Multi-modal intelligent situation awareness underground cable detection and positioning system and method

Abstract

This invention discloses a multimodal intelligent situational awareness system and method for underground cable detection and location, relating to the field of underground pipeline detection technology. The system includes synchronously acquiring raw magnetic field data using a three-dimensional magnetic field sensor array deployed on the ground, and acquiring surface vibration signals using a distributed acoustic wave sensor network buried in shallow soil. After filtering noise from the magnetic field data, spatial three-dimensional interpolation and gradient calculation are performed to construct a magnetic field intensity isosurface cloud map and a magnetic field gradient tensor field. Characteristic frequencies are extracted from the vibration signals to generate a cable characteristic vibration spectrum map. Finally, by fusing the cable characteristic vibration spectrum map and the magnetic field gradient tensor field, a feature space matching algorithm is used to identify the spatial direction of potential cable current channels, generating preliminary candidate cable paths. This invention integrates electromagnetic and acoustic vibration dual physical field characteristics, achieving accurate and robust detection and location of underground cable paths.

Description

Technical Field

[0001] This invention belongs to the field of underground pipeline detection technology, specifically a multimodal intelligent situational awareness underground cable detection and positioning system and method. Background Technology

[0002] Underground cable path detection technologies are primarily based on magnetic induction or acoustic vibration principles. Magnetic induction methods track cables by sensing the magnetic field generated by the cable current. However, in urban environments, significant power frequency electromagnetic interference and geomagnetic background noise reduce the signal-to-noise ratio of the magnetic field signal. For deeply buried cables or low-current applications, the magnetic field characteristics are weak, making it difficult to extract effective information from the noise and failing to characterize the gradient changes of the magnetic field in three-dimensional space. Acoustic vibration methods rely on detecting vibration signals caused by cable operation. However, surface environment vibration and noise are complex, traditional ground-deployed sensors have poor coupling with the soil medium, resulting in significant energy loss at the interface and insensitivity to capturing micro-vibration signals transmitted underground, making it difficult to separate the cable's power frequency-related characteristic frequencies.

[0003] Existing technologies typically process magnetic fields and vibration signals independently or simply superimpose them, lacking a deep fusion mechanism at the feature level. This results in insufficient reliability of the detection results under interference environments, as the ambiguity of magnetic field information cannot be corrected by vibration information, and cross-validation is impossible. Limitations in vibration signal acquisition methods lead to low signal-to-noise ratios in the raw data and insufficient feature extraction, hindering the application effectiveness of multimodal information fusion. This invention aims to solve the problem of achieving high-precision cable path detection in high-noise and complex electromagnetic environments. It involves constructing spatial distribution features from disturbed magnetic field data and extracting cable vibration frequency features from the noise background, and improving the accuracy and robustness of the positioning through fusion. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, this invention proposes a multimodal intelligent situational awareness method for underground cable detection and location, including: The original magnetic field strength data is synchronously collected by a three-dimensional magnetic field sensing array deployed on the ground of the detection area. The three-dimensional magnetic field sensing array consists of multiple three-axis fluxgate sensors distributed in a grid, which are used to obtain the magnetic field vector information of the three-dimensional space of the detection area. The distributed acoustic wave sensor network deployed on the ground in the detection area is used to collect surface vibration acoustic signals. The distributed acoustic wave sensor network consists of an array of fiber optic microphones buried in the shallow soil layer, which is used to capture micro-vibration signals of the strata caused by underground cable flow or external interference. The synchronously acquired raw magnetic field strength data is input into an adaptive noise suppression processor to filter out power frequency interference and geomagnetic background noise, generating a pure time-varying magnetic field strength data sequence. The collected surface vibration acoustic signal is input into a feature frequency extractor to separate the vibration components related to the power frequency harmonics of the cable, and generate a characteristic vibration spectrum of the cable. Spatial three-dimensional interpolation and gradient calculation are performed on the purified time-varying magnetic field intensity data sequence to construct the magnetic field intensity isosurface cloud map and magnetic field gradient tensor field of the underground space of the detection area. By integrating the characteristic vibration spectrum of the cable with the magnetic field gradient tensor field, the spatial orientation of potential cable current channels is identified through a feature space matching algorithm, generating preliminary candidate cable paths.

[0005] Further, the step of inputting the synchronously acquired raw magnetic field strength data into an adaptive noise suppression processor to filter out power frequency interference and geomagnetic background noise includes: A fast Fourier transform is performed on the raw magnetic field strength data from each sensor of the three-dimensional magnetic field sensing array to obtain the magnetic field strength spectrum; In the magnetic field strength spectrum, identify the spectral peaks centered at 50 Hz and its integer multiples to construct a spectral template for power frequency interference; An adaptive spectral subtraction algorithm is used to subtract the component corresponding to the spectral template of the power frequency interference from the spectrum of the original magnetic field strength data; Empirical mode decomposition is performed on the remaining spectral components after spectral subtraction to separate the low-frequency trend term reflecting the diurnal variation of geomagnetism. The low-frequency trend term is then removed as geomagnetic background noise. The signal after removing the geomagnetic background noise is subjected to inverse Fourier transform to reconstruct the pure time-varying magnetic field intensity data sequence in the time domain.

[0006] Further, the step of inputting the collected surface vibration acoustic signal to a characteristic frequency extractor to separate the vibration component related to the power frequency harmonic characteristics of the cable includes: The surface vibration acoustic signal of each node in the distributed acoustic wave sensing network is bandpass filtered to retain the frequency band from 100 Hz to 2 kHz. Perform a short-time Fourier transform on the filtered signal to generate a time-frequency spectrum. In the aforementioned time-spectrum diagram, spectral lines whose energy is continuous in time and whose frequency is an integer multiple of fifty Hz are searched. Extract all spectral lines that meet the frequency conditions found in the search, and calculate the average energy of each spectral line within the analysis time window; The frequency components corresponding to spectral lines with average energy exceeding a preset threshold are reconstructed and combined to form the characteristic vibration spectrum of the cable.

[0007] Further, the step of performing spatial three-dimensional interpolation and gradient calculation on the purified time-varying magnetic field intensity data sequence to construct the magnetic field intensity isosurface cloud map and magnetic field gradient tensor field of the underground space of the detection area includes: Establish a three-dimensional regular grid spatial model covering the detection area and the preset depth range; The pure time-varying magnetic field intensity data sequence of each of the three-axis fluxgate sensors at the sampling time is used as the magnetic field vector observation value at the geographic coordinates of the three-axis fluxgate sensor; The Kriging space interpolation algorithm is used to interpolate the discrete magnetic field vector observations onto each grid node of the three-dimensional regular grid space model to obtain magnetic field vector data that are continuously distributed throughout the field. Based on the interpolated magnetic field vector data of the continuous distribution across the entire field, the spatial rate of change of the magnetic field vector at each grid node in the three orthogonal directions is calculated, i.e., the magnetic field gradient components. All magnetic field gradient components of each grid node are combined to form the magnetic field gradient tensor field describing the rate of change of the magnetic field in space, and the magnetic field intensity isosurface contour map of the magnetic field intensity scalar is plotted.

[0008] Furthermore, the step of fusing the characteristic vibration spectrum of the cable with the magnetic field gradient tensor field, and identifying the spatial orientation of potential cable current channels through a feature space matching algorithm, includes: Extract a preset number of characteristic frequency points with the strongest vibration energy from the characteristic vibration spectrum of the cable, and record the amplitude distribution of the signal corresponding to each extracted characteristic frequency point on each node of the distributed acoustic wave sensor network. From the magnetic field gradient tensor field, extract the three-dimensional spatial coordinates of the magnetic field gradient amplitude points where local maxima occur to form a set of magnetic field anomaly points; Construct a joint feature space based on geographic coordinates and incorporating vibration amplitude and magnetic field gradient amplitude as feature vectors; The set of magnetic field anomalies is mapped to the joint feature space, and the vibration amplitude distribution information of the distributed acoustic wave sensing network nodes is mapped to the same joint feature space. Within the joint feature space, a density clustering algorithm is used to find spatial clustering regions that simultaneously possess high vibration amplitude and high magnetic field gradient amplitude features. The three-dimensional spatial curve connecting the center points of the spatial clustering regions is identified as the preliminary candidate cable path.

[0009] Furthermore, it also includes the steps of performing current phase inversion and path authenticity verification on the preliminary candidate cable paths: On the initial candidate cable route, a series of verification points with intervals are selected; From the purified time-varying magnetic field strength data sequence, extract the waveform of the magnetic field strength changing with time corresponding to each verification point location; Cross-correlation analysis was performed on the magnetic field intensity waveforms at all verification points to calculate the time delay between waveforms, and the apparent propagation speed of electromagnetic waves along the preliminary candidate cable path was estimated based on the known sensor spacing. Based on the Biot-Savart law in electromagnetic field theory, a positive magnetic field calculation model is established with the cable position, current amplitude and phase as unknowns. Using the measured magnetic field strength waveform at the verification point as a constraint, an iterative optimization algorithm is used to adjust the unknowns in the positive magnetic field calculation model until the magnetic field waveform calculated by the model matches the measured waveform within a preset error range. The optimized cable current phase information is checked for consistency along the preliminary cable path candidate line. If the phase change is continuous and conforms to the law of a single current-carrying conductor, the cable path candidate line is determined to be the real cable path.

[0010] Furthermore, the step of adjusting the unknowns in the positive magnetic field calculation model using an iterative optimization algorithm, constrained by the measured magnetic field strength waveform at the verification point, includes: Initialize the positive magnetic field calculation model and assign initial current amplitude, initial phase, and initial three-dimensional spatial position fine-tuning amount to the preliminary cable path candidate lines; Using the initial parameters, the theoretical waveform of the synthetic magnetic field strength at each verification point is calculated using the positive magnetic field calculation model; Calculate the overall root mean square error between the theoretical waveform of the synthesized magnetic field strength and the corresponding measured magnetic field strength waveform at all verification points; A gradient descent optimization strategy is adopted to iteratively adjust the current amplitude, initial phase, and three-dimensional spatial position fine-tuning amount based on the overall root mean square error. After each iteration, the overall root mean square error is recalculated until the decrease in the overall root mean square error is less than the preset convergence threshold, or the number of iterations reaches the preset upper limit of the number of iterations. The model parameters at this point are then used as the optimization result.

[0011] Furthermore, it also includes a dynamic detection sensitivity adjustment step based on adaptive environmental parameters: Real-time monitoring of the background noise level in the detection area, including the intensity of environmental electromagnetic noise and the intensity of ground vibration noise; The filtering bandwidth and threshold of the adaptive noise suppression processor are dynamically adjusted based on the monitored environmental electromagnetic noise intensity. Based on the monitored surface vibration noise intensity, the window function length and spectral energy threshold of the feature frequency extractor for short-time Fourier transform are dynamically adjusted. The adjusted filtering parameters are then applied to the subsequent processing flow of the original magnetic field strength data and the surface vibration acoustic signal. Based on the changes in the background noise level, the required calculation accuracy of the magnetic field gradient tensor field and the clustering radius parameter in the feature space matching algorithm are updated synchronously.

[0012] Furthermore, it also includes a step of performing multimodal joint estimation of the burial depth and fault location of the identified cable path: For a verified real cable path, extract the maximum value of the vertical magnetic gradient at the ground surface directly above the path from the magnetic field gradient tensor field. By combining the cable current amplitude obtained through current phase inversion, the theoretical burial depth of the cable is calculated using an infinitely long straight conductor magnetic field model. Simultaneously, the vibration signal energy attenuation characteristics distributed along the actual cable path are analyzed in the characteristic vibration spectrum diagram of the cable. A coupled relationship model between vibration energy attenuation, propagation distance, and cable burial depth is established, and the equivalent burial depth of the cable is independently estimated by fitting the vibration energy attenuation curve. By comparing the theoretical burial depth obtained from the back calculation of the vertical gradient of the magnetic field with the equivalent burial depth estimated by the vibration energy attenuation curve, if the absolute value of the difference between the theoretical burial depth and the equivalent burial depth is less than the preset burial depth error threshold, the arithmetic mean of the theoretical burial depth and the equivalent burial depth is calculated as the final burial depth estimate; if the absolute value of the difference between the theoretical burial depth and the equivalent burial depth is greater than or equal to the preset burial depth error threshold, the corresponding cable section is marked as a potential insulation damage section or an external interference section, and is identified as a suspected fault point.

[0013] Furthermore, the present invention also includes a multimodal intelligent situational awareness underground cable detection and positioning system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the multimodal intelligent situational awareness underground cable detection and positioning method described above.

[0014] Compared with the prior art, the beneficial effects of the present invention are: The magnetic field data is interpolated in three dimensions and gradients are calculated to construct a magnetic field gradient tensor field for the detection area. This tensor field describes the rate of change and directional characteristics of the magnetic field vector at various points in space. Components related to the cable's power frequency harmonics are separated from the vibration signal to generate a characteristic vibration spectrum of the cable. A feature space matching algorithm is used to fuse and match these two heterogeneous features: the magnetic field gradient tensor and the vibration spectrum. The fusion process establishes a deep correlation between electromagnetic and acoustic / vibration physical quantities in characterizing the cable's spatial properties. This method overcomes the incompleteness of single physical field information under strong interference; when the magnetic field signal is blurred by environmental interference, the vibration spectrum provides an independent feature criterion. The consistent matching of the two features in spatial distribution enhances the robustness and accuracy of identifying the true cable path from complex backgrounds and reduces misjudgments caused by the failure of a single field.

[0015] Fiber optic microphone arrays are embedded in the shallow soil layer of the detection area in a distributed network. This embedded deployment allows for direct and tight mechanical coupling between the sensor and the soil medium, reducing energy loss and signal distortion during vibration signal transmission at the medium interface. This method can more efficiently capture micro-vibrations in the strata caused by underground cable current flow or external disturbances and directly transmitted through the soil and rock medium. Directly acquiring signals from within the vibration propagation path effectively suppresses the intrusion of surface environmental vibration noise, improving the signal-to-noise ratio of the original vibration signal. The higher signal-to-noise ratio allows subsequent signal processing to more clearly and completely separate the vibration frequency components that are strictly related to the cable current characteristics, providing a reliable data foundation for generating highly discriminative cable characteristic vibration spectrum maps. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of the multimodal intelligent situational awareness method for detecting and locating underground cables as described in this invention. Figure 2 A flowchart for noise suppression and reconstruction of the original magnetic field strength data; Figure 3 A flowchart for constructing the magnetic field intensity isosurface contour map and gradient tensor field; Figure 4 This is a diagram showing the influence of burial depth on magnetic field gradient and vibration energy. Figure 5 This is a diagram for estimating the burial depth of multimode cables. Detailed Implementation

[0017] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See Figure 1 A three-dimensional magnetic field sensing array, consisting of multiple grid-distributed triaxial fluxgate sensors, is deployed on the ground in the detection area to synchronously acquire raw magnetic field strength data in the three-dimensional space of the detection area. This raw magnetic field strength data is vector information. A distributed acoustic wave sensing network, composed of an array of fiber optic microphones buried in the shallow soil layer, is also deployed on the ground in the detection area to acquire acoustic signals of surface vibrations caused by underground cable current flow or external interference, i.e., ground micro-vibration signals. The synchronously acquired raw magnetic field strength data is input to an adaptive noise suppression processor to filter out power frequency interference and geomagnetic background noise, generating a clean time-varying magnetic field strength data sequence. The acquired surface vibration acoustic signals are input to a characteristic frequency extractor to separate the vibration components related to the power frequency harmonics of the cable, generating a characteristic vibration spectrum of the cable. The clean time-varying magnetic field strength data sequence is then subjected to spatial three-dimensional interpolation and gradient calculation to construct a magnetic field strength isosurface cloud map and a magnetic field gradient tensor field in the underground space of the detection area. By integrating the characteristic vibration spectrum of the cable with the magnetic field gradient tensor field, the spatial orientation of potential cable current channels is identified through a feature space matching algorithm, generating preliminary candidate lines for cable paths.

[0019] See Figure 2 In one embodiment of the present invention, the raw magnetic field strength data acquired by the three-dimensional magnetic field sensing array contains multi-component information. The raw magnetic field strength data of each sensing node is transmitted separately to an adaptive noise suppression processor for processing. The adaptive noise suppression processor performs a Fast Fourier Transform (FFT) on each raw magnetic field strength data stream. The output of the FFT is the magnetic field strength spectrum, which is a frequency domain representation. In the magnetic field strength spectrum, the adaptive noise suppression processor identifies spectral peaks centered at 50 Hz and its integer multiples, constructing a spectral template for power frequency interference. The spectral template is centered at a specific frequency and covers the main peak and sidelobe features. Using an adaptive spectral subtraction algorithm, the adaptive noise suppression processor subtracts the component corresponding to the spectral template of power frequency interference from the spectrum of the raw magnetic field strength data. The core of the adaptive spectral subtraction algorithm is the subtraction of frequency domain amplitudes, and the weighting factor of the subtraction is adjusted in real time according to the estimation of signal and noise. The adaptive noise suppression processor performs empirical mode decomposition (EMD) on the remaining spectral components after spectral subtraction. EMD decomposes the signal into a series of intrinsic mode functions (EMFs) and a residual component, separating out the low-frequency trend term reflecting the diurnal variation of the geomagnetic field. The low-frequency trend term is characterized by the last EMF or residual component and is removed as geomagnetic background noise. An inverse Fourier transform is then performed on the signal after removing geomagnetic background noise to reconstruct a clean time-varying magnetic field strength data sequence in the time domain. This clean time-varying magnetic field strength data sequence serves as the input for subsequent spatial analysis.

[0020] In some embodiments, the adaptive noise suppression processor constructs the power frequency interference spectrum template using the following method: After identifying spectral peaks at 50 Hz and their integer multiples, these peaks appear as local maxima on the spectrum. The adaptive noise suppression processor extracts the spectral amplitudes within a preset bandwidth on both sides of each peak frequency, and normalizes these amplitudes to form a reference spectral profile corresponding to the center frequency. This operation is repeated for all identified power frequency harmonics, and all reference spectral profiles are superimposed to finally form a complete power frequency interference spectrum template. The adaptive spectral subtraction algorithm performs spectral component subtraction according to the following formula:

[0021] in: This represents the complex spectrum value of the original magnetic field strength data at frequency point k. This represents the normalized amplitude of the power frequency interference spectrum template at frequency point k. This represents the adaptive attenuation factor related to the signal-to-noise ratio of the input signal at frequency k. This represents the complex spectral value of the output at frequency point k after spectral subtraction. The empirical mode decomposition process is adaptive. The decomposition process does not pre-set basis functions. It separates the intrinsic mode functions from high frequency to low frequency from the signal through an iterative screening process until the decomposed residual becomes a monotonic function, representing the low-frequency trend term of the diurnal variation of the geomagnetic field.

[0022] In practical implementation, the surface vibration acoustic signals collected by the distributed acoustic wave sensor network, specifically the surface vibration acoustic signals from each fiber optic microphone node, are transmitted to a characteristic frequency extractor. The characteristic frequency extractor performs bandpass filtering on the surface vibration acoustic signals from each node of the distributed acoustic wave sensor network. The filter is a digital filter, retaining the frequency band from 100 Hz to 2 kHz. A short-time Fourier transform is performed on the filtered signal to generate a time-spectrum graph. The horizontal axis of the time-spectrum graph represents time, the vertical axis represents frequency, and the numerical values ​​represent energy intensity. In the time-spectrum graph, the characteristic frequency extractor searches for spectral lines whose energy is continuous in time and whose frequency is an integer multiple of 50 Hz. The search process is performed frame-by-frame. The characteristic frequency extractor extracts all the spectral lines that meet the frequency conditions and calculates the average energy of each spectral line within the analysis time window. The characteristic frequency extractor reconstructs the signal of the frequency components corresponding to the spectral lines whose average energy exceeds a preset threshold. The reconstruction methods include inverse short-time Fourier transform or filter bank synthesis, which are combined to form a characteristic vibration spectrum of the cable. The characteristic vibration spectrum of the cable is a time-frequency representation of multiple characteristic frequency components.

[0023] In some embodiments, the feature frequency extractor needs to select a window function and window length when performing a short-time Fourier transform. The feature frequency extractor uses a Hanning window as the window function, and the length of the analysis window is dynamically set according to the signal stability and frequency resolution requirements to ensure sufficient resolution in both the time and frequency domains. In the generated time-spectrum graph, the feature frequency extractor tracks the continuity of energy on the time axis for each 50 Hz integer multiple of frequency. Continuity is defined as the fluctuation of spectral line energy between adjacent time frames being less than a preset proportion, and the frequency center offset of the spectral line being less than a preset bandwidth. For each spectral line that meets the conditions, the feature frequency extractor calculates its average energy within the analysis time window. The average energy calculation formula is the integral of energy over time divided by the time window length. The feature frequency extractor reconstructs the signal by combining the average energy with the frequency components corresponding to the spectral lines at preset thresholds, and finally combines them to form a cable characteristic vibration spectrum graph. The cable characteristic vibration spectrum graph can be represented as a set of time-varying frequency components and their amplitudes.

[0024] Optionally, the adaptive noise suppression processor can update the spectral template for power frequency interference periodically. For example, whenever the load on the environmental power grid changes significantly, the adaptive noise suppression processor automatically triggers a new spectral peak identification and template reconstruction process to ensure effective suppression of dynamic power frequency interference. Optionally, when the characteristic frequency extractor extracts the characteristic vibration spectrum of the cable, a preset spectral energy threshold can be used. This preset spectral energy threshold can be set based on the statistical value of the average energy of all spectral lines during the current monitoring period, for example, set to 1.5 times the average energy value of all spectral lines.

[0025] See Figure 3In one embodiment of the present invention, a three-dimensional regular grid spatial model covering the detection area and a preset depth range needs to be established based on the geographical boundary of the detection area, the expected maximum burial depth of the cable, and the deployment range of the ground sensors. The spacing between grid nodes in the horizontal direction of the three-dimensional regular grid spatial model is usually related to the deployment spacing of the ground three-dimensional magnetic field sensing array to ensure the effectiveness of spatial interpolation. In specific implementations, the number of layers in the vertical direction of the three-dimensional regular grid spatial model, i.e., the discrete division of the preset depth range, needs to be determined based on geological survey data or the preset maximum detection depth, with each layer representing a depth plane. After establishing the three-dimensional regular grid spatial model, the pure time-varying magnetic field intensity data sequence of each three-axis fluxgate sensor at the sampling time is used as the magnetic field vector observation value at the geographical coordinates of the three-axis fluxgate sensor. The magnetic field vector observation value includes three components: eastward, northward, and vertical. These data constitute the discrete observation dataset of known points. The Kriging spatial interpolation algorithm is used to interpolate the discrete magnetic field vector observation values ​​to each grid node of the three-dimensional regular grid spatial model to obtain continuously distributed magnetic field vector data across the entire field. The Kriging spatial interpolation algorithm is based on the principles of spatial statistics. It uses a variogram model to describe the spatial correlation structure of magnetic field vectors and assigns the optimal unbiased estimate to each grid node to be interpolated by solving the Kriging equations. Finally, it generates magnetic field vector data with a continuous full-field distribution covering the entire three-dimensional regular grid spatial model.

[0026] In some embodiments, establishing a three-dimensional regular grid spatial model requires specifying model parameters. In a specific implementation, the geographical boundary of the detection area is defined by measurement coordinate points. The upper limit of the preset depth range is the ground surface, and the lower limit is a pre-set maximum detection depth, such as five meters underground. The three-dimensional regular grid spatial model has a grid resolution of 0.5 meters in the horizontal direction and a layer spacing of 0.2 meters in the vertical direction, thereby generating a three-dimensional data structure containing tens of thousands or even hundreds of thousands of grid nodes. When executing the Kriging spatial interpolation algorithm, interpolation calculations need to be performed independently for each magnetic field component. The spatial correlation of the eastward, northward, and vertical components is described by their respective variograms. The variogram model used in the Kriging spatial interpolation algorithm is typically a spherical model or a Gaussian model, and the model parameters are determined by fitting the spatial semivariance of known observation points. The Kriging spatial interpolation algorithm works as follows: For each grid node in a 3D regular grid spatial model, the algorithm searches for all triaxial fluxgate sensor observation points within a certain radius of its vicinity. Using the observed magnetic field vector values ​​from these observation points and the fitted variogram model, the algorithm constructs and solves the Kriging equations, calculating the interpolated estimate of the magnetic field vector for that grid node and its variance. The resulting continuously distributed magnetic field vector data is a 3D vector field defined on a 3D regular grid spatial model, with each grid node storing one magnetic field vector.

[0027] In practice, based on the interpolated, continuously distributed magnetic field vector data, the spatial rate of change of the magnetic field vector at each grid node in three orthogonal directions is calculated, i.e., the magnetic field gradient components. The method for calculating the magnetic field gradient components is the central difference method within a three-dimensional regular grid spatial model. For any grid node within the three-dimensional regular grid spatial model, the spatial rate of change of its eastward magnetic field component along the east direction is obtained by centrally differencing the eastward component values ​​of the adjacent grid nodes on its east and west sides. Similarly, nine spatial rates of change of the northward and vertical components along the east, north, and vertical directions can be obtained. These nine spatial rates of change together constitute the magnetic field gradient tensor at that grid node. Combining all the magnetic field gradient components of each grid node forms a magnetic field gradient tensor field describing the spatial rate of change of the magnetic field. The magnetic field gradient tensor field is a third-order tensor field defined on the three-dimensional regular grid spatial model. Simultaneously, the magnetic field strength scalar of each grid node, i.e. the magnitude of the magnetic field vector, is extracted from the continuously distributed magnetic field vector data. The magnetic field strength isosurface cloud map of the magnetic field strength scalar is drawn. The magnetic field strength isosurface cloud map forms a series of curved surfaces in three-dimensional space by connecting grid nodes with equal magnetic field strength scalars, which intuitively shows the spatial distribution of magnetic field strength.

[0028] It is understandable that the accuracy of magnetic field gradient component calculation depends on the grid resolution of the three-dimensional regular grid spatial model. The higher the grid resolution, the closer the magnetic field gradient components calculated by the central difference method can be to the actual local spatial rate of change. It is also understandable that each component of the magnetic field gradient tensor field contains rich spatial information, especially the vertical gradient component, which is indicative of shallow linear current sources. When plotting the magnetic field intensity isosurface contour map, it is necessary to set a series of magnetic field intensity scalar thresholds for the isosurfaces. These thresholds can be determined based on the maximum and minimum values ​​of the magnetic field intensity scalars and their distribution histogram in the continuously distributed magnetic field vector data.

[0029] Optionally, before interpolation, the Kriging space interpolation algorithm can preprocess the discrete magnetic field vector observations to remove obvious outliers caused by instantaneous sensor malfunctions. The criterion for outlier removal can be that the observed value deviates from the local spatial mean by more than three standard deviations. Optionally, when calculating the magnetic field gradient components, for grid nodes located on the boundary of a three-dimensional regular grid spatial model, since central differencing is not possible, one-sided differencing methods such as forward or backward differencing can be used to approximate the magnetic field gradient components to ensure the spatial integrity of the magnetic field gradient tensor field.

[0030] In one embodiment of the present invention, a predetermined number of characteristic frequency points with the strongest vibration energy are extracted from the characteristic vibration spectrum of the cable. The predetermined number of characteristic frequency points is determined based on the size of the detection area and the data quality; for example, the top five characteristic frequency points with the highest vibration energy are selected. The amplitude distribution of the signal corresponding to each extracted characteristic frequency point at each node of the distributed acoustic wave sensing network is recorded. The amplitude distribution is stored in the form of a vector, and each element of the vector corresponds to the amplitude of the signal at that characteristic frequency point received by a node of the distributed acoustic wave sensing network. From the magnetic field gradient tensor field, three-dimensional spatial coordinates of the magnetic field gradient amplitude that exhibit local maxima are extracted, forming a set of magnetic field anomalies. The criterion for judging local maxima is that, in a three-dimensional regular grid space model, if the magnetic field gradient amplitude of a certain grid node is greater than the magnetic field gradient amplitudes of all its directly adjacent grid nodes, then the coordinates of that grid node are determined to be a local maximum point. The coordinates of all grid nodes that meet the condition together constitute the set of magnetic field anomalies. Construct a joint feature space with geographic coordinates as the basis and vibration amplitude and magnetic field gradient amplitude as the feature vectors. Each point in the joint feature space, i.e. the feature vector, consists of three parts: spatial location coordinates, normalized vibration amplitude, and normalized magnetic field gradient amplitude.

[0031] In some embodiments, the calculation process for extracting local maxima from the magnetic field gradient tensor field is as follows: For each internal grid node in the three-dimensional regular mesh space model, its magnetic field gradient magnitude is calculated. The formula for calculating the magnetic field gradient magnitude is:

[0032] in: This represents the magnitude of the magnetic field gradient at that grid node. These represent the nine components of the magnetic field gradient tensor field. The values ​​of each grid node are compared sequentially. The value is related to the grid nodes that are directly adjacent to it in the six directions of up, down, left, right, front, and back in three-dimensional space. Value, if the current grid node's The value is greater than that of all six adjacent grid nodes. If the value is positive, the three-dimensional spatial coordinates of the current grid node are determined to be a local maximum of the magnetic field gradient amplitude. The three-dimensional spatial coordinates of all local maximum points are extracted and stored in a list to form a set of magnetic field anomalies. When constructing the joint feature space, the multi-dimensional features from different sensors need to be scaled and normalized. For example, the vibration signal amplitude recorded at each node of the distributed acoustic wave sensor network is first mapped to geographic coordinates, and then, together with the magnetic field gradient amplitude at the corresponding location, is divided by the maximum value of each component within the entire detection area, thereby normalizing all feature components to between zero and one. Using geographic coordinates as the basis means that the first three dimensions of each feature vector are the east, north, and vertical coordinate values ​​of the three-dimensional spatial coordinates, and the last two dimensions are the normalized vibration amplitude feature and the normalized magnetic field gradient amplitude feature, respectively.

[0033] In practice, the set of magnetic field anomalies is mapped to a joint feature space, and the vibration amplitude distribution information of the distributed acoustic wave sensing network nodes is also mapped to the same joint feature space. For each three-dimensional coordinate point in the set of magnetic field anomalies, its eigenvector in the joint feature space has its first three dimensions given by its own coordinates, and its fifth eigenvalue, i.e., the normalized magnetic field gradient amplitude, is directly derived from that point. The normalized value is obtained by querying or interpolating its fourth-dimensional feature component, i.e., the normalized vibration amplitude. In specific implementations, the vibration amplitude distribution information of the distributed acoustic wave sensor network nodes is recorded for a preset number of feature frequency points, and a set of amplitude distributions is generated for each feature frequency point. Mapping the vibration amplitude distribution information of the distributed acoustic wave sensor network nodes to the joint feature space means generating a corresponding feature vector in the joint feature space for the geographic coordinates of each distributed acoustic wave sensor network node. The first three dimensions of this feature vector are the three-dimensional coordinates of the node, the fourth dimension is the normalized value of the average vibration amplitude of the node at all feature frequency points, and the fifth dimension is the normalized value of the magnetic field gradient amplitude at the node location, which is obtained by spatial interpolation of the magnetic field gradient tensor field at the node coordinates. In specific implementations, after the mapping is completed, the joint feature space contains feature points from the set of magnetic field anomalies and feature points from the distributed acoustic wave sensor network nodes. Within the joint feature space, a density clustering algorithm is used to find spatial clusters of feature points that simultaneously possess high vibration amplitude and high magnetic field gradient amplitude. The density clustering algorithm identifies densely packed regions of feature points in high-dimensional space as clusters and sparse regions as noise. In specific implementation, since the fourth and fifth feature components are normalized vibration amplitude and magnetic field gradient amplitude, the density clustering algorithm can identify dense regions formed by feature points with high values ​​in both feature dimensions. These feature points also exhibit clustering in the spatial dimension. The three-dimensional spatial curve connecting the center points of these spatial clusters is identified as a preliminary candidate line for the cable path. The center point of the spatial cluster can be taken as the average of the first three-dimensional spatial coordinates of all feature points belonging to the same cluster. The three-dimensional spatial polyline or curve formed by connecting these average coordinate points is the preliminary candidate line for the cable path.

[0034] It is understandable that the vibration amplitude and magnetic field gradient amplitude are comparable in magnitude after normalization, which facilitates the setting of uniform neighborhood radius and density threshold parameters for density clustering algorithms. It is also understandable that density clustering algorithms do not require pre-specifying the number of clusters and can automatically discover clusters of arbitrary shapes based on data distribution, making them suitable for the potentially curved shapes of underground cable paths in three-dimensional space. Optionally, the density clustering algorithm can be the DBSCAN algorithm, whose neighborhood radius and minimum number of points parameters need to be set before implementation based on the scale and data density of the detection scene. Optionally, when extracting local maxima of the magnetic field gradient amplitude, an absolute value threshold can be set, including only local maxima points with magnetic field gradient amplitudes greater than this absolute threshold in the magnetic field anomaly point set to exclude minor fluctuations caused by background noise.

[0035] In one embodiment of the invention, a series of spaced verification points are selected along the initial candidate cable path. The selection of verification points follows an equal-spacing principle, and the distance between adjacent verification points is determined based on the length of the candidate cable path and the expected spatial resolution; for example, a verification point is selected every two meters. From the clean time-varying magnetic field strength data sequence, the waveform of the magnetic field strength changing over time corresponding to each verification point is extracted. This extraction is achieved through spatial location mapping. For each verification point, the nearest ground-based three-dimensional magnetic field sensing array sensor node is found, or spatial interpolation is performed at the verification point coordinates using the constructed, continuously distributed magnetic field vector data to obtain the waveform of the magnetic field strength changing over time at that verification point. Cross-correlation analysis is performed on the magnetic field strength waveforms of all verification points to calculate the time delay between waveforms, and the apparent propagation speed of electromagnetic waves along the initial candidate cable path is estimated based on the known sensor spacing. Cross-correlation analysis calculates the similarity of the magnetic field strength waveforms of two verification points in the time domain; the time shift corresponding to the peak value is the time delay between the two waveforms. Dividing the spatial distance between adjacent verification points by the time delay yields the apparent propagation speed of electromagnetic waves between these two verification points. Based on the Biot-Savart law in electromagnetic field theory, a positive magnetic field calculation model is established with cable location, current amplitude, and phase as unknowns. The model treats the initial candidate cable path as a spatial polygonal line composed of a series of short, connected straight segments, each of which is a current-carrying segment. The magnetic field generated at any verification point in space by this segment is calculated according to the Biot-Savart law. Using the measured magnetic field strength waveform at the verification point as a constraint, an iterative optimization algorithm is employed to adjust the unknowns in the positive magnetic field calculation model until the calculated magnetic field waveform matches the measured waveform within a preset error range. The optimized cable current phase information is then checked for consistency along the initial candidate cable path. If the phase change is continuous and conforms to the characteristics of a single current-carrying conductor, the candidate cable path is determined to be the true cable path. The consistency check method involves examining the current phase value corresponding to each short straight segment after optimization. These phase values ​​should monotonically change along the candidate cable path or remain constant within a reasonable tolerance range.

[0036] In some embodiments, the forward magnetic field calculation model is initialized, assigning initial current amplitude, initial phase, and initial three-dimensional spatial position fine-tuning to the preliminary cable path candidate lines. The initial current amplitude can be roughly estimated based on the peak value of the measured magnetic field strength waveform at the verification point. The initial phase can be set to zero. The initial three-dimensional spatial position fine-tuning can be set to zero, meaning that the cable position in the first iteration is assumed to be the preliminary cable path candidate line itself. The forward magnetic field calculation model is based on the magnetic field generated by each short straight-line current-carrying conductor at a spatial point, calculated using the following formula:

[0037] in: Represents a single short line segment at a point in space. The magnetic field increment vector generated at that location, It is the vacuum permeability. It is the complex form of the current flowing through the line segment (including amplitude and phase). It is a line element vector on a line segment. From the position of the line element to the spatial point vector, yes The modulus length, the integral along the line segment from the starting point To the finish line The process is as follows: The magnetic field increment vectors generated by all short straight segments on the initial candidate cable path are vector-superimposed to obtain the theoretical waveform of the synthetic magnetic field strength at the verification point. Using initial parameters, the theoretical waveform of the synthetic magnetic field strength at each verification point is calculated using a forward magnetic field calculation model. The overall root mean square error (RMSE) between the theoretical waveform of the synthetic magnetic field strength at all verification points and the corresponding measured magnetic field strength waveform is calculated. The formula for calculating the overall RMSE is the root mean square of the sum of the squares of the differences between the theoretical waveform values ​​and the measured waveform values ​​at all verification points and all sampling time points. A gradient descent optimization strategy is adopted, iteratively adjusting the current amplitude, initial phase, and three-dimensional spatial position fine-tuning amount based on the overall RMSE. The gradient descent optimization strategy requires calculating the gradient of the overall RMSE with respect to each parameter to be optimized (current amplitude, initial phase, and three-dimensional position fine-tuning amount of each short straight segment node), and then adjusting the parameter values ​​along the opposite direction of the gradient with a certain learning rate to reduce the overall RMSE. After each iteration, the overall root mean square error is recalculated until the decrease in the overall root mean square error is less than the preset convergence threshold, or the number of iterations reaches the preset upper limit of the number of iterations. The model parameters at this point are taken as the optimization result, as shown in Table 1.

[0038] Table 1: Example of time delay results from cross-correlation analysis

[0039] It is understandable that the apparent propagation speed should be close to the propagation speed of electromagnetic waves in the surrounding medium. If the calculated apparent propagation speed deviates significantly from the normal range, it suggests that the initial candidate cable path may be inaccurate. It is also understandable that the gradient descent optimization strategy may get stuck in local optima. Therefore, in some implementations, multiple random initializations or strategies combined with simulated annealing can be used to increase the probability of finding the global optimum. Optionally, in the forward magnetic field calculation model, when discretizing the cable path into short straight segments, the segment length should be much smaller than the detection depth and the distance from the verification point to the cable to ensure the accuracy of the numerical calculation. Optionally, the preset error range can be set to five percent of the amplitude of the measured magnetic field strength waveform. When the root mean square error between the magnetic field waveform calculated by the model and the measured waveform is lower than this threshold, it is considered to have achieved a good match.

[0040] See Figure 4 This is a graph showing the impact of burial depth on the magnetic field gradient and vibration energy, illustrating the influence of cable burial depth on the vertical magnetic field gradient and the average energy of the vibration signal. The maximum value of the vertical magnetic field gradient decreases significantly with increasing cable burial depth. This is because the magnetic field strength generated by the current-carrying cable attenuates with increasing distance; the greater the burial depth, the smaller the magnetic field gradient at the surface. The average energy of the vibration signal also attenuates with increasing burial depth, but the attenuation rate is relatively gradual. This is because the propagation attenuation characteristics of vibration signals in the strata differ from those of the magnetic field. At a burial depth of approximately 2.0 meters, a localized rebound in vibration signal energy occurs, which may indicate changes in the strata structure or potential fault points at this depth, leading to an abnormal increase in vibration energy. When the burial depth exceeds 2.5 meters, the numerical difference between the two curves increases, which affects the accuracy of burial depth estimation, necessitating cross-validation using both methods.

[0041] In one embodiment of the invention, the background noise level of the detection area is monitored in real time. The background noise level includes the intensity of ambient electromagnetic noise and the intensity of ground vibration noise. The intensity of ambient electromagnetic noise is evaluated by calculating the total root mean square value of the remaining signal after removing the power frequency component from the raw magnetic field intensity data collected by multiple sensors in a three-dimensional magnetic field sensing array. The intensity of ground vibration noise is evaluated by calculating the average value of the energy spectral density of the ground vibration acoustic signal collected by multiple fiber optic microphone nodes in a distributed acoustic wave sensing network, outside the 50 Hz power frequency and its octave bands. Based on the monitored ambient electromagnetic noise intensity, the filtering bandwidth and threshold of the adaptive noise suppression processor are dynamically adjusted. For example, when the monitored ambient electromagnetic noise intensity is high, the adaptive noise suppression processor automatically widens the bandwidth used in its spectral subtraction algorithm to construct the power frequency interference spectrum template and increases the energy threshold for identifying power frequency peaks to more strictly filter out noise. Based on the monitored ground vibration noise intensity, the window function length and spectral line energy threshold of the short-time Fourier transform in the feature frequency extractor are dynamically adjusted. For example, when the surface vibration noise intensity is high, the feature frequency extractor uses a longer window function length to improve frequency resolution and correspondingly increases the spectral energy threshold used to determine cable characteristic spectral lines to resist noise interference. The adjusted filtering parameters are then applied to the processing flow of subsequently acquired raw magnetic field strength data and surface vibration acoustic signals. Based on changes in background noise levels, the required calculation accuracy of the magnetic field gradient tensor field and the clustering radius parameter in the feature space matching algorithm are updated synchronously. For example, under high background noise levels, the required signal-to-noise ratio for the magnetic field gradient tensor field calculation can be reduced, and the neighborhood radius of density clustering in the feature space matching algorithm can be increased to tolerate sparser feature point distributions.

[0042] In some embodiments, the quantitative assessment of environmental electromagnetic noise intensity follows a fixed time window, for example, calculating the statistical characteristics of all sensor background noise data within the past minute every minute. For monitoring environmental electromagnetic noise intensity, in specific implementations, the adaptive noise suppression processor, after eliminating power frequency interference and geomagnetic background noise, retains a wide-band noise reference signal. By calculating the total root mean square value of this noise reference signal within a specific analysis frequency band, the current quantitative index of environmental electromagnetic noise intensity is obtained. Based on the numerical range of this quantitative index, the adaptive noise suppression processor searches for and dynamically adjusts the adaptive attenuation factor α in the adaptive spectral subtraction algorithm and the bandwidth parameters when constructing the power frequency interference spectrum template from a preset parameter mapping table. For monitoring surface vibration noise intensity, the distributed acoustic wave sensor network continuously acquires raw acoustic signals. Before feature extraction, the feature frequency extractor performs a bandpass filter on the signal that does not contain 50 Hz or its harmonics, and then calculates the mean power spectral density of the filtered signal as an evaluation value of the surface vibration noise intensity. Based on this evaluation value, the feature frequency extractor dynamically selects the window function length of the short-time Fourier transform. A higher evaluation value results in a longer window function length, improving the ability to identify weak feature frequency lines in noise. Simultaneously, the energy threshold used by the feature frequency extractor to determine the validity of spectral lines also increases linearly with the evaluation value. The computational accuracy requirements for the magnetic field gradient tensor field are mainly reflected in the variogram fitting of the Kriging space interpolation algorithm and the variance threshold of the interpolation estimation. When the background noise level increases, the constraint on the variance of the interpolation results can be appropriately relaxed. The clustering radius parameter in the feature space matching algorithm refers to the neighborhood search radius of the density clustering algorithm. In high-noise environments, the distribution of feature points in the joint feature space may be more dispersed; therefore, it is necessary to appropriately increase the clustering radius to form effective clusters.

[0043] In practical implementation, for verified real cable paths, the maximum vertical magnetic gradient at the ground surface directly above the path is extracted from the magnetic field gradient tensor field. Specifically, along the projection line of the real cable path onto the ground surface, at each projection point, the vertical magnetic field gradient component (Gzz component) directly below it is read from the magnetic field gradient tensor field. The absolute maximum value of the Gzz component at all projection points is then identified as the maximum vertical magnetic gradient. Combined with the cable current amplitude obtained through current phase inversion, the theoretical burial depth of the cable is calculated using an infinitely long straight conductor magnetic field model. The vertical component of the magnetic field generated by an infinitely long straight conductor at a point on the ground surface is related to the current, burial depth, and horizontal distance. When the observation point is directly above the conductor, the horizontal distance is zero, and the relationship between the vertical magnetic gradient and the burial depth and current can be simplified. The formula for calculating the theoretical burial depth of the cable using the infinitely long straight conductor magnetic field model is:

[0044] in: This represents the theoretical burial depth of the cable calculated by inversely using the vertical gradient of the magnetic field. It is the vacuum permeability. The cable current amplitude is obtained through current phase inversion. This is the absolute value of the maximum vertical gradient of the magnetic field at the ground surface directly above the path, extracted from the magnetic field gradient tensor field. Simultaneously, the energy attenuation characteristics of the vibration signal distributed along the actual cable path are analyzed in the cable characteristic vibration spectrum diagram. In specific implementation, the cable characteristic vibration spectrum diagram provides the distribution of signal amplitude at each characteristic frequency point across nodes of the distributed acoustic wave sensor network. Using the projection line of the actual cable path onto the ground surface as a reference, nodes of the distributed acoustic wave sensor network near the projection line are selected. The variation of the signal amplitude at a specific characteristic frequency point on these nodes with the increase of the distance from the node to the starting point of the path projection line is analyzed, resulting in the vibration signal energy attenuation curve. A coupling relationship model between vibration energy attenuation, propagation distance, and cable burial depth is established. The equivalent burial depth of the cable is independently estimated by fitting the vibration energy attenuation curve. The coupling relationship model typically considers the stratum as a homogeneous half-space elastic medium. The propagation attenuation of vibration energy in the medium is related to the propagation distance and the wave propagation characteristics in the medium (related to the medium properties and wave type), while the cable burial depth affects the excitation efficiency and propagation path of the vibration. By fitting the measured vibration signal energy attenuation curve with the theoretical attenuation curve under different burial depth assumptions, the burial depth corresponding to the theoretical curve with the highest fitting degree is found, which is the equivalent burial depth estimated by the vibration energy attenuation curve.

[0045] By comparing the theoretical burial depth calculated from the vertical gradient of the magnetic field with the equivalent burial depth estimated from the vibration energy attenuation curve, if the absolute value of the difference between the theoretical and equivalent burial depths is less than a preset burial depth error threshold, the arithmetic mean of the two depths is calculated as the final burial depth estimate. The preset burial depth error threshold can be set according to measurement accuracy and engineering requirements, for example, 0.2 meters. If the absolute value of the difference between the theoretical and equivalent burial depths is greater than or equal to the preset burial depth error threshold, the corresponding cable section is marked as a potential insulation failure section or an external interference section, and identified as a suspected fault point. In this case, the significant difference between the burial depths estimated by the two independent methods suggests that the electromagnetic field radiation characteristics or vibration propagation characteristics of the cable in that section may be abnormal.

[0046] See Figure 5This is a multimodal cable burial depth estimation map, showcasing the core analysis results of the burial depth and fault point estimation stages in underground cable detection and location. The burial depth of eight cable sections was compared using three methods: magnetic field back-calculation, vibration back-calculation, and final burial depth estimation. The three burial depth estimates show an upward trend from sections 1 to 5, peaking in section 5 and gradually decreasing from sections 6 to 8. In sections 1 to 4, the results of the three methods are highly consistent, indicating reliable burial depth estimation. In section 5, the magnetic field back-calculated burial depth (1.80m) is slightly higher than the vibration back-calculated burial depth (1.70m), but the difference remains within a reasonable range. In sections 6 to 8, the results of the three methods remain consistent, with no significant differences, indicating that the cable condition in these sections is normal.

[0047] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A multimodal intelligent situational awareness method for detecting and locating underground cables, characterized in that, The method includes: The original magnetic field strength data is synchronously collected by a three-dimensional magnetic field sensing array deployed on the ground of the detection area. The three-dimensional magnetic field sensing array consists of multiple three-axis fluxgate sensors distributed in a grid, which are used to obtain the magnetic field vector information of the three-dimensional space of the detection area. The distributed acoustic wave sensor network deployed on the ground in the detection area is used to collect surface vibration acoustic signals. The distributed acoustic wave sensor network consists of an array of fiber optic microphones buried in the shallow soil layer, which is used to capture micro-vibration signals of the strata caused by underground cable flow or external interference. The synchronously acquired raw magnetic field strength data is input into an adaptive noise suppression processor to filter out power frequency interference and geomagnetic background noise, generating a pure time-varying magnetic field strength data sequence. The collected surface vibration acoustic signal is input into a feature frequency extractor to separate the vibration components related to the power frequency harmonics of the cable, and generate a characteristic vibration spectrum of the cable. Spatial three-dimensional interpolation and gradient calculation are performed on the purified time-varying magnetic field intensity data sequence to construct the magnetic field intensity isosurface cloud map and magnetic field gradient tensor field of the underground space of the detection area. By integrating the characteristic vibration spectrum of the cable with the magnetic field gradient tensor field, the spatial orientation of potential cable current channels is identified through a feature space matching algorithm, generating preliminary candidate cable paths.

2. The multimodal intelligent situational awareness method for detecting and locating underground cables according to claim 1, characterized in that, The step of inputting the synchronously acquired raw magnetic field strength data into an adaptive noise suppression processor to filter out power frequency interference and geomagnetic background noise includes: A fast Fourier transform is performed on the raw magnetic field strength data from each sensor of the three-dimensional magnetic field sensing array to obtain the magnetic field strength spectrum; In the magnetic field strength spectrum, identify the spectral peaks centered at 50 Hz and its integer multiples to construct a spectral template for power frequency interference; An adaptive spectral subtraction algorithm is used to subtract the component corresponding to the spectral template of the power frequency interference from the spectrum of the original magnetic field strength data; Empirical mode decomposition is performed on the remaining spectral components after spectral subtraction to separate the low-frequency trend term reflecting the diurnal variation of geomagnetism. The low-frequency trend term is then removed as geomagnetic background noise. The signal after removing the geomagnetic background noise is subjected to inverse Fourier transform to reconstruct the pure time-varying magnetic field intensity data sequence in the time domain.

3. The multimodal intelligent situational awareness method for detecting and locating underground cables according to claim 2, characterized in that, The step of inputting the collected surface vibration acoustic signal into a characteristic frequency extractor to separate the vibration component related to the power frequency harmonic characteristics of the cable includes: The surface vibration acoustic signal of each node in the distributed acoustic wave sensing network is bandpass filtered to retain the frequency band from 100 Hz to 2 kHz. Perform a short-time Fourier transform on the filtered signal to generate a time-frequency spectrum. In the aforementioned time-spectrum diagram, spectral lines whose energy is continuous in time and whose frequency is an integer multiple of fifty Hz are searched. Extract all spectral lines that meet the frequency conditions found in the search, and calculate the average energy of each spectral line within the analysis time window; The frequency components corresponding to spectral lines with average energy exceeding a preset threshold are reconstructed and combined to form the characteristic vibration spectrum of the cable.

4. The multimodal intelligent situational awareness method for detecting and locating underground cables according to claim 3, characterized in that, The step of performing spatial three-dimensional interpolation and gradient calculation on the purified time-varying magnetic field intensity data sequence to construct the magnetic field intensity isosurface cloud map and magnetic field gradient tensor field of the underground space of the detection area includes: Establish a three-dimensional regular grid spatial model covering the detection area and the preset depth range; The pure time-varying magnetic field intensity data sequence of each of the three-axis fluxgate sensors at the sampling time is used as the magnetic field vector observation value at the geographic coordinates of the three-axis fluxgate sensor; The Kriging space interpolation algorithm is used to interpolate the discrete magnetic field vector observations onto each grid node of the three-dimensional regular grid space model to obtain magnetic field vector data that are continuously distributed throughout the field. Based on the interpolated magnetic field vector data of the continuous distribution across the entire field, the spatial rate of change of the magnetic field vector at each grid node in the three orthogonal directions is calculated, i.e., the magnetic field gradient components. All magnetic field gradient components of each grid node are combined to form the magnetic field gradient tensor field describing the rate of change of the magnetic field in space, and the magnetic field intensity isosurface contour map of the magnetic field intensity scalar is plotted.

5. The multimodal intelligent situational awareness method for detecting and locating underground cables according to claim 4, characterized in that, The process of fusing the characteristic vibration spectrum of the cable with the magnetic field gradient tensor field, and identifying the spatial orientation of potential cable current channels using a feature space matching algorithm, includes: Extract a preset number of characteristic frequency points with the strongest vibration energy from the characteristic vibration spectrum of the cable, and record the amplitude distribution of the signal corresponding to each extracted characteristic frequency point on each node of the distributed acoustic wave sensor network. From the magnetic field gradient tensor field, extract the three-dimensional spatial coordinates of the magnetic field gradient amplitude points where local maxima occur to form a set of magnetic field anomaly points; Construct a joint feature space based on geographic coordinates and incorporating vibration amplitude and magnetic field gradient amplitude as feature vectors; The set of magnetic field anomalies is mapped to the joint feature space, and the vibration amplitude distribution information of the distributed acoustic wave sensing network nodes is mapped to the same joint feature space. Within the joint feature space, a density clustering algorithm is used to find spatial clustering regions that simultaneously possess high vibration amplitude and high magnetic field gradient amplitude features. The three-dimensional spatial curve connecting the center points of the spatial clustering regions is identified as the preliminary candidate cable path.

6. The multimodal intelligent situational awareness method for underground cable detection and positioning according to claim 5, characterized in that, It also includes the steps of performing current phase inversion and path authenticity verification on the preliminary cable path candidates: On the initial candidate cable route, a series of verification points with intervals are selected; From the purified time-varying magnetic field strength data sequence, extract the waveform of the magnetic field strength changing with time corresponding to each verification point location; Cross-correlation analysis was performed on the magnetic field intensity waveforms at all verification points to calculate the time delay between waveforms, and the apparent propagation speed of electromagnetic waves along the preliminary candidate cable path was estimated based on the known sensor spacing. Based on the Biot-Savart law in electromagnetic field theory, a positive magnetic field calculation model is established with the cable position, current amplitude and phase as unknowns. Using the measured magnetic field strength waveform at the verification point as a constraint, an iterative optimization algorithm is used to adjust the unknowns in the positive magnetic field calculation model until the magnetic field waveform calculated by the model matches the measured waveform within a preset error range. The optimized cable current phase information is checked for consistency along the preliminary cable path candidate line. If the phase change is continuous and conforms to the law of a single current-carrying conductor, the cable path candidate line is determined to be the real cable path.

7. The multimodal intelligent situational awareness method for detecting and locating underground cables according to claim 6, characterized in that, The step of adjusting the unknowns in the positive magnetic field calculation model using an iterative optimization algorithm, constrained by the measured magnetic field strength waveform at the verification point, includes: Initialize the positive magnetic field calculation model and assign initial current amplitude, initial phase, and initial three-dimensional spatial position fine-tuning amount to the preliminary cable path candidate lines; Using the initial parameters, the theoretical waveform of the synthetic magnetic field strength at each verification point is calculated using the positive magnetic field calculation model; Calculate the overall root mean square error between the theoretical waveform of the synthesized magnetic field strength and the corresponding measured magnetic field strength waveform at all verification points; A gradient descent optimization strategy is adopted to iteratively adjust the current amplitude, initial phase, and three-dimensional spatial position fine-tuning amount based on the overall root mean square error. After each iteration, the overall root mean square error is recalculated until the decrease in the overall root mean square error is less than the preset convergence threshold, or the number of iterations reaches the preset upper limit of the number of iterations. The model parameters at this point are then used as the optimization result.

8. The multimodal intelligent situational awareness method for detecting and locating underground cables according to claim 7, characterized in that, It also includes a dynamic detection sensitivity adjustment step based on environmental parameters: Real-time monitoring of the background noise level in the detection area, including the intensity of environmental electromagnetic noise and the intensity of ground vibration noise; The filtering bandwidth and threshold of the adaptive noise suppression processor are dynamically adjusted based on the monitored environmental electromagnetic noise intensity. Based on the monitored surface vibration noise intensity, the window function length and spectral energy threshold of the feature frequency extractor for short-time Fourier transform are dynamically adjusted. The adjusted filtering parameters are then applied to the subsequent processing flow of the original magnetic field strength data and the surface vibration acoustic signal. Based on the changes in the background noise level, the required calculation accuracy of the magnetic field gradient tensor field and the clustering radius parameter in the feature space matching algorithm are updated synchronously.

9. The multimodal intelligent situational awareness method for detecting and locating underground cables according to claim 8, characterized in that, It also includes a step of performing multimodal joint estimation of the burial depth and fault location of the identified cable path: For a verified real cable path, extract the maximum value of the vertical magnetic gradient at the ground surface directly above the path from the magnetic field gradient tensor field. By combining the cable current amplitude obtained through current phase inversion, the theoretical burial depth of the cable is calculated using an infinitely long straight conductor magnetic field model. Simultaneously, the vibration signal energy attenuation characteristics distributed along the actual cable path are analyzed in the characteristic vibration spectrum diagram of the cable. A coupled relationship model between vibration energy attenuation, propagation distance, and cable burial depth is established, and the equivalent burial depth of the cable is independently estimated by fitting the vibration energy attenuation curve. By comparing the theoretical burial depth obtained from the back calculation of the vertical gradient of the magnetic field with the equivalent burial depth estimated by the vibration energy attenuation curve, if the absolute value of the difference between the theoretical burial depth and the equivalent burial depth is less than the preset burial depth error threshold, the arithmetic mean of the theoretical burial depth and the equivalent burial depth is calculated as the final burial depth estimate; if the absolute value of the difference between the theoretical burial depth and the equivalent burial depth is greater than or equal to the preset burial depth error threshold, the corresponding cable section is marked as a potential insulation damage section or an external interference section, and is identified as a suspected fault point.

10. A multimodal intelligent situational awareness underground cable detection and positioning system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the underground cable detection and positioning method for multimodal intelligent situational awareness as described in any one of claims 1 to 9.

Images