A fixed-gap lightning protection insulator flashover fault identification system

Abstract

The application relates to the technical field of flashover fault identification, and discloses a fixed-gap lightning protection insulator flashover fault identification system, which comprises the following steps: synchronously collecting multi-source discharge observation and unifying to a dimensionless time axis; extracting device entity key nodes to expand a three-dimensional geometric contour into a one-dimensional normalized development path, and projecting visible arc light and temperature rise distribution onto the development path to reconstruct time-varying arc light density distribution and thermal trace density distribution; calculating hysteresis invasion ring area and invasion irreversibility according to a mass center trajectory, fusing discharge current and radio frequency information to calculate a flashover deviation energy; adaptively distributing evaluation weights by using each time sequence information distribution, deriving a posterior probability of a real flashover fault without a threshold, and finally locking a specific fault position and an evolution stage. The application eliminates misjudgment caused by normal protective arcing and real fault initial homogenization, effectively overcomes interference caused by individual aging drift in long-term operation, and improves the precision of online monitoring of a power transmission line.

Description

Technical Field

[0001] This invention relates to the field of flashover fault identification technology, and more specifically, to a flashover fault identification system for fixed-gap lightning protection insulators. Background Technology

[0002] In the external insulation protection of power transmission networks, fixed-gap lightning protection insulators often use parallel metal discharge electrodes to guide lightning overvoltages to preferentially discharge in the gap area, thus protecting the integrity of the creepage path of the main insulation. Therefore, the initial discharge of the gap is a self-protection process as expected. However, under complex conditions such as high altitude, heavy pollution, or icing, the electric field distribution after arc initiation is easily deteriorated, causing the arc, which should be confined to the vicinity of the gap, to jump and spread unexpectedly. When environmental pollution and humidity cause changes in the local insulation medium, the arc, which should have extinguished quickly, can easily cross the boundary of the first layer of sheds and irreversibly invade adjacent insulation areas in a unidirectional direction, thus evolving into a flashover fault in the main insulation. Current monitoring schemes mostly rely on a single discharge current peak or a local brightness transient peak for threshold triggering. This method ignores the extremely similar spatial origin of the safety discharge action and the actual flashover intrusion in the early stage of discharge, easily mistaking normal protective actions for equipment failure. As insulation devices age, electrode tip ablation and passivation, along with uneven aging of the skirt surface, cause a delayed shift in the discharge initiation boundary, resulting in significantly different signal spectra for the same type of fault at different times. Relying on fixed thresholds and isolated indicators to address this shift not only leads to severe false alarms and missed alarms but also derails the accuracy of grid condition-based maintenance. Summary of the Invention

[0003] This invention provides a flashover fault identification system for fixed-gap lightning protection insulators, which solves the technical problems mentioned in the background art.

[0004] This invention provides a fixed-gap lightning protection insulator flashover fault identification system, applied to a lightning protection device comprising a discharge electrode, a discharge gap, a first shed, a subsequent shed, grounding side hardware, and a multi-source sensing unit. The multi-source sensing unit is used to acquire discharge current, radio frequency amplitude, visible image, and temperature. The fixed-gap lightning protection insulator flashover fault identification system includes a processor configured to perform the following steps: The multi-source observation signals of the lightning protection device are synchronously acquired using the same event window, and the multi-source observation signals are uniformly compressed to a dimensionless time axis. The multi-source observation signals include the original discharge current, the original radio frequency amplitude, the original visible image, and the original fiber grating temperature. Extract the spatial key nodes of the lightning protection device to construct a one-dimensional normalized expansion path, and project the multi-source observation signal onto the one-dimensional normalized expansion path; The time-varying arc light density distribution and thermal trace density distribution are reconstructed on the one-dimensional normalized expansion path, respectively. Based on the time-varying trajectories of the arc light density distribution and the thermal trace density distribution, the area of ​​the hysteresis intrusion ring and the intrusion irreversibility are calculated. The dimensionless current and dimensionless radio frequency amplitude generation time excitation weights obtained by the compression mapping of the original discharge current and the original radio frequency amplitude are used, and the gap deviation energy is calculated by combining the arc density distribution and the thermal trace density distribution. The discreteness of the time-series evolution sequence of parameters within each single event is obtained to adaptively allocate evaluation weights. The posterior probability of the true flashover fault can be obtained by fusing the hysteresis intrusion ring area, the intrusion irreversibility, and the gap deviation. When a real flashover fault is determined based on the posterior probability, the location and evolution stage of the fault are output according to the spatial density distribution and temporal change characteristics at the final moment.

[0005] The beneficial effects of this invention include: by synchronously projecting the visible arc front and the residual temperature rise memory onto the normalized unfolded device path; and by deeply analyzing the unidirectional propagation ratio along the surface after the arc breaks through the initial boundary, this invention can screen out dangerous flashovers that have already undergone malignant extension. Combined with an adaptive probability weighting mechanism based on the internal evolution dispersion of events, this invention eliminates the hybrid artifacts of the early arc initiation stage, directly penetrating the drift fog caused by device surface aging and positional deformation without requiring repeated manual adjustments to the monitoring threshold, achieving highly robust qualitative discrimination and accurate spatial tracing of various discharge processes. Attached Figure Description

[0006] Figure 1 This is a block diagram of a fixed-gap lightning protection insulator flashover fault identification system according to the present invention. Detailed Implementation

[0007] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0008] like Figure 1As shown, a fixed-gap lightning protection insulator flashover fault identification system is applied to a lightning protection device comprising a discharge electrode, a discharge gap, a first shed, a subsequent shed, grounding side hardware, and a multi-source sensing unit. The multi-source sensing unit is used to acquire discharge current, radio frequency amplitude, visible image, and temperature. The fixed-gap lightning protection insulator flashover fault identification system includes a processor configured to perform the following steps: The multi-source observation signals of the lightning protection device are synchronously acquired using the same event window, and the multi-source observation signals are uniformly compressed to a dimensionless time axis. The multi-source observation signals include the original discharge current, the original radio frequency amplitude, the original visible image, and the original fiber grating temperature. Extract the spatial key nodes of the lightning protection device to construct a one-dimensional normalized expansion path, and project the multi-source observation signal onto the one-dimensional normalized expansion path; The time-varying arc light density distribution and thermal trace density distribution are reconstructed on the one-dimensional normalized expansion path, respectively. Based on the time-varying trajectories of the arc light density distribution and the thermal trace density distribution, the area of ​​the hysteresis intrusion ring and the intrusion irreversibility are calculated. The dimensionless current and dimensionless radio frequency amplitude generation time excitation weights obtained by the compression mapping of the original discharge current and the original radio frequency amplitude are used, and the gap deviation energy is calculated by combining the arc density distribution and the thermal trace density distribution. The discreteness of the time-series evolution sequence of parameters within each single event is obtained to adaptively allocate evaluation weights. The posterior probability of the true flashover fault can be obtained by fusing the hysteresis intrusion ring area, the intrusion irreversibility, and the gap deviation. When a real flashover fault is determined based on the posterior probability, the location and evolution stage of the fault are output according to the spatial density distribution and temporal change characteristics at the final moment.

[0009] In particular, the dimensions of the numerically stable term ε appearing in all formulas in this application are consistent with the dimensions of the main operand directly added to it in the expression, and its value is the typical minimum value of the corresponding main operand under normal operating conditions. times.

[0010] Specifically, the coordinates of the one-dimensional normalized expansion path described in this application The dimensionless cumulative path length is calculated as follows: First, key spatial nodes are selected sequentially along the path from the tip of the high-voltage side discharge electrode, the center of the discharge gap, the root of the first umbrella skirt, the edge of the first umbrella skirt, and the subsequent umbrella skirts to the grounding side fitting entrance, constructing a continuous spatial polygonal path; then, the total physical length of this polygonal path is calculated. For any observation point in space First, the interpolation parameters on the spatial polyline path are determined using the nearest neighbor projection method. Then calculate the cumulative physical length from the path start point to the projection point. Finally, normalization operation Obtain dimensionless path coordinates .

[0011] Specifically, the dimensionless arc image obtained through background subtraction and nonlinear compression in this application needs to undergo full-field area normalization before being used for arc density distribution reconstruction calculations. That is, the normalized arc image... Therefore, the full-field dual-space integral value of the dimensionless arc diagram is always 1.

[0012] Specifically, this application uses intermediate features (including the area of ​​the hysteresis intrusion loop) to calculate the posterior probability of a true flashover fault. Irreversible intrusion Arc gap deviation energy thermal trace gap deviation energy Before being substituted into the logarithmic evidence calculation formula, all parameters must be normalized to dimensionless parameters.

[0013] Preferably, the multi-source observation signals of the lightning protection device are simultaneously acquired using the same event window, and the multi-source observation signals are uniformly compressed onto a dimensionless time axis, including: The original discharge current was collected. The original radio frequency amplitude The original visible image and the original fiber grating temperature ,in and These are the original pixel coordinates. Original time, For measurement point index; The sum of the absolute value of the original discharge current and the absolute value of the original radio frequency amplitude after compression mapping is calculated to form a composite excitation. : in, This is a preset nonlinear compression function; The nearest local minima on the left and right sides of the main peak of the composite excitation are respectively taken as the start times of the event window. and the end time ; Current time Subtract the difference from the start time, divide by the difference between the end time and the start time, and a preset numerical stability term. The sum of these values ​​yields the dimensionless time on the dimensionless time axis. : Substitute the absolute values ​​of the original discharge current and the original radio frequency amplitude into a preset nonlinear compression function. The dimensionless currents were obtained respectively. and the dimensionless radio frequency amplitude : The image prior to the start time is used as the reference frame. After performing background subtraction absolute value processing on the original visible image, the result is fed into the nonlinear compression function. , thus obtaining a dimensionless arc light pattern : in, and The dimensions are the pixel coordinates after image width and height processing; the temperature rise of each measuring point in the multi-source sensing unit relative to the starting time is extracted and substituted into the nonlinear compression function. The dimensionless temperature rise sequence at each measuring point was obtained. : The same event window is a unified analysis period established around the same discharge event.

[0014] Multi-source observation signals are a collection of synchronous observation data from discharge current, radio frequency amplitude, visible images, and temperature channels, used to characterize discharge behavior from multiple dimensions including electrical, optical, and thermal aspects.

[0015] A dimensionless time axis is a time representation that compresses the original time of different sampling frequencies and different physical units into a standard time interval, thereby eliminating the difficulty of comparison caused by the different durations of different events.

[0016] The initial discharge current is the original current waveform flowing through the discharge circuit when a discharge event occurs. It can be obtained using a high-frequency Rogowski coil, a broadband current transformer, or a low-inductance shunt in conjunction with a high-speed data acquisition card.

[0017] The original radio frequency amplitude is the original amplitude of the high-frequency electromagnetic signal radiated during the discharge process, output through the receiving link. It can be obtained using a broadband radio frequency antenna, an ultra-high frequency antenna, or an electromagnetic coupling sensor in conjunction with an envelope detection circuit and a high-speed acquisition card.

[0018] The raw visible image is the original image sequence of the discharge process in the visible light channel. It can be acquired using a high-speed industrial camera or a high-speed visible light camera.

[0019] The raw fiber Bragg grating temperature is the original temperature sequence output from fiber Bragg grating measuring points deployed at different locations within the lightning protection device. It can be obtained using fiber Bragg grating sensors and fiber Bragg grating demodulators.

[0020] The original pixel coordinates are the position markers of each pixel in the original visible image within the image plane, used to locate the spatial distribution of the arc light in the image.

[0021] The original time t is the time stamp of each channel sample under the original acquisition clock.

[0022] The measurement point index k is a numbering marker used to distinguish different temperature measurement points in the multi-source sensing unit.

[0023] The composite excitation is a combined excitation quantity formed by superimposing the original discharge current and the original radio frequency amplitude after compression mapping, and is used to stably characterize the electrical excitation intensity of a discharge event.

[0024] Nonlinear compression functions are monotonic mapping functions that compress original signals with large dynamic ranges into a uniform dimensionless interval. Logarithmic compression functions are preferred, followed by arctangent compression functions, thereby reducing the peak-dominant effects of discharge current and radio frequency amplitude while preserving weak discharge differences.

[0025] The composite excitation peak is the point of maximum response of the composite excitation within the current candidate event segment, used to anchor the core moment of a discharge event.

[0026] Local minima are the nearest troughs on either side of the main peak of the composite excitation that satisfy the minimum condition, and are used to determine the boundaries of the event window.

[0027] The starting time is the time position corresponding to the left boundary of the event window.

[0028] The termination time is the time position corresponding to the right boundary of the event window.

[0029] The numerical stability term is a small positive number used to prevent the denominator from being zero, the logarithm from being singular, or the normalization from becoming unstable. A preferred value is 0.000001, which ensures numerical stability without altering the dominant order of the input.

[0030] Dimensionless time is a time coordinate system that compresses the original time into a unified standard interval based on the start and end times, and is used to achieve same-scale comparisons between different events.

[0031] Dimensionless current is a standardized current characterization quantity formed by absolute value processing and nonlinear compression of the original discharge current, and is used to participate in the subsequent calculation of time excitation weights.

[0032] The dimensionless radio frequency amplitude is a standardized radio frequency characterization quantity formed by absolute value processing and nonlinear compression of the original radio frequency amplitude, which is used to participate in the subsequent calculation of time excitation weights.

[0033] The reference frame is a reference image selected before the start time.

[0034] The dimensionless arc light map is a standardized arc light distribution map obtained by performing background difference absolute value processing on the original visible image relative to the reference frame and then performing nonlinear compression. It is used to characterize the intensity and position of the arc light in the image at different times.

[0035] Dimensionless pixel coordinates are pixel position coordinates that are normalized from the width and height of the original image, and are used to eliminate scale differences caused by different image resolutions.

[0036] Temperature rise is the increment of the baseline temperature at each measurement point at the current moment relative to the starting moment, used to characterize the degree of propagation of discharge heat accumulation on the device surface.

[0037] The dimensionless temperature rise sequence at each measuring point is a standardized time series obtained by nonlinear compression of the temperature rise at each measuring point.

[0038] In practice, for synchronous acquisition within the same event window, the discharge current channel is used as the main trigger channel. When two consecutive sampling points of the original discharge current exceed five times the root mean square value of the silent baseline, it is determined to enter the candidate event. The main trigger signal is simultaneously sent to the RF acquisition card, high-speed camera, and fiber optic demodulator, and each channel is written with a unified timestamp. When the acquisition devices are deployed in a distributed manner, the same hardware clock or unified timing pulse is used for time synchronization to ensure that the clock deviation between channels is no more than 1 microsecond. Each channel extracts a candidate segment from 2 milliseconds before arc ignition to 20 milliseconds after arc ignition as the original input for the same event window.

[0039] In practice, for multi-source signal time, it is necessary to first convert each channel to a relative time coordinate with the main trigger time as the zero point, then retain the original sampling points of high sampling rate signals such as current and radio frequency in chronological order, and project the image frame time and temperature sampling time onto the same relative time axis; before entering dimensionless time compression, current and radio frequency are resampled using linear interpolation or piecewise cubic interpolation, the image uses the nearest neighbor mapping of frame time, and the temperature uses linear interpolation.

[0040] In practical implementation, for nonlinear compression functions, the input needs to be nonnegated first, then the amplitude is normalized according to the representative upper limit within the event or the rated range of the device, and then the result is mapped to the interval of 0 to 1 using a monotonically increasing compression function; logarithmic compression is preferred, followed by arctangent compression.

[0041] In practice, for the construction of composite excitation, the original discharge current and the original radio frequency amplitude need to be de-DC, anti-spiking and short-window smoothing processed before entering nonlinear compression; median filtering is first performed to remove isolated spikes, and then sliding smoothing with a length of 5 to 11 points is performed; when the signal noise floor is large, the mean of the silent segment can be used as the baseline and baseline subtraction can be performed.

[0042] In practice, for the main peak of the composite excitation and the nearest local minima on the left and right, it is necessary to first determine the global maximum point of the composite excitation within the candidate event segment as the main peak of the composite excitation, and then search point by point to the left and right from the main peak. When the derivative sign changes from negative to positive and the current value is less than 10% to 20% of the peak value, it is identified as the nearest local minima on the left and right. If no local minima that meets the conditions appears on one side, the first sampling point on that side that is less than 10% of the peak value is used as the alternative boundary.

[0043] In practice, for numerical stability terms, positive stability terms of the same order of magnitude should be used in all operations involving division, logarithms, and probability normalization. When the input has been compressed to the range of 0 to 1, the numerical stability term should preferably be 0.000001. When the input has not been fully normalized and still retains a large value difference, the numerical stability term can be increased to 0.0001 or 0.001.

[0044] In practice, for the reference frame and dimensionless pixel coordinates, the reference frame should be the average frame of the 1 to 5 nearest arc-free images before the start time to suppress camera noise and environmental flicker. The dimensionless pixel coordinates are obtained by dividing the pixel horizontal coordinate by the image width minus 1 and the pixel vertical coordinate by the image height minus 1, so that both the horizontal and vertical coordinates fall within the range of 0 to 1.

[0045] In practice, for temperature rise and fiber optic grating measurement point baseline, the baseline temperature of each measurement point should be taken as the average of 5 to 20 consecutive sampling points before the start time. The temperature rise is obtained by subtracting the baseline temperature from the current temperature. When the ambient temperature drifts slowly, at least one reference measurement point that is not directly affected by the arc light should be set, and the simultaneous change of the reference measurement point should be used to perform common mode compensation for all working measurement points.

[0046] Preferably, extracting the spatial key nodes of the lightning protection device to construct a one-dimensional normalized unfolded path, and projecting the multi-source observation signals onto the one-dimensional normalized unfolded path, includes: The key spatial nodes are selected sequentially along the path from the tip of the high-voltage side discharge electrode, the center of the discharge gap, the root of the first parachute skirt, the edge of the first parachute skirt, and the subsequent parachute skirts to the grounding side fitting entrance. Constructing spatial polyline paths ; The total path length is obtained by summing the spatial Euclidean distances between all adjacent spatial key nodes. : in, The total number of line segments in the spatial polyline path, i.e., the total number of key nodes, is: ; For any two-dimensional image pixel coordinates or measurement point coordinates in space Calculate the shortest distance from it to each line segment of the spatial polyline path to determine the nearest projection point; The cumulative distance along the stated direction to the nearest projection point is accumulated. Divide by the total path length to obtain the dimensionless path coordinates of the point on the one-dimensional normalized expanded path. : in, These are the interpolation parameters.

[0047] Spatial critical nodes are key structural locations selected sequentially along the possible flashover propagation paths of a fixed-gap lightning protection insulator. They are used to abstract the complex shape of the actual device into a calculable path skeleton.

[0048] One-dimensional normalized expansion path maps the actual propagation path on the device surface to a one-dimensional path expression within a continuous interval of 0 to 1.

[0049] A spatial polyline path is a polyline trajectory formed by sequentially connecting adjacent key spatial nodes.

[0050] The total path length is the cumulative physical length of the spatial polyline path from the starting point to the ending point.

[0051] The total number of line segments is the number of line segments formed between adjacent key nodes in a spatial polyline path.

[0052] The pixel coordinates or measurement point coordinates of a two-dimensional image are the coordinate representations of the position of the arc light pixel or the spatial position of the temperature measurement point in the image.

[0053] The nearest projection point is the point on the spatial polyline path that is closest to a given coordinate point.

[0054] The cumulative distance is the path length accumulated from the starting point of the path along the direction to the nearest projection point, used to characterize the progress of that point on the path.

[0055] Dimensionless path coordinates are the ratio of cumulative distance to total path length.

[0056] Interpolation parameters are parameters that describe the relative position of the nearest projected point within a path segment.

[0057] In practical implementation, for key spatial nodes and coordinate systems, it is necessary to first establish a unified spatial coordinate system for the lightning protection device based on equipment design drawings, 3D models or physical measurements. Priority should be given to using the tip of the high-voltage side discharge electrode as the starting point reference and the direction of the grounding side fitting entrance as the positive direction of the path. Key spatial nodes should include at least the tip of the discharge electrode, the center of the discharge gap, the root of the first umbrella skirt, the edge of the first umbrella skirt, and the grounding side fitting entrance. If there are many subsequent umbrella skirts, nodes should be added at the root and edge of each subsequent umbrella skirt.

[0058] In practice, for one-dimensional normalized unfolding paths, it is necessary to connect adjacent key spatial nodes according to the physical direction of discharge intrusion from the high-voltage side to the ground side to form a spatial polygonal path. Then, the total path length is obtained by accumulating the Euclidean distances of each segment. Finally, the cumulative path distance from any projection point to the starting point is divided by the total path length. When the equipment structure is asymmetrical, the path should be established along the outer surface contour where flashover is most likely to occur, rather than along the geometric center line.

[0059] In practice, for the calibration of image pixel coordinates to spatial coordinates, it is necessary to first use a calibration board to complete the camera intrinsic parameter calibration, and then use at least 4 spatial reference markers installed near the device to complete the extrinsic parameter calibration to obtain the mapping relationship between pixel points and spatial rays; then, the pixel rays are intersected with the device's three-dimensional surface model or a pre-established approximate outer contour surface to obtain the corresponding spatial points.

[0060] In practice, for obtaining the coordinates of the measuring points, the physical position along the surface of the device should be recorded when the fiber optic measuring points are installed. The coordinates of the design drawings and the on-site laser ranging should be used for joint verification. For attached measuring points, the specific mounting point at the root, edge or near the hardware entrance of the corresponding umbrella skirt should be recorded and the point should be projected onto the same spatial broken line path.

[0061] In practice, for the calculation of the nearest projection point and the selection of endpoints, it is necessary to calculate the orthogonal projection of any point to be projected to each path segment segment. When the orthogonal projection falls inside the line segment, the projection point is directly selected as a candidate. When the orthogonal projection falls outside the line segment, the candidate point is selected as the endpoint of the line segment closer to the point. Finally, the distances from all candidate points to the origin are compared, and the smallest distance is selected as the nearest projection point.

[0062] In practice, for the normalization direction and zero point of dimensionless path coordinates, the zero point of the path coordinates needs to be fixedly defined at the starting point of the path where the tip of the discharge electrode on the high-voltage side is located, and the path coordinate 1 needs to be fixedly defined at the ending point of the path where the hardware entrance on the grounding side is located. The path coordinates increase monotonically as the flashover spreads to the main insulation and the grounding side. When there are multiple local bends in the same equipment, the path length accumulated along the physical intrusion direction is still used as the only coordinate reference.

[0063] Preferably, reconstructing the time-varying arc light density distribution and thermal trace density distribution on the one-dimensional normalized expansion path includes: Multiply the dimensionless arc image by a two-dimensional sampling kernel function. Then, a double spatial integration is performed, followed by division by the sum of the superposition of the corresponding dimensionless arc light map full-field double spatial integration and the sampling path integration, and the sum of the numerical stability term, to obtain the arc light density distribution that integrals to one on the one-dimensional normalized expansion path. : in, A sampling kernel function that maps two-dimensional pixel space to one-dimensional path coordinates. It is an integral dummy variable; Using the interpolation basis functions of each measurement point on the one-dimensional normalized expansion path The dimensionless temperature rise sequence at each measuring point is weighted and summed, then divided by the integral of the weighted sum over the entire path interval and the sum of the numerical stability term, to obtain the heat trace density distribution whose integral is one on the one-dimensional normalized expansion path. : in, This represents the total number of measurement points on the fiber optic grating. Multiplying the dimensionless path coordinates by the arc light density distribution and the heat trace density distribution respectively, and integrating over the entire path interval, yields the corresponding arc light centroid coordinates. and the coordinates of the centroid of the thermal trace : Arc density distribution is the normalized spatial distribution of a dimensionless arc diagram along a one-dimensional normalized expansion path, used to describe the location where arc energy accumulates along the path at a certain moment.

[0064] A two-dimensional sampling kernel function is a weighted mapping function that projects arc light information in a two-dimensional image space onto one-dimensional path coordinates. A triangular kernel or a Gaussian kernel is preferred, with a bandwidth of 1% to 3% of the total path length, thus smoothing discrete pixel noise while maintaining path position resolution.

[0065] Dumb variables for integration are temporary variables used to complete normalized integration.

[0066] The heat trace density distribution is the normalized spatial distribution of the dimensionless temperature rise sequence at each measuring point after reconstruction along a one-dimensional normalized expansion path. It is used to describe the distribution location of heat accumulation along the path.

[0067] The interpolation basis function is a weighting function that extends the temperature rise at discrete measurement points into a continuous path distribution. The preferred method is a piecewise linear hat-shaped basis function, followed by a cubic spline basis function, which offers simplicity, strong locality, and the ability to stably maintain the transition relationship between adjacent measurement points.

[0068] The total number of fiber Bragg grating measurement points is the total number of all fiber Bragg grating measurement points involved in the reconstruction of the thermal trace density distribution.

[0069] The centroid coordinates of the arc are the weighted average position of the arc density distribution along the one-dimensional normalized expansion path, used to characterize the dominant spatial position of the arc front.

[0070] The centroid coordinates of the heat trace are the weighted average position of the heat trace density distribution along the one-dimensional normalized expansion path, used to characterize the dominant spatial position of the heat accumulation center.

[0071] In practical implementation, for the discrete implementation of the two-dimensional sampling kernel function, ideal point sampling should not be used directly in the discrete implementation. Instead, the dimensionless path coordinates of each pixel should be distributed to several adjacent path sampling points. Triangular kernels or Gaussian kernels should be preferred, with a bandwidth of 1 to 3 times the path sampling interval.

[0072] In practice, for the integral domain, mask, and normalization of the arc light density distribution, it is necessary to calculate the arc light density distribution only within the effective field of view of the device. First, a device region mask is established on the original visible image to remove the background, support, and strong reflection areas. Then, pixel-weighted projection and summation are performed on the dimensionless arc light map within the mask. Finally, normalization is performed using the sum of the projections along the entire path.

[0073] In practice, depending on the type of interpolation basis function and the setting of the support interval, each measurement point needs to have a locally supported interpolation basis function, the center of which is located at the dimensionless path coordinate of the measurement point, and the support interval covers the path segment between the measurement point and the adjacent measurement points; piecewise linear hat-shaped basis functions are preferred, so that any path sampling point is determined by only 2 or 3 adjacent measurement points.

[0074] In practice, for the discrete reconstruction of the heat trace density distribution, it is necessary to first set uniform discrete sampling points on the one-dimensional normalized expansion path, then calculate the basis function values ​​of all measurement points for each sampling point, and then sum the dimensionless temperature rise at the corresponding time according to the basis function weight to obtain the heat trace intensity of the path point at that time. Finally, normalize the entire path.

[0075] In practice, for normalization processing under weak or zero signal conditions, if the total intensity of the arc light or the total intensity of the thermal trace is less than the preset weak signal threshold at a certain moment, the density distribution of the most recent valid moment should be used as a substitute. If there is no valid signal in the entire event period, the density distribution at that moment should be set to uniform distribution throughout the entire path, and the confidence level of that moment should be lowered.

[0076] In practice, for the numerical integration of the centroid coordinates of the arc and the thermal trace, trapezoidal integration is required to obtain the centroid on the discrete path. First, the path coordinates of each sampling point are multiplied by the corresponding density value, and then the summation is performed over the entire path and multiplied by the sampling interval. The number of discrete points on the path is preferably 128 or 256 to balance speed and accuracy.

[0077] Preferably, the calculation of the hysteresis intrusion ring area and intrusion irreversibility based on the time-varying trajectories of the arc light density distribution and the thermal trace density distribution includes: The coordinates of the arc centroid and the coordinates of the thermal trace centroid are used to construct a two-dimensional parametric state curve. The curve is closed using its first and last points, and the absolute value of the integral area enclosed by the closed curve is calculated as the area of ​​the hysteresis intrusion ring. : Calculate the first-order natural cubic spline derivative of the arc centroid coordinates with respect to the dimensionless time, and extract the positive propagation component that is greater than zero. ; The intrusion irreversibility is obtained by dividing the full-time integral of the positive propagation component by the sum of the full-time integral of the absolute value of the first derivative of the natural cubic spline and the numerical stability term. : in, For positive part operators, satisfying .

[0078] The two-dimensional parametric state curve is a state evolution trajectory formed by connecting the coordinates of the arc centroid and the thermal trace centroid in chronological order, and is used to characterize the linkage between the arc front and thermal hysteresis memory.

[0079] The hysteresis intrusion ring area is the absolute value of the area enclosed by the two-dimensional parametric state curve after it is closed. It is used to quantify the degree of hysteresis between arc propulsion and thermal trace following.

[0080] The first derivative of a natural cubic spline is a smoothed first-order rate of change obtained by fitting the curve of the centroid coordinates of the arc light with dimensionless time using a natural cubic spline. It is used to characterize the speed at which the arc light propagates along the path.

[0081] The positive propulsion component is the part of the first derivative of the natural cubic spline that is greater than 0, and is used to retain the effective propulsion along the discharge intrusion direction.

[0082] The positive part operator is an operation rule that sets all non-positive parts of the input to 0 and retains only the positive parts, and is used to filter out the actual forward-moving behavior.

[0083] Intrusion irreversibility is the ratio of the integral of the forward propulsion component to the integral of the total absolute propulsion, used to quantify whether arc intrusion has a unidirectional irreversible characteristic.

[0084] In practice, for the closure of the two-dimensional parameter state curve, it is necessary to first arrange all the sample points of the arc centroid coordinates and thermal trace centroid coordinates in ascending order of dimensionless time, then connect the sample points of adjacent time points in sequence to form an open curve, and finally use a straight line connection between the end point and the beginning point to achieve closure; when the distance between the end point and the beginning point is extremely small and they are approximately coincident, it can be directly regarded as a natural closure.

[0085] In practical implementation, for solving the area of ​​the hysteresis intrusion loop, the polygonal directed area method or trapezoidal line integral method should be used first in the discrete implementation. The areas of small trapezoids formed by adjacent state points are accumulated in time sequence, and the absolute value is finally taken as the area of ​​the hysteresis intrusion loop. When the number of sampling points is small, the state curve can be linearly densified first before calculating the area.

[0086] In practice, for the input and boundary condition settings of natural cubic spline fitting, the input data needs to be set as the arc centroid coordinate sequence on a unified time grid, and the boundary condition adopts the natural boundary condition with the second derivative at both ends being 0. Before fitting, obvious outliers can be removed. Outliers are preferentially defined as points whose deviation from the mean of two adjacent points exceeds 3 times the local standard deviation.

[0087] In practice, for the extraction of forward propagation components and noise suppression, it is necessary to first set a small dead zone for the first derivative of the natural cubic spline, treat derivatives with an absolute value less than 0.001 as 0, retain derivatives with a value greater than 0, and clear derivatives less than or equal to 0, thereby suppressing false forward and backward propagation caused by numerical micro-oscillations.

[0088] In practice, for discretization of the integral of intrusion irreversibility and handling of abnormal situations, trapezoidal integrals should be used for both the positive propulsion component and the absolute value of the first derivative. When the total absolute propulsion is less than the threshold corresponding to the numerical stability term, the intrusion irreversibility is directly set to 0. If the arc centroid hardly moves throughout the entire event, it should be considered that there is no effective intrusion propulsion.

[0089] Preferably, the time excitation weight is generated using the dimensionless current and the dimensionless radio frequency amplitude, and the gap deviation energy is calculated by combining the arc density distribution and the thermal trace density distribution, including: The instantaneous dimensionless current is added to the dimensionless radio frequency amplitude, and then divided by the sum of the integral of this sum over the entire time period and the sum of the numerical stability term to obtain the time excitation weight. : in, For integration time; The second moment of the arc is obtained by multiplying the square of the dimensionless path coordinates by the arc light density distribution and taking the total path integral. Multiply the square of the dimensionless path coordinates by the heat trace density distribution and calculate the total path integral to obtain the second moment of the heat trace. The time excitation weight is multiplied by the second moment of the arc and the second moment of the thermal trace, and the integral is calculated over the entire time interval to obtain the arc gap deviation energy. Deviation energy from thermal trace gap Both together constitute the gap deviation energy: The time excitation weight is a time weight distribution obtained by normalizing the dimensionless current and the dimensionless radio frequency amplitude. It is used to emphasize the contribution of the period of strong electrical excitation to the gap deviation energy.

[0090] The integration time variable is a temporary time variable used when performing time integration.

[0091] The second moment of the arc is the weighted integral of the square of the dimensionless path coordinates over the arc density distribution, and is used to characterize the degree of dispersion and outward expansion of the arc distribution relative to the initial segment of the path.

[0092] The second moment of the thermal trace is the weighted integral of the square of the dimensionless path coordinates over the thermal trace density distribution. It is used to characterize the degree of dispersion and outward expansion of thermal accumulation relative to the initial segment of the path.

[0093] The arc gap deviation energy is a quantity obtained by integrating the time excitation weight and the second moment of the arc over the entire time period. It is used to characterize the degree to which the arc deviates from the gap initiation region under strong electric excitation conditions.

[0094] The thermal trace gap deviation energy is a quantity obtained by integrating the time excitation weight and the second moment of the thermal trace over the entire time period. It is used to characterize the degree to which the thermal trace deviates from the gap initiation region under strong electric excitation conditions.

[0095] Gap deviation energy is a comprehensive deviation descriptor composed of arc gap deviation energy and thermal trace gap deviation energy, used to characterize the strength of fault energy transfer to the main insulation region.

[0096] In practice, for the discrete integration and normalization of the time excitation weight, it is necessary to first sum the dimensionless current and dimensionless radio frequency amplitude at each moment on a unified time grid, then use trapezoidal integration on the sum over the entire time to obtain the normalized denominator, and finally divide the instantaneous sum at each moment by the denominator to obtain the time excitation weight with an area of ​​1; when the denominator is less than the weak excitation threshold, the time excitation weight is changed to a uniform distribution over the entire time.

[0097] In practical implementation, for the reference origin setting of the second moment of arc and the second moment of thermal trace, the starting path segment where the tip of the high-voltage side discharge electrode and the center of the discharge gap are located needs to be regarded as the gap reference starting segment. The larger the path coordinate, the further away from the starting segment and the closer to the downstream area of ​​the main insulation. The square of the path coordinate in the second moment is used to emphasize the far-end distribution extending towards the subsequent shed and grounding side hardware.

[0098] In practical implementation, regarding the correspondence between the physical meaning of the second moment and the gap deviation, it is necessary to clarify that when the arc or heat trace is mainly distributed at the front end of the path, the square of the path coordinate is small and the value of the second moment is low, indicating that the discharge is still limited to the vicinity of the gap; when the distribution gradually extends to the subsequent umbrella skirt and the grounding side hardware inlet, the square of the path coordinate increases rapidly and the second moment increases accordingly, indicating that the fault has deviated from the originally allowed gap discharge area.

[0099] In practice, to achieve the time integration of the arc gap deviation energy and the thermal track gap deviation energy, it is necessary to calculate the second moment of the arc and the second moment of the thermal track on a unified time grid, multiply them by the time excitation weight at the same time, and finally sum them over the entire time period using trapezoidal integration.

[0100] In practice, regarding the combination relationship of gap deviation energy in subsequent determination, the arc gap deviation energy and thermal track gap deviation energy need to be retained as two parallel features and entered into subsequent weight allocation and posterior probability calculation separately, instead of being merged into a single scalar before participating in the determination. When displayed externally, the two can be collectively referred to as gap deviation energy, but they should be retained separately in internal calculations to avoid the mutual cancellation of optical anomaly expansion and thermal anomaly expansion.

[0101] Preferably, the dispersion of the time-series evolution sequence of parameters within each single event is obtained to adaptively allocate evaluation weights, and the posterior probability of the true flashover fault is obtained by fusing the hysteresis intrusion ring area, the intrusion irreversibility, and the gap deviation, including: In a unified time grid Next, extract the absolute difference sequence between the coordinates of the arc centroid and the coordinates of the thermal trace centroid at the same moment. The positive incremental sequence calculated based on the arc centroid The time-series evolution sequences of the arc second moment and the thermal trace second moment constitute the time-series evolution sequence. and : Dividing each of the four sequences by the sum of its total sequence value and the numerical stability term yields four independent discrete probability distributions. : Calculate the information entropy of each of the aforementioned independent discrete probability distributions. The difference is obtained by subtracting the entropy of each parameter from the entropy of the given parameters. Each difference is then divided by the sum of all differences and the numerical stability term to obtain the evaluation weight corresponding to each parameter. : For the area of ​​the hysteresis intrusion ring The irreversibility of the intrusion The arc gap deviation energy and the thermal trace gap deviation energy Assigning intermediate agent volume ,in , , , Calculate the logarithm of the sum of each intermediate proxy quantity and the numerical stability term, divided by one minus the sum of the sum of each intermediate proxy quantity and the numerical stability term, to obtain the logarithmic dominance evidence for each; multiply each piece of logarithmic dominance evidence by its corresponding evaluation weight and sum them to obtain the comprehensive logarithmic evidence. : The posterior probability is obtained by dividing 1 by the sum of 1 and the negative exponent of the natural constant. The exponent of the negative exponent of the natural constant is the composite logarithmic evidence: in, For sequence dimension indexing, This represents the total number of samples in the time grid.

[0102] A unified time grid is a common time frame that resamples all features within the same event onto the same discrete time series. A preferred grid is 128 equally spaced sampling points, thus balancing the evolution resolution of short-duration flashover events with the computational overhead of online processing.

[0103] The absolute difference sequence is a sequence of absolute distances between the centroid coordinates of the arc and the centroid coordinates of the thermal trace at the same moment, formed over time, and is used to reflect the spatial difference between the light front and the thermal hysteresis.

[0104] The positive increment sequence is a sequence formed by retaining only the positive increase in the centroid coordinates of the arc light between adjacent time points, and is used to characterize the incremental rhythm of the progression toward the fault.

[0105] The temporal evolution sequence is a sequence formed by the change of the second moment of the arc light with a uniform time grid, and is used to characterize the temporal evolution of the spatial expansion of the arc light.

[0106] The temporal evolution sequence is a sequence formed by the variation of the second moment of the thermal trace with a uniform time grid, used to characterize the temporal evolution of the spatial expansion of thermal accumulation.

[0107] Independent discrete probability distributions are discrete probability expressions obtained by normalizing each type of time series sequence to a sum of 1.

[0108] Information entropy is an indicator that characterizes the degree of dispersion and concentration of a certain type of independent discrete probability distribution.

[0109] The evaluation weights are adaptively generated from various types of information entropy.

[0110] The intermediate proxy quantity is a normalized quantity that converts the hysteresis intrusion ring area, intrusion irreversibility, arc gap deviation energy, and thermal trace gap deviation energy into a quantity suitable for logarithmic advantage calculation.

[0111] Log-dominant evidence is the amount of evidence obtained by transforming intermediate proxy quantities through log-dominant transformation, and is used to unify the contribution scale of different features.

[0112] The comprehensive logarithmic evidence is a comprehensive criterion obtained by weighting and summing the logarithmic dominance evidence according to the evaluation weights.

[0113] The posterior probability is the output of the probability of a true flashover fault under the current single-event condition.

[0114] Sequence dimension indexes are index markers used to distinguish different feature sequence categories.

[0115] The total number of time grid samples N is the number of sampling points on a uniform time grid.

[0116] In specific implementation, for the unified time grid construction and interpolation strategy, the dimensionless time interval needs to be evenly divided into 64 to 256 sampling points, and then the coordinates of the centroid of the arc light, the centroid of the thermal trace, the second moment of the arc light, and the second moment of the thermal trace are all resampled onto the grid. Continuous quantities are preferentially interpolated using piecewise cubic interpolation, while quantities derived from frame-level images are preferentially interpolated using linear interpolation.

[0117] In specific implementation, for the resampling and missing value handling of the four types of time series evolution sequences, if a single sample of a sequence is missing at a certain time, it should be linearly filled with the adjacent valid samples; if the length of continuous missing values ​​exceeds 10% of the total length of the unified time grid, the event should be directly marked as a low confidence event and the output of the posterior probability should be stopped; for the positive incremental sequence at the start point of the event, its first point is fixed at 0, and the remaining points are obtained only by the difference between the two adjacent resampling times.

[0118] In practice, for the processing of independent discrete probability distributions in the case of zero and near zero, each sequence should be checked for its total sequence sum before sum normalization. If it is less than the threshold, the independent discrete probability distribution corresponding to the sequence should be uniformly distributed, and its subsequent evaluation weight should be reduced, so as to avoid extremely weak sequences being mistakenly regarded as having high concentration.

[0119] In practice, to stabilize information entropy and evaluation weights, the minimum value of each discrete probability distribution needs to be truncated to the lower limit corresponding to the numerical stability term before calculating information entropy, in order to prevent instability of logarithmic operations. After obtaining information entropy, one minus information entropy is used as the discrimination index, and all discrimination indices are renormalized to obtain the evaluation weights. When the discrimination indices of the four sequences are extremely small and the total discrimination index is close to 0, the evaluation weights are uniformly degenerated into equal weight allocation.

[0120] In practice, for the normalization mapping of intermediate proxy quantities, the hysteresis intrusion ring area, arc gap deviation energy, and thermal trace gap deviation energy need to be compressed into the open interval of 0 to 1 through zero-to-one saturation normalization, and then respectively correspond to their respective intermediate proxy quantities; the intrusion irreversibility itself falls within the 0 to 1 interval and can be directly used as its corresponding intermediate proxy quantity; the monotonic compression method of dividing the quantity by the sum of the quantity and 1 is preferred, or the saturation normalization method based on the median value of healthy samples is adopted.

[0121] In practice, for the prior assumptions set for the logarithmic dominance evidence, it is necessary to assume that real flashover faults and non-faults have equal prior status at system startup. Therefore, the comprehensive logarithmic evidence does not have additional prior bias. If the application scenario emphasizes conservative detection, a fixed positive bias can be pre-added to the comprehensive logarithmic evidence in the engineering implementation to improve the fault output tendency.

[0122] In practice, for the numerical truncation and effective interval constraints of the posterior probability output, the posterior probability needs to be limited to the range of 0.001 to 0.999 before output to avoid the subsequent display and statistics modules becoming unstable due to extreme value saturation. When the comprehensive logarithmic evidence is too large or too small, the upper and lower limits are first truncated before the exponential calculation, and then the posterior probability conversion is performed.

[0123] Preferably, when a real flashover fault is determined based on the posterior probability, the location and evolution stage of the fault are output according to the spatial density distribution and temporal variation characteristics at the final moment, including: When the posterior probability Greater than or equal to the non-fault posterior probability When a true flashover fault is determined to have occurred, the posterior probability of the non-fault is the difference between the posterior probability and the actual fault. At the final moment of the event, i.e., dimensionless time At that time, the arc light density distribution and the thermal trace density distribution are added together to obtain the superimposed density distribution; Multiply the dimensionless path coordinates by the stacking density distribution to obtain the total path integral, and divide by the sum of the total path integrals of the stacking density distribution to obtain the locked stacking centroid. : The locked superposition centroid is mapped to the specific component interval divided by the spatial key node, and the corresponding component name is output as the location of the fault. Calculate the time second derivative of the first derivative of the natural cubic spline with respect to the dimensionless time; Within the open interval from zero to one, the moment when the maximum point of the second derivative of time is located is defined as the moment when the intrusion acceleration begins. The moment when the minimum point of the second derivative of time is located is defined as the start time of the intrusion lock. The evolutionary stage is identified by combining the start time and the end time: The non-fault posterior probability is the non-fault probability output that is complementary to the posterior probability.

[0124] The superimposed density distribution is the combined spatial distribution formed by adding the arc light density distribution and the thermal trace density distribution at the final moment of the event.

[0125] The locked centroid is the weighted average position of the stacking density distribution along the one-dimensional normalized expansion path.

[0126] The specific component interval is a path sub-interval divided by key nodes in adjacent spaces, and each sub-interval corresponds to a part on the device.

[0127] The location of the fault is the name of the component output based on the specific component range into which the locked superimposed centroid falls.

[0128] The second derivative of time is the rate of change obtained by differentiating the first derivative of the natural cubic spline with respect to dimensionless time, and is used to characterize the acceleration of intrusion propulsion.

[0129] The acceleration of the intrusion begins at the point where the second derivative of time reaches its maximum value in the time interval from zero to one, and is used to indicate the starting point when the intrusion changes from slow to accelerated.

[0130] The initiation of invasion and locking is the point where the second derivative of time reaches its minimum value in the time interval from zero to one, and is used to indicate the starting point where the invasion changes from accelerated advancement to locked deposition.

[0131] The evolution stage is a fault process stage divided according to the start time, the start time of intrusion acceleration, the start time of intrusion locking, and the termination time. It is used to describe the time sequence of fault occurrence and locking.

[0132] In specific implementation, the judgment strategy for the non-fault posterior probability comparison threshold needs to be clarified that the non-fault posterior probability is equal to 1 minus the posterior probability. Therefore, a posterior probability greater than or equal to the non-fault posterior probability is equivalent to a posterior probability greater than or equal to 0.5. 0.5 is taken as the default judgment threshold, and it is stipulated that when it is equal to 0.5, it is treated as a fault, so as to meet the principle of online monitoring that it is better to report early than to miss a report.

[0133] In specific implementation, for the processing of superposition density distribution and locking superposition centroid in weak signal scenarios, when the sum of arc light density distribution and thermal trace density distribution at the final moment of the event is less than the weak signal threshold, the result of the final single frame is not used directly. Instead, the average superposition density distribution in the last 10% time interval of the event is used to obtain the locking superposition centroid. If there is still no valid signal at the end, the process is backtracked to the most recent valid moment.

[0134] In practice, for the mapping from the locked superposition centroid to the specific component interval, it is necessary to first divide the one-dimensional normalized expansion path into continuous intervals using adjacent spatial key nodes, and then assign a corresponding component name to each interval; the high-voltage side discharge electrode tip to the discharge gap center corresponds to the gap start area, the discharge gap center to the root of the first umbrella skirt corresponds to the first crossing area, the root of the first umbrella skirt to the edge of the first umbrella skirt corresponds to the first umbrella skirt area, and the subsequent intervals correspond to the subsequent umbrella skirt areas and the grounding side hardware entrance area in sequence.

[0135] In practice, for the calculation and smoothing of the time second derivative, it is necessary to first construct the natural cubic spline first derivative based on the coordinates of the centroid of the arc on the unified time grid, and then continue to differentiate the smoothed first derivative to obtain the time second derivative; before differentiating, a short window smoothing can be performed again, with the window length preferably taken as 3 to 7 points, in order to suppress local jitter.

[0136] In practice, for the search and conflict resolution of maxima and minima, all local maxima and minima need to be searched in the open interval from zero to one. Then, the maxima with the largest amplitude is selected as the start time of intrusion acceleration, and the minima with the smallest amplitude after it is selected as the start time of intrusion locking. If the minima appears before the maxima, the minima is discarded and the search continues.

[0137] In specific implementation, the division of the evolution stages should be defined as follows: the time from the start of the intrusion acceleration to the start of the intrusion is defined as the arc initiation stage; the time from the start of the intrusion acceleration to the start of the intrusion locking is defined as the intrusion acceleration stage; and the time from the start of the intrusion locking to the end of the intrusion locking is defined as the intrusion locking stage. When the interval between the start of the intrusion acceleration and the start of the intrusion locking is less than 10% of the total event time, the latter two stages can be merged into the rapid locking stage.

[0138] It should be noted that the input and output parameters in the calculation formulas of this application are all dimensionless calculations performed through normalization processing. The formulas are all derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0139] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A fixed-gap lightning protection insulator flashover fault identification system, applied to a lightning protection device comprising a discharge electrode, a discharge gap, a first shed, a subsequent shed, grounding side hardware, and a multi-source sensing unit, wherein the multi-source sensing unit is used to acquire discharge current, radio frequency amplitude, visible image, and temperature; the fixed-gap lightning protection insulator flashover fault identification system includes a processor, characterized in that... The processor is configured to perform the following steps: The multi-source observation signals of the lightning protection device are synchronously acquired using the same event window, and the multi-source observation signals are uniformly compressed to a dimensionless time axis. The multi-source observation signals include the original discharge current, the original radio frequency amplitude, the original visible image, and the original fiber grating temperature. Extract the spatial key nodes of the lightning protection device to construct a one-dimensional normalized expansion path, and project the multi-source observation signal onto the one-dimensional normalized expansion path; The time-varying arc light density distribution and thermal trace density distribution are reconstructed on the one-dimensional normalized expansion path, respectively. Based on the time-varying trajectories of the arc light density distribution and the thermal trace density distribution, the area of ​​the hysteresis intrusion ring and the intrusion irreversibility are calculated. The dimensionless current and dimensionless radio frequency amplitude generation time excitation weights obtained by the compression mapping of the original discharge current and the original radio frequency amplitude are used, and the gap deviation energy is calculated by combining the arc density distribution and the thermal trace density distribution. The discreteness of the time-series evolution sequence of parameters within each single event is obtained to adaptively allocate evaluation weights. The posterior probability of the true flashover fault can be obtained by fusing the hysteresis intrusion ring area, the intrusion irreversibility, and the gap deviation. When a real flashover fault is determined based on the posterior probability, the location and evolution stage of the fault are output according to the spatial density distribution and temporal change characteristics at the final moment.

2. The fixed-gap lightning protection insulator flashover fault identification system according to claim 1, characterized in that, Multi-source observation signals of the lightning protection device are simultaneously acquired using the same event window, and these multi-source observation signals are uniformly compressed onto a dimensionless time axis, including: The original discharge current, the original radio frequency amplitude, the original visible image, and the original fiber grating temperature are collected. The sum of the absolute value of the original discharge current and the absolute value of the original radio frequency amplitude after compression mapping is calculated to form a composite excitation; The nearest local minima on the left and right sides of the composite excitation peak are respectively used as the start and end times of the event window; Subtract the difference between the current time and the start time, divide by the sum of the difference between the end time and the start time and a preset numerical stability term, and obtain the dimensionless time on the dimensionless time axis. Substituting the absolute values ​​of the original discharge current and the original radio frequency amplitude into a preset nonlinear compression function, the dimensionless current and the dimensionless radio frequency amplitude are obtained respectively. Using the image before the start time as the reference frame, the original visible image is processed by background difference absolute value processing and then input into the nonlinear compression function to obtain a dimensionless arc light image; The temperature rise of each measuring point in the multi-source sensing unit relative to the starting time is extracted and substituted into the nonlinear compression function to obtain the dimensionless temperature rise sequence of each measuring point.

3. A fixed-gap lightning protection insulator flashover fault identification system according to claim 2, characterized in that, Extracting the spatial key nodes of the lightning protection device to construct a one-dimensional normalized expansion path, and projecting the multi-source observation signals onto the one-dimensional normalized expansion path, including: A spatial polygonal path is constructed by sequentially selecting the key spatial nodes along the direction of the high-voltage side discharge electrode tip, the center of the discharge gap, the root of the first umbrella skirt, the edge of the first umbrella skirt, the subsequent umbrella skirts to the grounding side fitting entrance; The total path length is obtained by summing the spatial Euclidean distances between all adjacent spatial key nodes. For any two-dimensional image pixel coordinates or measurement point coordinates in space, calculate the shortest distance from it to each line segment of the spatial polyline path to determine the nearest projection point; The dimensionless path coordinates of the point are obtained by dividing the cumulative distance along the stated direction to the nearest projection point by the total path length.

4. A fixed-gap lightning protection insulator flashover fault identification system according to claim 3, characterized in that, Reconstructing the time-varying arc light density distribution and heat trace density distribution along the one-dimensional normalized expansion path, respectively, includes: Multiply the dimensionless arc light map by a two-dimensional sampling kernel function and then perform double spatial integration. Divide the result by the sum of the superposition of the double spatial integration of the full field of view of the dimensionless arc light map and the sampling path integration, and the sum of the numerical stability term, to obtain the arc light density distribution with an integral of one on the one-dimensional normalized expansion path. Using the interpolation basis functions of each measuring point on the one-dimensional normalized expansion path, the dimensionless temperature rise sequence of each measuring point is weighted and summed, and then divided by the integral of the weighted sum over the entire path interval and the sum of the numerical stability term, the heat trace density distribution with an integral of one on the one-dimensional normalized expansion path is obtained. Multiply the dimensionless path coordinates by the arc light density distribution and the thermal trace density distribution respectively, and integrate them over the entire path interval to obtain the corresponding arc light centroid coordinates and thermal trace centroid coordinates respectively.

5. A fixed-gap lightning protection insulator flashover fault identification system according to claim 4, characterized in that, Based on the time-varying trajectories of the arc light density distribution and the thermal trace density distribution, the area of ​​the hysteresis intrusion ring and the intrusion irreversibility are calculated, including: The coordinates of the arc centroid and the coordinates of the thermal trace centroid are used to construct a two-dimensional parametric state curve. The first and last points of the two-dimensional parametric state curve are closed, and the absolute value of the integral area enclosed by the closed curve is calculated as the area of ​​the hysteresis intrusion ring. Calculate the first-order natural cubic spline derivative of the arc centroid coordinates with respect to the dimensionless time, and extract the positive propagation component that is greater than zero; The intrusion irreversibility is obtained by dividing the full-time integral of the positive propulsion component by the sum of the full-time integral of the absolute value of the first derivative of the natural cubic spline and the numerical stability term.

6. A fixed-gap lightning protection insulator flashover fault identification system according to claim 5, characterized in that, The time excitation weight is generated using the dimensionless current and the dimensionless radio frequency amplitude, and the gap deviation energy is calculated by combining the arc density distribution and the thermal trace density distribution, including: The instantaneous dimensionless current is added to the dimensionless radio frequency amplitude, and then divided by the sum of the integral of the sum over the entire time period and the sum of the numerical stability term to obtain the time excitation weight. The second moment of the arc is obtained by multiplying the square of the dimensionless path coordinates by the arc light density distribution and taking the total path integral. Multiply the square of the dimensionless path coordinates by the heat trace density distribution and calculate the total path integral to obtain the second moment of the heat trace. The time excitation weight is multiplied by the second moment of the arc and the second moment of the thermal trace, and the integral over the entire time period is obtained to obtain the arc gap deviation energy and the thermal track gap deviation energy, which together constitute the gap deviation energy.

7. A fixed-gap lightning protection insulator flashover fault identification system according to claim 6, characterized in that, The discreteness of the time-series evolution sequence of internal parameters of each single event is obtained to adaptively allocate evaluation weights. The posterior probability of the true flashover fault is obtained by fusing the hysteresis intrusion ring area, the intrusion irreversibility, and the gap deviation, including: Under a unified time grid, the absolute difference sequence between the coordinates of the arc centroid and the coordinates of the thermal trace centroid at the same moment, the positive increment sequence calculated based on the arc centroid, and the temporal evolution sequence of the second moment of the arc and the second moment of the thermal trace are extracted to form the temporal evolution sequence. The absolute difference sequence, the positive increment sequence, the temporal evolution sequence of the arc second moment, and the temporal evolution sequence of the thermal trace second moment are each divided by the sum of their entire sequences and the sum of the numerical stability term, and then converted into four sets of independent discrete probability distributions. The information entropy of each independent discrete probability distribution is calculated, and the difference is obtained by subtracting the information entropy from each one. The difference is then divided by the sum of all the differences and the numerical stability term to obtain the evaluation weight corresponding to each parameter. Intermediate surrogate quantities are assigned to the hysteresis intrusion ring area, the intrusion irreversibility, the arc gap deviation energy, and the thermal trace gap deviation energy. The sum of each intermediate surrogate quantity and the numerical stability term is calculated and divided by the sum of the difference between each intermediate surrogate quantity and the numerical stability term. The logarithm of the quotient is then obtained to obtain the logarithmic dominance evidence for each. The logarithmic evidence is obtained by multiplying each piece of evidence by its corresponding evaluation weight and summing the results. The posterior probability is obtained by dividing one by the sum of one and the negative exponent of the natural constant, where the exponent of the negative exponent of the natural constant is the composite logarithmic evidence.

8. A fixed-gap lightning protection insulator flashover fault identification system according to claim 7, characterized in that, When a real flashover fault is determined based on the posterior probability, the location and evolution stage of the fault are output according to the spatial density distribution and temporal variation characteristics at the final moment, including: When the posterior probability is greater than or equal to the non-fault posterior probability, it is determined that the real flashover fault has occurred, and the non-fault posterior probability is the difference obtained by subtracting the posterior probability. At the final moment of the event, the arc light density distribution and the thermal trace density distribution are added together to obtain the superimposed density distribution; Multiply the dimensionless path coordinates by the superposition density distribution to obtain the total path integral, and divide by the sum of the total path integrals of the superposition density distribution to obtain the locked superposition centroid. The locked superposition centroid is mapped to the specific component interval divided by the spatial key node, and the corresponding component name is output as the location of the fault. Calculate the time second derivative of the first derivative of the natural cubic spline with respect to the dimensionless time; Within the open interval from zero to one, the moment when the maximum value of the second derivative of time is located is defined as the start time of the invasion acceleration, and the moment when the minimum value of the second derivative of time is located is defined as the start time of the invasion locking. The evolution stage is identified by combining the start time and the end time.

Images