Soft soil foundation deformation prediction method and system based on big data

Abstract

The invention discloses a soft soil foundation deformation prediction method and system based on big data, and relates to the technical field of rock and soil monitoring. InSAR satellite data, Beidou GNSS displacement data, optical fiber strain data and meteorological and geological parameters are integrated through a multi-source sensing network. ERA5 reanalysis data is adopted to establish an atmospheric delay compensation function, and dynamic sliding window filtering and strain gradient constraint are combined to realize data space-time alignment and anomaly cleaning. Based on a generalized Kelvin creep constitutive model, environment coupling functions of temperature, humidity and pore water pressure are fused to dynamically correct model parameters, and the creep response characterization capability in a complex environment is enhanced. And performing distributed joint training on the regionalized geological data through a federated learning framework, fusing differential privacy encryption and a node credibility verification mechanism, realizing safety aggregation and migration optimization of cross-regional data, and generating a geological partition adaptive deformation prediction result. The method effectively improves the reliability of soft soil foundation deformation prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of geotechnical monitoring, and particularly to a method and system for predicting soft soil foundation deformation based on big data. Background Art

[0002] The prediction of soft soil foundation deformation is a core technical problem in the field of geotechnical engineering safety monitoring, and its accuracy is directly related to the long-term stability of major infrastructure and the effectiveness of disaster warning. Traditional prediction methods mostly rely on single monitoring means, such as GNSS displacement measurement or geological exploration data, and it is difficult to comprehensively capture the multi-dimensional coupling characteristics of soft soil creep. Especially in complex geological regions such as coastal areas and river networks, the foundation deformation is dynamically affected by multiple environmental parameters such as temperature, humidity, and pore water pressure. Existing constitutive models often use static parameter correction, resulting in significant lag deviation in long-term prediction. In addition, although new monitoring technologies such as InSAR and fiber optic sensing can provide high-density spatio-temporal data, problems such as inconsistent spatio-temporal benchmarks of multi-source heterogeneous data, atmospheric delay interference, and abnormal noise seriously restrict the data fusion accuracy. At the model construction level, the traditional centralized training mode faces contradictions such as regional sensitivity of geological data, cross-regional sharing barriers, and privacy protection, and it is difficult to achieve distributed learning that takes into account both data security and model generalization ability. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention provides a method and system for predicting soft soil foundation deformation based on big data.

[0004] The present invention provides a method for predicting soft soil foundation deformation based on big data, including: S1. Obtain deformation monitoring data of the soft soil foundation through a multi-source sensing network, including InSAR satellite phase data, Beidou GNSS displacement time series data, distributed fiber optic strain data, meteorological data, and geological exploration parameters; S2. Perform spatio-temporal alignment and outlier cleaning on the deformation monitoring data, and construct an atmospheric delay compensation function using ERA5 reanalysis data to generate a preprocessed standardized data set; S3. Based on the standardized data set, construct a generalized Kelvin creep constitutive model, introduce an environmental coupling function to dynamically correct the model parameters, where T is temperature, W is humidity, and P is pore water pressure; S4. Use the federated learning framework to perform distributed joint training on the generalized Kelvin creep constitutive model, and optimize the data characteristics of different geological zones through transfer learning to generate regionally adaptive deformation prediction results.

[0005] Optionally, the method for constructing the atmospheric delay compensation function in S2 includes: Extract the total atmospheric water vapor parameter and the total electron content parameter of the ionosphere in the monitoring area based on ERA5 reanalysis data; Construct the atmospheric delay compensation function by combining the residual phase noise, the total water vapor parameter, the total electron content parameter, and the corresponding regional atmospheric parameter compensation coefficient.

[0006] Optionally, the outlier cleaning in S2 includes the following steps: Perform wavelet multi-scale decomposition on the Beidou GNSS displacement time series data to separate the low-frequency deformation signal and the high-frequency noise component; Dynamically adjust the length of the sliding filter window based on the ratio relationship between the signal-to-noise ratio of the current window and the reference signal-to-noise ratio; Perform adaptive threshold filtering on the high-frequency noise component using the dynamically adjusted sliding filter window to retain the low-frequency deformation signal.

[0007] Optionally, the outlier cleaning in S2 also includes a real-time dynamic threshold adjustment mechanism: Based on the spatial distribution characteristics of the strain gradient of adjacent distributed fiber optic sensing nodes, establish a strain gradient safety constraint threshold by combining the node spacing, soil shear strength, and elastic modulus; When it is detected that the strain change rate of adjacent nodes exceeds the safety constraint threshold, perform the following processing: automatically extend the time coverage range of the current sliding filter window, and dynamically adjust the window expansion amplitude based on the duration of the anomaly; synchronously fuse the Beidou GNSS displacement data and the InSAR deformation inversion results in the corresponding period, and compensate the strain measurement values in the anomaly area through a multi-source data joint filtering algorithm.

[0008] Optionally, the environmental coupling function in S3 is defined as: ; where, f env (T, W, P) is the environmental coupling function, T represents temperature, W represents humidity, P represents pore water pressure, W sat is the saturated humidity, m1, m2, m3 are coupling coefficients dynamically updated through the LSTM network, t represents time, represents the pore water pressure gradient.

[0009] Optionally, the implementation of the federated learning framework in S4 includes: S41. Each edge node trains a generalized Kelvin model sub-model based on local data, and performs noise perturbation processing on the model gradient parameters using the differential privacy encryption algorithm; S42. The cloud aggregation node calculates the weighted average of the global model parameters based on the encrypted gradient parameters uploaded by each node, with the proportion of the monitoring data volume of each node in the total data volume as the weighting coefficient, to achieve distributed joint update of cross-regional model parameters.

[0010] Optionally, before S42, edge node credibility verification is also included: When each edge node uploads gradient parameters, it synchronously submits the mean value, variance, and difference degree index from the local dataset to the global data distribution; Based on the ratio threshold of the difference degree index to the data dispersion degree and the deviation tolerance range of the local mean value from the global mean value distribution, the cloud selects a credible node set that meets the consistency constraints; When performing weighted aggregation on the gradient parameters of the credible node set, the data volume weight is exponentially attenuated and corrected based on the data distribution difference degree of each node to generate a global model parameter update amount.

[0011] In a second aspect, the present invention also provides a soft soil foundation deformation prediction system based on big data, including: An acquisition module, configured to acquire deformation monitoring data of the soft soil foundation through a multi-source sensing network, including InSAR satellite phase data, Beidou GNSS displacement time series data, distributed optical fiber strain data, meteorological data, and geological exploration parameters; A generation module, configured to perform spatio-temporal alignment and outlier cleaning on the deformation monitoring data, and construct an atmospheric delay compensation function using ERA5 reanalysis data to generate a preprocessed standardized dataset; A correction module, configured to construct a generalized Kelvin creep constitutive model based on the standardized dataset, and introduce an environmental coupling function to dynamically correct the model parameters, where T is temperature, W is humidity, and P is pore water pressure; A prediction module, configured to perform distributed joint training on the generalized Kelvin creep constitutive model using a federated learning framework, optimize data characteristics in different geological zones through transfer learning, and generate a regionally adaptive deformation prediction result.

[0012] The present invention has the following technical effects: The present invention significantly improves the spatial coverage density and temporal resolution of soft soil foundation deformation monitoring through multi-dimensional data fusion of a multi-source sensing network, overcoming the perception limitations of traditional single monitoring means. The atmospheric delay compensation function constructed based on ERA5 reanalysis data effectively eliminates the tropospheric and ionospheric interference in InSAR satellite phase data. Combining with the dynamic sliding window filtering algorithm, it realizes the adaptive noise reduction processing of Beidou GNSS displacement data, ensuring the spatio-temporal consistency of multi-source heterogeneous data. The environmental coupling function is introduced to dynamically correct the parameters of the generalized Kelvin creep constitutive model, enhancing the response sensitivity of the model to temperature and humidity changes and pore water pressure fluctuations, and solving the hysteresis problem of the static parameter model in complex environments. The federal learning framework is adopted to break through the cross-regional geological data sharing barrier. While ensuring data security through differential privacy encryption, the node credibility verification mechanism is used to screen training nodes that conform to the global distribution characteristics, improving the generalization ability of the regional adaptive model. Further, the abnormal cross-verification of distributed optical fiber data is realized through the strain gradient constraint equation. Combining with the multi-source data compensation mechanism of Kalman filtering, the fault tolerance and robustness of the monitoring system are strengthened. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0014] Figure 1 It is a schematic flow chart of a method for predicting soft soil foundation deformation based on big data provided by an embodiment of the present invention; Figure 2 It is a schematic structural diagram of a system for predicting soft soil foundation deformation based on big data provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope protected by the present invention.

[0016] Figure 1 It is a schematic flow chart of a method for predicting soft soil foundation deformation based on big data provided by an embodiment of the present invention, including: S1. Obtain the deformation monitoring data of the soft soil foundation through a multi-source sensing network, including InSAR satellite phase data, Beidou GNSS displacement time series data, distributed optical fiber strain data, meteorological data, and geological exploration parameters; S2. Perform spatio-temporal alignment and outlier cleaning on the deformation monitoring data, construct an atmospheric delay compensation function using ERA5 reanalysis data, and generate a preprocessed standardized data set; S3. Based on the standardized data set, construct a generalized Kelvin creep constitutive model, introduce an environmental coupling function to dynamically correct the model parameters, where T is temperature, W is humidity, and P is pore water pressure; S4. Use the federated learning framework to perform distributed joint training on the generalized Kelvin creep constitutive model, optimize the data characteristics of different geological zones through transfer learning, and generate regional adaptive deformation prediction results.

[0017] When implementing this method, first establish a multi-source sensing network monitoring system covering the soft soil foundation area. Install Beidou GNSS reference stations and monitoring stations according to the geological structure characteristics in the monitoring area to form a satellite positioning observation array, and synchronously access the InSAR phase data acquisition channel of the spaceborne synthetic aperture radar. Lay distributed optical fiber sensing cables along the key sections of the roadbed, and use Brillouin optical time domain reflectometry technology to obtain the foundation strain distribution in real time. Install meteorological monitoring stations to collect atmospheric temperature, precipitation, and air pressure data in real time, and extract geological parameters such as soil porosity and permeability coefficient in combination with the geological exploration report. The multi-source data is connected to the edge computing node through the Internet of Things gateway to form a four-dimensional monitoring matrix including deformation displacement, strain distribution, environmental status, and geological characteristics.

[0018] In the data preprocessing stage, a spatio-temporal reference unification method is used to solve the problem of multi-sensor heterogeneity. Aiming at the time sampling interval difference between InSAR phase data and Beidou GNSS displacement data, minute-level time alignment is achieved through cubic spline interpolation. The space coordinate system conversion module projects the geocoding coordinate system of InSAR data, the WGS84 coordinate system of Beidou GNSS, and the engineering coordinate system of optical fiber strain data onto the regional construction coordinate system. The outlier cleaning process adopts a multi-level filtering strategy, performs wavelet multi-scale decomposition on the Beidou GNSS displacement time series data, separates the low-frequency deformation trend term and the high-frequency vibration noise, and dynamically adjusts the sliding window length according to the signal energy ratio to achieve adaptive noise reduction. The distributed optical fiber strain data eliminates local abnormal jump points through spatial continuity verification, and uses the strain gradient constraint conditions of adjacent sensing nodes to identify the distorted data caused by equipment failures or external interferences.

[0019] When establishing a generalized Kelvin creep constitutive model, an environmental parameter coupling term is added to the classical viscoelastic theory. The model input layer integrates preprocessed displacement and strain time series data, and the hidden layer introduces temperature, humidity, and pore water pressure as dynamic correction factors. A long-short-term memory neural network is used to learn the correlation between environmental parameters and creep rate online, adjusting the viscoelastic coefficient and delay time parameters in real time. The model training utilizes a transfer learning strategy, using training weights from typical geological regions as initialization parameters. Fine-tuning and optimization are performed based on local monitoring data to enhance the model's adaptability to regional geological characteristics.

[0020] When deploying the federated learning framework, a cloud-based aggregation server and multiple edge computing nodes are set up. Each edge node stores local monitoring data and trains a local creep model, using a homomorphic encryption algorithm to protect the privacy of the model's gradient parameters. The cloud server periodically initiates model aggregation instructions, receives the encrypted gradients uploaded by each node, performs a weighted average operation, generates global model parameters, and issues updates. During training, a data distribution feature comparison mechanism is established. Anomalous nodes are identified through statistical analysis of the KL divergence between node data and the global model, and the aggregation weight distribution strategy is dynamically adjusted. The resulting regional adaptive prediction model can simultaneously output the immediate deformation state and long-term creep trend of the soft soil foundation, providing a decision-making basis for engineering safety assessments.

[0021] In some embodiments, the method of constructing the atmospheric delay compensation function in S2 includes: The total atmospheric water vapor parameters and ionospheric total electron content parameters of the monitoring area were extracted based on ERA5 reanalysis data; The atmospheric delay compensation function is constructed by combining the residual phase noise, total water vapor parameter, total electron content parameter and the corresponding regional atmospheric parameter compensation coefficient.

[0022] Specifically, the tropospheric temperature and humidity profiles and ionospheric TEC maps can be extracted based on the ERA5 reanalysis data to construct a phase delay compensation function: ; in, is the phase delay, PWV is the total amount of atmospheric water vapor, TEC is the total electron content of the ionosphere, k1 and k2 are compensation coefficients, is the residual phase noise.

[0023] When implementing atmospheric delay compensation, first obtain ERA5 reanalysis data from the weather forecast center and extract spatio-temporally continuous atmospheric parameters covering the monitoring area. By parsing the geopotential height and relative humidity vertical profiles in the reanalysis dataset, integrate along the satellite radar wave propagation path to calculate the total atmospheric water vapor content PWV value. Synchronously obtain the total electron content TEC grid data of the ionosphere, and use the bilinear interpolation algorithm to map it to the InSAR satellite imaging geometric space to eliminate the influence of the gradient change of ionospheric delay with spatial position.

[0024] During the construction of the phase delay compensation function, aiming at the physical correlation between the atmospheric delay component and meteorological elements in the InSAR interferometric phase, establish a regional parameter calibration mechanism. Through regression analysis of historical monitoring data, perform multiple linear fitting on the InSAR phase residuals in the rainy and dry seasons of the same region and the ERA5 meteorological parameters, and solve to obtain the regional atmospheric parameter compensation coefficients k1 and k2. Among them, the PWV component reflects the tropospheric wet delay effect, and the TEC component characterizes the ionospheric group delay characteristics. The two form a deterministic model compensation amount of atmospheric delay through weighted superposition.

[0025] Residual phase noise The processing of is carried out by combining spatial filtering and time series analysis. Perform wavelet multi-resolution decomposition on the compensated phase residuals, extract the low-frequency phase components related to terrain height for terrain-related atmospheric correction. The high-frequency residual components are suppressed by constructing an adaptive filter, and the filter parameters are dynamically adjusted according to the atmospheric state stability within the satellite revisit period. For ionospheric sudden disturbances, establish a TEC change rate threshold monitoring mechanism. When it is detected that the time derivative of TEC exceeds the preset warning value, automatically trigger the real-time update process of the compensation coefficient k2.

[0026] The dynamic optimization of the regional atmospheric parameter compensation coefficient is realized through an online learning module. In the area covered by the Beidou GNSS reference station, cross-validate the zenith tropospheric delay retrieved by GNSS and the PWV value calculated by ERA5, and calculate the linear correlation coefficient between the two as the correction factor of k1. Synchronously use the TEC observation values of the ionospheric piercing points of the dual-frequency GNSS receiver to perform spatial matching degree analysis with the TEC map provided by ERA5, and adjust the regional weighted weight of k2 according to the matching error distribution. Through the above mechanism, ensure that the compensation model can adapt to the seasonal and sudden changes of the atmospheric environment.

[0027] The application of the compensation model adopts a phased processing strategy. At the front end of InSAR data processing, use Δφ atmThe function performs the initial atmospheric correction on the original interferometric phase to eliminate the large-scale spatially correlated delay components. At the back end of deformation inversion, the residual phase is iteratively optimized by combining the prediction results of the deformation model, and the coupling parameters of the deformation rate and the residual atmospheric phase are solved through least squares adjustment. The finally output deformation monitoring data is verified for accuracy by comparing with the measured displacement values of the Beidou GNSS reference station, forming a closed-loop quality control system for atmospheric delay compensation.

[0028] In some embodiments, the outlier cleaning in S2 includes the following steps: Perform wavelet multi-scale decomposition on the Beidou GNSS displacement time series data to separate the low-frequency deformation signal and the high-frequency noise component; Based on the ratio relationship between the signal-to-noise ratio of the current window and the reference signal-to-noise ratio, dynamically adjust the length of the sliding filter window; Use the dynamically adjusted sliding filter window to perform adaptive threshold filtering on the high-frequency noise component and retain the low-frequency deformation signal.

[0029] Specifically, the outlier cleaning in S2 uses a dynamic sliding window filtering algorithm: Decompose the Beidou GNSS displacement data into a low-frequency deformation signal and high-frequency noise through wavelet transform, and dynamically adjust the length of the sliding window according to the signal-to-noise ratio: ; where, L w represents the length of the sliding window, SNR current is the current signal-to-noise ratio, SNR base is the reference signal-to-noise ratio, and L base is the initial window length.

[0030] When performing outlier cleaning on the Beidou GNSS displacement data, first perform preprocessing of multipath effect suppression and cycle slip repair on the original observation data. Use dual-frequency carrier phase observations to construct a geometry-free combination observation equation, detect cycle slip events through statistical analysis of the carrier phase change rate within the sliding window, and use the Kalman smoothing algorithm to repair the phase breakpoints. After completing the data integrity check, input the displacement time series into the wavelet transform module, select the Daubechies wavelet basis function for multi-scale decomposition, and decompose the signal into a low-frequency component representing the creep trend of the ground and a high-frequency component containing measurement noise and high-frequency vibrations.

[0031] The number of wavelet decomposition levels is determined based on the relationship between the signal sampling frequency and the geological creep characteristic frequency. For the Beidou GNSS displacement data sampled at the minute level, a five-layer decomposition structure is used to divide the signal frequency band into the main creep frequency band below 0.5 Hz and the noise frequency band above 0.5 Hz. By calculating the proportion of signal energy in each frequency band, the current signal-to-noise ratio level is dynamically evaluated. When it is detected that the proportion of energy in the high-frequency sub-band exceeds the preset ratio threshold, it is determined that the current data is greatly affected by instantaneous interference noise.

[0032] The calculation of the dynamic sliding window length follows the formula The implementation logic. The initial window length L base is set according to the typical vibration period of the engineering site, and the reference signal-to-noise ratio SNR base is calibrated by the ratio of the energy of the low-frequency component to the energy of the high-frequency component in the historical pure data set. The instantaneous ratio of the energy of the low-frequency component to the high-frequency noise energy in the current sliding window is calculated in real time as SNR current . When the signal-to-noise ratio decreases, the window length is extended proportionally to enhance the filtering and smoothing effect. When the signal-to-noise ratio rebounds, the window is shrunk to retain the effective high-frequency details.

[0033] The sliding window filtering adopts an improved Savitzky-Golay convolution algorithm to fit a quadratic polynomial within the window to achieve trend smoothing. During the adjustment of the window length, the boundary extension algorithm is used to avoid signal distortion at the head and tail ends of the data sequence. For the extended long window, a segmented weighting strategy is adopted to reduce the weight ratio of the edge data points and suppress the filtering ringing effect caused by window mutations. After the high-frequency noise component is processed by threshold filtering, it is fused with the low-frequency trend component through the wavelet reconstruction algorithm to generate the denoised displacement time series data.

[0034] The filtering effect is verified by a multi-sensor cross-comparison mechanism, and the consistency test of the time-domain integration operation is carried out on the processed Beidou GNSS displacement data and the distributed fiber optic strain data at the same point. When the difference in the accumulated deformation amount within the sliding window between the two exceeds the allowable range, the re-optimization process of the window length parameter is triggered, and L base The reference value is dynamically corrected through the feedback adjustment mechanism. At the same time, a noise energy spectrum feature library is established to perform pattern recognition and classification on the high-frequency noise component, distinguish different types of noise sources such as equipment inherent noise, vehicle vibration interference, and rainfall impact, and provide environmental context information for the window adjustment strategy.

[0035] In some embodiments, the outlier cleaning in S2 further includes a real-time dynamic threshold adjustment mechanism: Based on the spatial distribution characteristics of the strain gradient of adjacent distributed fiber optic sensing nodes, a strain gradient safety constraint threshold is established in combination with the node spacing, soil shear strength, and elastic modulus; When it is detected that the strain change rate of an adjacent node exceeds the safety constraint threshold, the following processing is performed: the time coverage of the current sliding filter window is automatically extended, and the window expansion amplitude is dynamically adjusted based on the duration of the anomaly; the Beidou GNSS displacement data and InSAR deformation inversion results of the corresponding period are synchronously integrated, and the strain measurement values in the abnormal area are compensated through a multi-source data joint filtering algorithm.

[0036] Specifically, outlier cleaning in S2 also includes a real-time dynamic threshold adjustment mechanism: Based on the spatial continuity characteristics of distributed optical fiber strain data, the strain gradient constraint equation of adjacent sensor nodes is constructed: ; in, represents the strain gradient, ε i , ε j is the strain value of the adjacent node, d i,j is the node spacing, σ ult is the ultimate shear strength of soil, E soil is the elastic modulus, α is the safety factor, represents the strain gradient threshold; When the strain gradient exceeds the threshold, the following actions are triggered: 1) Automatically extend the sliding window length to , where β is the time decay factor, Δt is the duration of the abnormality, Indicates the adjusted sliding window length; 2) Aggregate BeiDou GNSS displacement data and InSAR phase data, and perform data compensation through Kalman filtering.

[0037] When implementing anomaly detection for distributed fiber-optic strain data, a strain gradient constraint network based on spatial continuity is first established. Sensing nodes are deployed at equal intervals along the fiber-optic sensing cable. The spacing between adjacent nodes is determined based on the uniformity of the engineering geology, and the node density is increased in areas of sudden soil changes. Each node collects axial strain values in real time. Data exchange channels are established between adjacent nodes via internal communication links within the fiber-optic cable, forming a local topological structure for strain gradient calculation.

[0038] The physical meaning of the strain gradient constraint equation is to quantify the strain change rate of adjacent nodes. When calculating, the average strain of three consecutive sampling moments is taken as the current ε i , ε j Input value, node spacing d i,j The precise value is obtained through the construction record of the optical cable laying. The threshold setting stage introduces a dynamic correction mechanism for soil mechanical parameters, and the ultimate shear strength of the soil obtained from geological survey is converted to ult and elastic modulus E soilThe input calculation module adjusts the value of the safety factor α in combination with the real-time pore water pressure monitoring value. When the pore water pressure rises and causes the effective stress of the soil to decrease, the value of α is automatically reduced to tighten the strain gradient threshold.

[0039] The anomaly triggering mechanism adopts a multi-condition joint decision strategy. When the strain gradients of a single node and more than two adjacent nodes exceed the limit simultaneously, it is determined that there is a real abnormal deformation in this area; if only the gradient of the unilateral node exceeds the limit, the device self-check program is started to verify the working state of the fiber optic demodulator. After confirming the data anomaly, the sliding window length extension operation is performed. The time decay factor β in the window increment ΔL = β·Δt is dynamically adjusted according to the duration of the anomaly: a larger β value is adopted in the initial anomaly stage to achieve rapid window expansion, and the β value decays exponentially with time in the continuous anomaly state to avoid excessive window expansion.

[0040] The extended sliding window A multi-source data fusion compensation mechanism is introduced in the data filtering process. The Beidou GNSS displacement data and InSAR deformation inversion results in the same area during the abnormal period are synchronously retrieved, and the three types of data are unified to the same spatio-temporal reference through the spatio-temporal registration module. The state equation of the Kalman filter is constructed as a deformation rate-acceleration model, and the observation equation integrates the three observation values of the fiber optic strain integral displacement, GNSS displacement, and InSAR deformation. The filtering gain matrix dynamically allocates weights according to the confidence levels of each data source: the weight ratio of the fiber optic data is reduced during the abnormal period of the fiber optic data, and the compensation effects of the GNSS and InSAR data are enhanced.

[0041] A residual convergence monitoring mechanism is established during the data compensation process. When the norm of the Kalman filter innovation vector exceeds the stable range for multiple consecutive periods, the sensor cross-calibration process is triggered. The redundant node data of the distributed optical fiber is used to reconstruct the strain field of the abnormal area, and the optimal strain distribution is solved by the least squares fitting method. The reconstructed result is compared with the original observation data to generate the calibration compensation amount. The calibrated strain data is re-input into the strain gradient constraint equation for secondary verification until the system determines that the anomaly is eliminated and the normal sliding window length is restored. The closed-loop control logic formed in this process effectively maintains the consistency of the multi-source data system and avoids the cascade error diffusion caused by local anomalies.

[0042] In some embodiments, the environmental coupling function in S3 is defined as: ; where, f env (T, W, P) is the environmental coupling function, T represents temperature, W represents humidity, P represents pore water pressure, W sat is the saturated humidity, m1, m2, and m3 are coupling coefficients dynamically updated through the LSTM network, t represents time, represents the pore water pressure gradient.

[0043] Environmental coupling function f env In it, through the coefficients m1 (s - ¹), m2 (dimensionless), m3 (m -1 ), the dimensional consistency of each component is achieved, and the final output unit is strain rate (s - ¹), which is compatible with the constitutive equation of the generalized Kelvin model.

[0044] When constructing the environmental coupling function, a multi-physical field parameter synchronous acquisition system is first established. The temperature parameter is obtained through a thermistor array buried at different depths in the soil. A vertical temperature measurement chain is arranged according to the layered structure of the soft soil foundation to synchronously measure the surface atmospheric temperature and the underground temperature gradient. The humidity parameter is jointly observed by a dielectric constant sensor and a matrix suction sensor. The former monitors the volumetric water content of the soil, and the latter measures the water potential energy in the unsaturated zone through a tensiometer. The pore water pressure is obtained through a vibrating wire piezometer network, forming a three-dimensional observation system of the groundwater level at the monitoring section, and performing spatial interpolation correction in combination with the water content distribution map inverted by the ground penetrating radar.

[0045] The physical meaning of the environmental coupling function is to quantify the time-varying influence of environmental excitation on soil creep. The temperature change rate term is calculated by the difference method, and the five-point central difference format is used to suppress the interference of measurement noise, reflecting the transient influence of the thermal expansion and contraction effect on the viscoelasticity of the soil. The humidity ratio term ln(W / W sat ) The saturated humidity W sat is dynamically calculated through the soil-water characteristic curve and non-linearly corrected in combination with the porosity parameter monitored in real time, accurately characterizing the regulation effect of the water content change on the effective stress. The pore water pressure gradient is calculated by the finite element interpolation method, and the pressure field contour map is constructed based on the piezometer network data, and the maximum principal stress direction gradient value at the monitoring point is extracted.

[0046] The long short-term memory neural network is used to dynamically update the coupling coefficients m1, m2, and m3. The input layer of the network includes the creep rate observation value, the time series of environmental parameters, and the soil constitutive parameters. The hidden layer is designed as a two-layer gated recurrent unit structure. The first layer extracts the short-term fluctuation correlation characteristics between environmental parameters and creep rate, and the second layer captures the long-term dependence relationship between geological parameters and coupling effects. The output layer generates the normalized weight distribution of the coupling coefficients through the Softmax function, and maps it to the actual physical dimension range through anti-normalization processing. In the training stage, a transfer learning strategy is adopted, and the network weights pre-trained with laboratory triaxial creep test data are used as the initial values, and online fine-tuning is performed in combination with on-site monitoring data.

[0047] The real-time update mechanism is implemented through a sliding time window. In each prediction cycle, the latest environmental monitoring data and creep observation values are input into the LSTM network to calculate the incremental coupling coefficients Δm1, Δm2, and Δm3. The momentum factor is introduced in the update process to suppress parameter mutations, and the exponential weighted moving average algorithm is used to smooth the coefficient change curve to prevent the model from oscillating caused by local abnormal data. At the same time, a coefficient change trajectory tracking module is established. When it is detected that the m2 value continuously deviates from the historical mean by more than three standard deviations, the humidity sensor calibration process is automatically triggered to ensure the dimensional consistency of the environmental coupling term.

[0048] The model parameter correction process is deeply coupled with the constitutive equation solver. In each iteration step of the generalized Kelvin model, the creep rate calculated at the current moment is compared with the output value of the environmental coupling function, and the relaxation time parameter of the viscoelastic element is adjusted through an adaptive step size algorithm. When the environmental mutation causes a jump in the output magnitude of the coupling function, the implicit Euler method is automatically switched to solve the differential equation to enhance the numerical stability of the model in a non-steady environment. The corrected model parameters are verified in a closed loop by comparing the integrated results of the fiber optic strain data with the GNSS displacement monitoring values. When the residual exceeds the limit, the retraining process of the LSTM network is triggered.

[0049] In some embodiments, the implementation of the federated learning framework in S4 includes: S41. Each edge node trains a sub-model of the generalized Kelvin model based on local data, and performs noise perturbation processing on the model gradient parameters using the differential privacy encryption algorithm; S42. The cloud aggregation node calculates the weighted average of the global model parameters based on the encrypted gradient parameters uploaded by each node, using the proportion of the monitoring data volume of each node in the total data volume as the weighting coefficient, to achieve distributed joint update of the cross-regional model parameters.

[0050] Specifically, the implementation of the federated learning framework in S4 includes: S41. Each edge node trains a sub-model based on local data and encrypts the gradient parameters using differential privacy; S42. The cloud aggregation node updates the global model parameters by weighted averaging the aggregated gradients: ; where, is the updated global parameter model, i represents the i-th node, N is the total number of nodes, is the local model gradient parameter, n i is the data volume of the i-th node, n total is the total data volume.

[0051] In the deployment stage of the federated learning framework, a distributed architecture of a cloud coordination server and multiple edge computing nodes is established. Each edge node is deployed in the engineering area data center, equipped with an independent data storage unit and a model training engine, storing the deformation monitoring data of the local soft soil foundation and the historical database of environmental parameters. When the training cycle starts, the edge node downloads the current global model parameters from the cloud as the initial weights, and uses the local dataset for forward propagation and backpropagation calculations to generate a gradient tensor matrix. The gradient parameters are processed by differential privacy before leaving the station, and privacy protection is achieved by adding noise perturbations that satisfy the Laplace distribution. The noise amplitude is dynamically adjusted according to the data sensitivity level to ensure that attackers cannot reverse-derive the original monitoring data.

[0052] The cloud server sets the model aggregation trigger condition. When the number of online edge nodes reaches the minimum collaborative training threshold or reaches the preset time window, a global aggregation instruction is initiated. Each node synchronously uploads the encrypted local gradient parameters and the data distribution feature vector. The feature vector includes the mean vector, covariance matrix of the local dataset, and the KL divergence value from the global model prediction result. The cloud verification module parses the feature vector and performs a two-level node screening: at the first level, calculate the ratio of the KL divergence value to the variance of each node to filter out abnormal nodes with significantly deviated data distributions from the mainstream; at the second level, compare the node data mean with the three-standard deviation range of the global distribution to eliminate outlier nodes with systematic biases.

[0053] When performing weighted average operations on the gradient parameters of the trusted node set, an exponential decay adjustment strategy is used to optimize the weight allocation. The initial weights are determined according to the proportion of each node's data volume, and at the same time, a KL divergence penalty factor is introduced to dynamically decay the weights. The weight coefficient in the aggregation formula is modified to a product function of the original data volume weight and the divergence exponent, ensuring that nodes with good data quality and consistent distributions play a dominant role in model updates. The aggregation process uses a tensor operation protocol in the homomorphic encryption state to complete the weighted summation operation in the ciphertext space, avoiding the interception and parsing of gradient plaintext during transmission.

[0054] After the global model parameters are updated, they are sent to all edge nodes participating in the training through a version control mechanism. After receiving the new parameters, the node performs a local model hot update, retaining some historical parameters as the base model for ensemble learning, and balancing global consistency and regional characteristics through a voting mechanism. For edge nodes with special geological conditions, a regularization constraint term is introduced in the local training stage to limit the model update amplitude not to exceed the elastic boundary of the global parameters, preventing overfitting to regional noise. At the same time, a model degradation monitoring mechanism is established. When the accuracy of the local validation set continuously decreases, the global model rollback process is triggered, and the historical optimal parameters are loaded to restore prediction stability.

[0055] During the training process, a multi-party secure computing channel is constructed. The cloud only obtains the gradient statistical features, and the edge nodes cannot peek at the data distributions of other participants. The comparison of data feature vectors adopts a secure multi-party computing protocol, and the joint calculation of the KL divergence value is realized through a garbled circuit. The participants complete the screening of trusted nodes without disclosing the details of local data. For abnormal situations such as communication interruptions, a model parameter caching and breakpoint resuming mechanism is designed to ensure that the federated learning process has the ability to resist network fluctuations. The finally formed regional adaptive deformation prediction model, while ensuring the privacy of monitoring data, integrates multi-regional engineering experience and knowledge, and significantly improves the generalization prediction accuracy for regions not participating in training.

[0056] In some embodiments, before S42, it further includes the verification of the credibility of edge nodes: When each edge node uploads the gradient parameters, it synchronously submits the mean, variance of the local data set, and the difference degree index from the global data distribution; The cloud selects a set of trusted nodes that meet the consistency constraints based on the ratio threshold of the difference degree index to the data dispersion degree, and the deviation tolerance range of the local mean from the global mean distribution; When weighted aggregating the gradient parameters of the set of trusted nodes, the data volume weight is exponentially decayed and corrected based on the data distribution difference degree of each node to generate the global model parameter update amount.

[0057] Specifically, before the gradient aggregation in S42, the verification of the credibility of edge nodes is added: 1) Each node uploads the gradient parameters and synchronously submits the local data distribution feature vector , where is the local data mean, is the local data variance, and KL i is the KL divergence between the local data distribution and the global data distribution; 2) The cloud selects a set of trusted nodes through the following formula : ; where δ is the threshold of the ratio of KL divergence to variance, is the tolerance multiple of the mean deviation, μ g , σ g are the mean and standard deviation of the global data distribution; 3) Aggregate the gradients of the nodes in and dynamically adjust the weighting coefficient to , where λ here is the attenuation coefficient, KL i is the KL divergence, and the final global model parameter formula is .

[0058] When implementing the credibility verification of edge nodes, first establish a standardized generation process for the data distribution feature vector. After the local training cycle of each edge node ends, extract the time-series statistics of displacement, strain, and environmental parameters from the preprocessed monitoring dataset. Mean vector Calculate through moving window mean filtering to retain the data trend characteristics; variance Calculate after eliminating the influence of seasonal variations using detrended fluctuation analysis; KL divergence value KL i By comparing the local data distribution with the global distribution histogram sent from the cloud, use kernel density estimation method to generate the probability density function and calculate the relative entropy value.

[0059] Feature vector is transmitted using a lightweight encapsulation protocol, and an independent data segment is opened in the gradient parameter upload channel for carrying. After receiving the feature vector, the cloud verification module starts a multi-dimensional joint analysis process: first, normalize the KL divergence value by dividing it by the local variance to eliminate the influence of data scale differences on the discrimination threshold. The preset δ threshold is dynamically adjusted according to the 95th percentile of the KL / σ² ratio distribution of normal nodes in the historical training cycle to ensure that the screening conditions adapt to the gradual changes in the data distribution.

[0060] The mean deviation check uses a sliding reference window mechanism. The cloud maintains rolling update queues for the global data mean μ g and standard deviation σ g . After each aggregation, retain the statistics of the last N cycles, and calculate the current reference value using the exponential weighting method. When the absolute difference between the node mean μ[[ID=2Y]] i and μ g exceeds[[ID=2Y]] times σ g , trigger the deviation root cause analysis: if the deviation direction is consistent with the geological zoning law, it is determined as a reasonable regional characteristic and relax the value; if the deviation shows randomness, it is classified as data anomaly.

[0061] The generation of the trusted node set adopts a combination strategy of hard and soft decisions. Initially, retain candidate nodes through threshold comparison, and then perform clustering analysis on the candidate node set to identify whether there are regional node clusters. Conduct a secondary verification on isolated nodes, and perform a spatial correlation test on their data distribution characteristics with adjacent geographical nodes to exclude pseudo-abnormal nodes caused by communication failures. The finally formed trusted node list is stored on the blockchain to ensure the traceability of the screening process.

[0062] In the dynamic weight adjustment link, introduce a divergence decay factor, and combine the original data volume weight n i with exp(-λ·KL i)Multiplication is used to achieve punitive weight reduction. The attenuation coefficient λ is adaptively adjusted according to the convergence state of the global model: a larger λ value is adopted at the initial stage of training to accelerate convergence, and the λ value is reduced in the later stage to retain regional characteristics. When performing weight normalization, weight compensation is applied to regions with fewer edge nodes to prevent the model from being biased towards high-frequency monitoring regions due to data volume differences.

[0063] The fault-tolerant synchronization mechanism is adopted in the process of gradient aggregation of trusted nodes, and a gradient reception timeout window and a data verification retransmission request are set. For nodes with partial gradient loss, the parameter matrix is completed through the historical gradient interpolation algorithm to ensure the integrity of aggregation. At the same time, a blacklist mechanism is established, and nodes that fail to pass verification for multiple consecutive cycles trigger an offline diagnosis process, and the operation and maintenance terminal intervenes for device calibration or data traceability. Through the above mechanism, a closed-loop quality control system is formed to maintain the reliability of model updates while ensuring the efficiency of federated learning.

[0064] In some embodiments, to further improve the reliability of data fusion, the embodiments of the present invention introduce a multi-source data cross-verification method based on a variational autoencoder: A deep generative model is constructed, with the Beidou GNSS displacement time series, the spatial distribution of fiber optic strain, and the InSAR deformation inversion result as the combined input. After extracting the latent feature vector through the encoder, the decoder reconstructs the physical quantities of each data source. Define the reconstruction loss function: ; where is the reconstruction loss function, and x here i can be x1, x2, x3, respectively representing three types of data (Beidou GNSS displacement time series, spatial distribution of fiber optic strain, and InSAR deformation inversion result), is the reconstructed output data vector, is the original output data vector, is the dynamic weight coefficient, is the squared error term. When the reconstruction error of any data source exceeds 2 times the historical standard deviation, the weight reduction process of this data source is automatically triggered, and the redundant sensor data compensation process is started.

[0065] In some embodiments, the physical-data dual-drive correction of the creep model can also be introduced. Specifically: Embed a physics-informed neural network (PINN) based on the generalized Kelvin model: Integrate the constitutive equation as a hard constraint into the network structure to construct a hybrid loss function: ; where is the hybrid loss function, is the mean square error between the predicted value and the measured value, It is the residual term of the control equation, and λ1 and λ2 are coefficients. The ratios of λ1 and λ2 are dynamically adjusted by the adaptive weight algorithm. When the environmental parameters mutate and cause an increase in data noise, the physical constraint weight is increased to more than 0.7 to ensure that the model evolution conforms to the basic laws of soil mechanics.

[0066] Figure 2 The following is a schematic structural diagram of a soft soil foundation deformation prediction system based on big data provided by an embodiment of the present invention, including: An acquisition module 100, configured to acquire deformation monitoring data of the soft soil foundation through a multi-source sensing network, including InSAR satellite phase data, Beidou GNSS displacement time series data, distributed optical fiber strain data, meteorological data, and geological exploration parameters; A generation module 200, configured to perform spatio-temporal alignment and outlier cleaning on the deformation monitoring data, construct an atmospheric delay compensation function using ERA5 reanalysis data, and generate a preprocessed standardized data set; A correction module 300, configured to construct a generalized Kelvin creep constitutive model based on the standardized data set, introduce an environmental coupling function to dynamically correct the model parameters, where T is temperature, W is humidity, and P is pore water pressure; A prediction module 400, configured to perform distributed joint training on the generalized Kelvin creep constitutive model using a federated learning framework, optimize the data characteristics of different geological zones through transfer learning, and generate a regionally adaptive deformation prediction result.

[0067] The system provided by the embodiment of the present invention has the same technical features as the above method, so it can solve the same technical problems and achieve the same technical effects, which will not be elaborated here.

[0068] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the deformation of soft soil foundation based on big data, characterized in that, Including: S1. Obtain deformation monitoring data of soft soil foundation through a multi-source sensing network, including InSAR satellite phase data, Beidou GNSS displacement time series data, distributed optical fiber strain data, meteorological data, and geological exploration parameters; S2. Perform spatio-temporal alignment and outlier cleaning on the deformation monitoring data, construct an atmospheric delay compensation function using ERA5 reanalysis data, and generate a preprocessed standardized data set; S3. Construct a generalized Kelvin creep constitutive model based on the standardized dataset and introduce an environmental coupling function Dynamically correct the model parameters, where T is temperature, W is humidity, and P is pore water pressure; S4. Use a federated learning framework to perform distributed joint training on the generalized Kelvin creep constitutive model, optimize data characteristics of different geological zones through transfer learning, and generate regional adaptive deformation prediction results.

2. The method according to claim 1, wherein The method for constructing the atmospheric delay compensation function in S2 includes: Extract the total atmospheric water vapor parameter and the total electron content parameter of the ionosphere in the monitoring area based on ERA5 reanalysis data; Combine the residual phase noise, the total water vapor parameter, the total electron content parameter, and the corresponding regional atmospheric parameter compensation coefficient to construct the atmospheric delay compensation function.

3. The method according to claim 1, wherein The outlier cleaning in S2 includes the following steps: Perform wavelet multi-scale decomposition on the Beidou GNSS displacement time series data to separate the low-frequency deformation signal and the high-frequency noise component; Dynamically adjust the sliding filter window length based on the ratio relationship between the signal-to-noise ratio of the current window and the reference signal-to-noise ratio; Perform adaptive threshold filtering on the high-frequency noise component using the dynamically adjusted sliding filter window to retain the low-frequency deformation signal.

4. The method according to claim 3, wherein The outlier cleaning in S2 also includes a real-time dynamic threshold adjustment mechanism: Based on the strain gradient spatial distribution characteristics of adjacent distributed optical fiber sensing nodes, establish a strain gradient safety constraint threshold in combination with the node spacing, soil shear strength, and elastic modulus; When it is detected that the strain change rate of adjacent nodes exceeds the safety constraint threshold, perform the following processing: automatically extend the time coverage range of the current sliding filter window, and dynamically adjust the window expansion amplitude based on the abnormal duration; synchronously fuse the Beidou GNSS displacement data and the InSAR deformation inversion result of the corresponding period, and compensate the strain measurement value of the abnormal area through a multi-source data joint filtering algorithm.

5. The method according to claim 1, characterized in that The environmental coupling function in S3 is defined as: ； Among them, f env (T, W, P) is the environmental coupling function, where T represents temperature, W represents humidity, P represents pore water pressure, and W sat is the saturation humidity, m1, m2, and m3 are coupling coefficients dynamically updated through the LSTM network, t represents time, represents the pore water pressure gradient.

6. The method according to claim 1, wherein The implementation of the federated learning framework in S4 includes: S41. Each edge node trains a generalized Kelvin model sub-model based on local data, and performs noise perturbation processing on the model gradient parameters using the differential privacy encryption algorithm; S42. The cloud aggregation node calculates the weighted average value of the global model parameters based on the encrypted gradient parameters uploaded by each node, with the proportion of the monitoring data volume of each node in the total data volume as the weighting coefficient, to achieve distributed joint update of cross-regional model parameters.

7. The method according to claim 6, wherein Before S42, it also includes edge node credibility verification: Each edge node synchronously submits the mean value, variance, and the difference degree index from the global data distribution of the local data set when uploading the gradient parameters; The cloud selects a credible node set that meets the consistency constraints based on the ratio threshold between the difference degree index and the data dispersion degree, and the deviation tolerance range between the local mean value and the global mean value distribution; When performing weighted aggregation on the gradient parameters of the trusted node set, an exponential decay correction is made to the data volume weight based on the difference degree of the data distribution of each node to generate a global model parameter update amount.

8. A soft soil foundation deformation prediction system based on big data, characterized in that, Including: An acquisition module, configured to acquire deformation monitoring data of a soft soil foundation through a multi-source sensing network, including InSAR satellite phase data, Beidou GNSS displacement time series data, distributed optical fiber strain data, meteorological data, and geological exploration parameters; A generation module, configured to perform spatio-temporal alignment and outlier cleaning on the deformation monitoring data, construct an atmospheric delay compensation function by using ERA5 reanalysis data, and generate a preprocessed standardized data set; A correction module, configured to construct a generalized Kelvin creep constitutive model based on the standardized data set and introduce an environmental coupling function dynamically correct model parameters, where T is temperature, W is humidity, and P is pore water pressure; A prediction module, configured to perform distributed joint training on the generalized Kelvin creep constitutive model by using a federated learning framework, optimize the data characteristics of different geological zones through transfer learning, and generate a regionally adaptive deformation prediction result.

Images