A vertical retaining structure displacement automatic monitoring device and a monitoring method thereof

Abstract

This invention discloses an automatic displacement monitoring device and method for vertical retaining structures, belonging to the field of automatic displacement monitoring technology. It includes: a data acquisition unit for configuring monitoring nodes, acquiring sensor data from the top of the retaining wall, and preprocessing the sensor data to obtain effective tilt angle data; a displacement conversion unit for using a dynamic Kalman filter algorithm to identify and optimize the attitude angle, and calculating the displacement value of the vertical retaining structure based on the vertical distance from the bottom to the top of the retaining wall; a monitoring and early warning unit for uploading the vertical retaining structure displacement value to a remote data platform, constructing a monitoring network, analyzing and predicting the deformation trend of the retaining wall, and triggering an early warning based on the analysis results; and a battery management unit for providing battery status monitoring and management. This invention can accurately sense and monitor the displacement of the top of the retaining wall through tilt angle data, and employs an attitude solver and a dynamic Kalman filter algorithm to ensure comprehensive and accurate monitoring.

Description

Technical Field

[0001] This invention relates to the field of automatic displacement monitoring technology, and more specifically, to an automatic displacement monitoring device and method for vertical retaining structures. Background Technology

[0002] With the continuous development of urban construction, vertical retaining structures are widely used in civil engineering. In order to ensure the safety and stability of the structure, it is particularly important to monitor the displacement and deformation of the retaining wall in real time. However, existing monitoring methods usually require complex manual measurement or rely on fixed sensors, which cannot meet the needs of real-time, accurate and long-term online monitoring. Furthermore, traditional monitoring systems lack intelligent processing capabilities, cannot automatically analyze data and issue early warnings, and have very limited support for the needs of long-term online monitoring.

[0003] Traditional manual measurement methods require significant manpower and time. Furthermore, due to complex and variable field conditions, measurement results are often affected by environmental factors, leading to poor data accuracy and consistency. Manual measurement can only provide data at specific moments, failing to achieve continuous monitoring. This allows potential problems to be overlooked until they develop into serious safety hazards. While monitoring methods relying on fixed sensors reduce some manual intervention, they lack intelligent analysis capabilities, cannot automatically process data or provide early warnings, and have limited support for long-term online monitoring. Therefore, there is an urgent need for an automated monitoring device that can achieve intelligent monitoring, real-time data transmission, remote management, and possesses high precision and long-endurance capabilities.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] To address the problems in related technologies, this invention proposes an automatic displacement monitoring device and method for vertical retaining structures, thereby overcoming the aforementioned technical problems existing in the prior art.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows:

[0007] In a first aspect, the present invention provides an automatic displacement monitoring device for vertical retaining structures, comprising:

[0008] The data acquisition unit is used to configure monitoring nodes, acquire sensor data monitored at the top of the retaining wall, and preprocess the sensor data to obtain effective tilt angle data.

[0009] The displacement conversion unit is used to identify and optimize the attitude angle using the dynamic Kalman filter algorithm, and calculate the displacement value of the vertical retaining structure by combining the vertical distance from the bottom to the top of the retaining wall.

[0010] The monitoring and early warning unit is used to upload the displacement values ​​of vertical retaining structures to a remote data platform, build a monitoring network, analyze and predict the deformation trend of the retaining wall, and trigger early warnings based on the analysis results.

[0011] Furthermore, the data acquisition unit includes:

[0012] The data acquisition module is used to acquire real-time sensor data from various types of sensors; the sensor data includes gyroscope data, accelerometer data, and geomagnetic sensor data.

[0013] The gyroscope preprocessing module is used to remove static offset errors from gyroscope data based on an adaptive high-pass digital filter to obtain effective gyroscope data.

[0014] The acceleration preprocessing module is used to smooth the accelerometer data in the time domain based on the moving average filter, and to separate the smoothed accelerometer data using a second-order Butterworth low-pass filter to obtain effective accelerometer data.

[0015] The geomagnetic data preprocessing module is used to eliminate noise from geomagnetic sensor data based on a windowed median filter, identify the environmental magnetic field interference frequency band through fast Fourier transform frequency domain analysis, and suppress noise in the interference frequency band by combining a notch filter to obtain effective magnetic sensor data.

[0016] The data integration module is used to integrate the processed effective gyroscope data, effective accelerometer data, and effective magnetic sensor data as the effective tilt angle data of the retaining wall.

[0017] Furthermore, frequency bands of environmental magnetic field interference can be identified through fast Fourier transform frequency domain analysis, including:

[0018] The noise-reduced geomagnetic sensor data is normalized, and the normalized geomagnetic sensor data is transformed in the frequency domain using the fast Fourier transform function to obtain the geomagnetic sensor data in the frequency domain.

[0019] Based on the geomagnetic sensor data represented in the frequency domain, obtain the frequency corresponding to each complex point, and calculate the amplitude spectrum of each frequency point;

[0020] Based on the amplitude spectrum of each frequency point, an amplitude spectrum diagram is plotted, and the frequency corresponding to the peak point with higher amplitude is obtained as the frequency band of environmental magnetic field interference.

[0021] Furthermore, the displacement transformation unit includes:

[0022] The error accumulation processing module is used to obtain the tilt optimization angle of the previous moment, and calculate the tilt optimization angle of the current moment by combining it with the effective gyroscope data of the current moment.

[0023] The optimized angle correction module is used to correct the effective accelerometer data using Kalman gain, and combined with the tilt optimization calculation at the current moment, the optimized attitude angle is obtained through dynamic weighting calculation.

[0024] The displacement conversion module is used to obtain the vertical distance from the bottom to the top of the retaining wall measured by the laser rangefinder. By optimizing the attitude angle conversion, the displacement value of the vertical retaining structure is calculated.

[0025] Furthermore, the formula for calculating the tilt optimization angle at the current moment is:

[0026]

[0027] In the formula, This indicates the optimal tilt angle at time k; ω represents the tilt optimization angle at the previous moment; k This represents the gyroscope angular velocity corresponding to the valid gyroscope data; Δt represents the time interval.

[0028] The formula for calculating the optimized attitude angle is:

[0029]

[0030] In the formula, Indicates the optimized attitude angle; Z accel，k K represents the tilt angle of the accelerometer at time k; k Indicates the Kalman gain at time k;

[0031] The formula for calculating the displacement of a vertical retaining structure is:

[0032]

[0033] In the formula, ΔL represents the displacement value of the vertical retaining structure; h represents the vertical distance from the bottom to the top of the retaining wall.

[0034] Furthermore, the monitoring and early warning unit includes:

[0035] The edge communication computing unit is used to aggregate edge computing data to a remote data platform using a wireless communication protocol, thereby realizing data aggregation of distributed monitoring nodes;

[0036] The digital modeling and display unit is used to construct a three-dimensional virtual model of the retaining wall. By inputting the vertical retaining displacement values ​​of each monitoring node, the deformation state of the physical entity is mapped in real time.

[0037] The spatiotemporal correlation prediction unit is used to predict the future deformation trend of retaining walls and identify potential risk points based on historical and real-time data through spatiotemporal information fusion and mining.

[0038] The early warning decision support unit is used to set dynamic multi-level early warning strategies, trigger multi-level early warnings based on analysis and prediction results, and generate emergency response and maintenance suggestions.

[0039] Furthermore, based on historical and real-time data, through spatiotemporal information fusion and mining, the future deformation trend of retaining walls is predicted and potential risk points are identified, including:

[0040] The historical and real-time vertical retaining structure displacement values, sensor data and environmental data of the retaining wall are obtained. All monitoring nodes are mapped to a unified spatiotemporal coordinate system. The Pearson correlation coefficient is used to quantify the spatial correlation between each monitoring node, and strongly correlated neighbor nodes are selected to construct an adjacency matrix.

[0041] Based on the adjacency matrix, the features of adjacent monitoring nodes are aggregated through a multi-layer graph convolutional network, the adjacency matrix is ​​normalized, and the node features are passed layer by layer to capture the synergistic effect of local deformation of the retaining wall.

[0042] The spatially aggregated features are input into a bidirectional gated recurrent network to model the long-term dependencies of displacement sequences along the time axis. A spatiotemporal cross-attention mechanism is introduced to calculate the association weights between spatial features and temporal hidden states, generating a coupled state vector that integrates spatiotemporal dynamics.

[0043] Based on the spatiotemporal fusion characteristics, a time-varying Gaussian analysis model is constructed by using the mean and variance of the displacement prediction values ​​output by the linear layer. The maximum likelihood loss function is used to optimize the model parameters, and the time-varying Gaussian analysis model is used to predict the mean and variance of the deformation at future moments.

[0044] Based on the mean and variance prediction results, the deformation trend and potential risk points of the retaining wall are assessed, and an online learning mechanism is introduced to optimize the model using the latest monitoring data.

[0045] Furthermore, the parameter prediction formula for the time-varying Gaussian analysis model is as follows:

[0046]

[0047] In the formula, μ t W represents the mean value of the predicted displacement of the vertical retaining structure. μ c represents the linear layer weights of the mean. t b represents the spatiotemporal fusion feature vector; μ This represents the mean bias term; W represents the variance of the predicted displacement values ​​of vertical retaining structures. σ Linear layer weights representing variance; b σ This represents the variance bias term;

[0048] The formula for the maximum likelihood loss function is:

[0049]

[0050] In the formula, L represents the maximum likelihood loss function; yt represents the actual observed displacement of the vertical retaining structure at time t; and T represents the total length of the time series.

[0051] Furthermore, based on the mean and variance prediction results, the deformation trend and potential risk points of the retaining wall are assessed, and an online learning mechanism is introduced to optimize the model using the latest monitoring data, including:

[0052] The future deformation of the retaining wall follows a Gaussian distribution. Based on the mean and variance prediction results, the probability interval of the future deformation is determined, and risk points are marked according to the deformation of each monitoring node.

[0053] Calculate the mutual information value between environmental factors and the displacement value of vertical retaining structures, dynamically adjust the input weights of the time-varying Gaussian distribution model, and construct an environment-threshold mapping table to match the current environmental adjustment standard;

[0054] Set an update cycle to periodically collect the latest monitoring data, calculate the loss value of the new data using the maximum likelihood loss function, and update the parameters of the time-varying Gaussian distribution model if the loss value is greater than the preset threshold.

[0055] The mean prediction results are mapped to a three-dimensional virtual model to visualize the risk distribution of the retaining wall.

[0056] Secondly, the present invention also provides an automatic monitoring method for the displacement of vertical retaining structures, the monitoring method comprising:

[0057] Distributed monitoring nodes are configured on the top of the retaining wall to acquire sensor data from the top of the retaining wall, and the sensor data is preprocessed to obtain effective tilt angle data.

[0058] Based on the effective tilt angle data, the dynamic Kalman filter algorithm is used to identify and optimize the attitude angle, and combined with the vertical distance from the bottom to the top of the retaining wall, the displacement value of the vertical retaining structure is calculated.

[0059] The displacement values ​​of vertical retaining structures calculated from all monitoring nodes are uploaded to a remote data platform to build a monitoring network, analyze and predict the deformation trend of the retaining wall, and trigger early warnings based on the analysis results.

[0060] The beneficial effects of this invention are as follows:

[0061] 1. This invention accurately senses and monitors the displacement of the top of the retaining wall using tilt angle data, overcoming the shortcomings of traditional monitoring methods that cannot obtain displacement data in real time and accurately. It employs an attitude solver and a dynamic Kalman filter algorithm, and transmits and controls data remotely through a remote data platform, thereby achieving automatic monitoring with more comprehensive and accurate data, avoiding complex manual measurements. Simultaneously, edge computing is used for real-time noise reduction and spatiotemporal alignment of the data, significantly improving data reliability. A spatiotemporal prediction model based on graph convolutional networks and recurrent neural networks can dynamically analyze the overall deformation trend and local risk points of the retaining wall, quantifying uncertainty through probabilistic prediction and triggering tiered early warnings. It supports long-term continuous operation in outdoor environments without power grids, and combined with a self-inspection mechanism and remote maintenance interface, it significantly reduces the frequency of manual inspections. The modular design balances deployment convenience and system scalability, providing a high-precision, fully automatic, and low-maintenance closed-loop solution for engineering safety monitoring.

[0062] 2. This invention achieves accurate monitoring of the displacement of vertical retaining structures by acquiring the tilt angle data of the retaining structure, ensuring accurate measurement under complex working conditions. At the same time, it uses digital filtering technology to preprocess the tilt angle data, which can effectively remove noise and interference from the data and further improve the measurement accuracy. Combined with the dynamic Kalman filter algorithm, it can accurately calculate the current motion attitude angle of the module in complex dynamic environments to achieve high-precision monitoring.

[0063] 3. This invention uploads the displacement values ​​and motion angles of vertical retaining structures to a remote data platform, transmitting monitoring data to a remote server in real time and stably. Engineering management personnel can access the monitoring data anytime, anywhere via computers or mobile phones, without needing to be physically present on-site. Furthermore, the remote data platform has intelligent risk prediction and early warning functions; once the monitoring data exceeds a preset safety threshold, it will immediately notify relevant personnel via SMS, email, and other means to facilitate timely action. The remote data platform also allows management personnel to remotely adjust the device's parameter settings, enabling refined management of the monitoring process and further enhancing the level of intelligent engineering management. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is a schematic diagram of an automatic displacement monitoring device for vertical retaining structures according to an embodiment of the present invention.

[0066] Figure 2 This is one of the schematic diagrams illustrating a specific implementation of an automatic displacement monitoring device for vertical retaining structures according to an embodiment of the present invention;

[0067] Figure 3 This is a second schematic diagram illustrating a specific implementation of an automatic displacement monitoring device for vertical retaining structures according to an embodiment of the present invention;

[0068] Figure 4 This is a flowchart of an automatic displacement monitoring method for vertical retaining structures according to an embodiment of the present invention.

[0069] In the picture:

[0070] 1. Data acquisition unit; 2. Displacement conversion unit; 3. Monitoring and early warning unit; 4. Battery management unit. Detailed Implementation

[0071] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0072] According to an embodiment of the present invention, an automatic displacement monitoring device for vertical retaining structures is provided.

[0073] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the automatic displacement monitoring device for vertical retaining structures according to an embodiment of the present invention includes: a data acquisition unit 1, a displacement conversion unit 2, a monitoring and early warning unit 3, and a battery management unit 4.

[0074] Data acquisition unit 1 is used to configure monitoring nodes, acquire sensor data monitored at the top of the retaining wall, and preprocess the sensor data to obtain effective tilt angle data.

[0075] In the description of this invention, the data acquisition unit 1 includes: a data acquisition module, a gyroscope preprocessing module, an acceleration preprocessing module, a geomagnetic data preprocessing module, and a data integration module.

[0076] The data acquisition module is used to acquire sensor data monitored in real time by various types of sensors; the sensor data includes gyroscope data, accelerometer data and geomagnetic sensor data.

[0077] Specifically, high-precision gyroscopes, accelerometers, and geomagnetic sensors are used to perceive the tilt angle data of the top of the retaining wall in real time, namely gyroscope data, accelerometer data, and geomagnetic sensor data. Among them, the gyroscope data is used to measure the dynamic angular velocity; the accelerometer data calculates the static tilt angle through the gravity component; and the geomagnetic sensor data is used for orientation angle calibration. The measurement accuracy is 0.1 degrees in dynamic environment and 0.05 degrees in static environment.

[0078] For example, when heavy rain causes the soil to loosen and the top of the wall begins to slowly tilt to the right, the gyroscope immediately senses "0.01 degrees to the right per second", the accelerometer detects the shift in the direction of gravity and confirms "static tilt of 0.5 degrees", the geomagnetic sensor finds that "the overall direction is 0.5 degrees east of the original record", and the geomagnetic field sensor locks the absolute direction to detect the absolute orientation of the retaining wall.

[0079] The gyroscope preprocessing module is used to remove static offset errors from gyroscope data based on an adaptive high-pass digital filter, thereby obtaining effective gyroscope data.

[0080] Specifically, the gyroscope data is processed by an adaptive high-pass digital filter to remove static offset errors and retain the effective signal of dynamic angular velocity. This effectively smooths out random noise, reduces the fluctuation of angular velocity data, improves measurement stability, and ensures that tilt measurement noise is suppressed to within 0.1 degrees in dynamic environments.

[0081] The accelerometer preprocessing module is used to smooth the accelerometer data in the time domain based on the moving average filter, and then use a second-order Butterworth low-pass filter to separate the smoothed accelerometer data to obtain effective accelerometer data.

[0082] Specifically, a moving average filter is used to calculate the average value of the accelerometer data as the current acceleration value. The raw data is smoothed in the time domain to filter out short-term high-frequency interference. Furthermore, a second-order Butterworth low-pass filter is used with a cutoff frequency of 5Hz to effectively separate the gravity component from vibration noise and ensure that the static tilt angle calculation accuracy reaches 0.05 degrees.

[0083] The geomagnetic data preprocessing module is used to eliminate noise from geomagnetic sensor data based on a windowed median filter, identify the environmental magnetic field interference frequency band through fast Fourier transform frequency domain analysis, and suppress the noise in the interference frequency band by combining a notch filter to obtain effective magnetic sensor data.

[0084] Specifically, a windowed median filter is used to eliminate impulse noise from the geomagnetic sensor data, and the frequency bands of environmental magnetic field interference are identified through fast Fourier transform frequency domain analysis. Notch filters are used to suppress noise in specific frequency bands, while retaining the low-frequency geomagnetic signal required for azimuth calibration.

[0085] In the description of this invention, identifying environmental magnetic field interference frequency bands through fast Fourier transform frequency domain analysis includes:

[0086] Step S101: Normalize the noise-reduced geomagnetic sensor data, and use the fast Fourier transform function to perform frequency domain transformation on the normalized geomagnetic sensor data to obtain the frequency domain representation of the geomagnetic sensor data.

[0087] Step S102: Based on the geomagnetic sensor data represented in the frequency domain, obtain the frequency corresponding to each complex point and calculate the amplitude spectrum of each frequency point.

[0088] Step S103: Based on the amplitude spectrum of each frequency point, draw an amplitude spectrum diagram and obtain the frequency corresponding to the peak point with higher amplitude as the environmental magnetic field interference frequency band.

[0089] It should be noted that after removing impulse noise using a windowed median filter, the noise-reduced geomagnetic sensor data is normalized, mapping the data to a specific range from -1 to 1 for easier subsequent analysis and processing. The Fast Fourier Transform (FFT) function is then used to perform a frequency domain transformation on the preprocessed data, obtaining a complex frequency domain representation. Each complex point corresponds to a specific frequency. Then, the amplitude spectrum of each frequency point is calculated, i.e., the magnitude of the FFT result is taken to determine the intensity of different frequency components. The amplitude spectrum reflects the magnitude of the signal at each frequency. By plotting the amplitude spectrum, the horizontal axis represents frequency, and the vertical axis represents the amplitude at the corresponding frequency. The frequency corresponding to the peak point with the higher amplitude is the interference frequency band.

[0090] The data integration module is used to integrate the processed effective gyroscope data, effective accelerometer data, and effective magnetic sensor data as the effective tilt angle data of the retaining wall.

[0091] Displacement conversion unit 2 is used to identify and optimize the attitude angle using a dynamic Kalman filter algorithm, and calculate the displacement value of the vertical retaining structure by combining the vertical distance from the bottom to the top of the retaining wall.

[0092] In the description of this invention, the displacement conversion unit 2 includes: an error accumulation processing module, an optimized angle correction module, and a displacement conversion module.

[0093] The error accumulation processing module is used to obtain the tilt optimization angle of the previous moment, and calculate the tilt optimization angle of the current moment by combining it with the effective gyroscope data of the current moment.

[0094] In the description of this invention, the tilt angle data is preprocessed through data processing and filtering to obtain the angle measured by the gyroscope. The gyroscope is good at capturing dynamic changes and dominating dynamic response, and has high short-term prediction accuracy, but it will accumulate drift error over the long term. Therefore, it is necessary to calculate the optimized tilt angle at the current moment. The formula for calculating the optimized tilt angle at the current moment is:

[0095]

[0096] In the formula, This indicates the optimal tilt angle at time k; ω represents the tilt optimization angle at the previous moment; k This represents the gyroscope angular velocity corresponding to the valid gyroscope data; Δt represents the time interval.

[0097] The optimized angle correction module is used to correct the effective accelerometer data using Kalman gain, and combined with the tilt optimization calculation at the current moment, the optimized attitude angle is obtained through dynamic weighting calculation.

[0098] Specifically, the optimized angle correction module has a built-in attitude solver that, combined with a dynamic Kalman filter algorithm, can accurately calculate the current motion attitude angle of the module in complex dynamic environments and convert the attitude angle into the displacement value of the top of the retaining wall.

[0099] The dynamic Kalman filter algorithm estimates the noise covariance matrix of the gyroscope, accelerometer and geomagnetic sensor in real time, and dynamically adjusts the fusion weights of the multi-source data to obtain the optimized attitude angle.

[0100] The noise covariance matrix includes a process noise covariance matrix Q, which describes the uncertainty of gyroscope measurements, and an observation noise covariance matrix R, which describes the uncertainty of accelerometer and geomagnetic sensor measurements. Both the process noise covariance matrix Q and the observation noise covariance matrix R, which describes the uncertainty of geomagnetic sensor measurements, are fixed values. The observation noise covariance matrix R, which describes the uncertainty of accelerometer measurements, is adjusted in real time according to the environmental vibration intensity, as shown in the following formula:

[0101]

[0102] In the formula, R a R represents the observation noise covariance matrix of the accelerometer data. base γ represents a fixed static reference noise value; α represents a fixed sensitivity value; rms This represents the root mean square value of the acceleration vibration;

[0103] The noise covariance matrix determines the weight of the measurement data in the state update through the Kalman gain K. The formula is as follows:

[0104]

[0105] In the formula, H represents the observation matrix H=

[10] ; R represents the observation noise covariance matrix; Q represents the process noise covariance matrix.

[0106] α under high vibration environment rms >0.2g, resulting in a Kalman gain K k Reduce the weight of accelerometer measurements to increase the weight of gyroscope measurements in order to optimize attitude angles.

[0107] In the description of this invention, the data correction stage incorporates Kalman gain to correct the accelerometer measurements. The accelerometer and geomagnetic sensor provide static references and orientation information, but these are susceptible to interference in dynamic environments. Therefore, they need to be dynamically weighted to obtain optimized attitude angles, which are affected by vibration and noise. The formula for calculating the optimized attitude angle is as follows:

[0108]

[0109] In the formula, Indicates the optimized attitude angle; Z accel，k K represents the tilt angle of the accelerometer at time k; k This represents the Kalman gain at time k.

[0110] The displacement conversion module is used to obtain the vertical distance from the bottom to the top of the retaining wall measured by the laser rangefinder. By optimizing the attitude angle conversion, the displacement value of the vertical retaining structure is calculated.

[0111] In the description of this invention, the formula for calculating the displacement value of a vertical retaining structure is as follows:

[0112]

[0113] In the formula, ΔL represents the displacement value of the vertical retaining structure; h represents the vertical distance from the bottom to the top of the retaining wall.

[0114] In the description of this invention, the monitoring and early warning unit 3 includes: an edge communication computing unit, a digital modeling and display unit, a spatiotemporal correlation prediction unit, and an early warning decision support unit.

[0115] The edge communication computing unit is used to aggregate edge computing data to a remote data platform using wireless communication protocols, thereby realizing data aggregation of distributed monitoring nodes.

[0116] Specifically, this unit integrates low-power wide-area network (LoRa) and cellular network (4G / 5G) dual-mode communication protocols to achieve bidirectional data transmission between monitoring nodes and remote platforms.

[0117] In outdoor scenarios without stable network coverage, the LoRa protocol is prioritized for long-distance, low-speed data uploads. When abnormal displacement is detected or an alert is triggered, the system automatically switches to a high-speed cellular network to transmit critical data. The edge computing terminal incorporates a lightweight data compression algorithm (such as Huffman coding) to increase the original data compression rate to over 70%, and uses an encryption chip (supporting national cryptographic algorithms) to ensure transmission security. During network outages, the local storage module can cache at least 30 days of data, which is then retransmitted via a breakpoint resume mechanism after network recovery, ensuring continuous monitoring.

[0118] For example, in the monitoring of retaining walls in mountainous areas, this unit can aggregate displacement and tilt data from multiple nodes in real time, compress them, and upload them to the cloud, significantly reducing communication energy consumption and latency.

[0119] The digital modeling and display unit is used to construct a three-dimensional virtual model of the retaining wall. By inputting the vertical displacement values ​​of each monitoring node, the deformation state of the physical entity can be mapped in real time.

[0120] Specifically, based on the fusion engine of Building Information Modeling (BIM) and Geographic Information System (GIS), this unit maps the displacement values ​​of each monitoring node to the corresponding positions in the three-dimensional virtual model, generating a deformation heat map with millimeter-level accuracy.

[0121] The model supports dynamic incremental updates, reconstructing only the local meshes where deformation has occurred, thus reducing computational resource consumption. By combining it with UAV oblique photogrammetry data, a comparison between the actual terrain and the virtual deformation trend can be overlaid and displayed.

[0122] For example, when the vertical displacement of a section of retaining wall exceeds 5mm, the model automatically marks that area in red and generates a historical displacement curve. Furthermore, it supports augmented reality (AR) interaction, allowing engineers to scan the site with a mobile device to view deformation data superimposed on the real-world scene, aiding in the rapid location of risk points.

[0123] The spatiotemporal correlation prediction unit is used to predict the future deformation trend of retaining walls and identify potential risk points based on historical and real-time data through spatiotemporal information fusion and mining.

[0124] It should be noted that spatiotemporal fusion consists of spatial fusion and temporal fusion. Spatial fusion is based on graph convolutional networks (GCN) to adaptively aggregate displacement, strain and other features of adjacent monitoring points to capture local synergistic effects in the overall deformation of the retaining wall. Temporal fusion uses bidirectional gated recurrent units (Bi-GRU) to model the long-term dependencies of displacement sequences along the time axis and combines an attention mechanism to achieve dynamic coupling of spatiotemporal features.

[0125] Assuming the monitoring data contains time-varying Gaussian noise, the model outputs the mean and variance of displacement predictions, replacing traditional point predictions. It quantifies prediction uncertainty through a maximum likelihood estimation loss function, improving robustness to noise. Leveraging a full-process training mechanism, the hidden state at each time step participates in loss calculation during model training, avoiding information loss caused by traditional truncated training and strengthening the model's overall learning ability regarding the deformation accumulation process.

[0126] In displacement monitoring of vertical retaining structures, spatial correlation modeling is used, and Geometric Networks (GCNs) are employed to dynamically assign weights to features from multiple measuring points, thus addressing the problem of traditional single-point models neglecting spatial correlation. For example, when a section of the retaining wall experiences cascading deformation due to local settlement, GCNs automatically enhance the feature transmission of relevant nodes through adjacency matrix weights, accurately capturing the deformation propagation path.

[0127] For time-series dependencies and long-term predictions, Bi-GRU and full-process training are used to integrate historical displacement data with external environmental parameters (such as rainfall and groundwater levels) to address the problem of inaccurate predictions of long-term deformation trends by traditional models. For example, the soil softening effect caused by continuous heavy rainfall can be predicted 72 hours in advance through time-series dependencies. In addition, to improve noise robustness, the probabilistic prediction framework directly models sensor noise in the monitoring data (such as fiber optic wavelength drift and visual target recognition errors), quantifies the prediction reliability through variance output, and avoids false alarms caused by noise interference.

[0128] In the description of this invention, predicting the future deformation trend of retaining walls and identifying potential risk points based on historical and real-time data through spatiotemporal information fusion and mining includes:

[0129] Step S31: Obtain historical and real-time vertical retaining structure displacement values, sensor data, and environmental data (such as rainfall, groundwater level, temperature, etc.) of the retaining wall, map all monitoring nodes to a unified spatiotemporal coordinate system, and use the Pearson correlation coefficient to quantify the spatial correlation between each monitoring node, filter strongly correlated neighbor nodes, and construct an adjacency matrix.

[0130] Specifically, a sensor network deployed on the retaining wall collects real-time data on vertical displacement, strain, and tilt angle, simultaneously accessing environmental parameters from meteorological stations (rainfall) and hydrological monitoring equipment (groundwater level, temperature). Historical data is extracted from a database, covering monitoring records for at least one complete hydrological year (12 months).

[0131] Missing values ​​are handled by using linear interpolation to fill in short-term missing values ​​(such as temporary sensor malfunctions), while long-term missing values ​​are marked as invalid. Noise filtering applies a moving average filter to the displacement data (window width Δt = 1 hour), ultimately normalizing data of different dimensions such as displacement and rainfall to the interval [0, 1].

[0132] All monitoring nodes are equipped with BeiDou RTK positioning modules (accuracy ±1cm) to obtain latitude and longitude coordinates (EPSG: 4326) and convert them into a local engineering coordinate system (such as UTM projection); and the NTP protocol is used to synchronize the clocks of each node to ensure that the data timestamp error is <1ms and to eliminate timing misalignment caused by asynchronous sampling.

[0133] For each pair of measuring points (i,j), calculate the Pearson correlation coefficient of its historical displacement sequence, and set a threshold ρ. thres =0.8, if ρ ij ≥ρ thres The points are identified as strongly correlated neighbors. For each measurement point, the top k==5 neighbors with the highest correlation coefficients are retained.

[0134] Step S32: Based on the adjacency matrix, aggregate the features of adjacent monitoring nodes through a multi-layer graph convolutional network, normalize the adjacency matrix, and pass the node features layer by layer to capture the synergistic effect of local deformation of the retaining wall.

[0135] Specifically, by aggregating features from adjacent monitoring nodes through a multi-layer graph convolutional network and normalizing the adjacency matrix, the feature propagation bias caused by differences in node degree in traditional graph convolution can be resolved, avoiding gradient explosion or vanishing gradients. An identity matrix IN needs to be added to the diagonal of the original adjacency matrix A. This ensures that each node retains its own features during feature aggregation (e.g., the displacement value of a measurement point is not completely overwritten by its neighbors). Then, the adjacency matrix is ​​normalized using the degree matrix D (where the diagonal elements are the degree of the nodes, i.e., the number of neighbors). If measurement point i has 3 neighbors and measurement point j has 5 neighbors, the feature contribution weights of the two will be balanced after normalization, preventing nodes with more neighbors from dominating the entire network.

[0136] By stacking multiple GCN layers, the receptive field of features is gradually expanded, capturing cooperative deformation from local to global perspectives. Single-layer GCN operations include the following aspects:

[0137] Input: Node feature matrix (N is the number of measurement points, F is the feature dimension);

[0138] Output: Updated feature matrix H (l+1) The formula is: H (l+1) =σ(AH (l) W (l) );

[0139] in, Let represent the trainable weight matrix of the l-th layer.

[0140] Additionally, for multi-layered stacking examples: First layer: Aggregate direct neighbor features to capture local deformation near the measuring point (e.g., synchronous displacement changes of adjacent points caused by cracks in a section of wall). Second layer: Indirectly aggregate through neighbors' neighbors to perceive deformation correlations over a larger range (e.g., upstream displacement triggering downstream cascading settlement).

[0141] By leveraging the coupling effect of adjacency matrix weights and feature propagation, node features are passed layer by layer to capture the synergistic effect of local deformation in retaining walls. This can be used to demonstrate local heave deformation in a retaining wall area caused by a sudden increase in earth pressure. The implementation process includes:

[0142] 1. Adjacency matrix weights: Strongly correlated neighbors (ρ ij Weights greater than 0.8 are higher. For example, if measuring points A and B are physically adjacent and have the same displacement trend, the weight a is higher. AB =0.9.

[0143] 2. Feature aggregation: The features (displacement, strain) of measuring point A are transferred to the feature expression of measuring point B with high weight, which enhances the correlation between the two in the coordinated deformation.

[0144] 3. Multi-layer transmission: The second layer GCN further transmits the characteristics of measuring point B to the more distant measuring point C, thereby spreading the influence of local deformation to the entire monitoring network.

[0145] Step S33: Input the spatially aggregated features into a bidirectional gated recurrent network, model the long-term dependency of the displacement sequence along the time axis, and introduce a spatiotemporal cross-attention mechanism to calculate the association weights between spatial features and temporal hidden states, generating a coupled state vector that integrates spatiotemporal dynamics.

[0146] Specifically, the Bi-Gated Recurrent Network (Bi-GRU) models temporal dependencies, which can capture the long-term positive and negative dependencies of displacement sequences and solve the gradient vanishing problem of traditional RNNs.

[0147] By dynamically associating spatial features with temporal latent states, the feature representation of key spatiotemporal nodes is enhanced. The computational steps include the following aspects:

[0148] 1. Feature mapping: Map spatial features and temporal latent states to query, key, and value spaces respectively.

[0149] 2. Attention Score: Calculate the association weight of each measurement point i at time t. The calculation formula is as follows:

[0150]

[0151] In the formula, α ti d represents the association weight of each monitoring node i at time t; k Q represents the attention dimension;i K represents the query vector of the monitoring node. t The key vector representing time t.

[0152] 3. Feature weighted fusion: Generates a spatiotemporally coupled feature vector, calculated using the following formula:

[0153]

[0154] In the formula, c t V represents the spatiotemporal coupling eigenvector; i This represents the value vector of monitoring node i.

[0155] Example: When a measuring point experiences a surge in displacement due to a rise in groundwater level, its correlation weight α... ti Increase the size of the node to enhance its impact on the overall prediction.

[0156] Step S34: Based on the spatiotemporal fusion characteristics, a time-varying Gaussian analysis model is constructed by using the mean and variance of the displacement prediction values ​​output by the linear layer. The model parameters are optimized using the maximum likelihood loss function, and the mean and variance of the deformation at future moments are predicted using the time-varying Gaussian analysis model.

[0157] In the description of this invention, the parameter prediction formula of the time-varying Gaussian analysis model is as follows:

[0158]

[0159] In the formula, μ t W represents the mean value of the predicted displacement of the vertical retaining structure. μ The linear layer weights represent the mean; c t b represents the spatiotemporal fusion feature vector; μ This represents the mean bias term; W represents the variance of the predicted displacement values ​​of vertical retaining structures. σ Linear layer weights representing variance; b σ This represents the variance bias term.

[0160] The formula for the maximum likelihood loss function is:

[0161]

[0162] In the formula, L represents the maximum likelihood loss function; yt represents the actual observed displacement of the vertical retaining structure at time t; and T represents the total length of the time series.

[0163] Step S35: Based on the mean and variance prediction results, assess the deformation trend and potential risk points of the retaining wall, and introduce an online learning mechanism to optimize the model using the latest monitoring data.

[0164] In the description of this invention, the deformation trend and potential risk points of the retaining wall are evaluated based on the mean and variance prediction results, and an online learning mechanism is introduced to optimize the model using the latest monitoring data, including:

[0165] Step S351: The future deformation of the retaining wall follows a Gaussian distribution. Based on the mean and variance prediction results, the probability interval of the future deformation is determined, and risk points are marked according to the deformation of each monitoring node.

[0166] Specifically, the future deformation follows a Gaussian distribution. The model outputs probability distribution parameters rather than single numerical values. For example, when the model predicts that μt = 5.2 mm at a certain measurement point, At that time, its 95% confidence interval is 5.2 ± 1.96 × 0.8 mm, i.e., [3.6 mm, 6.8 mm]. If the upper limit of this interval exceeds the preset safety threshold (e.g., 6 mm), it is identified as a potential risk point. The risk level is dynamically classified based on the closeness of the mean to the threshold.

[0167] Yellow alert: μt is between 50% and 80% of the threshold (e.g., 3mm to 4.8mm);

[0168] Orange alert: μt is between 80% and 100% of the threshold (e.g., 4.8 mm to 6 mm);

[0169] Red alert: μt exceeds the threshold and

[0170] Risk point identification needs to be combined with spatial correlation. If three or more adjacent measuring points trigger an early warning at the same time, it is judged as a regional systemic risk.

[0171] Step S352: Calculate the mutual information value between environmental factors and the displacement value of the vertical retaining structure, dynamically adjust the input weights of the time-varying Gaussian distribution model, and construct an environment-threshold mapping table to match the current environmental adjustment standard.

[0172] Specifically, environmental factors (such as rainfall R) are quantified using weighted mutual information. t Groundwater level W t Correlation with displacement:

[0173]

[0174] In the formula, the weight w(x,y) is defined according to the spatiotemporal proximity (e.g., more recent data has higher weight). p(x,y) represents the joint probability distribution of X and Y (e.g., the joint distribution of displacement and environmental factors); p(x) and p(y) represent the marginal probability distributions (individual probability distributions) of X and Y.

[0175] Construct an "environment-threshold" mapping table, for example: when R tWhen W > 50 mm, the displacement threshold is lowered from 6 mm to 4.5 mm; t When the water level exceeds the warning level, the early warning sensitivity of the associated monitoring points increases by 20%.

[0176] Step S353: Set an update cycle to periodically collect the latest monitoring data, calculate the loss value of the newly added data using the maximum likelihood loss function, and update the parameters of the time-varying Gaussian distribution model if the loss value is greater than the preset threshold.

[0177] Step S354: Map the mean prediction results to a three-dimensional virtual model to visualize the risk distribution of the retaining wall.

[0178] The early warning decision support unit is used to set dynamic multi-level early warning strategies, trigger multi-level early warnings based on analysis and prediction results, and generate emergency response and maintenance suggestions.

[0179] Specifically, based on a dynamic three-level early warning mechanism, this unit matches preset thresholds with prediction results and real-time data. For example, the thresholds can be set as follows: a yellow warning (displacement rate > 2 mm / day) triggers a platform pop-up and SMS reminder; an orange warning (cumulative displacement > 10 mm and abnormal soil pressure) activates an audible and visual alarm and generates preliminary reinforcement suggestions (such as adding drainage ditches); and a red warning (topological instability risk) links with the emergency management system to push anchor reinforcement plans and evacuation routes.

[0180] Meanwhile, the contingency plan matching engine filters similar case records from the historical case database (such as the anchor cable reinforcement scheme in a landslide treatment in Fujian), and combines blockchain technology to store evidence of early warning events and the handling process, so as to achieve accountability.

[0181] For example, when the predicted probability of a flood wall collapsing exceeds 85%, the emergency material allocation plan of a nearby project is automatically invoked, and the disposal instructions are synchronized to the mobile terminals of on-site personnel.

[0182] In this invention, the spatiotemporal fusion model achieves millimeter-level displacement monitoring accuracy. Combined with probabilistic prediction, it outputs confidence intervals and supports three-level early warning systems: yellow, orange, and red. For example, when the mean displacement in a certain area exceeds a threshold and the variance is low, a red early warning is directly triggered, and a reinforcement plan is pushed out. It possesses long-term stability and adaptive capabilities. Solar power supply and intelligent power management units ensure continuous operation of the device in environments without a power grid. An online learning mechanism periodically fine-tunes model parameters using the latest data to adapt to long-term evolution such as retaining wall material aging and changes in earth pressure.

[0183] It supports full automation and low maintenance. The self-testing system calibrates the sensor zero point daily and automatically switches to the redundant channel when an anomaly occurs. The maintenance alarm unit generates maintenance work orders in conjunction with GPS positioning, reducing the frequency of manual inspections and significantly reducing operation and maintenance costs in complex scenarios such as remote mountainous areas and river flood walls.

[0184] In the description of this invention, the battery management unit 4 includes: an energy supply module, a status monitoring module, a power consumption control module, and a maintenance alarm module.

[0185] The energy supply module is used to provide continuous power to the monitoring device through a combination of solar power generation and battery energy storage, and to adjust the energy input and output in real time to prevent overcharging or over-discharging from damaging the battery.

[0186] The status monitoring module is used to continuously track the real-time battery charge, health status, and operating temperature. It collects data through voltage and current sensors to analyze the battery's remaining lifespan and performance degradation trend.

[0187] The power consumption control module is used to dynamically adjust the device power consumption according to the current power level and monitoring needs. In the low power state, it automatically shuts down non-core functions and keeps only the basic sensors running.

[0188] The maintenance alarm module is used to establish a multi-level early warning mechanism, which prompts maintenance needs in a tiered manner through platform pop-ups, SMS notifications, etc., and automatically generates maintenance work orders and pushes them to designated personnel.

[0189] Please see Figure 4 The present invention also provides an automatic monitoring method for the displacement of vertical retaining structures, the monitoring method comprising:

[0190] S1. Configure distributed monitoring nodes on the top of the retaining wall to acquire sensor data monitored on the top of the retaining wall, and preprocess the sensor data to obtain effective tilt angle data;

[0191] S2. Based on the effective tilt angle data, the dynamic Kalman filter algorithm is used to identify and optimize the attitude angle, and combined with the vertical distance from the bottom to the top of the retaining wall, the displacement value of the vertical retaining structure is calculated.

[0192] S3. Upload the displacement values ​​of the vertical retaining structure calculated by all monitoring nodes to the remote data platform, build a monitoring network, analyze and predict the deformation trend of the retaining wall, and trigger early warning based on the analysis results.

[0193] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0194] like Figures 2 to 3 As shown, this embodiment provides an automatic displacement monitoring device for vertical retaining structures, installed on the top of the retaining wall. In this technical solution, high-precision gyroscopes, accelerometers, and geomagnetic field sensors are used to detect the tilt angle of the top of the retaining wall in real time.

[0195] The gyroscope data is used to measure dynamic angular velocity; the accelerometer data calculates the static tilt angle using the gravitational component; and the geomagnetic sensor data is used for orientation angle calibration. The measurement accuracy is 0.1 degrees in dynamic environments and 0.05 degrees in static environments. Furthermore, the sensors utilize MEMS technology, offering high-precision monitoring, environmental adaptability, long battery life, and resistance to vibration interference.

[0196] Specifically, the gyroscope can be a BMI088 gyroscope, which has high dynamic response capabilities and is suitable for capturing instantaneous vibrations or deformations of retaining walls; it has an industrial-grade temperature range (-40℃~85℃) and is suitable for extreme outdoor environments. The three-axis accelerometer can be an ADI ADXL355 accelerometer, which features ultra-low noise characteristics, ensuring a static tilt angle calculation accuracy of 0.05°, built-in anti-aliasing filtering to suppress high-frequency vibration interference, and supports solar power generation to extend battery life. The geomagnetic sensor can be an MMC5983MA geomagnetic sensor, with high resolution supporting sub-degree heading angle calibration (0.1° accuracy); its anti-interference design makes it suitable for retaining wall structures.

[0197] like Figures 2 to 3 As shown, the data preprocessing and filtering in this technical solution preprocesses the tilt angle data measured by the sensor to reduce measurement noise and improve measurement accuracy, making it suitable for various environments.

[0198] Specifically, an adaptive high-pass digital filter is used to eliminate static offset errors from gyroscope data, preserving the effective signal of dynamic angular velocity. This effectively smooths random noise, reduces the volatility of angular velocity data, improves measurement stability, and ensures that tilt measurement noise is suppressed to within 0.1 degrees in dynamic environments. A moving average filter is used for accelerometer data, calculating its average value as the current acceleration value. Time-domain smoothing is applied to the raw data to filter out short-term high-frequency interference. Furthermore, a second-order Butterworth low-pass filter is used, with a cutoff frequency set at 5Hz, effectively separating the gravity component from vibration noise, ensuring a static tilt angle calculation accuracy of 0.05 degrees. A windowed median filter is used to eliminate impulse noise, and a fast Fourier transform frequency domain analysis identifies the environmental magnetic field interference frequency band. A notch filter is used to suppress noise in specific frequency bands, preserving the low-frequency geomagnetic signal required for azimuth angle calibration.

[0199] In this technical solution, the monitoring data obtained by the attitude solver after preprocessing and filtering is obtained, and the current optimized attitude angle (motion attitude angle) of the module is calculated by combining the dynamic Kalman filter algorithm. The attitude angle is then converted into the displacement value of the top of the retaining wall by a specific formula and a programming script.

[0200] The attitude solver connects to a high-precision microelectromechanical gyroscope, a three-axis accelerometer, and a geomagnetic sensor via a dedicated digital bus to achieve multi-channel synchronous data acquisition and transmission. It utilizes a dynamic Kalman filter algorithm to optimize attitude angles. In vibration environments, it automatically improves the reliability of the gyroscope, while in static scenarios, it prioritizes calculations using the accelerometer's gravity component, ensuring a measurement accuracy of 0.1 degrees in dynamic environments and 0.05 degrees in static environments.

[0201] The dynamic Kalman filter algorithm dynamically adjusts the fusion weights of multi-source data by estimating the noise covariance matrix of the gyroscope, accelerometer, and geomagnetic sensor in real time, thereby obtaining the optimized attitude angle and accurately calculating the current motion attitude angle of the module in a complex dynamic environment.

[0202] As an alternative implementation, programming scripts or data processing software can be used to calculate the trigonometric function of the optimized current attitude angle (motion attitude angle) using a specific formula, so as to calculate the displacement value of the top of the retaining wall.

[0203] Furthermore, this technical solution features communication and remote management capabilities. It incorporates a built-in 4G communication device, enabling secure and stable transmission of real-time monitoring data to a remote data platform for cloud-based data management and analysis, used to monitor the displacement value of the retaining wall top. The communication device also has command receiving capabilities, allowing for remote calibration of sensors and adjustment of device settings via a built-in microcontroller, ensuring the accuracy and reliability of the monitoring data.

[0204] The remote data platform supports real-time alarm functionality, instantly alerting users or maintenance personnel when abnormal displacement or tilt angles exceed safe limits. The device's internal remote management function requires a built-in microcontroller to parse command content, identify the type of sensor to be operated, and remotely modify sensor sampling frequency, sensitivity, or alarm thresholds to adapt to different operating conditions.

[0205] Furthermore, the battery management module in this technical solution is equipped with an ultra-long-lasting battery that supports solar power generation, making it suitable for long-term monitoring in outdoor environments and ensuring the long-term stable operation of the monitoring device. It also features battery status monitoring, periodically reporting battery level to ensure timely reminders to maintenance personnel for maintenance or battery replacement when the battery is low. The battery management module supports solar charging, optimizes photovoltaic panel efficiency through an MPPT controller, employs a battery management system (BMS) to prevent overcharging, over-discharging, short circuits, and abnormal temperatures, and periodically sends battery status data to the cloud platform via a 4G module.

[0206] As an alternative implementation, the ultra-long-lasting battery can be a lithium thionyl chloride battery, which has extremely high energy density, is suitable for extreme temperatures, and has an extremely low self-discharge rate. It is suitable for long-term outdoor deployment for more than 10 years and is widely used in IoT nodes and remote monitoring equipment.

[0207] In summary, by utilizing the above-mentioned technical solutions of this invention, the present invention can accurately sense and monitor the displacement of the top of the retaining wall through tilt angle data, solving the deficiency of traditional monitoring methods in not being able to obtain displacement data in real time and accurately. It employs an attitude solver and a dynamic Kalman filter algorithm, and transmits and controls data remotely through a remote data platform, thereby achieving automatic monitoring with more comprehensive and accurate data, avoiding complex manual measurements. Simultaneously, edge computing is combined to perform real-time noise reduction and spatiotemporal alignment of the data, significantly improving data reliability. Based on a spatiotemporal prediction model using graph convolutional networks and recurrent neural networks, it can dynamically analyze the overall deformation trend and local risk points of the retaining wall, quantify uncertainty through probabilistic prediction, and trigger graded early warnings. It can support long-term continuous operation in outdoor environments without power grids, and combined with a self-inspection mechanism and remote maintenance interface, it significantly reduces the frequency of manual inspections. The modular design balances deployment convenience and system scalability, providing a high-precision, fully automatic, and low-maintenance closed-loop solution for engineering safety monitoring. This invention achieves precise monitoring of the displacement of vertical retaining structures by acquiring their tilt angle data, ensuring accurate measurement even under complex working conditions. Furthermore, digital filtering technology is used to preprocess the tilt angle data, effectively removing noise and interference and further improving measurement accuracy. Combined with a dynamic Kalman filter algorithm, the current motion attitude angle of the module can be accurately calculated in complex dynamic environments, achieving high-precision monitoring. This invention uploads the displacement values ​​and motion attitude angles of the vertical retaining structure to a remote data platform, transmitting the monitoring data to a remote server in real time and stably. Engineering management personnel can access the monitoring data anytime, anywhere via computers or mobile phones, without needing to be physically present on-site. Moreover, the remote data platform has intelligent risk prediction and early warning functions; once the monitoring data exceeds a preset safety threshold, it will immediately notify relevant personnel through SMS, email, and other means to facilitate timely action. The remote data platform also allows management personnel to remotely adjust the device's parameter settings, enabling refined management of the monitoring process and further enhancing the level of intelligent engineering management.

[0208] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An automatic displacement monitoring device for vertical retaining structures, characterized in that, include: The data acquisition unit is used to configure monitoring nodes, acquire sensor data monitored at the top of the retaining wall, and preprocess the sensor data to obtain effective tilt angle data. The displacement conversion unit is used to identify and optimize the attitude angle using the dynamic Kalman filter algorithm, and calculate the displacement value of the vertical retaining structure by combining the vertical distance from the bottom to the top of the retaining wall. The monitoring and early warning unit is used to upload the displacement value of the vertical retaining structure to a remote data platform, build a monitoring network, analyze and predict the deformation trend of the retaining wall, and trigger an early warning based on the analysis results. The battery management unit is used to provide battery status monitoring and management, and to periodically report battery energy. The data acquisition unit includes: The data acquisition module is used to acquire sensor data monitored in real time by various types of sensors; the sensor data includes gyroscope data, accelerometer data and geomagnetic sensor data; The gyroscope preprocessing module is used to remove static offset errors from gyroscope data based on an adaptive high-pass digital filter to obtain effective gyroscope data. The acceleration preprocessing module is used to smooth the accelerometer data in the time domain based on the moving average filter, and to separate the smoothed accelerometer data using a second-order Butterworth low-pass filter to obtain effective accelerometer data. The geomagnetic data preprocessing module is used to eliminate noise from geomagnetic sensor data based on a windowed median filter, identify the environmental magnetic field interference frequency band through fast Fourier transform frequency domain analysis, and suppress noise in the interference frequency band by combining a notch filter to obtain effective magnetic sensor data. The data integration module is used to integrate the processed effective gyroscope data, effective accelerometer data, and effective magnetic sensor data as the effective tilt angle data of the retaining wall. The displacement conversion unit includes: The error accumulation processing module is used to obtain the tilt optimization angle of the previous moment, and calculate the tilt optimization angle of the current moment by combining it with the effective gyroscope data of the current moment. The optimized angle correction module is used to correct the effective accelerometer data using Kalman gain, and combined with the tilt optimization calculation at the current moment, the optimized attitude angle is obtained through dynamic weighting calculation. The displacement conversion module is used to obtain the vertical distance from the bottom to the top of the retaining wall measured by the laser rangefinder. By optimizing the attitude angle conversion, the displacement value of the vertical retaining structure is calculated. The formula for calculating the tilt optimization angle at the current moment is: In the formula, This indicates the optimal tilt angle at time k; ω represents the tilt optimization angle at the previous moment; k This represents the gyroscope angular velocity corresponding to the valid gyroscope data; Δt represents the time interval. The formula for calculating the optimized attitude angle is: In the formula, Indicates the optimized attitude angle; Z accel，k K represents the tilt angle of the accelerometer at time k; k Indicates the Kalman gain at time k; The formula for calculating the displacement value of the vertical retaining structure is as follows: In the formula, ΔL represents the displacement value of the vertical retaining structure; h represents the vertical distance from the bottom to the top of the retaining wall.

2. The automatic displacement monitoring device for vertical retaining structures according to claim 1, characterized in that, The identification of environmental magnetic field interference frequency bands through fast Fourier transform frequency domain analysis includes: The noise-reduced geomagnetic sensor data is normalized, and the normalized geomagnetic sensor data is transformed in the frequency domain using the fast Fourier transform function to obtain the geomagnetic sensor data in the frequency domain. Based on the geomagnetic sensor data represented in the frequency domain, obtain the frequency corresponding to each complex point, and calculate the amplitude spectrum of each frequency point; Based on the amplitude spectrum of each frequency point, an amplitude spectrum diagram is plotted, and the frequency corresponding to the peak point with higher amplitude is obtained as the frequency band of environmental magnetic field interference.

3. The automatic displacement monitoring device for vertical retaining structures according to claim 1, characterized in that, The monitoring and early warning unit includes: The edge communication computing unit is used to aggregate edge computing data to a remote data platform using a wireless communication protocol, thereby realizing data aggregation of distributed monitoring nodes; The digital modeling and display unit is used to construct a three-dimensional virtual model of the retaining wall. By inputting the vertical retaining displacement values ​​of each monitoring node, the deformation state of the physical entity is mapped in real time. The spatiotemporal correlation prediction unit is used to predict the future deformation trend of retaining walls and identify potential risk points based on historical and real-time data through spatiotemporal information fusion and mining. The early warning decision support unit is used to set dynamic multi-level early warning strategies, trigger multi-level early warnings based on analysis and prediction results, and generate emergency response and maintenance suggestions.

4. The automatic displacement monitoring device for vertical retaining structures according to claim 3, characterized in that, The method of predicting the future deformation trend of retaining walls and identifying potential risk points based on historical and real-time data through spatiotemporal information fusion and mining includes: The historical and real-time vertical retaining structure displacement values, sensor data and environmental data of the retaining wall are obtained. All monitoring nodes are mapped to a unified spatiotemporal coordinate system. The Pearson correlation coefficient is used to quantify the spatial correlation between each monitoring node, and strongly correlated neighbor nodes are selected to construct an adjacency matrix. Based on the adjacency matrix, the features of adjacent monitoring nodes are aggregated through a multi-layer graph convolutional network, the adjacency matrix is ​​normalized, and the node features are passed layer by layer to capture the synergistic effect of local deformation of the retaining wall. The spatially aggregated features are input into a bidirectional gated recurrent network to model the long-term dependencies of displacement sequences along the time axis. A spatiotemporal cross-attention mechanism is introduced to calculate the association weights between spatial features and temporal hidden states, generating a coupled state vector that integrates spatiotemporal dynamics. Based on the spatiotemporal fusion characteristics, a time-varying Gaussian analysis model is constructed by using the mean and variance of the displacement prediction values ​​output by the linear layer. The maximum likelihood loss function is used to optimize the model parameters, and the time-varying Gaussian analysis model is used to predict the mean and variance of the deformation at future moments. Based on the mean and variance prediction results, the deformation trend and potential risk points of the retaining wall are assessed, and an online learning mechanism is introduced to optimize the model using the latest monitoring data.

5. The automatic displacement monitoring device for vertical retaining structures according to claim 4, characterized in that, The parameter prediction formula for the time-varying Gaussian analysis model is as follows: In the formula, μ t W represents the mean value of the predicted displacement of the vertical retaining structure. μ c represents the linear layer weights of the mean. t b represents the spatiotemporal fusion feature vector; μ This represents the mean bias term; W represents the variance of the predicted displacement values ​​of vertical retaining structures. σ Linear layer weights representing variance; b σ This represents the variance bias term; The formula for the maximum likelihood loss function is: In the formula, L represents the maximum likelihood loss function; y t The value represents the actual observed displacement of the vertical retaining structure at time t; T represents the total length of the time series.

6. The automatic displacement monitoring device for vertical retaining structures according to claim 5, characterized in that, The process of assessing the deformation trend and potential risk points of the retaining wall based on the mean and variance prediction results, and introducing an online learning mechanism to optimize the model using the latest monitoring data, includes: The future deformation of the retaining wall follows a Gaussian distribution. Based on the mean and variance prediction results, the probability interval of the future deformation is determined, and risk points are marked according to the deformation of each monitoring node. Calculate the mutual information value between environmental factors and the displacement value of vertical retaining structures, dynamically adjust the input weights of the time-varying Gaussian distribution model, and construct an environment-threshold mapping table to match the current environmental adjustment standard; Set an update cycle to periodically collect the latest monitoring data, calculate the loss value of the new data using the maximum likelihood loss function, and update the parameters of the time-varying Gaussian distribution model if the loss value is greater than the preset threshold. The mean prediction results are mapped to a three-dimensional virtual model to visualize the risk distribution of the retaining wall.

7. An automatic displacement monitoring method for vertical retaining structures, used to implement the automatic displacement monitoring device for vertical retaining structures as described in any one of claims 1-6, characterized in that, The monitoring method includes: Distributed monitoring nodes are configured on the top of the retaining wall to acquire sensor data from the top of the retaining wall, and the sensor data is preprocessed to obtain effective tilt angle data. Based on the effective tilt angle data, the dynamic Kalman filter algorithm is used to identify and optimize the attitude angle, and combined with the vertical distance from the bottom to the top of the retaining wall, the displacement value of the vertical retaining structure is calculated. The displacement values ​​of vertical retaining structures calculated from all monitoring nodes are uploaded to a remote data platform to build a monitoring network, analyze and predict the deformation trend of the retaining wall, and trigger early warnings based on the analysis results.

Images