Lightning identification method, lightning detection system, lightning detection device and electronic equipment

Abstract

The invention relates to the field of lightning detection, and discloses a lightning identification method, a lightning detection system, a lightning detection device and electronic equipment, and the method comprises the steps: receiving a lightning identification result, confidence and original lightning data sent by each edge detection end; obtaining a lightning positioning theoretical value based on the analysis of the original lightning data, and obtaining a lightning positioning error through the inversion of the lightning positioning theoretical value; for each edge detection end, the weight of the lightning identification result is calculated by using the lightning positioning error and the confidence coefficient of the edge detection end; and performing weighted voting on the lightning identification result of each edge detection end to obtain a final lightning identification result. In the application, even if a certain edge detection end has a fault or data is inaccurate, the data of other detection ends which work normally can still influence a final result. Through a weighted voting mechanism, data abnormity of an individual detection end can be offset to a certain degree, and the reliability of a lightning identification result is ensured.

Description

Technical Field

[0001] This invention relates to the field of lightning detection, and in particular to a method for lightning identification, a system, apparatus, and electronic device for lightning detection. Background Technology

[0002] In practical lightning detection applications, multiple lightning observation devices are set up to record the electromagnetic pulse signals generated when lightning occurs. When lightning strikes, the electromagnetic pulse signals generated by the lightning travel at the speed of light to each device, resulting in a slight time difference (the distance and time of arrival vary depending on the device). After receiving and processing the electromagnetic signals, the observation devices at each location transmit the lightning reception time to remote positioning and processing software via a communication network. The positioning and processing software receives real-time data, calculates the time difference of the same lightning strike collected by each observation device, and thus calculates the specific location of the lightning strike. Furthermore, by using altitude information, it can identify the type of lightning, such as cloud-to-ground lightning or ground-to-ground lightning (located near the ground).

[0003] However, in actual deployment and application, observation networks often suffer from relatively sparse distribution of observation equipment and insufficient effective detection stations. As a result, only four or fewer edge detectors can successfully capture the signal of a single lightning event. In this situation, the location results are reduced to a two-dimensional plane (lacking altitude information), making it impossible to accurately determine the lightning type. This hinders accurate forest fire risk warnings and reduces the ability to assess aircraft safety threats, thus weakening its role in disaster prevention and control. Summary of the Invention

[0004] The purpose of this invention is to provide at least one method for lightning identification, a system, device, and electronic equipment for lightning detection, which can at least solve the technical problem of being unable to accurately determine the type of lightning and accurately guide disaster prevention due to the relatively sparse distribution of observation equipment and the lack of effective detection sites, and can at least achieve more accurate determination of the type of lightning and more accurate guidance for disaster prevention.

[0005] To address the aforementioned technical problems, at least one embodiment of this application provides a method for lightning identification, applied to the central end of a lightning detection system. The lightning detection system includes multiple edge detection ends and the central end, comprising: Receive lightning identification results, the confidence level corresponding to the lightning identification results, and the original lightning data sent from each of the edge detection terminals; Based on the original lightning data analysis, the theoretical value of lightning location is obtained, and the theoretical time for the lightning signal to reach each of the edge detection terminals is obtained by inversion using the theoretical value of lightning location. Based on the measurement error between the theoretical time and the actual time of the lightning signal arrival at the corresponding edge detection terminal, the lightning location error is calculated. For each edge detection terminal, the weight of the lightning identification result of the edge detection terminal is calculated using the lightning positioning error and the confidence level corresponding to the edge detection terminal, and the sum of the weights corresponding to each edge detection terminal is 1; The lightning identification results of each edge detection terminal are weighted and voted to obtain the final lightning identification result.

[0006] At least one embodiment of this application also provides a lightning identification device applied to the central end of a lightning detection system, the lightning detection system including multiple edge detection ends and the central end, the device comprising: The receiving module is used to receive lightning identification results, the confidence level corresponding to the lightning identification results, and the original lightning data sent from each of the edge detection terminals; The error calculation module is used to parse the original lightning data to obtain the theoretical value of lightning location, and use the theoretical value of lightning location to invert the theoretical time of the lightning signal reaching each of the edge detection terminals. The lightning location error is obtained based on the measurement error between the theoretical time and the actual time of the lightning signal arrival at the corresponding edge detection terminal. The weight calculation module is used to calculate the weight of the lightning identification result of each edge detection end by using the lightning positioning error and the confidence level corresponding to the edge detection end. The sum of the weights corresponding to each edge detection end is 1. The identification module is used to perform weighted voting on the lightning identification results of each edge detection end to obtain the final lightning identification result.

[0007] At least one embodiment of this application also provides a lightning detection system, comprising: Multiple edge detection terminals, each of which is used to analyze and obtain a lightning identification result and a confidence level corresponding to the lightning identification result based on the raw lightning data it receives; The central terminal receives the lightning identification results, corresponding confidence levels, and raw lightning data sent from each of the edge detection terminals. Based on the raw lightning data, it parses to obtain the theoretical value for lightning location and uses this theoretical value to invert the theoretical time for the lightning signal to reach each of the edge detection terminals. Based on the measurement error between the theoretical time and the actual arrival time of the lightning signal at the corresponding edge detection terminal, it calculates the lightning location error. For each edge detection terminal, it uses the lightning location error and the corresponding confidence level to calculate the weight of the lightning identification result for that edge detection terminal, with the sum of the weights for each edge detection terminal being 1. Finally, it performs a weighted voting process on the lightning identification results of each edge detection terminal to obtain the final lightning identification result.

[0008] At least one embodiment of this application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the lightning recognition method described above.

[0009] At least one embodiment of this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described lightning recognition method.

[0010] The lightning identification method, lightning detection system, apparatus, and electronic device provided in the embodiments of this application simultaneously acquire preliminary identification results and raw data from each edge detection end, providing a basis for subsequent cross-validation. Retaining the raw data allows the central end to independently verify the data, avoiding complete reliance on edge end judgment, fully utilizing the real-time processing capabilities of the edge end, and reducing the computational burden on the central end. The weight of the lightning identification result of each edge detection end is calculated based on the lightning positioning error and the corresponding confidence level of the edge detection end. Edge detection ends with smaller lightning positioning errors and higher confidence levels have larger weights and occupy a more important position in the final weighted voting. This dynamic weight adjustment method makes the data from detection ends with higher positioning accuracy have a greater impact on the final result, helping to improve the overall lightning positioning accuracy. That is, because this application uses data from multiple edge detection ends for comprehensive judgment, even if one edge detection end malfunctions or its data is inaccurate, the data from other normally functioning detection ends can still affect the final result. Through the weighted voting mechanism, the abnormality of individual detection end data can be offset to a certain extent, improving the system's fault tolerance to faults and errors and ensuring the reliability of the lightning identification results.

[0011] In some optional embodiments, the calculation of the lightning location error based on the measurement error between the theoretical time and the measured arrival time of the lightning signal at the corresponding edge detection end includes: The chi-square error is calculated based on the measurement error corresponding to each edge detection end, and the chi-square error is used as the lightning positioning error.

[0012] In this embodiment, the chi-square error reflects the degree of consistency between these errors and theoretical expectations. If the measurement errors of each edge detector end are in good agreement with the theoretical model, the chi-square error will be relatively small; conversely, if the measurement errors deviate significantly from the theoretical expectations, the chi-square error will increase. This measurement of error consistency helps to more accurately assess the error situation of lightning positioning and determine whether the positioning results are reliable.

[0013] In some optional embodiments, the weighted voting of the lightning identification results of each edge detection terminal to obtain the final lightning identification result includes: The lightning identification results are quantified according to different types, and the quantified lightning identification results are weighted and summed. Determine the numerical range to which the summation result belongs, and determine the final lightning identification result based on the preset mapping relationship between the numerical range and the lightning identification type.

[0014] In this embodiment, the numericalized lightning identification results are weighted and summed to fully consider the different importance and reliability of each edge detection end. By assigning corresponding weights to the identification results of each detection end, those detection ends with high positioning accuracy and strong data reliability play a greater role in the final decision. The numerical range to which the summed result belongs is determined, and the final lightning identification result is determined according to the preset mapping relationship between the numerical range and the lightning identification type, providing clear and unambiguous rules for the final result determination. This rule-based determination method avoids the subjectivity and arbitrariness of human judgment, making the determination of the final result repeatable and consistent.

[0015] In some optional embodiments, it also includes: Based on the accuracy of lightning identification by the edge detection terminal in the current identification task, adjust the confidence level of the edge detection terminal in the next lightning identification task.

[0016] In this embodiment, the role of high-quality detectors is enhanced and the influence of inefficient detectors is weakened by adjusting the confidence level, so as to ensure that the final lightning identification result is closer to the real situation.

[0017] In some optional embodiments, adjusting the confidence level of the edge detection terminal in the next lightning identification task based on the accuracy of lightning identification by the edge detection terminal in the current identification task includes: The lightning identification result of the edge detection end corresponding to the measurement error that contributes the least to the lightning positioning error is used as the correction standard; For edge detectors that differ from the correction standard, reduce the confidence level of the edge detector during the next lightning detection.

[0018] In this embodiment, the edge detection end corresponding to the measurement error that contributes the least to the lightning location error usually means that the data acquired by that detection end in this identification task is of higher quality and the detection result is more reliable. Using its lightning identification result as a correction standard can provide an accurate and reliable reference for the entire identification system, which helps to reduce the overall identification error caused by the data deviation of individual detection ends, thereby improving the accuracy of the next lightning identification.

[0019] In some optional embodiments, the method for determining the measurement error that contributes the least to the lightning location error includes: Calculate the chi-square residual corresponding to the measurement error of each edge detection end. The smaller the chi-square residual, the smaller its contribution to the lightning positioning error.

[0020] In this embodiment, the chi-square residual provides a precise quantitative value for the contribution of measurement error to lightning location error. Compared to traditional subjective judgment or simple qualitative analysis, the chi-square residual can clearly and intuitively reflect the correlation between the measurement error of each edge detector and the lightning location error with a specific numerical value. It can intuitively determine which detector's data is more reliable and contributes less to the location error. Attached Figure Description

[0021] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.

[0022] Figure 1 This is a flowchart of a lightning identification method provided in one embodiment of this application; Figure 2 This is a schematic diagram illustrating the workflow of an edge detection terminal provided in one embodiment of this application; Figure 3 This is a schematic diagram of the confidence correction process provided in another embodiment of this application; Figure 4 This is a schematic diagram of a lightning recognition device provided in another embodiment of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been provided in the various embodiments of this application to help readers better understand this application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and modifications based on the following embodiments. The division of the various embodiments below is for the convenience of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.

[0024] To facilitate understanding of the embodiments of this application, relevant content regarding lightning detection will be introduced first.

[0025] In practical lightning detection applications, a single lightning observation device can record the electromagnetic pulse signal generated when lightning occurs. However, to obtain precise information such as the location and time of lightning strikes, multiple lightning observation devices need to be set up at different locations. When lightning occurs, the electromagnetic pulse signal generated by the lightning travels at the speed of light to each device, resulting in a slight time difference (the lightning travels different distances and takes different times to reach each device). After receiving and processing the electromagnetic signals, the observation devices at each location transmit the lightning reception time to remote positioning and processing software via a communication network. The positioning and processing software receives real-time data, calculates the time difference of the same lightning strike collected by each observation device, and thus inversely calculates the specific location of the lightning strike.

[0026] The method described above is called the Time of Arrival (TOA) method. TOA calculates the coordinates (longitude and latitude) and altitude of lightning, forming a three-dimensional lightning location. The altitude information allows for the identification of lightning types, such as cloud-to-ground lightning and ground-to-ground lightning (at altitudes close to the ground). By identifying the location and type of lightning, meteorological departments can improve their ability to issue timely warnings for forest disasters, power supply disruptions, and flight scheduling.

[0027] However, achieving accurate 3D lightning localization typically requires five or more observation devices within an observation network to work collaboratively for effective inversion calculations. In practical deployment and application, observation networks are often affected by the following conditions: 1. Equipment Sparsity and Insufficient Effective Detection Stations: Current observation networks exhibit relatively sparse equipment distribution, with station spacing typically ranging from 60 to 100 km. Due to limitations in equipment detection efficiency, complex terrain (such as hills obstructing the view), and the influence of electromagnetic interference, a single lightning event often requires only four or fewer edge detectors to successfully capture its signal. In this situation, the location results are reduced to a two-dimensional plane (lacking height information), making it impossible to accurately determine the lightning type (e.g., cloud-to-ground lightning or ground-to-ground lightning). 2. Data Transmission Bandwidth and Content Limitations: Edge detection devices deployed in remote outdoor environments primarily rely on public wireless network communication (such as 4G / 5G base stations) to transmit data. However, due to network bandwidth bottlenecks in edge areas (typically less than 1 Mbps) and high data traffic costs, the devices can only upload highly compressed metadata (such as time of arrival (TOA) and peak signal strength) in real time, and cannot transmit high-resolution raw waveform data. This limitation directly prevents the positioning processing software from analyzing key temporal / frequency domain morphological characteristics of lightning waveforms (including rise edge steepness, oscillation decay patterns, and pulse duration), making it impossible for the system to determine the lightning type based on other information in the two-dimensional positioning results under sparse site conditions.

[0028] Therefore, the current observation network has a high proportion of two-dimensional positioning results (lacking altitude information), which severely limits the practical application value of the system: for example, it cannot accurately support forest fire risk warnings triggered by ground lightning (CG) or effectively assess the safety threat of cloud lightning (IC) to aircraft, which weakens the decision-making guidance role of the lightning positioning system in the disaster prevention chain.

[0029] To address the aforementioned technical problem of inaccurate lightning type determination and precise disaster prevention guidance due to the relatively sparse distribution of observation equipment and insufficient effective detection sites, this embodiment proposes a lightning identification method. The implementation details of the lightning identification method in this embodiment are described below. The following content is only for ease of understanding and is not necessary for implementing this solution.

[0030] Example 1: The lightning identification method of this embodiment can be applied to the central end of a lightning detection system, which includes multiple edge detection ends and the central end. The specific process can be as follows: Figure 1 As shown, it includes: Step 110: Receive lightning identification results, the confidence level corresponding to the lightning identification results, and the original lightning data sent from each of the edge detection terminals; In this embodiment, a lightweight deep learning model is deployed at the edge detection edge to directly and intelligently identify the collected raw lightning data in real time, achieving rapid differentiation of lightning types. Edge identification is completed as soon as the data is generated, thereby reducing the latency between data transmission and central processing.

[0031] like Figure 2 The diagram shows the workflow of the edge detection terminal. In its implementation, the edge detection terminal first acquires broadband electromagnetic signals through a high-speed analog-to-digital converter module and preprocesses them, including DC bias removal, normalization, and bandpass filtering. The preprocessed waveform data is then fed into an embedded deep learning model for real-time inference. The model can employ a structure combining a one-dimensional convolutional neural network with a temporal attention mechanism to consider both local waveform features and overall temporal trends. During the training phase, this embodiment utilizes a large dataset of labeled lightning waveforms for supervised learning and incorporates unlabeled data using a semi-supervised training strategy to enhance the model's generalization ability. During the inference phase, the model can output the probability of whether the event belongs to cloud-based lightning or ground-based lightning within milliseconds, while also providing the corresponding confidence index.

[0032] To ensure the feasibility of this method on embedded hardware, this embodiment performs quantization and structural pruning optimization on the model to ensure stable operation on low-power processors or neural network acceleration units. In terms of system architecture, the edge detection end acts as an edge computing node; its output classification results can be stored and displayed locally, and can also be uploaded to the central server via a communication link. During the localization calculation stage, when the central end cannot accurately obtain altitude information due to insufficient station numbers, it can directly use the classification results from the edge devices as constraints, thereby improving the reliability and completeness of the final localization result.

[0033] Step 120: Based on the original lightning data, obtain the theoretical value of lightning location, and use the theoretical value of lightning location to invert the theoretical time of the lightning signal reaching each edge detection end. Calculate the lightning location error based on the measurement error between the theoretical time and the measured time of the lightning signal arrival at the corresponding edge detection end. Specifically, lightning location error primarily measures the degree of agreement between observed data and theoretical models. Specifically, the central monitoring station calculates a theoretical location result based on the arrival times of lightning signals received by each edge detection station. This result is then used to extrapolate the theoretical time difference that each station should have observed. Comparing these theoretical time differences with the actual measured time differences reveals a certain measurement error. Lightning location error is an index calculated by weighting these measurement errors according to noise levels. A smaller value indicates a closer fit between the observation and the model, resulting in a more reliable location result; a larger value indicates a significant discrepancy between the observation and the theory, potentially suggesting anomalies in station data or location errors.

[0034] Step 130: For each edge detection terminal, the weight of the lightning identification result of the edge detection terminal is calculated using the lightning positioning error and the confidence level corresponding to the edge detection terminal, and the sum of the weights corresponding to each edge detection terminal is 1. Specifically, the weighting needs to consider both confidence level (reflecting the reliability of the lightning identification result) and positioning error (reflecting the positioning accuracy of the detector). For example, a linear weighted model can be used: Weight = Confidence / (Location error + Normalization constant); Then, normalization is performed to ensure that the sum of the weights is 1: After calculating the initial weights for all edge endpoints, normalization is performed to ensure that the sum of all weights is 1, as follows: =ω_i / (ω_1 +ω_2 + ... + ω_n); in, The weights obtained by normalizing the weights ω_i will have higher weights for probes with high confidence and low error.

[0035] Step 140: Perform a weighted vote on the lightning identification results of each edge detection end to obtain the final lightning identification result.

[0036] Specifically, the identification results of all edge points in the same lightning event are used as "votes". The value of each "vote" is not 1 vote, but its corresponding normalized weight W_i.

[0037] For each possible identification category (such as "cloud flash" or "ground flash"), the weights assigned to that category are summed to obtain the total weight score for that category.

[0038] The final result is generated, and the category with the highest total weight score can be selected as the final lightning identification result. For example, for three edge points, the identification results and weights are [ground lightning: 0.6], [cloud lightning: 0.3], and [ground lightning: 0.1]. Then, the total weight of "ground lightning" is 0.7, and "cloud lightning" is 0.3. The final result is "ground lightning". In some other embodiments, the total weight score of each category can also be output as its probability to provide richer information. For example, the output is {"ground lightning": 0.7, "cloud lightning": 0.3}. Users can set their own thresholds for judgment or use it for risk probability assessment.

[0039] In summary, the lightning identification method provided in this embodiment simultaneously acquires the preliminary identification results and raw data from each edge detector, providing a foundation for subsequent cross-validation. Retaining the raw data allows the central detector to independently verify the results, avoiding complete reliance on edge detector judgments and fully utilizing the real-time processing capabilities of the edge detectors, thus reducing the computational burden on the central detector. The weight of each edge detector's lightning identification result is calculated based on the lightning positioning error and the corresponding confidence level. Edge detectors with smaller lightning positioning errors and higher confidence levels have larger weights and play a more significant role in the final weighted voting. This dynamic weight adjustment method allows data from detectors with higher positioning accuracy to have a greater impact on the final result, helping to improve the overall accuracy of lightning positioning. That is, because this application uses data from multiple edge detectors for comprehensive judgment, even if one edge detector malfunctions or its data is inaccurate, the data from other normally functioning detectors can still influence the final result. Through the weighted voting mechanism, abnormal data from individual detectors can be offset to a certain extent, improving the system's fault tolerance to faults and errors and ensuring the reliability of the lightning identification results.

[0040] In some optional embodiments, the step of calculating the lightning positioning error based on the measurement error between the theoretical time and the measured time of the lightning signal arrival at the corresponding edge detection end includes: calculating the chi-square error based on the measurement error corresponding to each edge detection end, and using the chi-square error as the lightning positioning error.

[0041] In this embodiment, the chi-square error reflects the degree of consistency between these errors and theoretical expectations. If the measurement errors of each edge detector end are in good agreement with the theoretical model, the chi-square error will be relatively small; conversely, if the measurement errors deviate significantly from the theoretical expectations, the chi-square error will increase. This measurement of error consistency helps to more accurately assess the error situation of lightning positioning and determine whether the positioning results are reliable.

[0042] Specifically, the arrival time t_obs, i of the lightning signal actually measured by the edge observation terminal i, and the theoretical arrival time t_calc, i that should have been observed by the edge observation terminal i, calculated based on the lightning location theory result from the central terminal, are used to inversely determine the measurement error: Measurement error i = t_obs, i - t_calc, i; Furthermore, the chi-square error is calculated using the following formula. : ; Where N is the number of edge detection terminals. Let be the variance of the time measurement error of the i-th edge detector.

[0043] In some alternative embodiments, other error metrics may also be used, such as calculating the mean absolute error of the measurement errors at all edge observation points, or the root mean square error as the lightning location error.

[0044] In some alternative embodiments, a consistency check method can also be used to determine the lightning location error. For example, cross-validation can be used, where the location is performed using data from N-1 sites, the prediction error is verified by the Nth site, and this process is repeated for all sites. The average prediction error is then calculated as the lightning location error.

[0045] In some optional embodiments, the step of weighted voting on the lightning identification results of each edge detection terminal to obtain the final lightning identification result includes: quantifying the lightning identification results according to different types, performing weighted summation on the quantified lightning identification results; determining the numerical range to which the summation result belongs, and determining the final lightning identification result according to a preset mapping relationship between numerical range and lightning identification type.

[0046] In this embodiment, the numericalized lightning identification results are weighted and summed to fully consider the different importance and reliability of each edge detection end. By assigning corresponding weights to the identification results of each detection end, those detection ends with high positioning accuracy and strong data reliability play a greater role in the final decision. The numerical range to which the summed result belongs is determined, and the final lightning identification result is determined according to the preset mapping relationship between the numerical range and the lightning identification type, providing clear and unambiguous rules for the final result determination. This rule-based determination method avoids the subjectivity and arbitrariness of human judgment, making the determination of the final result repeatable and consistent.

[0047] Specifically, in obtaining normalized weights Next, the positioning processing software performs a weighted vote on the edge discrimination type and the positioning inference type based on the weight calculation results, and obtains the final lightning type output. The specific voting calculation method is as follows:

[0048] In the formula, The edge-end indicates whether it is a cloud-based lightning or a ground-based lightning (usually 0 indicates a cloud-based lightning and 1 indicates a ground-based lightning). H refers to the step function, which outputs 1 when the weighted result is greater than 0.5 and 0 when it is less than 0.5 (as shown in the formula below). This is the final result of the cloud-to-ground flash.

[0049]

[0050] In some optional embodiments, the method further includes: adjusting the confidence level of the edge detection terminal in the next lightning identification task based on the accuracy of lightning identification by the edge detection terminal in the current identification task.

[0051] Specifically, considering that in practical applications: 1) the confidence distribution of the edge model may change under different geographical regions, seasons, and electromagnetic noise environments; and 2) the statistical characteristics of the positioning chi-square error will also change under different station network densities and hardware performance, this embodiment introduces a feedback mechanism into the system to allow the system to adaptively adjust the confidence of each edge detector. By adjusting the confidence, the role of high-quality detectors is strengthened, the influence of inefficient detectors is weakened, and the final lightning identification result is ensured to be closer to the real situation.

[0052] In some optional embodiments, adjusting the confidence level of the edge detection terminal in the next lightning identification task based on the accuracy of lightning identification by the edge detection terminal in the current identification task includes: using the lightning identification result of the edge detection terminal corresponding to the measurement error that contributes the least to the lightning positioning error as the correction standard; and reducing the confidence level of the edge detection terminal in the next lightning identification for edge detection terminals that are different from the correction standard.

[0053] In this embodiment, the edge detection end corresponding to the measurement error that contributes the least to the lightning location error usually means that the data acquired by that detection end in this identification task is of higher quality and the detection result is more reliable. Using its lightning identification result as a correction standard can provide an accurate and reliable reference for the entire identification system, which helps to reduce the overall identification error caused by the data deviation of individual detection ends, thereby improving the accuracy of the next lightning identification.

[0054] Specifically, sites A, B, C, and D all had an initial confidence level of 0.8. Analysis following a single lightning event: Site A: Minimum contribution of measurement error, identification result = Cloud Flash; Site B: Consistent with A (Cloud Flash), with a moderate contribution to error; Site C: Inconsistent with A (ground flash), contributing significantly to the error; Site D: Inconsistent with A (ground flash), contributing the most to the error.

[0055] The adjustments are as follows: Site A: As a reference standard, the confidence level is increased to 0.84; Site B: Consistent with the reference, confidence level remains at 0.8; Site C: Inconsistent with the reference, confidence level drops to 0.72; Site D: Inconsistent with the reference, confidence level drops to 0.64.

[0056] In some alternative embodiments, when adjusting the confidence level, the confidence level can be adjusted based on discrete levels. For example, if it is consistent with the reference standard, the confidence level can be slightly increased or maintained; if it is inconsistent with the reference, it can be decreased according to the level. For example, for sites with high expectations, the disappointment level is large and the confidence level decreases more, while for sites with low expectations, the penalty is relatively small.

[0057] In some optional embodiments, the confidence level can be adjusted based on a continuous function. Specifically: 1) First, calculate the error factor: ; in, Let e ​​be the error factor, and e be the error contribution. The smaller the contribution, the larger the factor. 2) Then calculate the adjustment factor: ; in, The adjustment factor is s, which is the consistency score. s ranges from 0 to 1, where 1 indicates complete consistency with the reference result and 0 indicates complete inconsistency. 3) Next, calculate the new confidence level: ; in, For the new confidence level, This represents the current confidence level. This calculation ensures that the confidence level is adjusted smoothly, avoiding drastic changes.

[0058] 4) Finally, truncate:

[0059] In some optional embodiments, the method for determining the measurement error that contributes the least to the lightning positioning error includes: calculating the chi-square residual corresponding to the measurement error of each edge detection end; the smaller the chi-square residual, the smaller its contribution to the lightning positioning error.

[0060] In this embodiment, the chi-square residual provides a precise quantitative value for the contribution of measurement error to lightning location error. Compared to traditional subjective judgment or simple qualitative analysis, the chi-square residual can clearly and intuitively reflect the correlation between the measurement error of each edge detector and the lightning location error with a specific numerical value. It can intuitively determine which detector's data is more reliable and contributes less to the location error.

[0061] Specifically, the chi-square residual of edge detection terminal i = .

[0062] In some alternative embodiments, error contribution can also be calculated using other quantification methods, such as relative error contribution method, influence function method, relative entropy contribution, etc.

[0063] Example 2: Based on the above embodiments, this embodiment provides an application example. This embodiment provides a lightning detection system, including: Multiple edge detection terminals, each of which is used to analyze and obtain a lightning identification result and a confidence level corresponding to the lightning identification result based on the raw lightning data it receives; The central terminal receives the lightning identification results, corresponding confidence levels, and raw lightning data sent from each of the edge detection terminals. Based on the raw lightning data, it parses to obtain the theoretical value for lightning location and uses this theoretical value to invert the theoretical time for the lightning signal to reach each of the edge detection terminals. Based on the measurement error between the theoretical time and the actual arrival time of the lightning signal at the corresponding edge detection terminal, it calculates the lightning location error. For each edge detection terminal, it uses the lightning location error and the corresponding confidence level to calculate the weight of the lightning identification result for that edge detection terminal, with the sum of the weights for each edge detection terminal being 1. Finally, it performs a weighted voting process on the lightning identification results of each edge detection terminal to obtain the final lightning identification result.

[0064] Specifically, by deploying a lightweight deep learning model on the detection device, the acquired raw lightning waveforms are intelligently identified in real time, enabling rapid differentiation of lightning types. Unlike existing methods that rely on centralized processing, the technical solution in this embodiment can complete edge recognition as soon as the data is generated, thereby reducing the latency of data transmission and centralized processing.

[0065] In the specific implementation process, the detection device first acquires broadband electromagnetic signals through a high-speed analog-to-digital converter module and preprocesses them, including DC bias removal, normalization, and bandpass filtering. The preprocessed waveform data is then fed into an embedded deep learning model for real-time inference. The model employs a structure combining a one-dimensional convolutional neural network and a temporal attention mechanism to consider both local waveform features and overall temporal trends. During the training phase, this embodiment utilizes a large dataset of labeled lightning waveforms for supervised learning and incorporates unlabeled data using a semi-supervised training strategy to enhance the model's generalization ability. During the inference phase, the model can output the probability of whether the event belongs to cloud-based lightning or ground-based lightning within milliseconds, while also providing the corresponding confidence index.

[0066] To ensure the feasibility of this method on embedded hardware, this embodiment performs quantization and structural pruning optimization on the model to ensure stable operation on low-power processors or neural network acceleration units. In terms of system architecture, the detection device acts as an edge computing node; its output classification results can be stored and displayed locally, and can also be uploaded to the central server via a communication link. During the localization calculation phase, when the central terminal cannot accurately obtain altitude information due to insufficient station numbers, it can directly use the classification results from the edge device as constraints, thereby improving the reliability and completeness of the final localization result.

[0067] Through the above technical solution, this embodiment achieves an organic combination of edge intelligence and central computation, enabling the lightning detection system to obtain stable lightning type identification results even when site coverage is sparse or observation conditions are limited. This method effectively improves the real-time performance and robustness of the overall system, providing more reliable technical support for ground lightning monitoring, cloud lightning activity research, and related disaster prevention and mitigation applications.

[0068] This embodiment proposes an optimized lightning type discrimination scheme based on multi-source information fusion, namely a dynamic weighting identification mechanism for the localization processing software. Its core idea is to fuse the intelligent waveform identification results at the edge end with the localization solution results at the center end, and then optimize the calculation through a dynamic weighting allocation model, thereby improving the overall reliability of lightning type identification. The specific scheme is as follows: Data input: The central server uses the lightweight deep learning model embedded in each observation device to identify the lightning type (cloud lightning / ground lightning) and the corresponding confidence score in real time, as well as the chi-square error of the two-dimensional / three-dimensional coordinate information of the lightning obtained by the localization processing algorithm.

[0069] Basic dynamic weight calculation method for weight calculation model: Construct a weight function when the system determines the lightning type.

[0070] In the formula, The confidence level of the waveform identification result at the corrected i-th station is... It is the inference confidence of the waveform discrimination model for the i-th station. This represents the chi-square error in the positioning calculation. This is the dynamic correction factor obtained after calibration based on the chi-square test.

[0071] After obtaining the weights, the positioning processing software performs a weighted vote on the edge discrimination type and the positioning inference type based on the weight calculation results, and obtains the final lightning type output. The specific voting calculation method is as follows:

[0072] In the formula, The edge-end indicates whether it is a cloud-based lightning or a ground-based lightning (usually 0 indicates a cloud-based lightning and 1 indicates a ground-based lightning). H refers to the step function, which outputs 1 when the weighted result is greater than 0.5 and 0 when it is less than 0.5 (as shown in the formula below). This is the final result of the cloud-to-ground flash.

[0073]

[0074] Dynamic weight feedback mechanism: Considering that in practical applications: 1) the confidence distribution of the edge model may change under different geographical regions, seasons, and electromagnetic noise environments; 2) the statistical characteristics of the positioning chi-square error will also change under different station network densities and hardware performance, this invention introduces a feedback mechanism into the system to allow the system to adaptively adjust the confidence calculation method. The specific method is as follows: The confidence correction process is as follows: Figure 3 As shown, after acquiring multi-site observation data at the edge computing end, the model quickly outputs preliminary positioning results and simultaneously provides a confidence index based on waveform features and model inference. This confidence index is uploaded to the central processing end along with the results.

[0075] Subsequently, at the central processing end, when results from multiple sites are received, the system will introduce a dynamic weighting factor generated by the chi-square error test based on time difference positioning, on the basis of the confidence level at the edge end. This dynamic weighting factor will be applied to the model inference confidence level at the edge computing end of each site to generate the final lightning type identifier after comprehensive judgment.

[0076]

[0077] In the formula, The confidence level of the waveform identification result at the corrected i-th station is... The waveform discrimination model infers the current confidence level for the i-th station. This is the dynamic correction factor obtained after calibration based on the chi-square test.

[0078] Furthermore, the central endpoint will use the result obtained from time-of-flight positioning, where the chi-square residual meets a preset threshold, as a high-confidence reference result. This result has higher statistical reliability and can serve as a benchmark for weight calibration. When the initial judgment at the edge endpoint is consistent with this high-confidence reference result, the system will increase the weight of this edge result in subsequent events, i.e., by adding a dynamic correction factor. To increase confidence; conversely, if there is a discrepancy between the two, the weight of the intelligent waveform identification result of the site is reduced accordingly, that is, by reducing the dynamic correction factor. This reduces the confidence level. Through continuous feedback and iterative calibration, the dynamic correction factor will gradually converge, thereby maintaining stability and reliability in complex environments.

[0079] Ultimately, the mechanism forms a dynamic, closed-loop weight update system: the edge ensures real-time inference and rapid response, while the central end provides statistical correction based on large-scale observations. The two work together to ensure that the overall positioning results are both real-time and accurate, and can be automatically and continuously optimized according to the site conditions in long-term applications.

[0080] The purpose of this embodiment is to address the shortcomings of existing lightning location systems in practical applications by proposing a lightning type discrimination method that combines edge intelligent recognition and central collaborative optimization, thereby improving the application value and reliability of the system under sparse observation site conditions. Specifically, it includes: 1. Solve the problem of missing height information: Overcome the limitations of the existing time difference method, which can only obtain two-dimensional positioning results when the number of detection stations is insufficient (≤4) and cannot accurately distinguish between cloud lightning and ground lightning, so that the system still has reliable lightning type identification capability under sparse observation network conditions.

[0081] 2. Overcoming the data transmission bandwidth bottleneck: By embedding a lightweight deep learning model at the observation device, real-time edge intelligent identification can be performed directly using the original waveform data, avoiding the problem of being unable to upload waveforms to the central station due to insufficient network bandwidth, and reducing the computational pressure and dependence of the central station.

[0082] 3. Improve the robustness and reliability of the discrimination results: Design a dynamic weight labeling mechanism in the positioning processing software to integrate the confidence level of the edge waveform identification with parameters such as the chi-square error of the positioning solution, so as to realize the collaborative optimization discrimination of multi-source information and improve the accuracy and stability of the lightning type output results.

[0083] In summary, the purpose of this invention is to propose a lightning type identification method and system based on edge intelligence and center collaborative optimization, which can achieve high-confidence discrimination of lightning type under sparse observation network conditions and break through the application bottleneck of traditional positioning methods in two-dimensional results.

[0084] Example 3: Another embodiment of this application relates to a lightning detection device. The implementation details of the lightning detection device in this embodiment are described below. The following implementation details are provided for ease of understanding and are not essential for implementing this solution. A schematic diagram of the lightning detection device in this embodiment can be seen as follows: Figure 4 As shown, the central end of the lightning detection system is applied. The lightning detection system includes multiple edge detection ends and the central end, including a receiving module 410, an error calculation module 420, a weight calculation module 430 and an identification module 440.

[0085] The receiving module 410 is used to receive lightning identification results, the confidence level corresponding to the lightning identification results, and the original lightning data sent from each of the edge detection terminals; The error calculation module 420 is used to obtain the theoretical value of lightning location based on the original lightning data, and to invert the theoretical time of the lightning signal reaching each of the edge detection ends using the theoretical value of lightning location. The lightning location error is obtained based on the measurement error between the theoretical time and the measured time of the lightning signal arrival at the corresponding edge detection end. The weight calculation module 430 is used to calculate the weight of the lightning identification result of each edge detection end by using the lightning positioning error and the confidence level corresponding to the edge detection end. The sum of the weights corresponding to each edge detection end is 1. The identification module 440 is used to perform weighted voting on the lightning identification results of each edge detection end to obtain the final lightning identification result.

[0086] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed in this application; however, this does not mean that other units are absent in this embodiment.

[0087] In some optional embodiments, the error calculation module is used for: The chi-square error is calculated based on the measurement error corresponding to each edge detection end, and the chi-square error is used as the lightning positioning error.

[0088] In some optional embodiments, the identification module includes: The weighted summation unit is used to quantify the lightning identification results according to different types and perform weighted summation on the quantified lightning identification results; The result determination unit is used to determine the numerical range to which the summed result belongs, and to determine the final lightning identification result according to the preset mapping relationship between the numerical range and the lightning identification type.

[0089] In some optional embodiments, it also includes: The confidence adjustment module is used to adjust the confidence level of the edge detection terminal in the next lightning identification task based on the accuracy of lightning identification by the edge detection terminal in the current identification task.

[0090] In some optional embodiments, the confidence adjustment module includes: The correction standard determination unit is used to take the lightning identification result of the edge detection end corresponding to the measurement error that contributes the least to the lightning positioning error as the correction standard; The comparison and adjustment unit is used to reduce the confidence level of the edge detection end during the next lightning identification for edge detection ends that are different from the correction standard.

[0091] In some optional embodiments, the correction criterion determination unit is used for: Calculate the chi-square residual corresponding to the measurement error of each edge detection end. The smaller the chi-square residual, the smaller its contribution to the lightning positioning error.

[0092] Example 4: Another embodiment of this application relates to an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the lightning recognition method in the above embodiments.

[0093] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0094] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0095] Example 5: Another embodiment of this application relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.

[0096] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0097] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.

Claims

1. A method for lightning recognition, characterized in that, The method, applied to the central end of a lightning detection system, which includes multiple edge detection ends and the central end, comprises: Receive lightning identification results, the confidence level corresponding to the lightning identification results, and the original lightning data sent from each of the edge detection terminals; Based on the original lightning data analysis, the theoretical value of lightning location is obtained, and the theoretical time for the lightning signal to reach each of the edge detection terminals is obtained by inversion using the theoretical value of lightning location. Based on the measurement error between the theoretical time and the actual time of the lightning signal arrival at the corresponding edge detection terminal, the lightning location error is calculated. For each edge detection terminal, the weight of the lightning identification result of the edge detection terminal is calculated using the lightning positioning error and the confidence level corresponding to the edge detection terminal, and the sum of the weights corresponding to each edge detection terminal is 1; The lightning identification results of each edge detection terminal are weighted and voted to obtain the final lightning identification result.

2. The lightning identification method according to claim 1, characterized in that, The lightning location error is calculated based on the measurement error between the theoretical time and the measured arrival time of the lightning signal at the corresponding edge detection end, including: The chi-square error is calculated based on the measurement error corresponding to each edge detection end, and the chi-square error is used as the lightning positioning error.

3. The lightning recognition method according to claim 1, characterized in that, The weighted voting of the lightning identification results from each edge detection terminal to obtain the final lightning identification result includes: The lightning identification results are quantified according to different types, and the quantified lightning identification results are weighted and summed. Determine the numerical range to which the summation result belongs, and determine the final lightning identification result based on the preset mapping relationship between the numerical range and the lightning identification type.

4. The lightning identification method according to claim 1, characterized in that, Also includes: Based on the accuracy of lightning identification by the edge detection terminal in the current identification task, adjust the confidence level of the edge detection terminal in the next lightning identification task.

5. The lightning identification method according to claim 4, characterized in that, The step of adjusting the confidence level of the edge detection terminal in the next lightning identification task based on the accuracy of lightning identification by the edge detection terminal in the current identification task includes: The lightning identification result of the edge detection end corresponding to the measurement error that contributes the least to the lightning positioning error is used as the correction standard; For edge detectors that differ from the correction standard, reduce the confidence level of the edge detector during the next lightning detection.

6. The lightning identification method according to claim 5, characterized in that, The method for determining the measurement error that contributes the least to the lightning positioning error includes: Calculate the chi-square residual corresponding to the measurement error of each edge detection end. The smaller the chi-square residual, the smaller its contribution to the lightning positioning error.

7. A lightning detection system, characterized in that, include: Multiple edge detection terminals, each of which is used to analyze and obtain a lightning identification result and a confidence level corresponding to the lightning identification result based on the raw lightning data it receives; The central terminal is used to receive the lightning identification results, corresponding confidence levels, and raw lightning data sent from each of the edge detection terminals; it parses the raw lightning data to obtain the theoretical value of lightning location, and uses the theoretical value of lightning location to invert the theoretical time for the lightning signal to reach each of the edge detection terminals; it calculates the lightning location error based on the measurement error between the theoretical time and the measured time of the lightning signal arrival at the corresponding edge detection terminal; for each edge detection terminal, it calculates the weight of the lightning identification result of the edge detection terminal using the lightning location error and the confidence level corresponding to the edge detection terminal, and the sum of the weights corresponding to each edge detection terminal is 1; The lightning identification results of each edge detection terminal are weighted and voted to obtain the final lightning identification result.

8. A device for lightning detection, characterized in that, A device for use at the central end of a lightning detection system, the lightning detection system comprising multiple edge detection ends and the central end, the device comprising: The receiving module is used to receive lightning identification results, the confidence level corresponding to the lightning identification results, and the original lightning data sent from each of the edge detection terminals; The error calculation module is used to parse the original lightning data to obtain the theoretical value of lightning location, and use the theoretical value of lightning location to invert the theoretical time of the lightning signal reaching each of the edge detection terminals. The lightning location error is obtained based on the measurement error between the theoretical time and the actual time of the lightning signal arrival at the corresponding edge detection terminal. The weight calculation module is used to calculate the weight of the lightning identification result of each edge detection end by using the lightning positioning error and the confidence level corresponding to the edge detection end. The sum of the weights corresponding to each edge detection end is 1. The identification module is used to perform weighted voting on the lightning identification results of each edge detection end to obtain the final lightning identification result.

9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the lightning recognition method as described in any one of claims 1 to 6.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the lightning recognition method according to any one of claims 1 to 6.

Images