A method for extracting lightning precursor features by array electric field signal denoising fusion

Abstract

The application discloses a thunderstorm precursor feature extraction method based on array electric field signal noise reduction fusion, and belongs to the technical field of atmospheric electric field monitoring signal processing, thunderstorm precursor identification and array monitoring data fusion. The method synchronously collects original time sequence electric field data of array atmospheric electric field probes, removes power frequency harmonic, vehicle electrostatic, dust friction and temperature and humidity drift interference through wavelet transform layered noise reduction, removes fault or distortion measuring points through multi-probe cross fault tolerance verification, further utilizes weak electric field change adaptive amplification processing to enhance low charge initial thunderstorm cloud cluster features, carries out segmented fitting on continuous electric field time sequence, extracts 10-60 minute thundercloud charge slow lifting precursor curve, and finally outputs a standardized thunderstorm precursor feature vector. The application can improve signal signal-to-noise ratio, prolong thunderstorm early warning lead time, enhance array monitoring fault tolerance, and reduce remote weak thunderstorm missing report probability.

Description

Technical Field

[0001] This invention belongs to the fields of atmospheric electric field monitoring signal processing, lightning precursor feature identification, array monitoring data fusion, weak lightning electric signal extraction, and fault tolerance technology for monitoring equipment. Specifically, it relates to an array-type electric field signal denoising and fusion lightning precursor feature extraction system and method that uses wavelet transform for global noise reduction, extraction of long-period charge rise curves in thunderclouds, multi-probe cross-fault tolerance verification, and weak electric field change amplification identification. It is suitable for long-term lightning prediction scenarios in wind power, photovoltaic, substations, ancient building clusters, and chemical industrial parks. Background Technology

[0002] Existing array atmospheric electric field monitoring data processing technologies suffer from drawbacks such as severe noise interference, delayed precursor identification, easy system failure due to single-point faults, and inability to identify weak charge clouds.

[0003] 1. Various types of noise interference superimpose, drowning out effective lightning signals. Harmonics from power frequency transmission lines around the site, static electricity from passing vehicles, dust or vegetation dust friction, and sudden changes in temperature and humidity bring stray electric fields that continuously interfere with the original electric field waveform. Conventional mean filtering and simple threshold filtering can only filter out a small amount of impulse noise and cannot separate continuous power frequency harmonics from slowly varying static electricity interference. The true precursor signal of thunderclouds is easily masked by noise.

[0004] 2. It only identifies instantaneous electric field changes and cannot capture long-term lightning precursors. Traditional algorithms mainly identify drastic changes in the electric field, ignoring the gradual rise of the precursor curve during the accumulation of charge in thunderclouds over a period of 10 to 60 minutes. It can only issue an alarm when lightning is approaching, and cannot achieve ultra-long lead time warnings, leaving insufficient time for response.

[0005] 3. A single probe failure directly leads to the failure of the area's early warning system. When any electric field probe in the array network loses power, drifts, or is damaged, the traditional algorithm lacks a cross-validation mechanism and directly uses distorted data in the calculation, which can easily cause positioning errors, false alarms, or missed alarms, thus reducing the reliability of the entire monitoring system.

[0006] 4. Low-charge, weak thundercloud signals cannot be identified. Small, nascent clouds and distant, weak thunderstorm clouds have low charge accumulation and tiny electric field changes. Conventional signal processing algorithms have insufficient resolution, and these subtle changes are easily filtered out by noise, leading to the missed detection of distant, nascent thunderstorms.

[0007] In summary, traditional electric field signal processing methods have weak noise reduction capabilities, short lead time for identification, lack of equipment fault tolerance mechanisms, and poor weak thunderstorm identification capabilities, making them unable to support a high-precision, long-term, and highly reliable array-based short-term lightning early warning system. Summary of the Invention

[0008] I. Purpose of the invention.

[0009] To address the shortcomings of existing electric field monitoring signals, such as severe clutter interference, inability to extract long-period lightning precursors, poor fault tolerance of single probes, and difficulty in identifying weak, low-charge clouds, this invention provides a method for extracting lightning precursor features through array-based electric field signal denoising and fusion. By employing four core technologies—wavelet transform layered denoising, extraction of precursor curves for slow charge rise in thunderclouds, multi-probe cross-verification fault-tolerance mechanism, and a weak electric field change amplification and identification algorithm—this method achieves high-precision purification of the original array electric field signal, extraction of ultra-long-term lightning precursors, uninterrupted early warning in the event of single-device failures, and accurate identification of low-charge nascent thunderclouds, thereby improving the lead time for lightning warnings and the overall system stability.

[0010] II. Technical Solution.

[0011] This invention provides an array-type electric field signal denoising and fusion lightning precursor feature extraction system, including a wavelet layered denoising processing module, a long-period charge rise precursor extraction module, an array multi-probe cross-tolerance verification module, a weak electric field signal amplification and identification module, and a lightning precursor feature vector output unit.

[0012] 1. Wavelet layered noise reduction module. This module receives the original electric field time-series data from all probes in the array, performs multi-layer wavelet decomposition, distinguishes between different frequency band noises such as power frequency harmonics, dust friction static electricity, and temperature and humidity drift, removes interference components layer by layer, and reconstructs the pure original waveform of the lightning electric field.

[0013] 2. Long-period charge rise precursor extraction module. This module performs piecewise fitting on the noise-reduced continuous time-series electric field values, extracts the characteristic curve of slowly and continuously rising thundercloud charge accumulation, and outputs lightning precursor characteristic parameters on a scale of 10 to 60 minutes, which serve as the input basis for long-term early warning.

[0014] 3. Multi-probe cross-tolerance verification module. This module synchronously compares the average electric field value, trend, and fluctuation range of multiple probes in the array over the same time period, identifies faulty, drifting, and distorted measurement points and automatically removes them, and uses the data from the remaining normal measurement points for fusion calculation to ensure that the system continues to output effective features even in the event of a single probe failure.

[0015] 4. Weak Electric Field Signal Amplification and Recognition Module. This module is equipped with an adaptive differential amplification algorithm to perform feature enhancement processing on small-amplitude, slowly varying electric field signals, amplifying the weak electric field variation characteristics of low-charge clouds and improving the sensitivity of long-distance, nascent small thunderstorm cloud recognition.

[0016] 5. Lightning Precursor Feature Vector Output Unit. This unit standardizes and combines the noise-reduced waveform, effective measurement point set, consistency parameters, weak signal enhancement parameters, long-period rise slope, duration, and risk level to generate a lightning precursor feature vector, which is then transmitted to the lightning location, short-term warning, and hierarchical linkage control system.

[0017] III. Methods and Steps.

[0018] The present invention provides a method for extracting lightning precursor features by array-type electric field signal denoising and fusion, which includes the following steps.

[0019] S1. Synchronously acquire raw time-series electric field data from all atmospheric electric field probes in the array, collect raw mixed noise signals, and record probe number, sampling timestamp, sampling frequency, and measurement point location.

[0020] S2. Wavelet transform layered noise reduction processing: The original signal is decomposed into multiple frequency domains, and various interferences such as power frequency harmonics, vehicle static electricity, dust friction, temperature and humidity drift are filtered out in layers to reconstruct a pure electric field time sequence waveform.

[0021] S3. Cross-tolerance verification of array multi-probes: compare synchronous data from multiple measurement points in the same area, identify and remove faulty, drifting, and distorted probe data, and retain only the valid measurement point dataset.

[0022] S4. Adaptive amplification and identification of weak electric field changes: Differential feature amplification is performed on the time-series signal after noise reduction and verification to enhance the characteristics of small electric field changes in low-charge nascent clouds.

[0023] S5. Extract the precursor curve of long-period thundercloud charge rise, fit the continuous electric field time series in segments, extract the slow charge accumulation gradient curve from 10 to 60 minutes, and output the characteristic parameters of thunderstorm precursors.

[0024] S6. Standardize and output lightning precursor feature vectors, and transmit them to the lightning location, short-term early warning, and hierarchical linkage control system to complete the feature extraction closed loop.

[0025] IV. Data Processing Logic.

[0026] In the wavelet layered denoising process, let the original electric field value collected by the i-th electric field probe at time t be Ei(t). Perform multi-scale wavelet decomposition on Ei(t) to obtain the low-frequency approximate component A and multiple detail components D1 to Dn. Set thresholds for detail components at different scales to suppress impulse noise, power frequency harmonics and intermediate frequency electrostatic interference, and reconstruct the denoised electric field sequence Eci(t).

[0027] During the cross-tolerance process of multi-probe array, the mean, slope, variance and correlation coefficient of adjacent measuring points are calculated for each measuring point within the same time window. When the deviation of a measuring point from the array median continuously exceeds the set threshold, or the trend correlation coefficient is lower than the set threshold, the measuring point is marked as an abnormal measuring point and removed from the fusion calculation.

[0028] In the process of amplifying and identifying weak electric field changes, the difference value of adjacent windows and the sliding slope are calculated for the denoised electric field sequence, and the gain coefficient is adaptively determined according to the background noise level. Feature enhancement is performed on electric field changes with small amplitude but strong persistence to avoid low-charge nascent clouds being directly filtered out.

[0029] During the extraction of long-period thundercloud charge rise precursor curves, piecewise linear fitting is performed on the noise-reduced fused electric field sequence within a continuous time window to obtain the rise slope, duration, goodness of fit, measurement point consistency, and enhanced response intensity. When the above parameters meet the preset combination criteria, the corresponding level of thunderstorm precursor characteristic parameters are output.

[0030] V. Beneficial Effects.

[0031] 1. Comprehensive filtering of multiple types of interference significantly improves signal-to-noise ratio. Wavelet layered noise reduction accurately separates clutter interference across all scenarios, addressing the industry pain point of power frequency, dust, and vehicle static electricity masking the effective lightning signal, and significantly improving the reliability of raw data.

[0032] 2. Significantly extended warning time. It can detect signs of slow power generation in thunderclouds up to 60 minutes in advance, breaking through the limitations of traditional alarms that can only be triggered near the source, and allowing sufficient time for graded risk avoidance and power reduction operations of equipment in unmanned sites.

[0033] 3. High fault-tolerant operation of array monitoring. The system does not interrupt operation and early warning output even if a single probe is damaged, drifts, or loses power, greatly improving the stability of the entire lightning monitoring system under complex field conditions.

[0034] 4. Significantly enhanced ability to identify weak thunderstorms. The algorithm for amplifying weak electric field signals can identify distant, low-charge nascent clouds, greatly reducing the probability of missed reports of distant thunderstorms.

[0035] 5. Adapt to multi-source fusion AI lightning early warning system. Provides high-purity, long-term, and highly reliable underlying feature data for multi-source fusion early warning, accurate lightning strike location, and hierarchical linkage control, and improves the entire chain of AI lightning protection system, including signal processing, precursor extraction, location prediction, and hierarchical linkage.

[0036] 6. Wide range of applications. It can be used in conjunction with array-type atmospheric electric field monitoring systems for all scenarios, including wind power, photovoltaic energy storage, substations, chemical industrial parks, ancient buildings, and high-voltage transmission corridors. Attached Figure Description

[0037] Figure 1 Flowchart of a method for array-based electric field signal noise reduction and fusion for lightning precursor feature extraction.

[0038] Figure 2 Schematic diagram of an array-type electric field signal noise reduction and fusion lightning precursor feature extraction system.

[0039] Figure 3 : 3D schematic diagram of array atmospheric electric field probe networking and fault-tolerant fusion.

[0040] Figure 4 Schematic diagram of noise reduction, enhancement and curve extraction of lightning precursor signals.

[0041] Abstract and Figure Selection Figure 1 .

[0042] Figure labeling: 100, Array monitoring area; 101, Atmospheric electric field probe; 102, Thundercloud or charge cloud; 103, Data communication link; 104, Edge computing terminal; 105, Early warning linkage terminal; 106, Fault probe; 201, Wavelet hierarchical noise reduction module; 202, Long-period charge rise precursor extraction module; 203, Array multi-probe cross-fault tolerance verification module; 204, Weak electric field signal amplification and identification module; 205, Lightning precursor feature vector output unit; 206, Lightning location, short-term early warning and hierarchical linkage control system. Detailed Implementation

[0043] The present invention will be further described below with reference to the accompanying drawings. These specific embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0044] See Figures 1 to 4 Multiple atmospheric electric field probes 101 are deployed within the array monitoring area 100. Each atmospheric electric field probe 101 synchronously collects the electric field time-series data of its measuring point according to a fixed sampling period and transmits it to the edge computing terminal 104 through the data communication link 103. The atmospheric electric field probes 101 can be deployed around wind farms, photovoltaic areas, substations, chemical industrial parks, ancient building complexes, or high-voltage transmission corridors to form an array monitoring network covering the target area.

[0045] After receiving raw time-series electric field data from multiple atmospheric electric field probes 101, the edge computing terminal 104 first performs timestamp alignment and initial screening for abnormal missing measurements, so that data from different measurement points enter the same fusion time window. For short-term packet loss data, adjacent time windows can be used for interpolation; for probe data with continuous power outages or significant drift, it is included in the fault candidate set.

[0046] In the wavelet hierarchical denoising stage, the edge computing terminal 104 calls the wavelet hierarchical denoising module 201 to perform multi-scale decomposition on the original electric field sequence of each measurement point. High-frequency detail components are used to identify and filter out high-frequency noise such as spike pulses, communication jitter, and power frequency harmonic coupling; mid-frequency detail components are used to suppress mid-frequency interference such as vehicle friction static electricity and dust friction static electricity; low-frequency approximation components are used to preserve the slow changing trend of thundercloud charge while suppressing spurious trends caused by temperature and humidity drift. After thresholding, the denoised electric field sequence is reconstructed.

[0047] During the cross-tolerance verification phase of the multi-probe array, the cross-tolerance verification module 203 compares the mean electric field, slope of change, fluctuation range, and trend correlation coefficient of different measurement points within the same time window. When a probe deviates too much from the array median in multiple consecutive windows, or when its trend correlation coefficient with adjacent probes remains consistently low, the probe is marked as a faulty probe 106 and automatically removed. The system continues to use data from the remaining normal probes for fusion, ensuring that a single-point fault does not cause the entire early warning system to be interrupted.

[0048] In the weak electric field signal amplification and identification stage, the weak electric field signal amplification and identification module 204 determines the adaptive gain coefficient based on the background noise intensity and the consistency of effective measurement points, and performs differential amplification and feature enhancement on the low-amplitude, slowly changing signal after noise reduction. For the small electric field changes caused by distant weak thunderstorm clouds or nascent thunderclouds 102, this module can improve the distinguishability of the change slope and persistence features.

[0049] In the long-period charge rise precursor extraction stage, the long-period charge rise precursor extraction module 202 performs sliding window piecewise fitting on the fused denoised electric field sequence, calculating the rise slope, duration, goodness of fit, measurement point consistency, and enhanced response intensity within a timescale of 10 to 60 minutes. When the above parameters meet the combined criterion of continuous accumulation of thundercloud charge, the lightning precursor features are output.

[0050] The lightning precursor feature vector output unit 205 standardizes and encapsulates the effective measurement point number, noise-reduced electric field value, fused electric field value, rise slope, duration, goodness of fit, weak signal enhancement coefficient, probe consistency score, and risk level to form a lightning precursor feature vector, which is then sent to the lightning location, short-term warning, and hierarchical linkage control system 206. The warning linkage terminal 105 can then use this vector to execute audible and visual warnings, reduce equipment power, prompt personnel evacuation, or implement lightning protection linkage control.

[0051] The processing flow of the present invention can be performed according to the following steps.

[0052] S1. Synchronously acquire raw, noisy, time-series electric field data from all electric field probes in the array.

[0053] S2. Wavelet transform is used for layered noise reduction to filter out various interferences such as power frequency, dust, vehicle static electricity, and temperature and humidity drift, and reconstruct a pure electric field waveform.

[0054] S3. Perform multi-probe cross-validation to remove faulty, drifting, or distorted measurement point data, and retain valid monitoring datasets.

[0055] S4. Enhance the characteristics of weak electric field changes through adaptive amplification processing to identify low-charge nascent thunderstorm cloud clusters.

[0056] S5. Piecewise fitting of time-series electric fields to extract precursory characteristic curves of slow charge rise in thunderclouds over 10 to 60 minutes.

[0057] S6. Standardize the output of lightning precursor characteristics for use in lightning early warning, positioning, and linkage control systems.

Claims

1. A method for extracting lightning precursor features by array-based electric field signal denoising and fusion, characterized in that, Includes the following steps: S1. Synchronously acquire the raw, noisy, time-series electric field data of all atmospheric electric field probes in the array, and align the timestamps of the data from different probes; S2. Perform wavelet transform layered noise reduction on the original noisy time-series electric field data, filter out power frequency harmonics, vehicle static electricity, dust friction and temperature and humidity drift interference in layers, and reconstruct the noise-reduced electric field sequence; S3. Perform cross-tolerance verification on the synchronous data of multiple probes in the array, identify and remove faulty, drifting or distorted probe data, and form a valid measurement point dataset; S4. Perform adaptive amplification processing on the denoised electric field sequence in the effective measurement point dataset to enhance the characteristics of small electric field changes in low-charge nascent thunderstorm clouds. S5. Perform piecewise fitting on the amplified continuous electric field time series to extract the precursor characteristic curves of slow charge rise in thunderclouds on a scale of 10 to 60 minutes. S6. The precursor characteristic curves and corresponding characteristic parameters are standardized and output as lightning precursor characteristic vectors for use by lightning location, short-term early warning or hierarchical linkage control systems.

2. The method for extracting lightning precursor features according to claim 1, characterized in that, The wavelet transform layered noise reduction includes: performing multi-scale wavelet decomposition on the original electric field sequence collected by each atmospheric electric field probe to obtain low-frequency approximate components and multiple high-frequency and mid-frequency detail components; setting thresholds for different scale components to suppress high-frequency impulse noise, mid-frequency electrostatic interference and low-frequency temperature and humidity drift clutter, and reconstructing the noise-reduced electric field sequence.

3. The method for extracting lightning precursor features according to claim 1, characterized in that, The cross-tolerance verification includes: calculating the mean electric field, slope of change, fluctuation range and trend correlation coefficient of multiple measuring points within the same time window; when the deviation of any measuring point from the array median continuously exceeds the preset deviation threshold, or the trend correlation coefficient with the adjacent measuring point is lower than the preset correlation threshold, the probe corresponding to that measuring point is marked as an abnormal probe and removed from the fusion calculation.

4. The method for extracting lightning precursor features according to claim 1, characterized in that, The adaptive amplification processing of weak electric field changes includes: determining the adaptive gain coefficient based on the background noise intensity and the consistency of effective measurement points, and amplifying the gain of the electric field difference value and the electric field change slope between adjacent sliding windows to enhance the low-amplitude and continuously changing thundercloud charge accumulation signal.

5. The method for extracting lightning precursor features according to claim 1, characterized in that, The piecewise fitting includes: performing piecewise linear fitting on the fused noise-reduced electric field sequence within a sliding time window of 10 to 60 minutes, and outputting the rise slope, duration, goodness of fit, measurement point consistency, and enhanced response intensity; when the above parameters meet the preset combination criteria, it is determined that there is a precursor to a slow rise in thundercloud charge.

6. The method for extracting lightning precursor features according to claim 1, characterized in that, The lightning precursor feature vector includes one or more of the following: effective measurement point number, noise-reduced electric field value, fused electric field value, rise slope, duration, goodness of fit, weak signal enhancement coefficient, probe consistency score, and lightning strike risk level.

7. An array-type electric field signal denoising and fusion lightning precursor feature extraction system, characterized in that, It includes a wavelet layered noise reduction processing module, a long-period charge rise precursor extraction module, an array multi-probe cross-tolerance verification module, a weak electric field signal amplification and identification module, and a lightning precursor feature vector output unit; The wavelet layered noise reduction module is used to filter out power frequency harmonics, vehicle static electricity, dust friction and temperature and humidity drift interference in layers using wavelet transform. The long-period charge rise precursor extraction module is used to extract the precursor curve of slow charge rise in thunderclouds over a period of 10 to 60 minutes. The array multi-probe cross-tolerance verification module is used to eliminate abnormal probe data through cross-verification of data from multiple measurement points, so that the overall early warning output is not interrupted when a single probe fails. The weak electric field signal amplification and identification module is used to amplify the small electric field changes of low-charge clouds to achieve early identification of weak thunderstorms. The lightning precursor feature vector output unit is used to output a standardized lightning precursor feature vector.

8. The lightning precursor feature extraction system according to claim 7, characterized in that, The system also includes an array of atmospheric electric field probes and an edge computing terminal. Multiple arrays of atmospheric electric field probes are used to synchronously collect raw time-series electric field data from different measurement points within the target area. The edge computing terminal is used to perform noise reduction, fault tolerance verification, weak signal enhancement, and precursor curve extraction.

9. The lightning precursor feature extraction system according to claim 7, characterized in that, The array multi-probe cross-tolerance verification module is configured to automatically remove abnormal measurement points when it detects that the probe is powered off, drifting, damaged, or has data distortion, and to perform fusion calculation using the remaining normal probe data.

10. The lightning precursor feature extraction system according to claim 7, characterized in that, The lightning precursor feature vector output unit is communicatively connected to the lightning location, short-term early warning and hierarchical linkage control system, and is used to transmit standardized lightning precursor feature vectors to audible and visual early warning, equipment power reduction, personnel evacuation prompts or lightning protection linkage control equipment.

Images