction signal; and adjusting the fault judgment thresholds of the subsequent two sampling cycles according to the prediction signal.

[0010] Optionally, the generation of a fault location instruction with a confidence weight includes: obtaining the switch status data of the feeder terminal, and reversely inferring the fault current propagation path based on the switch status data; performing an overlap analysis on the propagation path and the abnormal area of ​​the time-space fusion feature matrix, and increasing the confidence weight when the overlapping area ratio is greater than a preset ratio threshold; if there are multiple candidate sections, calculating the correlation coefficient between the rate of change of the dynamic entropy change parameter and the spatial displacement of the conductor, and selecting the section according to the size of the correlation coefficient.

[0011] Optionally, the method also includes: collecting the line capacitance current difference before and after the fault is removed, and performing similarity matching with the standard deviation value in a preset fault feature database; generating a fault type label based on the matching result, associating the fault type label with the environmental parameters and adding it to a preset historical fault case library; and updating the subsequent initial threshold default value based on the fault frequency distribution at the same geographic coordinates in the historical fault case library.

[0012] Optionally, the subsequent update of the initial threshold default value includes: extracting the environmental characteristic parameters and action delay data corresponding to the fault type label to generate a four-dimensional space-time feature vector; using a time series analysis algorithm to perform trend prediction on the four-dimensional space-time feature vector to generate a dynamic threshold recommended value within a future preset time period; after superimposing the dynamic threshold recommended value with the real-time meteorological warning data, the final initial threshold default value is determined by the gradient descent method.

[0013] Optionally, the method for acquiring the spatial displacement of the conductor includes: scanning the spatial coordinates of the distribution line corridor in real time through a laser radar to generate a three-dimensional displacement trajectory of the conductor suspension point; performing time synchronization matching on the displacement mutation of the three-dimensional displacement trajectory and the fluctuation period of the space-time fusion feature matrix; when the correlation coefficient between the displacement mutation and the electrical parameter fluctuation is greater than a preset correlation threshold, triggering a high-risk warning signal for line mechanical deformation.

[0014] Optionally, the high-risk warning signal that triggers mechanical deformation of the line includes: inputting the time and space coordinates corresponding to the high-risk warning signal into a preset insulator aging prediction model to calculate the residual mechanical strength decay rate; adjusting the inspection cycle of the section according to the decay rate, and synchronously increasing the threshold value.

[0015] Based on the same inventive concept, the present invention also provides an intelligent diagnosis system for distribution line faults based on multi-source information fusion, the system comprising: a data acquisition module, used to obtain current signals, voltage signals and temperature signals of the distribution line, and synchronously collect the humidity gradient change rate in the meteorological data and the conductor spatial displacement in the geographic coordinate data to generate a multi-dimensional heterogeneous data set; a feature construction module, used to calculate the entropy value of the time domain distortion characteristics of the current signal and the voltage signal, and superimpose the humidity gradient change rate to generate a dynamic entropy change parameter, and at the same time perform spatial morphology calibration on the dynamic entropy change parameter based on the conductor spatial displacement to generate a spatiotemporal fusion feature matrix; an initial threshold adjustment module, used to use the spatiotemporal fusion feature matrix to adjust the preset The initial threshold is dynamically adjusted; wherein the adjustment process includes calculating the conductor resistance change rate according to the real-time temperature data in the meteorological data, and weightedly compensating the resistance change rate through a sliding window algorithm to generate an adaptive fault judgment threshold; a fault judgment module is used to verify the abnormal fluctuation of the electrical parameters of the line for multiple rounds according to the fault judgment threshold, and when the fluctuation amplitude of three consecutive sampling cycles exceeds the fault judgment threshold and the proportion of high-frequency harmonics reaches a preset threshold, a fault location instruction with a confidence weight is generated; a fault processing module is used to match the fault location instruction with the line topology in real time, and correct the fault section priority queue based on the switch action feedback data of the feeder terminal, and trigger a fault removal action matching the priority.

[0016] Compared with the prior art, the present invention has the following advantages: 1. The present invention improves the accuracy and efficiency of distribution line fault diagnosis; by integrating electrical parameter data such as current, voltage, temperature and meteorological information, a dynamic entropy change parameter and time-space fusion feature matrix are established, which solves the limitation of a single data source and can effectively distinguish environmental interference from real fault signals; the high-frequency harmonic detection and humidity compensation mechanism combined with multi-dimensional data analysis improves the anti-interference ability under complex meteorological conditions such as humidity and high temperature, and significantly reduces the misjudgment rate; 2. The threshold adaptive adjustment method combining sliding window algorithm and dynamic Bayesian network is adopted to realize real-time compensation of the change rate of wire resistance, overcoming the problem of lack of sensitivity of traditional fixed threshold due to temperature fluctuation; through three sampling cycle verification and confidence weight determination, the reliability and real-time performance of fault identification are ensured, avoiding erroneous actions caused by instantaneous disturbances; 3. Through real-time matching of fault location instructions with line topology, combined with switch feedback data at the feeder terminal, the fault section priority queue is dynamically corrected; this method optimizes the fault removal logic and shortens the positioning time, especially in complex branch line scenarios. It can accurately lock the fault point and give priority to removing high-risk sections, thereby improving the power supply restoration speed.

[0017] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a flow chart of a method for intelligent diagnosis of distribution line faults based on multi-source information fusion according to an embodiment of the present invention.

[0020] Figure 2 It is a heat map of the spatiotemporal fusion feature matrix of an embodiment of the present invention.

[0021] Figure 3 is a curve diagram of the dynamic threshold adjustment process of an embodiment of the present invention.

[0022] Figure 4 It is a three-dimensional wire displacement trajectory diagram of an embodiment of the present invention.

[0023] Figure 5 It is a structural schematic diagram of a distribution line fault intelligent diagnosis system based on multi-source information fusion according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] Reference Figure 1 An embodiment of the present invention proposes an intelligent diagnosis method for distribution line faults based on multi-source information fusion, which adopts the spatiotemporal feature fusion and dynamic threshold adjustment mechanism of current, voltage, meteorological and spatial displacement data, can accurately identify the fault type and locate the fault section in real time, and significantly improve the diagnosis accuracy and response speed in complex environments.

[0026] The method of this embodiment specifically includes: Obtain the current signal, voltage signal and temperature signal of the distribution line, and simultaneously collect the humidity gradient change rate in the meteorological data and the conductor spatial displacement in the geographic coordinate data to generate a multi-dimensional heterogeneous data set; Specifically, the humidity gradient change rate is the rate of change of ambient humidity per unit time, reflecting the risk of condensation on the surface of the insulating material. The wire spatial displacement is the amplitude of the position change of the wire in three-dimensional space, which represents the mechanical stability of the line under the action of external forces. The heterogeneous data set is a time-aligned multi-physics field data set covering four dimensions: electrical, thermal, wet, and mechanical.

[0027] The time domain distortion characteristics of the current signal and the voltage signal are entropy calculated, and the humidity gradient change rate is superimposed to generate a dynamic entropy change parameter. At the same time, the dynamic entropy change parameter is spatially calibrated based on the wire spatial displacement to generate a time-space fusion feature matrix; Specifically, the time domain distortion feature is the degree to which the waveform deviates from the standard sine wave, which is used to quantify the current distortion caused by overload or short circuit. The entropy value calculation reflects the signal complexity through information entropy, effectively distinguishing between steady-state load and fault transient. The dynamic entropy change parameter is a composite indicator that integrates the influence of electrical anomalies and environmental humidity. Figure 2 As shown in Figure 2, the spatiotemporal fusion feature matrix shows the changes of dynamic entropy parameters with time and spatial displacement through color mapping. The light-colored area indicates a high correlation between electrical anomalies and environmental factors.

[0028] The preset initial threshold is dynamically adjusted using the time-space fusion feature matrix; wherein the adjustment process includes calculating the conductor resistance change rate according to the real-time temperature data in the meteorological data, and performing weighted compensation on the resistance change rate through a sliding window algorithm to generate an adaptive fault discrimination threshold; Specifically, the sliding window algorithm performs moving average processing on the resistance changes in continuous time periods to smooth the impact of instantaneous temperature fluctuations. The adaptive fault discrimination threshold is a dynamic threshold value that changes with temperature to avoid the failure of the fixed threshold in extreme weather. Figure 3 As shown, after the initial threshold is adjusted by the sliding window algorithm, the fault judgment criterion is accurately triggered in the 12-15 hour period to avoid misjudgment of transient disturbances.

[0029] Perform multiple rounds of verification on abnormal fluctuations of the electrical parameters of the line according to the fault discrimination threshold, and generate a fault location instruction with a confidence weight when the fluctuation amplitude of three consecutive sampling cycles exceeds the fault discrimination threshold and the proportion of high-frequency harmonics reaches a preset threshold; The fault location instruction is matched with the line topology in real time, and the fault section priority queue is corrected based on the switch action feedback data of the feeder terminal, triggering a fault removal action matching the priority.

[0030] Specifically, the suspected fault node in the positioning instruction is matched with the line topology of the SCADA system by the shortest path, and the three candidate sections with the closest distance are selected first. The switch opening and closing status uploaded by the feeder terminal is received, and if the adjacent switch of a section has been operated, its priority is reduced. Finally, the section with the highest priority is selected to trigger the circuit breaker to open, and a positioning report is sent to the operation and maintenance terminal.

[0031] The fault diagnosis model is constructed through the fusion of spatiotemporal multi-dimensional data, and the threshold is dynamically adjusted to adapt to complex environmental changes. The coupled calibration of electrical parameters and mechanical displacement effectively suppresses interference signals, and the humidity data correction improves the sensitivity to insulation moisture faults. After implementation, the fault identification accuracy can be improved, the positioning time can be shortened, and the risk of false operation in extreme weather conditions can be reduced.

[0032] Optionally, generating a spatiotemporal fusion feature matrix includes: Extract the time domain waveform distortion characteristics of the current signal and the third harmonic mutation rate of the voltage signal, and calculate the composite distortion entropy value of the two; Specifically, firstly, the original current signal of the distribution line is collected through the current transformer on the high-voltage side, and the signal is input into the bandpass filter to remove low-frequency interference other than the 50Hz power frequency. The time domain waveform analysis of the filtered current signal is performed using a differential operator, and the distortion amplitude of each sampling point and the change in the time interval between adjacent peaks and troughs are calculated to obtain the current time domain distortion characteristics. At the same time, the voltage signal collected by the voltage transformer is subjected to a fast Fourier transform to extract the ratio of the amplitude of the third harmonic component to the fundamental wave, and the rising slope of the ratio per unit time is used as the third harmonic mutation rate. Composite distortion entropy value The calculation is achieved through the following formula: , where is the number of sampling points, For the The deviation between the instantaneous current amplitude at each sampling point and the standard sine wave, is the effective value of current, is the time-varying derivative of the third harmonic voltage. This formula combines the energy distribution of the current waveform distortion with the dynamic characteristics of the voltage harmonic change to characterize the degree of abnormal power quality of the line.

[0033] The composite distortion entropy value is superimposed on the logarithmic attenuation coefficient of the humidity gradient change rate to generate a dynamic entropy change parameter; The composite distortion entropy value is superimposed on the logarithmic attenuation coefficient of the humidity gradient change rate to generate a dynamic entropy change parameter; Specifically, the time series data of the environmental humidity sensor is collected, the absolute value of the humidity change rate in the past 30 minutes is calculated, and the humidity gradient change curve is generated by piecewise linear interpolation. The natural logarithm function is used to perform nonlinear scaling on the humidity gradient to generate the logarithmic attenuation coefficient. ,have: , where is the humidity change rate. The coefficient is proportionally superimposed with the composite distortion entropy value to obtain the dynamic entropy change parameter. ,have: , where is the adjustment coefficient of the humidity change rate. This operation maintains the original entropy value to reflect the characteristics of electrical parameters, and adds dynamic correction of the effect of humidity change on the conductivity of the insulating material, effectively improving the sensitivity to surface leakage current in high humidity scenarios.

[0034] The dynamic entropy change parameter is subjected to morphological corrosion processing based on the fluctuation trajectory of the spatial displacement of the wire, and a spatiotemporal fusion feature matrix is ​​output.

[0035] Specifically, the three-dimensional coordinates of the wire suspension point are recorded every 0.1 second using a laser radar scanning device, and the spatial displacement components of adjacent time points are calculated by difference to generate the wire fluctuation trajectory. The dynamic entropy change parameters are mapped to the displacement trajectory points according to the sampling time, and a time-domain spatial joint distribution matrix is ​​constructed. The 3×3 structure check matrix in the morphological corrosion algorithm is traversed and processed. The specific operations include: for each matrix element, it is compared with the values ​​of the eight adjacent positions, and the element value is replaced with the minimum value in the neighborhood. After corrosion, isolated noise peaks (such as abnormal entropy values ​​caused by wire swings caused by short-term strong winds) are suppressed, and the characteristic areas consistent with the physical deformation trend of the line are retained, and finally a time-space fusion feature matrix is ​​generated.

[0036] For example, monitoring equipment is installed on a 10kV distribution line in a coastal typhoon-prone area. When the instantaneous wind speed reaches 15m / s, the conductor will produce spatial displacement due to violent swinging, and the strong wind carrying salt mist will cause a sudden increase in humidity. The current transformer detects the current distortion caused by the sudden increase in load, and the voltage transformer records that the amplitude of the third harmonic increases from 1% to 5% within 2 seconds. The current waveform distortion feature extraction shows that the distortion rate is 12%, the voltage third harmonic mutation rate is 4% / s, the composite distortion entropy value is calculated to be 2.3, the ambient humidity increases from 60% to 95% within 5 minutes, the logarithmic attenuation coefficient of the humidity gradient change rate is 0.8, and the corrected dynamic entropy change parameter increases to 2.3×1.24≈2.85. The laser radar measured that the horizontal displacement trajectory of the conductor fluctuates periodically in the same period. After morphological corrosion treatment to eliminate accidental displacement spikes, only the displacement area synchronized with the current harmonic surge is retained in the space-time matrix. This method effectively distinguishes the conductor swing interference caused by typhoons from actual insulation degradation faults. Traditional methods would misjudge current distortion caused by strong winds as short circuit faults, but through humidity correction and spatial corrosion processing, the system filters out the instantaneous disturbance effect of salt fog humidity on electrical parameters, and confirms the real fault based on the time synchronization of displacement trajectory and harmonic changes. Operators can accurately locate a section of line with reduced insulation due to salt fog corrosion, avoiding the waste of resources caused by power outages for maintenance.

[0037] Optionally, generating an adaptive fault discrimination threshold comprises: Obtain historical average data of the wire resistance change rate, and fit the resistance change curve based on the real-time temperature data; Specifically, retrieve the wire resistance value and corresponding ambient temperature records of the target line in the last 30 days from the cloud storage module, calculate the correlation coefficient between the daily average temperature and the resistance fluctuation, and generate the historical resistance change mean sequence. ,in, is the number of historical resistance change means, is the temperature range (each 5℃ is a segment), In the same temperature range The offset of the wire resistance relative to the nominal value. Real-time data from the wire surface temperature sensor is collected, and the resistance change rate is calculated in real time according to the resistance temperature coefficient formula of the wire material. Resistance change rate at time ,have: , in, The wire is at the reference temperature The nominal resistance value under is the temperature coefficient of resistance, for The temperature of the wire surface at the moment. The least square method is used to calculate the real-time The temperature-resistance change cubic polynomial curve is fitted with the historical mean value to predict the theoretical change range of the wire resistance under the current environment.

[0038] Combined with the time-space fusion feature matrix, a dynamic Bayesian network is used to identify abnormal sections of the resistance change curve and generate compensation weight factors; Specifically, the spatiotemporal fusion feature matrix is ​​input into the dynamic Bayesian network in a time slice manner. The network contains hidden state nodes (indicates normal / abnormal status of the wire) and observation node (Consists of parameters such as the entropy value of electrical parameters and displacement correlation in the space-time matrix). The state transition probability is defined as: , where is the first probability value, is the second probability value. Observation probability By calculating the deviation between the data of each dimension of the space-time matrix and the current resistance change curve, Deviation between the data of each dimension of the time-space matrix and the current resistance change curve ,have:

[0039] , In the formula, For Dimension The expected value of is used to calculate the deviation between the observed value and the expected value. for The dynamic entropy change parameter at each moment, It is a cubic polynomial curve function of temperature-resistance change.

[0040] like If it is abnormal, then: , In the formula, is the sensitivity coefficient, is a natural constant. According to the Viterbi algorithm, the maximum probability state sequence is inferred when there are three consecutive time slices. When it is abnormal, mark the resistance change curve as an abnormal section and generate a compensation weight factor : , In the formula, is the initial time of the abnormal segment, is the end time of the abnormal segment, For the moment from arrive The definite integral of .

[0041] The compensation weight factor is convolved with the initial threshold to generate a fault discrimination threshold including a dynamic coupling relationship of environmental parameters.

[0042] Specifically, enter the initial threshold (determined by 120% of the line rated current), and convolved with the compensation weight factor, Fault discrimination threshold at time ,have: , In the formula, Relative to the current time slice The offset of is the Gaussian kernel function, is the sliding window width. This operation weights and fuses the thresholds in the current time window and its surrounding time windows with the weight factor, so that the thresholds dynamically reflect the coupling effects of the abnormal wire resistance and environmental parameters (temperature, humidity).

[0043] For example, when a cold wave hits the north in winter, the temperature of a 110kV line drops from -5°C to -20°C within 1 hour. Through historical data, it is found that the average resistance of the line at -20°C is 8% higher than the standard value, and the fitting curve predicts 7.5%. However, the actual sensor measured 14%, which deviates from the predicted value. The time-space fusion matrix shows that the increase in current entropy value is abnormal and the displacement of the conductor is stable (no signs of icing). The dynamic Bayesian network determines that the resistance change in this section is inconsistent with the normal expansion caused by low temperature, triggering the abnormal generation of compensation weight factors . Initial threshold After convolution operation, it is adjusted to , 0.95 is the Gaussian kernel function value, which significantly improves the sensitivity of overcurrent detection in this section. The real reason for the abnormal increase in wire resistance during the cold wave is that the oxidation of a certain joint causes the contact resistance to increase, rather than a simple drop in temperature. The traditional method relies solely on temperature compensation to set the threshold to 2300A, which may not be able to detect the hidden danger (the actual fault current is 2500A). After using the dynamic Bayesian network to identify the irreversible anomaly of resistance change, the system accurately adjusts the threshold to 2593A, sensitively triggers the alarm and locates the hidden danger point. This avoids fuse accidents caused by overheating of the joint, and effectively distinguishes the different impact modes of environmental factors and equipment degradation.

[0044] Optionally, the performing multiple rounds of verification on abnormal fluctuations of electrical parameters of the line according to the fault discrimination threshold comprises: When the first fluctuation amplitude exceeds the fault discrimination threshold, collecting hydrogen gas volume fraction data of the distribution transformer oil chromatogram; Specifically, when the electrical parameter fluctuation breaks through the fault discrimination threshold after dynamic adjustment for the first time, the system immediately triggers the oil chromatography online monitoring device to start gas sampling. The hydrogen dissolved in the transformer oil is precipitated by an oil-gas separation membrane, and its volume fraction in the mixed gas is measured by a gas sensor. This process continuously collects data for three consecutive cycles, and the median is taken as the current hydrogen gas integral score. If at least two of the three consecutive samplings have a hydrogen gas integral score greater than the score threshold, it is determined to be an abnormal state.

[0045] If the hydrogen gas integral exceeds the preset fractional threshold and the high-frequency harmonics account for the preset threshold, a partial discharge prediction signal is triggered; Specifically, the current hydrogen gas volume is With preset threshold By comparison, the total energy proportion of high-frequency harmonics above 2kHz in the voltage signal is calculated by Fourier transform. Assume that the threshold 5%, if satisfied and , generate partial discharge prediction signal. Prediction signal strength The classification is based on the following formula: , In the formula, is the minimum value function. The high-frequency harmonic proportion is the ratio of the high-frequency component to the total harmonic distortion, reflecting the change in the spectrum characteristics caused by the discharge pulse. The partial discharge prediction signal is a risk level indicator that characterizes the early discharge activity caused by insulation defects.

[0046] The fault discrimination thresholds of two subsequent sampling cycles are adjusted according to the prediction signal.

[0047] Specifically, an exponential decay model is used to dynamically modify the subsequent threshold. Corrected fault discrimination threshold at time ,have: , where is an adjustment factor, which can be 0.15. This adjustment allows the system to lower the threshold when a partial discharge risk is detected, increasing sensitivity to weak fault features. The exponential decay model is a mathematical correction method that proportionally reduces the threshold based on the predicted signal strength, ensuring that the system enters a highly sensitive monitoring mode in high-risk scenarios.

[0048] For example, the oil chromatographic monitoring of a transformer in a suburban substation shows that the hydrogen gas volume fraction is continuously between 0.6% and 0.7%, and the voltage high-frequency harmonics account for 7%. Exceeding the standard; high-frequency harmonics account for 7% and exceed the 5% threshold, triggering a partial discharge prediction signal The original dynamic threshold of 300A was adjusted to 300×(1-0.15×1)=255A. In the subsequent two samplings, an abnormal current with an amplitude of 265A was captured, which was confirmed to be a casing surface creepage fault. The hidden creepage fault that cannot be detected by the traditional current threshold is identified in advance to avoid power outages of the entire station caused by insulation breakdown.

[0049] 5 Optionally, the generating of the fault location instruction with confidence weight includes: Acquiring switch status data of the feeder terminal, and reversely calculating the fault current propagation path based on the switch status data; Specifically, the current open and close status and action timestamp of each section switch are obtained through the FTU (feeder terminal unit), and the adjacency matrix is ​​established in combination with the line topology. Using the reverse path tracing algorithm: , in, is the adjacency matrix, is the fault current vector. The algorithm recursively searches upstream from the end node where the fault current is detected to generate a set of possible propagation paths The reverse path tracing algorithm is a graph theory method that reversely analyzes the possible sources of fault current through switch states. The adjacency matrix is ​​a two-dimensional Boolean array that describes the connection relationship between line nodes.

[0050] Performing overlap analysis on the propagation path and the abnormal area of ​​the spatiotemporal fusion feature matrix, and increasing the confidence weight when the overlapping area ratio is greater than a preset ratio threshold; Specifically, the coordinates of the area where the entropy value exceeds the safety critical value in the time-space fusion feature matrix are extracted and mapped to the line topology coordinate system to form a polygonal anomaly area. The spatial overlap between each propagation path and the polygonal anomaly area is calculated. Spatial overlap of the propagation paths ,have:

[0051] In the formula, No. The propagation paths, is the polygonal abnormal area. For the proportion threshold , can be 70%, if , then increase the confidence weight of the path, for the The confidence weight of the propagation path ,have: , In the formula, is the adjustment coefficient of the propagation path, the overlap area ratio is a quantitative indicator of the geometric overlap between the path and the abnormal area, and represents the spatial correlation of the fault location. The confidence weight is a dynamic weighting coefficient based on the overlap, which is used for priority sorting.

[0052] If there are multiple candidate sections, the correlation coefficient between the change rate of the dynamic entropy change parameter and the wire spatial displacement is calculated, and the section is selected according to the size of the correlation coefficient.

[0053] Specifically, there are multiple candidate sections, that is, there are candidate paths with the same weight, and the correlation coefficient between the wire spatial displacement of the corresponding section of each path and the change rate of the dynamic entropy change parameter of the electrical parameter is calculated. Correlation coefficient of candidate paths ,have: , In the formula, is the covariance function, For the The wire spatial displacement of the corresponding segment of the candidate path, is the rate of change of the dynamic entropy change parameter of the electrical parameter, is the standard deviation of the conductor spatial displacement, is the standard deviation of the rate of change of the dynamic entropy change parameter of the electrical parameter, select The largest section is the final positioning target, and its priority is raised to the top of the queue. The correlation coefficient is a statistical indicator to measure the synchronization between the mechanical deformation of the conductor and the electrical anomaly, which prevents the false fault path caused by the switch misoperation from being selected. Avoid positioning deviation caused by a single criterion, integrate space and physical characteristics to lock the real fault source, and reduce invalid inspections.

[0054] Optionally, the method further comprises: Collect the line capacitance current difference before and after the fault is removed, and perform similarity matching with the standard deviation value in the preset fault feature database; Specifically, after the line is powered off, the capacitor current transformer is used to record the capacitor current waveforms of the 5 cycles before and 10 cycles after the fault is removed, and the difference is calculated by integration. ,have: , in, is a standard time window, such as a 0.1 second time window, to is the period before fault removal, to The post-resection period, is the capacitor current waveform before fault removal, is the capacitor current waveform after the fault is removed. The nominal values ​​of each fault type in the database, such as short circuit, grounding, and disconnection Perform cosine similarity calculation, for similarity ,have: , If the similarity When it is greater than the preset threshold, it is determined to be the same type of fault mode. The capacitance current difference is the energy change of the capacitive current before and after the fault, reflecting the change in the line charge distribution state. The cosine similarity is used to quantify the matching degree between the current difference and the historical mode.

[0055] Generate a fault type label according to the matching result, associate the fault type label with the environmental parameter and then add it to a preset historical fault case library; Specifically, if the capacitor current difference matches the short-circuit nominal value, a "short-circuit line fault" label is generated, and the label is combined with the ambient temperature, humidity, and wind speed data at the time of the fault into a tuple. This tuple is stored in the NoSQL database of the historical case library by timestamp, and a sharding storage strategy is adopted to improve retrieval efficiency. The fault type label is a classification code that identifies the specific physical cause of the fault. Environmental parameter association is to bind meteorological conditions and fault types for storage for subsequent pattern analysis.

[0056] Based on the fault frequency distribution at the same geographic coordinates in the historical fault case library, the subsequent initial threshold default value is updated.

[0057] Specifically, the line area is divided into geographical grids, and the average monthly occurrence of similar faults in each grid is counted. The initial threshold is adjusted using the exponentially weighted moving average model. , for the updated initial threshold ,have: , In the formula, is the attenuation coefficient of the average monthly occurrence frequency, is the average number of occurrences per month, is a natural constant.

[0058] For example, a coastal industrial area grid has experienced 6 "wet flashover grounding faults" in the past three months. Calculate the capacitor current difference Wet flash nominal value Similarity , the match is successful; when adding tags, the humidity is 92% and the temperature is 28℃. Update the initial threshold every month to get When a similar fault occurs the next month, the alarm is triggered 200ms in advance due to the lower threshold. This significantly improves the fault warning speed in high-incidence areas and avoids the expansion of faults caused by long-term exposure of equipment to corrosive and humid environments.

[0059] Optionally, the updating of the subsequent initial threshold default value includes: Extracting environmental characteristic parameters and action delay data corresponding to the fault type label to generate a four-dimensional spatiotemporal characteristic vector; Specifically, the environmental parameters and action delays corresponding to the fault labels are extracted from the historical case library to construct a vector. The environmental parameters include temperature, humidity, and wind speed. The action delay is the time from detection to removal. After normalization, they are sorted in time series to generate a spatiotemporal feature matrix. The action delay is an indicator of system response efficiency, reflecting the action speed of the protection device. The four-dimensional spatiotemporal feature vector is a multidimensional data expression that integrates time, space, environment, and performance.

[0060] Using a time series analysis algorithm to perform trend prediction on the four-dimensional space-time feature vector, and generating a dynamic threshold recommendation value within a future preset time period;

[0061] Specifically, we predict each dimension separately based on the ARIMA model, set the autoregressive order to 3, the difference order to 1, and the moving average order to 2. After predicting the values ​​of each parameter for the next 24 hours, we generate the dynamic threshold recommendation value by weighted fusion of temperature, humidity, wind speed, and action delay. ,have: , In the formula, is the weight coefficient, determined by ridge regression training, is the temperature factor, is the humidity factor, is the wind speed factor, is the action delay factor. ARIMA model is an autoregressive integrated moving average model, which is suitable for non-stationary time series prediction. Ridge regression is a linear regression method with L2 regularization to prevent overfitting of weight coefficients.

[0062] After the dynamic threshold recommendation value is superimposed on the real-time meteorological warning data, the final initial threshold default value is determined by the gradient descent method.

[0063] Specifically, receive the rainfall probability and lightning warning level for the next 6 hours issued by the meteorological station, construct correction items, and ,have: , In the formula, is the weight coefficient, is the probability of rainfall, is the lightning warning level. The optimal threshold is iteratively solved using the gradient descent method. The default value of the final initial threshold is ,have: , in, is the theoretical safety threshold calculated based on the line current carrying capacity, The gradient descent iterative function with the default value of the initial threshold.

[0064] For example, a mountain line is predicted to have a humidity rise of 85% and a lightning level of 3 in the next 24 hours. ARIMA predicts that the humidity factor is 86% and the wind speed factor is 8m / s. Combined with the lightning warning calculation , dynamic threshold recommended value , calculated after gradient descent optimization , which is 3% higher than the original threshold of 1000A. In actual thunderstorms, the line inrush current reaches 1015A, and the threshold adjustment avoids the accidental disconnection of normal loads. Through the dynamic optimization of thresholds driven by meteorological warnings, real faults and instantaneous overcurrents caused by lightning can be effectively distinguished to maintain power supply continuity.

[0065] Optionally, the method for acquiring the wire spatial displacement includes: The spatial coordinates of the distribution line corridor are scanned in real time by LiDAR to generate the three-dimensional displacement trajectory of the wire suspension point; Specifically, a rotating laser radar device is set up on both sides of the corridor where the conductor is located. A laser pulse is emitted every 0.05 seconds. After receiving the reflected signal, the distance between the conductor suspension point and the radar station is calculated by the time-of-flight method. The observation data of multiple radar stations are combined to reconstruct the real-time spatial coordinates of the suspension point based on the three-dimensional point cloud triangulation algorithm. The coordinate sequence of 20 consecutive cycles (within 1 second) constitutes a displacement trajectory curve. The laser radar time-of-flight method is a precise measurement technology that calculates distance based on the time delay between the emission and reception of light pulses. Three-dimensional point cloud triangulation is a geometric method that uses multi-station observation data to solve three-dimensional space coordinates. The displacement trajectory curve describes the continuous movement path of the conductor in space. Figure 4 As shown, the marked points show the key displacement nodes of the wire, reflecting the spatiotemporal synchronization of mechanical vibration and electrical signals.

[0066] Performing time synchronization matching on the displacement mutation amount of the three-dimensional displacement trajectory and the fluctuation period of the spatiotemporal fusion feature matrix; Specifically, the maximum absolute value of the coordinate change of adjacent periods on the displacement trajectory is extracted. If the change of the x-axis coordinate is the largest, then the displacement mutation It is equal to the absolute value of the change in the x-axis coordinate; when the maximum absolute value of the coordinate change is greater than the preset value, such as 0.5m, it is marked as a displacement mutation period. Synchronously read the mutation timestamps in the spatiotemporal fusion feature matrix where the entropy fluctuation exceeds the safety benchmark. If there is a time period overlap area where the difference between the displacement mutation period and the mutation timestamp is less than 0.1s, it is determined to be a spatiotemporal correlation event. The spatiotemporal fusion feature matrix is ​​a data structure containing electrical parameters, environment and displacement characteristics. Time synchronization matching is the comparison process of mechanical displacement and electrical anomaly occurrence time, and the error window is 0.1 seconds.

[0067] When the correlation coefficient between the displacement mutation and the electrical parameter fluctuation is greater than a preset correlation threshold, a high-risk warning signal of line mechanical deformation is triggered.

[0068] Specifically, calculate the displacement mutation sequence The parameter sequence of the entropy change of the electrical parameters The Pearson correlation coefficient is: , in, For the The displacement mutation amount, For the The entropy change parameters of electrical parameters, is the mean value of displacement mutation, is the mean value of the entropy change parameter of the electrical parameter. If the Pearson relationship number If the duration exceeds 5 seconds, a Level III high-risk warning signal will be triggered and the corresponding line section will be located.

[0069] For example, a line in a mountainous area was vibrating under strong winds, and the laser radar measured a sudden change in displacement of 6m for 8 seconds. The spatiotemporal fusion matrix showed a surge in entropy values ​​during the same period, and the Pearson correlation coefficient was calculated. . A high-risk warning was triggered and the section between tower 3 and tower 4 was locked. The hidden danger of the suspension wire clamp loosening due to the dancing of the wire was discovered 12 hours in advance, avoiding the accident of tower collapse due to breakage. Traditional line inspection requires visual inspection and it is difficult to find such gradual defects.

[0070] Optionally, the high-risk warning signal triggering line mechanical deformation includes: Input the time and space coordinates corresponding to the high-risk warning signal into a preset insulator aging prediction model to calculate the residual mechanical strength attenuation rate; Specifically, the insulator aging prediction model adopts the LSTM neural network structure. The input parameters include the historical data of ambient temperature and humidity corresponding to the time and space coordinates, the mean displacement mutation within 30 days before the warning occurs, and the cumulative power frequency flashover times of insulators at the same position. The output is the mechanical strength attenuation rate in the next 30 days. ,have: , in, is the mechanical strength attenuation rate adjustment coefficient, is the displacement mutation mean adjustment coefficient, is the cumulative power frequency flashover adjustment coefficient, is the average displacement mutation amount within 30 days before the warning occurs, is the cumulative power frequency flashover times of insulators at the same position, is the historical data of ambient temperature and humidity corresponding to the time and space coordinates, is a natural constant.

[0071] The inspection period of the section is adjusted according to the attenuation rate, and the threshold value is increased synchronously.

[0072] Specifically, the inspection cycle adjustment formula is: , In the formula, For the adjusted inspection cycle, is the base attenuation coefficient, which can be 0.3. At the same time, the high-frequency harmonic threshold of partial discharge detection is increased from 5% to 3% to improve sensitivity.

[0073] For example, after a warning is triggered in a coastal salt fog section, the input model is obtained (Standard value 0.1). The inspection cycle was adjusted from 28 days to On the 7th day, the threshold was reduced to 3%. The subsequent inspection on the 7th day found that the salt density on the surface of the insulator exceeded the standard and was cleaned in time. Avoid insulator brittle fracture accidents caused by salt spray corrosion. At the same time, the high-frequency threshold adaptive adjustment captures early signs of corona discharge. Sub-meter deformation monitoring is achieved through lidar and is aligned with the electrical characteristics in time and space, so that the system can determine whether mechanical disturbances such as conductor dancing and icing cause safety hazards. Combined with the LSTM model to predict the attenuation path of the insulator life, the detection parameters can be adjusted dynamically in stages, so that operation and maintenance resources can be accurately deployed to high-risk points. After implementation, the prevention capability of hidden mechanical defects is significantly improved, and the service life of key equipment is extended.

[0074] Based on the same inventive concept, Figure 5 As shown, the present invention also provides a distribution line fault intelligent diagnosis system based on multi-source information fusion, the system comprising: The data acquisition module is used to obtain the current signal, voltage signal and temperature signal of the distribution line, and simultaneously collect the humidity gradient change rate in the meteorological data and the conductor spatial displacement in the geographic coordinate data to generate a multi-dimensional heterogeneous data set; A feature construction module, used to calculate the entropy value of the time domain distortion features of the current signal and the voltage signal, and to superimpose the humidity gradient change rate to generate a dynamic entropy change parameter, and to perform spatial morphology calibration on the dynamic entropy change parameter based on the spatial displacement of the conductor to generate a spatiotemporal fusion feature matrix; An initial threshold adjustment module is used to dynamically adjust the preset initial threshold using the time-space fusion feature matrix; wherein the adjustment process includes calculating the conductor resistance change rate according to the real-time temperature data in the meteorological data, and performing weighted compensation on the resistance change rate through a sliding window algorithm to generate an adaptive fault discrimination threshold; A fault judgment module is used to perform multiple rounds of verification on abnormal fluctuations of electrical parameters of the line according to the fault judgment threshold, and generate a fault location instruction with a confidence weight when the fluctuation amplitude of three consecutive sampling periods exceeds the fault judgment threshold and the proportion of high-frequency harmonics reaches a preset threshold; The fault processing module is used to match the fault location instruction with the line topology in real time, and to correct the fault section priority queue based on the switch action feedback data of the feeder terminal, and trigger the fault removal action matching the priority.

[0075] It should be noted that the electrical connection between the above-mentioned units does not necessarily mean direct connection of the lines, and the indirect connection mode can be applied to the embodiments of the present invention as long as the purpose of the present invention is achieved. The above is only an exemplary embodiment of the present invention and cannot be used to limit the scope of the present invention.

[0076] That is, any equivalent changes and modifications made according to the teachings of the present invention are still within the scope of the present invention. After considering the disclosure of the specification and the truth of practice, those skilled in the art will easily think of other embodiments of the present invention. This application is intended to cover any variation, use or adaptive change of the present invention, which follows the general principles of the present invention and includes common knowledge or customary technical means in the art that are not described in the present invention.

Claims

1. An intelligent diagnosis method for distribution line faults based on multi-source information fusion, characterized in that: The method comprises: Obtain the current signal, voltage signal and temperature signal of the distribution line, and simultaneously collect the humidity gradient change rate in the meteorological data and the conductor spatial displacement in the geographic coordinate data to generate a multi-dimensional heterogeneous data set; The time domain distortion characteristics of the current signal and the voltage signal are entropy calculated, and the humidity gradient change rate is superimposed to generate a dynamic entropy change parameter. At the same time, the dynamic entropy change parameter is spatially calibrated based on the wire spatial displacement to generate a time-space fusion feature matrix; The preset initial threshold is dynamically adjusted using the time-space fusion feature matrix; wherein the adjustment process includes calculating the conductor resistance change rate according to the real-time temperature data in the meteorological data, and performing weighted compensation on the resistance change rate through a sliding window algorithm to generate an adaptive fault discrimination threshold; Perform multiple rounds of verification on abnormal fluctuations of the electrical parameters of the line according to the fault discrimination threshold, and generate a fault location instruction with a confidence weight when the fluctuation amplitude of three consecutive sampling cycles exceeds the fault discrimination threshold and the proportion of high-frequency harmonics reaches a preset threshold; The fault location instruction is matched with the line topology in real time, and the fault section priority queue is corrected based on the switch action feedback data of the feeder terminal, triggering a fault removal action matching the priority.

2. The intelligent diagnosis method for distribution line fault based on multi-source information fusion according to claim 1 is characterized in that: The generating of the spatiotemporal fusion feature matrix comprises: Extract the time domain waveform distortion characteristics of the current signal and the third harmonic mutation rate of the voltage signal, and calculate the composite distortion entropy value of the two; The composite distortion entropy value is superimposed on the logarithmic attenuation coefficient of the humidity gradient change rate to generate a dynamic entropy change parameter; The dynamic entropy change parameter is subjected to morphological corrosion processing based on the fluctuation trajectory of the spatial displacement of the wire, and a spatiotemporal fusion feature matrix is ​​output.

3. The intelligent diagnosis method for distribution line fault based on multi-source information fusion according to claim 2 is characterized in that: The generating of the adaptive fault discrimination threshold comprises: Obtain historical average data of the wire resistance change rate, and fit the resistance change curve based on the real-time temperature data; Combined with the time-space fusion feature matrix, a dynamic Bayesian network is used to identify abnormal sections of the resistance change curve and generate compensation weight factors; The compensation weight factor is convolved with the initial threshold to generate a fault discrimination threshold including a dynamic coupling relationship of environmental parameters.

4. The intelligent diagnosis method for distribution line fault based on multi-source information fusion according to claim 3 is characterized in that: The performing multiple rounds of verification on abnormal fluctuations of electrical parameters of the line according to the fault discrimination threshold comprises: When the first fluctuation amplitude exceeds the fault discrimination threshold, collecting hydrogen gas volume fraction data of the distribution transformer oil chromatogram; If the hydrogen gas integral exceeds the preset fractional threshold and the high-frequency harmonic ratio reaches the preset threshold, a partial discharge prediction signal is triggered; The fault discrimination thresholds of two subsequent sampling cycles are adjusted according to the prediction signal.

5. The intelligent diagnosis method for distribution line fault based on multi-source information fusion according to claim 4 is characterized in that: The generating of the fault location instruction with confidence weight comprises: Acquiring switch status data of the feeder terminal, and reversely calculating the fault current propagation path based on the switch status data; Performing overlap analysis on the propagation path and the abnormal area of ​​the spatiotemporal fusion feature matrix, and increasing the confidence weight when the overlapping area ratio is greater than a preset ratio threshold; If there are multiple candidate sections, the correlation coefficient between the change rate of the dynamic entropy change parameter and the wire spatial displacement is calculated, and the section is selected according to the size of the correlation coefficient.

6. The intelligent diagnosis method for distribution line fault based on multi-source information fusion according to claim 5 is characterized in that: The method further comprises: Collect the line capacitance current difference before and after the fault is removed, and perform similarity matching with the standard deviation value in the preset fault feature database; Generate a fault type label according to the matching result, associate the fault type label with the environmental parameter and then add it to a preset historical fault case library; Based on the fault frequency distribution at the same geographic coordinates in the historical fault case library, the subsequent initial threshold default value is updated.

7. The intelligent diagnosis method for distribution line fault based on multi-source information fusion according to claim 6 is characterized in that: The initial threshold value default values ​​after the update include: Extracting environmental characteristic parameters and action delay data corresponding to the fault type label to generate a four-dimensional spatiotemporal characteristic vector; Using a time series analysis algorithm to perform trend prediction on the four-dimensional space-time feature vector, and generating a dynamic threshold recommendation value within a future preset time period; After the dynamic threshold recommendation value is superimposed on the real-time meteorological warning data, the final initial threshold default value is determined by the gradient descent method.

8. The intelligent diagnosis method for distribution line faults based on multi-source information fusion according to claim 4 is characterized in that: The method for obtaining the wire spatial displacement includes: The spatial coordinates of the distribution line corridor are scanned in real time by LiDAR to generate the three-dimensional displacement trajectory of the wire suspension point; Performing time synchronization matching on the displacement mutation amount of the three-dimensional displacement trajectory and the fluctuation period of the spatiotemporal fusion feature matrix; When the correlation coefficient between the displacement mutation and the electrical parameter fluctuation is greater than a preset correlation threshold, a high-risk warning signal of line mechanical deformation is triggered.

9. The intelligent diagnosis method for distribution line fault based on multi-source information fusion according to claim 8 is characterized in that: The high-risk warning signals that trigger line mechanical deformation include: Input the time and space coordinates corresponding to the high-risk warning signal into a preset insulator aging prediction model to calculate the residual mechanical strength attenuation rate; The inspection period of the section is adjusted according to the attenuation rate, and the threshold value is increased synchronously.

Images