Lightning detection method based on traffic signal perturbation analysis

Abstract

This invention discloses a lightning detection method based on traffic signal disturbance analysis, belonging to the field of lightning monitoring technology. The method includes the following steps: Ⅰ. Acquiring traffic signals and performing multi-channel synchronous preprocessing; Ⅱ. Identifying pre-breakdown lightning based on the signal polarization disturbance tensor. Compared to traditional methods that only detect lightning that has already occurred, this invention can detect potential lightning events in advance, improving the timeliness of early warnings. It eliminates the need for dedicated active transmitting equipment, resulting in low construction costs, wide coverage, reduced false alarms and misjudgments, improved reliability and engineering application value of lightning detection results, and the ability to achieve three-dimensional spatial positioning of lightning discharge channels, improving positioning accuracy. It provides more comprehensive and detailed lightning activity information, supporting meteorological analysis and disaster assessment. This avoids the difficulties, high costs, and significant safety risks associated with traditional direct measurement methods, greatly enhancing the information acquisition and comprehensive analysis capabilities of lightning monitoring systems.

Description

Technical Field

[0001] This invention relates to the field of lightning monitoring technology, and more specifically to a lightning detection method based on operational signal disturbance analysis. Background Technology

[0002] Lightning is one of the most intense electrical discharge phenomena in nature. The transient high currents and strong electromagnetic pulses it generates can damage critical infrastructure such as power transmission lines, communication base stations, broadcasting and television systems, wind farms, airport navigation equipment, and petrochemical facilities. It can also cause equipment failures, data interruptions, and personal injury accidents. Therefore, real-time detection, location, and early warning of lightning activity are of great significance.

[0003] Existing lightning detection technologies mainly include ground electric field meters, lightning location networks, VLF / LF lightning monitoring, and weather radar monitoring. Ground electric field meters primarily determine lightning activity by monitoring changes in the electrostatic field; however, their detection range is limited and they are easily affected by terrain obstruction and local environmental interference. Lightning location systems typically rely on a network of dedicated detection stations, measuring the time difference of arrival of lightning radiation signals to achieve location, but this requires the construction of a large number of dedicated monitoring devices, resulting in high construction and maintenance costs. While weather radar can monitor the development of strong convective clouds, its ability to accurately locate lightning discharge channels and acquire lightning parameters is limited. Traditional VLF / LF lightning monitoring systems mainly rely on the lightning's own radiation signals for detection, and in complex electromagnetic environments, they are easily affected by electromagnetic interference from industrial equipment, power electronic devices, and communication equipment, thus reducing detection accuracy.

[0004] Meanwhile, with the widespread construction of broadcast television systems, mobile communication systems, private network communication systems, and Internet of Things (IoT) systems, a large number of UHF / VHF band service signals have covered most areas. These service signals are affected by electromagnetic field disturbances generated by lightning discharges during propagation, resulting in observable changes in signal amplitude, phase, polarization, and propagation delay. However, most existing technologies treat these changes as communication noise or propagation anomalies and suppress them, failing to fully exploit the lightning activity information contained within. This has led to existing wireless service network resources not effectively serving the field of lightning detection. Therefore, inventing and proposing a lightning detection method based on service signal disturbance analysis has become a crucial technical problem urgently needing to be solved in the field of lightning monitoring.

[0005] Existing lightning detection methods cannot detect potential lightning events in advance, require dedicated active transmitting equipment, which is costly, has a small coverage area, and suffers from a high incidence of false alarms and misjudgments. They also cannot achieve three-dimensional spatial positioning of lightning discharge channels, reducing the information acquisition and comprehensive analysis capabilities of lightning monitoring systems. Therefore, we propose a lightning detection method based on operational signal disturbance analysis. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies and provide a lightning detection method based on traffic signal disturbance analysis.

[0007] This invention proposes a lightning detection method based on traffic signal disturbance analysis. The technical solution adopted to solve the technical problem is as follows:

[0008] Ⅰ. Acquire service signals and perform multi-channel synchronous preprocessing: Deploy multiple receiving nodes in the monitoring area, and synchronously acquire UHF / VHF service signals and environmental electromagnetic disturbance signals through dual-polarized antennas and auxiliary sensing antennas. Then, preprocess the acquired signals to construct a multi-channel service signal dataset.

[0009] II. Identification of lightning pre-breakdown based on signal polarization perturbation tensor: Based on the preprocessed multi-channel service signal dataset, a polarization perturbation tensor is constructed, the perturbation energy in the corresponding frequency band is analyzed, and lightning pre-breakdown events are identified.

[0010] III. Construct a multi-source disturbance correlation matrix and eliminate false disturbances: Based on the identified lightning pre-breakdown events, construct a multi-source disturbance correlation matrix between the service signal and the auxiliary channel, extract the matrix feature parameters, and distinguish between real lightning disturbances and non-lightning events to eliminate false alarms.

[0011] IV. Statistical analysis of cross-entropy delay spectrum between each receiving channel to locate the three-dimensional space of lightning: Based on the selected real lightning disturbances, a passive positioning network is constructed through multiple service transmission sources and receiving nodes to calculate the cross-entropy delay spectrum between each receiving channel and reconstruct the location, spatial morphology, and development direction of the lightning discharge channel.

[0012] V. Intelligent classification of lightning types based on reconstructed lightning spatial information: Based on the reconstructed lightning spatial information, a complete disturbance feature evolution trajectory is constructed, and the trajectory is analyzed in time series. At the same time, the model parameters are dynamically corrected by combining GPS timing information and meteorological background data, and the probability distribution of different lightning types is output.

[0013] VI. Analyze the inversion characteristics of each service signal and estimate the peak intensity of lightning current: Analyze the transient frequency response characteristics of each service signal change and extract each disturbance parameter. Then, use the pre-established lightning current mapping model to convert each disturbance parameter into an estimated value of the peak intensity of lightning current.

[0014] VII. Lightning Event Fusion Assessment and Result Output: By integrating the lightning pre-breakdown identification results, authenticity verification results, spatial positioning results, lightning type identification results, and lightning current intensity estimation results, a complete lightning event information model is constructed; then, combined with risk assessment rules, the lightning hazard level is calculated, and the lightning occurrence time, spatial location, discharge direction, type, intensity, and early warning information are output.

[0015] As a further aspect of the present invention, the specific steps of constructing the polarization perturbation tensor, analyzing the perturbation energy in the corresponding frequency band, and identifying lightning pre-breakdown events in step II are as follows:

[0016] S1.1: Select the horizontal polarization channel and the vertical polarization channel under the same receiving node from the multi-channel service signal dataset. Then, convert the selected two service signal data into complex analytical signals by Hilbert transform. Extract the instantaneous phase of the two complex analytical signals and calculate the polarization phase difference between the two instantaneous phases. Then, based on the amplitude of the two complex analytical signals, calculate the polarization ellipticity.

[0017] S1.2: Extract the sampling interval of each service signal data in the multi-channel service signal data set, then count the first and second order variation features of polarization phase difference and polarization ellipticity between two sets of adjacent sampling points, and then combine the polarization phase difference, polarization ellipticity and its first and second order variation features in a fixed order to establish the corresponding polarization perturbation tensor.

[0018] S1.3: A sliding window Kalman filter is used to separate the high-frequency polarization abrupt component in the polarization perturbation tensor of each sampling point, and the separated high-frequency polarization abrupt component is decomposed in the short time frequency domain to obtain the frequency band energy in each time window. Then, the mean and standard deviation of the frequency band energy of multiple historical non-abnormal windows are statistically analyzed, and a corresponding adaptive threshold is established based on the mean and standard deviation of the frequency band energy.

[0019] S1.4: If the frequency band energy is greater than the adaptive threshold, the lightning pre-breakdown event is determined to be established, and the corresponding time window is marked as an abnormal window; if the frequency band energy is less than or equal to the adaptive threshold, it means that the pre-breakdown criterion has not been met, and the corresponding time window is marked as a non-abnormal window; and the adaptive threshold is updated in real time based on the latest determined non-abnormal window.

[0020] As a further aspect of the present invention, the specific steps of constructing the multi-source disturbance correlation matrix between the service signal and the auxiliary channel in step III, extracting the matrix feature parameters, and distinguishing between real lightning disturbances and non-lightning events to eliminate false alarms are as follows:

[0021] S2.1: Based on the identified lightning pre-breakdown events, determine the corresponding event trigger time, and then extract a complete event time window based on each event trigger time. This event time window includes the background disturbance before the event, the transient disturbance during the event, and the fallback process after the event ends. Then, calculate the average disturbance value and standard deviation of each group of data within each event time window.

[0022] S2.2: Based on the obtained average disturbance value and standard deviation, standardize the data of each group within the event time window to generate the corresponding disturbance sequence. Then, based on the disturbance sequences in each auxiliary channel, calculate the normalized correlation strength between each auxiliary channel in the event time window, and combine the correlation strength between the business channel and the three auxiliary channels into a multi-source disturbance correlation matrix.

[0023] S2.3: Extract the sum of the main diagonal elements from the multi-source perturbation correlation matrix and establish the corresponding matrix trace. Then extract the matrix eigenvalue set and calculate the eigenvalue dispersion. After that, calculate the average absolute value of the off-diagonal elements of the multi-source perturbation correlation matrix and integrate the four sets of data into a discriminant vector.

[0024] S2.4: Input the real-time acquired discrimination vectors into the offline trained support vector machine for discrimination. At the same time, the support vector machine outputs the discrimination score of the corresponding lightning pre-breakdown event based on each discrimination vector. If the discrimination score is >0, it means that the corresponding lightning pre-breakdown event is a real lightning disturbance and is marked as "passed the authenticity verification"; otherwise, it is a non-lightning disturbance and is removed, and will not enter the subsequent process.

[0025] As a further aspect of the present invention, the specific steps of step IV, which involve calculating the cross-entropy delay spectrum between each receiving channel and reconstructing the location, spatial morphology, and development direction of the lightning discharge channel, are as follows:

[0026] S3.1: Based on the lightning pre-breakdown event marked "passed authenticity verification", multiple service transmission sources and multiple receiving nodes are selected from the same area to form a passive positioning observation network, and each receiving node is set to synchronously collect the disturbance amplitude of lightning disturbance on different propagation paths around the same trigger time.

[0027] S3.2: The perturbation amplitude within each sampling window is processed by short-time framing to obtain the local energy in different frames and frequency bands. Then, the frequency band energy within the entire sampling window is normalized to obtain the probability quality of each receiving node in different frames and frequency bands and construct the probability distribution. Then, two sets of perturbation sequences of receiving nodes are randomly selected, and the time offset and frequency offset between the two sets of perturbation sequences are calculated. Time-frequency compensation is performed on the perturbation sequence of any receiving node.

[0028] S3.3: Establish a corresponding probability distribution for the time-frequency compensated perturbation sequence, then calculate the cross-entropy value between the uncompensated perturbation sequence and the time-frequency compensated perturbation sequence, and establish a corresponding cross-entropy time delay spectrum based on each cross-entropy value. Then, locate the joint peak position on the established cross-entropy time delay spectrum, and calculate the spectral quality in the neighborhood of each peak. Integrate the peak parameters of all receiving node pairs to form a joint observation vector.

[0029] S3.4: Establish a three-dimensional spatial grid within the target monitoring area and divide the entire space into multiple groups of candidate voxel elements. Simultaneously, determine the spatial coordinates of each receiving node in the three-dimensional spatial grid. Then, calculate the theoretical time difference between each voxel element and any receiving node pair based on the propagation geometry. Next, obtain the corresponding frequency difference characteristics through a pre-calibrated frequency difference mapping function. Based on the theoretical time difference and frequency difference characteristics of all grid voxel pairs and all receiving node pairs, establish a dictionary matrix.

[0030] S3.5: Statistically calculate the matching strength between each voxel unit and the actual discharge channel. Based on the observation vector, dictionary matrix, and matching strength, perform sparse reconstruction on each voxel unit to obtain the main distribution area of ​​the discharge channel in space. Integrate voxel units with matching strength exceeding the threshold into an active voxel set. Calculate the spatial centroid of the discharge channel based on the active voxels and their weights. Then, construct a weighted covariance matrix based on the spatial coordinates of each active voxel and its corresponding spatial centroid. Perform eigenvalue decomposition on the weighted covariance matrix and take the eigenvector corresponding to the largest eigenvalue as the main development direction of the discharge channel.

[0031] As a further aspect of the present invention, the specific steps of constructing a complete disturbance feature evolution trajectory based on the reconstructed lightning spatial information in step V are as follows:

[0032] S4.1: Based on the spatial location, main distribution area and main development direction of lightning, collect the corresponding perturbation sequence of each receiving node after alignment, and count the spatial distance from each receiving node to the reconstructed discharge center. At the same time, based on the obtained spatial distance, obtain the fusion weight of each receiving node. Then, construct the main perturbation sequence by using the aligned perturbation sequence and the fusion weight.

[0033] S4.2: Set a sliding background window and perform local background suppression on the main disturbance sequence based on the set sliding background window to obtain the net disturbance sequence after deducting the local background. Then, obtain the envelope sequence by analytical signal method, set the foreground trigger threshold, and record the rising edge starting point of the envelope sequence that first exceeds the foreground trigger threshold. Then, select the sampling point index corresponding to the highest envelope peak as the envelope peak point.

[0034] S4.3: Based on the envelope peak point and the rising edge start index, calculate the average steepness of the rising segment, and at the same time, perform linear fitting on the rising segment to obtain the corresponding slope coefficient. Then set the fall threshold, and the fall threshold < the foreground trigger threshold. Then, take the first fall of the envelope sequence to a position below the fall threshold as the end point of the perturbation fall. Then, calculate the total duration of the perturbation based on the end point of the perturbation fall and the rising edge start index.

[0035] S4.4: Identify each return stroke peak in the envelope sequence, calculate the time interval between them, then calculate the average return stroke interval in the current lightning event, then extract the local signal before and after the peak or during the fall-off phase, and perform discrete Fourier transform on it to obtain the complex spectrum values ​​on different frequency grids, and extract the amplitude at each frequency point, then calculate the dominant oscillation frequency of the disturbance based on each amplitude, and calculate the corresponding spectral centroid.

[0036] S4.5: After completing the analysis of rising edge, duration, return stroke interval and frequency, the current lightning event window is divided into multiple continuous sub-windows, and the rising edge steepness, duration, dominant oscillation frequency and average return stroke interval in each continuous sub-window are obtained. At the same time, each feature dimension is standardized, and the feature vectors of all sub-windows are arranged in chronological order to form a complete disturbance feature evolution trajectory.

[0037] As a further aspect of the present invention, the specific steps of step V, which involves dynamically correcting the model parameters by combining GPS timing information and meteorological background data to output the probability distribution of different lightning types, are as follows:

[0038] S5.1: Organize the feature vectors of each sub-window in the perturbation feature evolution trajectory into a unified time-series input in chronological order, and then stack all time segments in sequence to form an input sequence. Then, collect the current GPS timing information and meteorological background information and encode them into context vectors. Using the generated context vectors, establish a parameter correction representation jointly generated by GPS timing and meteorological background. Then, use this parameter correction representation to correct the gate biases of the input gate, forget gate, output gate and candidate state gate.

[0039] S5.2: After completing the parameter correction, based on the gating mechanism, obtain the output vectors of the input sequence at the input gate, forget gate, output gate and candidate state gate respectively, and update the current memory unit state based on the candidate state gate vector and forget gate vector. Then, generate the corresponding hidden state based on the output gate vector and the current memory unit state.

[0040] S5.3: After the entire input sequence has been processed based on the gating mechanism, the sequence-level representation of the current lightning event is obtained by averaging and converging each hidden state. Based on the obtained sequence-level representation and context vector, the discrimination scores of various lightning types are calculated. Then, the Softmax function is used to convert it into a probability distribution, and the lightning type with the highest probability is selected as the final recognition type of the current lightning event.

[0041] As a further aspect of the present invention, the lightning types mentioned in S5.3 specifically include cloud-to-ground lightning, cloud-to-cloud lightning, and air-to-air lightning.

[0042] As a further aspect of the present invention, the specific steps of step VI, which involves analyzing the transient frequency response characteristics of each service signal change and extracting each disturbance parameter, and then using a pre-established lightning current mapping model to convert each disturbance parameter into an estimated value of the lightning current peak intensity, are as follows:

[0043] S6.1: Based on the determination result of the lightning event, the event signal containing the lightning disturbance and the reference signal are extracted from the corresponding receiving channel. Then, the two signals are weighted by window functions to generate the windowing sequence of the event window and the reference window. Then, the event signal and the reference signal are analyzed in short time frequency domain to obtain the complex spectrum values ​​of each signal on different frequency grids. Based on the obtained complex spectrum values ​​on the frequency grids, the amplitude spectrum and phase spectrum of the event signal and the reference signal are obtained.

[0044] S6.2: After obtaining the spectrum of the event segment and the reference segment, when the event signal exists as a lightning disturbance, the output of the reference signal after a transient response is constructed, and the corresponding transient frequency response function is constructed. Then, based on the complex spectrum values ​​of various signals, the transient frequency response on the frequency grid is obtained through the transient frequency response function, and the transient frequency response is decomposed into frequency response amplitude and frequency response phase.

[0045] S6.3: Based on the frequency response amplitude and frequency response phase, the average amplitude compression caused by lightning disturbance and the additional phase shift depth in each frequency band are statistically analyzed. Then, the frequency response phase is frequency-differentiated to obtain the group delay estimate at each frequency point. Based on the group delay estimate, the average group delay of the target frequency band is obtained. Then, the additional phase shift depth, average amplitude compression, and average group delay are integrated into a unified input vector, and the input vector is normalized.

[0046] S6.4: Using a pre-established lightning current mapping model, the normalized input vector is converted into the peak intensity of lightning current. Based on the output peak intensity of lightning current, the estimated fluctuation degree of the current lightning current mapping model on the calibration data, and the confidence coefficient, a confidence interval for the corresponding peak intensity of lightning current is established. The model outputs the peak intensity of lightning current, the corresponding confidence interval, and the corresponding additional phase shift depth, average amplitude compression degree, and average group delay parameters.

[0047] The beneficial effects of this invention are:

[0048] This invention extracts horizontally and vertically polarized service signals from the same receiving node from a multi-channel service signal dataset. Complex analytic signals are obtained through Hilbert transform, and polarization phase difference, polarization ellipticity, and their first and second-order variation characteristics are calculated to construct a polarization perturbation tensor. Subsequently, a sliding window Kalman filter is used to separate high-frequency polarization abrupt components, and short-time frequency domain analysis is performed. An adaptive threshold is established using historical non-anomaly windows to identify lightning pre-breakdown events. Then, for the identified lightning events, a passive localization network is constructed using multiple service transmitters and receiving nodes. Time difference and frequency difference features are extracted through cross-entropy delay spectrum, and the spatial location of the lightning discharge channel is reconstructed using spatial grid compressed sensing. The model first identifies the location, distribution area, and development direction of lightning. Then, it integrates multi-node disturbance sequences to construct the main disturbance sequence. Features such as rising edge steepness, duration, return stroke interval, and dominant oscillation frequency are extracted, and a disturbance feature evolution trajectory is formed. The disturbance feature evolution trajectory is then input into a long short-term memory network. At the same time, GPS timing information and meteorological background information are introduced to dynamically correct the model parameters. Through time series analysis, the probability distribution of lightning types such as cloud-to-ground lightning, cloud-to-cloud lightning, and air-to-air lightning is output to achieve automatic identification of lightning types. Finally, the event signal and reference signal are extracted to invert the transient frequency response of the service propagation channel. Parameters such as additional phase shift depth, amplitude compression degree, and average group delay are extracted and an input vector is constructed. By using a pre-established lightning current mapping model to estimate the peak intensity of lightning current and generate corresponding confidence intervals, the system ultimately outputs the peak lightning current, confidence intervals, and related disturbance parameters. This enables passive intensity assessment of lightning events. Compared to traditional methods that only detect lightning that has already occurred, this approach can detect potential lightning events in advance, improving the timeliness of early warnings. It eliminates the need for dedicated active transmitting equipment, resulting in low construction costs, wide coverage, reduced false alarms and misjudgments, and improved reliability and engineering application value of lightning detection results. It also enables three-dimensional spatial positioning of lightning discharge channels, improving positioning accuracy and providing more comprehensive and detailed lightning activity information to support meteorological analysis and disaster assessment. This approach avoids the difficulties, high costs, and significant safety risks associated with traditional direct measurement methods, greatly enhancing the information acquisition and comprehensive analysis capabilities of lightning monitoring systems. Attached Figure Description

[0049] The present invention will now be further described with reference to the accompanying drawings.

[0050] Figure 1 This is a framework diagram of a lightning detection method based on traffic signal disturbance analysis. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] See Figure 1 This invention provides a lightning detection method based on traffic signal disturbance analysis. The method includes the following steps:

[0054] Acquire service signals and perform multi-channel synchronous preprocessing: Deploy multiple receiving nodes within the monitoring area to synchronously acquire UHF / VHF service signals and environmental electromagnetic disturbance signals through dual-polarized antennas and auxiliary sensing antennas. Then, preprocess the acquired signals to construct a multi-channel service signal dataset.

[0055] It should be further explained that multiple receiving nodes were deployed near broadcast television towers, communication base stations, wind turbine booster stations, and substations within the monitoring area. Each node is equipped with a dual-polarized receiving antenna, an auxiliary sensing antenna, a GPS timing module, and a signal acquisition terminal. The dual-polarized receiving antenna is used to simultaneously receive the horizontal and vertical polarization components of the same UHF / VHF service signal, the auxiliary sensing antenna is used to collect environmental electromagnetic disturbance signals, and the GPS timing module is used to provide a unified time reference.

[0056] After startup, each node synchronously receives broadcast television and FM radio signals within its area and continuously collects service signals and environmental disturbance signals. During the collection process, each set of data is accompanied by a GPS time stamp to achieve unified time synchronization across multiple nodes. Subsequently, the GPS second pulse signal is used to verify and correct the time stamps of each node, eliminating equipment clock deviations and ensuring that all node data are in a unified time coordinate system. Then, a node with higher signal quality is selected as a reference node, and the time delay between each node and the reference node is calculated through correlation analysis. Time shift compensation is then applied to the data to achieve time alignment of service signals across multiple nodes.

[0057] After time alignment, frequency calibration is performed on the service signals of each node. The system uses the stable pilot in the service signal as a frequency reference to detect frequency offsets in the received spectrum of each node, and eliminates local oscillator errors through digital down-conversion compensation, ensuring that the spectrum of each node is consistent. Subsequently, amplitude normalization processing is performed. The average received power of each channel under thunderstorm-free weather conditions is statistically analyzed and used as a unified reference to proportionally correct the signal amplitude of different nodes, thereby eliminating the effects of differences in antenna gain, feeder loss, and receiver parameters.

[0058] After normalization, a notch filter is used to suppress power frequency interference and its harmonic components, and a bandpass filter is used to retain the effective information of the service frequency band. In addition, wavelet denoising method is combined to remove random background noise and impulse interference, thereby improving signal quality.

[0059] Finally, the preprocessed horizontal polarization signals, vertical polarization signals, and auxiliary sensing signals from each node are fused and recombined along a unified time axis to construct a unified multi-channel service signal dataset. This dataset fully preserves the service signal propagation characteristics and environmental electromagnetic disturbance information, providing a reliable data foundation for subsequent polarization disturbance analysis, lightning pre-breakdown identification, and lightning event detection.

[0060] Identifying Lightning Pre-Breakdown Based on Signal Polarization Perturbation Tensor: Based on the preprocessed multi-channel service signal dataset, a polarization perturbation tensor is constructed, the perturbation energy in the corresponding frequency band is analyzed, and lightning pre-breakdown events are identified.

[0061] Specifically, from the multi-channel service signal dataset, the horizontally polarized channel and the vertically polarized channel under the same receiving node are selected. The selected two service signal data are then converted into complex analytic signals using Hilbert transform. The instantaneous phases of the two complex analytic signals are extracted, and the polarization phase difference between the two instantaneous phases is calculated. Then, based on the amplitudes of the two complex analytic signals, the polarization ellipticity is statistically analyzed. The sampling interval of each service signal data in the multi-channel service signal dataset is extracted. Next, the first-order and second-order variation characteristics of the polarization phase difference and polarization ellipticity between two sets of adjacent sampling points are statistically analyzed. Finally, the polarization phase difference, polarization ellipticity, and their first-order and second-order variation characteristics are combined in a fixed order to establish a corresponding polarization ellipticity distribution. The perturbation tensor is processed, and a sliding window Kalman filter is used to separate the high-frequency polarization abrupt component in the polarization perturbation tensor of each sampling point. The separated high-frequency polarization abrupt component is then decomposed in the short-time frequency domain to obtain the frequency band energy within each time window. Subsequently, the mean and standard deviation of the frequency band energy of multiple historical non-abnormal windows are statistically analyzed, and a corresponding adaptive threshold is established based on the mean and standard deviation of the frequency band energy. If the frequency band energy is greater than the adaptive threshold, the lightning pre-breakdown event is determined to be established, and the corresponding time window is marked as an abnormal window; if the frequency band energy is less than or equal to the adaptive threshold, it indicates that the pre-breakdown criterion has not been met, and the corresponding time window is marked as a non-abnormal window. The adaptive threshold is updated in real time based on the latest non-abnormal window.

[0062] Construct a multi-source disturbance correlation matrix and eliminate false disturbances: Based on the identified lightning pre-breakdown events, construct a multi-source disturbance correlation matrix between the service signal and the auxiliary channel, extract the matrix feature parameters, and distinguish between real lightning disturbances and non-lightning events to eliminate false alarms.

[0063] Specifically, based on the identified lightning pre-breakdown events, the corresponding event trigger times are determined. Then, a complete event time window is extracted based on each event trigger time. This event time window includes background disturbances before the event, transient disturbances during the event, and the settling process after the event. Next, the average disturbance value and standard deviation of each set of data within each event time window are calculated. Based on the obtained average disturbance value and standard deviation, the data within each event time window are standardized to generate corresponding disturbance sequences. Then, based on the disturbance sequences in each auxiliary channel, the normalized correlation strength between each auxiliary channel within the event time window is calculated. Finally, the correlation strength between the service channel and the three auxiliary channels is combined into a multi-source disturbance. The correlation matrix is ​​obtained by extracting the sum of the main diagonal elements from the multi-source disturbance correlation matrix and establishing the corresponding matrix trace. Then, the matrix eigenvalue set is extracted, and the eigenvalue dispersion is calculated. Subsequently, the average absolute value of the off-diagonal elements of the multi-source disturbance correlation matrix is ​​calculated. At the same time, the four sets of data are integrated into a discriminant vector. The discriminant vectors obtained in real time are input into the offline trained support vector machine for discrimination. The support vector machine outputs the discrimination score of the corresponding lightning pre-breakdown event based on each discriminant vector. If the discrimination score is > 0, it means that the corresponding lightning pre-breakdown event is a real lightning disturbance and is marked as "passed the authenticity verification"; otherwise, it is a non-lightning disturbance and is removed, and will not enter the subsequent process.

[0064] To locate lightning in three-dimensional space, the cross-entropy delay spectrum between each receiving channel is statistically analyzed: Based on the selected real lightning disturbances, a passive positioning network is constructed through multiple service transmitters and receiving nodes to calculate the cross-entropy delay spectrum between each receiving channel and reconstruct the location, spatial morphology, and development direction of the lightning discharge channel.

[0065] Specifically, based on lightning pre-breakdown events marked as "verified for authenticity," multiple service transmitters and multiple receiver nodes are selected from the same area to form a passive positioning and observation network. Each receiver node is configured to synchronously collect the disturbance amplitude of lightning along different propagation paths around the same trigger time. The disturbance amplitude within each sampling window is processed through short-time framing to obtain the local energy in different frames and frequency bands. Subsequently, the frequency band energy within the entire sampling window is normalized to obtain the probabilistic quality of each receiver node in different frames and frequency bands. The process involves constructing a probability distribution, then randomly selecting two sets of perturbation sequences from receiving nodes, simultaneously calculating the time and frequency offsets between the two sets of perturbation sequences, and performing time-frequency compensation on the perturbation sequence of any receiving node. A corresponding probability distribution is then established for the time-frequency compensated perturbation sequence. Next, the cross-entropy values ​​between the uncompensated and time-frequency compensated perturbation sequences are calculated, and a corresponding cross-entropy time delay spectrum is established based on each cross-entropy value. Finally, the joint peak position is located on the established cross-entropy time delay spectrum, and the spectral quality within the neighborhood of each peak is calculated. All receiving nodes are then... The peak parameters are integrated to form a joint observation vector. A three-dimensional spatial grid is established within the target monitoring area, and the entire space is divided into multiple groups of candidate voxel units. The spatial coordinates of each receiving node in the three-dimensional spatial grid are determined. Then, the theoretical time difference between each voxel unit and any receiving node pair is calculated according to the propagation geometry. The corresponding frequency difference features are obtained through a pre-calibrated frequency difference mapping function. Based on the theoretical time difference and frequency difference features of all grid voxels to all receiving node pairs, a dictionary matrix is ​​established. The matching strength between each voxel unit and the actual discharge channel is statistically analyzed. Based on the observation vector, dictionary matrix, and matching strength, each voxel unit is sparsely reconstructed to obtain the main distribution area of ​​the discharge channel in space. Voxel units with matching strength exceeding the threshold are integrated into an active voxel set. The spatial centroid of the discharge channel is calculated based on the active voxels and their weights. Then, based on the spatial coordinates of each active voxel and the corresponding spatial centroid, a weighted covariance matrix is ​​constructed. The weighted covariance matrix is ​​then eigenvalued, and the eigenvector corresponding to the largest eigenvalue is taken as the main development direction of the discharge channel.

[0066] Based on the reconstructed lightning spatial information, lightning types are intelligently classified: Based on the reconstructed lightning spatial information, a complete disturbance feature evolution trajectory is constructed, and the trajectory is analyzed in time series. At the same time, the model parameters are dynamically corrected by combining GPS timing information and meteorological background data, and the probability distribution of different lightning types is output.

[0067] Specifically, based on the spatial location, main distribution area, and main development direction of lightning, the aligned perturbation sequence of each receiving node is collected, and the spatial distance from each receiving node to the reconstructed discharge center is calculated. Simultaneously, based on the acquired spatial distance, the fusion weight of each receiving node is obtained. Then, using the aligned perturbation sequence and the fusion weight, a main perturbation sequence is constructed. A sliding background window is set, and local background suppression is applied to the main perturbation sequence based on the set sliding background window to obtain the net perturbation sequence after deducting the local background. Subsequently, the envelope sequence is obtained using an analytical signal method. A foreground trigger threshold is set, and the starting point of the rising edge when the envelope sequence first exceeds the foreground trigger threshold is recorded. Then, the sampling point index corresponding to the highest envelope peak is selected as the envelope peak point. Based on the envelope peak point and the rising edge starting point index, the average steepness of the rising segment is calculated. Simultaneously, linear fitting is performed on the rising segment to obtain the corresponding slope coefficient. Then, a fallback threshold is set, and the fallback threshold is less than the foreground trigger threshold. Finally, the... The first drop of the envelope sequence below the drop threshold is taken as the end point of the disturbance drop. Then, based on the index of the end point of the disturbance drop and the rising edge start point, the total duration of the disturbance is calculated. Each return stroke peak in the envelope sequence is identified, and the time interval between them is calculated. Then, the average return stroke interval in the current lightning event is calculated. After that, local signals before and after the peak or in the drop phase are extracted and subjected to discrete Fourier transform to obtain complex spectrum values ​​on different frequency grids. The amplitude at each frequency point is extracted. Then, based on each amplitude, the dominant oscillation frequency of the disturbance is calculated, and the corresponding spectral centroid is calculated. After completing the analysis of rising edge, duration, return stroke interval, and frequency, the current lightning event window is divided into multiple continuous sub-windows. The rising edge steepness, duration, dominant oscillation frequency, and average return stroke interval in each continuous sub-window are obtained. At the same time, each feature dimension is standardized. Then, the feature vectors of all sub-windows are arranged in chronological order to form a complete disturbance feature evolution trajectory.

[0068] Specifically, the feature vectors of each sub-window in the perturbation feature evolution trajectory are organized into a unified temporal input in chronological order. Then, all time segments are stacked sequentially to form an input sequence. Next, the current GPS timing information and meteorological background information are collected and encoded into context vectors. Using the generated context vectors, a parameter correction representation jointly generated by GPS timing and meteorological background is established. Subsequently, the parameter correction representation is used to correct the gate biases of the input gate, forget gate, output gate, and candidate state gate. After the parameter correction is completed, based on the gating mechanism, the output vectors of the input sequence at the input gate, forget gate, output gate, and candidate state gate are obtained respectively. The current memory unit state is updated based on the candidate state gate vector and forget gate vector. Then, the corresponding hidden state is generated based on the output gate vector and the current memory unit state. After the entire input sequence is processed based on the gating mechanism, the sequence-level representation of the current lightning event is obtained by averaging and converging each hidden state. Based on the obtained sequence-level representation and context vector, the discrimination scores of various lightning types are calculated. Then, the Softmax function is used to convert them into a probability distribution, and the lightning type with the highest probability is selected as the final identification type of the current lightning event.

[0069] Analyze the inversion characteristics of each service signal and estimate the peak intensity of lightning current: Analyze the transient frequency response characteristics of each service signal change and extract each disturbance parameter. Then, use the pre-established lightning current mapping model to convert each disturbance parameter into an estimated value of the peak intensity of lightning current.

[0070] Specifically, based on the lightning event determination result, the event signal containing the lightning disturbance and the reference signal are extracted from the corresponding receiving channel. Then, the two signal segments are weighted using window functions to generate windowing sequences for the event window and the reference window. Short-time frequency domain analysis is then performed on the event signal and the reference signal to obtain the complex spectrum values ​​of each signal on different frequency grids. Based on the obtained complex spectrum values ​​on the frequency grids, the amplitude and phase spectra of the event signal and the reference signal are obtained. After obtaining the spectra of the event segment and the reference segment, the output of the reference signal after a transient response when the event signal exists as a lightning disturbance is considered, and a corresponding transient frequency response function is constructed. Then, based on the complex spectrum values ​​of various signals, the transient frequency response on the frequency grid is obtained through the transient frequency response function, and the transient frequency response is decomposed into frequency response amplitude and frequency response phase. Based on the frequency response amplitude... The system calculates the average amplitude compression caused by lightning disturbance and the additional phase shift depth in each frequency band, along with the frequency response phase. It then performs frequency difference analysis on the frequency response phase to obtain group delay estimates at each frequency point. Based on these group delay estimates, it obtains the average group delay of the target frequency band. The additional phase shift depth, average amplitude compression, and average group delay are then integrated into a unified input vector. This input vector is normalized, and a pre-established lightning current mapping model is used to convert the normalized input vector into peak lightning current intensity. Based on the output peak lightning current intensity, the estimated fluctuation of the current lightning current mapping model on the calibration data, and the confidence coefficient, a confidence interval for the corresponding peak lightning current is established. The model outputs the peak lightning current intensity, the corresponding confidence interval, and the corresponding additional phase shift depth, average amplitude compression, and average group delay parameters.

[0071] Furthermore, as a further explanation of the present invention, the specific construction process of its lightning current mapping model is as follows: a large amount of historical lightning event data with known lightning current parameters are collected synchronously within the target monitoring area. The lightning current parameters can be obtained from lightning location networks, lightning current measurement devices, or authoritative meteorological databases. At the same time, UHF / VHF service signal data at the corresponding time are acquired. Subsequently, frequency response inversion analysis is performed on the service signals before and after each lightning event to extract channel disturbance parameters such as additional phase shift depth, amplitude compression coefficient, and propagation delay change, and these parameters are combined into feature vectors to form a "disturbance feature - lightning current peak value" sample library.

[0072] Next, outlier, missing, and noisy samples in the sample library are screened and corrected, and standardization is used to eliminate differences in the dimensions of different features. Then, the samples are divided into training, validation, and test sets according to a preset ratio. During the training phase, a lightning current mapping relationship is established using the perturbation feature vector as input and the corresponding measured peak lightning current as output. For scenarios with a small sample size, a multivariate nonlinear regression model can be used; for complex nonlinear scenarios, a BP neural network or a fully connected neural network can be used to construct the mapping model. The model parameters are continuously adjusted through error backpropagation to gradually reduce the mean square error between the predicted and measured lightning currents.

[0073] After model training, the model parameters are optimized using the validation set to determine the optimal network structure and hyperparameter configuration. Then, the model's generalization ability is evaluated using the test set. Once the prediction error meets the preset requirements, the final model parameters are saved to form a lightning current mapping model.

[0074] Meanwhile, in practical applications, the real-time acquired additional phase shift depth, amplitude compression coefficient, and propagation delay change are input into the trained lightning current mapping model to output the corresponding estimated peak lightning current intensity. Furthermore, the system can continuously update the sample library based on newly acquired calibrated lightning events and periodically perform incremental training on the model, continuously optimizing the lightning current mapping relationship and improving the estimation accuracy under different regional and seasonal conditions.

[0075] Lightning event fusion assessment and result output: By integrating the lightning pre-breakdown identification results, authenticity verification results, spatial positioning results, lightning type identification results, and lightning current intensity estimation results, a complete lightning event information model is constructed; then, combined with risk assessment rules, the lightning hazard level is calculated, and the lightning occurrence time, spatial location, discharge direction, type, intensity, and early warning information are output.

[0076] It should be further explained that after a lightning event occurs, service signal data within the range of 5ms before and 20ms after the lightning event are extracted from multiple receiving nodes closest to the discharge area as the analysis object, and reference service signal data under the lightning-free state before the event is simultaneously captured. Subsequently, frequency domain transformations are performed on the event signal and the reference signal respectively, and the transient frequency response characteristics of the service propagation channel during the lightning disturbance are inverted by comparing the spectral differences between the two.

[0077] During frequency response analysis, the system extracts key parameters such as additional phase shift depth, amplitude compression factor, and propagation delay variation within the target frequency band. These parameters are then combined to form a lightning current estimation eigenvector. Subsequently, each parameter is standardized to eliminate the influence between different dimensions.

[0078] After feature construction is completed, the system calls a pre-established lightning current mapping model. This model is trained on a large number of historical lightning event samples. Its inputs are perturbation parameters such as additional phase shift depth, amplitude compression coefficient, and propagation delay variation, and its output is the corresponding peak lightning current intensity. The system inputs the feature vector of the current event into the model for calculation to obtain the peak lightning current estimation result. At the same time, the peak lightning current, the confidence interval, and the perturbation parameters involved in the calculation are recorded in the lightning event database and stored in association with the spatial location, occurrence time, and lightning type information of the event, thus completing the passive estimation process of the peak lightning current intensity.

[0079] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A lightning detection method based on traffic signal disturbance analysis, characterized in that, Includes the following steps: Ⅰ. Acquire service signals and perform multi-channel synchronous preprocessing: Deploy multiple receiving nodes in the monitoring area, and synchronously acquire UHF / VHF service signals and environmental electromagnetic disturbance signals through dual-polarized antennas and auxiliary sensing antennas. Then, preprocess the acquired signals to construct a multi-channel service signal dataset. II. Identification of lightning pre-breakdown based on signal polarization perturbation tensor: Based on the preprocessed multi-channel service signal dataset, a polarization perturbation tensor is constructed, the perturbation energy in the corresponding frequency band is analyzed, and lightning pre-breakdown events are identified. III. Construct a multi-source disturbance correlation matrix and eliminate false disturbances: Based on the identified lightning pre-breakdown events, construct a multi-source disturbance correlation matrix between the service signal and the auxiliary channel, extract the matrix feature parameters, and distinguish between real lightning disturbances and non-lightning events to eliminate false alarms. IV. Statistical analysis of cross-entropy delay spectrum between each receiving channel to locate the three-dimensional space of lightning: Based on the selected real lightning disturbances, a passive positioning network is constructed through multiple service transmission sources and receiving nodes to calculate the cross-entropy delay spectrum between each receiving channel and reconstruct the location, spatial morphology, and development direction of the lightning discharge channel. V. Intelligent classification of lightning types based on reconstructed lightning spatial information: Based on the reconstructed lightning spatial information, a complete disturbance feature evolution trajectory is constructed, and the trajectory is analyzed in time series. At the same time, the model parameters are dynamically corrected by combining GPS timing information and meteorological background data, and the probability distribution of different lightning types is output. VI. Analyze the inversion characteristics of each service signal and estimate the peak intensity of lightning current: Analyze the transient frequency response characteristics of each service signal change and extract each disturbance parameter. Then, use the pre-established lightning current mapping model to convert each disturbance parameter into an estimated value of the peak intensity of lightning current. VII. Lightning Event Fusion Assessment and Result Output: By integrating the lightning pre-breakdown identification results, authenticity verification results, spatial positioning results, lightning type identification results, and lightning current intensity estimation results, a complete lightning event information model is constructed; then, combined with risk assessment rules, the lightning hazard level is calculated, and the lightning occurrence time, spatial location, discharge direction, type, intensity, and early warning information are output.

2. The lightning detection method based on traffic signal disturbance analysis according to claim 1, characterized in that, The specific steps for constructing the polarization perturbation tensor, analyzing the perturbation energy in the corresponding frequency band, and identifying lightning pre-breakdown events as described in Step II are as follows: S1.1: Select the horizontal polarization channel and the vertical polarization channel under the same receiving node from the multi-channel service signal dataset. Then, convert the selected two service signal data into complex analytical signals by Hilbert transform. Extract the instantaneous phase of the two complex analytical signals and calculate the polarization phase difference between the two instantaneous phases. Then, based on the amplitude of the two complex analytical signals, calculate the polarization ellipticity. S1.2: Extract the sampling interval of each service signal data in the multi-channel service signal data set, then count the first and second order variation features of polarization phase difference and polarization ellipticity between two sets of adjacent sampling points, and then combine the polarization phase difference, polarization ellipticity and its first and second order variation features in a fixed order to establish the corresponding polarization perturbation tensor. S1.3: A sliding window Kalman filter is used to separate the high-frequency polarization abrupt component in the polarization perturbation tensor of each sampling point, and the separated high-frequency polarization abrupt component is decomposed in the short time frequency domain to obtain the frequency band energy in each time window. Then, the mean and standard deviation of the frequency band energy of multiple historical non-abnormal windows are statistically analyzed, and a corresponding adaptive threshold is established based on the mean and standard deviation of the frequency band energy. S1.4: If the frequency band energy is greater than the adaptive threshold, the lightning pre-breakdown event is determined to be established, and the corresponding time window is marked as an abnormal window; if the frequency band energy is less than or equal to the adaptive threshold, it means that the pre-breakdown criterion has not been met, and the corresponding time window is marked as a non-abnormal window; and the adaptive threshold is updated in real time based on the latest determined non-abnormal window.

3. The lightning detection method based on traffic signal disturbance analysis according to claim 2, characterized in that, Step III involves constructing a multi-source disturbance correlation matrix between service signals and auxiliary channels, extracting matrix feature parameters, and distinguishing between real lightning disturbances and non-lightning events to eliminate false alarms. The specific steps are as follows: S2.1: Based on the identified lightning pre-breakdown events, determine the corresponding event trigger time, and then extract a complete event time window based on each event trigger time. This event time window includes the background disturbance before the event, the transient disturbance during the event, and the fallback process after the event ends. Then, calculate the average disturbance value and standard deviation of each group of data within each event time window. S2.2: Based on the obtained average disturbance value and standard deviation, standardize the data of each group within the event time window to generate the corresponding disturbance sequence. Then, based on the disturbance sequences in each auxiliary channel, calculate the normalized correlation strength between each auxiliary channel in the event time window, and combine the correlation strength between the business channel and the three auxiliary channels into a multi-source disturbance correlation matrix. S2.3: Extract the sum of the main diagonal elements from the multi-source perturbation correlation matrix and establish the corresponding matrix trace. Then extract the matrix eigenvalue set and calculate the eigenvalue dispersion. After that, calculate the average absolute value of the off-diagonal elements of the multi-source perturbation correlation matrix and integrate the four sets of data into a discriminant vector. S2.4: Input the real-time acquired discrimination vectors into the offline trained support vector machine for discrimination. At the same time, the support vector machine outputs the discrimination score of the corresponding lightning pre-breakdown event based on each discrimination vector. If the discrimination score is > 0, it means that the corresponding lightning pre-breakdown event is a real lightning disturbance and is marked as "passed the authenticity verification". Conversely, if the disturbance is not lightning-induced, it is considered a non-lightning disturbance and is excluded from further processing.

4. The lightning detection method based on traffic signal disturbance analysis according to claim 3, characterized in that, The specific steps for calculating the cross-entropy delay spectrum between each receiving channel and reconstructing the location, spatial morphology, and development direction of the lightning discharge channel in step IV are as follows: S3.1: Based on the lightning pre-breakdown event marked "passed authenticity verification", multiple service transmission sources and multiple receiving nodes are selected from the same area to form a passive positioning observation network, and each receiving node is set to synchronously collect the disturbance amplitude of lightning disturbance on different propagation paths around the same trigger time. S3.2: The perturbation amplitude within each sampling window is processed by short-time framing to obtain the local energy in different frames and frequency bands. Then, the frequency band energy within the entire sampling window is normalized to obtain the probability quality of each receiving node in different frames and frequency bands and construct the probability distribution. Then, two sets of perturbation sequences of receiving nodes are randomly selected, and the time offset and frequency offset between the two sets of perturbation sequences are calculated. Time-frequency compensation is performed on the perturbation sequence of any receiving node. S3.3: Establish a corresponding probability distribution for the time-frequency compensated perturbation sequence, then calculate the cross-entropy value between the uncompensated perturbation sequence and the time-frequency compensated perturbation sequence, and establish a corresponding cross-entropy time delay spectrum based on each cross-entropy value. Then, locate the joint peak position on the established cross-entropy time delay spectrum, and calculate the spectral quality in the neighborhood of each peak. Integrate the peak parameters of all receiving node pairs to form a joint observation vector. S3.4: Establish a three-dimensional spatial grid within the target monitoring area and divide the entire space into multiple groups of candidate voxel elements. Simultaneously, determine the spatial coordinates of each receiving node in the three-dimensional spatial grid. Then, calculate the theoretical time difference between each voxel element and any receiving node pair based on the propagation geometry. Next, obtain the corresponding frequency difference characteristics through a pre-calibrated frequency difference mapping function. Based on the theoretical time difference and frequency difference characteristics of all grid voxel pairs and all receiving node pairs, establish a dictionary matrix. S3.5: Statistically calculate the matching strength between each voxel unit and the actual discharge channel. Based on the observation vector, dictionary matrix, and matching strength, perform sparse reconstruction on each voxel unit to obtain the main distribution area of ​​the discharge channel in space. Integrate voxel units with matching strength exceeding the threshold into an active voxel set. Calculate the spatial centroid of the discharge channel based on the active voxels and their weights. Then, construct a weighted covariance matrix based on the spatial coordinates of each active voxel and its corresponding spatial centroid. Perform eigenvalue decomposition on the weighted covariance matrix and take the eigenvector corresponding to the largest eigenvalue as the main development direction of the discharge channel.

5. The lightning detection method based on traffic signal disturbance analysis according to claim 4, characterized in that, The specific steps for constructing the complete perturbation feature evolution trajectory based on the reconstructed lightning spatial information, as described in step V, are as follows: S4.1: Based on the spatial location, main distribution area and main development direction of lightning, collect the corresponding perturbation sequence of each receiving node after alignment, and count the spatial distance from each receiving node to the reconstructed discharge center. At the same time, based on the obtained spatial distance, obtain the fusion weight of each receiving node. Then, construct the main perturbation sequence by using the aligned perturbation sequence and the fusion weight. S4.2: Set a sliding background window and perform local background suppression on the main disturbance sequence based on the set sliding background window to obtain the net disturbance sequence after deducting the local background. Then, obtain the envelope sequence by analytical signal method, set the foreground trigger threshold, and record the rising edge starting point of the envelope sequence that first exceeds the foreground trigger threshold. Then, select the sampling point index corresponding to the highest envelope peak as the envelope peak point. S4.3: Based on the envelope peak point and the rising edge start index, calculate the average steepness of the rising segment, and at the same time, perform linear fitting on the rising segment to obtain the corresponding slope coefficient. Then set the fall threshold, and the fall threshold < the foreground trigger threshold. Then, take the first fall of the envelope sequence to a position below the fall threshold as the end point of the perturbation fall. Then, calculate the total duration of the perturbation based on the end point of the perturbation fall and the rising edge start index. S4.4: Identify each return stroke peak in the envelope sequence, calculate the time interval between them, then calculate the average return stroke interval in the current lightning event, then extract the local signal before and after the peak or during the fall-off phase, and perform discrete Fourier transform on it to obtain the complex spectrum values ​​on different frequency grids, and extract the amplitude at each frequency point, then calculate the dominant oscillation frequency of the disturbance based on each amplitude, and calculate the corresponding spectral centroid. S4.5: After completing the analysis of rising edge, duration, return stroke interval and frequency, the current lightning event window is divided into multiple continuous sub-windows, and the rising edge steepness, duration, dominant oscillation frequency and average return stroke interval in each continuous sub-window are obtained. At the same time, each feature dimension is standardized, and the feature vectors of all sub-windows are arranged in chronological order to form a complete disturbance feature evolution trajectory.

6. The lightning detection method based on traffic signal disturbance analysis according to claim 5, characterized in that, The specific steps for dynamically correcting model parameters and outputting the probability distribution of different lightning types by combining GPS timing information and meteorological background data, as described in step V, are as follows: S5.1: Organize the feature vectors of each sub-window in the perturbation feature evolution trajectory into a unified time-series input in chronological order, and then stack all time segments in sequence to form an input sequence. Then, collect the current GPS timing information and meteorological background information and encode them into context vectors. Using the generated context vectors, establish a parameter correction representation jointly generated by GPS timing and meteorological background. Then, use this parameter correction representation to correct the gate biases of the input gate, forget gate, output gate and candidate state gate. S5.2: After completing the parameter correction, based on the gating mechanism, obtain the output vectors of the input sequence at the input gate, forget gate, output gate and candidate state gate respectively, and update the current memory unit state based on the candidate state gate vector and forget gate vector. Then, generate the corresponding hidden state based on the output gate vector and the current memory unit state. S5.3: After the entire input sequence has been processed based on the gating mechanism, the sequence-level representation of the current lightning event is obtained by averaging and converging each hidden state. Based on the obtained sequence-level representation and context vector, the discrimination scores of various lightning types are calculated. Then, the Softmax function is used to convert it into a probability distribution, and the lightning type with the highest probability is selected as the final recognition type of the current lightning event.

7. The lightning detection method based on traffic signal disturbance analysis according to claim 1, characterized in that, Step VI involves analyzing the transient frequency response characteristics of each service signal change, extracting each disturbance parameter, and then using a pre-established lightning current mapping model to convert each disturbance parameter into an estimated value of the lightning current peak intensity. The specific steps are as follows: S6.1: Based on the determination result of the lightning event, the event signal containing the lightning disturbance and the reference signal are extracted from the corresponding receiving channel. Then, the two signals are weighted by window functions to generate the windowing sequence of the event window and the reference window. Then, the event signal and the reference signal are analyzed in short time frequency domain to obtain the complex spectrum values ​​of each signal on different frequency grids. Based on the obtained complex spectrum values ​​on the frequency grids, the amplitude spectrum and phase spectrum of the event signal and the reference signal are obtained. S6.2: After obtaining the spectrum of the event segment and the reference segment, when the event signal exists as a lightning disturbance, the output of the reference signal after a transient response is constructed, and the corresponding transient frequency response function is constructed. Then, based on the complex spectrum values ​​of various signals, the transient frequency response on the frequency grid is obtained through the transient frequency response function, and the transient frequency response is decomposed into frequency response amplitude and frequency response phase. S6.3: Based on the frequency response amplitude and frequency response phase, the average amplitude compression caused by lightning disturbance and the additional phase shift depth in each frequency band are statistically analyzed. Then, the frequency response phase is frequency-differentiated to obtain the group delay estimate at each frequency point. Based on the group delay estimate, the average group delay of the target frequency band is obtained. Then, the additional phase shift depth, average amplitude compression, and average group delay are integrated into a unified input vector, and the input vector is normalized. S6.4: Using a pre-established lightning current mapping model, the normalized input vector is converted into the peak intensity of lightning current. Based on the output peak intensity of lightning current, the estimated fluctuation degree of the current lightning current mapping model on the calibration data, and the confidence coefficient, a confidence interval for the corresponding peak intensity of lightning current is established. The model outputs the peak intensity of lightning current, the corresponding confidence interval, and the corresponding additional phase shift depth, average amplitude compression degree, and average group delay parameters.

Images