Electromagnetic pulse data transmission method

Abstract

The invention relates to the technical field of electromagnetic pulses, and discloses an electromagnetic pulse data transmission method. The method comprises the following steps: collecting natural electromagnetic pulse signal data of a detection target area through a six-dimensional complementary electrode array structure, and executing Kalman filtering peak extraction and characteristic frequency locking to obtain frequency-optimized signal data; executing first-order and second-order gradient difference matrix calculation and space vector synthesis to obtain signal source position coordinate data and signal intensity data; performing spectral domain reconstruction and multilayer dynamic threshold analysis to obtain a spatial clustering abnormal data packet; and selecting an optimal transmission channel to transmit the spatial clustering abnormal data packet to a cloud server. According to the invention, the influence of an external interference source is effectively eliminated, accurate extraction and characteristic frequency locking of effective frequency band signals are realized, power frequency and harmonic interference thereof are effectively suppressed, the signal-to-noise ratio of the effective frequency band of the signals is improved, and the reliability and timeliness of data transmission in a severe field environment are ensured.

Description

Technical Field

[0001] The present application relates to the field of electromagnetic pulse technology, and in particular to an electromagnetic pulse data transmission method. Background Art

[0002] As a non-invasive geological detection method, natural electromagnetic pulse signal technology has broad application prospects in the fields of geological disaster early warning, mine safety monitoring and underground structure analysis. Traditional electromagnetic pulse signal monitoring systems mainly rely on a single receiver or a simple array structure to collect data. The signal acquisition accuracy is limited and the anti-interference ability is insufficient, which makes it difficult to meet the needs of high-precision detection in complex geological environments. Especially in the monitoring process of ground stress concentration areas, due to the complex electromagnetic characteristics of the geological body and multi-source electromagnetic interference, the signal-to-noise ratio of the collected signal is low and the characteristics are fuzzy, which seriously affects the accuracy of subsequent analysis and processing.

[0003] The electromagnetic pulse data transmission system in the existing technology generally has problems such as low data processing efficiency, large transmission bandwidth occupancy, and poor real-time performance. Traditional methods usually use fixed parameter filtering algorithms and simple spectrum analysis techniques, which cannot effectively deal with the non-stationary and time-varying characteristics of geological electromagnetic signals. At the same time, due to the lack of effective spatial positioning means, most systems find it difficult to accurately lock the signal source position and intensity distribution, resulting in insufficient accuracy in identifying abnormal areas. In addition, the communication conditions in the field detection environment limit a large amount of raw data, making it impossible to transmit a large amount of raw data to the processing center in a timely and efficient manner, restricting the real-time response capability of the geological disaster early warning system. Summary of the invention

[0004] The present application provides an electromagnetic pulse data transmission method, which effectively eliminates the influence of external interference sources, realizes the precise extraction of effective frequency band signals and characteristic frequency locking, effectively suppresses the interference of industrial frequency and its harmonics, improves the signal-to-noise ratio of the signal in the effective frequency band, and ensures the reliability and timeliness of data transmission in harsh outdoor environments.

[0005] In a first aspect, the present application provides an electromagnetic pulse data transmission method, the electromagnetic pulse data transmission method comprising: The natural electromagnetic pulse signal data of the detection target area is collected through a six-dimensional complementary electrode array structure, and Kalman filter peak extraction and characteristic frequency locking are performed to obtain frequency-optimized signal data; Performing first-order and second-order gradient difference matrix calculation and space vector synthesis on the frequency-optimized signal data to obtain signal source position coordinate data and signal strength data; Performing spectral domain reconstruction and multi-layer dynamic threshold analysis on the signal source position coordinate data and the signal strength data to obtain spatial clustering abnormal data packets; Block-level differential compression is performed on the spatial clustering abnormal data packets to obtain compressed data, and the optimal transmission channel is selected through the heterogeneous network self-organizing switching technology to transmit the compressed data to the cloud server.

[0006] In the technical solution provided by the present application, the six-dimensional complementary electrode array structure adopts an equilateral triangle grid layout to achieve high-sensitivity acquisition of omnidirectional electromagnetic pulse components, and cooperates with high-precision GPS clock synchronization and reference site settings to effectively eliminate the influence of external interference sources; spectrum adaptive tracking is combined with Kalman filtering technology to achieve accurate extraction of effective frequency band signals and characteristic frequency locking, effectively suppressing power frequency and its harmonic interference, and improving the signal-to-noise ratio of the signal effective frequency band; the first-order and second-order gradient difference matrix calculation is combined with space vector synthesis technology to improve the signal source positioning accuracy, and the accurate inversion of the signal source is achieved through principal component analysis and multilateral positioning algorithm, making the stress field distribution characteristics clearer. The signal enhancement method combining wavelet packet transform with singular value decomposition realizes high-resolution reconstruction of signal time-frequency characteristics, effectively removes random noise components while retaining key information, and enhances the weak signal detection capability; the anomaly recognition scheme combining multi-layer dynamic threshold analysis with density clustering algorithm improves the accuracy of abnormal area recognition and reduces the false alarm rate by considering multiple threshold conditions such as global statistics, regional relative and time series changes; block-level differential compression and heterogeneous network self-organizing transmission strategy achieve high compression rate while keeping key feature information intact, and combines forward error correction and network automatic switching technology to ensure data transmission reliability and timeliness in harsh field environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0008] Figure 1 This is a schematic diagram of an embodiment of the electromagnetic pulse data transmission method in the embodiment of the present application. DETAILED DESCRIPTION

[0009] The embodiment of the present application provides an electromagnetic pulse data transmission method. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0010] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the electromagnetic pulse data transmission method in the embodiment of the present application includes: Step S101, collecting natural electromagnetic pulse signal data of the detection target area through a six-dimensional complementary electrode array structure, and performing Kalman filter peak extraction and characteristic frequency locking to obtain frequency-optimized signal data; It is understandable that the execution subject of the present application may be an electromagnetic pulse data transmission system, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0011] Specifically, according to the topographic structure and geological target distribution of the detection area, the detection area is gridded using an equilateral triangle arrangement pattern, and the deployment plan of the six-dimensional complementary electrode array is determined accordingly. At each set electrode receiving point, a six-dimensional complementary electrode array unit with a consistent structure is installed. The array structure consists of three pairs of mutually orthogonal vector electromagnetic induction coils, which are used to collect electric field and magnetic field vector information in the three axes of X, Y, and Z, respectively, so as to achieve full-space, high-resolution, and multi-dimensional reception coverage of natural electromagnetic pulse signals. The induction coil structure is made of highly sensitive magnetoelectric materials and is equipped with silver-silver chloride composite electrodes to synchronously collect electric field changes. At the same time, multiple electrode receiving devices arranged at equal intervals are embedded in each receiving unit to construct a gradient difference measurement structure to enhance the subsequent spatial positioning accuracy. In order to ensure the synchronization and coordination of all six-dimensional complementary electrode arrays in large-area layout, a high-precision GPS clock synchronization module is integrated in each array unit, which achieves a time alignment accuracy better than 10 nanoseconds, so that the subsequent phase difference calculation and gradient difference fitting are not disturbed by time errors; at the same time, each electrode array node is equipped with a low-power wireless transmission module, which supports the construction of a self-organizing mesh network architecture, realizes adaptive routing, data forwarding and state coordination between nodes, builds an edge-level data acquisition network through a distributed real-time communication system, and enables the raw data of the entire detection area to be gathered to the main control processing platform without delay. In order to enhance the system's ability to identify and suppress external environmental interference signals, reference sites are deployed at E typical locations on the periphery of the entire target area. Each reference site is equipped with a six-dimensional electrode unit equivalent to the detection node, which is used to record the background electromagnetic interference input from the outside of the area in real time, including broadband noise information generated by industrial sources, communication systems, meteorological changes, etc. The interference data obtained is used as a noise calibration benchmark in subsequent electromagnetic signal analysis. On the basis of the above, the entire detection target area forms an original electromagnetic data acquisition system, and real-time communication is achieved through the mesh network. Each node uploads the electromagnetic vector signal (including instantaneous electric field, current change rate and spatial gradient) collected locally to the central processing unit to form a preliminary set of original electromagnetic pulse vector data. The background interference data collected by the reference station is introduced, and the correlation analysis of each channel and frequency band with the original data of the target area node is carried out. Through the joint comparison of the frequency domain and the time domain, the background interference components with high correlation are identified and eliminated, and the purified natural electromagnetic pulse signal data is obtained by calibration.After the calibration is completed, the processed electromagnetic vector signal data is input into the spectrum adaptive tracking unit, and a fixed-length window (such as 2048 points) fast Fourier transform is performed to obtain a complete spectrum diagram of the electromagnetic signal in the range of 0.1Hz to 10kHz. The peak extraction algorithm based on Kalman filtering is applied to the spectrum diagram to automatically identify the main frequency components. The algorithm continuously tracks the energy change trajectory of each frequency component by constructing a dynamic state estimation model, and extracts the stable main frequency characteristics based on the sliding superposition of multiple windows. After the stable main frequency is identified, the frequency is matched with the preset geological characteristic frequency template, and the matching threshold (such as 85%) is set to ensure that the identified frequency has a clear geological directivity; when the match is successful, the system automatically locks the frequency interval and dynamically adjusts the subsequent data sampling frequency to eight times the characteristic frequency, thereby achieving frequency optimization on the basis of satisfying the Nyquist sampling theorem. Output a set of frequency-optimized signal data.

[0012] The natural electromagnetic pulse signal data is preprocessed to enhance the identifiability of its spectral characteristics and suppress noise interference. Since power frequency interference (such as 50Hz or 60Hz) is ubiquitous in the natural electromagnetic environment and has strong periodicity, the power frequency notch filtering technology is applied to the input signal data, and the interference frequency of the main power system and its harmonic components are effectively removed from the signal using a digital notch filter with adjustable center frequency. On this basis, in order to eliminate high-frequency noise and non-stationary background disturbances, the wavelet threshold denoising algorithm is used to perform multi-scale decomposition on the signal after notch processing. By selecting a suitable wavelet basis function (such as Daubechies or Symlet), the signal is decomposed into multiple frequency bands and a soft threshold function is set. The coefficients of each frequency band are denoised in an adaptive manner, and a set of denoised electromagnetic pulse signals with a higher signal-to-noise ratio are reconstructed. The denoised signal is input into the spectrum adaptive tracking unit and converted to the frequency domain. The signal in the entire time window is spectrally unfolded using discrete Fourier transform, and its power spectrum density is calculated to obtain the energy distribution of the signal at each frequency point. The power spectrum curve is analyzed based on the principle of local maximum, and the frequency points with significant energy peaks in the spectrum are marked and their positions, amplitudes and spectrum widths are recorded to form the initial spectrum feature data for subsequent modeling processing. In order to enhance the stability analysis capability of the frequency component, the extracted initial spectrum features are used as input to construct a state space model, the core of which includes the state transfer matrix and the observation matrix to describe the evolution law of the frequency component in the time series and the influence of observation noise. The model is embedded in the Kalman filter algorithm to perform state estimation and prediction in real time, eliminate background disturbances and system errors, and obtain a set of continuous and smooth filtered spectrum trajectory data. Based on the filtered spectrum trajectory data, the drift rate of the spectrum peak in three adjacent time windows is calculated, which is defined as the ratio of the peak frequency difference to the time difference in adjacent windows, and compared with the preset stability target value. When the drift rate of a certain frequency point is lower than the threshold and stable features appear continuously, it is considered to have long-term traceability and physical representativeness. The system marks the frequency point as a "locked frequency" and records its characteristic parameters in the frequency locking parameter set. The parameter set contains the frequency value, as well as its energy, stability coefficient and bandwidth characteristics, which are used to guide subsequent analysis and processing. Based on the above frequency locking parameter set, the Hilbert transform of the denoised signal is processed with an adaptive window, and a refined joint time-frequency analysis is performed by constructing an analytical signal and calculating its instantaneous amplitude and instantaneous frequency. The sliding window mechanism is used to perform local analysis on different time periods, so that the signal is analyzed in both time and frequency dimensions, forming a set of enhanced time-frequency feature matrices containing energy, frequency and time information, showing the evolution of frequency components, and distinguishing the main part of the signal from the noise interference part.The frequency selective attenuation mechanism is implemented on the basis of the enhanced time-frequency characteristic matrix, that is, the non-characteristic frequency components are uniformly subjected to 50% energy attenuation, while the signal components marked as locked frequencies are completely retained. The adjusted time-frequency characteristic matrix is ​​inversely processed, including inverse Hilbert transform and time domain signal reconstruction, and a set of frequency-optimized electromagnetic pulse signal data is output.

[0013] Step S102, performing first-order and second-order gradient difference matrix calculation and space vector synthesis on the frequency-optimized signal data to obtain signal source position coordinate data and signal strength data; Specifically, based on the signal data after frequency optimization, a multi-point measurement framework is established in three-dimensional space, and based on the distribution characteristics of the six-dimensional complementary electrode array in the three axes of X, Y, and Z, according to the geometric structure between adjacent measuring points in each axis, the intensity difference of the signal at these measuring points is calculated. The intensity difference reflects the gradient change generated during the propagation of electromagnetic waves in different directions, and a three-dimensional first-order gradient difference matrix composed of multiple measuring point differences is constructed in each direction. Each element in the first-order gradient difference matrix is ​​divided by the actual distance between the corresponding measuring points, that is, spatial normalization is performed to obtain the electromagnetic signal intensity change rate per unit distance, which constitutes the standardized first-order gradient field distribution data. Based on the first-order gradient field distribution data, its spatial change rate is calculated, that is, the gradient of the gradient is calculated in each direction to form a second-order gradient quantity. The first-order gradient field is differentially operated along each spatial dimension using the central difference method to construct a three-dimensional second-order gradient difference matrix to represent the degree of curvature change of the electromagnetic signal in space. The Gaussian smoothing filter algorithm is applied to the second-order gradient difference matrix, and the spatial convolution operation is used to perform weighted averaging on the neighboring points, so as to effectively suppress local peaks and edge distortions and obtain more continuous and physically reasonable second-order gradient field distribution data. The first-order gradient field distribution data and the second-order gradient field distribution data are combined into gradient field synthesis data, which reflects the whole process of electromagnetic signal evolution from first-order linear change to second-order curvature in three-dimensional space. In order to extract the spatial variation characteristics with physical dominant significance from the synthetic gradient field, principal component analysis is performed on the data. By calculating the covariance matrix of the data and solving its eigenvectors and eigenvalues, dimensionality reduction processing is achieved, and the spatial vector field characteristic parameters containing the main energy information are extracted. Based on the spatial vector field characteristic parameters, the inversion mathematical model of the electromagnetic signal source is constructed. According to the basic relationship of the electromagnetic field spatial propagation in the Maxwell equations, the mapping function between the signal intensity gradient and the signal source position is established, and the principal component obtained by the principal component analysis is introduced as a constraint term into the inversion equation to form a restricted optimization problem. The model is solved by the least squares fitting algorithm to deduce the position coordinate data (X, Y, Z) of the signal source in three-dimensional space, and the spatial distance of each point in the entire measurement area relative to the signal source is calculated based on this. Using the attenuation model of electromagnetic waves propagating in different media, the signal strength is associated with parameters such as distance, dielectric constant, conductivity and magnetic permeability, and the change function of the signal strength in three-dimensional space is derived to generate a signal strength distribution map in the entire space.

[0014] Step S103, performing spectrum domain reconstruction and multi-layer dynamic threshold analysis on the signal source position coordinate data and the signal strength data to obtain spatial clustering abnormal data packets; Specifically, the spatial positioning result of the signal source and its corresponding electromagnetic pulse signal strength data are fused in spatial correspondence. Taking each positioning moment as the benchmark, the corresponding three-dimensional coordinate data (X, Y, Z) is mapped one by one with the signal strength value at that moment, and a time-space domain signal block sequence with clear geographic spatial significance is constructed. Wavelet packet transform is performed on the time-space domain signal block sequence to extract the local characteristics of the signal in different frequency bands. Wavelet packet transform further decomposes the high-frequency and low-frequency sub-bands at the same time, so that the entire signal is unfolded on a fine-grained frequency domain structure to construct a multi-level frequency band tree structure. After completing the multi-level decomposition, the coefficient energy calculation results of each sub-band are summarized to generate a frequency band weight coefficient table, which reflects the energy distribution of the signal in different frequency bands. Based on the frequency band weight coefficient table, soft threshold shrinkage processing is performed on all wavelet packet coefficients, that is, an adaptive soft threshold function is used to modify each coefficient, while retaining the main information, the low-amplitude component is weakened, and a set of optimized wavelet packet coefficients are obtained. After reorganizing the optimized wavelet packet coefficients, a two-dimensional signal matrix is ​​constructed. Each row or column of the matrix represents the spectrum structure under different time or space dimensions, forming a high-dimensional data set reflecting the characteristics of spatiotemporal evolution. In order to extract the most representative principal component structure in the matrix, the singular value decomposition algorithm is introduced to decompose the signal matrix into three parts: the left singular matrix, the singular value diagonal matrix and the right singular matrix. Each singular value in the singular value diagonal matrix represents the contribution of the corresponding principal component to the overall signal. These singular values ​​are sorted and the retention threshold is set. For example, the first 5 to 10 main singular value components that occupy more than 90% of the total energy are retained, and the remaining small singular values ​​are set to zero, thereby suppressing the secondary noise information and retaining only the most significant principal component structure in the signal. By reconstructing the singular value diagonal matrix, combining the left singular matrix and the right singular matrix, the denoising and reconstruction process of the signal is completed, and a set of signal reconstruction data with dominant characteristics after denoising is obtained. On this basis, the denoised signal data is subjected to frequency band decomposition, and the signal is remapped to each frequency sub-band according to the original decomposition structure in the wavelet packet tree, and weighted fusion is performed according to the energy distribution in the sub-band to form an enhanced time-frequency distribution signal with high signal-to-noise ratio and time-frequency resolution. A multi-layer dynamic threshold analysis mechanism is introduced to perform hierarchical screening and classification of the enhanced time-frequency distribution signal. The key statistical indicators such as the mean, standard deviation, kurtosis and skewness of the enhanced signal at each time and space point are counted, and a dynamic threshold system is set accordingly, where the first layer of thresholds corresponds to anomaly detection that slightly deviates from the baseline, the second layer of thresholds is used to identify significant abnormal signals, and the third layer of thresholds focuses on capturing strong anomalies with extremely strong energy mutation. Each layer of thresholds is dynamically updated according to local statistical characteristics and the global baseline, and has environmental self-adaptation capabilities, and can automatically adapt to changes in background conditions such as seasonal changes, geological disturbances or human interference.After the multi-layer threshold analysis is completed, the signal points that meet the abnormal conditions are spatially clustered, and the density clustering algorithm (such as DBSCAN) is used to divide the adjacent abnormal points into spatial aggregates, namely spatial abnormal clusters. Each abnormal cluster contains key attributes such as its center position, average abnormal level, spatial extension range and duration. The description information of each abnormal cluster is encapsulated in a standardized format to construct a spatial clustering abnormal data package, which contains spatial coordinates, time-frequency characteristics, abnormal level, energy trend, first appearance time and duration.

[0015] The enhanced time-frequency distribution signal is structured to accurately express the change law of the signal in three dimensions: space, time and frequency. The enhanced time-frequency distribution signal is projected into a three-dimensional coordinate system, which constructs a three-dimensional information volume with spatial position, time index and frequency component as three dimensions. By setting the spatial resolution, time sampling interval and frequency bandwidth step, the entire three-dimensional continuous signal field is divided into multiple discrete units of finite size. Each discrete unit is defined as a voxel, which contains the electromagnetic signal characteristic values ​​in the spatial position, time period and frequency range, such as instantaneous energy, spectral density, frequency stability coefficient or coherence amplitude, so that the system obtains the signal feature voxel data matrix, which is organized in the form of a regular three-dimensional grid. Multi-layer threshold discrimination conditions are constructed based on the signal feature voxel data matrix. In order to ensure the accuracy and robustness of anomaly detection, a global statistical analysis is performed on the entire voxel matrix, and key parameters such as the mean, standard deviation, kurtosis and skewness of each type of signal feature are extracted. Based on these statistical indicators, a dynamic threshold model containing multiple levels is constructed. The model consists of three layers, which are used to identify slight anomalies, significant anomalies, and strong anomalies respectively; its threshold expression is formalized as , , , where μ is the average value of the feature and σ is the standard deviation. The feature value of each voxel is input into the multi-layer threshold discrimination model, and the voxels are evaluated point by point against the thresholds at each level to determine whether it meets the abnormal judgment conditions of a certain level. Through this process, all voxel units that meet the abnormal conditions are screened out in the entire three-dimensional voxel space to form a preliminary abnormal voxel set, which exists in the form of discontinuous discrete voxel point groups and represents the spatial electromagnetic anomaly area. Spatial density clustering is performed on the preliminary abnormal voxel set, and a density-based clustering algorithm, such as the DBSCAN algorithm, is selected to classify the voxel set. In this process, two core clustering parameters are set, namely the scanning radius ε and the minimum number of points MinPts. Then, based on the relative position relationship of the voxels in the three-dimensional space and the abnormal intensity weight, density accessibility analysis is performed on all preliminary abnormal voxels, and voxels with adjacent distances and similar abnormal degrees are aggregated into the same spatial clustering unit to obtain a spatial abnormal clustering set. Each clustering unit represents an electromagnetic anomaly structure that is continuous, locally concentrated, and significantly deviates from the normal background in the geophysical sense, with a clear spatial distribution morphology and signal intensity characteristics. On this basis, in order to improve the geometric expression ability and engineering applicability of clustering results, the geometric features of each clustering result are calculated, including the boundary voxel envelope, spatial centroid position, maximum external boundary size, number of voxels, voxel density distribution, etc., and the boundary data of the target cluster is constructed accordingly. The boundary geometric data of each spatial abnormal cluster is matched and integrated with the signal characteristic parameters corresponding to its internal voxels (such as average energy, peak frequency, duration, change trend, etc.), and a complete spatial clustering abnormal data packet is generated through structured packaging. Each data packet contains the spatial boundary description, three-dimensional position center, voxel feature statistics, multi-layer threshold level, formation time and duration of the abnormal cluster, while keeping the data structure compact for efficient transmission and distributed storage.

[0016] Step S104: perform block-level differential compression on the spatial clustering abnormal data packets to obtain compressed data, and select the optimal transmission channel through the heterogeneous network self-organizing switching technology to transmit the compressed data to the cloud server.

[0017] Specifically, a structured block partitioning operation is performed on the spatial clustering abnormal data packet. The data packet contains multi-dimensional information such as the spatial coordinates, signal characteristics, boundary description, time evolution, and abnormal level of the abnormal area. Therefore, based on data attributes and data usage frequency, the entire data packet is divided into several functionally complete but independent logical data blocks according to attribute categories. In order to improve compression efficiency and reduce redundancy, these blocks are further optimized and rearranged to make the data storage structure more suitable for the compression model, and an optimized data block set is constructed. The spatiotemporal differential encoding operation is performed on the optimized blocks, that is, the difference between continuous blocks is encoded according to the change trend of information in the time dimension and the numerical gradient between spatial adjacent voxels, so that only the necessary incremental information is retained. This process can significantly reduce the amount of data and constitute the first data stream. The first data stream is input into the adaptive quantizer module to perform data importance grading operations. This module assigns different quantization accuracy levels to different fragments according to the criticality of each data fragment in the overall signal expression, combined with its abnormal level, spatial location centrality, and correlation with other regions and other multi-dimensional parameters. Data with higher importance will retain higher precision, while data with lower importance will use a coarser quantization level, thereby further compressing the data volume while ensuring the integrity of key information. This operation generates a second data stream and introduces a clear data hierarchical structure. A standardized data message format is constructed based on the second data stream, and complete message header information is added on this basis. The message header contains multiple key fields, including a high-precision timestamp indicating the acquisition time, a geographic location identifier output by the positioning module, a data priority flag generated by the aforementioned quantizer, and a checksum field for subsequent data integrity verification. These fields ensure the traceability and distribution scheduling capabilities of the data during transmission, and also improve the robustness and scheduling flexibility of the system in the face of heterogeneous network environments. In order to improve the reliable reception rate of data in a low-quality network environment, Reed-Solomon forward error correction coding is added to the entire data message, and the data error resistance is improved by adding redundant symbols, so that the receiving end can still accurately restore the original data in the face of partial data loss or bit errors. After completing the above operations, a compressed data unit with self-description capabilities, integrity assurance capabilities, and redundant fault tolerance capabilities is formed. Before entering the actual transmission, the system detects multiple heterogeneous wireless communication networks available in the current environment in real time and in parallel, including 4G / 5G cellular communication networks, LoRa long-distance IoT communication links, and long-distance wireless communication systems based on satellite links. For each communication channel, the real-time parameters of multiple dimensions such as signal strength, transmission delay, current available bandwidth, and frequent interruption probability are monitored separately, and these parameters are organized into a unified standard network performance vector. Based on these vectors, a network quality scoring matrix is ​​constructed. Each element in the matrix represents the performance of a certain network on a certain transmission indicator, which has time dynamics and spatial correlation.The scoring matrix is ​​comprehensively calculated using fuzzy logic inference algorithms, and the multidimensional indicators are converted into fuzzy membership functions. Through weighted synthesis and rule base evaluation, the comprehensive score of each available network is output, and the optimal transmission channel at the current moment is selected accordingly. The compressed data is packetized according to the optimal transmission channel, and the compressed data is divided into several target data packets according to the maximum transmission unit size of the channel. Each data packet contains a complete message header and segment identification information. These data packets are scheduled and sent through heterogeneous network self-organizing switching technology. This technology has the ability to dynamically switch between multiple links. It can automatically detect the channel status during transmission and automatically reroute according to changes in network conditions to ensure the reliability and continuity of data during transmission. After sending, the cloud server completes data reorganization, differential restoration, error correction decoding and integrity verification through the preset parsing protocol, and finally obtains spatial clustering anomaly data with complete structure and content.

[0018] In the embodiment of the present application, the six-dimensional complementary electrode array structure adopts an equilateral triangle grid layout to achieve high-sensitivity acquisition of omnidirectional electromagnetic pulse components, and cooperates with high-precision GPS clock synchronization and reference site settings to effectively eliminate the influence of external interference sources; spectrum adaptive tracking is combined with Kalman filtering technology to achieve accurate extraction of effective frequency band signals and characteristic frequency locking, effectively suppressing power frequency and its harmonic interference, and improving the signal-to-noise ratio of the signal effective frequency band; the first-order and second-order gradient difference matrix calculations are combined with space vector synthesis technology to improve the signal source positioning accuracy, and the accurate inversion of the signal source is achieved through principal component analysis and multilateral positioning algorithms, making the stress field distribution characteristics clearer. The signal enhancement method combining wavelet packet transform and singular value decomposition realizes high-resolution reconstruction of signal time-frequency characteristics, effectively removes random noise components while retaining key information, and enhances the weak signal detection capability; the anomaly recognition scheme combining multi-layer dynamic threshold analysis with density clustering algorithm improves the accuracy of abnormal area recognition and reduces the false alarm rate by considering multiple threshold conditions such as global statistics, regional relative and time series changes; block-level differential compression and heterogeneous network self-organizing transmission strategy achieve high compression rate while keeping key feature information intact, and combines forward error correction and network automatic switching technology to ensure data transmission reliability and timeliness in harsh field environments.

[0019] In a specific embodiment, the process of executing step S101 may specifically include the following steps: Determine the electrode array deployment plan for the detection target area according to an equilateral triangle arrangement pattern, and install a six-dimensional complementary electrode array including three pairs of orthogonal vector electromagnetic induction coils at each set receiving point; Install high-precision GPS clock modules on all six-dimensional complementary electrode arrays and configure low-power wireless transmission modules to build a mesh network architecture to obtain a real-time data transmission system; Set up E reference sites in the detection target area, and record the external electromagnetic interference signal of the area through each reference site to obtain noise calibration benchmark data; The detection target area is monitored through a real-time data transmission system to obtain raw electromagnetic pulse vector data; Based on the original electromagnetic pulse vector data and the noise calibration reference data, correlation calculation and calibration are performed to obtain natural electromagnetic pulse signal data; The natural electromagnetic pulse signal data is input into the spectrum adaptive tracking unit for Kalman filtering peak extraction and characteristic frequency locking to obtain frequency optimized signal data.

[0020] Specifically, the area to be measured is geometrically modeled and the detection points are deployed and optimized in an equilateral triangle arrangement mode according to the complexity of the geological structure, the distribution of the target body and the size of the space, so as to minimize the blind area and provide uniform coverage on the two-dimensional plane, thereby improving the spatial resolution capability of the electromagnetic field gradient change. Each set receiving point is used as a collection node, and a set of standardized six-dimensional complementary electrode array devices is installed. The array device consists of three pairs of mutually orthogonal vector electromagnetic induction coils, which synchronously sense the electromagnetic field components on the three orthogonal spatial axes of X, Y, and Z. At the same time, it is equipped with a high-sensitivity silver-silver chloride electrode and magnetic induction coil combination assembly to ensure the full-band response capability of low-frequency and medium-frequency electromagnetic pulse signals, thereby realizing high-resolution stereoscopic acquisition of electromagnetic vector fields in three-dimensional space. Time synchronization and data communication configuration are performed for each collection node. In order to ensure the timing consistency between multiple array nodes and reduce the error of subsequent signal alignment, all six-dimensional complementary electrode array units are integrated with high-precision GPS timing modules, which have nanosecond time synchronization capabilities. Through the satellite clock source, unified timing control is achieved for all nodes in the entire region, eliminating the time drift problem caused by terrain differences, distance or device asynchrony. At the same time, in order to ensure data intercommunication between long-distance and multi-nodes, low-power wireless transmission modules are configured in each node. These modules support automatic networking, self-repair and node discovery functions, and all nodes are linked into a distributed wireless transmission network with flexible topology and scalable structure through a self-organizing mesh network architecture. The mesh architecture can automatically adjust the routing path according to the network environment, and has anti-interference and self-healing capabilities. It can automatically find alternative paths to maintain the overall network connectivity when any node fails, thereby building a real-time data acquisition and transmission system with edge computing pre-functions. After completing the node layout and network initialization in the target area, several reference sites are set up outside the detection area to collect real-time background electromagnetic environment information outside the area. The number of reference sites is set to E, distributed around the boundaries of the area. The same electrode array configuration as the detection node is used to ensure that the data structure is consistent with the signal characteristics. It is used to record non-geological interference sources such as industrial interference, power system noise, communication base station signals and natural background fields. These interference signals are collected by the reference site to form an independent noise benchmark library. As all nodes are put into operation, the mesh network starts the data link and maintains a stable communication channel with the edge processing server or the central acquisition unit. Each six-dimensional electrode array unit collects natural electromagnetic pulse signals distributed in three-dimensional space in real time. The data obtained contains multi-dimensional features such as electric field strength, magnetic induction intensity and its gradient change in vector form, and is digitized with high precision through a 24-bit analog-to-digital converter. The sampling rate is set between 1kHz and 100kHz to cover natural signals in different frequency bands, and then sent to the main control unit after local processing to form an original electromagnetic pulse vector data set.At the same time, E reference sites upload their collected noise data synchronously through the network to form a background noise database that is aligned with the main observation area in time and space. In the signal processing process, the original electromagnetic vector data is dynamically compared and correlated with the reference noise data. Based on the joint correlation measurement of time domain and frequency domain, the electromagnetic pulse signal collected by each detection node is subjected to signal coherence calculation, energy similarity matching and frequency relative offset analysis with the reference site data to determine which frequency components or time period signals are mainly derived from external interference and which components have natural geological response characteristics. After correlation calculation and noise removal, the signal part without the influence of external background disturbance is extracted to construct natural electromagnetic pulse signal data with clear geological and physical significance, and its time series characteristics, frequency component distribution and spatial corresponding coordinates are retained in vector format. The natural electromagnetic pulse signal data is input into the spectrum adaptive tracking unit. The input signal is subjected to a fixed-length fast Fourier transform operation to extract its power spectrum density map and identify the local peak characteristics in the spectrum. In order to dynamically extract stable main frequency information from non-stationary signals, a frequency tracking algorithm based on Kalman filtering is used to establish a frequency drift state space model, and the frequency distribution is dynamically estimated and predicted and corrected. The Kalman filter continuously adjusts the error between the predicted value and the measured value, so that the main frequency tracking process has time continuity and mutation suppression capabilities, thereby locking a group of frequency points with high stability on the time axis as the main frequencies related to geological targets. The sampling parameters are readjusted according to the locked frequency to ensure that the subsequent acquisition meets the Nyquist condition and generates electromagnetic pulse signal data with final frequency optimization.

[0021] In a specific embodiment, the execution step of inputting the natural electromagnetic pulse signal data into the spectrum adaptive tracking unit to perform Kalman filter peak extraction and characteristic frequency locking to obtain the frequency optimized signal data may specifically include the following steps: Perform power frequency notch filtering and wavelet threshold denoising on the natural electromagnetic pulse signal data to obtain a de-noised electromagnetic pulse signal; The electromagnetic pulse signal after noise reduction is converted into the frequency domain by a spectrum adaptive tracking unit, the power spectrum density is calculated by discrete Fourier transform, and the characteristic frequency points are marked according to the local maximum principle to obtain the initial spectrum characteristic data; A state space model is constructed based on the initial spectrum feature data, and the state is estimated and predicted through the state space model to obtain the filtered spectrum trajectory data; Based on the filtered spectrum trajectory data, the drift rate of the spectrum peak in three adjacent time windows is calculated, and the frequency points whose stability coefficient is greater than the preset target value are marked as locked frequencies according to the drift rate to obtain a frequency locking parameter set; Based on the frequency locking parameter set, an adaptive window Hilbert transform is performed to obtain the instantaneous frequency and instantaneous amplitude, and a time-frequency joint analysis is performed through the instantaneous frequency and instantaneous amplitude to obtain an enhanced time-frequency feature matrix; The non-characteristic frequency components in the enhanced time-frequency characteristic matrix are attenuated by 50%, the signal integrity of the locked frequency is retained, and the frequency-optimized signal data is obtained through inverse transformation.

[0022] Specifically, an effective noise reduction operation is performed on the original signal to improve the spectral clarity of the signal and the accuracy of subsequent processing. In the actual natural electromagnetic environment, the signal is affected by a large amount of background interference, especially the low-frequency power system interference signal represented by the power frequency is the most common, usually in the form of 50Hz or 60Hz and its higher harmonics. In the preprocessing stage, the power frequency notch filtering technology is used to construct a band-stop filter with a center frequency that accurately matches the power frequency interference frequency, filter out the main frequency interference and its adjacent frequency band signals, thereby significantly reducing the impact of strong interference components on the subsequent analysis process. In order to suppress high-frequency oscillation, random noise and local mutation interference, a wavelet threshold denoising method is used on the basis of notch filtering. This method decomposes the signal into sub-signals of different frequency levels by performing multi-scale wavelet decomposition on the signal, and then performs shrinkage processing on the wavelet coefficients according to the preset soft threshold function, setting the low amplitude coefficients to zero or compression, thereby retaining the main components and eliminating noise disturbances. Finally, the signal is reconstructed to obtain a set of denoised electromagnetic pulse signals with high signal-to-noise ratio and structural stability. The denoised electromagnetic pulse signal is input into the spectrum adaptive tracking unit to enter the frequency domain analysis stage. The task of this stage is to identify and extract the main frequency components in the signal and their changing trends over time. Therefore, the discrete Fourier transform is used to convert the time domain signal to the frequency domain, calculate its power spectrum density, and draw a complete spectrum diagram with frequency as the horizontal axis and energy as the vertical axis. In the spectrum diagram, according to the principle of local maximum, the frequency points with significant energy prominence are identified. These points correspond to the main frequency components of the electromagnetic pulse signal or its high-order harmonic characteristics. By scanning the spectrum diagram point by point and comparing it with the neighborhood, a series of frequency peaks are marked, and their characteristic information such as frequency value, peak amplitude and peak width are extracted to form an initial spectrum feature data set. Based on the initial spectrum feature data, a state space model is constructed to realize dynamic estimation and trend prediction of frequency trajectory. The model consists of a state transfer equation and an observation equation, in which the state variable represents the time series evolution behavior of the frequency peak, and the observation variable is the frequency value and amplitude calculated in the current window. The Kalman filter method is used to fuse the observed value and the predicted value, and the error covariance matrix is ​​introduced to dynamically adjust the estimation accuracy to form a spectrum tracking mechanism with noise suppression capability and time continuity. Under the action of this model, the original spectrum peak is smoothed and a set of filtered spectrum trajectory data is output. This data not only shows continuous changes on the time axis, but also reflects the dominant trend of the real signal component on the frequency axis, and eliminates short-term disturbances or false peaks. Frequency stability analysis is performed based on the filtered spectrum trajectory data, and the change rate of each frequency peak in three consecutive time windows is analyzed in a sliding window manner, that is, the drift rate is calculated. By calculating the ratio of the maximum change amplitude of the same frequency component in three adjacent windows to the time span, the drift rate of each peak is obtained, and the definition of the stability coefficient is introduced, that is, the normalized value of 1 minus the drift rate is used to represent the frequency stability.If the stability coefficient of a frequency point in multiple time windows is higher than the target threshold preset by the system, the frequency is judged to have strong physical continuity and geological response stability, and it is marked as a locked frequency. The frequency value, average amplitude, main peak width and time stability range constitute a frequency locking parameter set. Based on the frequency locking parameter set, an adaptive window Hilbert transform is performed to extract fine-grained instantaneous features of the signal. The Hilbert transform constructs an analytical signal and directly extracts the instantaneous amplitude and instantaneous frequency from it to describe the non-stationary change trend of the signal on a microscopic time scale. The adaptive window mechanism automatically adjusts the window length according to the waveform period characteristics of the frequency band where the locked frequency is located, so that short-period high-frequency signals obtain sufficient frequency resolution, while long-period low-frequency signals obtain stronger time resolution. By traversing the entire signal sequence through a sliding window, the corresponding instantaneous frequency and amplitude are extracted in each window, and a two-dimensional matrix covering the entire time domain and the entire frequency band is constructed, namely the enhanced time-frequency feature matrix. The matrix takes time as the horizontal axis and frequency as the vertical axis. The matrix value represents the instantaneous energy or amplitude, which describes the dynamic process of the signal evolving over time in different frequency channels and highlights the energy aggregation characteristics near the locked frequency. In order to enhance the frequency selectivity and suppress the interference components of the non-feature frequency band, the frequency channel suppression operation is performed on the enhanced time-frequency feature matrix. All frequency channels in the matrix that are not included in the frequency locking parameter set are uniformly attenuated, with a specific attenuation ratio of 50%, that is, the amplitude of the channel is halved, while the frequency within a certain bandwidth near the locked frequency retains its full signal strength. The selective suppression operation makes the locked frequency show a relative advantage in the matrix, significantly improves the recognition of the main frequency component of the signal, and reduces the interference of redundant noise on the spectrum structure. After completing feature retention and non-feature suppression, the time-frequency matrix is ​​subjected to inverse Hilbert transform and inverse time-frequency transform operations, and the processed matrix is ​​restored to a time domain waveform to form the final frequency optimized electromagnetic pulse signal data.

[0023] In a specific embodiment, the process of executing step S102 may specifically include the following steps: The signal strength differences between adjacent measuring points in the three spatial directions of X, Y, and Z are calculated for the frequency-optimized signal data, a three-dimensional first-order gradient difference matrix is ​​constructed, and the three-dimensional first-order gradient difference matrix is ​​normalized by dividing it by the actual distance between the measuring points to obtain the first-order gradient field distribution data; Based on the first-order gradient field distribution data, the rate of change of the gradient field in each direction is calculated, and a three-dimensional second-order gradient difference matrix is ​​constructed by the central difference method. The three-dimensional second-order gradient difference matrix is ​​subjected to Gaussian smoothing filtering to eliminate singular points, and the second-order gradient field distribution data is obtained. Combining the first-order gradient field distribution data with the second-order gradient field distribution data into gradient field synthesis data, and performing principal component analysis on the gradient field synthesis data to obtain spatial vector field characteristic parameters; The signal source inversion equation is calculated based on the spatial vector field characteristic parameters to obtain the signal source position coordinate data, and the distribution of the signal strength in three-dimensional space is calculated based on the signal source position coordinate data to obtain the signal strength data.

[0024] Specifically, with the three spatial directions of X, Y, and Z as the main axes, the three-dimensional measuring point grid deployed in the target area is directional analyzed. Each direction contains multiple measuring point nodes with known physical coordinates and obtained frequency optimization signals. In order to extract the changing trend of signal strength in the spatial direction, the frequency optimization signal amplitude between adjacent measuring points in each direction is differentially operated to obtain the corresponding signal strength difference, which reflects the intensity attenuation, interference enhancement, or spatial gradient change characteristics of geological boundaries caused by the propagation of electromagnetic signals in space. These intensity differences are combined into a matrix structure in the three directions of X, Y, and Z according to the actual arrangement of the measuring points in three-dimensional space, and merged to construct a three-dimensional first-order gradient difference matrix. The three-dimensional first-order gradient difference matrix is ​​normalized by dividing the actual distance between the measuring points to eliminate the influence of different measuring point spacing on the gradient amplitude. According to the actual spatial distance of each pair of adjacent measuring points, the corresponding gradient difference is divided and normalized to obtain the gradient strength under unit distance, forming standardized first-order gradient field distribution data. Based on the first-order gradient field distribution data, the rate of change of the gradient field in each direction is calculated, that is, the second-order derivative structure is constructed. Detect the degree of change of the first-order gradient field itself in space, so as to reveal whether the signal intensity changes rapidly or tends to be flat in a certain direction, and then identify the gradient mutation caused by the geological interface, the edge of the abnormal body or the complex geological structure. The central difference method is adopted, that is, two points with equal distances upstream and downstream of a measuring point are selected, and the difference of their first-order gradient is calculated and divided by the distance between the two points as the estimated value of the second-order gradient at the point. By performing central difference calculation on all valid points in the three-dimensional grid, a three-dimensional second-order gradient difference matrix is ​​constructed. The second-order gradient matrix is ​​subjected to noise reduction. The Gaussian smoothing filter technology is introduced. By applying an isotropic Gaussian convolution kernel to the three-dimensional gradient data and performing a weighted average operation, the influence of singular points is effectively suppressed, the continuity and physical rationality of the spatial curvature field are improved, and the second-order gradient field distribution data after structural smoothing and abnormal point removal is obtained. The first-order gradient field distribution data is merged with the smoothed second-order gradient field data to construct the gradient field synthetic data, which forms a high-dimensional vector tensor field containing first-order information (linear change) and second-order information (curvature change). In order to extract the most dominant change direction and spatial structure characteristics, principal component analysis is performed on the gradient field synthetic data. By constructing its covariance matrix and solving the eigenvalues ​​and eigenvectors, the main directions of the most significant gradient changes in three-dimensional space and their weight distribution are identified, providing the optimal coordinate projection direction and gradient change mapping structure for constructing the physical model in the subsequent inversion calculation, forming a set of spatial vector field characteristic parameters with a compact structure but clear physical meaning.Based on the characteristic parameters of the spatial vector field, an electromagnetic signal source inversion mathematical model is constructed. The model is based on the Maxwell equations, introduces the main direction of the gradient field, the intensity change rate and the spatial distribution function as the model input variables, and establishes a quantitative relationship between the signal intensity and the spatial position through the least squares fitting method or the Bayesian estimation method, so as to perform an inversion calculation on the real position of the signal source in three-dimensional space. According to the estimation results output by the model, the spatial position coordinate data of the signal source is generated, specifically in the form of X, Y, and Z three-dimensional coordinate values, and the result is fed back to the overall model for iterative correction to ensure that its error converges within the preset range. When obtaining the signal source coordinates, based on the known signal source position, the spatial geometric relationship between the measuring points and the electromagnetic wave propagation attenuation law are combined to construct a signal intensity spatial diffusion model. Based on the propagation path of the electromagnetic wave in the medium, the impedance matching coefficient, the reflection and refraction boundary conditions and other parameters, the model calculates the energy loss of the signal in the process of diffusion from the source point to each measuring point, and generates a set of signal intensity distribution diagrams in the three-dimensional space grid. The diagram shows the expansion characteristics of the signal decreasing from the source point to the outside, and at the same time reveals the modulation effect of the geological structure on the propagation path, forming the signal intensity data.

[0025] Among them, the signal strength difference between adjacent measuring points in the three spatial directions of X, Y, and Z is calculated for the frequency-optimized signal data, a three-dimensional first-order gradient difference matrix is ​​constructed, and the difference is normalized by dividing it by the actual distance between the measuring points to obtain the spatial first-order gradient field distribution data, including: constructing a three-dimensional coordinate system for the frequency-optimized signal data according to the topological structure of the monitoring network, converting the geographical coordinates of each measuring point into standard rectangular coordinates, creating a four-dimensional data index table in combination with the measuring point ID and the acquisition timestamp, and obtaining spatial structured signal data; performing neighborhood measuring point identification on the spatial structured signal data, constructing a measuring point adjacency relationship graph based on the Delaunay triangulation algorithm, determining the nearest neighbor measuring point in the three directions of X, Y, and Z for each measuring point, calculating the Euclidean distance between each measuring point, and obtaining the measuring point neighborhood relationship matrix; Based on the measurement point neighborhood relationship matrix, the signal strength difference ΔS is calculated for the adjacent measurement point pairs (i, j) in the X direction. x (i, j) = S (i) - S (j), where S (i) and S (j) represent the signal strength values ​​at the measuring points i and j respectively. All the X-direction differences are combined into a matrix ΔS x , and obtain the signal strength gradient matrix in the X direction; the signal strength difference ΔS of adjacent measuring point pairs in the Y direction is calculated using the same method as that in the X direction. y (i, j) and the signal strength difference ΔS of the adjacent measuring points in the Z direction z (i, j), all Y-direction and Z-direction differences are combined into matrices ΔS y and ΔS z, and obtain the signal intensity gradient matrix in the Y and Z directions; the signal intensity gradient matrix in the X direction ΔS x , Y direction signal intensity gradient matrix ΔS y and the Z-direction signal intensity gradient matrix ΔS z The matrix is ​​integrated into a three-dimensional tensor, where i, j, and k represent the indexes of the measuring points in the X, Y, and Z directions, respectively, to obtain the initial three-dimensional first-order gradient difference matrix. Each element in the initial three-dimensional first-order gradient difference matrix is ​​divided by the actual physical distance d(i, j, k) between the corresponding measuring point pairs, and normalization is performed. Gaussian smoothing filtering is applied to eliminate noise interference to obtain the spatial first-order gradient field distribution data.

[0026] Among them, based on the first-order gradient field distribution data, the rate of change of the gradient field in each direction is calculated, and a three-dimensional second-order gradient difference matrix is ​​constructed by the central difference method. The Gaussian smoothing filter is applied to eliminate singular points to obtain stable second-order gradient field distribution data, including: The first-order gradient field distribution data is processed by boundary extension. Virtual grid points are added outside the boundary of the calculation domain by the mirror reflection method to maintain the continuity of the first-order gradient field at the boundary. Boundary condition constraints are set to obtain the expanded first-order gradient field calculation grid; the first-order gradient value in the X direction of the expanded first-order gradient field calculation grid is calculated by applying the five-point central difference format to obtain the second-order gradient component in the X direction; the second-order derivative of the first-order gradient in the Y direction and the second-order derivative of the first-order gradient in the Z direction are calculated by the same central difference method as that for calculating the second-order gradient in the X direction, and the cross-second-order derivatives are calculated at the same time to form a complete second-order gradient tensor and obtain the original three-dimensional second-order gradient. degree difference matrix; perform singular point detection on the original three-dimensional second-order gradient difference matrix, calculate the spatial aggregation degree of each outlier and the continuity index of the surrounding gradient field by setting threshold conditions, and obtain the singular point distribution map; based on the singular point distribution map, generate a three-dimensional Gaussian smoothing convolution kernel, adaptively adjust the threshold according to the intensity of the gradient field change, set differentiated smoothing intensity for different regions, and obtain the adaptive Gaussian filter parameters; apply the adaptive Gaussian filter parameters to the original three-dimensional second-order gradient difference matrix, perform three-dimensional convolution operation, and perform physical consistency check on the processed results to ensure that the gradient field satisfies the continuity equation and boundary conditions, and obtain stable second-order gradient field distribution data.

[0027] In a specific embodiment, the process of executing step S103 may specifically include the following steps: The signal source position coordinate data and the signal strength data are fused according to the spatial correspondence to obtain a time-space domain signal block sequence; Performing wavelet packet transformation on the time-space domain signal block sequence to obtain a frequency band weight coefficient table, and performing soft threshold shrinkage processing on the wavelet packet coefficients based on the frequency band weight coefficient table to obtain optimized wavelet packet coefficients; The optimized wavelet packet coefficients are reorganized into a signal matrix, and the signal matrix is ​​decomposed into a left singular matrix, a singular value diagonal matrix and a right singular matrix by a singular value decomposition algorithm; Sorting and reconstructing the singular values ​​in the singular value diagonal matrix to obtain denoised signal reconstruction data, and performing frequency band decomposition on the denoised signal reconstruction data to obtain an enhanced time-frequency distribution signal; Multi-layer dynamic threshold analysis is performed based on the enhanced time-frequency distribution signal to obtain spatially clustered abnormal data packets.

[0028] Specifically, the signal source position coordinate data is paired with the corresponding signal strength data. According to the uniqueness of each coordinate point in the three-dimensional space, it is used as the index association strength value to establish a spatial point array signal representation model. On this basis, the time dimension is introduced, and the signal change sequence of each spatial point at continuous moments is combined according to the measurement time step to form a time-space domain signal block sequence containing a ternary structure of spatial position, time sequence and signal strength value. Each signal block is the electromagnetic response waveform of a certain spatial point in a certain period of time. The sequence reflects the evolution of the electromagnetic field in time and space in the entire monitoring area. Wavelet packet transform is performed on the time-space domain signal block sequence, and a compactly supported wavelet basis function (such as Symlet or Daubechies) suitable for electromagnetic pulse analysis is used to perform multi-level, wavelet packet structure frequency domain expansion on each group of signal blocks, and the original time domain signal is completely decomposed at different scales and frequency resolutions. Wavelet packet transform not only decomposes the low-frequency component, but also recursively disassembles the high-frequency part, so it can more comprehensively capture the local energy change characteristics of the signal in multiple frequency sub-bands. By traversing all wavelet packet subbands, the sum of squares of wavelet coefficients in each frequency band, i.e., the energy value of the frequency band, is calculated, and a frequency band weight coefficient table is constructed to record the contribution of each frequency band in the entire signal. According to the frequency band weight coefficient table, a soft threshold shrinkage strategy is implemented to adjust the energy of all coefficients obtained after wavelet packet decomposition, and a higher shrinkage intensity is introduced in the frequency band with smaller weight to achieve weak information suppression and background noise removal; in the high-weight frequency band corresponding to the main frequency band, more original energy structures are retained, thereby optimizing the entire wavelet packet coefficient set, and obtaining a set of optimized wavelet packet coefficient sets that retain key features and filter out non-structural disturbances. The optimized wavelet packet coefficients are structured and reorganized to generate a two-dimensional signal matrix, in which each row represents the signal feature vector of a spatial point, and each column represents the response characteristics at a certain frequency scale, thereby constructing a signal matrix integrating the triple features of time, space and frequency. In order to extract the most representative structural components in the signal and eliminate residual redundant information, the singular value decomposition algorithm is applied to the signal matrix to decompose it into three parts: the left singular matrix, the singular value diagonal matrix and the right singular matrix. The left singular matrix represents the main direction of the spatial structure, the right singular matrix reflects the main components in the frequency domain, and each singular value in the singular value diagonal matrix represents the contribution of the corresponding main component to the overall signal energy. By sorting these singular values ​​in descending order and setting the energy retention threshold (such as retaining the first 90% of the cumulative energy), the secondary components with high redundancy and low signal-to-noise ratio are identified, and the corresponding singular values ​​are set to zero, thereby completing the singular value reconstruction of the signal. The updated singular value diagonal matrix is ​​used to perform matrix remultiplication with the left and right singular matrices to generate the denoised signal reconstruction data.Based on the denoised signal, the frequency band decomposition is performed again. The reconstructed signal is decomposed into multi-band scales using the wavelet packet structure consistent with the above, and the time-frequency distribution diagram is constructed based on the frequency center, bandwidth and local amplitude information of each sub-band. The diagram has the dual characteristics of time continuity and frequency resolution, and can clearly reflect the instantaneous energy change trajectory of each spatial point in different frequency bands, thereby generating an enhanced time-frequency distribution signal. Based on the enhanced time-frequency distribution signal, a multi-layer dynamic threshold analysis mechanism is introduced to realize the automatic identification and clustering of potential electromagnetic anomalies. The signal amplitude of each frequency channel, time window and spatial point is statistically analyzed, and statistical indicators such as mean μ, standard deviation σ, skewness and kurtosis are calculated, and a three-layer progressive dynamic threshold model is established based on this. The first-layer threshold is defined as μ+2σ, which is used to identify mild abnormal areas; the second-layer threshold is μ+3σ, which is used to judge significant abnormal behavior; and the third-layer threshold μ+4σ is used to locate strong mutation characteristic signals. The multi-layer discriminant model is applied point by point on the entire time-frequency spectrum, and all signal points that meet any threshold condition are marked with abnormal labels to form a high-dimensional abnormal point set. A density clustering algorithm, such as DBSCAN, is performed on all abnormal points. The abnormal points that are close to each other in space and have similar spectral characteristics are aggregated into independent abnormal area clusters using parameters such as the scanning radius ε and the minimum number of neighboring points MinPts. Each abnormal cluster represents an abnormal event with physical uniformity and geological response consistency, and its characteristics include spatial coordinate range, abnormal energy center, dominant frequency band, duration, etc. Each abnormal cluster area is described, encoded and packaged to finally form a structured spatial clustering abnormal data packet. Each data packet contains key features such as the boundary description of the abnormal area, abnormal level, corresponding frequency band, first appearance time, maximum duration, abnormal energy density, etc., and is stored in a unified data format.

[0029] In a specific embodiment, the execution step performs multi-layer dynamic threshold analysis according to the enhanced time-frequency distribution signal to obtain spatially clustered abnormal data packets, which may specifically include the following steps: Perform three-dimensional voxel division on the enhanced time-frequency distribution signal to obtain a signal feature voxel data matrix; Based on the signal feature voxel data matrix, a multi-layer threshold discrimination condition is constructed, and the signal feature voxel data matrix is ​​input into the multi-layer threshold discrimination condition for point-by-point evaluation to obtain a preliminary abnormal voxel set; Density clustering is performed on the preliminary abnormal voxel set to obtain a spatial abnormal cluster set, and based on the spatial abnormal cluster set, the geometric features of each cluster are calculated to obtain the target cluster boundary data; The target cluster boundary data and the corresponding abnormal voxel signal features are packaged and integrated to obtain the spatial cluster abnormal data packet.

[0030] Specifically, the enhanced time-frequency distribution signal is structurally reconstructed to have three-dimensional divisibility. The enhanced signal data is derived from the time-frequency matrix enhanced by frequency locking, Hilbert transform and wavelet analysis in the previous module. The matrix itself is a three-dimensional domain composed of time, frequency and spatial position as three axes, which can be naturally mapped to three-dimensional space and discretized. According to the preset time step Δt, frequency resolution Δf and spatial grid scale Δs, the entire enhanced signal space is divided into several three-dimensional voxel units with fixed sizes by equally spaced segments along the time axis, frequency axis and spatial axis. Each voxel unit corresponds to a specific time window, a frequency bandwidth and a geographical area, and stores multiple statistics such as the average signal strength, main frequency characteristics, spectral energy density or instantaneous change rate within the range, thereby constructing a signal feature voxel data matrix, which is essentially a multidimensional array with a position index, whose index is the voxel coordinate (x, y, z) and whose value is the corresponding feature vector or scalar value. Based on the signal feature voxel data matrix, a multi-layer threshold discrimination condition is constructed. A global statistical analysis is performed on the entire voxel data matrix to extract statistical parameters such as the mean μ, standard deviation σ, kurtosis K, and skewness S of all voxels. The triple anomaly judgment interval is determined by combining historical samples with regional electromagnetic background empirical data. The first layer of discrimination conditions is used to identify slight disturbances, and its threshold is set to ; The second layer of conditions is used to mark moderate anomalies, and the threshold is set to ; The third layer is dedicated to identifying high-intensity mutation areas, and its threshold is . A dynamic adjustment mechanism is introduced to make a small compensation correction to the above threshold according to the background drift of the electromagnetic environment in different time periods to adapt to the influence of external factors such as seasonal changes, weather disturbances or power frequency interference. The signal feature voxel data matrix is ​​used as input, and each layer of threshold conditions is evaluated point by point in turn. For each voxel, it is judged whether its characteristic value exceeds the judgment standard of a certain level. If any condition is met, it is marked as an abnormal voxel, and its abnormal level and the discrimination layer to which it belongs are recorded to form a preliminary abnormal voxel set. Density clustering is performed on the preliminary abnormal voxel set to identify abnormal clusters with concentrated distribution characteristics in three-dimensional space. A density-based spatial clustering algorithm (such as DBSCAN) is used. The algorithm does not need to preset the number of clusters, but automatically identifies the clustering structure based on the spatial distance and density distribution between voxels. The scanning radius ε is set, that is, any voxel can be regarded as a "neighboring point" within this radius, and the minimum number of points MinPts is set at the same time, indicating that when several voxels in a region meet the neighboring conditions at the same time, it can be determined that they constitute a type of clustered area. In the abnormal voxel set, DBSCAN clustering is performed according to the spatial position of the voxels, and the neighborhood of each voxel identified as a core point is expanded to form multiple spatial abnormal cluster sets. Each cluster set represents a spatially continuous and energy-concentrated electromagnetic anomaly area in a physical sense, which is caused by geological faults, structural boundaries, aquifer activities, ore body changes or external electromagnetic sources and has clear abnormalities. In order to make each cluster have identifiable boundary attributes and geometric morphological characteristics, the geometric parameters of the clustering results are extracted. The extraction process of geometric parameters includes calculating key indicators such as the spatial convex hull volume, maximum boundary length, voxel density distribution function, spatial centroid, boundary irregularity (i.e., surface area to volume ratio) of the cluster voxels, so as to describe the morphological structure, boundary range and diffusion capacity of each abnormal cluster in three-dimensional space. At the same time, according to the statistical average and peak values ​​of the abnormal level of each voxel in the cluster, the energy level evaluation index of the cluster is generated, and together with the boundary parameters, a complete cluster boundary description structure is formed. The boundary data is stored in the form of polygonal boundaries or boundary boxes, which is convenient for subsequent visualization and GIS platform access, and supports subsequent data transmission packaging. The target cluster boundary data generated above is associated and integrated with the corresponding abnormal voxel signal feature data inside it to complete the construction of the spatial cluster abnormal data packet. Each data packet is encapsulated in a structured manner, including core fields such as cluster number, spatial boundary information, abnormal level statistics, main frequency characteristics, number of voxels, time span, energy density, average amplitude change rate, cluster center coordinates and first appearance time. The data packet size is automatically adjusted according to the cluster scale and controlled within a few KB to more than ten KB. The system supports packaging the data packet in JSON, XML or a custom binary structure, and directly transmits it to the upper-level data fusion platform or cloud-based abnormality identification system through an interface, realizing the automatic identification, spatial clustering analysis and efficient aggregation of large-scale, multi-source heterogeneous electromagnetic monitoring data.

[0031] In a specific embodiment, the process of executing step S104 may specifically include the following steps: Block-dividing the spatially clustered abnormal data packets to obtain an optimized data block set, and performing spatiotemporal differential encoding on the optimized data block set to obtain a first data stream; Inputting the first data stream into an adaptive quantizer to perform data importance classification to obtain a second data stream; Based on the second data stream, construct a data message header including a timestamp, a geographic location identifier, a data priority flag, and an integrity check code, and add Reed-Solomon forward error correction coding to obtain compressed data; For compressed data, the signal strength, transmission delay and bandwidth resources of 4G / 5G cellular networks, LoRa long-distance IoT and satellite communication networks are detected in parallel to establish a network quality scoring matrix, and the fuzzy decision algorithm is used to select the optimal transmission channel; The compressed data is packetized according to the optimal transmission channel to obtain multiple target data packets, and then the multiple target data packets are sent to the cloud server through heterogeneous network self-organizing switching technology.

[0032] Specifically, the multi-dimensional structured information contained in the spatial clustering anomaly data packet is functionally partitioned. Considering the high-dimensional parameters such as the geographic spatial coordinates of the abnormal area, anomaly level, voxel statistics, abnormal signal time-frequency characteristics, duration, first appearance time, spatial range, and change trend contained in the data packet, the data packet content is divided into blocks according to the logical association and access frequency between the data. The division dimensions include time-related information blocks, spatial structure information blocks, signal feature information blocks, statistical analysis information blocks, and redundant check information blocks. By storing each type of data in a centralized manner and encapsulating it in a fixed format, an optimized data block set is generated. In order to reduce the redundancy of data in the time dimension and space dimension, a spatiotemporal differential encoding operation is performed on the above optimized data block set. A reference value is selected between or within the data block, and each target value is replaced by the difference between it and the reference value, forming a data expression format based on the change amount. The time difference is used to express the change trend of the abnormal data between consecutive moments, while the spatial difference is used to express the attribute difference of adjacent areas at the same moment. Through these two differential methods, the storage bit number of repeated or gradually changing information is effectively reduced. The output result of spatiotemporal differential coding is the first data stream, which consists of a data sequence composed of difference values, and has a high data compression rate and good structural sparsity in most scenarios. The first data stream is taken as input, and the adaptive quantizer module is introduced to evaluate the importance and compress the precision of the data units therein. The quantizer assigns an importance level to each data segment based on multiple indicators such as the abnormality level, signal strength, spatial centrality, and time-frequency structure rarity of the data. High-importance data will be given a higher quantization precision to maximize the retention of the original information, while low-importance data will be mapped to a low-bit precision interval to save bit width overhead. The process outputs the second data stream, whose content structure is basically the same as the first data stream, but the values ​​have been compressed according to the quantization level and have an additional importance flag bit, forming a data content structure optimized for transmission. After the second data stream is constructed, a data message header is constructed for it to support cross-network transmission, message integrity verification, and priority scheduling. The message header contains multiple key fields: the first is the timestamp field, which is used to mark the time when the data was generated, so that the receiving end can sort and analyze the timeliness of the data; the second is the geographic location identification field, which encodes the spatial region or grid number to which the data belongs, to support spatial distribution modeling and regional analysis; the third is the data priority flag field, which is directly derived from the classification result of the aforementioned adaptive quantizer and is used to guide the resource allocation of the network scheduling strategy; the fourth is the integrity check code field, which is generated by the system using cyclic redundancy check or hash function calculation to ensure that the data integrity of the message can be verified by the receiving end during the transmission process. In order to enhance the system's error-resistant capability in a weak network environment, the Reed-Solomon forward error correction coding algorithm is used to embed redundant information in the entire data segment, so that the receiving end can automatically recover lost or damaged data fragments within a certain range.Through this step, a data compression result with structural compression, importance level, location identification, error correction, fault tolerance and integrity protection is obtained. Before data transmission, the optimal transmission channel is dynamically selected to adapt to the variability and heterogeneity of the current communication environment. The system detects three different types of link resources in parallel, including 4G / 5G cellular networks, LoRa long-distance low-power IoT communication networks and satellite communication networks, and obtains key indicators such as signal strength, bandwidth resources, transmission delay, link stability and data congestion of each channel in real time, and constructs a network quality scoring matrix. The scoring matrix is ​​a three-dimensional table, each row corresponds to an available transmission channel, each column corresponds to a network performance dimension, and each cell stores the performance indicator of the network at the current moment. The system uses a fuzzy decision algorithm to comprehensively evaluate the scoring matrix, convert each network performance dimension into a membership function, and select the channel with the largest comprehensive membership as the current optimal transmission path through weighted average and rule reasoning. After determining the transmission path, the compressed data is packetized according to the maximum transmission unit standard of the selected channel. Each data packet contains a complete message header, data body segmentation and packet identification information to ensure that the cloud server reorganizes the received data and restores the original structure. These target data packets are sent to the cloud server through the heterogeneous network self-organizing switching technology, which supports dynamic switching between multiple networks. For example, when the packet loss rate of the LoRa network increases or the 5G network is congested, the data transmission link is automatically migrated without interrupting the transmission process, ensuring the continuity and integrity of the data. In addition, an intelligent caching mechanism is introduced to temporarily cache data packets when the network is interrupted and automatically retransmit them after the network is restored; when high-priority data arrives, preemptive scheduling is supported, and data with higher importance is inserted into the front of the current transmission queue for transmission, ensuring that the system has high availability and data hierarchical response capabilities. All data packets are uploaded to the cloud server safely and efficiently, and after a series of processing such as decoding, dequantization, inverse differential restoration, singular value reconstruction and voxel fusion, they are restored to complete spatial anomaly data.

[0033] Among them, the optimal transmission channel is selected through the self-organizing switching technology of heterogeneous networks to transmit the compressed data to the cloud server, including: grading the transmission priority of the compressed data, dividing the data into three priorities of emergency, important and regular based on the stress intensity, change rate and spatial range of the abnormal clustering data, assigning a transmission resource weight coefficient to each priority, and obtaining a data transmission priority list; deploying the solar power supply system energy management module at the monitoring node, collecting the current energy data through the light intensity sensor and the battery status monitor, and establishing an energy availability prediction model in combination with the historical energy consumption records and weather forecast data to obtain the energy distribution prediction data for the next 24 hours; integrating the energy distribution prediction data with the data transmission priority list, constructing a multi-level transmission management architecture including environmental perception, energy prediction, task scheduling and execution control, setting the initial transmission strategy through heuristic rules, and obtaining a transmission parameter configuration set; applying the DSAC-CAL (Distributed Soft Actor Critic Conservative Augmented Lagrangian) algorithm to the transmission parameter configuration set, integrating the energy constraint conditions and the data transmission goal into the deep reinforcement learning framework, and training the transmission strategy network by setting the state space, action space and reward function to obtain the optimal transmission strategy model under energy constraints; based on the optimal transmission strategy The proposed method dynamically adjusts the wireless transmission module parameters, including transmission power, data transmission interval, compression rate adjustment and processor frequency, sets four-level energy management mode according to the battery power status, and automatically starts the extreme power saving mode when the power is lower than 20%, so as to obtain the adaptive transmission control parameters. According to the adaptive transmission control parameters, the real-time performance evaluation is performed on various network interfaces, and the transmission efficiency index of the current 4G / 5G, LoRa and satellite communication networks is calculated. The current optimal transmission channel combination is selected by a multi-objective optimization algorithm in combination with the transmission data volume and real-time network quality, and the multi-channel collaborative transmission scheme is obtained. The compressed data is intelligently packetized according to the multi-channel collaborative transmission scheme, and a redundant transmission strategy is implemented for the emergency priority data. The same data packet is sent in parallel through multiple channels. The main and backup channel mechanism is adopted for the important priority data. The most economical channel is selected for the conventional data according to the bandwidth cost, so as to obtain the differentiated transmission data packet. The transmission status monitoring is performed on the differentiated transmission data packet, and the sending status, confirmation reception and transmission delay of each data packet are recorded. When the transmission abnormality is detected, the automatic retransmission mechanism is triggered, and the transmission quality feedback information is used as the reward signal of the reinforcement learning system for online learning, so as to continuously optimize the transmission strategy and realize the high-reliability data transmission under the condition of energy limitation.

[0034] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0035] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling an electromagnetic pulse data transmission device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0036] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An electromagnetic pulse data transmission method, characterized in that: include: The natural electromagnetic pulse signal data of the detection target area is collected through a six-dimensional complementary electrode array structure, and Kalman filter peak extraction and characteristic frequency locking are performed to obtain frequency-optimized signal data; Performing first-order and second-order gradient difference matrix calculation and space vector synthesis on the frequency-optimized signal data to obtain signal source position coordinate data and signal strength data; Performing spectral domain reconstruction and multi-layer dynamic threshold analysis on the signal source position coordinate data and the signal strength data to obtain spatial clustering abnormal data packets; Block-level differential compression is performed on the spatial clustering abnormal data packets to obtain compressed data, and the optimal transmission channel is selected through the heterogeneous network self-organizing switching technology to transmit the compressed data to the cloud server.

2. The electromagnetic pulse data transmission method according to claim 1, characterized in that: The method collects the natural electromagnetic pulse signal data of the detection target area through the six-dimensional complementary electrode array structure, performs Kalman filter peak extraction and characteristic frequency locking, and obtains frequency-optimized signal data, including: Determine the electrode array deployment plan for the detection target area according to an equilateral triangle arrangement pattern, and install a six-dimensional complementary electrode array including three pairs of orthogonal vector electromagnetic induction coils at each set receiving point; Install high-precision GPS clock modules on all six-dimensional complementary electrode arrays and configure low-power wireless transmission modules to build a mesh network architecture to obtain a real-time data transmission system; Setting E reference sites in the detection target area, and recording the external electromagnetic interference signal of the area through each reference site to obtain noise calibration benchmark data; The detection target area is monitored by the real-time data transmission system to obtain original electromagnetic pulse vector data; Perform correlation calculation and calibration based on the original electromagnetic pulse vector data and the noise calibration reference data to obtain natural electromagnetic pulse signal data; The natural electromagnetic pulse signal data is input into a spectrum adaptive tracking unit to perform Kalman filtering peak extraction and characteristic frequency locking to obtain frequency-optimized signal data.

3. The electromagnetic pulse data transmission method according to claim 2, characterized in that: The step of inputting the natural electromagnetic pulse signal data into a spectrum adaptive tracking unit for Kalman filtering peak extraction and characteristic frequency locking to obtain frequency-optimized signal data includes: Performing power frequency notch filtering and wavelet threshold denoising processing on the natural electromagnetic pulse signal data to obtain a denoised electromagnetic pulse signal; The electromagnetic pulse signal after noise reduction is converted into the frequency domain by a spectrum adaptive tracking unit, the power spectrum density is calculated by discrete Fourier transform, and the characteristic frequency points are marked according to the local maximum principle to obtain initial spectrum characteristic data; Constructing a state space model according to the initial spectrum feature data, and performing state estimation and prediction through the state space model to obtain filtered spectrum trajectory data; Based on the filtered spectrum trajectory data, the drift rate of the spectrum peaks in three adjacent time windows is calculated, and the frequency points whose stability coefficients are greater than a preset target value are marked as locked frequencies according to the drift rate to obtain a frequency locking parameter set; Performing adaptive window Hilbert transform based on the frequency locking parameter set to obtain instantaneous frequency and instantaneous amplitude, and performing time-frequency joint analysis through the instantaneous frequency and the instantaneous amplitude to obtain an enhanced time-frequency feature matrix; The non-characteristic frequency components in the enhanced time-frequency characteristic matrix are attenuated by 50%, the signal integrity of the locked frequency is retained, and frequency-optimized signal data is obtained through inverse transformation.

4. The electromagnetic pulse data transmission method according to claim 1, characterized in that: The performing first-order and second-order gradient difference matrix calculation and space vector synthesis on the frequency-optimized signal data to obtain signal source position coordinate data and signal strength data includes: Calculating the signal strength differences between adjacent measuring points in the three spatial directions of X, Y and Z for the frequency-optimized signal data, constructing a three-dimensional first-order gradient difference matrix, and dividing the three-dimensional first-order gradient difference matrix by the actual distance between the measuring points for normalization processing to obtain first-order gradient field distribution data; Based on the first-order gradient field distribution data, the rate of change of the gradient field in each direction is calculated, a three-dimensional second-order gradient difference matrix is ​​constructed by a central difference method, and a Gaussian smoothing filter is applied to the three-dimensional second-order gradient difference matrix to eliminate singular points, thereby obtaining second-order gradient field distribution data; Combining the first-order gradient field distribution data with the second-order gradient field distribution data into gradient field synthesis data, and performing principal component analysis on the gradient field synthesis data to obtain spatial vector field characteristic parameters; The signal source inversion equation is calculated based on the spatial vector field characteristic parameters to obtain signal source position coordinate data, and the distribution of signal strength in three-dimensional space is calculated according to the signal source position coordinate data to obtain signal strength data.

5. The electromagnetic pulse data transmission method according to claim 1, characterized in that: The performing spectral domain reconstruction and multi-layer dynamic threshold analysis on the signal source position coordinate data and the signal strength data to obtain a spatial clustering abnormal data packet includes: The signal source position coordinate data and the signal strength data are merged according to a spatial correspondence to obtain a time-space domain signal block sequence; Performing wavelet packet transformation on the time-space domain signal block sequence to obtain a frequency band weight coefficient table, and performing soft threshold shrinkage processing on the wavelet packet coefficients based on the frequency band weight coefficient table to obtain optimized wavelet packet coefficients; Reorganizing the optimized wavelet packet coefficients into a signal matrix, and decomposing the signal matrix into a left singular matrix, a singular value diagonal matrix and a right singular matrix by a singular value decomposition algorithm; Sorting and reconstructing the singular values ​​in the singular value diagonal matrix to obtain denoised signal reconstruction data, and performing frequency band decomposition on the denoised signal reconstruction data to obtain an enhanced time-frequency distribution signal; Multi-layer dynamic threshold analysis is performed according to the enhanced time-frequency distribution signal to obtain spatially clustered abnormal data packets.

6. The electromagnetic pulse data transmission method according to claim 5, characterized in that: The performing of multi-layer dynamic threshold analysis according to the enhanced time-frequency distribution signal to obtain spatially clustered abnormal data packets includes: Performing three-dimensional voxel division on the enhanced time-frequency distribution signal to obtain a signal feature voxel data matrix; Based on the signal feature voxel data matrix, a multi-layer threshold discrimination condition is constructed, and the signal feature voxel data matrix is ​​input into the multi-layer threshold discrimination condition for point-by-point evaluation to obtain a preliminary abnormal voxel set; Density clustering is performed on the preliminary abnormal voxel set to obtain a spatial abnormal cluster set, and based on the spatial abnormal cluster set, geometric features of each cluster are calculated to obtain target cluster boundary data; The target cluster boundary data and the corresponding abnormal voxel signal features are packaged and integrated to obtain a spatial cluster abnormal data packet.

7. The electromagnetic pulse data transmission method according to claim 1, characterized in that: The block-level differential compression is performed on the spatial clustering abnormal data packets to obtain compressed data, and the optimal transmission channel is selected through the heterogeneous network self-organizing switching technology to transmit the compressed data to the cloud server, including: Dividing the spatially clustered abnormal data packets into blocks to obtain an optimized data block set, and performing spatiotemporal differential coding on the optimized data block set to obtain a first data stream; Inputting the first data stream into an adaptive quantizer to perform data importance classification to obtain a second data stream; Based on the second data stream, construct a data message header including a timestamp, a geographic location identifier, a data priority flag, and an integrity check code, and add Reed-Solomon forward error correction coding to obtain compressed data; For the compressed data, a network quality scoring matrix is ​​established by parallelly detecting the signal strength, transmission delay and bandwidth resources of 4G / 5G cellular networks, LoRa long-distance IoT and satellite communication networks, and a fuzzy decision algorithm is used to select the optimal transmission channel; The compressed data is packetized according to the optimal transmission channel to obtain multiple target data packets, and the multiple target data packets are sent to the cloud server through heterogeneous network self-organizing switching technology.

Images