A Multimodal Slope Condition Monitoring Method Based on Infrasound and BeiDou Signals

Abstract

This invention relates to the field of geological disaster prevention and control, specifically disclosing a multimodal slope condition monitoring method based on infrasound and BeiDou signals. The method includes: determining potential slip surfaces based on a three-dimensional geological model of the slope and numerical simulation; collaboratively deploying an infrasound sensor array and a BeiDou monitoring station; achieving reliable data transmission through a hybrid self-organizing network and BeiDou short messages; extracting infrasound features using wavelet packet transform and spectral analysis; processing BeiDou signals using a state threshold least squares algorithm; constructing a spatiotemporal convolutional network based on an attention mechanism to dynamically fuse multimodal features; triggering graded early warnings based on a dynamic threshold model; and adaptively adjusting system operating parameters. This invention solves the problems of unscientific deployment, low fusion degree, and poor environmental adaptability in existing monitoring methods, achieving early and accurate early warning and adaptive intelligent monitoring of slope conditions. It can be widely applied to slope stability monitoring in mining, highway, and railway fields.

Description

Technical Field

[0001] This invention relates to the field of geological disaster prevention and control technology, and in particular to a multimodal slope condition monitoring method based on infrasound and BeiDou signals. Background Technology

[0002] Slope stability monitoring is a major technical challenge in the field of geological disaster prevention and control, especially in engineering fields such as mining and transportation infrastructure, where it has significant safety implications. Traditional slope monitoring technologies mainly include displacement measurement methods (such as total stations and GNSS surface displacement monitoring), physical sensor methods (such as inclinometers and stress-strain gauges), and emerging remote sensing monitoring methods (such as InSAR). Although these methods are widely used, they have obvious limitations: surface displacement monitoring often only triggers early warnings when significant slope deformation occurs, resulting in a warning lag; single-point sensor monitoring range is limited, making it difficult to comprehensively reflect the overall slope condition; and InSAR technology is susceptible to atmospheric interference and has a long revisit period, failing to meet real-time monitoring requirements.

[0003] In recent years, multimodal sensing technology has been increasingly applied to slope monitoring, particularly the combination of infrasound monitoring and BeiDou satellite navigation technology, which has demonstrated significant potential value. Infrasound monitoring can capture low-frequency acoustic emission signals generated by micro-fractures within soil and rock masses. Studies have shown that infrasound signal anomalies can appear hours or even days before surface displacement, offering a significant early warning advantage. The BeiDou satellite navigation system can provide surface displacement monitoring with millimeter-level accuracy, forming an integrated space-ground monitoring capability.

[0004] However, existing multimodal monitoring methods based on infrasound and BeiDou still face several technical bottlenecks: First, the deployment of monitoring points lacks scientific rigor, often simply following geometric patterns without organically integrating with the geological structure of the slope and the characteristics of potential slip surfaces, resulting in monitoring blind spots; second, the degree of multimodal data fusion is low, typically employing simple data overlay or weighted averaging, failing to deeply explore the spatiotemporal correlation between infrasound characteristics and displacement changes; third, in complex mountainous environments, data transmission reliability is insufficient, especially in areas without public network coverage, where critical early warning information cannot be uploaded in a timely manner; finally, existing analysis methods lack sufficient intelligence, making it difficult to adapt to different geological conditions and environmental interference, resulting in high false alarm and false negative rates.

[0005] Patent CN114397015A proposes a slope monitoring scheme based on BeiDou and acoustic emission, but it only simply combines data from two sensors and fails to address the issues of deep fusion of multimodal data and adaptive early warning. While CN113689668A employs multi-sensor information fusion, its deployment strategy does not consider geological model guidance and lacks a reliable transmission scheme in weak network environments. Therefore, there is an urgent need for a scientifically deployable, deeply fused, and intelligently analyzed infrasound and BeiDou multimodal slope monitoring method to achieve early and accurate warnings of slope instability. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a multimodal slope condition monitoring method based on infrasound and BeiDou signals.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A multimodal slope condition monitoring method based on infrasound and BeiDou signals includes:

[0009] Based on the potential slip surface determined by the three-dimensional geological model and numerical simulation of the slope, infrasound sensor arrays and Beidou monitoring stations are deployed in coordination at the top of the slope, the shoulder of the slope and the potential slip surface area to ensure that the monitoring network covers key risk areas.

[0010] Infrasound sensor arrays collect infrasound signals from rock fractures, and BeiDou monitoring stations acquire millimeter-level displacement data; reliable data transmission in complex environments is achieved through a hybrid self-organizing network of ZigBee, LoRa, and BeiDou short messages.

[0011] The original signal is denoised using wavelet packet transform; the short-time Fourier transform (STFT) of the signal is calculated to obtain the time spectrum; the main frequency, frequency band energy ratio, spectral entropy, and azimuth features based on time difference of arrival (TDOA) are extracted; the state threshold least squares algorithm is used to process the BeiDou signal to suppress errors and improve the accuracy and stability of the solution.

[0012] The infrasound feature vector is spatiotemporally aligned with the BeiDou displacement sequence. The infrasound feature vector is a set of multidimensional, qualitative features describing the microscopic damage inside the slope, while the BeiDou displacement sequence is direct, quantitative measurement data describing the overall macroscopic deformation of the slope. The data is then input into an attention-based spatiotemporal convolutional network (AT-STCN) to dynamically weight and fuse multimodal features, and output the slope stability probability.

[0013] Based on the fused output, a dynamic threshold model is used to trigger tiered early warnings (blue, yellow, orange, and red); the sampling frequency of the sensor array is adaptively adjusted according to the early warning level and the on-site conditions.

[0014] Regularly feed back new monitoring data and early warning results to form a closed loop, and incrementally learn the spatiotemporal convolutional network (AT-STCN) to optimize its performance in specific slope environments.

[0015] As a further technical solution of the present invention, the potential slip surface determined based on the three-dimensional geological model of the slope and numerical simulation is deployed in conjunction with an infrasound sensor array and a Beidou monitoring station at the top of the slope, the shoulder of the slope, and the potential slip surface area. Specifically, this includes: establishing a three-dimensional geological model of the slope through oblique photography using terminal equipment; analyzing the stability under extreme working conditions using numerical simulation methods to determine the potential slip surface; the infrasound sensors adopt an equilateral triangle layout, with the center located at the top of the slope or the shoulder of the slope, a spacing of 20 to 50 meters, and a frequency response range of 0.01 to 20 Hz; and the Beidou monitoring station is deployed along the projection line of the potential slip surface and at key locations, including existing deformation areas, areas with weak geological structures, areas with geometric shapes and stress concentrations, and areas affected by engineering activities.

[0016] As a further technical solution of the present invention, the step of establishing a three-dimensional geological model of the slope through oblique photography of a terminal device specifically includes:

[0017] High-resolution image data of the slope surface is acquired through terminal equipment. Using the bundle adjustment algorithm, an observation equation is established to minimize the error, resulting in the three-dimensional point cloud data of the slope. The formula is: ,in: The total number of 3D points participating in the adjustment calculation. For the first Two-dimensional coordinates of each observation point Let these be the coordinates of the observation point in three-dimensional space. For the camera's imaging model, including focal length Principal point coordinates Rotation matrix Translation vector ;

[0018] The 3D point cloud data is triangulated using Delaunay triangulation to generate a triangular mesh, which is then used to construct a 3D geological model of the slope.

[0019] As a further technical solution of the present invention, the method of analyzing stability under extreme working conditions and determining potential slip surfaces using numerical simulation specifically includes:

[0020] The stress distribution of a slope under extreme working conditions is analyzed using the equations of elasticity; the formula is: ,in: It is normal stress. For shear stress, Let Lamé constant be . For strain components, For shear strain components;

[0021] The safety factor of the slope is calculated and the potential slip surface is determined using the limit equilibrium method; the formula is: ,in: For safety reasons, The cohesion of soil and rock. The length of the slip surface, For normal stress, It is the internal friction angle. The weight of the sliding body. The angle of inclination of the sliding surface;

[0022] Assuming the slip surface is a circular arc, use the coordinates of the center of the circle. and radius The parameterized slip surface is defined by the following formula: ,in: Let be the two-dimensional planar coordinates of a point on the slip surface, which is assumed to be a circular arc. These are the coordinates of a point on the arc.

[0023] As a further technical solution of the present invention, the short-time Fourier transform (STFT) for obtaining the time spectrum specifically includes:

[0024] The signal is divided into frames, and a Hamming window is used. • Calculate the Fourier Transform (FFT) for each frame to obtain the time-frequency spectrum:

[0025]

[0026] in: Indicates the first Frame at frequency amplitude, It's a signal. Let be the discrete position of the signal in the time domain. It is the number of pixels per frame. It is the imaginary unit of complex numbers.

[0027] As a further technical solution of the present invention, the extraction of the dominant frequency, frequency band energy ratio, spectral entropy, and azimuth features based on time difference of arrival (TDOA) specifically includes:

[0028] clock speed ;

[0029] Bandwidth energy ratio ,in and These are the start and end points of the frequency band, respectively.

[0030] Spectral entropy ,in ;

[0031] Based on the azimuth characteristics of Time Difference of Address (TDOA), assuming that the two sensors receive the signals at the following times... and The speed of sound is Then the azimuth angle Calculated using the following formula: ,in It is the distance between the two sensors.

[0032] As a further technical solution of the present invention, the process of processing BeiDou signals using the state threshold least squares algorithm specifically includes:

[0033] Assuming the BeiDou signal is The solution result is The algorithm is then expressed as:

[0034]

[0035] in, It is a design matrix. It is an adaptive convergence threshold used to prevent the algorithm iteration from failing to converge;

[0036] Solution results Calculated using the following formula: ,in: It is a matrix transpose, It is an identity matrix.

[0037] As a further technical solution of the present invention, the infrasound feature vector is spatiotemporally aligned with the BeiDou displacement sequence, wherein the infrasound feature vector is a set of multidimensional, qualitative features describing the microscopic damage inside the slope, and the BeiDou displacement sequence is direct, quantitative measurement data describing the overall macroscopic deformation of the slope, specifically including:

[0038] Assume the infrasound feature sequence is The BeiDou displacement sequence is Their sampling times are respectively and By aligning the time using interpolation methods, and assuming BeiDou data as the quasi-reference, the new aligned infrasound feature sequence is: With BeiDou displacement sequence One-to-one correspondence.

[0039] As a further technical solution of the present invention, the input to the attention-based spatiotemporal convolutional network (AT-STCN) dynamically weights and fuses multimodal features to output the slope stability probability, specifically including:

[0040] Infrasound characteristics and Beidou features The inputs are respectively fed into a 1D CNN (for capturing temporal characteristics) and a 2D CNN (for capturing spatiotemporal characteristics): ,in: and The feature extraction results are for infrasound and BeiDou, respectively.

[0041] Define an attention module that obtains weights by calculating the correlation between features: ,in: It is a learnable parameter matrix. It is the weight of infrasound features. Given the length of the time series, the weights of the BeiDou features are: ;

[0042] Dynamically weighted fusion of feature vectors ;

[0043] The fused features are then input into a classifier (such as Softmax) to output the slope state probability. ,in: This represents the probability of slope stability. These represent stable and unstable slopes, respectively. and It is the weight vector of the classifier.

[0044] A multimodal slope condition monitoring system based on infrasound and BeiDou signals is used to implement a multimodal slope condition monitoring method based on infrasound and BeiDou signals, including:

[0045] The infrasound signal acquisition module deploys an array of infrasound sensors in key areas of the slope to capture infrasound signals generated by rock mass fracturing.

[0046] The Beidou displacement monitoring module deploys Beidou monitoring stations along the projection line of the potential slip surface and at key locations on the slope to acquire millimeter-level displacement data and monitor the slope displacement in real time.

[0047] The data transmission and processing module constructs a hybrid self-organizing network of ZigBee, LoRa, and BeiDou short message service to achieve reliable data transmission in complex environments.

[0048] The feature extraction and noise reduction module is used to denoise infrasound signals using wavelet packet transform and to extract the dominant frequency, frequency band energy ratio, spectral entropy, and azimuth features based on time difference of arrival.

[0049] The multimodal feature fusion module aligns the infrasound feature vector with the BeiDou displacement sequence in time and space. The infrasound feature vector is a set of multidimensional and qualitative features describing the microscopic damage inside the slope, while the BeiDou displacement sequence is direct and quantitative measurement data describing the overall macroscopic deformation of the slope. The input is a spatiotemporal convolutional network (AT-STCN) based on the attention mechanism, which dynamically weights and fuses the multimodal features to output the probability of the slope's stable state.

[0050] The slope condition classification and early warning module, based on fusion output, uses a dynamic threshold model to trigger graded early warnings (blue, yellow, orange, red). According to the early warning level and the site conditions, it adaptively adjusts the sampling frequency of the sensor array to ensure the real-time performance and accuracy of slope monitoring.

[0051] The online model update module regularly feeds back new monitoring data and early warning results to form a closed loop, enabling incremental learning of the spatiotemporal convolutional network (AT-STCN) to optimize its performance in specific slope environments and improve the model's adaptability and prediction accuracy.

[0052] The beneficial effects of this invention are as follows:

[0053] 1. Improved scientific rigor and targeted approach to monitoring: By employing a collaborative deployment method based on three-dimensional geological models and numerical simulations of slopes, the infrasound sensor array and BeiDou monitoring stations can accurately cover potential slip surfaces and key risk areas. This solves the monitoring blind spot problem inherent in traditional uniform deployment and improves the spatial representativeness and risk capture capability of the monitoring network.

[0054] 2. Significantly enhanced early warning capability: By utilizing the sensitivity of infrasound to micro-fractures in rock mass, combined with wavelet packet transform and spectral feature extraction algorithms, it is possible to effectively identify early signs of slope instability.

[0055] 3. Deep fusion of multimodal data: The innovative attention-based spatiotemporal convolutional network (AT-STCN) can dynamically weight the importance of different modal features, fully explore the spatiotemporal correlation between infrasound features and BeiDou displacement sequences, solve the information loss problem in simple data superposition or weighted average fusion, and improve the accuracy of state recognition.

[0056] 4. Strong adaptability to complex environments: The multi-transmission scheme of ZigBee and LoRa hybrid self-organizing network combined with Beidou short message effectively solves the data transmission problem in mountainous areas without public network coverage, ensuring the integrity of monitoring data and the timeliness of early warning information. Attached Figure Description

[0057] Figure 1 This is a flowchart of a multimodal slope condition monitoring method based on infrasound and BeiDou signals proposed in this invention.

[0058] Figure 2This is the equipment layout diagram in Example 1.

[0059] Figure 3 This is a comparison diagram of the effects of the present invention and the conventional method in Example 1. Detailed Implementation

[0060] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0061] Please see the appendix Figure 1 A multimodal slope condition monitoring method based on infrasound and BeiDou signals, comprising:

[0062] Based on the potential slip surface determined by the three-dimensional geological model and numerical simulation of the slope, infrasound sensor arrays and BeiDou monitoring stations are deployed in a coordinated manner at the slope top, slope shoulder, and potential slip surface area to ensure that the monitoring network covers key risk areas, specifically including:

[0063] A three-dimensional geological model of the slope was created using oblique photography with terminal equipment.

[0064] High-resolution image data of the slope surface is acquired through terminal equipment. Using the bundle adjustment algorithm, an observation equation is established to minimize the error, resulting in the three-dimensional point cloud data of the slope. The formula is: ,in: The total number of 3D points participating in the adjustment calculation. For the first Two-dimensional coordinates of each observation point Let these be the coordinates of the observation point in three-dimensional space. For the camera's imaging model, including focal length Principal point coordinates Rotation matrix Translation vector ;

[0065] The 3D point cloud data is subjected to Delaunay triangulation to generate a triangular mesh, and a 3D geological model of the slope is constructed, specifically including:

[0066] First, the point cloud data is segmented into a regular Voxel grid, with the size of each Voxel adjustable according to specific needs. Next, 2D Delaunay triangulation is performed on the points within each Voxel, typically projected onto the XY plane for calculation. Then, the 2D triangulation results are extended to 3D by adding a corresponding Z-coordinate to each triangle. Finally, the triangulation results of all Voxels are merged to form a complete 3D triangular mesh, thus constructing a 3D geological model of the slope. In this process, algorithms such as the Bowyer-Watson algorithm can be used to implement Delaunay triangulation, and it is important to adjust the Voxel size according to the point cloud density to optimize computational efficiency.

[0067] Stability under extreme conditions was analyzed using numerical simulation methods to determine the potential slip surface:

[0068] The stress distribution of a slope under extreme working conditions is analyzed using the equations of elasticity; the formula is: ,in: It is normal stress. For shear stress, Let Lamé constant be . For strain components, For shear strain components;

[0069] The safety factor of the slope is calculated and the potential slip surface is determined using the limit equilibrium method; the formula is: ,in: For safety reasons, The cohesion of soil and rock. The length of the slip surface, For normal stress, It is the internal friction angle. The weight of the sliding body. The angle of inclination of the sliding surface;

[0070] Assuming the slip surface is a circular arc, use the coordinates of the center of the circle. and radius The parameterized slip surface is defined by the following formula: ,in: Let be the two-dimensional planar coordinates of a point on the slip surface, which is assumed to be a circular arc. These are the coordinates of a point on the arc.

[0071] The infrasound sensors adopt an equilateral triangle layout, with the center located at the top or shoulder of the slope and a spacing of 20 to 50 meters, with a frequency response range of 0.01 to 20 Hz; the Beidou monitoring stations are deployed along the projection line of the potential slip surface and in key locations, including existing deformation areas, areas with weak geological structures, areas with geometric shapes and stress concentrations, and areas affected by engineering activities.

[0072] Infrasound sensor arrays collect infrasound signals from rock fractures, and BeiDou monitoring stations acquire millimeter-level displacement data; reliable data transmission in complex environments is achieved through a hybrid self-organizing network of ZigBee, LoRa, and BeiDou short messages.

[0073] The hybrid self-organizing network specifically includes: using ZigBee for short-range, low-power communication within the monitoring network; using LoRa for medium-range data transmission in areas without public network coverage; and enabling BeiDou short message service to transmit critical early warning information in emergency situations or when regular communication is interrupted.

[0074] Denoising of the original signal is performed using wavelet packet transform; the short-time Fourier transform (STFT) of the signal is calculated to obtain the time spectrum; the dominant frequency, bandwidth energy ratio, spectral entropy, and azimuth features based on time difference of arrival (TDOA) are extracted; the state threshold least squares algorithm is used to process the BeiDou signal to suppress errors and improve the accuracy and stability of the solution; specifically including:

[0075] Wavelet packet transform denoising: Selecting a suitable wavelet basis and decomposition level (e.g., ...) or The signal is decomposed into multi-resolution components to obtain wavelet packet coefficients. Then, an adaptive soft thresholding strategy is used to process the coefficients and remove noise:

[0076]

[0077] in: These are the coefficients after noise reduction; For the sign function, when the wavelet packet coefficients When greater than 0, Output +1 when When the value is less than 0, output -1; if If the value is 0, then output 0; The threshold is adaptively calculated based on the standard deviation of the signal and noise.

[0078] Short-Time Fourier Transform (STFT) for obtaining the time spectrum: The signal is divided into frames, and a Hamming window is used. • Calculate the Fourier Transform (FFT) for each frame to obtain the time-frequency spectrum:

[0079]

[0080] in: Indicates the first Frame at frequency amplitude, It's a signal. Let be the discrete position of the signal in the time domain. It is the number of pixels per frame. It is the imaginary unit of complex numbers.

[0081] Extract the dominant frequency, bandwidth energy ratio, spectral entropy, and azimuth features based on time difference of arrival (TDOA):

[0082] clock speed ;

[0083] Bandwidth energy ratio ,in and These are the start and end points of the frequency band, respectively.

[0084] Spectral entropy ,in ;

[0085] Based on the azimuth characteristics of Time Difference of Address (TDOA), assuming that the two sensors receive the signals at the following times... and The speed of sound is Then the azimuth angle Calculated using the following formula: ,in It is the distance between the two sensors.

[0086] The state-threshold least squares algorithm is used for BeiDou signal processing. It introduces an adaptive convergence threshold to prevent algorithm iteration from failing to converge, thereby improving positioning accuracy and program robustness, and ensuring the reliability of displacement monitoring data. Specifically, it includes:

[0087] Assuming the BeiDou signal is The solution result is The algorithm is then expressed as:

[0088]

[0089] in, It is a design matrix. It is an adaptive convergence threshold used to prevent the algorithm iteration from failing to converge;

[0090] Solution results Calculated using the following formula: ,in: It is a matrix transpose, It is an identity matrix.

[0091] The infrasound feature vector is spatiotemporally aligned with the BeiDou displacement sequence. The infrasound feature vector is a set of multidimensional, qualitative features describing the microscopic damage within the slope, while the BeiDou displacement sequence is direct, quantitative measurement data describing the overall macroscopic deformation of the slope. The data is then input into an attention-based spatiotemporal convolutional network (AT-STCN) for dynamic weighted fusion of multimodal features, outputting the slope stability probability. Specifically, this includes:

[0092] Assume the infrasound feature sequence is The BeiDou displacement sequence is Their sampling times are respectively and By aligning the time using interpolation methods, and assuming BeiDou data as the quasi-reference, the new aligned infrasound feature sequence is: With BeiDou displacement sequence One-to-one correspondence.

[0093] The Attention-Based Spatiotemporal Convolutional Network (AT-STCN) includes: a one-dimensional CNN branch for extracting temporal patterns of infrasound features; a two-dimensional CNN branch for mining the spatiotemporal correlation of BeiDou displacement; an attention fusion module that dynamically calculates the weights of each modality feature to achieve fusion focused on key information; and a fully connected layer that outputs the classification probability of slope stability.

[0094] The Spatiotemporal Convolutional Network (AT-STCN) employs a multi-task learning mechanism during training. The main task is slope state classification, while the auxiliary tasks are displacement prediction and infrasound signal reconstruction, in order to improve the model's feature extraction and generalization capabilities.

[0095] Infrasound characteristics and Beidou features The inputs are respectively fed into a 1D CNN (for capturing temporal characteristics) and a 2D CNN (for capturing spatiotemporal characteristics): ,in: and The feature extraction results are for infrasound and BeiDou, respectively.

[0096] Define an attention module that obtains weights by calculating the correlation between features: ,in: It is a learnable parameter matrix. It is the weight of infrasound features. Given the length of the time series, the weights of the BeiDou features are: ;

[0097] Dynamically weighted fusion of feature vectors ;

[0098] The fused features are then input into a classifier (such as Softmax) to output the slope state probability. ,in: This represents the probability of slope stability. These represent stable and unstable slopes, respectively. and It is the weight vector of the classifier.

[0099] Based on the fused output, a dynamic threshold model is used to trigger tiered early warnings (blue, yellow, orange, and red); the sampling frequency of the sensor array is adaptively adjusted according to the early warning level and the on-site conditions.

[0100] The dynamic threshold model adaptively adjusts the warning threshold based on real-time rainfall, ground motion and other environmental parameters, as well as historical slip data. The specific definitions of the graded warning are: blue warning (strengthen manual patrols), yellow warning (push information to management personnel), orange warning (stop on-site work and prepare for reinforcement), and red warning (activate emergency response and evacuate personnel).

[0101] The adaptive adjustment is as follows: under a blue alert, the system operates at the base frequency; under a yellow alert, the infrasound sampling frequency and the BeiDou calculation frequency are increased to 1.5 times the base frequency; under an orange or higher alert, all computing resources are activated, backup sensors are woken up, and the system enters a full power consumption monitoring mode.

[0102] Regularly feed back new monitoring data and early warning results to form a closed loop, and incrementally learn the spatiotemporal convolutional network (AT-STCN) to optimize its performance in specific slope environments.

[0103] A multimodal slope condition monitoring system based on infrasound and BeiDou signals is used to implement a multimodal slope condition monitoring method based on infrasound and BeiDou signals, including:

[0104] Infrasound signal acquisition module: An array of infrasound sensors is deployed in key areas of the slope to capture infrasound signals generated by rock mass fracturing;

[0105] Beidou displacement monitoring module: Beidou monitoring stations are deployed along the projection line of the potential slip surface and at key locations on the slope to acquire millimeter-level displacement data and monitor the slope displacement in real time;

[0106] Data transmission and processing module: Constructs a hybrid self-organizing network of ZigBee, LoRa and BeiDou short message to achieve reliable data transmission in complex environments;

[0107] Feature extraction and noise reduction module: used to denoise infrasound signals using wavelet packet transform, and extract the dominant frequency, frequency band energy ratio, spectral entropy, and azimuth features based on time difference of arrival;

[0108] Multimodal feature fusion module: Spatiotemporally aligns infrasound feature vectors with BeiDou displacement sequences. Infrasound feature vectors are a set of multidimensional, qualitative features describing microscopic damage inside the slope, while BeiDou displacement sequences are direct, quantitative measurement data describing the overall macroscopic deformation of the slope. The input is a spatiotemporal convolutional network (AT-STCN) based on an attention mechanism, which dynamically weights and fuses multimodal features to output the slope stability probability.

[0109] Slope condition classification and early warning module: Based on fusion output, a dynamic threshold model is used to trigger graded early warnings (blue, yellow, orange, red). According to the early warning level and the site conditions, the sampling frequency of the sensor array is adaptively adjusted to ensure the real-time performance and accuracy of slope monitoring.

[0110] Online model update module: Regularly feeds back new monitoring data and early warning results to form a closed loop, incrementally learns the spatiotemporal convolutional network (AT-STCN), optimizes its performance in specific slope environments, and improves the model's adaptability and prediction accuracy.

[0111] Example 1: Verification Case of Slope Monitoring in an Iron Mine

[0112] 1. Site Overview: The north slope of a large open-pit iron mine in Northwest my country was selected as the verification object. The slope is approximately 280 meters high with an overall slope angle of about 38°. The strata are mainly composed of weathered basalt and tuff, and two sets of dominant structural planes exist. This area has experienced several local collapses in recent years, and recent monitoring data shows an increase in displacement rate, making it an ideal location to verify the effectiveness of this invention.

[0113] 2. Monitoring System Deployment

[0114] According to the method of the present invention, the following deployment work was first carried out:

[0115] (1) Geological modeling and numerical simulation: A high-precision three-dimensional model of the slope (accuracy up to 5cm) was obtained using oblique photography technology of terminal equipment. Combined with geological survey data, a refined three-dimensional geological model including rock mass structural planes and weak interlayers was established. The stability under heavy rainfall and earthquake conditions was analyzed using FLAC3D numerical simulation software to determine the location and extent of potential slip surfaces.

[0116] (2) Sensor Co-deployment: Based on simulation results, monitoring equipment is deployed in the potential slip surface influence area:

[0117] Infrasound sensor array: Four infrasound sensors (frequency range 0.01–20 Hz) are arranged in a diamond pattern on the slope crest and shoulder, with a sensor spacing of 30 meters, accurately covering the potential slip surface exit area.

[0118] Beidou monitoring stations: Six Beidou monitoring points (supporting Beidou-3 B2b signals) are deployed, three of which are located at key locations along the projection line of the potential slip surface, two are located in the fracture development zone, and one is located in stable bedrock as a reference station.

[0119] (3) Communication network construction: A hybrid network of ZigBee (within 100m transmission distance) and LoRa (within 3km transmission distance) is adopted, and Beidou short message communication modules are deployed in areas without 4G signal coverage as emergency communication support.

[0120] 3. Data Acquisition and Processing

[0121] The system began operation on August 1, 2023, and continuously monitored data until November 30, 2023, during which time it experienced several heavy rainfall events. Data collection and processing included:

[0122] (1) Infrasound signal processing: The acquired raw infrasound signal is first processed by wavelet packet transform to remove environmental interference such as wind noise and mechanical vibration. Then, the short-time Fourier transform is calculated to extract the main frequency characteristics, frequency band energy ratio and spectral entropy value of the daily infrasound signal.

[0123] (2) BeiDou displacement calculation: The state threshold least squares algorithm is used to process BeiDou observation data. The plane positioning accuracy reaches 2.5mm, the elevation direction accuracy reaches 5.2mm, and the data integrity rate reaches 99.3%.

[0124] (3) Environmental parameter monitoring: Simultaneously monitor environmental parameters such as rainfall and temperature changes for subsequent dynamic threshold adjustment.

[0125] 4. Multimodal data fusion and early warning

[0126] The processed multi-source data is then input into an attention-based spatiotemporal convolutional network (AT-STCN) for fusion analysis.

[0127] (1) Model training: The AT-STCN model was pre-trained using historical monitoring data of the mining area (including data of 3 known landslide events). A multi-task learning framework was adopted, with the main task being stability classification and the auxiliary task being displacement prediction.

[0128] (2) Real-time analysis: The system analyzes the spatiotemporal correlation between infrasound characteristics and displacement sequence in real time. On October 15, the model monitored that the main frequency of the infrasound signal decreased from the normal 3-5 Hz to 1.2 Hz, while the spectral entropy value increased from 0.68 to 0.92, indicating that the rock mass fracturing intensified; at the same time, Beidou monitoring showed that accelerated displacement occurred in the middle and upper part of the slope, with the maximum displacement rate reaching 5.2 mm / d.

[0129] (3) Warning Trigger: At 8:30 on October 16, the fusion analysis results showed that the slope stability probability dropped to 0.63, and the system triggered a yellow warning and automatically increased the monitoring frequency to 1.5 times the benchmark. At 14:20 on October 17, the stability probability further dropped to 0.41, and the system triggered an orange warning, notifying the mine to stop operations in the affected area and take reinforcement measures.

[0130] 5. Verification of Early Warning Effectiveness

[0131] On November 5th, a partial collapse occurred on the slope, with a volume of approximately 800 m³. Thanks to advance warning, there were no casualties or equipment losses. This embodiment verifies the effectiveness of the invention:

[0132] (1) Early warning period: From the first yellow warning (October 16) to the collapse (November 5), the early warning period is 20 days, which is far longer than traditional monitoring methods (which can usually only give warnings 2 to 3 days in advance).

[0133] (2) Accuracy verification: The actual collapse location and the predicted slip surface match 85%, indicating the scientific nature of the layout plan.

[0134] (3) System reliability: During the period of multiple heavy rainfalls that caused communication interruptions, the key early warning information was successfully transmitted through Beidou short message, with a communication success rate of 100%.

[0135] (4) False alarm rate control: During the entire monitoring period, the system only issued one yellow warning (true warning) and one orange warning (true warning), and no false alarms occurred.

[0136] Table 1 below compares the early warning effect of the present invention with that of traditional methods in this embodiment:

[0137] Table 1: Comparison of the effects of the present invention and traditional methods

[0138]

[0139] 6. Conclusion: This embodiment fully verifies the excellent performance of the present invention in slope monitoring. Through deep fusion and intelligent analysis of multimodal data, early and accurate warnings of slope instability are achieved; through a scientific collaborative deployment scheme and reliable communication transmission, the integrity and reliability of the monitoring system are ensured; and through an adaptive early warning mechanism, the false alarm rate is effectively reduced. This invention provides an effective technical means for slope monitoring and has significant promotional value.

[0140] As can be seen from the above description, the above embodiments of the present invention achieve the following technical effects: Improved scientificity and targeting of monitoring: By using a collaborative deployment method based on a three-dimensional geological model of the slope and numerical simulation, the infrasound sensor array and Beidou monitoring station can accurately cover potential slip surfaces and key risk areas, solving the monitoring blind spot problem existing in traditional uniform deployment, and improving the spatial representativeness and risk capture capability of the monitoring network.

[0141] Significantly enhanced early warning capabilities: Utilizing the sensitivity of infrasound to micro-fractures in rock masses, combined with wavelet packet transform and spectral feature extraction algorithms, this invention can effectively identify early signs of slope instability. Practical applications show that this invention can detect slope anomalies 5–15 hours earlier than traditional displacement monitoring methods, providing ample time for emergency response.

[0142] Deep fusion of multimodal data: The innovative attention-based spatiotemporal convolutional network (AT-STCN) can dynamically weight the importance of different modal features, fully explore the spatiotemporal correlation between infrasound features and BeiDou displacement sequences, solve the information loss problem in simple data superposition or weighted average fusion, and improve the state recognition accuracy to over 97%.

[0143] High adaptability to complex environments: Employing a multi-modal transmission scheme combining ZigBee and LoRa hybrid self-organizing networks with BeiDou short message service, it effectively solves the data transmission problem in mountainous areas without public network coverage, ensuring the integrity of monitoring data and the timeliness of early warning information. Experiments show that the data upload integrity rate can reach 99.2% even in weak network environments.

[0144] High level of adaptability and intelligence: The dynamic threshold model can automatically adjust the warning threshold according to real-time environmental parameters (such as rainfall and ground motion), reducing the false alarm rate; the adaptive sampling mechanism intelligently adjusts the monitoring frequency according to the warning level, balancing the relationship between monitoring accuracy and energy consumption; the online model update mechanism enables the system to have continuous learning and optimization capabilities, making it more adaptable to specific slope environments.

[0145] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.

[0146] This invention is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this specification. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A multimodal slope condition monitoring method based on infrasound and BeiDou signals, characterized in that, include: Based on the potential slip surface determined by the three-dimensional geological model and numerical simulation of the slope, an array of infrasound sensors and a Beidou monitoring station were deployed in coordination at the top of the slope, the shoulder of the slope and the potential slip surface area. Infrasound sensor arrays collect infrasound signals from rock fractures, and BeiDou monitoring stations acquire millimeter-level displacement data; data transmission is achieved through a hybrid self-organizing network of ZigBee, LoRa, and BeiDou short messages. Denoising of the original signal was performed using wavelet packet transform. The original signal included infrasound signals from rock mass fracture and millimeter-level displacement data. The time spectrum was obtained by calculating the short-time Fourier transform of the original signal. The dominant frequency, frequency band energy ratio, spectral entropy, and azimuth features based on time difference of arrival were extracted. The state threshold least squares algorithm is used to process BeiDou signals; The infrasound feature vector is spatiotemporally aligned with the BeiDou displacement sequence. The infrasound feature vector is a set of multidimensional, qualitative features describing the microscopic damage inside the slope, while the BeiDou displacement sequence is direct, quantitative measurement data describing the overall macroscopic deformation of the slope. The data is then input into a spatiotemporal convolutional network based on an attention mechanism, where multimodal features are dynamically weighted and fused to output the probability of the slope's stable state. Based on the fused output, a dynamic threshold model is used to trigger tiered early warnings; the sampling frequency of the sensor array is adaptively adjusted according to the early warning level and the on-site conditions. Regularly feed back the newly added monitoring data and early warning results to form a closed loop, and perform incremental learning on the spatiotemporal convolutional network to optimize its performance in specific slope environments; The potential slip surface, determined based on the three-dimensional geological model of the slope and numerical simulation, involves the coordinated deployment of an infrasound sensor array and a BeiDou monitoring station at the slope top, slope shoulder, and potential slip surface area. Specifically, this includes: establishing a three-dimensional geological model of the slope through oblique photography using terminal equipment; analyzing stability under extreme conditions using numerical simulation methods to determine the potential slip surface; employing an equilateral triangle layout for the infrasound sensors, with the center located at the slope top or slope shoulder; and deploying BeiDou monitoring stations along the projection line of the potential slip surface and at key locations, including existing deformation areas, areas with weak geological structures, areas with geometric shapes and stress concentrations, and areas affected by engineering activities. The process of establishing a three-dimensional geological model of the slope through oblique photography using a terminal device specifically includes: High-resolution image data of the slope surface is acquired through terminal equipment. Using the bundle adjustment algorithm, an observation equation is established to minimize the error, resulting in the three-dimensional point cloud data of the slope. The formula is: ,in: The total number of 3D points participating in the adjustment calculation. For the first Two-dimensional coordinates of each observation point Let these be the coordinates of the observation point in three-dimensional space. For the camera's imaging model, including focal length Principal point coordinates Rotation matrix Translation vector ; The 3D point cloud data is triangulated using Delaunay triangulation to generate a triangular mesh, which is then used to construct a 3D geological model of the slope. The method of analyzing stability under extreme conditions using numerical simulation to determine potential slip surfaces specifically includes: The stress distribution of a slope under extreme working conditions is analyzed using the equations of elasticity; the formula is: ,in: It is normal stress. For shear stress, Let Lamé constant be . For strain components, For shear strain components; The safety factor of the slope is calculated and the potential slip surface is determined using the limit equilibrium method; the formula is: ,in: For safety reasons, The cohesion of soil and rock. The length of the slip surface, For normal stress, It is the internal friction angle. The weight of the sliding body. The angle of inclination of the sliding surface; Assuming the slip surface is a circular arc, use the coordinates of the center of the circle. and radius The parameterized slip surface is defined by the following formula: ,in: These are the two-dimensional planar coordinates of points on the sliding surface; The process involves spatiotemporally aligning the infrasound feature vector with the BeiDou displacement sequence. The infrasound feature vector is a set of multidimensional, qualitative features describing the microscopic damage within the slope, while the BeiDou displacement sequence is direct, quantitative measurement data describing the overall macroscopic deformation of the slope. Specifically, this includes: Assume the infrasound feature sequence is The BeiDou displacement sequence is Their sampling times are respectively and By aligning the time using interpolation methods, and assuming BeiDou data as the quasi-reference, the new aligned infrasound feature sequence is: , with BeiDou displacement sequence One-to-one correspondence; The input is fed into a spatiotemporal convolutional network based on an attention mechanism, dynamically weighted and fused with multimodal features, and outputs the probability of slope stability, specifically including: Infrasound characteristics and Beidou features The inputs are fed into a one-dimensional CNN and a two-dimensional CNN, respectively: ,in: and The feature extraction results are for infrasound and BeiDou, respectively. Define an attention module that obtains weights by calculating the correlation between features: ,in: It is a learnable parameter matrix. It is the weight of infrasound features. Given the length of the time series, the weights of the BeiDou features are: ; Dynamically weighted fusion of feature vectors ; The fused features are input into a classifier to output the slope state probability: ,in: This represents the probability of slope stability. These represent stable and unstable slopes, respectively. and It is the weight vector of the classifier.

2. The multimodal slope condition monitoring method based on infrasound and BeiDou signals according to claim 1, characterized in that, The calculation of the short-time Fourier transform of the original signal to obtain the time spectrum specifically includes: The signal is divided into frames, and a Hamming window is used. • Calculate the Fourier transform for each frame to obtain the time spectrum: in: Indicates the first Frame at frequency amplitude, It's a signal. This represents the discrete position of the signal in the time domain. It is the number of pixels per frame. It is the imaginary unit of complex numbers.

3. The multimodal slope condition monitoring method based on infrasound and BeiDou signals according to claim 2, characterized in that, The extraction of the dominant frequency, frequency band energy ratio, spectral entropy, and azimuth features based on time difference of arrival specifically includes: clock speed ; Bandwidth energy ratio ,in and These are the start and end points of the frequency band, respectively. Spectral entropy ,in ; Based on the azimuth characteristics of the time difference of arrival (TDOA), assuming that the two sensors receive the signals at the following times: and The speed of sound is Then the azimuth angle Calculated using the following formula: ,in It is the distance between the two sensors.

4. The multimodal slope condition monitoring method based on infrasound and BeiDou signals according to claim 3, characterized in that, The process of using the state threshold least squares algorithm to process BeiDou signals specifically includes: Assuming the BeiDou signal is The solution result is The algorithm is then expressed as: in, It is a design matrix. It is an adaptive convergence threshold used to prevent the algorithm from failing to converge during iteration; Solution results Calculated using the following formula: ,in: It is a matrix transpose, It is an identity matrix.

5. A multimodal slope condition monitoring system based on infrasound and BeiDou signals, characterized in that, A method for implementing a multimodal slope condition monitoring method based on infrasound and BeiDou signals as described in any one of claims 1-4 includes: The infrasound signal acquisition module deploys an array of infrasound sensors in key areas of the slope to capture infrasound signals generated by rock mass fracturing. The Beidou displacement monitoring module deploys Beidou monitoring stations along the projection line of the potential slip surface and at key locations on the slope to acquire millimeter-level displacement data and monitor the slope displacement in real time. The data transmission and processing module constructs a hybrid self-organizing network of ZigBee, LoRa, and BeiDou short message service to achieve reliable data transmission in complex environments. The feature extraction and noise reduction module is used to denoise infrasound signals using wavelet packet transform and to extract the dominant frequency, frequency band energy ratio, spectral entropy, and azimuth features based on time difference of arrival. The multimodal feature fusion module aligns the infrasound feature vector with the BeiDou displacement sequence in a spatiotemporal manner. The infrasound feature vector is a set of multidimensional and qualitative features describing the microscopic damage inside the slope, while the BeiDou displacement sequence is direct and quantitative measurement data describing the overall macroscopic deformation of the slope. The input is a spatiotemporal convolutional network based on an attention mechanism, which dynamically weights and fuses the multimodal features to output the probability of the slope's stable state. The slope condition classification and early warning module, based on fusion output, uses a dynamic threshold model to trigger graded early warnings and adaptively adjusts the sensor array sampling frequency according to the early warning level and the site conditions. The online model update module regularly feeds back new monitoring data and early warning results to form a closed loop, enabling incremental learning of the spatiotemporal convolutional network and optimizing its performance in specific slope environments.

Images