Three-dimensional lightning TOA positioning method based on pulse signal efficient matching

Abstract

The invention discloses a three-dimensional lightning TOA positioning method based on pulse signal efficient matching, and the method comprises the following steps: 1, sequentially selecting each detection station as a current reference station, and carrying out the front and back expansion of the wave band data of all detection stations; step 2, carrying out cross-correlation operation on wave band data of the reference station and the non-reference station, and carrying out coarse alignment on wave band time; step 3, carrying out second time delay fine alignment on the coarse alignment data; step 4, realizing pulse matching optimization of multiple detection stations, and outputting a time cluster set after clustering; step 5, verifying whether each time cluster is a pulse set of the same radiation source, and rejecting unreasonable time clusters to obtain effective time clusters; step 6, collection and duplicate removal of effective time clusters of the whole system; and 7, carrying out space-time positioning and precision optimization on the radiation source. According to the invention, full utilization of wave band signals of each detection station, accurate time delay alignment and decentration matching of pulse signals can be realized, and the radiation source positioning number and the lightning channel continuity are improved.

Description

Technical Field

[0001] This invention belongs to the field of lightning detection and signal processing technology, and in particular relates to a three-dimensional lightning TOA localization method based on efficient pulse signal matching, which is suitable for high-precision localization of lightning radiation sources and fine structural reconstruction of lightning discharge channels. Background Technology

[0002] High spatiotemporal resolution three-dimensional fine-scale structure research of lightning is the core approach to revealing the physical mechanism of lightning. Three-dimensional fine-scale localization of VHF (30-300MHz) lightning radiation sources is the core means of analyzing lightning discharge processes (such as initial breakdown, K-process, and direct-drive leaders). The TOA (Time of Arrival) algorithm, as a key technology for three-dimensional lightning localization, has become one of the mainstream technologies for lightning localization due to its simple principle and high positioning accuracy. Since Proctor (1971) first applied TOA technology to lightning research, the algorithm has gradually combined with GPS clock synchronization, cross-correlation matching, and other technologies to achieve performance upgrades, forming mature systems such as the Lightning Location Array (LMA, Rison et al., 1999) and the Beijing Lightning Network (BLNET, Wang et al., 2016).

[0003] While progress has been made in existing TOA-related improvement techniques—for example, Lyu et al. (2014) combined a broadband interferometer with the TOA algorithm to increase the number of radiation sources, Liu et al. (2020) optimized pulse matching through Hilbert transform and three-step cross-correlation, and Wang et al. (2021) used empirical mode decomposition (EMD) to reduce noise interference—existing TOA methods have three key limitations that restrict their ability to detect the spatiotemporal continuity of lightning channels:

[0004] 1. Imbalanced signal utilization due to a single reference station: Existing TOA (Transit-Oriented Alignment) positioning technologies for three-dimensional lightning radiation sources all rely on a "single reference station" model. If the reference station does not receive a signal from a certain radiation source, positioning cannot be achieved even if five or more other detection stations successfully receive the signal.

[0005] 2. Fixed-time window omission of boundary pulses: Existing TOA algorithms use non-overlapping sliding windows (e.g., cross-correlation every 100μs in the 1ms band, Fan et al., 2018), but due to the rapid development of lightning (e.g., arrow leader velocities reaching 1.93×10⁻⁶), boundary pulses are missed. 7 The change in the position of the radiation source caused by m / s will produce a residual time delay, which will cause the pulse at the edge of the window to be submerged and unable to participate in subsequent pulse matching, resulting in the loss of pulse information.

[0006] 3. Reference station reliance on limited pulse matching results: Traditional pulse matching searches for pulses from other detectors centered on the pulse peak of the reference station. If the radiation source is only received by multiple detectors other than the reference station (e.g., received by the other 5 detectors but not by the reference station), its position and time information cannot be retrieved, resulting in insufficient radiation source capture capability.

[0007] Therefore, there is an urgent need for a three-dimensional lightning TOA localization method that can improve the utilization of band signals, avoid pulse omissions, and not rely on pulse matching from a single reference station. Summary of the Invention

[0008] The purpose of this invention is to provide a three-dimensional lightning TOA localization method based on efficient pulse signal matching. By using cyclical reference stations, sliding overlapping windows, and DBSCAN time clustering technology, the method achieves full utilization of band signals from each detection station, precise time delay alignment, and decentralized pulse signal matching, ultimately improving the number of radiation source locations and the continuity of lightning channels.

[0009] The technical solution adopted in this invention is:

[0010] A three-dimensional lightning TOA localization method based on efficient pulse signal matching includes the following steps:

[0011] Step 1: Using a cyclical reference station mechanism, select each detection station as the current reference station in sequence, and expand the band data of all detection stations before and after.

[0012] Step 2: Perform cross-correlation on the band data of the reference station and the non-reference station to coarsely align the band times between them;

[0013] Step 3: Taking into account the rapid development characteristics of lightning, for each reference station S i The k-th band and all non-reference stations S j nth j For each band of coarse alignment data, a sliding overlap window is used to perform a second time-delay fine alignment.

[0014] Step 4: Use the DBSCAN time clustering algorithm to optimize pulse matching among multiple detector stations, and output a set of time clusters after clustering;

[0015] Step 5: For each time cluster Two layers of screening are performed to verify whether they are pulse sets from the same radiation source, and unreasonable time clusters are eliminated to obtain valid time clusters;

[0016] Step 6: Collection and deduplication of valid time clusters across the entire system;

[0017] Step 7: Spatiotemporal positioning and accuracy optimization of radiation source.

[0018] Further, step 1 specifically involves: Let the set of detection stations be S = {S1, S2, ... S}. i …,S M}, i = 1, 2, ..., M, and M ≥ 6, cyclically select each detection station as a reference station, when S is selected i When the current reference station is used, it contains N raw bands, each corresponding to 1 ms (1,000,000 ns) of raw data. The sampling interval is Δt = 50 ns, therefore each raw band contains 20,000 sampling points, and the indices of each sampling point are n = 1, 2, ..., 20,000; the nth sampling point of the kth raw band corresponds to the GPS time t. k +(n-1)×50ns,t k The GPS timing start time for this original band is denoted as .

[0019] The k-th original band of the reference station is expanded, and the expanded band is denoted as . The expansion rule is to add 100μs sampling points before and after the band. 100μs corresponds to a duration of 100,000ns. Based on the sampling interval, 2,000 sampling points are added on each side. Therefore, the total number of sampling points in the expanded band is 24,000, corresponding to a total duration of 1.2ms, or 1,200,000ns.

[0020] During the extended processing, the indices of the 2000 sampling points corresponding to the first 100μs are m=1,2,...,2000. If there exists an adjacent and temporally continuous preceding band, i.e. k>1 and the end time of the (k-1)th original band seamlessly connects with the start time of the kth original band, specifically t k =t k-1 +1ms means that the signal values ​​of the 2000 sampling points are taken from the last 2000 sampling points of the (k-1)th original band, that is... If no such preceding band exists, including cases where k=1 or k>1 but the preceding and following bands are not time-discontinuous, then the signal values ​​of these 2000 sampling points are all 0; the indices of the 20000 sampling points corresponding to the original 1ms are m=2001,2002,...,22000, and their signal values ​​are directly taken from all sampling points of the k-th original band, i.e. The indices of the 2000 sampling points corresponding to the last 100μs are m = 22001, 22002, ..., 24000. If there exists an adjacent and temporally continuous next band, i.e., k < N and the end time of the kth original band is seamlessly connected to the start time of the (k+1)th original band, specifically t k+1 =t k+1ms means that the signal values ​​of the 2000 sampling points are taken from the first 2000 sampling points of the (k+1)th original band, i.e. If there is no such back band, including k=N or k<N but the time of the preceding and following bands is not continuous, then the signal values ​​of the 2000 sampling points are all 0;

[0021] For all non-reference stations S j Select band numbers n from the original band set that meet the following conditions. j This band With reference station S i If the starting time difference ΔT of the k-th original band is less than 1 ms, and the condition is met, then n j If there are two, then only keep n corresponding to the minimum value of ΔT. j (i.e., the band with the best time synchronization);

[0022] Subsequently, following the original expansion rules, the unique original band after filtering was selected. Expanded to:

[0023]

[0024] The corresponding sampling interval is 50ns, and the total duration is 1.2ms.

[0025] Further, step 2 specifically involves: based on GPS timing tags, using reference station S... i The kth extended band The GPS time corresponding to the starting sampling point is used as the reference time point to obtain the non-reference station S. j The nth j Extended bands The GPS time corresponding to the starting sampling point is determined, and the GPS time difference Δt between the two is calculated, where Δt = non-reference station S. j Extended band start GPS time - reference station S i The starting GPS time of the extended band, and then based on Δt... The time axis is shifted and adjusted: if Δt>0, that is, X j If the start time is later than the reference time point, then... Fill the gaps to the left of the reference time point on the time axis with zeros, and discard signal data to the right that exceeds the extended band length of the reference station; if Δt < 0, i.e. X j If the start time is earlier than the baseline time, it is discarded. The portion exceeding the reference time point on the time axis is filled with zeros, and the missing portion corresponding to the extended band length of the reference station on the right is filled with zeros, ultimately making... and The starting sampling points are all aligned with the reference time point, and the data lengths of the two are exactly the same;

[0026] by Using this as a baseline, calculate S for each non-reference station. j nth j Extended bands Generalized cross-correlation function

[0027]

[0028] Where d is the sampling point offset, which is an integer representing the number of offset sampling points of the non-reference station band relative to the reference station band: when d > 0, the non-reference station band is relatively lagging; when d < 0, the non-reference station band is relatively leading. This represents the signal value after offset by d sampling points in the non-reference station extended band;

[0029] turn up The offset d corresponding to the maximum value 1,ij The coarse alignment delay is τ. 1,ij =d 1,ij ×50ns, to obtain the non-reference station S j The nth j The coarsely aligned bands:

[0030]

[0031] Reference Station S i It does not require translation for alignment; the coarse alignment result of its k-th band is...

[0032]

[0033] Furthermore, step 3 specifically involves: based on the Euclidean distance d between the detection stations... ij Construct a fully connected graph G = (S, E), where nodes are probe stations and edge weights are the Euclidean distances between probe stations. Solve the minimum spanning tree (MST) using Kruskal's algorithm, and select the probe station pair with the furthest distance in the MST (S, E). P ,S q ), and their spacing d max =d pq ;

[0034] The Monte Carlo method was used to simulate the residual delay caused by rapid lightning development. Within a time window of 120 μs, the residual delay caused by rapid lightning development was less than 20 μs. The sliding overlap window parameters were set as follows: window length T. win =120μs, corresponding to the number of sampling points w = 120000ns / 50ns = 2400, and the overlap length of adjacent windows T over =20μs, corresponding to the number of sampling points o = 20000ns / 50ns = 400, the sliding step size corresponds to the number of sampling points sstep =wo=2400-400=2000, that is, slide forward 2000 sampling points each time, with adjacent windows overlapping by 400 sampling points, to achieve repeated extraction of 20μs data, and extract subbands according to window number p=1,2,...,10; subband extraction starts at point 1801 of the coarse alignment band of the reference station and extends to point 22200 to ensure complete coverage of the effective data of the coarse alignment band;

[0035] The sub-band extraction rules for reference stations are as follows:

[0036]

[0037] The subband extraction rules for non-reference stations are as follows:

[0038]

[0039] Where s = 1, 2, ..., 2400 are the sampling point indices within the sub-band window, corresponding to a total window length of 120 μs; when p = 1, 2, ..., 10, the sampling point index is determined by the sliding step size S. step In conjunction with the number of overlapping sampling points o, 400 sampling points are repeatedly extracted from adjacent sub-bands, and all sub-bands together cover the entire coarsely aligned effective band.

[0040] For the k-th band of the reference station and the n-th band of the non-reference station j Calculate the cross-correlation function for the p-th sub-band window of each band.

[0041]

[0042] Where d is the sampling point offset within the sub-band window, and the corresponding time delay τ = d × 50 ns;

[0043] turn up The offset d corresponding to the maximum value 2,ij Its corresponding fine delay τ 2,ij =d 2,ij ×50ns, based on offset d 2,ij A second fine-grained time delay compensation is performed on the non-reference station subband:

[0044]

[0045] The precise alignment result of the reference station subband is as follows:

[0046]

[0047] Furthermore, step 4 specifically involves: adjusting the reference station S. i The k-th band and the p-th precisely aligned sub-band All non-reference stations Sj The nth j The p-th precisely aligned sub-band of the band. Extract the pulses for each subband, s = 1, 2, ..., 2400, corresponding to a sampling interval Δt = 50 ns and a window length of 120 μs. Record the corresponding peak characteristics: refer to the peak sampling point index of the subband at the reference station. With peak amplitude Peak sampling point index for each non-reference station With peak amplitude

[0048] Reference Station S i The set of indices of all window pulse sampling points in the k-th band is All non-reference stations S j The nth j The set of indices of all window pulse sampling points in each band is Merge the pulse peak sampling point indices of all detection stations to form the dataset to be clustered:

[0049]

[0050] in, Reference station S i The set of pulse peak sampling point indices of all detection stations associated with the k-th band, including both reference and non-reference stations;

[0051] Using the DBSCAN algorithm Temporal clustering is performed with a neighborhood radius ε = 2 μs to ensure that pulses with a sampling point index difference less than or equal to 40 are grouped into the same time cluster, representing the set of sampling point indices corresponding to the peak pulse signals observed by different detectors from the same radiation source. The minimum number of points is set to MinPts = 3, and each time cluster contains at least 3 pulse peak sampling point indices from different detectors. The time cluster set is output after clustering.

[0052]

[0053] Where Q is the total number of time clusters obtained by clustering, and the q-th time cluster is... The definition of is:

[0054]

[0055] Here, l represents the number of pulses contained in the q-th time cluster, i.e., the total number of detection stations participating in this time cluster, a1, a2, ..., a l Indexes for different probe stations, including reference station S i Non-reference station S j , Indicates the probe station Index of the pulse peak sampling point in this time cluster.

[0056] Furthermore, step 5 specifically includes:

[0057] 1) Selection criteria for the number of detectors: retain clusters with a total number of detectors l≥5 within the retention time cluster, including reference stations and non-reference stations;

[0058] 2) Time-amplitude consistency test screening criteria: in time clusters In the middle, the peak times corresponding to all pulses are The corresponding sampling point index is Find the minimum time The peak amplitude corresponding to the detection station where the electromagnetic wave first arrives is Find the maximum moment The peak amplitude corresponding to the detection station where the electromagnetic wave arrives latest is [value missing].

[0059] Verification of the laws of physical propagation: Since the detection station closer to the radiation source receives the signal earlier and with a larger amplitude, it must meet the following conditions.

[0060] Pulse time clusters that pass the above two screening criteria are denoted as effective time clusters.

[0061] Furthermore, step 6 specifically includes:

[0062] 1) Collection of effective time clusters across the entire system:

[0063] Let the reference station S be... i The total number of bands that can be processed is N i Reference station S i The number of effective time clusters for the k-th band after filtering in step 5 is set to Q. i,k k = 1, 2, ..., N i All reference stations and all bands' valid time clusters are collected and combined to form a total list of valid time clusters:

[0064]

[0065] in, Reference station S i The q-th effective time cluster of the k-th band, with the superscript (i,k) clearly distinguishing the cluster origins of different reference stations and different bands;

[0066] 2) Deduplication of repeating time clusters:

[0067] For the total effective time cluster list C total Any two valid time clusters and Perform the following deduplication operation:

[0068] If at least one identical detection station S exists in two valid time clusters m S m A set of detectors that simultaneously belong to two valid time clusters, and whose pulse peak sampling point indices in both valid time clusters satisfy the following conditions: Then it is determined to be a repeating time cluster corresponding to the same radiation source;

[0069] For time clusters determined to be repetitive, calculate the sum of the peak pulse amplitudes of all detectors within each time cluster: Where l and l' represent the number of detectors contained in the two time clusters, respectively, the cluster with the larger sum of pulse peak amplitudes is retained. Then retain Otherwise, retain Remove the remaining duplicate time clusters;

[0070] 3) Output of the unique set of valid time clusters:

[0071] After deduplication, a unique set of valid time clusters without duplicate radiation sources is obtained:

[0072]

[0073] Among them, Q' i,k For reference station S i The number of effective time clusters after deduplication in the k-th band, where Q' i,k ≤Q i,k C unique Each time cluster corresponds to a different radiation source.

[0074] Furthermore, 7 specifically includes:

[0075] Extract C unique Each unique valid time cluster Key data from all probe stations: Cluster-wide probe station set {S m |m=1,2,…l}、Detection Station S m Sampling point index The peak pulse time corresponding to the sampling point index The three-dimensional coordinates (x, y) of the known probe station m ,y m ,z m ), l≥5;

[0076] Using the coordinates (x, y, z) of the radiation source and the occurrence time t0 as the unknowns, a system of nonlinear equations is established:

[0077]

[0078] Where m = 1, 2, ..., l, c = 3 * 10 8 m / s is the speed of light, t m The peak pulse time of the radiation source received by each detection station

[0079] The system of equations is solved using ordinary least squares. The objective function is constructed by minimizing the sum of squared residuals. The Levenberg-Marquardt algorithm is then used to iteratively optimize (x,y,z,t0) to minimize the sum of squared residuals. The chi-square test is then used to calculate χ². 2 Value, retaining χ² 2 Results ≤5 were discarded to remove values ​​with large positioning errors, ultimately yielding high-precision spatiotemporal coordinates of the radiation source.

[0080] Compared with the prior art, the beneficial effects of the present invention are:

[0081] 1. In view of the shortcomings of existing technologies, such as fixed reference stations not making full use of signals from other detection stations and rigid time window settings that easily lead to the omission of effective signals, this invention fully explores the effective signal resources of all detection stations through the collaborative optimization design of "dynamic rotating reference station mechanism and sliding overlapping time window", which greatly increases the number of effective locations of radiation sources and breaks through the signal utilization limitations of traditional algorithms.

[0082] 2. Compared with the problems of insufficient rationality and easy omission of matching in the pulse matching criteria of the prior art, the present invention proposes to perform cluster analysis on the pulse signals of each detection station based on DBSCAN time clustering, and constructs an efficient pulse matching criterion, which fundamentally optimizes the logic and process of pulse matching and realizes the efficient utilization of pulse signals from different detection stations.

[0083] 3. Relying on the collaborative technical support of the aforementioned dynamic reference station rotation, sliding overlap time window setting, and DBSCAN time clustering pulse matching, the algorithm channel reconstruction capability of this invention is significantly enhanced. Compared with the existing technology, it can more clearly restore the fine discharge evolution process of lightning multi-channel and multi-stage, with better spatial resolution and temporal continuity of the discharge structure, and significantly improve the accuracy and integrity of the restoration of lightning physical structure.

[0084] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0085] Figure 1 This is a flowchart of the algorithm of this invention;

[0086] Figure 2These are Hilbert envelope signal diagrams from different detection stations after the first time delay compensation.

[0087] Figure 3 This is a schematic diagram of the site distribution and lightning propagation within a small time window based on the minimum spanning tree;

[0088] Figure 4 It is a pulse matching waveform diagram based on DBSCAN time clustering and traditional algorithms;

[0089] Figure 5 This is a 3-D localization result image of the IC234819 lightning obtained using the traditional TOA algorithm;

[0090] Figure 6 This is a 3-D localization result image of the IC234819 lightning obtained using the TOA algorithm of this invention;

[0091] Figure 7 This is a comparison diagram of the TOA algorithm of this invention and the traditional algorithm for 3-D positioning in the 17ms band of Lightning IC234819. Detailed Implementation

[0092] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0093] A three-dimensional lightning TOA localization method based on efficient pulse signal matching includes the following steps:

[0094] Step 1: Using a cyclical reference station mechanism, select each detection station as the current reference station in sequence, and expand the band data of all detection stations before and after.

[0095] Step 2: Perform cross-correlation on the band data of the reference station and the non-reference station to coarsely align the band times between them;

[0096] Step 3: Taking into account the rapid development characteristics of lightning, for each reference station S i The k-th band and all non-reference stations S j nth j For each band of coarse alignment data, a sliding overlap window is used to perform a second time-delay fine alignment.

[0097] Step 4: Use the DBSCAN time clustering algorithm to optimize pulse matching among multiple detector stations, and output a set of time clusters after clustering;

[0098] Step 5: For each time cluster Two layers of screening are performed to verify whether they are pulse sets from the same radiation source, and unreasonable time clusters are eliminated to obtain valid time clusters;

[0099] Step 6: Collection and deduplication of valid time clusters across the entire system;

[0100] Step 7: Spatiotemporal positioning and accuracy optimization of radiation source.

[0101] Further, step 1 specifically involves: Let the set of detection stations be S = {S1, S2, ... S}. i …,S M}, i = 1, 2, ..., M, and M ≥ 6, cyclically select each detection station as a reference station, when S is selected i When the current reference station is used, it contains N raw bands, each corresponding to 1 ms (1,000,000 ns) of raw data. The sampling interval is Δt = 50 ns, therefore each raw band contains 20,000 sampling points, and the indices of each sampling point are n = 1, 2, ..., 20,000; the nth sampling point of the kth raw band corresponds to the GPS time t. k +(n-1)×50ns,t k The GPS timing start time for this original band is denoted as .

[0102] The k-th original band of the reference station is expanded, and the expanded band is denoted as .

[0103] The expansion rule is to add 100μs sampling points before and after the band. 100μs corresponds to a duration of 100,000ns. Based on the sampling interval, 2,000 sampling points are added on each side. Therefore, the total number of sampling points in the expanded band is 24,000, corresponding to a total duration of 1.2ms, or 1,200,000ns.

[0104] During the extended processing, the indices of the 2000 sampling points corresponding to the first 100μs are m=1,2,...,2000. If there exists an adjacent and temporally continuous preceding band, i.e. k>1 and the end time of the (k-1)th original band seamlessly connects with the start time of the kth original band, specifically t k =t k-1 +1ms means that the signal values ​​of the 2000 sampling points are taken from the last 2000 sampling points of the (k-1)th original band, that is... If no such preceding band exists, including cases where k=1 or k>1 but the preceding and following bands are not time-discontinuous, then the signal values ​​of these 2000 sampling points are all 0; the indices of the 20000 sampling points corresponding to the original 1ms are m=2001,2002,...,22000, and their signal values ​​are directly taken from all sampling points of the k-th original band, i.e. The indices of the 2000 sampling points corresponding to the last 100μs are m = 22001, 22002, ..., 24000. If there exists an adjacent and temporally continuous next band, i.e., k < N and the end time of the kth original band is seamlessly connected to the start time of the (k+1)th original band, specifically t k+1 =t k +1ms means that the signal values ​​of the 2000 sampling points are taken from the first 2000 sampling points of the (k+1)th original band, i.e. If there is no such back band, including k=N or k<N but the time of the preceding and following bands is not continuous, then the signal values ​​of the 2000 sampling points are all 0;

[0105] For all non-reference stations S j Select band numbers n from the original band set that meet the following conditions. j This band With reference station S i If the starting time difference ΔT of the k-th original band is less than 1 ms, and the condition is met, then n j If there are two, then only keep n corresponding to the minimum value of ΔT. j (i.e., the band with the best time synchronization);

[0106] Subsequently, following the original expansion rules, the unique original band after filtering was selected. Expanded to:

[0107]

[0108] The corresponding sampling interval is 50ns, and the total duration is 1.2ms.

[0109] Further, step 2 specifically involves: based on GPS timing tags, using reference station S... i The kth extended band The GPS time corresponding to the starting sampling point is used as the reference time point to obtain the non-reference station S. j The nth j Extended bands The GPS time corresponding to the starting sampling point is determined, and the GPS time difference Δt between the two is calculated, where Δt = non-reference station S. j Extended band start GPS time - reference station S i The starting GPS time of the extended band, and then based on Δt... The time axis is shifted and adjusted: if Δt>0, that is, X j If the start time is later than the reference time point, then... Fill the gaps to the left of the reference time point on the time axis with zeros, and discard signal data to the right that exceeds the extended band length of the reference station; if Δt < 0, i.e. X jIf the start time is earlier than the baseline time, it is discarded. The portion exceeding the reference time point on the time axis is filled with zeros, and the missing portion corresponding to the extended band length of the reference station on the right is filled with zeros, ultimately making... and The starting sampling points are all aligned with the reference time point, and the data lengths of the two are exactly the same;

[0110] by Using this as a baseline, calculate S for each non-reference station. j nth j Extended bands Generalized cross-correlation function

[0111]

[0112] Where d is the sampling point offset, which is an integer representing the number of offset sampling points of the non-reference station band relative to the reference station band: when d > 0, the non-reference station band is relatively lagging; when d < 0, the non-reference station band is relatively leading. This represents the signal value after offset by d sampling points in the non-reference station extended band;

[0113] turn up The offset d corresponding to the maximum value 1,ij The coarse alignment delay is τ. 1,ij =d 1,ij ×50ns, to obtain the non-reference station S j The nth j The coarsely aligned bands:

[0114]

[0115] Reference Station S i It does not require translation for alignment; the coarse alignment result of its k-th band is...

[0116] Furthermore, step 3 specifically involves: based on the Euclidean distance d between the detection stations... ij Construct a fully connected graph G = (S, E), where nodes are probe stations and edge weights are the Euclidean distances between probe stations. Solve the minimum spanning tree (MST) using Kruskal's algorithm, and select the probe station pair with the furthest distance in the MST (S, E). P ,S q ), and their spacing d max =d pq ;

[0117] The Monte Carlo method was used to simulate the residual delay caused by rapid lightning development. Within a time window of 120 μs, the residual delay caused by rapid lightning development was less than 20 μs. The sliding overlap window parameters were set as follows: window length T.win =120μs, corresponding to the number of sampling points w = 120000ns / 50ns = 2400, and the overlap length of adjacent windows T over =20μs, corresponding to the number of sampling points o = 20000ns / 50ns = 400, the sliding step size corresponds to the number of sampling points s step =wo=2400-400=2000, that is, slide forward 2000 sampling points each time, with adjacent windows overlapping by 400 sampling points, to achieve repeated extraction of 20μs data, and extract subbands according to window number p=1,2,...,10; subband extraction starts at point 1801 of the coarse alignment band of the reference station and extends to point 22200 to ensure complete coverage of the effective data of the coarse alignment band;

[0118] The sub-band extraction rules for reference stations are as follows:

[0119]

[0120] The subband extraction rules for non-reference stations are as follows:

[0121]

[0122] Where s = 1, 2, ..., 2400 are the sampling point indices within the sub-band window, corresponding to a total window length of 120 μs; when p = 1, 2, ..., 10, the sampling point index is determined by the sliding step size S. step In conjunction with the number of overlapping sampling points o, 400 sampling points are repeatedly extracted from adjacent sub-bands, and all sub-bands together cover the entire coarsely aligned effective band.

[0123] For the k-th band of the reference station and the n-th band of the non-reference station j Calculate the cross-correlation function for the p-th sub-band window of each band.

[0124]

[0125] Where d is the sampling point offset within the sub-band window, and the corresponding time delay τ = d × 50 ns;

[0126] turn up The offset d corresponding to the maximum value 2,ij Its corresponding fine delay τ 2,ij =d 2,ij ×50ns, based on offset d 2,ij A second fine-grained time delay compensation is performed on the non-reference station subband:

[0127]

[0128] The precise alignment result of the reference station subband is as follows:

[0129]

[0130] Furthermore, step 4 specifically involves: adjusting the reference station S. i The k-th band and the p-th precisely aligned sub-band All non-reference stations S j The nth j The p-th precisely aligned sub-band of the band. Extract the pulses for each subband, s = 1, 2, ..., 2400, corresponding to a sampling interval Δt = 50 ns and a window length of 120 μs. Record the corresponding peak characteristics: refer to the peak sampling point index of the subband at the reference station. With peak amplitude Peak sampling point index for each non-reference station With peak amplitude

[0131] Reference Station S i The set of indices of all window pulse sampling points in the k-th band is All non-reference stations S j The nth j The set of indices of all window pulse sampling points in each band is Merge the pulse peak sampling point indices of all detection stations to form the dataset to be clustered:

[0132]

[0133] in, Reference station S i The set of pulse peak sampling point indices of all detection stations associated with the k-th band, including both reference and non-reference stations;

[0134] Using the DBSCAN algorithm Temporal clustering is performed with a neighborhood radius ε = 2 μs to ensure that pulses with a sampling point index difference less than or equal to 40 are grouped into the same time cluster, representing the set of sampling point indices corresponding to the peak pulse signals observed by different detectors from the same radiation source. The minimum number of points is set to MinPts = 3, and each time cluster contains at least 3 pulse peak sampling point indices from different detectors. The time cluster set is output after clustering.

[0135]

[0136] Where Q is the total number of time clusters obtained by clustering, and the q-th time cluster is... The definition of is:

[0137]

[0138] Here, l represents the number of pulses contained in the q-th time cluster, i.e., the total number of detection stations participating in this time cluster, a1, a2, ..., a l Indexes for different probe stations, including reference station S i Non-reference station S j , Indicates the probe station Index of the pulse peak sampling point in this time cluster.

[0139] Furthermore, step 5 specifically includes:

[0140] 1) Selection criteria for the number of detectors: retain clusters with a total number of detectors l≥5 within the retention time cluster, including reference stations and non-reference stations;

[0141] 2) Time-amplitude consistency test screening criteria: in time clusters In the middle, the peak times corresponding to all pulses are The corresponding sampling point index is Find the minimum time The peak amplitude corresponding to the detection station where the electromagnetic wave first arrives is Find the maximum moment The peak amplitude corresponding to the detection station where the electromagnetic wave arrives latest is [value missing].

[0142] Verification of the laws of physical propagation: Since the detection station closer to the radiation source receives the signal earlier and with a larger amplitude, it must meet the following conditions.

[0143] Pulse time clusters that pass the above two screening criteria are denoted as effective time clusters.

[0144] Furthermore, step 6 specifically includes:

[0145] 1) Collection of effective time clusters across the entire system:

[0146] Let the reference station S be... i The total number of bands that can be processed is N i Reference station S i The number of effective time clusters for the k-th band after filtering in step 5 is set to Q. i,k k = 1, 2, ..., N i All reference stations and all bands' valid time clusters are collected and combined to form a total list of valid time clusters:

[0147]

[0148] in, Reference station S iThe q-th effective time cluster of the k-th band, with the superscript (i,k) clearly distinguishing the cluster origins of different reference stations and different bands;

[0149] 2) Deduplication of repeating time clusters:

[0150] For the total effective time cluster list C total Any two valid time clusters and Perform the following deduplication operation:

[0151] If at least one identical detection station S exists in two valid time clusters m S m A set of detectors that simultaneously belong to two valid time clusters, and whose pulse peak sampling point indices in both valid time clusters satisfy the following conditions: Then it is determined to be a repeating time cluster corresponding to the same radiation source;

[0152] For time clusters determined to be repetitive, calculate the sum of the peak pulse amplitudes of all detectors within each time cluster: Where l and l' represent the number of detectors contained in the two time clusters, respectively, the cluster with the larger sum of pulse peak amplitudes is retained. Then retain Otherwise, retain Remove the remaining duplicate time clusters;

[0153] 3) Output of the unique set of valid time clusters:

[0154] After deduplication, a unique set of valid time clusters without duplicate radiation sources is obtained:

[0155]

[0156] Among them, Q' i,k For reference station S i The number of effective time clusters after deduplication in the k-th band, where Q' i,k ≤Q i,k C unique Each time cluster corresponds to a different radiation source.

[0157] Furthermore, 7 specifically includes:

[0158] Extract C unique Each unique valid time cluster Key data from all probe stations: Cluster-wide probe station set {S m |m=1,2,…l}、Detection Station S m Sampling point index The peak pulse time corresponding to the sampling point index The three-dimensional coordinates (x, y) of the known probe stationm ,y m ,z m ), l≥5;

[0159] Using the coordinates (x, y, z) of the radiation source and the occurrence time t0 as the unknowns, a system of nonlinear equations is established:

[0160]

[0161] Where m = 1, 2, ..., l, c = 3 * 10 8 m / s is the speed of light, t m The peak pulse time of the radiation source received by each detection station

[0162] The system of equations is solved using ordinary least squares. The objective function is constructed by minimizing the sum of squared residuals. The Levenberg-Marquardt algorithm is then used to iteratively optimize (x,y,z,t0) to minimize the sum of squared residuals. The chi-square test is then used to calculate χ². 2 Value, retaining χ² 2 Results ≤5 were discarded to remove values ​​with large positioning errors, ultimately yielding high-precision spatiotemporal coordinates of the radiation source.

[0163] Example

[0164] The positioning system used in this embodiment consists of 6 detection stations (referred to as S1 to S6), which are located in Xuyi County and Hongze County of Huai'an City, with a distance between the detection stations ranging from 10 to 20 km.

[0165] Each detection station uses a uniform VHF receiver, with the following core parameters adjusted: center frequency 63MHz; bandwidth 5MHz; sampling frequency f. s =20MHz (1ms corresponds to 20,000 sampling points); GPS timing module accuracy ≤50ns (ensuring time synchronization between all detection stations); data storage is performed by dividing the raw data into "1ms per band".

[0166] This embodiment presents a three-dimensional lightning TOA localization method based on efficient matching of VHF pulse signals, combined with... Figure 1 This includes the following steps:

[0167] Step 1: This positioning system contains 6 stations, namely S1 to S6. Reference stations are selected in the rotation order of S1→S2→S3→S4→S5→S6. Taking S6 as the current reference station (i=6) as an example, the raw bands from k=1 to 749 are processed (the S6 detection station received a total of 749 1ms detection data).

[0168] For the kth original band of S6 (n = 1, 2, ..., 20000, corresponding to 20000 sampling points), the front end is stitched with the last 100μs of data from the (k-1)th band of S6 (padded with zeros if there is no preceding continuous band, corresponding to 2000 sampling points), and the back end is stitched with the initial 100μs of data from the (k+1)th band of S6 (padded with zeros if there is no following continuous band, corresponding to 2000 sampling points), to obtain the extended band. (m = 1, 2, ..., 24000, corresponding to 24000 sampling points);

[0169] For non-reference station S j (j = 1~5), select its nth digit. j n original bands j The condition is satisfied that "the time difference ΔT between the start of the k-th band of S6 and the band is less than 1 ms", and only the n corresponding to the minimum value of ΔT is retained. j Expanded according to the same rules to Ensure with Alignment with the relative time axis.

[0170] Step 2: Based on GPS time tags, The GPS time corresponding to the initial sampling point (m=1) is taken as the reference time point, and the non-reference station S is sampled using the method described above. j of With reference station S6 Time alignment and matching of data lengths; calculation of cross-correlation function. (Formula 2) By finding The maximum value yields the coarse alignment delay τ. 1,6j (e.g., S2 corresponds to τ) 1,62 = -14.4μs, S3 corresponds to τ 1,63 =-454.4μs); for Time delay compensation is performed to obtain the coarsely aligned band. Reference station S6 itself does not need to be translated for alignment, that is...

[0171] Step 3: After coarse band alignment (the Hilbert envelope of one band after coarse alignment corresponds to...) Figure 2 The coarse alignment result performs the following operations:

[0172] MST Construction: Calculate the Euclidean distance between each probe station, where the distance between S2 and S6 is the largest (d max =14.8km), which is the longest side of MST;

[0173] Monte Carlo simulation: Based on the characteristics of lightning development (simulation diagram corresponding to...) Figure 3 ), using arrow-shaped leader velocity V lead =1.93×107 m / s, generating 10 4 Simulation results for several radiation source events show that the maximum residual time delay is less than τ. max =20μs. Therefore, the window overlap length T is set. over =20μs.

[0174] Window parameter settings: Window length T win =120μs, the overlap length of adjacent windows is T over =20μs, the sliding step size corresponds to the number of sampling points s step =wo=2400-400=2000, and a total of 10 windows are extracted for each coarsely aligned band (p=1,2,...,10);

[0175] Sub-band extraction: for S6 Extract the reference station sub-band according to formula (4) For S j of Extract the non-reference station sub-bands according to formula (5).

[0176] Secondary cross-correlation: calculation (Formula 6) yields the precise alignment delay τ. 2,6j (e.g., the 5th window τ in S3) 2,63 =6.4μs), after compensation, we get

[0177] Step 4: Extract the peak sampling point index for each sub-band of S6. With peak amplitude Each non-reference station S j Peak sampling point index With peak amplitude Generate a dataset of all probe stations to be clustered in the k-th band of S6. (Formula 9);

[0178] The neighborhood radius ε = 2μs is set, meaning pulses with a sampling point index difference less than or equal to 40 are grouped into the same time cluster; the minimum number of points MinPts = 3, and DBSCAN is used to cluster the pulse clusters. After clustering, a time amplitude consistency check is performed on the successfully matched pulse clusters. One 120μs time window corresponds to the pulse matching waveform diagram based on DBSCAN time clustering and traditional algorithms. Figure 4 A total of 15 time clusters (Q=15) were obtained, marked with black pentagrams, corresponding to the pulse matching effect of the TOA optimized algorithm. The 7 pulse clusters successfully matched by the traditional TOA algorithm were marked with gray rhombuses. Clearly, the TOA algorithm used in this invention has better pulse matching capabilities than the traditional algorithm.

[0179] Step 5: Loop through all reference stations (S1~S6), collect valid time clusters, and remove duplicates according to the deduplication criteria. After deduplication, a unique cluster set C is obtained. unique (Formula 13).

[0180] Step 6: For C unique Extract the pulse peak time corresponding to the sampling point index of each detection station S1 to S6. And based on the three-dimensional coordinates (x) of the probe station m ,y m ,z m Establish a nonlinear system of equations (14), solve the system of equations using the ordinary least squares method, and use the Levenberg-Marquardt algorithm to iteratively optimize the coordinates of the radiation source and the occurrence time (x,y,z,t0), retaining χ. 2 The result ≤5 represents the final spatiotemporal coordinates of the radiation source.

[0181] This embodiment is based on the typical lightning event Cloud Flash IC234819 during a severe thunderstorm on August 15, 2023. By following the algorithm implementation steps described above, the localization results of the traditional TOA algorithm and the algorithm used in this invention are compared, and the effectiveness of the algorithm is verified from the dimensions of the number of radiation sources and the reconstruction of fine structure.

[0182] Cloud flashes are characterized by dispersed pulses and complex discharge structures. This embodiment uses this event to verify the algorithm's ability to locate the fine structure of cloud flashes. The location results correspond to... Figure 5 (Traditional TOA algorithm) Figure 6 (Algorithm of this invention) Figure 7 (Comparison of key positioning time periods at 17ms).

[0183] Traditional TOA algorithm: only identifies 1768 radiation sources. Due to the limitations of existing technology, it is easy to misjudge and miss lightning structures with weak radiation source signals (such as the initial breakdown pulse PB1).

[0184] The optimized algorithm of this invention successfully identified 5054 radiation sources, which is 2.9 times the number of traditional algorithms. It fully utilizes the full-band signal and significantly improves the ability to identify the fine structure of lightning channels (such as the initial breakdown pulse PB1).

[0185] contrast Figure 5 and Figure 6 This invention uses an algorithm to clearly reconstruct the complete discharge structure of the cloud flash, including the two initial breakdown pulses (PB1, PB2), the negative leader (C-NL), the K process, and other structures. It is particularly effective in reconstructing PB1 and C-NL.

[0186] PB1 (isolated initial breakdown): Occurs at approximately 16ms, and its location expands from east to west. Compared to the algorithm of this invention, the traditional algorithm, due to its lower signal utilization efficiency, only locates it to approximately [missing information]. The algorithm of this invention can effectively capture the propagation trajectory of the channel, with a propagation speed of approximately 3.5 × 10⁻⁶. 6 m / s, consistent with the initial breakdown velocity of cloud flash reported in publicly available literature; band comparison at the 17ms time interval shows ( Figure 7 The traditional algorithm can only locate 20 radiation sources, while the optimized algorithm can locate 56 (2.8 times) and cover all radiation sources of the traditional algorithm. The channel continuity is stronger, realizing the structural restoration of PB1 from "fragmentation to continuity".

[0187] C-NL (Negative Leader): Traditional algorithms can only identify "part" of the leader. The new algorithm restores its "C"-shaped path (first extending to the northwest, then turning to the northeast), forming a continuous and complete discharge channel, which conforms to the typical dipole charge structure of IC flash (upper negative leader, lower positive leader).

[0188] The optimized algorithm identifies 2.9 times more radiation sources than the traditional algorithm, and improves the localization capability of weak radiation sources (such as PB1) by more than 2.8 times, solving the core pain point of "low pulse signal utilization" of the traditional algorithm. The localization results of this algorithm can clearly restore the PB branch, C-NL, and "C"-shaped path lightning channel of the cloud lightning, and the structural continuity is better than that of the traditional algorithm.

[0189] In summary, this embodiment, through specific parameters, steps, and verification data, fully demonstrates that the present invention can effectively improve the utilization rate of lightning TOA positioning signals. The algorithm of the present invention is significantly better than the traditional TOA algorithm in terms of the number of lightning locations and the reconstruction of fine structures, and has engineering application value.

[0190] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A three-dimensional lightning TOA localization method based on efficient pulse signal matching, characterized in that, Includes the following steps: Step 1: Using a cyclical reference station mechanism, select each detection station as the current reference station in sequence, and expand the band data of all detection stations before and after. Step 2: Perform cross-correlation on the band data of the reference station and the non-reference station to coarsely align the band times between them; Step 3: Taking into account the rapid development characteristics of lightning, for each reference station S i The k-th band and all non-reference stations S j nth j For each band of coarse alignment data, a sliding overlap window is used to perform a second time-delay fine alignment. Step 4: Use the DBSCAN time clustering algorithm to optimize pulse matching among multiple detector stations, and output a set of time clusters after clustering; Step 5: For each time cluster Two layers of screening are performed to verify whether they are pulse sets from the same radiation source, and unreasonable time clusters are eliminated to obtain valid time clusters; Step 6: Collection and deduplication of valid time clusters across the entire system; Step 7: Spatiotemporal positioning and accuracy optimization of radiation source.

2. The three-dimensional lightning TOA localization method based on efficient pulse signal matching according to claim 1, characterized in that, Step 1 specifically involves: Let the set of detection stations be S = {S1, S2, ... S} i …,S M }, i = 1, 2, ..., M, and M ≥ 6, cyclically select each detection station as a reference station, when S is selected i When the current reference station is used, it contains N raw bands, each corresponding to 1 ms (1,000,000 ns) of raw data. The sampling interval is Δt = 50 ns, therefore each raw band contains 20,000 sampling points, and the indices of each sampling point are n = 1, 2, ..., 20,000; the nth sampling point of the kth raw band corresponds to the GPS time t. k +(n-1)×50ns,t k The GPS timing start time for this original band is denoted as . The k-th original band of the reference station is expanded, and the expanded band is denoted as . The expansion rule is to add 100μs sampling points before and after the band. 100μs corresponds to a duration of 100,000ns. Based on the sampling interval, 2,000 sampling points are added on each side. Therefore, the total number of sampling points in the expanded band is 24,000, corresponding to a total duration of 1.2ms, or 1,200,000ns. During the extended processing, the indices of the 2000 sampling points corresponding to the first 100μs are m=1,2,...,2000. If there exists an adjacent and temporally continuous preceding band, i.e. k>1 and the end time of the (k-1)th original band seamlessly connects with the start time of the kth original band, specifically t k =t k-1 +1ms means that the signal values ​​of the 2000 sampling points are taken from the last 2000 sampling points of the (k-1)th original band, that is... m = 1, 2, ..., 2000; if no such front band exists, including k = 1 or k > 1 but the time intervals of the preceding and following bands are discontinuous, then the signal values ​​of these 2000 sampling points are all 0; the indices of the 20000 sampling points corresponding to the original 1ms are m = 2001, 2002, ..., 22000, and their signal values ​​are directly taken from all sampling points of the kth original band, i.e. m = 2001, 2002, ..., 22000; the indices of the 2000 sampling points corresponding to the last 100μs are m = 22001, 22002, ..., 24000. If there exists an adjacent and time-continuous subsequent band, i.e., k < N and the end time of the kth original band is seamlessly connected to the start time of the (k+1)th original band, specifically t k+1 =t k +1ms means that the signal values ​​of the 2000 sampling points are taken from the first 2000 sampling points of the (k+1)th original band, i.e. m = 22001, 22002, ..., 24000. If there is no such back band, including k = N or k < N but the time of the front and back bands is not continuous, then the signal values ​​of the 2000 sampling points are all 0. For all non-reference stations S j Select band numbers n from the original band set that meet the following conditions. j This band With reference station S i If the starting time difference ΔT of the k-th original band is less than 1 ms, and the condition is met, then n j If there are two, then only keep n corresponding to the minimum value of ΔT. j That is, the band with the best time synchronization; Subsequently, following the original expansion rules, the unique original band after filtering was selected. Expanded to: The corresponding sampling interval is 50ns, and the total duration is 1.2ms.

3. The three-dimensional lightning TOA localization method based on efficient pulse signal matching according to claim 2, characterized in that, Step 2 specifically involves: based on GPS timing tags, using reference station S... i The kth extended band The GPS time corresponding to the starting sampling point is used as the reference time point to obtain the non-reference station S. j The nth j Extended bands The GPS time corresponding to the starting sampling point is determined, and the GPS time difference Δt between the two is calculated, where Δt = non-reference station S. j Extended band start GPS time - reference station S i The starting GPS time of the extended band, and then based on Δt... The time axis is shifted and adjusted: if Δt>0, that is, X j If the start time is later than the reference time point, then... Fill the gaps to the left of the reference time point on the time axis with zeros, and discard signal data to the right that exceeds the extended band length of the reference station; if Δt < 0, i.e. X j If the start time is earlier than the baseline time, it is discarded. The portion exceeding the reference time point on the time axis is filled with zeros, and the missing portion corresponding to the extended band length of the reference station on the right is filled with zeros, ultimately making... and The starting sampling points are all aligned with the reference time point, and the data lengths of the two are exactly the same; by Using this as a baseline, calculate S for each non-reference station. j nth j Extended bands Generalized cross-correlation function Where d is the sampling point offset, which is an integer representing the number of offset sampling points of the non-reference station band relative to the reference station band: when d > 0, the non-reference station band is relatively lagging; when d < 0, the non-reference station band is relatively leading. This represents the signal value after offset by d sampling points in the non-reference station extended band; turn up The offset d corresponding to the maximum value 1,ij The coarse alignment delay is τ. 1,ij =d 1,ij ×50ns, to obtain the non-reference station S j The nth j The coarsely aligned bands: Reference Station S i It does not require translation for alignment; the coarse alignment result of its k-th band is...

4. The three-dimensional lightning TOA localization method based on efficient pulse signal matching according to claim 3, characterized in that, Step 3 specifically involves: based on the Euclidean distance d between the detection stations ij Construct a fully connected graph G = (S, E), where nodes are probe stations and edge weights are the Euclidean distances between probe stations. Solve the minimum spanning tree (MST) using Kruskal's algorithm, and select the probe station pair with the furthest distance in the MST (S, E). P ,S q ), and their spacing d max =d pq ; The Monte Carlo method was used to simulate the residual delay caused by rapid lightning development. Within a time window of 120 μs, the residual delay caused by rapid lightning development was less than 20 μs. The sliding overlap window parameters were set as follows: window length T. win =120μs, corresponding to the number of sampling points w = 120000ns / 50ns = 2400, and the overlap length of adjacent windows T over =20μs, corresponding to the number of sampling points o = 20000ns / 50ns = 400, the sliding step size corresponds to the number of sampling points s step =wo=2400-400=2000, that is, slide forward 2000 sampling points each time, with adjacent windows overlapping by 400 sampling points, to achieve repeated extraction of 20μs data, and extract subbands according to window number p=1,2,…,10; subband extraction starts at point 1801 of the coarse alignment band of the reference station and extends to point 22200 to ensure complete coverage of the effective data of the coarse alignment band; The sub-band extraction rules for reference stations are as follows: The subband extraction rules for non-reference stations are as follows: Where s = 1, 2, ..., 2400 are the sampling point indices within the sub-band window, corresponding to a total window length of 120 μs; when p = 1, 2, ..., 10, the sampling point index is determined by the sliding step size S. step In conjunction with the number of overlapping sampling points o, 400 sampling points are repeatedly extracted from adjacent sub-bands, and all sub-bands together cover the entire coarsely aligned effective band. For the k-th band of the reference station and the n-th band of the non-reference station j Calculate the cross-correlation function for the p-th sub-band window of each band. Where d is the sampling point offset within the sub-band window, and the corresponding time delay τ = d × 50 ns; turn up The offset d corresponding to the maximum value 2,ij Its corresponding fine delay τ 2,ij =d 2,ij ×50ns, based on offset d 2,ij A second fine-grained time delay compensation is performed on the non-reference station subband: The precise alignment result of the reference station subband is as follows:

5. The three-dimensional lightning TOA localization method based on efficient pulse signal matching according to claim 4, characterized in that, Step 4 specifically involves: adjusting the reference station S... i The k-th band and the p-th precisely aligned sub-band All non-reference stations S j The nth j The p-th precisely aligned sub-band of the band. Extract the pulses for each subband, s = 1, 2, ..., 2400, corresponding to a sampling interval Δt = 50 ns and a window length of 120 μs. Record the corresponding peak characteristics: refer to the peak sampling point index of the subband at the reference station. With peak amplitude Peak sampling point index for each non-reference station With peak amplitude Reference Station S i The set of indices of all window pulse sampling points in the k-th band is All non-reference stations S j The nth j The set of indices of all window pulse sampling points in each band is Merge the pulse peak sampling point indices of all detection stations to form the dataset to be clustered: in, Reference station S i The set of pulse peak sampling point indices of all detection stations associated with the k-th band, including both reference and non-reference stations; Using the DBSCAN algorithm Temporal clustering is performed with a neighborhood radius ε = 2 μs to ensure that pulses with a sampling point index difference less than or equal to 40 are grouped into the same time cluster, representing the set of sampling point indices corresponding to the peak pulse signals observed by different detectors from the same radiation source. The minimum number of points is set to MinPts = 3, and each time cluster contains at least 3 pulse peak sampling point indices from different detectors. The time cluster set is output after clustering. Where Q is the total number of time clusters obtained by clustering, and the q-th time cluster is... The definition of is: Here, l represents the number of pulses contained in the q-th time cluster, i.e., the total number of detection stations participating in this time cluster, a1, a2, ..., a l Indexes for different probe stations, including reference station S i Non-reference station S j , Indicates the probe station Index of the pulse peak sampling point in this time cluster.

6. The three-dimensional lightning TOA localization method based on efficient pulse signal matching according to claim 5, characterized in that, Step 5 specifically includes: 1) Selection criteria for the number of detectors: retain clusters with a total number of detectors l≥5 within the retention time cluster, including reference stations and non-reference stations; 2) Time-amplitude consistency test screening criteria: in time clusters In the middle, the peak times corresponding to all pulses are The corresponding sampling point index is Find the minimum time The peak amplitude corresponding to the detection station where the electromagnetic wave first arrives is Find the maximum moment The peak amplitude corresponding to the detection station where the electromagnetic wave arrives latest is [value missing]. Verification of the laws of physical propagation: Since the detection station closer to the radiation source receives the signal earlier and with a larger amplitude, it must meet the following conditions. Pulse time clusters that pass the above two screening criteria are denoted as effective time clusters.

7. The three-dimensional lightning TOA localization method based on efficient pulse signal matching according to claim 6, characterized in that, Step 6 specifically includes: 1) Collection of effective time clusters across the entire system: Let the reference station S be... i The total number of bands that can be processed is N i Reference station S i The number of effective time clusters for the k-th band after filtering in step 5 is set to Q. i,k k = 1, 2, ..., N i All reference stations and all bands' valid time clusters are collected and combined to form a total list of valid time clusters: in, Reference station S i The q-th effective time cluster of the k-th band, with the superscript (i,k) clearly distinguishing the cluster origins of different reference stations and different bands; 2) Deduplication of repeating time clusters: For the total effective time cluster list C total Any two valid time clusters and Perform the following deduplication operation: If at least one identical detection station S exists in two valid time clusters m S m A set of detectors that simultaneously belong to two valid time clusters, and whose pulse peak sampling point indices in both valid time clusters satisfy the following conditions: Then it is determined to be a repeating time cluster corresponding to the same radiation source; For time clusters determined to be repetitive, calculate the sum of the peak pulse amplitudes of all detectors within each time cluster: Where l and l' represent the number of detectors contained in the two time clusters, respectively, the cluster with the larger sum of pulse peak amplitudes is retained. Then retain Otherwise, retain Remove the remaining duplicate time clusters; 3) Output of the unique set of valid time clusters: After deduplication, a unique set of valid time clusters without duplicate radiation sources is obtained: Among them, Q' i,k For reference station S i The number of effective time clusters after deduplication in the k-th band, where Q' i,k ≤Q i,k C unique Each time cluster corresponds to a different radiation source.

8. The three-dimensional lightning TOA localization method based on efficient pulse signal matching according to claim 7, characterized in that, 7 specifically includes: Extract C unique Each unique valid time cluster Key data from all probe stations: Cluster-wide probe station set {S m |m=1,2,…l}、Detection Station S m Sampling point index The peak pulse time corresponding to the sampling point index The three-dimensional coordinates (x, y) of the known probe station m ,y m ,z m ), l≥5; Using the coordinates (x, y, z) of the radiation source and the occurrence time t0 as the unknowns, a system of nonlinear equations is established: Where m = 1, 2, ..., l, c = 3 * 10 8 m / s is the speed of light, t m The peak pulse time of the radiation source received by each detection station The system of equations is solved using ordinary least squares. The objective function is constructed by minimizing the sum of squared residuals. The Levenberg-Marquardt algorithm is then used to iteratively optimize (x,y,z,t0) to minimize the sum of squared residuals. Finally, x is calculated using a chi-square test. 2 Value, retain x 2 Results ≤5 were discarded to remove values ​​with large positioning errors, ultimately yielding high-precision spatiotemporal coordinates of the radiation source.

Images