A multi-partial discharge source fingerprint positioning method based on wideband radio frequency detection

Abstract

The present application relates to the technical field of partial discharge on-line monitoring of power equipment, and specifically discloses a multi-partial discharge source fingerprint positioning method based on wideband radio frequency detection, which comprises: judging whether there is a multi-path space-time coupling problem by calculating the flying point proportion and amplitude ratio variation coefficient; if so, identifying a jump event and analyzing the spatial regularity and change mode regularity to determine whether the multi-path coupling mode is stable; if so, positioning a hot spot area to encrypt a partial fingerprint library, and constructing a phase sub-fingerprint library through phase correlation; if not, constructing a time difference of arrival and particle filtering fusion positioning model, and analyzing the jump frequency stability to determine whether to adopt a periodic switching strategy. The present application can cover various discharge sources from quasi-static to high-speed movement by only a small amount of preset parameters through hierarchical judgment and adaptive processing, thereby improving the robustness and accuracy of multi-partial discharge source fingerprint positioning in a multi-path environment.

Description

Technical Field

[0001] This invention relates to the field of online monitoring technology for partial discharge in power equipment, and specifically to a multi-partial discharge source fingerprinting method based on broadband radio frequency detection. Background Technology

[0002] Partial discharge fingerprint localization achieves precise location of the discharge source by pre-establishing a location-signal feature mapping database, online acquisition and matching of measured features. Wideband radio frequency signals are widely used in fingerprint localization due to their rich time-frequency information.

[0003] However, within enclosed metal containers (such as GIS pipelines and transformer tanks), radio frequency propagation exhibits strong multipath characteristics. Furthermore, for mobile discharge sources (such as free metal particles), even minor changes in their motion and position can cause drastic, discontinuous jumps in multipath patterns (i.e., multipath spatiotemporal coupling). This leads to a severe mismatch between real-time fingerprint acquisition and the static fingerprint database, resulting in the following prominent problems: location flying points (random jumps in the location results of adjacent discharge pulses far exceeding physically possible speeds), trajectory breaks (the continuous motion trajectory of the mobile source is fragmented into discontinuous pieces), misclassification as multiple fixed sources (clustering algorithms mistakenly identify the mobile source's dwell points at different time periods as multiple independent fixed defects), and weak source flooding. In existing technologies, static fingerprint databases cannot adapt to multipath spatiotemporal coupling, while simple time difference of arrival (TDOA) positioning struggles with first-wave extraction in cases of severe multipath. Some improved solutions, such as particle filtering and blind source separation, are often designed for single scenarios, have numerous parameters, require manual intervention, and cannot adaptively handle a wide dynamic range from quasi-static to high-speed movement.

[0004] Therefore, the present invention provides a multi-partition source fingerprint localization method based on broadband radio frequency detection. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-partition discharge source fingerprint localization method based on broadband radio frequency detection to solve the above-mentioned background problems.

[0006] The objective of this invention can be achieved through the following technical solution: a multi-partition source fingerprint localization method based on broadband radio frequency detection, comprising the following steps: Spatial continuity analysis and multipath mode fluctuation analysis are performed on the historical fingerprint localization results of multiple local emission sources to determine whether there is a multipath spatiotemporal coupling problem. If such events exist, identify the transition events and perform spatial regularity analysis and change pattern regularity analysis on the multipath spatiotemporal coupling patterns of the transition events to determine whether the multipath spatiotemporal coupling patterns are stable. If the multipath spatiotemporal coupling mode is stable, hotspot areas are located based on the spatial regularity analysis results, local fingerprint databases are encrypted, and multipath jump modes are identified based on the change mode regularity analysis results. Through the correlation analysis between multipath jump modes and power frequency phases, jump stable power frequency phases are identified, and corresponding phase sub-fingerprint databases are constructed. If the multipath spatiotemporal coupling mode is unstable, a time difference of arrival (TDOA) and particle filter fusion positioning model is constructed, and the stability of the multipath jump frequency is analyzed. If it is stable, the stable occurrence period of the multipath jump is determined as the switching period, and the positioning model is switched according to the switching period. If it is unstable, the TDOA and particle filter fusion positioning model is used for direct positioning.

[0007] Furthermore, the method for determining whether a multipath spatiotemporal coupling problem exists is as follows: The location result sequence of multiple discharge events was continuously collected, and the coefficient of variation of the flying point ratio and amplitude ratio was calculated. The positioning results are sorted by time, the Euclidean distance and time interval between two adjacent positioning points are calculated, and the apparent velocity is obtained by proportionally calculating the Euclidean distance and time interval. The number of adjacent point pairs with apparent speeds greater than the physical limit speed is counted and compared with the total number of adjacent point pairs to obtain the flying point ratio. Select the two sensors closest to the two ends of the device, extract the peak amplitude of each discharge event of the two sensors, calculate the peak amplitude ratio of each discharge event of the two sensors, and integrate them into an amplitude ratio sequence according to time sequence; If the proportion of flying points is greater than the preset proportion and the amplitude ratio variation coefficient is greater than the preset variation coefficient, then it is determined that there is a multipath spatiotemporal coupling problem.

[0008] Furthermore, the method for determining whether the multipath spatiotemporal coupling mode is stable is as follows: For each discharge event: Calculate the change in amplitude ratio between each discharge event and the previous discharge event, and compare it with the global average amplitude ratio. If the change in amplitude ratio is greater than the global average amplitude ratio, mark the discharge event as a jump event, and record the spatial coordinates of each jump event. Density clustering is performed on the spatial coordinates of all jump events to analyze spatial regularity, and fingerprint feature change vectors before and after the jump are divided and clustered to analyze the regularity of change patterns. If the jumps have spatial regularity and the change patterns are regular, the multipath spatiotemporal coupling mode is stable; if the jumps do not have spatial regularity or the change patterns are not regular, the multipath spatiotemporal coupling mode is unstable.

[0009] Furthermore, the process of spatial regularity analysis is as follows: The neighborhood radius is set to λ / 4, where λ is the wavelength corresponding to the center frequency, and the minimum number of points is set to 3. For any two positioning points, calculate the Euclidean distance between the two positioning points; Iterate through each location point and count the number of location points within the neighborhood radius of the location point. If the number of location points within the neighborhood radius is greater than or equal to the minimum number of points, then mark the location point as a core point; otherwise, mark it as a boundary point. Initialize cluster number c=0, create a list of unvisited points, initially containing all localization points; Randomly select a location from the list of unvisited locations: If the location point p is a core point, create a new cluster, add the location point p to the new cluster, obtain all the location points within the neighborhood radius of the location point p, and add them to the new cluster in sequence. For each location point q, if q is a core point, add all the unvisited points within its neighborhood radius to the new cluster, and mark all the location points added to the cluster as visited. If the location point p is not a core point, temporarily mark it as noise and mark p as visited. The process of generating clusters is repeated until all locations have been visited, eventually forming multiple clusters. If the number of clusters formed is less than or equal to the predetermined number and the radius of each cluster is less than λ / 2, then the jump is determined to have spatial regularity.

[0010] Furthermore, the process of analyzing the regularity of the aforementioned change patterns is as follows: For each transition event, extract the average fingerprint feature vector of multiple events before the transition and the average fingerprint feature vector of multiple events after the transition, and calculate the change vector; The change vectors of all transition events are standardized and used as clustering features; The number of clusters K is determined using the elbow method. K cluster features are randomly selected as initial cluster centers. The Euclidean distance between each cluster feature and the K cluster centers is calculated. The cluster features are then assigned to the change vector cluster with the closest Euclidean distance. After all clustering features are assigned, the center of each cluster is recalculated. The assignment and update are repeated until the change in the cluster center is less than or equal to a preset threshold. The clustering ends, and multiple change vector clusters are obtained. The ratio of the average distance within a cluster to the distance between clusters of change vectors is calculated to obtain the profile coefficient. If the profile coefficient meets the requirements, the change pattern is determined to be regular.

[0011] Furthermore, the process of encrypting the local fingerprint database is as follows: The geometric center of each cluster obtained from spatial regularity analysis is used as the hot spot center, and the hot spot region is defined with the hot spot center as the center and a radius of λ / 4. Within the hotspot area, the spacing between fingerprint reference points is encrypted to λ / 8, and virtual fingerprints of newly added reference points are generated using inverse distance weighted interpolation, where the weight is the reciprocal of the square of the distance.

[0012] Furthermore, the identification method for the transition stable power frequency phase is as follows: The power frequency cycle is divided into multiple equally spaced phase intervals, and the occurrence frequency of each transition mode in each phase interval is statistically analyzed. The chi-square test is used to determine whether the frequency of occurrence in each phase interval is significantly higher than the global average occurrence rate. If so, it is marked as a strongly correlated interval. Collect the union of all strongly correlated intervals as the transition-stable power frequency phase of this transition mode.

[0013] Furthermore, the construction process of the time difference of arrival and particle filter fusion localization model is as follows: After wavelet denoising of the waveforms of each channel, the arrival time of the first wave is detected by constant false alarm rate. The time difference is calculated with the channel with the highest signal-to-noise ratio as a reference. The position of the single arrival time difference is solved by two-stage weighted least squares method. Initialize the particle set, with each particle containing three-dimensional position and velocity. Predict the position using a uniform motion model, update the particle weights based on the arrival time difference observations, and output the weighted average position after resampling.

[0014] Furthermore, the process of performing stability analysis on the multipath jump frequency is as follows: Mark each discharge event as a jump event, record the timestamp of the jump, calculate the time interval between adjacent jumps and integrate them into an interval sequence, and calculate the coefficient of variation of the interval sequence; If the coefficient of variation meets the requirements, the frequency of jumps is considered stable; otherwise, it is considered unstable.

[0015] Furthermore, the switching cycle is determined as follows: Take the mean of the jump time interval sequence, round it up to an integer number of power frequency cycles, and use it as the switching period; Maintain a counter that increments by one every power frequency cycle. When the counter is less than half a cycle, use static fingerprint positioning. When the counter is greater than or equal to half a cycle or a jump event is detected, switch to the time difference of arrival and particle filter fusion positioning model. Use a Kalman filter for smooth transition during the switch.

[0016] The beneficial effects of this invention are as follows: By employing a three-tiered approach—multipath coupling existence determination, mode stability analysis, and frequency stability analysis—the system automatically selects the most suitable localization strategy, avoiding the failure of a single method in complex environments and achieving hierarchical adaptive processing. The local fingerprint database is encrypted to reduce the feature gap, and the phase sub-fingerprint database decouples the multipath mode under different phases, which significantly reduces flying points caused by jumps, while maintaining the high accuracy of fingerprint positioning, and achieves accurate positioning in stable mode. The time difference of arrival and particle filter fusion localization model does not rely on a fingerprint database. It utilizes physical time difference and motion continuity to output a continuous trajectory even when the fingerprint is completely invalid. The frequency adaptive switching further balances accuracy and reliability in predictable jump scenarios, achieving accurate localization in unstable modes. It eliminates flying points and trajectory breaks, avoids the misjudgment of moving sources as multiple fixed sources, and eliminates interference from hardware errors and real high-speed motion through joint criteria, effectively solving the moving source problem. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart of a multi-partition discharge source fingerprint localization method based on broadband radio frequency detection according to the present invention; Figure 2 This is the logic diagram for determining whether a multipath spatiotemporal coupling problem exists in this invention. Detailed Implementation

[0019] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0020] Example: Please refer to Figure 1 As shown, the present invention provides a multi-partition discharge source fingerprint localization method based on broadband radio frequency detection. Addressing issues such as static fingerprint database failure, location flying points, and trajectory breaks caused by multipath spatiotemporal coupling of mobile discharge sources within a metal-enclosed container, the invention proposes a comprehensive method encompassing multipath coupling problem determination, mode stability analysis, adaptive processing, and fusion localization. Specifically, the method includes the following steps: Step 1: Perform spatial continuity analysis and multipath mode fluctuation analysis on the historical fingerprint localization results of multiple local discharge sources to determine whether there is a multipath spatiotemporal coupling problem; In step one, the process of determining whether a multipath spatiotemporal coupling problem exists includes: Obtain historical fingerprint location results for multiple partial discharge sources, including a sequence of location results for multiple discharge events, including timestamps, location coordinates, and original waveforms or waveform features of each channel; The first point to clarify is that the calculation of the flying point ratio is as follows: A location flypoint is defined as an unreasonable jump in the location position between two consecutive discharges that is much greater than the physically possible speed. The positioning results are sorted by time, the Euclidean distance and time interval between two adjacent positioning points are calculated, and the apparent velocity is obtained by proportionally calculating the Euclidean distance and time interval. Set the physical limit speed according to the equipment type, for example, 10 m / s in GIS and 1 m / s in transformer oil. Count the number of adjacent point pairs whose apparent speed is greater than the physical limit speed, and calculate the ratio with the total number of adjacent point pairs to obtain the flying point ratio. It is understandable that the physical meaning of the flying point ratio is: reflecting the degree of spatial jump in the positioning result. In the real physical world, the speed of the discharge power source (especially moving particles) is limited by the electric field force and the resistance of the medium, and there is an upper limit. When the positioning system outputs an incorrect position due to multipath abrupt changes, it often leads to the distance between adjacent points being much larger than the actual possible distance of movement, thus resulting in flying points. The higher the flying point ratio, the worse the temporal consistency of the positioning result, and the less reliable the system is. Secondly, it should be noted that the calculation of the amplitude ratio coefficient of variation is as follows: Select the two sensors closest to the two ends of the device, extract the peak amplitude of each discharge event of the two sensors, calculate the peak amplitude ratio of each discharge event of the two sensors, and integrate them into an amplitude ratio sequence according to time sequence; Calculate the coefficient of variation of the amplitude ratio sequence to obtain the amplitude ratio coefficient of variation; Understandably, the physical meaning of the amplitude ratio variation coefficient is that it reflects the severity of fluctuations in the multipath mode. The essence of multipath spatiotemporal coupling is that slight changes in the discharge attitude or position cause discontinuous jumps in the multipath superposition mode. These jumps are directly reflected in the amplitude ratio of the signals received by different sensors. When the multipath mode is stable, even if the discharge quantity changes, the amplitude ratio of the two sensors should remain basically constant (because the relative attenuation of the propagation path remains unchanged). When the amplitude ratio variation coefficient is large, it indicates that the multipath mode is changing rapidly. The flying point ratio is compared with the preset ratio, and the amplitude ratio variation coefficient is compared with the preset variation coefficient. If the flying point ratio is greater than the preset ratio and the amplitude ratio variation coefficient is greater than the preset variation coefficient, then it is determined that there is a multipath spatiotemporal coupling problem. Among them, the preset ratio is the threshold for judging the flying point ratio, which is set based on the actual motion characteristics of mobile sources in typical power equipment (GIS, transformer) and the error statistics of the positioning system. The preset coefficient of variation is the threshold for judging the amplitude ratio coefficient of variation, which is set based on the statistical distribution of a large amount of measured data from fixed and mobile sources; It should be noted that the criteria for determining the multipath spatiotemporal coupling problem are: multipath spatiotemporal coupling inevitably leads to both spatial discontinuity (flying point) in the positioning results and drastic fluctuations in the multipath pattern (amplitude ratio change), thereby eliminating misjudgments caused by a single factor such as sensor synchronization error or real high-speed motion. Step 2: If it exists, identify the jump event, and perform spatial regularity analysis and change pattern regularity analysis on the multipath spatiotemporal coupling mode of the jump event to determine whether the multipath spatiotemporal coupling mode is stable; In step two, the process of determining whether the multipath spatiotemporal coupling mode is stable includes: The location results of multiple discharge events were collected to obtain the location coordinate sequence, as well as the power frequency phase, multi-channel peak amplitude, and fingerprint feature vector corresponding to each discharge event; It should be noted that the stability of the multipath spatiotemporal coupling mode refers to whether the multipath jump phenomenon has a repeatable spatial regularity and a repeatable change pattern during the movement of the mobile power supply. A stable mode means that the jump mainly occurs in a fixed spatial region, and the fingerprint change characteristics before and after each jump are similar; an unstable mode means that the jump is random, without a fixed location or fixed pattern. The first point to clarify is that extracting transition events specifically involves: For each discharge event: Calculate the change in amplitude ratio between each discharge event and the previous discharge event, and compare it with the global average amplitude ratio. If the change in amplitude ratio is greater than the global average amplitude ratio, mark the discharge event as a jump event, and record the spatial coordinates of each jump event. Secondly, it should be noted that spatial regularity analysis and abrupt pattern clustering are specifically as follows: Perform DBSCAN clustering on the spatial coordinates of all transition events. Parameter settings: neighborhood radius is λ / 4, where λ is the wavelength corresponding to the center frequency. The center frequency refers to the main frequency center point of the working frequency band of the broadband RF sensor. For example, the wavelength corresponding to the center frequency at 1GHz is 0.3m, and the neighborhood radius is 0.075m. The minimum number of points is 3. The clustering process includes: For any two positioning points, calculate the Euclidean distance between the two positioning points; Iterate through each location point and count the number of location points within the neighborhood radius of the location point. If the number of location points within the neighborhood radius is greater than or equal to the minimum number of points, then mark the location point as a core point; otherwise, mark it as a boundary point. Initialize cluster number c=0, create a list of unvisited points, initially containing all localization points; Randomly select a location point p from the list of unvisited points: If the location point p is a core point, create a new cluster, add the location point p to the new cluster, obtain all the location points within the neighborhood radius of the location point p, and add them to the new cluster in sequence. For each location point q, if q is a core point, add all the unvisited points within its neighborhood radius to the new cluster, and mark all the location points added to the cluster as visited. If the location point p is not a core point, temporarily mark it as noise (it may be included by other core points later, but if it is not eventually covered by the neighborhood of any core point, it will remain as noise), and mark p as visited; Repeat this cluster generation process until all locations have been visited, eventually forming multiple clusters; If the number of clusters formed is less than or equal to the predetermined number, where the predetermined number is 3, and the radius (maximum interior point distance) of each cluster is less than λ / 2, then the jump is judged to have spatial regularity (i.e. concentrated in a few hot spots); otherwise, the spatial distribution is considered to be scattered and the pattern is unstable. It should be noted that the predetermined number is based on the following: In independent power equipment section application scenarios, the equipment length in these sections is usually 3 to 10 meters, and the number of possible electric field or structural abrupt change points is limited. Field measurement data shows that the spatial distribution of jump events caused by multipath spatiotemporal coupling usually does not exceed 3 concentrated areas. In the clustering algorithm, the setting of the minimum number of points will naturally filter out low-density isolated jump points. If the number of clusters exceeds 3, it usually means that the jump event forms 4 or more mutually separated high-density areas in space. At the algorithm level, this shows that the spatial distribution of data points is highly dispersed and cannot be explained by a few fixed hot spots. Classifying it as an unstable mode has statistical rationality. Thirdly, it should be noted that the analysis of the patterns of change shows consistency in fingerprint feature changes before and after the transition, specifically: For each transition event, extract the average fingerprint feature vector ffront of the 5 events before the transition and the average fingerprint feature vector fback of the 5 events after the transition, and calculate the change vector Δf = fback - ffront. K-means clustering is performed on the change vectors of all transition events. The clustering process includes: The change vectors of all transition events are standardized and used as clustering features; The number of clusters K is determined using the elbow method. K cluster features are randomly selected as initial cluster centers. The Euclidean distance between each cluster feature and the K cluster centers is calculated. The cluster features are then assigned to the change vector cluster with the closest Euclidean distance. After all clustering features are assigned, the center of each cluster is recalculated. The assignment and update are repeated until the change in the cluster center is less than or equal to a preset threshold. The clustering ends, and multiple change vector clusters are obtained. The silhouette coefficient is obtained by calculating the ratio of the average distance within a cluster to the distance between clusters of change vectors. If the silhouette coefficient is >0.5, the change pattern is considered to be regular; otherwise, the change pattern is considered to be random. If the jump has spatial regularity and the change pattern has regularity, then the multipath spatiotemporal coupling mode is determined to be stable; otherwise, if the jump does not have spatial regularity or the change pattern does not have regularity, then the multipath spatiotemporal coupling mode is determined to be unstable. It should be noted that the logic for determining the stability of multipath spatiotemporal coupling modes is as follows: spatial regularity reflects whether the jump occurs at a fixed geometric location (such as a standing wave node or a reflection boundary), which is an important characteristic that distinguishes multipath coupling from random noise; consistency of the change pattern reflects whether each jump has the same physical mechanism (for example, when a particle crosses a standing wave node, the amplitude ratio always changes from high to low). When both conditions are met, it indicates that the multipath jump is predictable and modelable. Step 3: If the multipath spatiotemporal coupling mode is stable, locate hotspot areas based on the spatial regularity analysis results, encrypt the local fingerprint database, identify multipath jump modes based on the change mode regularity analysis results, identify the jump stable power frequency phase through the correlation analysis between the multipath jump mode and the power frequency phase, and construct the corresponding phase sub-fingerprint database. In step three, the process of locating hotspot areas based on spatial regularity analysis results includes: Hotspot regions are spatial regions where multipath mutations occur frequently. The geometric center of each cluster obtained from spatial regularity analysis is the average value of the coordinates of all positioning points within the cluster, which is used as the hotspot center of the jump. With the hot spot center as the center and a radius of λ / 4, where λ is the wavelength corresponding to the center frequency, the hot spot region is defined. Within the hot spot region, multipath events occur frequently. In step three, the local fingerprint database encryption process includes: Within the hotspot area, the spacing is refined to λ / 8, where λ is the wavelength corresponding to the center frequency, to ensure that the maximum characteristic change between adjacent reference points is less than half of the standing wave node spacing. The positions of the refined reference points can be generated by using equally spaced grids within the hotspot area. Virtual fingerprints are generated using interpolation. For each new reference point: In the existing fingerprint database, find the k reference points (k=4-6) closest to the new reference point, and record their coordinates and fingerprint feature vectors as P. i and f i , where P i Let f be the coordinates of the i-th reference point out of k reference points. i Let i be the fingerprint feature vector of the i-th reference point out of k reference points; Using inverse distance weighting, the weight ω of the i-th reference point i =1 / d i 2 , where d i Let be the distance between the i-th reference point among the k reference points and the newly added reference point; The interpolated fingerprint is: Add the new reference points and their interpolated fingerprints to the fingerprint database; It should be noted that local fingerprint database encryption reduces the gap in the feature space, making it easier for the measured fingerprint to find reference points with similar features when the mobile source undergoes multipath transitions near hotspot areas. This reduces the probability of being forced to match distant locations (flying points) due to the lack of matching points. This method only adds a small number of reference points, has low computational overhead, but can effectively improve the localization continuity in stable mode. In step three, the process of identifying multipath jump patterns based on the regularity analysis results of change patterns includes: The K change vector clusters obtained from the regularity analysis of change patterns are defined as different jump patterns. Each jump pattern corresponds to a multipath jump behavior, such as the amplitude ratio changing from large to small or the time delay difference changing from positive to negative. Assign a label m=1,2,...,K to each transition pattern, record the transition pattern to which each transition event belongs, and calculate the average phase interval of each transition pattern. In step three, the identification process of the transition stable power frequency phase includes: For each transition event, record the power frequency phase at the time of its occurrence, and also record the mode label to which the transition event belongs; The phase interval from 0° to 360° is divided into 36 equally spaced intervals (one every 10°). Calculate the occurrence frequency of each transition mode within each phase interval, specifically: For any given phase interval: Count the number of transitions of each transition mode m that falls within the phase interval, and count the total number of discharges within the phase interval (regardless of whether a transition occurs). Calculate the ratio of the number of transitions of transition mode m that falls within the phase interval to the total number of discharges within the phase interval to obtain the occurrence frequency of transition mode m within the phase interval. For each transition pattern: Calculate the global average occurrence rate. For each phase interval, use the chi-square test to determine whether the occurrence frequency of the transition mode m in each phase interval is significantly higher than the global average occurrence rate. If so, the transition mode is strongly correlated with the phase interval and is marked as a strongly correlated interval. Collect all strongly correlated intervals and take the union (which may be multiple discontinuous intervals) as the transition stable power frequency phase of this transition mode; It should be noted that different transition modes correspond to different physical mechanisms (for example, particles are stable near the voltage zero-crossing point due to the minimum electric field force, and flip due to the maximum electric field force near the peak point). Through correlation analysis, it is possible to identify which phases are prone to which transition, thus providing a basis for phase grouping. In step three, the process of constructing the corresponding phase sub-fingerprint library includes: Using the boundary of the stable power frequency phase interval of each switching mode as the dividing point, 0°-360° is divided into several continuous sub-intervals. For example, if the stable power frequency phase interval is concentrated in 80°-100° and 260°-280°, the intervals are: [0°, 80°), [80°, 100°], [100°, 260°), [260°, 280°], [280°, 360°). When the equipment is shut down, a standard pulse is injected at each reference position, and the power frequency phase at the time of injection is recorded (which can be arbitrarily set by the synchronization signal). The fingerprint of each reference position is stored separately according to the phase window. For phase windows with no injected data, they can be temporarily left empty. For each measured pulse, extract its power frequency phase and determine its corresponding phase window; Nearest neighbor search is performed only in the sub-fingerprint database corresponding to the phase window (if the sub-database is empty, the adjacent window or the global database is used as a fallback), and the matching position is output. It should be noted that the purpose of constructing the phase sub-fingerprint database is to decouple multipath patterns under different phases, avoiding the use of fingerprints from non-sensitive windows within sensitive phase windows (the latter may lead to incorrect matching). For stable patterns, this method can significantly reduce flying points caused by transitions without abandoning the fingerprint database. Step 4: If the multipath spatiotemporal coupling mode is unstable, construct a time difference of arrival (TDOA) and particle filter fusion positioning model, and perform stability analysis on the frequency of multipath jumps. If it is stable, determine the stable occurrence period of multipath jumps as the switching period, and switch the positioning model according to the switching period. If it is unstable, use the time difference of arrival (TDOA) and particle filter fusion positioning model for direct positioning. In step four, the construction process of the time difference of arrival and particle filter fusion localization model includes: The first point to clarify is that single-time positioning based on arrival time difference is as follows: For each discharge pulse, waveforms from all sensor channels are acquired synchronously, and wavelet denoising is performed on the waveforms of each channel. Then, a constant false alarm rate detection method is used to adaptively find the first peak exceeding the threshold in the noisy background. The time corresponding to this peak is regarded as the time when the first wave arrives at the sensor. The channel with the highest signal-to-noise ratio is selected as the reference channel, and the time difference between the other channels and the reference channel is calculated. If the signal-to-noise ratio of a certain channel is too low to reliably detect the first wave, the channel is discarded, and only the remaining channels are used for subsequent positioning. Using at least three effective time differences (at least four sensors are required for three-dimensional positioning), combined with the known spatial coordinates of the sensors and the propagation speed of electromagnetic waves in the device medium, a set of equations describing the distance difference between the discharge source and each sensor is listed. The set of equations is solved by the two-stage weighted least squares method to obtain a preliminary spatial coordinate of the current discharge pulse. Secondly, it should be noted that particle filter tracking specifically refers to: Particle filtering is used to smooth the random error of single positioning due to time difference of arrival and to fill the gaps of no observation using motion models; First, determine the total number of particles, for example, five hundred. These particles are then uniformly and randomly generated throughout the entire spatial range where the device might appear (e.g., inside a GIS pipeline). Each particle represents a hypothetical discharge state, including its three-dimensional position and three-dimensional velocity, with the initial velocity set to zero. All particles have equal initial weights. Whenever the next power frequency cycle begins or a new discharge pulse is detected, the position and velocity of each particle are propagated forward according to the state of all particles in the previous moment, following the uniform motion model. Random process noise is added during propagation to simulate the unpredictable acceleration or turning of particles under the action of electric field force. After obtaining the single-shot localization result based on the arrival time difference of the current pulse, this result is used to adjust the weight of each particle. The spatial distance between the position of each particle and the single-shot localization result is calculated. Particles that are closer to each other receive a higher likelihood value, and thus a larger weight. The adjustment range of the weight also depends on the reliability of the single-shot localization result: if the signal-to-noise ratio of the first wave is higher than the threshold, the single-shot localization is more reliable, and the weight adjustment is significant; if the signal-to-noise ratio is lower than the threshold, the adjustment range is reduced, making the filter more reliant on prediction. Subsequently, the weights of all particles are normalized so that the sum of the weights is one. After several updates, the weights of a few particles will become very large, while the weights of most particles will approach zero. In order to maintain the diversity of particles, resampling is required: based on the current weight distribution of particles, a new set of particles is drawn, particles with high weights are copied multiple times, particles with low weights are eliminated, and after resampling, the weights of all new particles are set to be equal again. At each moment, the positions of all particles are weighted and averaged according to their weights to obtain the optimal position estimate at that moment. At the same time, the covariance matrix of the particle set can be calculated, and the size of its trace reflects the uncertainty of the estimate. This weighted average position is the final output positioning result. In step four, the process of determining whether the multipath transition frequency is stable includes: For each of the multiple discharge events collected, mark whether each discharge event is a transition event and record the timestamp of each transition event. Sort all transition events by time to obtain the time of occurrence of the transition, calculate the time interval between adjacent transitions, and integrate them into an interval sequence according to time order; Calculate the coefficient of variation of the interval sequence and compare it with the preset coefficient of variation. If the coefficient of variation is less than the preset coefficient of variation, the frequency of multipath jump is stable; otherwise, it is unstable. It should be noted that frequency stability means that multipath jumps have a periodic pattern and may be modulated by the power frequency voltage (e.g., occurring once every half cycle or every cycle). In this case, this pattern can be used to utilize a static fingerprint database during the jump interval and temporarily switch to a more robust positioning method when a jump occurs, thus balancing accuracy and reliability. Frequency instability, on the other hand, means that the jumps are completely random and unpredictable. In step four, the process of determining the period during which the multipath transition stabilizes as the switching period includes: Calculate the mean of the interval sequence, round it up to the nearest integer number of power frequency cycles, and use it as the switching cycle; Maintain a counter, starting from 0 and incrementing by 1 every power frequency cycle. When the counter is less than half a switching cycle, use static fingerprint positioning because the multipath has not yet changed and fingerprint matching may still be reliable. When the counter approaches the moment of the jump (e.g., the last few cycles) or when a jump event is actually detected, the localization model is switched to a fusion localization model of time difference of arrival and particle filter. During the switch, a Kalman filter is used for a smooth transition to avoid position jumps. If the frequency of multipath jumps is unstable, a localization model that combines time difference of arrival and particle filtering can be used for direct localization. It should be noted that the role of multipath jump frequency stability analysis and model switching is to enable the system to output continuous and usable positioning results even under the worst-case scenario of unstable multipath spatiotemporal coupling mode, and to automatically optimize the switching strategy according to the regularity of the jump frequency, so as to retain the accuracy advantage of fingerprint positioning to the maximum extent.

[0021] The technical solution and advantages of this application are as follows: Spatial continuity analysis and multipath pattern fluctuation analysis are performed on the historical fingerprint positioning results of multiple partial discharge sources to determine whether a multipath spatiotemporal coupling problem exists. If so, jump events are identified, and spatial regularity and change pattern regularity analysis are performed on the multipath spatiotemporal coupling patterns of the jump events to determine whether the multipath spatiotemporal coupling patterns are stable. If the multipath spatiotemporal coupling patterns are stable, hotspot areas are located based on the spatial regularity analysis results, local fingerprint databases are encrypted, and multipath jump patterns are identified based on the change pattern regularity analysis results. Through correlation analysis between multipath jump patterns and power frequency phases, stable power frequency phases with jumps are identified, and corresponding phase sub-fingerprint databases are constructed. If the multipath spatiotemporal coupling patterns are unstable, a time difference of arrival (TDOA) and particle filter fusion positioning model is constructed, and the stability of the multipath jump occurrence frequency is analyzed. If stable, the stable occurrence period of multipath jumps is determined as the switching period, and the positioning model is switched according to the switching period. If unstable, the TDOA and particle filter fusion positioning model is used for direct positioning. This invention determines the existence of multipath spatiotemporal coupling by calculating the proportion of flying points and the coefficient of variation of amplitude ratio. If a multipath coupling problem exists, it identifies jump events and analyzes their spatial regularity and change pattern regularity to determine whether the multipath coupling mode is stable. If stable, it locates hotspot areas, encrypts the local fingerprint database, and constructs a phase sub-fingerprint database through phase correlation. If unstable, it constructs a localization model fused with time difference of arrival and particle filtering, and analyzes the stability of the jump frequency to determine whether to adopt a periodic switching strategy. Through hierarchical judgment and adaptive processing, this invention can cover various discharge sources from quasi-static to high-speed moving sources with only a few preset parameters, improving the robustness and accuracy of multi-partition source fingerprint localization in multipath environments.

[0022] The embodiments of the present invention have been described in detail above, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made in accordance with the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A multi-partition source fingerprint localization method based on broadband radio frequency detection, characterized in that: Includes the following steps: Spatial continuity analysis and multipath mode fluctuation analysis are performed on the historical fingerprint localization results of multiple local emission sources to determine whether there is a multipath spatiotemporal coupling problem. If such events exist, identify the transition events and perform spatial regularity analysis and change pattern regularity analysis on the multipath spatiotemporal coupling patterns of the transition events to determine whether the multipath spatiotemporal coupling patterns are stable. If the multipath spatiotemporal coupling mode is stable, hotspot areas are located based on the spatial regularity analysis results, local fingerprint databases are encrypted, and multipath jump modes are identified based on the change mode regularity analysis results. Through the correlation analysis between multipath jump modes and power frequency phases, jump stable power frequency phases are identified, and corresponding phase sub-fingerprint databases are constructed. If the multipath spatiotemporal coupling mode is unstable, a time difference of arrival (TDOA) and particle filter fusion positioning model is constructed, and the stability of the multipath jump frequency is analyzed. If it is stable, the stable occurrence period of the multipath jump is determined as the switching period, and the positioning model is switched according to the switching period. If it is unstable, the TDOA and particle filter fusion positioning model is used for direct positioning.

2. The multi-partition source fingerprint localization method based on broadband radio frequency detection according to claim 1, characterized in that: The method for determining whether a multipath spatiotemporal coupling problem exists is as follows: The location result sequence of multiple discharge events was continuously collected, and the coefficient of variation of the flying point ratio and amplitude ratio was calculated. The positioning results are sorted by time, the Euclidean distance and time interval between two adjacent positioning points are calculated, and the apparent velocity is obtained by proportionally calculating the Euclidean distance and time interval. The number of adjacent point pairs with apparent speeds greater than the physical limit speed is counted and compared with the total number of adjacent point pairs to obtain the flying point ratio. Select the two sensors closest to the two ends of the device, extract the peak amplitude of each discharge event of the two sensors, calculate the peak amplitude ratio of each discharge event of the two sensors, and integrate them into an amplitude ratio sequence according to time sequence; If the proportion of flying points is greater than the preset proportion and the amplitude ratio variation coefficient is greater than the preset variation coefficient, then it is determined that there is a multipath spatiotemporal coupling problem.

3. The multi-partition source fingerprint localization method based on broadband radio frequency detection according to claim 1, characterized in that: The method for determining whether the multipath spatiotemporal coupling mode is stable is as follows: For each discharge event: Calculate the change in amplitude ratio between each discharge event and the previous discharge event, and compare it with the global average amplitude ratio. If the change in amplitude ratio is greater than the global average amplitude ratio, mark the discharge event as a jump event, and record the spatial coordinates of each jump event. Density clustering is performed on the spatial coordinates of all jump events to analyze spatial regularity, and fingerprint feature change vectors before and after the jump are divided and clustered to analyze the regularity of change patterns. If the jumps have spatial regularity and the change patterns are regular, the multipath spatiotemporal coupling mode is stable; if the jumps do not have spatial regularity or the change patterns are not regular, the multipath spatiotemporal coupling mode is unstable.

4. The multi-partition source fingerprint localization method based on broadband radio frequency detection according to claim 3, characterized in that: The process of spatial regularity analysis is as follows: The neighborhood radius is set to λ / 4, where λ is the wavelength corresponding to the center frequency, and the minimum number of points is set to 3. For any two positioning points, calculate the Euclidean distance between the two positioning points; Iterate through each location point and count the number of location points within the neighborhood radius of the location point. If the number of location points within the neighborhood radius is greater than or equal to the minimum number of points, then mark the location point as a core point; otherwise, mark it as a boundary point. Initialize cluster number c=0, create a list of unvisited points, initially containing all localization points; Randomly select a location from the list of unvisited locations: If the location point p is a core point, create a new cluster, add the location point p to the new cluster, obtain all the location points within the neighborhood radius of the location point p, and add them to the new cluster in sequence. For each location point q, if q is a core point, add all the unvisited points within its neighborhood radius to the new cluster, and mark all the location points added to the cluster as visited. If the location point p is not a core point, temporarily mark it as noise and mark p as visited; The process of generating clusters is repeated until all locations have been visited, eventually forming multiple clusters. If the number of clusters formed is less than or equal to the predetermined number and the radius of each cluster is less than λ / 2, then the jump is determined to have spatial regularity.

5. The multi-partition source fingerprint localization method based on broadband radio frequency detection according to claim 4, characterized in that: The process of analyzing the regularity of the aforementioned change patterns is as follows: For each transition event, extract the average fingerprint feature vector of multiple events before the transition and the average fingerprint feature vector of multiple events after the transition, and calculate the change vector; The change vectors of all transition events are standardized and used as clustering features; The number of clusters K is determined using the elbow method. K cluster features are randomly selected as initial cluster centers. The Euclidean distance between each cluster feature and the K cluster centers is calculated. The cluster features are then assigned to the change vector cluster with the closest Euclidean distance. After all clustering features are assigned, the center of each cluster is recalculated. The assignment and update are repeated until the change in the cluster center is less than or equal to a preset threshold. The clustering ends, and multiple change vector clusters are obtained. The ratio of the average distance within a cluster to the distance between clusters of change vectors is calculated to obtain the profile coefficient. If the profile coefficient meets the requirements, the change pattern is determined to be regular.

6. The multi-partition source fingerprint localization method based on broadband radio frequency detection according to claim 1, characterized in that: The process of encrypting the local fingerprint database is as follows: The geometric center of each cluster obtained from spatial regularity analysis is used as the hot spot center, and the hot spot region is defined with the hot spot center as the center and a radius of λ / 4. Within the hotspot area, the spacing between fingerprint reference points is encrypted to λ / 8, and virtual fingerprints of newly added reference points are generated using inverse distance weighted interpolation, where the weight is the reciprocal of the square of the distance.

7. The multi-partial discharge source fingerprint localization method based on broadband radio frequency detection according to claim 6, characterized in that: The identification method for the transition stable power frequency phase is as follows: The power frequency cycle is divided into multiple equally spaced phase intervals, and the occurrence frequency of each transition mode in each phase interval is statistically analyzed. The chi-square test is used to determine whether the frequency of occurrence in each phase interval is significantly higher than the global average occurrence rate. If so, it is marked as a strongly correlated interval. Collect the union of all strongly correlated intervals as the transition-stable power frequency phase of this transition mode.

8. The multi-partition source fingerprint localization method based on broadband radio frequency detection according to claim 1, characterized in that: The construction process of the time difference of arrival and particle filter fusion positioning model is as follows: After wavelet denoising of the waveforms of each channel, the arrival time of the first wave is detected by constant false alarm rate. The time difference is calculated with the channel with the highest signal-to-noise ratio as a reference. The position of the single arrival time difference is solved by two-stage weighted least squares method. Initialize a particle set, with each particle containing three-dimensional position and velocity. Predict the position using a uniform motion model, update the particle weights based on the arrival time difference observations, and output the weighted average position after resampling.

9. The multi-partition source fingerprint localization method based on broadband radio frequency detection according to claim 8, characterized in that: The process of performing stability analysis on the multipath jump frequency is as follows: Mark each discharge event as a jump event, record the timestamp of the jump, calculate the time interval between adjacent jumps and integrate them into an interval sequence, and calculate the coefficient of variation of the interval sequence; If the coefficient of variation meets the requirements, the frequency of jumps is considered stable; otherwise, it is considered unstable.

10. A multi-partition discharge source fingerprint localization method based on broadband radio frequency detection according to claim 9, characterized in that: The switching cycle is determined as follows: Take the mean of the jump time interval sequence, round it up to an integer number of power frequency cycles, and use it as the switching period; Maintain a counter that increments by one every power frequency cycle. When the counter is less than half a cycle, use static fingerprint positioning. When the counter is greater than or equal to half a cycle or a jump event is detected, switch to the time difference of arrival and particle filter fusion positioning model. Use a Kalman filter for smooth transition during the switch.

Images