Intelligent monitoring method for mucky soil foundation based on multi-source data fusion

Abstract

The invention discloses a mucky soil foundation intelligent monitoring method based on multi-source data fusion, and particularly relates to the technical field of foundation monitoring and early warning. The method comprises the following steps: constructing an original soil body state data set of a mucky soil foundation, and carrying out anomaly elimination and time calibration based on a rheological constitutive model to generate a foundation state characteristic matrix; performing dynamic causal association tracking analysis on the feature matrix in a sliding window mode to generate a water-soil coupling drift parameter set; extracting a stress historical dependency coefficient set by using a nonlinear scale analysis method; and according to the water-soil coupling drift parameter set and the stress historical dependency coefficient set, the stability risk level of the mucky soil foundation is judged, and corresponding early warning information is selected from a preset graded early warning instruction database and output, so that the accuracy and adaptability of anomaly recognition and risk early warning are improved, and the accuracy and the adaptability of abnormal recognition and risk early warning are improved. The method is suitable for long-period stability monitoring and intelligent risk prevention and control of high-soft foundations or silt stratums.

Description

Technical Field

[0001] This invention relates to the field of foundation monitoring and early warning technology, and more specifically, to a method for intelligent monitoring of silty soil foundations based on multi-source data fusion. Background Technology

[0002] Existing monitoring methods for silty soil foundations mostly employ static criterion models based on a single parameter or rule setting.

[0003] The monitoring parameters such as pore water pressure, stratified settlement, and lateral displacement within silty foundations exhibit nonlinear, unsteady, and coupled drift characteristics. Traditional methods cannot capture these dynamic relationships, resulting in low accuracy in anomaly identification and insufficient early warning timeliness, making it difficult to meet the needs of long-term stability monitoring of soft foundations.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an intelligent monitoring method for silty soil foundations based on multi-source data fusion to address the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The intelligent monitoring method for silty soil foundations based on multi-source data fusion includes the following steps: S1: Simultaneously collect pore water pressure, stratified settlement, and lateral displacement data of silty soil foundation to generate original soil state dataset; S2: The foundation state feature matrix is ​​obtained by performing rheological constitutive cleaning and time calibration on the original soil state dataset; S3: Perform dynamic causal correlation tracing analysis on the foundation state characteristic matrix to generate a set of water-soil coupling drift parameters; S4: By performing nonlinear scaling analysis on the multi-parameter coordinated drift of the water-soil coupling drift parameters, a set of stress history dependence coefficients characterizing the reverse correlation of anomalies is obtained. S5: Perform dynamic response weight fusion analysis based on rheological state zoning on the set of water-soil coupling drift parameters and the set of stress history dependence coefficients to determine the stability risk level of silty soil foundation; S6: Construct graded early warning instructions based on the stability risk level of silty soil foundations and output early warning information.

[0007] In a preferred embodiment, S1 specifically refers to: A combined configuration of pore water pressure equipment, stratified settlement monitoring devices, and deep lateral displacement measurement equipment deployed at different depths and directions within silty soil foundations; By setting a unified sampling period, various monitoring devices are synchronously triggered to collect raw measurement data of pore water pressure, stratified settlement and deep lateral displacement respectively. The original measurement data of various types are divided and labeled according to their spatial location, and the spatial coordinates, sampling time and monitoring equipment number of each sampling point are recorded to form the original soil state dataset.

[0008] In a preferred embodiment, S2 specifically refers to: Data cleaning of the original soil state dataset was performed using a rheological constitutive model to remove outliers and interference terms from the pore water pressure data, stratified settlement data, and deep lateral displacement data. The sampling time of each sampling point is accurately calibrated using a unified standard time calibration method; After data cleaning and sampling time calibration, the pore water pressure data, stratified settlement data, and deep lateral displacement data are recombined according to spatial coordinates to generate a foundation state characteristic matrix.

[0009] In a preferred embodiment, S3 specifically refers to: Using the foundation state feature matrix as input, the multivariate time series is divided into sliding windows of fixed duration, so that each window contains equal amounts of pore water pressure data, stratified settlement data, and deep lateral displacement data. The two-way Granger causality test statistic is calculated sequentially for the monitoring data in each window, and the valid results are filtered by setting a significance threshold to obtain the causality weight matrix; The causal weight matrix obtained by the continuous window is subjected to first-order difference operation in the time dimension to extract the causal weight change rate vector and construct a candidate drift vector sequence. Principal component analysis is performed on the candidate drift vector sequence to reduce dimensionality, and principal components with a cumulative variance contribution rate not lower than a preset threshold are retained to form a set of water-soil coupling drift parameters.

[0010] In a preferred embodiment, S4 specifically refers to: A nonlinear scaling analysis method was used to perform multi-parameter collaborative drift analysis on the principal components in the water-soil coupling drift parameter set, and the collaborative drift index between the principal components was calculated. Based on the collaborative drift index among the principal components, abnormal reverse correlations between drift parameters are identified, and abnormal reverse correlation patterns are marked according to drift direction and amplitude. Statistical analysis was performed on the abnormal reverse correlation patterns, and the stress history dependence coefficients characterizing the abnormal reverse correlation patterns were calculated. The stress history dependence coefficients at each spatial coordinate were then labeled to form a set of stress history dependence coefficients.

[0011] In a preferred embodiment, S5 specifically refers to: By combining the set of water-soil coupling drift parameters with the set of stress history dependence coefficients, the silty soil foundation is divided into several rheological state regions by the rheological state zoning rule. For each rheological state region, the water-soil coupling drift parameters and stress history dependence coefficients within the region are extracted, and the dynamic response weights are calculated based on the extracted values. The dynamic response weights of each rheological state region are weighted and fused according to the weight coefficients of the sampling points in the region to obtain the overall dynamic response weight. The stability risk level of silty soil foundation is determined by comparing the overall dynamic response weight with the preset risk threshold.

[0012] In a preferred embodiment, S6 specifically refers to: Establish a tiered early warning instruction database in advance; Input the stability risk level of silty soil foundation into the graded early warning instruction database, and select the early warning instruction corresponding to the stability risk level of silty soil foundation according to the matching rules; The selected warning instructions are formatted to generate standardized warning information; Standardized early warning information is output to external monitoring terminals through a communication interface.

[0013] The technical effects and advantages of the intelligent monitoring method for silty soil foundation based on multi-source data fusion of this invention are as follows: By simultaneously collecting pore water pressure, stratified settlement, and lateral displacement data of silty soil foundations, unified acquisition of multiple monitoring information can be achieved, improving data integrity and timeliness. Introducing a rheological constitutive model to clean and time-calibrate the original soil state dataset improves data quality and comparability. Employing a dynamic causal correlation tracing method identifies the coupling relationships between monitoring parameters over time. Using nonlinear scaling analysis to extract a set of stress history dependence coefficients characterizing anomalies and reverse correlations, nonlinear reverse change features hidden in complex drift processes can be identified. By merging the water-soil coupled drift parameter set and the stress history dependence coefficient set through zoned response, the pertinence and dynamic adaptability of risk assessment are enhanced, enabling refined classification and judgment of silty foundation stability. By constructing graded early warning commands and outputting early warning information, early warning response efficiency and external system integration capabilities can be improved, comprehensively enhancing monitoring accuracy in complex foundation environments. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of an obstacle avoidance control method for an intelligent robot according to the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] Example Figure 1 The present invention provides an intelligent monitoring method for silty soil foundations based on multi-source data fusion, which includes the following steps: S1: Simultaneously collect pore water pressure, stratified settlement, and lateral displacement data of silty soil foundation to generate original soil state dataset; S2: The foundation state feature matrix is ​​obtained by performing rheological constitutive cleaning and time calibration on the original soil state dataset; S3: Perform dynamic causal correlation tracing analysis on the foundation state characteristic matrix to generate a set of water-soil coupling drift parameters; S4: By performing nonlinear scaling analysis on the multi-parameter coordinated drift of the water-soil coupling drift parameters, a set of stress history dependence coefficients characterizing the reverse correlation of anomalies is obtained. S5: Perform dynamic response weight fusion analysis based on rheological state zoning on the set of water-soil coupling drift parameters and the set of stress history dependence coefficients to determine the stability risk level of silty soil foundation; S6: Construct graded early warning instructions based on the stability risk level of silty soil foundations and output early warning information.

[0017] Simultaneously collect pore water pressure, stratified settlement, and lateral displacement data for silty soil foundations to generate an original soil state dataset, including: A combined configuration of pore water pressure equipment, stratified settlement monitoring devices, and deep lateral displacement measurement equipment deployed at different depths and directions within silty soil foundations; In the silty soil foundation site, geological survey results were used to determine the structural characteristics, aquifer distribution, and potential deformation zones at different depths and directions. Based on this information, pore water pressure monitoring devices, stratified settlement monitoring devices, and deep lateral displacement measurement devices were deployed at multiple depths.

[0018] Pore ​​water pressure equipment is mainly installed in representative saturated soft soil layers underground, ensuring multiple measuring points within the same monitoring profile to collect the vertical interlayer water pressure distribution characteristics; stratified settlement monitoring devices are deployed vertically along the strata, with one measuring point at each depth layer to capture changes in foundation stratified settlement; deep lateral displacement measurement equipment is deployed in areas where horizontal displacement may occur, typically by installing inclinometer tubes in vertical boreholes near slopes, foundation edges, etc., to obtain interlayer displacement information of foundation horizontal deformation.

[0019] All monitoring devices are uniformly numbered during the joint configuration process, and the spatial location coordinates (longitude, latitude, and burial depth) of each monitoring device, as well as basic configuration parameters such as installation angle and response range, are recorded to ensure spatial consistency of data collection.

[0020] By setting a unified sampling period, various monitoring devices are synchronously triggered to collect raw measurement data of pore water pressure, stratified settlement and deep lateral displacement respectively. To ensure temporal consistency among pore water pressure data, stratified settlement data, and deep lateral displacement data, a uniform sampling period is pre-set. For example, a synchronous sampling period in hours is selected based on the foundation response rate. Within the synchronous sampling period, a unified sampling command is issued, triggering all monitoring devices to start measurements at the preset sampling frequency.

[0021] The pore water pressure device records the pore water pressure value at the current depth as a data point; the stratified settlement monitoring device generates settlement data points by recording changes in inter-stratum height; and the deep lateral displacement measurement device collects the current horizontal displacement value at each depth segment in the inclinometer tube. All these data acquisition operations are completed under unified clock control, ensuring that all data records have a consistent sampling timestamp.

[0022] To prevent asynchronous transmission or trigger failure of sampling signals, a communication link connectivity test is performed before each synchronous acquisition, and a redundant triggering mechanism is configured to improve the stability of data synchronization.

[0023] The various types of raw measurement data are divided and labeled according to spatial location, and the spatial coordinates, sampling time and monitoring equipment number of each sampling point are recorded to form the raw soil state dataset; All collected raw measurement data are divided into spatial regions based on the spatial coordinates of the sampling points. The division method is as follows: the monitoring area is divided into multiple spatial partition units, each spatial partition unit corresponds to a coordinate grid block, and each raw data is assigned to the corresponding spatial partition unit according to its sampling point coordinates, forming a region identification label.

[0024] Add annotation fields to each original measurement data, including: spatial coordinate information, equipment number information, equipment type information, sampling time information, and spatial partition number, and summarize to generate the original soil state dataset.

[0025] For example, in a section of a silty soil foundation project, ten monitoring points were set up, located at different depths and in different directions. Each point was equipped with one pore water pressure device, one stratified settlement monitoring device, and one deep lateral displacement measurement device. The sampling cycle was uniformly set to ten minutes. Every ten minutes, thirty devices were simultaneously triggered to collect three types of raw measurement data. The thirty measurement data were sequentially assigned spatial coordinate information, device number information, device type information, sampling time information, and spatial zone number, forming a raw soil state data recording unit for a sampling cycle. Recording units were appended in continuous time sequence to construct the raw soil state dataset.

[0026] By performing rheological constitutive cleaning and time calibration on the original soil state dataset, the foundation state feature matrix is ​​obtained, including: Data cleaning of the original soil state dataset was performed using a rheological constitutive model to remove outliers and interference terms from the pore water pressure data, stratified settlement data, and deep lateral displacement data. Based on the creep and stress relaxation characteristics of silty soil, a rheological constitutive model incorporating elastic, viscous, and viscoelastic combined features was selected. The model describes the stress-strain relationship of the soil under constant and variable loads using a set of physical parameters. In the experimental phase, representative silty soil samples were first extracted in the laboratory, and the basic values ​​of each physical parameter of the model were obtained through triaxial shear tests, constant load stress relaxation tests, and creep tests. The physical parameters include, but are not limited to, elastic element parameters: instantaneous elastic modulus and delayed elastic modulus, used to describe the recoverable deformation capacity of the soil at the instant of load application and within a short period; viscous element parameters: shear viscosity coefficient or bulk viscosity coefficient, used to quantify the viscous flow rate of the soil under sustained load; and viscoelastic combined parameters: viscoelastic coupling modulus, viscoelastic damping coefficient, and the corresponding hysteresis (or relaxation) time constant, used to characterize the time-varying stress-strain response under the combined effects of elastic recovery and viscous hysteresis.

[0027] The above parameters are determined by back-calculation from the instantaneous stress-strain curve obtained from the triaxial shear test, the stress decay trajectory measured by the constant load stress relaxation test, and the strain growth curve obtained from the creep test. Together, they constitute the input of the rheological constitutive model, enabling the rheological constitutive model to accurately reproduce the creep and stress relaxation characteristics of silty soil under constant and variable load conditions.

[0028] For each original measurement data point, based on the corresponding spatial coordinates and sampling time, the model-driving load or boundary conditions at the corresponding time point are input into the rheological constitutive model to calculate the theoretical prediction value. The difference between the theoretical prediction value and the original measurement value is calculated to obtain the residual value. The residual value represents the degree of deviation between the actual soil response and the expected response of the rheological model.

[0029] Based on laboratory test results and engineering experience with silty soil, a reasonable residual threshold range was determined. This residual threshold range is based on residual quantile statistical analysis to ensure that most residuals fall within the threshold range in the absence of abnormal disturbances. If the absolute value of the residual corresponding to a certain measurement data point is greater than the threshold, the measurement value is determined to be an outlier or affected by external interference and is removed from the original dataset.

[0030] After removing residual abnormal values, the remaining data is smoothed using a smoothing filter to eliminate high-frequency noise interference that may have occurred during sampling. A weighted average smoothing algorithm based on three adjacent data points is employed: for three consecutive measurements, the new value at the midpoint is replaced with the sum of the three measurements weighted in a 1:2:1 ratio and then divided by the sum of the weighting coefficients (i.e., four). After smoothing, high-frequency fluctuations in pore water pressure data, stratified settlement data, and deep lateral displacement data are suppressed, reducing the risk of error propagation.

[0031] The sampling time of each sampling point is accurately calibrated using a unified standard time calibration method; Based on the distribution of sampling times at all sampling points, a unified time axis is established, with a predetermined unified sampling period as the smallest time unit. The unified sampling period is derived from the dynamic response characteristic analysis of silty soil and is typically used at ten-minute intervals. The global unified time axis, from the start time to the end time of monitoring, is segmented at equal intervals according to the unified sampling period to generate a continuous time series.

[0032] For each data point cleaned using the rheological constitutive model, its original sampling time is matched with the closest time node on the global unified time axis. The matching process adopts the principle of minimizing absolute time difference: the time difference between the sampling time of this data point and two adjacent time nodes on the global time axis is calculated, the node with the smaller time difference is taken as the mapping result, and the time label of this data point is replaced with the corresponding time node on the unified time axis.

[0033] After mapping, some time points may lack corresponding measurement data. A piecewise linear interpolation method is used to fill in the missing measurements: The two most recent mapped data points before and after the missing point are determined. The slope is obtained by dividing the difference between the latter and former data values ​​by the difference in their corresponding time intervals. This slope is then used to calculate the interpolation value for the missing point using the time interval between the missing points. The interpolation formula is as follows: Based on the known values ​​and time differences between two adjacent points, the missing point value is calculated and recorded according to the time ratio. After interpolation, it is ensured that each time point contains at least one pore water pressure data point, one stratified settlement data point, and one deep lateral displacement data point.

[0034] After data cleaning and sampling time calibration, the pore water pressure data, stratified settlement data, and deep lateral displacement data are recombined according to spatial coordinates to generate a foundation state characteristic matrix. Based on the recorded spatial coordinates of each sampling point, all pore water pressure data, stratified settlement data, and deep lateral displacement data after cleaning and calibration are projected onto pre-divided spatial grid cells according to their spatial coordinates. The spatial grid cells can be regular rectangular grids or irregular triangular grids, but the grid division principle must cover all monitoring points in the monitoring area, ensuring that each spatial cell contains at least one sampling point. The coordinate range of each spatial cell is clearly recorded.

[0035] For each time node within the same spatial unit, the pore water pressure measurement value, stratified settlement measurement value, and deep lateral displacement measurement value at the corresponding time node are aligned and combined. The combination method is as follows: at each time node, the pore water pressure value, stratified settlement value, and deep lateral displacement value are retrieved sequentially according to a fixed field order, and the three types of data values ​​are filled into a single record. The fields of the row record include, in order: spatial unit number, time node label, pore water pressure value, stratified settlement value, and deep lateral displacement value.

[0036] After merging the data from all spatial units, all row records are sorted in ascending order by time node and spatial unit number, forming a two-dimensional matrix structure. Matrix rows correspond to several sampling times and spatial unit combinations, while matrix columns correspond to three data columns: pore water pressure data, stratified settlement data, and deep lateral displacement data. The final generated matrix contains the following element fields: row identifiers include spatial unit number and time node label; column identifiers include the three monitoring values: pore water pressure, stratified settlement, and deep lateral displacement; and all elements within the matrix are cleaned, calibrated, and spatially combined standardized monitoring values.

[0037] Dynamic causal correlation tracing analysis was performed on the foundation state characteristic matrix to generate a set of water-soil coupling drift parameters, including: Using the foundation state feature matrix as input, the multivariate time series is divided into sliding windows of fixed duration, so that each window contains equal amounts of pore water pressure data, stratified settlement data, and deep lateral displacement data. The unified sampling period is set in step S1, with each sampling interval being ten minutes. Based on the rheological characteristics of silty soil and monitoring requirements, the sliding window length is set to six consecutive sampling periods, corresponding to a continuous one-hour data segment. The window step interval is set to one consecutive sampling period, corresponding to moving the window once every ten minutes. This ensures that each window contains six sets of pore water pressure data, six sets of stratified settlement data, and six sets of deep lateral displacement data, meeting the time series length requirements for causal analysis.

[0038] Extract all spatial unit numbers and time node labels from the foundation condition feature matrix. Starting from the first time node, slice the matrix in units of six time nodes with a sliding window length, according to the ascending order of the time node labels, to obtain the first time segment. The first time segment contains six rows corresponding to six time nodes, with each row including three monitoring values ​​for the same spatial unit: pore water pressure, stratified settlement, and deep lateral displacement. Then, slide forward one time node position at a step interval to generate the second time segment, and so on, until the end of the foundation condition feature matrix is ​​reached.

[0039] Each time segment contains six records for each spatial unit, including three monitoring values: pore water pressure, stratified settlement, and deep lateral displacement. If any monitoring value is missing for a spatial unit in a certain time segment, it has been removed or interpolated during the data cleaning process to ensure that each time segment contains six complete sets of monitoring data.

[0040] The two-way Granger causality test statistic is calculated sequentially for the monitoring data in each window, and the valid results are filtered by setting a significance threshold to obtain the causality weight matrix; For each sliding window, six sets of pore water pressure, six sets of stratified settlement, and six sets of deep lateral displacement are used to construct local multivariate time series data, each with three columns of values ​​under the corresponding spatial unit number. For any two types of monitoring values ​​within the same spatial unit, such as pore water pressure and stratified settlement, the Granger causality test statistic is first calculated using the two types of values ​​as the dependent and independent variables, respectively. Then, the order of the two types of values ​​is reversed, i.e., the calculation is repeated with stratified settlement as the dependent variable and pore water pressure as the independent variable. This yields a bidirectional causality test result, containing information on the strength of causal relationships in both directions.

[0041] Because there is a short-term lag effect between the two monitoring values, the optimal lag order is selected using the information criterion method. By comparing the numerical values ​​of the Akaike information criterion and the Bayesian information criterion at different lag orders, the lag order that minimizes the information criterion is determined as the test order. After verification with measured data, a lag order of two is selected. Thus, bidirectional tests are performed on each pair of monitoring values ​​sequentially, resulting in two sets of causal test statistics.

[0042] A significance level of 5% was set, and the p-values ​​of each Granger test were compared. A p-value less than the significance level indicates a significant Granger causal relationship between the explained variable and the explanatory variable at a given lag order. Weights were assigned to the two-way test results: when pore water pressure has a significant causal relationship with stratified settlement, the weight of the corresponding position (pore water pressure - stratified settlement) was set to one, otherwise to zero; when stratified settlement has a significant causal relationship with pore water pressure, the weight of the corresponding position (stratified settlement - pore water pressure) was set to one, otherwise to zero. Using this method, two-way tests were performed on all pairwise combinations of the three types of monitoring data, generating a 3x3 causal weight matrix, where diagonal elements were all set to zero, and off-diagonal elements were either zero or one.

[0043] The causal weight matrix obtained by the continuous window is subjected to first-order difference operation in the time dimension to extract the causal weight change rate vector and construct a candidate drift vector sequence. A local causal weight matrix is ​​calculated for each sliding window. The causal weight matrices of multiple consecutive windows are recorded in chronological order to form a time-series matrix sequence. Each time-series position corresponds to a 3x3 causal weight matrix.

[0044] Perform a first-order difference operation on the temporal sequence of causal weight matrices: for the causal weight matrices at two adjacent time points, perform a difference operation between the elements at the same position to generate a new 3x3 difference matrix. The element values ​​of the difference matrix range from -1 to positive 1. The difference result reflects the direction and magnitude of the change in the strength of the causal relationship. For example, the difference value is negative 1 when it changes from 1 to 0, and positive 1 when it changes from 0 to 1.

[0045] Each 3x3 difference matrix is ​​expanded into a nine-dimensional vector in a fixed order, which can be preset to either row-major or column-major order. This results in multiple nine-dimensional vectors representing the changes in causal relationships between adjacent time segments. These nine-dimensional vectors are then arranged in chronological order to form a candidate drift vector sequence.

[0046] Principal component analysis is performed on the candidate drift vector sequence to reduce dimensionality, and principal components with a cumulative variance contribution rate not lower than a preset threshold are retained to form a set of water-soil coupling drift parameters. To reduce the dimensionality and redundancy of the candidate drift vector sequence, principal component analysis (PCA) is employed. First, the candidate drift vector sequence is mean-centered: the mean value for each dimension is calculated, and the corresponding mean value for each dimension is subtracted from the mean value. After centering, a covariance matrix is ​​constructed. The directions of each principal component are obtained through eigenvalue analysis and eigenvector solving of the covariance matrix.

[0047] The cumulative variance contribution threshold for principal component retention is set at 90%. All eigenvalues ​​are arranged in descending order, and the cumulative proportion of each eigenvalue to the total sum of eigenvalues ​​is calculated. When the cumulative proportion reaches or exceeds 90%, all corresponding eigenvectors are retained as principal component directions. The cumulative variance contribution threshold ensures that the retained principal components sufficiently preserve information from the original candidate vector sequence while removing most noise and redundant information.

[0048] The centered candidate drift vector sequences are projected onto the directions of the retained principal components, resulting in a set of one-dimensional time series. These time series represent the projection coefficients of each retained principal component at different time segments. Each projected coefficient sequence generated for a retained principal component is assigned a unique number and a corresponding time window label. The projected coefficient sequences corresponding to all retained principal components are summarized in tabular form to form a water-soil coupling drift parameter set. This set includes the time window label, number, and projection coefficients for each drift parameter sequence, and is used for nonlinear scaling analysis.

[0049] By performing nonlinear scaling analysis on the multi-parameter coordinated drift of the water-soil coupling drift parameters, a set of stress history dependence coefficients characterizing the reverse correlation of anomalies is obtained, including: A nonlinear scaling analysis method was used to perform multi-parameter collaborative drift analysis on the principal components in the water-soil coupling drift parameter set, and the collaborative drift index between the principal components was calculated. The water-soil coupled drift parameter set includes several drift parameter sequences, each corresponding to a principal component. Each sequence consists of several continuous time window labels and corresponding projection coefficients, and each sequence is also accompanied by spatial coordinate identifiers.

[0050] Normalization is performed on each drift parameter sequence in the water-soil coupling drift parameter set: First, the minimum and maximum values ​​of the sequence are calculated in all time windows. Then, the minimum value is subtracted from the projection coefficient of each time window and divided by the difference between the maximum and minimum values ​​to obtain the normalized principal component sequence.

[0051] A similarity metric function based on intrinsic kernel mapping is used to calculate the nonlinear relationship between two principal component sequences: For the two normalized principal component sequences, construct two-dimensional point sets within the same time window label range, with the horizontal axis representing the projection coefficients of the first principal component and the vertical axis representing the projection coefficients of the second principal component.

[0052] A smooth curve is fitted onto a two-dimensional point set, representing the nonlinear correlation trend of the two principal component sequences at each time window label. The fitting process can employ a smooth spline curve fitting method, determining the smooth curve by minimizing a quadratic smoothing objective function.

[0053] Let any point on the smooth curve correspond to the normalized projection coefficients under a pair of time window labels. For this pair of time window labels, calculate the local curvature value along the smooth curve. The local curvature value reflects the degree of nonlinearity in the rate and direction of change of the two principal component sequences under this time window label.

[0054] The absolute values ​​of the local curvature values ​​on all time window labels are summed along the time axis to obtain the overall cumulative nonlinear drift of the two principal component sequences over the entire monitoring period. This cumulative drift is defined as the co-drift index between the principal component pairs. A larger co-drift index indicates a stronger nonlinear co-variation relationship between the two principal component sequences.

[0055] The co-drift indices corresponding to each pair of principal component sequences are recorded in tabular form, and each co-drift index is appended with two corresponding principal component numbers and corresponding spatial coordinate identifiers. All records together constitute a multi-parameter co-drift index table, which is used for anomaly reverse association pattern recognition.

[0056] Based on the collaborative drift index among the principal components, abnormal reverse correlations between drift parameters are identified, and abnormal reverse correlation patterns are marked according to drift direction and amplitude. When the smooth curves formed by two normalized principal component sequences within the same time window label range show an overall opposite trend and the local curvature signs are opposite, it can be considered that there is an inverse correlation between the two within the time window range.

[0057] Reverse trend standard: If, within a number of adjacent time windows, the projection coefficient of the first principal component first increases and then decreases over time, while the projection coefficient of the second principal component first decreases and then increases over time, then the two are considered to show reverse trends in the sub-segment.

[0058] Local curvature sign opposite criterion: The sign of the local curvature value of the smooth curve at each time window label point is judged. When the curvature value is positive, it means that the local change trend is in the same direction and is convex. When the curvature value is negative, it means that the local change trend is concave. If the local curvature signs of two principal component sequences are opposite at the same time window label point, then there is an inverse correlation at the time window label point.

[0059] For each pair of principal component sequences, a two-dimensional smoothing curve is constructed, iterating through all time window labels: For each time window label, determine whether the signs of the local curvature of the two sequences corresponding to that label are opposite; if they are opposite, then mark the time window label point.

[0060] Based on the marked time window labels, adjacent consecutive labels are merged into reverse-correlated sub-segments. A sub-segment composed of multiple adjacent time window labels represents a short-term reverse change process.

[0061] Extract all inverse correlation sub-segments, and record the start time window label and end time window label of each sub-segment, along with the corresponding principal component numbers, into the list of abnormal inverse correlation patterns.

[0062] For each extracted inversely correlated sub-segment, the magnitude index is calculated based on the average increase and average decrease of the two normalized principal component sequences within that segment: Amplitude index calculation method: Calculate the difference between the maximum and minimum values ​​of the two principal component sequences in the sub-segment to obtain two individual amplitudes; calculate the inverse correlation amplitude ratio as the ratio of the two individual amplitudes. If the ratio is greater than or equal to two or less than or equal to 0.5, the sub-segment amplitude is considered to be a significant inverse amplitude; otherwise, it is considered to be a normal inverse amplitude.

[0063] Directional labeling method: If, within a sub-segment, the first principal component sequence as a whole shows an upward trend followed by a downward trend, while the second principal component sequence as a whole shows an upward trend followed by a downward trend, then the reverse correlation pattern is labeled as "first upward, second downward"; if the opposite is true, it is labeled as "first downward, second upward".

[0064] The direction and magnitude markers of each inverse correlation sub-segment are appended to the corresponding item in the list of abnormal inverse correlation patterns.

[0065] Statistical analysis was performed on the abnormal reverse correlation patterns, and the stress history dependence coefficients characterizing the abnormal reverse correlation patterns were calculated. The stress history dependence coefficients at each spatial coordinate were then labeled to form a set of stress history dependence coefficients. Based on the spatial coordinate identifier attached to each record, all abnormal reverse association patterns are grouped to form a set of association patterns corresponding to different spatial coordinates.

[0066] For each spatial coordinate, count the number of all reverse-related sub-segments that appear under the spatial coordinate and the length of the corresponding sub-segment (measured by the number of time window labels).

[0067] For each spatial coordinate, the stress history dependence coefficient is defined as the ratio of the cumulative duration of all reverse-correlated sub-segments under the spatial coordinate to the total monitoring duration, multiplied by the average amplitude index coefficient: Cumulative duration calculation: The total number of reverse correlation time windows is obtained by summing the number of time window labels of all reverse correlation sub-segments under the same spatial coordinates.

[0068] Total monitoring duration calculation: Combine the set number of time windows with the sliding window movement scheme. For example, if the sliding window length is six time windows, the step interval is one time window, and the monitoring period contains a total of eighteen time window labels, then the total monitoring duration corresponds to a unit of ten minutes, that is, the total number of time windows is thirteen.

[0069] Average amplitude index calculation: The amplitude index is obtained by using the amplitude ratio calculation formula for all reverse correlation sub-segments under the same spatial coordinates, and the average value is used to obtain the average amplitude index coefficient.

[0070] Stress history dependence coefficient calculation: The stress history dependence coefficient is equal to the total number of reverse correlation time windows divided by the total number of time windows, and then multiplied by the average amplitude index coefficient. The result value ranges from zero to two, and is used to reflect the degree of influence of the reverse correlation strength and duration on spatial coordinates.

[0071] Record the stress history dependency coefficient and the three-dimensional spatial location of each spatial coordinate. The final set of stress history dependency coefficients includes the following fields: spatial coordinate identifier, dependency coefficient value, total number of reverse correlation time windows, total number of monitoring time windows, and average amplitude index coefficient. This is used for rheological state partitioning and dynamic weight fusion.

[0072] A dynamic response weighted fusion analysis based on rheological state zoning was performed on the set of water-soil coupling drift parameters and the set of stress history dependence coefficients to determine the stability risk level of the silty soil foundation, including: By combining the set of water-soil coupling drift parameters with the set of stress history dependence coefficients, the silty soil foundation is divided into several rheological state regions by the rheological state zoning rule. By combining the rheological properties of silty soil with its staged stress-strain model, stress history dependence coefficients and drift parameter values ​​are mapped to different ranges to delineate regions with similar rheological behaviors. The rheological state zoning relies on the following three types of indicators: Time window label set: represents the time range in which the drift parameter projection coefficients are generated; The set of stress history dependence coefficients reflects the intensity and persistence of the reverse correlation patterns occurring at each spatial coordinate. Water-soil coupling drift parameter set: the projection coefficient values ​​corresponding to each principal component, representing the deformation coupling characteristics of the foundation within a continuous time window.

[0073] Based on experimental results of shear strength and creep rate of cohesive soil, the stress history dependence coefficient and drift parameter projection coefficient are mapped to three different rheological state categories: Initial steady-state region: The stress history dependence coefficient is in a low range, and the fluctuation amplitude of the drift parameter projection coefficient is low, indicating that there is no obvious nonlinear cooperative drift of the foundation; Intermediate transition region: The stress history dependence coefficient is in the medium range, and the drift parameter projection coefficient shows relatively obvious fluctuations, indicating that the foundation is transitioning between gradual shear weakening and creep; Critical instability region: The stress history dependence coefficient is at a high level, and the drift parameter projection coefficient shows strong fluctuations and nonlinear jumps, indicating that the foundation may be entering a critical instability state.

[0074] The above low-level, medium-level, and high-level ranges were obtained through statistical analysis of historical monitoring data collected on-site. First, the historical stress dependence coefficients at each spatial coordinate point during each monitoring period were sorted, and their percentile positions were calculated. The low-level range was defined as the values ​​within the top 50%, the medium-level range as the values ​​between 50% and 80%, and the high-level range as the values ​​above 80%. The range of the drift parameter projection coefficient was also divided using the same method to ensure that the zoning rules matched the actual monitoring data. The drift parameter projection coefficient fluctuation range was defined as the values ​​within the top 50%, the drift parameter projection coefficient showing relatively significant fluctuations as the values ​​between 50% and 80%, and the drift parameter projection coefficient exhibiting strong fluctuations and nonlinear jumps as the values ​​above 80%.

[0075] For each spatial coordinate, extract the stress history dependence coefficient value corresponding to the stress history dependence coefficient set and the maximum absolute projection coefficient value among all principal component projection system sequences corresponding to the spatial coordinate in the soil-water coupling drift parameter set. Based on the stress history dependence coefficient value and the projection coefficient amplitude, determine which rheological state region the spatial coordinate should be assigned to. When the stress history dependence coefficient is in a low range and the magnitude of the maximum absolute projection coefficient is in a small range, the spatial coordinates are assigned to the initial steady-state region. When the stress history dependence coefficient is in the middle range and the amplitude of the maximum absolute projection coefficient is in the middle range, the spatial coordinates are assigned to the intermediate transition region. When the stress history dependence coefficient is in a high range or the projection coefficient amplitude is in a high range, the spatial coordinates are classified into the critical turbulence region.

[0076] For each rheological state region, the water-soil coupling drift parameters and stress history dependence coefficients within the region are extracted, and the dynamic response weights are calculated based on the extracted values. After completing the rheological state region division, perform the following operations on each rheological state region: Extract the set of water-soil coupling drift parameters corresponding to all spatial coordinates within the rheological state region, including the projection coefficient sequence and time window label for each drift parameter; Extract the set of stress history dependency coefficient entries corresponding to all spatial coordinates within the rheological state region, including the dependency coefficient value and associated sub-segment information for each spatial coordinate.

[0077] For the sequence of drift parameter projection coefficients corresponding to each spatial coordinate within the rheological state region, they are first aggregated within the same time window label: If different spatial coordinates within the same rheological state region have the same time window label range, the projection coefficients of the spatial coordinates are arithmetically averaged over that time window label to obtain the one-dimensional average projection coefficient of the time window label within the rheological state region. The projection coefficients of all spatial coordinates are then combined into a projection coefficient sequence representing the regional characteristics.

[0078] The arithmetic mean of the dependence coefficients corresponding to each spatial coordinate within the rheological state region is used to obtain a representative dependence coefficient value for the region. This value reflects the overall historical stress correlation intensity and persistence of the region.

[0079] Dynamic response weights are used to measure the combined impact of water-soil coupled drift parameters and stress history dependence coefficients on the regional stability assessment of the current rheological state. To reflect the coupling effect between drift change trends and the degree of historical dependence, the calculation of dynamic response weights is divided into two parts: Drift trend weight: The drift amplitude is obtained by taking the difference between the maximum and minimum values ​​of the representative projection coefficient sequence of the rheological state region; the drift amplitude is normalized by dividing the drift amplitude by the historical maximum amplitude of the region to obtain a normalized amplitude value between zero and one; the normalized amplitude value is the drift amplitude weight.

[0080] Historical dependency weight: A linear stretching mapping is applied to the representative dependency coefficient values ​​of the rheological state region, mapping the dependency coefficients from their original range to a range of zero to one. The mapping method is to subtract the lowest dependency coefficient from the original dependency coefficient and then divide by the difference between the highest and lowest dependency coefficients. The mapped value is the historical dependency weight.

[0081] The drift amplitude weight and the historical dependence weight are combined in a weighted average manner. The combination rule is: first add the two together and then divide by two to obtain the dynamic response weight of the region.

[0082] The dynamic response weights of each rheological state region are weighted and fused according to the weight coefficients of the sampling points in the region to obtain the overall dynamic response weight. To reflect the different importance of sampling points with different spatial coordinates in the regional stability evaluation, the weight coefficients of the sampling points are set based on the following three factors: Distance between spatial coordinates and key structural locations: If a sampling point is adjacent to a known potential slip surface or weak interlayer, it should be assigned a higher weight; otherwise, it should be assigned a lower weight. Sensor type and sensitivity: Monitoring data from high-precision pore water pressure sensors and high-sensitivity strain sensors are more valuable for drift and dependency analysis, and their sampling point weights should be higher than those of ordinary sensors. Data integrity and validity: If a sampling point has multiple missing or abnormal removals in the previous monitoring period, its weight should be appropriately reduced; if the data is complete and there is little noise, its weight should be appropriately increased.

[0083] The above three factors are quantified. For example, distance is linearly mapped to a value in meters between zero and one, sensor sensitivity level is mapped to a value between zero and one, and data integrity is mapped to a value between zero and one. The final sampling point weight coefficient is then calculated by averaging the three mapped values. The mapping method and proportional coefficients of each factor can be fine-tuned according to the actual engineering site conditions.

[0084] For all spatial coordinates within a certain rheological state region, the weighted fusion dynamic response weights are calculated sequentially using the following steps: For each spatial coordinate within the rheological state region, extract two values: the dynamic response weight value of the spatial coordinate and the sampling point weight coefficient of the spatial coordinate. For each spatial coordinate, its dynamic response weight value is multiplied by the corresponding sampling point weight coefficient to obtain the weighted contribution value of the spatial coordinate to the regional dynamic response. The weighted contribution values ​​of all spatial coordinates within the rheological state region are summed to obtain the total weighted contribution of the region. Divide the total weighted contribution of the region by the sum of the weight coefficients of all sampling points within the rheological state region to obtain the overall dynamic response weight of the rheological state region.

[0085] The stability risk level of silty soil foundation is determined by comparing the overall dynamic response weight with the preset risk threshold. A preset risk threshold is used to map continuous dynamic response weights to discrete risk levels. Based on historical ground monitoring data and the requirements of step-based risk grading, three risk levels and corresponding threshold ranges are defined: Low risk level: The overall dynamic response weight is less than or equal to 0.35; indicating that the foundation in the rheological state area is in a relatively stable state and no immediate intervention is required; Medium risk level: The overall dynamic response weight is greater than 0.35 and less than or equal to 0.7; indicating that there is moderate nonlinear drift and history dependence effect in the rheological state region, and monitoring should be strengthened and early warning should be prepared. High risk level: The overall dynamic response weight is greater than 0.7; this indicates that the rheological state region has entered a stage of severe nonlinear drift and historical inverse correlation, and thrust stop or drainage measures should be implemented in a timely manner.

[0086] The above threshold ranges were obtained through historical monitoring and engineering case analysis. For example, in previous similar silty soil foundation projects, when the dynamic response weight exceeded 0.7, most sections showed significant accelerated tilting and crack development within a few days; when it was below 0.35, the foundation changes were basically within a controllable range.

[0087] Compare the overall dynamic response weights of each rheological state region with the preset risk threshold: If the overall dynamic response weight of the rheological state region is less than or equal to 0.35, it is considered a low-risk level. If the overall dynamic response weight of the rheological state region is greater than 0.35 and less than or equal to 0.7, it is classified as a medium-risk level. If the overall dynamic response weight of the rheological state region is greater than 0.7, it is considered a high-risk level.

[0088] The risk levels of all rheological state regions are organized according to spatial location to form a stability risk level distribution table for silty soil foundations covering the entire monitoring area. The stability risk level distribution table includes the following fields: Region number: A unique code corresponding to each rheological state region, such as the initial steady state region being numbered as region one, the intermediate transition region being numbered as region two, and the critical turbulence region being numbered as region three; Overall dynamic response weight value: The dynamic response weight calculated for each rheological state region; Stability risk level: low risk, medium risk, or high risk.

[0089] The stability risk level distribution table provides a clear understanding of the risk situation in different regions and serves as a basis for the steps of constructing early warning instructions.

[0090] Based on the stability risk level of the silty soil foundation, a graded early warning instruction is constructed, and early warning information is output, including: Establish a tiered early warning instruction database in advance; A relational database management system is used to ensure the reliability and scalability of data storage. The database tables are stored in a structured tabular format, and the table name is "Hierarchical Early Warning Instruction Table".

[0091] The tiered early warning instruction table contains the following fields, each identified by its Chinese name, and the field order is as follows: Risk level identifier: Used to indicate the stability risk level, and can take text values ​​such as low risk, medium risk, high risk, etc., for matching with monitoring results; Warning Level: The warning level corresponds to the risk level. It can take text values ​​such as Level 1, Level 2, Level 3, etc., and is used for internal display and sorting of instructions. Command content text: A text type field that stores the description of the warning command that is finally sent to the monitoring terminal, such as the foundation being in good condition and no measures need to be taken, suggesting that drainage be strengthened and the frequency of observation be increased, and that reinforcement measures should be initiated immediately and personnel evacuated, etc. Recommended action measures: A text-type field that stores supplementary suggestions to the recipient in addition to the instruction content, such as suggestions for drainage ditch maintenance, real-time video monitoring, and evacuation of on-site personnel to a safe area; Output priority: A text type field used to sort the order in which commands are sent, with three possible values: high, medium, and low. Update timestamps: A date and time type field that records the last time each instruction was modified or approved.

[0092] Input the stability risk level of silty soil foundation into the graded early warning instruction database, and select the early warning instruction corresponding to the stability risk level of silty soil foundation according to the matching rules; Assuming the current stability risk level of the silty soil foundation is medium risk, the query condition sent to the graded early warning instruction table is "risk level identifier: medium risk". The query results include: The warning level is Level II; The instruction text states that the foundation is in a state of slow drift and recommends strengthening drainage and on-site inspections. Recommended measures include checking the drainage system for blockages and increasing monitoring frequency. The output priority is medium. Update the timestamps to the latest time automatically recorded by the database.

[0093] The selected warning instructions are formatted to generate standardized warning information; To ensure that the output warning information can be parsed by external monitoring terminals according to a unified protocol, standardized warning information is generated using the following field order and concatenation rules, including: Risk level: Consistent with the risk level identifier in the query results. Warning level: Consistent with the warning level in the query results. Command content: Consistent with the command content text in the query results. Recommended actions: Consistent with the recommended actions in the search results. Generation time: Current system date and time, expressed in year, month, day, hour, minute, and second.

[0094] Standardized early warning information is output to external monitoring terminals through a communication interface; Serial communication or LAN communication protocols are used to push early warning information to a pre-set external monitoring terminal via an industrial control bus. A verification mechanism is employed during communication to ensure the complete transmission of early warning information. Pre-set parsing logic in the external monitoring terminal maps standardized early warning information to corresponding graphical instructions, color codes, audible and visual alarm module control parameters, or SMS forwarding module content in the human-machine interface, ensuring that on-site operators can intuitively obtain and respond to early warning signals.

[0095] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0096] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0097] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0098] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0099] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0100] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0101] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0102] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0103] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0104] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent monitoring of silty soil foundations based on multi-source data fusion, characterized in that, The steps include: S1: Simultaneously collect pore water pressure, stratified settlement, and lateral displacement data of silty soil foundation to generate original soil state dataset; S2: The foundation state feature matrix is ​​obtained by performing rheological constitutive cleaning and time calibration on the original soil state dataset; S3: Perform dynamic causal correlation tracing analysis on the foundation state characteristic matrix to generate a set of water-soil coupling drift parameters; S4: By performing nonlinear scaling analysis on the multi-parameter coordinated drift of the water-soil coupling drift parameters, a set of stress history dependence coefficients characterizing the reverse correlation of anomalies is obtained. S5: Perform dynamic response weight fusion analysis based on rheological state zoning on the set of water-soil coupling drift parameters and the set of stress history dependence coefficients to determine the stability risk level of silty soil foundation; S6: Construct graded early warning instructions based on the stability risk level of silty soil foundations and output early warning information.

2. The intelligent monitoring method for silty soil foundation based on multi-source data fusion according to claim 1, characterized in that, S1, specifically: A combined configuration of pore water pressure equipment, stratified settlement monitoring devices, and deep lateral displacement measurement equipment deployed at different depths and directions within silty soil foundations; By setting a unified sampling period, various monitoring devices are synchronously triggered to collect raw measurement data of pore water pressure, stratified settlement and deep lateral displacement respectively. The original measurement data of various types are divided and labeled according to their spatial location, and the spatial coordinates, sampling time and monitoring equipment number of each sampling point are recorded to form the original soil state dataset.

3. The intelligent monitoring method for silty soil foundation based on multi-source data fusion according to claim 2, characterized in that, S2, specifically: Data cleaning of the original soil state dataset was performed using a rheological constitutive model to remove outliers and interference terms from the pore water pressure data, stratified settlement data, and deep lateral displacement data. The sampling time of each sampling point is accurately calibrated using a unified standard time calibration method; After data cleaning and sampling time calibration, the pore water pressure data, stratified settlement data, and deep lateral displacement data are recombined according to spatial coordinates to generate a foundation state characteristic matrix.

4. The intelligent monitoring method for silty soil foundation based on multi-source data fusion according to claim 3, characterized in that, S3, specifically: Using the foundation state feature matrix as input, the multivariate time series is divided into sliding windows of fixed duration, so that each window contains equal amounts of pore water pressure data, stratified settlement data, and deep lateral displacement data. The two-way Granger causality test statistic is calculated sequentially for the monitoring data in each window, and the valid results are filtered by setting a significance threshold to obtain the causality weight matrix; The causal weight matrix obtained by the continuous window is subjected to first-order difference operation in the time dimension to extract the causal weight change rate vector and construct a candidate drift vector sequence. Principal component analysis is performed on the candidate drift vector sequence to reduce dimensionality, and principal components with a cumulative variance contribution rate not lower than a preset threshold are retained to form a set of water-soil coupling drift parameters.

5. The intelligent monitoring method for silty soil foundation based on multi-source data fusion according to claim 4, characterized in that, S4, specifically: A nonlinear scaling analysis method was used to perform multi-parameter collaborative drift analysis on the principal components in the water-soil coupling drift parameter set, and the collaborative drift index between the principal components was calculated. Based on the collaborative drift index among the principal components, abnormal reverse correlations between drift parameters are identified, and abnormal reverse correlation patterns are marked according to drift direction and amplitude. Statistical analysis was performed on the abnormal reverse correlation patterns, and the stress history dependence coefficients characterizing the abnormal reverse correlation patterns were calculated. The stress history dependence coefficients at each spatial coordinate were then labeled to form a set of stress history dependence coefficients.

6. The intelligent monitoring method for silty soil foundation based on multi-source data fusion according to claim 5, characterized in that, S5, specifically: By combining the set of water-soil coupling drift parameters with the set of stress history dependence coefficients, the silty soil foundation is divided into several rheological state regions by the rheological state zoning rule. For each rheological state region, the water-soil coupling drift parameters and stress history dependence coefficients within the region are extracted, and the dynamic response weights are calculated based on the extracted values. The dynamic response weights of each rheological state region are weighted and fused according to the weight coefficients of the sampling points in the region to obtain the overall dynamic response weight. The stability risk level of silty soil foundation is determined by comparing the overall dynamic response weight with the preset risk threshold.

7. The intelligent monitoring method for silty soil foundation based on multi-source data fusion according to claim 6, characterized in that, S6, specifically: Establish a tiered early warning instruction database in advance; Input the stability risk level of silty soil foundation into the graded early warning instruction database, and select the early warning instruction corresponding to the stability risk level of silty soil foundation according to the matching rules; The selected warning instructions are formatted to generate standardized warning information; Standardized early warning information is output to external monitoring terminals through a communication interface.

Images