Dam deformation monitoring method based on multi-source heterogeneous data fusion

Abstract

The invention discloses a dam deformation monitoring method based on multi-source heterogeneous data fusion, and relates to the technical field of deformation monitoring. The method comprises the following steps: collecting dam deformation data containing remote sensing deformation data and ground monitoring data; extracting remote sensing deformation data features by using an interferometric synthetic aperture radar method; extracting ground monitoring data features and correcting the ground monitoring data features by using a multi-rate Kalman filtering method to obtain monitoring deformation correction data; carrying out difference analysis and processing on the two types of correction and feature data, and weighting through an entropy weight method to obtain dam deformation fusion data; constructing a space-time coupling prediction model, and inputting fusion data to obtain a space-time deformation prediction tensor; and calculating time and space risk components according to the prediction tensor, and combining to obtain a space-time coupling risk index for dam deformation monitoring. According to the method, a fusion algorithm of multi-source heterogeneous data is creatively adopted, the integrity of dam deformation monitoring is guaranteed, and the efficiency of dam disaster prevention and emergency response is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of deformation monitoring, and in particular to a dam deformation monitoring method based on multi-source heterogeneous data fusion. Background Art

[0002] Dam deformation monitoring is crucial for ensuring project safety and preventing dam failures. During operation, dams are subject to deformation due to various factors, including water pressure and changing geological conditions. Deformation monitoring, utilizing advanced measurement technologies and equipment, enables real-time and accurate data on dam displacement, settlement, and other deformations. This data provides engineers with critical information, helping them to understand the dam's operating status and determine its stability.

[0003] However, traditional dam deformation monitoring mainly relies on a single monitoring technology. Although it can provide monitoring data with a certain degree of accuracy, the single measurement data is easily affected by environmental interference, resulting in the inability to fully and accurately reflect the true deformation status of the dam. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a dam deformation monitoring method based on multi-source heterogeneous data fusion, which solves the problem that traditional dam deformation monitoring mainly relies on a single monitoring technology, is easily affected by environmental interference, and cannot fully and accurately reflect the actual deformation state of the dam.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A dam deformation monitoring method based on multi-source heterogeneous data fusion, comprising the following steps: Step S1: collecting dam deformation data to obtain dam deformation data, wherein the dam deformation data includes remote sensing deformation data and ground monitoring data; Step S2: extracting features from the remote sensing deformation data using an interferometric synthetic aperture radar method to obtain remote sensing deformation feature data; Step S3: extracting features from the ground monitoring data to obtain ground monitoring deformation feature data, correcting the ground monitoring deformation feature data using a multi-rate Kalman filter method, and calculating monitoring deformation correction data; Step S4: performing a difference analysis on the monitored deformation correction data and the remote sensing deformation characteristic data to calculate a difference index, processing the monitored deformation correction data and the remote sensing deformation characteristic data using the difference index to obtain processed monitored deformation correction data and remote sensing deformation characteristic data, performing a weight analysis on the processed monitored deformation correction data and remote sensing deformation characteristic data using an entropy weight method to obtain deformation data weights, and combining the deformation data weights, the processed monitored deformation correction data, and the remote sensing deformation characteristic data to obtain dam deformation fusion data; Step S5: constructing a spatiotemporal coupling prediction model, inputting the dam deformation fusion data into the spatiotemporal coupling prediction model, and outputting a spatiotemporal deformation prediction tensor; Step S6: Calculate the time risk component and the space risk component based on the space-time deformation prediction tensor, calculate the space-time coupling risk index based on the time risk component and the space risk component, and monitor the dam deformation based on the space-time coupling risk index.

[0006] Preferably, the collecting of dam deformation data to obtain the dam deformation data includes: Remote sensing deformation data: Use synthetic aperture radar satellites or drone-mounted radar to obtain time-series SAR images of the dam area; Ground monitoring data: 9-axis inertial sensors: installed at key monitoring points on the dam surface, collecting three-axis acceleration, angular acceleration and geomagnetic parameters in real time; Displacement meter and settlement meter: directly measure the local displacement and settlement of the dam; Total station: measures the three-dimensional coordinate changes of the dam surface through prism reflection; Dam meteorological data: Air pressure data: Sensor: MPS3115 air pressure sensor is deployed at the dam weather station, collecting data every 10 minutes; Temperature data: Sensor: PT100 thermal resistor, dam surface, every 10m 2 Point layout, 1 minute / time collection; Humidity data: Sensor: SHT31 capacitive sensor, placed 20m away from the dam, collecting data every 5 minutes, to assist InSAR meteorological correction.

[0007] Preferably, the step S2 includes: Data preprocessing input: remote sensing deformation data: master-slave SAR image pairs; meteorological data: air pressure, temperature, humidity; Perform sub-pixel registration of master and slave images to eliminate geometric offsets caused by track and terrain; Generate an interferogram: Interference phase generation formula:

[0008] in, is the interference phase difference, is the main image phase, The deformation-related phase difference is calculated from the image phase, that is, the phase of the later SAR image of the same area, compared with the main image; It's a model Operation, processing phase entanglement, to facilitate subsequent phase unentanglement; Deformation phase: caused by surface deformation; Flat ground phase: caused by terrain elevation and radar baseline; Meteorological phase: caused by meteorological delay; Flat Earth Phase Formula:

[0009] in, It is the flat earth phase is the vertical baseline, h is the height, R is the slope distance, is the incident angle, eliminating the influence of terrain height (h) and radar vertical baseline ( ) caused by the phase component. The flat ground phase is the theoretical phase when the terrain is flat. The actual terrain undulation will cause phase deviation. is the radar wavelength, which is determined by the transmission frequency of the SAR satellite; After removing the flat ground phase, the remaining phase It mainly reflects deformation and meteorological influences; Through the minimum cost flow algorithm, the optimal flow is found, the phase jump is corrected, and the remaining phase After unwrapping, the unwrapped phase is output ; Meteorological phase correction: Multiple linear regression model:

[0010] in, It is the meteorological phase, It's the change in air pressure. is the temperature change, is the humidity change; the coefficient Determined by fitting historical data using the least squares method, is the residual value; Correction operation: Subtract the meteorological phase component from the unwrapped phase:

[0011] in, is the phase data after multi-period correction, is the unwrapped phase diagram, It is the meteorological phase; meteorological correction is a key step to improve the accuracy of deformation monitoring. Uncorrected meteorological phase may lead to deformation rate errors of up to several millimeters per year. Deformation feature extraction: Phase data after multi-phase correction Perform a linear fit of a time series: Phase data after multi-phase correction , is the phase related to surface deformation after deducting errors such as flat ground phase and meteorological phase. According to the InSAR principle, the deformation of the radar line of sight is The relationship with phase is:

[0012] in, is the deformation of the radar line of sight, is the phase data after multi-period correction, is the radar wavelength, which is determined by the transmission frequency of the SAR satellite. This formula converts phase into deformation and is the core bridge connecting phase data with deformation velocity.

[0013] Time series deformation velocity calculation:

[0014] in, Indicates the time of acquiring the SAR image in period i deformation during the time is the acquisition time of the i-th SAR image; is the main image reference time; is the average annual deformation rate; is the residual term.

[0015] Speed ​​solution:

[0016] in, is the average annual deformation rate, is the SAR image period; From the 1st to the Period, image acquisition time for each period Deformation corresponding to time the sum of products; It is The acquisition time of SAR image, is the corresponding deformation; First, we calculated the period from the 1st to the Image acquisition time The sum of the deformations and the corresponding time , and then multiply these two sums together. This means the number of periods n multiplied by the time it takes to acquire images for each period from period 1 to period n. The sum of the squares of Indicates the image acquisition time from period 1 to period n The square of the sum.

[0017] Accumulate the deformation in each time period:

[0018] in, is the cumulative displacement, which indicates the total deformation of the dam over a period of time. It is The deformation speed at the moment, N represents the total number of time points in the monitoring period; It is an accumulation operation, for each time point Deformation speed Sum them up to get the total displacement; Spatial deformation distribution in the deformation area: Cluster analysis was used to identify significant deformation areas (such as dam deformation velocity >5mm / year), the area where the deformation is significant is regarded as the deformation area space; Remote sensing deformation characteristic data : Deformation speed ; Cumulative displacement ;Spatial deformation distribution in the deformation area.

[0019] Preferably, extracting features from the ground monitoring data to obtain ground monitoring deformation feature data includes: Extracted from the ground monitoring data, the ground monitoring deformation characteristic data is obtained: Ground monitoring data: 9-axis inertial sensor raw data: three-axis acceleration: , , ; Three-axis angular acceleration: , , ; Three-axis geomagnetic parameters: , , ; Deformation displacement: Directly measure the local deformation displacement d of the dam through the displacement meter; Generate deformation velocity: v(t)= ; Among them, v(t) is the deformation velocity at time t, is the deformation displacement at time t, is the time interval, ) is at the moment Deformation displacement Attitude calculation: Calculate the sensor deformation attitude angle by fusing acceleration and geomagnetic data , including the pitch angle , roll angle , yaw angle :

[0020] in, is the roll angle, which indicates the rotation angle of the object around the X axis. is the pitch angle, which reflects the degree of the object's forward and backward tilt , , are the accelerations on the x, y, and z axes respectively; The three-axis geomagnetic parameters Convert to a horizontal coordinate system: ; Yaw angle: The rotation angle of the object around the Z axis (relative to magnetic north):

[0021] The yaw angle reflects the direction of the object (magnetic north is 0°, increasing clockwise), and They are the parameters of the three-axis geomagnetic parameters converted to the x-axis and y-axis parameters, , , are the geomagnetic parameters on the x, y, and z axes respectively; Obtain ground monitoring deformation characteristic data, including: displacement , deformation speed v, deformation ; Preferably, the ground monitoring deformation feature data is corrected by a multi-rate Kalman filter method, and the monitoring deformation correction data is calculated to include: Multi-rate Kalman filter design: State vector:

[0022] in is the deformation displacement, is the deformation speed, Deformation angle (rad), is the deformation angular velocity; State equation: Consider the dynamic changes of acceleration:

[0023] in, They are displacement, velocity and angle respectively; is the displacement at the previous moment, is the velocity at the previous moment, is the time interval, is the acceleration at the previous moment, is the angle at the previous moment, is the angular acceleration at the previous moment, is the rate of change of angular velocity at the current moment, is the rate of change of angular velocity. Represents the process noise vector introduced by various unpredictable factors in the process from time k-1 to time k; Optimal state:

[0024] in, is the current optimal state estimate, is the current observation value, It is the Kalman gain formula, balancing the weights of prediction and observation; is the observation residual, reflecting the difference between the prediction and the actual measurement, H is the observation matrix, mapping the state to the observation space, is the previous state, k is the time step, and the predicted value is corrected using the observed value to obtain the optimal state estimate; The Kalman gain formula is used to calculate the optimal weights to balance the credibility of the predicted values ​​and the observed values:

[0025] in, is the Kalman gain formula, is the forecast state covariance matrix, representing the forecast uncertainty, is the observation matrix, which maps the state space to the observation space; is the transpose of H and is used for dimensionality matching in matrix operations; R is the observation noise covariance matrix, which describes the noise characteristics of sensor measurements; the Kalman gain formula determines whether to rely more on the prediction model or the actual measurement in the update step; According to the current optimal state estimate , take out the monitoring deformation correction data : Deformation Angle: , reflecting the local tilt or torsion of the dam; Deformation speed: , characterizes the urgency of the deformation trend; Deformation displacement: , the accumulated displacement is used to evaluate the overall stability.

[0026] Preferably, performing difference analysis on the monitored deformation correction data and the remote sensing deformation feature data to calculate the difference index includes: Remote sensing deformation characteristic data : deformation velocity, cumulative displacement, spatial deformation distribution; Monitor and correct deformation data : deformation angle, deformation speed, deformation displacement; Analytical methods: Spatiotemporal alignment: ensuring that remote sensing deformation feature data and monitoring correction deformation data strictly match in terms of timestamp and spatial coordinate system; Parameter comparison: At the same monitoring point, compare the deformation speed and displacement of remote sensing deformation characteristic data with the monitored and corrected deformation data; Difference index calculation formula:

[0027] in is the difference indicator, is the number of detection points; It is the deformation velocity and cumulative displacement in remote sensing deformation characteristic data; It is the deformation velocity and deformation displacement in the ground monitoring data. The smaller the MSE value is, the closer the predicted value is to the true value.

[0028] Preferably, the combining of deformation data weights, processed monitoring deformation correction data and remote sensing deformation feature data to obtain dam deformation fusion data includes: Calculate deformation data weights:

[0029] in, Indicates the The weight of the class data, Indicates the The entropy value of the class detection data, Indicates the The entropy value of the class detection data, is the number of data types, the denominator Indicates that all Class data The values ​​are summed; Data fusion and output fusion method: weighted average: ,A, ) in, is the dam fusion deformation data, The weights corresponding to the processed remote sensing deformation characteristic data, the deformation velocity data in the monitoring and correction deformation data, and the weights corresponding to the processed remote sensing deformation characteristic data and the dam deformation displacement data in the monitoring and correction deformation data; Remote sensing deformation characteristic data and deformation velocity data in monitoring and correction deformation data, and are the processed remote sensing deformation characteristic data and the dam deformation displacement data in the monitoring and correction deformation data, A is the spatial deformation distribution data, It is the deformation angle data of the dam.

[0030] Preferably, the constructing of a spatiotemporal coupling prediction model includes: Time prediction branch: Input layer: Fusion deformation data from the dam , feature extraction of two types of data to form a time series tensor with a dimension of ,in is the time step, and the 2D features are deformation velocity and cumulative displacement, which capture the deformation dynamics in the time dimension; ARIMA layer: Based on , determine the parameters through ACF / PACF analysis, fit the linear trend, and output (T, 1): linear trend tensor, which captures the long-term stable pattern of dam deformation; LSTM layer: uses a 2-layer LSTM network to process nonlinear fluctuations, learns short-term dynamic features through a gating mechanism, and outputs Nonlinear fluctuation tensor, enhancing adaptability to complex time-varying scenarios; Fusion layer: weighted average, generation Time prediction, output: future deformation speed and cumulative displacement, achieving full-cycle prediction in the time dimension; Spatial prediction branch: Input layer: Fusion deformation data from the dam , extract the spatial grid tensor: ground deformation angle, deformation displacement, and 1m resolution spatial grid data generated by remote sensing to form an (X, Y, 2) tensor; Kriging interpolation layer: spatially interpolate discrete monitoring data to generate a continuous global deformation field , intuitively showing "where the deformation is" and providing continuous spatial input for subsequent GNN; GNN layer: builds a spatial graph model, learns the deformation conduction law, and outputs Spatial prediction, achieving refined prediction of spatial dimensions; Spatiotemporal fusion layer: Input embedding: Flatten the temporal prediction (T', 2) and spatial prediction (X×Y, 2) into a sequence, and convert the timestamp and spatial coordinate (x, y) into a 64-dimensional vector for embedding, with a total dimension of 128, to capture the inherent characteristics and associations of spatiotemporal data; Output layer: Spatiotemporal deformation tensor: 7 days in the future, 10 minutes x 1m resolution, including displacement, velocity, angle, and support for dynamic 3D visualization; That is, the space-time deformation prediction tensor:

[0031] in, is the spatiotemporal deformation prediction tensor, It is the future time (10-minute intervals), spatial coordinates (1m grid) displacement prediction (unit: mm), is the deformation velocity prediction, is the deformation angle prediction.

[0032] Preferably, the calculating of the temporal risk component and the spatial risk component according to the spatiotemporal deformation prediction tensor includes: Prediction tensor based on spatiotemporal deformation , calculate the time risk component: The risk of displacement exceeding the limit over time is measured using the formula:

[0033] in, is the time risk component, 、 、 They correspond to the displacement, velocity, and angle peaks within the time window, 、 、 The corresponding safety thresholds for displacement, velocity, and angle respectively; 、 、 These are the displacement, velocity, and angle weight coefficients, respectively. Risk is assessed by the excess ratio of the weighted comprehensive three-dimensional parameters within a time window. Prediction tensor based on spatiotemporal deformation , calculate the spatial risk component: The displacement exceeding limit risk of the weighted average spatial grid is calculated as follows:

[0034] in, is the spatial risk component, , , is the weight coefficient of the space related to displacement, velocity and angle, , , are the displacement, velocity, and angle prediction values ​​of the i-th spatial grid point, 、 、 The corresponding safety thresholds of displacement, velocity and angle, is the geological risk weight of the ith grid point, and M is the total number of spatial grid points.

[0035] Preferably, the calculation of the distribution network status assessment index by combining the power generation side reliability index and the power grid side stability index includes: Spatiotemporal coupling risk index: Integrates time and space risks, the formula is:

[0036] in, is the spatiotemporal coupling risk index, and are the time risk component and the spatial risk component, is the weight of the time risk component, is the weight of the spatial risk component.

[0037] Beneficial effects The present invention provides a dam deformation monitoring method based on multi-source heterogeneous data fusion, involving machine learning and deep learning technologies, which has the following beneficial effects: (1) This dam deformation monitoring method based on multi-source heterogeneous data fusion ensures the accurate fusion of remote sensing and ground monitoring data through synchronous calibration and feature extraction of multi-source heterogeneous data. Especially when the deformation of the dam area is relatively complex, remote sensing deformation data and ground data can complement each other and provide more comprehensive deformation information. Remote sensing deformation data has a wide spatial coverage and can timely reflect the deformation trend of the area around the dam, while ground monitoring data provides high-precision deformation data in local areas. Through the adaptive adjustment of the weighted fusion algorithm, the errors and inconsistencies between different data sources are effectively eliminated, further improving the accuracy of deformation monitoring.

[0038] (2) This dam deformation monitoring method based on multi-source heterogeneous data fusion can predict the future deformation trend of the dam in real time by combining historical monitoring data with current data. In particular, when there are rapid changes or emergencies, such as seasonal water level fluctuations and extreme climate change, the system can automatically adjust the warning level. This function not only improves the response speed of the monitoring system, but also optimizes the accuracy of the warning system, allowing dam managers to obtain key deformation information in a timely manner and take corresponding preventive measures, greatly improving the safety of the dam.

[0039] (3) The dam deformation monitoring method based on multi-source heterogeneous data fusion can automatically identify the potential risk areas of dam deformation through intelligent analysis and weighted fusion of multi-source data. Combined with trend analysis technology, it can effectively predict possible structural problems, such as increased dam settlement and crack expansion, and automatically trigger an early warning when the deformation exceeds the safety threshold. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flow chart of a dam deformation monitoring method based on multi-source heterogeneous data fusion proposed by the present invention.

[0041] Figure 2 The present invention proposes a dam deformation monitoring method based on multi-source heterogeneous data fusion to obtain a flow chart of dam deformation fusion data.

[0042] Figure 3 This is a flow chart of obtaining the spatiotemporal coupling risk index from a dam deformation monitoring method based on multi-source heterogeneous data fusion proposed by the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] See also Figure 1 The present invention provides a technical solution: a dam deformation monitoring method based on multi-source heterogeneous data fusion. Specifically, the following dam deformation monitoring method based on multi-source heterogeneous data fusion is provided. Figure 1 , the method comprises the following steps: Step S1: collecting dam deformation data, wherein the dam deformation data includes remote sensing deformation data and ground monitoring data.

[0045] Remote sensing deformation data: Use synthetic aperture radar (SAR) satellites (such as Sentinel-1) or radar mounted on drones to obtain time-series SAR images of the dam area.

[0046] Data characteristics: wide coverage (kilometer level), high spatial resolution (meter level), but significantly affected by meteorological conditions.

[0047] Ground monitoring data: 9-axis inertial sensor: installed at key monitoring points on the dam surface, real-time collection of three-axis acceleration ( , , ), angular acceleration ( , , ) and geomagnetic parameters ( , , ).

[0048] Displacement meter and settlement meter: directly measure the local displacement (unit: mm) and settlement (unit: mm) of the dam.

[0049] Total station: measures the three-dimensional coordinate changes of the dam surface through prism reflection.

[0050] Dam meteorological data: Air pressure data (hPa): Sensor: Ground: An MPS3115 air pressure sensor (accuracy ±0.1 hPa) is deployed at the dam weather station (≥50 m from the dam to avoid local airflow interference). Data is collected every 10 minutes and transmitted to the data center via RS485.

[0051] Temperature data (℃): Sensor: Dam surface: PT100 thermal resistor (accuracy ±0.1℃), every 10m 2 Point distribution, 1 minute / time collection (capturing thermal expansion and contraction effects); Humidity data (%RH): Sensor: External environment: SHT31 capacitive sensor (IP68 protection, accuracy ±2%RH), 20m away from the dam, collecting data every 5 minutes, to assist InSAR weather correction.

[0052] Time synchronization instructions: GPS timing mechanism: All sensors have built-in GPS modules that receive satellite atomic clock signals and synchronize sensor time with the satellite clock, ensuring that timestamp deviations for multi-source data are within ±1 millisecond. Calibration using the satellite time reference eliminates local sensor clock drift, achieving high-precision time consistency and ensuring time consistency for backend fusion analysis.

[0053] Time synchronization: Using the formula Perform time calibration. The correction value of the sensor's internal clock bias is determined through pre-calibration or real-time dynamic calibration. This method unifies the sensor's measurement time to the GPS time scale, ensuring accurate temporal matching of multi-source data and supporting time series correlation analysis. Spatial Calibration (ICP Algorithm) Description: Point cloud matching: Extract the surface point cloud (remote sensing spatial data) from SAR images and the ground monitoring point cloud (ground monitoring measured data) measured by the total station. By matching spatial features (such as geometric contours and key point coordinates), a correspondence between the two types of point clouds is established to achieve preliminary spatial alignment and provide initial matching pairs for iterative optimization.

[0054] The ICP algorithm achieves spatial calibration unification (millimeter-level error) of remote sensing and ground monitoring data, providing a unified spatial framework for subsequent data fusion (such as the overlay of InSAR and ground monitoring sensor data), supporting multi-dimensional spatial analysis of large deformation monitoring, and improving data consistency and analysis reliability.

[0055] Time synchronization and spatial calibration, using GPS timing and the ICP algorithm, respectively, achieve high-precision synchronization of multi-source data in time (millisecond level) and spatial coordinate system alignment (millimeter level), laying the foundation for subsequent deformation analysis, data fusion, and intelligent early warning. This process ensures that data from different sources and with different characteristics can be compared and integrated in time and space. It is a core prerequisite for multi-source heterogeneous data processing and directly affects the consistency of large-scale monitoring data and the accuracy of analytical results.

[0056] Step S2: extracting features from the remote sensing deformation data using an interferometric synthetic aperture radar method to obtain remote sensing deformation feature data; Interferometric synthetic aperture radar method is used to extract features from the remote sensing deformation data to obtain time-series SAR images of the dam area: Data preprocessing input: Remote sensing deformation data: master-slave SAR image pairs; Meteorological data (weather datasets: air pressure, temperature, humidity).

[0057] Image registration: Sub-pixel registration is performed on the master and slave images to eliminate geometric offsets caused by track and terrain.

[0058] The registration error must be less than 0.1 pixel (e.g., at a 5m resolution for Sentinel-1, the error is <0.5m).

[0059] Generate an interferogram: Interference phase generation formula:

[0060] in, is the interference phase difference, It is the main image phase, that is, the radar echo phase of the reference SAR image (such as early data), which records the initial scattering characteristics of the surface. The deformation-related phase difference is calculated from the image phase, that is, the phase of the later SAR image of the same area, compared with the main image; It's a model Operation, handle phase winding (limit the phase difference to 0 to ), which facilitates subsequent phase unwrapping (restoring the true deformation phase).

[0061] Function: Generates an interferometric phase diagram by comparing the phase difference between the master image (Master) and the slave image (Slave). The phase difference mainly includes the following components: Deformation phase: caused by surface deformation (target signal); Flat ground phase: caused by terrain elevation and radar baseline (needs to be eliminated); Meteorological phase: caused by meteorological delay (needs to be corrected).

[0062] Flat Earth Phase Formula:

[0063] in, It is the flat earth phase is the vertical baseline, h is the height, R is the slope distance, is the angle of incidence. Eliminate the influence of terrain height (h) and radar vertical baseline ( ) caused by the phase component. The flat ground phase is the theoretical phase when the terrain is flat. The actual terrain undulation will cause phase deviation. is the radar wavelength, which is determined by the transmission frequency of the SAR satellite.

[0064] Correlation: After removing the flat ground phase, the remaining phase ( ) mainly reflects deformation and meteorological influences.

[0065] Phase Unwrapping: Method: Minimum Cost Flow Algorithm Target: Will Entangling Phase (Range: arrive ) is converted into a continuous absolute phase.

[0066] Implementation steps: Construct phase gradient map: calculate the phase difference between adjacent pixels.

[0067] Define the cost function:

[0068] in, is the cost function, is the phase gradient after unwrapping (ideal continuous value), is the winding phase gradient (including jumps). i represents the row index (vertical position) of the pixel in the image (or phase map), used to traverse each row of pixels in the image; j represents the column index (horizontal position) of the pixel in the image (or phase map), used to traverse each column of pixels in the image.

[0069] The cost function measures the difference between the unwrapped gradient and the original gradient, and the goal is to minimize the total cost, that is, to make the unwrapped phase gradient as close to the original gradient as possible (retaining the true deformation information).

[0070] The phase graph is modeled as a graph network (node ​​= pixel, edge = adjacent pixel, edge weight = cost function value), and the minimum cost flow algorithm is used to find the optimal flow, correct the phase jump, and calculate the remaining phase. After unwrapping, the unwrapped phase is output (Unit: radians).

[0071] Meteorological phase correction: Weather data integration: Time matching: The ERA5 data temporal resolution (hourly) is aligned with the SAR imaging time. Minute-level meteorological parameters are generated through linear interpolation.

[0072] Spatial interpolation: interpolate coarse-resolution meteorological data (0.25°×0.25°) to the SAR pixel level (e.g., 10m×10m) using Kriging interpolation; Meteorological phase modeling: Multiple linear regression model:

[0073] in, It is the meteorological phase, is the change in air pressure (hPa), is the temperature change (°C), is the humidity change (%); the coefficient Determined by fitting historical data using the least squares method, is the residual value.

[0074] Correction operation: Subtract the meteorological phase component from the unwrapped phase:

[0075] in, is the phase data after multi-period correction, is the unwrapped phase diagram, It is the meteorological phase; meteorological correction is a key step to improve the accuracy of deformation monitoring. Uncorrected meteorological phase may cause deformation data errors of up to several millimeters / year.

[0076] Deformation feature extraction: Core parameter calculation: Phase data after multi-phase correction Perform a linear fit of a time series: Phase data after multi-phase correction , is the phase related to surface deformation after deducting errors such as flat ground phase and meteorological phase. According to the InSAR principle, the deformation of the radar line of sight is The relationship with phase is:

[0077] in, is the deformation of the radar line of sight, is the phase data after multi-period correction, is the radar wavelength, which is determined by the transmission frequency of the SAR satellite. This formula converts phase into deformation and is the core bridge connecting phase data with deformation velocity.

[0078] Time series deformation velocity calculation (linear fitting):

[0079] in, Indicates the time of acquiring the SAR image in period i deformation during the time is the acquisition time of the i-th SAR image; is the main image reference time; is the average annual deformation rate (target value); is the residual term.

[0080] Speed ​​solution (least squares method):

[0081] in, is the average annual deformation rate (mm / year), is the SAR image period; From the 1st to the Period, image acquisition time for each period Deformation corresponding to time the sum of products; It is The acquisition time of SAR image, is the corresponding deformation; First, we calculated the period from the 1st to the Image acquisition time The sum of the deformations and the corresponding time , and then multiply these two sums together. This means the number of periods n multiplied by the time it takes to acquire images for each period from period 1 to period n. The sum of the squares of Indicates the image acquisition time from period 1 to period n The square of the sum.

[0082] This formula uses the least squares method to solve the annual average deformation velocity v, and uses deformation data at multiple time points for fitting to obtain the most accurate velocity value.

[0083] Accumulate the deformation in each time period:

[0084] in, is the cumulative displacement, which indicates the total deformation of the dam over a period of time. It is The deformation amount at the moment, N represents the total number of time points in the monitoring period; It is an accumulation operation, for each time point Deformation Sum them up to get the total displacement.

[0085] Spatial deformation distribution in the deformation area: Cluster analysis was used to identify significant deformation areas (such as dam deformation velocity >5mm / year), and the area with significant deformation is taken as the deformation area space, that is, the spatial deformation distribution A.

[0086] Remote sensing deformation characteristic data : Deformation speed ; Cumulative displacement ; Spatial deformation distribution A.

[0087] Step S3: extracting features from the ground monitoring data to obtain ground monitoring deformation feature data, correcting the ground monitoring deformation feature data using a multi-rate Kalman filter method, and calculating monitoring deformation correction data.

[0088] Extracted from the ground monitoring data, the ground monitoring deformation characteristic data is obtained: Ground monitoring data: 9-axis inertial sensor raw data: three-axis acceleration: , , (unit: ); Three-axis angular acceleration: , , (unit: ); Three-axis geomagnetic parameters: , , (unit: ).

[0089] Deformation displacement: Directly measure the local deformation displacement d of the dam through the displacement meter; Generate deformation velocity: v(t)= .

[0090] Among them, v(t) is the deformation velocity at time t, is the deformation displacement at time t, is the time interval, ) is at the moment Deformation displacement Attitude calculation: Calculate the sensor deformation attitude angle by fusing acceleration and geomagnetic data (Pitch angle , roll angle , yaw angle ):

[0091] in, is the roll angle, which indicates the rotation angle of the object around the X axis. is the pitch angle, which reflects the degree of the object's front-back tilt. , , are the accelerations on the x, y, and z axes respectively; The three-axis geomagnetic parameters Convert to a horizontal coordinate system: ; Yaw angle: The rotation angle of the object around the Z axis (relative to magnetic north):

[0092] The yaw angle reflects the direction of the object (magnetic north is 0°, increasing clockwise), and They are the parameters of the three-axis geomagnetic parameters converted to the x-axis and y-axis parameters, , , are the geomagnetic parameters on the x, y, and z axes respectively; Obtain ground monitoring deformation characteristic data, including: displacement , deformation speed v, deformation ; Multi-rate Kalman filter design (where multi-rate refers to the rate at different times) State vector:

[0093] in is the deformation displacement (mm), is the deformation rate (mm / year), Deformation angle (rad), .

[0094] State equation: Consider the dynamic changes of acceleration (such as sudden deformation ):

[0095] in, They are displacement, velocity and angle respectively; is the displacement at the previous moment, is the velocity at the previous moment, is the time interval, is the acceleration at the previous moment, is the angle at the previous moment, is the angular acceleration at the previous moment, is the rate of change of angular velocity at the current moment, is the rate of change of angular velocity. It represents the process noise vector introduced by various unpredictable factors (such as measurement error, environmental interference, etc.) in the process from time k-1 to time k.

[0096] A dynamic model used to describe the evolution of deformation parameters (displacement, velocity, angle) over time.

[0097] Optimal state (corrected value):

[0098] in, is the current optimal state estimate, is the current observation value, It is the Kalman gain formula, balancing the weights of prediction and observation; is the observation residual, reflecting the difference between the prediction and the actual measurement, H is the observation matrix, which maps the state (such as displacement, velocity) to the observation space, is the previous state, and k is the time step. The predicted value is modified using the observed value to obtain the optimal state estimate.

[0099] The Kalman gain formula is used to calculate the optimal weights to balance the credibility of the predicted values ​​and the observed values:

[0100] in, is the Kalman gain formula, is the forecast state covariance matrix, which represents the forecast uncertainty. is the observation matrix, which maps the state space to the observation space; is the transpose of H and is used for dimensionality matching in matrix operations; R is the observation noise covariance matrix, which describes the noise characteristics of sensor measurements; the Kalman gain formula determines whether to rely more on the prediction model or the actual measurement in the update step.

[0101] According to the current optimal state estimate , take out the monitoring deformation correction data : Deformation Angle: (Unit: degree), reflecting the local tilt or torsion of the dam; Deformation speed: (Unit: mm / s), characterizes the urgency of the deformation trend; Deformation displacement: (Unit: mm), the cumulative displacement is used to evaluate the overall stability.

[0102] Step S4: Perform difference analysis on the monitored deformation correction data and the remote sensing deformation characteristic data to calculate the difference index, process the monitored deformation correction data and the remote sensing deformation characteristic data using the difference index to obtain processed monitored deformation correction data and remote sensing deformation characteristic data, perform weight analysis on the processed monitored deformation correction data and remote sensing deformation characteristic data using the entropy weight method to obtain deformation data weight, and combine the deformation data weight, the processed monitored deformation correction data and the remote sensing deformation characteristic data to obtain dam deformation fusion data.

[0103] Remote sensing deformation characteristic data : Deformation speed , cumulative displacement , spatial deformation distribution A; Monitor and correct deformation data : Deformation angle , deformation speed , deformation displacement ; Analytical methods: Spatiotemporal alignment: Ensure that the remote sensing deformation feature data and the monitoring-corrected deformation data strictly match in terms of timestamp and spatial coordinate system.

[0104] Parameter comparison: At the same monitoring point, compare the deformation speed and displacement (cumulative displacement) of remote sensing deformation characteristic data and monitoring corrected deformation data.

[0105] Difference index calculation formula:

[0106] in is the difference indicator, is the number of detection points; It is the deformation velocity and cumulative displacement in remote sensing deformation characteristic data; It is the deformation velocity and deformation displacement in the ground monitoring data. The smaller the MSE value is, the closer the predicted value is to the true value.

[0107] If the difference index is less than the threshold (such as 10%), the monitoring point data is considered consistent and serves as the basis for subsequent data fusion; otherwise, it is marked as a conflict and the data of this monitoring point does not enter the subsequent data fusion. and and They are the processed remote sensing deformation feature data and the monitored corrected deformation data (the difference index is less than the threshold as the processing standard); Entropy Weight Method: Weights are dynamically assigned based on the information entropy of the data, with smaller entropy values ​​(lower uncertainty) giving higher weights.

[0108] Calculation steps:

[0109] in, Indicates the The entropy value of the class detection data. Entropy is an indicator used to measure the uncertainty or degree of confusion of data. It is the first In the data of The distribution probability of each data point. In practical applications, these probabilities are usually obtained by normalizing the data. In the case of remote sensing deformation data and ground monitoring data, the data must first be normalized and then the probability of each data point in the category to which it belongs is calculated.

[0110] Calculate deformation data weights:

[0111] in, Indicates the The weight of the class data, Indicates the The entropy value of the class detection data, Indicates the The entropy value of the class detection data, is the number of data types, the denominator Indicates that all Class data The values ​​are summed.

[0112] Data fusion and output fusion methods: weighted average and vector combination: ,A, ) in, is the dam fusion deformation data, The weights corresponding to the processed remote sensing deformation characteristic data, the deformation velocity data in the monitoring and correction deformation data, and the weights corresponding to the processed remote sensing deformation characteristic data and the dam deformation displacement data in the monitoring and correction deformation data; They are the processed remote sensing deformation feature data and the deformation velocity data in the monitoring and correction deformation data. and are the processed remote sensing deformation characteristic data and the dam deformation displacement data in the monitoring and correction deformation data, A is the spatial deformation distribution data, It is the deformation angle data of the dam.

[0113] Step S5: constructing a spatiotemporal coupling prediction model, inputting the dam deformation fusion data into the spatiotemporal coupling prediction model, and outputting a spatiotemporal deformation prediction tensor.

[0114] Building a spatiotemporal coupled prediction model Time prediction branch: Input layer: Fusion deformation data from the dam , features are extracted into two types of data ( 、 ), forming a time series tensor, where , , the dimension is ,in is the time step, and the two-dimensional features are deformation velocity and cumulative displacement, which capture the deformation dynamics in the time dimension (long-term trend + instantaneous fluctuation).

[0115] ARIMA layer: Based on , the parameters are determined by ACF / PACF analysis ( Autoregressive order, Difference order, Sliding average order), fitting a linear trend. Output (T, 1): Linear trend tensor that captures the long-term stable pattern of dam deformation (such as the linear growth of annual settlement rate).

[0116] LSTM layer: A 2-layer LSTM network (64 hidden units, ReLU activation) is used to process nonlinear fluctuations (such as accelerated settlement caused by water level changes during flood season, or instantaneous deformation mutations caused by earthquakes). Through the gating mechanism (input gate controls new information, forget gate retains history, and output gate generates predictions), it learns short-term dynamic features (such as changes in deformation speed within 10 minutes) and outputs Nonlinear fluctuation tensor, enhancing adaptability to complex time-varying scenarios (such as sudden increase in deformation rate during construction period).

[0117] Fusion layer: weighted average (trend 0.4, fluctuation 0.6, can be optimized through the validation set), generate Time prediction ( , 7 days × 144 10-minute time intervals), output: future deformation velocity, cumulative displacement, and achieve full-cycle prediction in the time dimension.

[0118] Spatial prediction branch (Kriging GNN fusion) Input layer: Fusion deformation data from the dam , extract the spatial grid tensor: dam deformation angle , deformation displacement , and 1m resolution spatial deformation distribution data A generated by remote sensing. The data are spatially registered) to form an (X, Y, 2) tensor (X=Y=100, 100m×100m dam area, 1m resolution).

[0119] Kriging interpolation layer: performs spatial interpolation on discrete monitoring data (ordinary Kriging method, taking into account spatial variability, such as the deformation difference between the dam foundation and the dam abutment) to generate a continuous global deformation field , intuitively showing “where deformation occurs” (e.g., the dam base area is dark red, representing high displacement), providing continuous spatial input for subsequent GNN.

[0120] GNN layer: build a spatial graph model (nodes: 10,000 grid points; features: angle, displacement; edges: distance weights) + geological similarity weight (0.8 for the same lithology, reflecting physical conduction). A 2-layer GCN (32 hidden units) is used to learn deformation conduction laws (such as how dam foundation settlement affects dam body tilt through geological structure, at a rate of 0.5mm / year), and output Spatial prediction (angle and displacement of each grid point in the next 7 days) enables refined prediction of spatial dimensions (1m resolution, capturing local deformation details).

[0121] Spatiotemporal Fusion Layer (Transformer) Input embedding: Flatten the temporal prediction ((T', 2)) and spatial prediction ((X×Y, 2)) into a sequence, embed the timestamp (e.g., 2025-05-28 10:00, converted to a 64-dimensional vector) and spatial coordinates ((x, y), converted to a 64-dimensional vector), with a total dimension of 128, to capture the inherent characteristics and associations of spatiotemporal data (e.g., a high deformation rate at a certain point in time is associated with the historical association of a specific spatial region).

[0122] The multi-head attention mechanism uses four attention heads (32 dimensions per head) to calculate spatiotemporal attention weights. For example, it focuses on moments of sudden rate increases (such as the third day of the flood season, when the velocity is >0.1mm / s or >0.3mm / s) and clusters of high deformation in space (such as a 10m×10m area downstream of the dam with a displacement >5mm). It learns spatiotemporal interaction patterns (such as the deformation force field patterns in this area during the sudden rate increase), enhancing the model's understanding of complex spatiotemporal scenarios.

[0123] Output layer: A fully connected layer generates spatiotemporal fusion predictions. The three-dimensional features are displacement (mm, 1m grid), velocity (mm / s, 10 minutes), and angle (°, 1m grid), with errors of ±1mm, ±0.08mm / s, and ±0.09°, respectively. The output is normalized and converted to actual physical quantities, supporting GIS visualization (a 3D model overlaid with a scene cloud map dynamically displays spatiotemporal deformations for the next seven days).

[0124] Training set: Time training set: 2018-2023 (80%, 5 years of data, including normal season, flood season, and construction period), sliding window (144×10 minutes, 7-day prediction), generating 1000+ samples.

[0125] Spatial training set: 1m grid data corresponding to time (80%, approximately 1.8×10^6 grid points), data augmentation (random cropping and rotation) to improve generalization ability.

[0126] Validation set: 2024 (20%), to evaluate the accuracy of the model in new scenarios (such as deformation after extreme weather).

[0127] Output: Spatiotemporal deformation tensor: for the next 7 days, 10 minutes x 1m resolution, including displacement, velocity, and angle, supporting dynamic 3D visualization (such as animation of dam deformation evolution over time, color-coded deformation, red for high-risk areas).

[0128] That is, the space-time deformation prediction tensor:

[0129] in, is the spatiotemporal deformation prediction tensor, It is the future time (10-minute intervals), spatial coordinates (1m grid) displacement prediction (unit: mm), is the deformation velocity prediction, is the deformation angle prediction.

[0130] Step S6: Calculate the time risk component and the space risk component based on the space-time deformation prediction tensor, calculate the space-time coupling risk index based on the time risk component and the space risk component, and monitor the dam deformation based on the space-time coupling risk index.

[0131] Prediction tensor based on spatiotemporal deformation , calculate the time risk component ( ): The risk of displacement exceeding the limit over time is measured using the formula:

[0132] in, is the time risk component, 、 、 Corresponding to the time window ( arrive ) of displacement, velocity, and angle peaks, 、 、 The corresponding safety thresholds for displacement, velocity, and angle respectively; 、 、 are the displacement, velocity, and angle weight coefficients (e.g. , , , which can be obtained through training with historical data). Risk is assessed by the excess ratio of weighted comprehensive three-dimensional parameters (displacement, velocity, angle) within a time window.

[0133] Prediction tensor based on spatiotemporal deformation , calculate the spatial risk component ( ): The displacement exceeding limit risk of the weighted average spatial grid is calculated as follows:

[0134] in, is the spatial risk component, , , is the weight coefficient of the space related to displacement, speed, and angle (experts in related fields are invited to score the importance of displacement, speed, and angle to the space based on their experience and expertise, and then the scores are processed to determine the weight.), , , are the displacement, velocity, and angle prediction values ​​of the i-th spatial grid point, 、 、 The corresponding safety thresholds of displacement, velocity and angle, is the geological risk weight of the ith grid point (e.g., larger in fault areas and smaller in normal areas).

[0135] Spatiotemporal coupling risk index ( ): Integrating time and space risks, the formula is:

[0136] in, is the spatiotemporal coupling risk index, and are the time risk component and the spatial risk component, is the weight of the time risk component, is the weight of the spatial risk component, (time weight), (Spatial weights can be optimized through cross-validation). , ,but (Green, safe).

[0137] Calculation process and threshold value division: Green (0< 30): Safety, routine monitoring (e.g. , displacement , space risk ).

[0138] Yellow (31< 60): Early warning, encrypted observation (such as , displacement 5-7mm, local high risk).

[0139] Orange (61< 80): Alarm, start the plan (such as , displacement 7-8mm, regional risk diffusion).

[0140] Red (81< 100): Emergency, evacuation response (eg , displacement , high risks are concentrated and transmitted quickly).

[0141] This invention innovatively employs a fusion algorithm for heterogeneous multi-source data, employing a dynamic weighting mechanism that adjusts the weights of each data source based on real-time changes. Specifically, it automatically optimizes the data fusion strategy based on factors such as deformation speed and degree at different stages of dam deformation. This adaptive mechanism maintains monitoring accuracy despite various environmental changes and ensures the usability and accuracy of fused data in complex situations, providing greater flexibility and reliability for practical applications.

[0142] It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions. The sentence "including an element defined by..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element."

[0143] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A dam deformation monitoring method based on multi-source heterogeneous data fusion, characterized in that: The method comprises the following steps: Step S1: collecting dam deformation data to obtain dam deformation data, wherein the dam deformation data includes remote sensing deformation data and ground monitoring data; Step S2: extracting features from the remote sensing deformation data using an interferometric synthetic aperture radar method to obtain remote sensing deformation feature data; Step S3: extracting features from the ground monitoring data to obtain ground monitoring deformation feature data, correcting the ground monitoring deformation feature data using a multi-rate Kalman filter method, and calculating monitoring deformation correction data; Step S4: performing a difference analysis on the monitored deformation correction data and the remote sensing deformation characteristic data to calculate a difference index, processing the monitored deformation correction data and the remote sensing deformation characteristic data using the difference index to obtain processed monitored deformation correction data and remote sensing deformation characteristic data, performing a weight analysis on the processed monitored deformation correction data and remote sensing deformation characteristic data using an entropy weight method to obtain deformation data weights, and combining the deformation data weights, the processed monitored deformation correction data, and the remote sensing deformation characteristic data to obtain dam deformation fusion data; Step S5: constructing a spatiotemporal coupling prediction model, inputting the dam deformation fusion data into the spatiotemporal coupling prediction model, and outputting a spatiotemporal deformation prediction tensor; Step S6: Calculate the time risk component and the space risk component based on the space-time deformation prediction tensor, calculate the space-time coupling risk index based on the time risk component and the space risk component, and monitor the dam deformation based on the space-time coupling risk index.

2. The dam deformation monitoring method based on multi-source heterogeneous data fusion according to claim 1 is characterized in that: The dam deformation data is collected to obtain the dam deformation data, including: Remote sensing deformation data: Use synthetic aperture radar satellites or drone-mounted radar to obtain time-series SAR images of the dam area; Ground monitoring data: 9-axis inertial sensors: installed at key monitoring points on the dam surface, collecting three-axis acceleration, angular acceleration and geomagnetic parameters in real time; Displacement meter and settlement meter: directly measure the local displacement and settlement of the dam; Total station: measures the three-dimensional coordinate changes of the dam surface through prism reflection; Dam meteorological data: Air pressure data: Sensor: MPS3115 air pressure sensor is deployed at the dam weather station, collecting data every 10 minutes; Temperature data: Sensor: PT100 thermal resistor, dam surface, every 10m 2 Point layout, 1 minute / time collection; Humidity data: Sensor: SHT31 capacitive sensor, placed 20m away from the dam, collecting data every 5 minutes, to assist InSAR meteorological correction.

3. The dam deformation monitoring method based on multi-source heterogeneous data fusion according to claim 2 is characterized in that: The step S2 includes: Data preprocessing input: remote sensing deformation data: master-slave SAR image pairs; meteorological data: air pressure, temperature, humidity; Perform sub-pixel registration of master and slave images to eliminate geometric offsets caused by track and terrain; Generate an interferogram: Interference phase generation formula: ； in, is the interference phase difference, is the main image phase, The deformation-related phase difference is calculated from the image phase, that is, the phase of the later SAR image of the same area, compared with the main image; It is a model Operation, processing phase entanglement, to facilitate subsequent phase unentanglement; Deformation phase: caused by surface deformation; Flat ground phase: caused by terrain elevation and radar baseline; Meteorological phase: caused by meteorological delay; Flat Earth Phase Formula: ； in, It is the flat earth phase is the vertical baseline, h is the height, R is the slope distance, is the incident angle, eliminating the influence of terrain height (h) and radar vertical baseline ( ) caused by the phase component. The flat ground phase is the theoretical phase when the terrain is flat. The actual terrain undulation will cause phase deviation. is the radar wavelength, which is determined by the transmission frequency of the SAR satellite; After removing the flat ground phase, the remaining phase It mainly reflects deformation and meteorological influences; Through the minimum cost flow algorithm, the optimal flow is found, the phase jump is corrected, and the remaining phase After unwrapping, the unwrapped phase is output ; Meteorological phase correction: Multiple linear regression model: ； in, It is the meteorological phase, It's the change in air pressure. is the temperature change, is the humidity change; the coefficient Determined by fitting historical data using the least squares method, is the residual value; Correction operation: Subtract the meteorological phase component from the unwrapped phase: ； in, is the phase data after multi-period correction, is the unwrapped phase diagram, It is the meteorological phase; meteorological correction is a key step to improve the accuracy of deformation monitoring. Uncorrected meteorological phase may lead to deformation rate errors of up to several millimeters per year. Deformation feature extraction: Phase data after multi-phase correction Perform a linear fit of a time series: Phase data after multi-phase correction , is the phase related to surface deformation after deducting errors such as flat ground phase and meteorological phase. According to the InSAR principle, the deformation of the radar line of sight is The relationship with phase is: ； in, is the deformation of the radar line of sight, is the phase data after multi-period correction, is the radar wavelength, which is determined by the transmission frequency of the SAR satellite. This formula converts phase into deformation and is the core bridge connecting phase data and deformation velocity; Time series deformation velocity calculation: ； in, Indicates the time of acquiring the SAR image in period i deformation during the time is the acquisition time of the i-th SAR image; is the main image reference time; is the average annual deformation rate; is the residual term; Speed ​​solution: ； in, is the average annual deformation rate, is the SAR image period; From the 1st to the Period, image acquisition time for each period Deformation corresponding to time the sum of products; It is The acquisition time of SAR image, is the corresponding deformation; First, we calculated the period from the 1st to the Image acquisition time The sum of the deformations and the corresponding time , and then multiply these two sums together. This means the number of periods n multiplied by the time it takes to acquire images for each period from period 1 to period n. The sum of the squares of Indicates the image acquisition time from period 1 to period n the square of the sum; Accumulate the deformation in each time period: ； in, is the cumulative displacement, which indicates the total deformation of the dam over a period of time. It is The deformation amount at the moment, N represents the total number of time points in the monitoring period; It is an accumulation operation, for each time point Deformation speed Sum them up to get the total displacement; Spatial deformation distribution in the deformation area: Cluster analysis was used to identify significant deformation areas (such as dam deformation velocity >5mm / year), and the area with significant deformation is regarded as the deformation area space; Remote sensing deformation characteristic data : Deformation speed ; Cumulative displacement ;Spatial deformation distribution in the deformation area.

4. The dam deformation monitoring method based on multi-source heterogeneous data fusion according to claim 3 is characterized in that: The extracting features from the ground monitoring data to obtain ground monitoring deformation feature data includes: Extracted from the ground monitoring data, the ground monitoring deformation characteristic data is obtained: Ground monitoring data: 9-axis inertial sensor raw data: three-axis acceleration: , , ; Three-axis angular acceleration: , , ; Three-axis geomagnetic parameters: , , ; Deformation displacement: Directly measure the local deformation displacement d of the dam through the displacement meter; Generate deformation velocity: v(t)= ; Among them, v(t) is the deformation velocity at time t, is the deformation displacement at time t, is the time interval, ) is at the moment Deformation displacement Attitude calculation: Calculate the sensor deformation attitude angle by fusing acceleration and geomagnetic data , including the pitch angle , roll angle , yaw angle : ； in, is the roll angle, which indicates the rotation angle of the object around the X axis. is the pitch angle, which reflects the degree of the object's forward and backward tilt , , are the accelerations on the x, y, and z axes respectively; The three-axis geomagnetic parameters Convert to a horizontal coordinate system: ; Yaw angle: The rotation angle of the object around the Z axis (relative to magnetic north): ； The yaw angle reflects the direction of the object (magnetic north is 0°, increasing clockwise), and They are the parameters of the three-axis geomagnetic parameters converted to the x-axis and y-axis parameters, , , are the geomagnetic parameters on the x, y, and z axes respectively; Obtain ground monitoring deformation characteristic data, including: displacement , deformation speed v, deformation .

5. The dam deformation monitoring method based on multi-source heterogeneous data fusion according to claim 4 is characterized in that: The method of correcting the ground monitoring deformation feature data by the multi-rate Kalman filter method and calculating the monitoring deformation correction data includes: Multi-rate Kalman filter design: State vector: ； in is the deformation displacement, is the deformation speed, Deformation angle, is the deformation angular velocity; State equation: Consider the dynamic changes of acceleration: ； in, They are displacement, velocity and angle respectively; is the displacement at the previous moment, is the velocity at the previous moment, is the time interval, is the acceleration at the previous moment, is the angle at the previous moment, is the angular acceleration at the previous moment, is the rate of change of angular velocity at the current moment, is the rate of change of angular velocity. Represents the process noise vector introduced by various unpredictable factors in the process from time k-1 to time k; Optimal state: ； in, is the current optimal state estimate, is the current observation value, It is the Kalman gain formula, balancing the weights of prediction and observation; is the observation residual, reflecting the difference between the prediction and the actual measurement, H is the observation matrix, mapping the state to the observation space, is the previous state, k is the time step, and the predicted value is corrected using the observed value to obtain the optimal state estimate; The Kalman gain formula is used to calculate the optimal weights to balance the credibility of the predicted values ​​and the observed values: ； in, is the Kalman gain formula, is the forecast state covariance matrix, representing the forecast uncertainty, is the observation matrix, which maps the state space to the observation space; is the transpose of H, used for dimensionality matching in matrix operations; R is the observation noise covariance matrix, which describes the noise characteristics of sensor measurements; the Kalman gain formula determines whether to rely more on the prediction model or the actual measurement in the update step; According to the current optimal state estimate , take out the monitoring deformation correction data : Deformation Angle: , reflecting the local tilt or torsion of the dam; Deformation speed: , characterizes the urgency of the deformation trend; Deformation displacement: , the accumulated displacement is used to evaluate the overall stability.

6. The dam deformation monitoring method based on multi-source heterogeneous data fusion according to claim 5 is characterized in that: The difference analysis of the monitored deformation correction data and the remote sensing deformation feature data is performed to calculate the difference index, including: Remote sensing deformation characteristic data : deformation velocity, cumulative displacement, spatial deformation distribution; Monitor and correct deformation data : deformation angle, deformation speed, deformation displacement; Analytical methods: Spatiotemporal alignment: ensuring that remote sensing deformation feature data and monitoring correction deformation data strictly match in terms of timestamp and spatial coordinate system; Parameter comparison: At the same monitoring point, compare the deformation speed and displacement of remote sensing deformation characteristic data with the monitored and corrected deformation data; Difference index calculation formula: ； in is the difference indicator, is the number of detection points; It is the deformation velocity and cumulative displacement in remote sensing deformation characteristic data; It is the deformation velocity and deformation displacement in the ground monitoring data. The smaller the MSE value is, the closer the predicted value is to the true value.

7. The dam deformation monitoring method based on multi-source heterogeneous data fusion according to claim 6 is characterized in that: The dam deformation fusion data is obtained by combining the deformation data weight, the processed monitoring deformation correction data and the remote sensing deformation feature data, including: Calculate deformation data weights: ； in, Indicates the The weight of the class data, Indicates the The entropy value of the class detection data, Indicates the The entropy value of the class detection data, is the number of data types, the denominator Indicates that all Class data The values ​​are summed; Data fusion and output fusion method: weighted average: ,A, )； in, is the dam fusion deformation data, The weights corresponding to the processed remote sensing deformation characteristic data, the deformation velocity data in the monitoring and correction deformation data, and the weights corresponding to the processed remote sensing deformation characteristic data and the dam deformation displacement data in the monitoring and correction deformation data; Remote sensing deformation characteristic data and deformation velocity data in monitoring and correction deformation data, and are the processed remote sensing deformation characteristic data and the dam deformation displacement data in the monitoring and correction deformation data, A is the spatial deformation distribution data, It is the deformation angle data of the dam.

8. The dam deformation monitoring method based on multi-source heterogeneous data fusion according to claim 7 is characterized in that: The constructing of the spatiotemporal coupling prediction model includes: Time prediction branch: Input layer: Fusion deformation data from the dam , feature extraction of two types of data to form a time series tensor with a dimension of ,in is the time step, and the 2D features are deformation velocity and cumulative displacement, which capture the deformation dynamics in the time dimension; ARIMA layer: Based on , determine the parameters through ACF / PACF analysis, fit the linear trend, and output (T, 1): linear trend tensor, which captures the long-term stable pattern of dam deformation; LSTM layer: uses a 2-layer LSTM network to process nonlinear fluctuations, learns short-term dynamic features through a gating mechanism, and outputs Nonlinear fluctuation tensor, enhancing adaptability to complex time-varying scenarios; Fusion layer: weighted average, generation Time prediction, output: future deformation speed and cumulative displacement, achieving full-cycle prediction in the time dimension; Spatial prediction branch: Input layer: Fusion deformation data from the dam , extract the spatial grid tensor: ground deformation angle, deformation displacement, and 1m resolution spatial grid data generated by remote sensing to form an (X, Y, 2) tensor; Kriging interpolation layer: spatially interpolate discrete monitoring data to generate a continuous global deformation field , intuitively showing "where the deformation is" and providing continuous spatial input for subsequent GNN; GNN layer: builds a spatial graph model, learns the deformation conduction law, and outputs Spatial prediction, achieving refined prediction of spatial dimensions; Spatiotemporal fusion layer: Input embedding: Flatten the temporal prediction (T', 3) and spatial prediction (X×Y, 2) into a sequence, and convert the timestamp and spatial coordinate (x, y) into a 64-dimensional vector for embedding, with a total dimension of 128, to capture the inherent characteristics and associations of spatiotemporal data; Output layer: Spatiotemporal deformation tensor: 7 days in the future, 10 minutes x 1m resolution, including displacement, velocity, angle, and support for dynamic 3D visualization; That is, the space-time deformation prediction tensor: ; in, is the spatiotemporal deformation prediction tensor, It is the future time (10-minute intervals), spatial coordinates (1m grid) displacement prediction, is the deformation velocity prediction, is the deformation angle prediction.

9. The dam deformation monitoring method based on multi-source heterogeneous data fusion according to claim 8 is characterized in that: The calculating of the temporal risk component and the spatial risk component according to the spatiotemporal deformation prediction tensor includes: Prediction tensor based on spatiotemporal deformation , calculate the time risk component: The risk of displacement exceeding the limit over time is measured using the formula: ； in, is the time risk component, 、 、 They correspond to the displacement, velocity, and angle peaks within the time window, 、 、 The corresponding safety thresholds for displacement, velocity, and angle respectively; 、 、 are the displacement, velocity, and angle weight coefficients, respectively, and the risk is assessed by the excess ratio of the weighted comprehensive three-dimensional parameters within a time window; Prediction tensor based on spatiotemporal deformation , calculate the spatial risk component: The displacement exceeding limit risk of the weighted average spatial grid is calculated as follows: ； in, is the spatial risk component, , , is the weight coefficient of the space related to displacement, velocity and angle, , , are the displacement, velocity, and angle prediction values ​​of the i-th spatial grid point, 、 、 The corresponding safety thresholds of displacement, velocity and angle, is the geological risk weight of the ith grid point, and M is the total number of spatial grid points.

10. The dam deformation monitoring method based on multi-source heterogeneous data fusion according to claim 9 is characterized in that: The calculation of the distribution network status assessment index by combining the power generation side reliability index and the power grid side stability index includes: Spatiotemporal coupling risk index: Integrates time and space risks, the formula is: ； in, is the spatiotemporal coupling risk index, and are the time risk component and the spatial risk component, is the weight of the time risk component, is the weight of the spatial risk component.

Images