Tidal river bank collapse early warning method based on stress response grading evolution characteristics of soil body

Abstract

The application discloses a kind of tidal river bank collapse early warning methods based on stress response grading evolution characteristics of soil mass.The method obtains the original online monitoring stress data in the internal soil mass of bank section, carries out abnormal pulse cleaning and low-pass filter smoothing processing, and obtains smooth stress data;Based on tidal harmonic model, separate the tidal stress component in the smooth stress data, extract the residual trend stress sequence removed from periodic interference;In the continuous sliding window, extract the multi-dimensional trend feature parameters representing the evolution of bank slope failure;Multi-index comprehensive comparison is carried out using multi-dimensional trend feature parameters and preset bank collapse risk threshold, and combined with early warning grade trigger and release logic, output grading early warning result.The application peels off the periodic hydrological interference in the process of bank collapse identification using soil stress change, improves the reliability of non-significant grading evolution trend extraction and the anti-disturbance ability of early warning grade discrimination.

Description

Technical Field

[0001] This invention relates to the field of river safety monitoring and disaster early warning technology, and in particular to a method for early warning of bank collapse in tidal river channels based on the graded evolution characteristics of soil stress response. Background Technology

[0002] Riverbank collapse is a physical evolution process caused by the continuous action of complex hydrodynamic conditions such as water erosion and water level fluctuations, leading to the gradual destruction and eventual instability of the riverbank soil structure. Continuous online monitoring and evolution trend prediction of the soil's internal mechanical state can quantify the cumulative characteristics of soil stress and accelerated deformation in real time. This is of significant engineering and technical importance for identifying early signs of instability in local or overall riverbanks and assessing the progression of collapse risk.

[0003] Current online monitoring methods for riverbank collapse typically rely on sensors deployed inside the bank slope to transmit stress or displacement data in real time. After conventional smoothing of the time series, the absolute change is used directly for risk warning. In practical applications, complex tidal river sections impose strong periodic hydraulic loads on the soil, and these multi-source external excitations are directly superimposed on the monitoring signal. Existing data processing workflows often use a single filter or fixed time window to calculate the average value to suppress physical clutter. The calculated rate of change is then directly compared with an absolute static threshold set based on expert experience, which serves as the sole criterion for triggering an early warning action.

[0004] The aforementioned conventional processing methods are insufficient to effectively decouple the coupling interference between broadband environmental noise and periodic hydrological loads, resulting in significant baseline drift in the extracted soil mechanical evolution trend. Furthermore, when faced with varying geological conditions and non-monotonic stress fluctuation signals, a single static threshold triggering mechanism often exhibits a high frequency of false triggers and alarm state jumps. Therefore, it is necessary to investigate an early warning method that can improve the robustness of data processing and the stability of risk state discrimination under complex hydrological environmental disturbances. Summary of the Invention

[0005] In view of this, this application provides a method for early warning of bank collapse in tidal river channels based on the hierarchical evolution characteristics of soil stress response, in order to solve the above-mentioned problems of the prior art.

[0006] According to one aspect of this application, a method for early warning of bank collapse in tidal channels based on the graded evolution characteristics of soil stress response is provided, comprising:

[0007] Obtain original online monitoring stress data of the soil inside the bank section at risk of bank collapse;

[0008] Abnormal pulse cleaning and low-pass filtering smoothing are performed on the raw online monitoring stress data to obtain smooth stress data with reduced high-frequency interference;

[0009] The tidal stress component in the smoothed stress data is calculated and separated based on the tidal harmonic model, and the residual trend stress sequence after removing tidal response interference is extracted.

[0010] Based on the residual trend stress sequence, multidimensional trend feature parameters for characterizing the evolution of bank slope failure are extracted within a continuous sliding window;

[0011] By comparing multi-dimensional trend feature parameters with preset bank collapse risk thresholds, and combining the warning level triggering and cancellation logic, the corresponding bank collapse risk classification warning results are output.

[0012] According to another aspect of this application, a tidal riverbank collapse early warning system based on the graded evolution characteristics of soil stress response is also provided, comprising:

[0013] At least one processor; and a memory communicatively connected to at least one of the processors;

[0014] The memory stores instructions that can be executed by the processor to implement the tidal riverbank collapse early warning method based on the graded evolution characteristics of soil stress response proposed in this application.

[0015] Based on the above technical solution, the present invention eliminates the periodic hydrological interference in the process of identifying bank collapses by utilizing soil stress changes, thereby improving the reliability of non-significant evolution trend extraction and the anti-disturbance capability of early warning level judgment. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0017] Figure 1 This is a schematic diagram of a tidal riverbank collapse early warning method based on the graded evolution characteristics of soil stress response, applicable to embodiments of this application.

[0018] Figure 2 This is a schematic diagram of the abnormal pulse cleaning process for the original online monitoring stress data applicable to the embodiments of this application.

[0019] Figure 3 This is a schematic diagram of the pre-built tidal harmonic model process applicable to embodiments of this application.

[0020] Figure 4 This is a schematic diagram illustrating the process of calculating and separating tidal stress components in smooth stress data based on a tidal harmonic model, and extracting the residual trend stress sequence after removing tidal response interference, applicable to embodiments of this application.

[0021] Figure 5 This is a schematic diagram illustrating the process of comprehensively comparing multiple indicators using multi-dimensional trend feature parameters and preset bank collapse risk thresholds, applicable to embodiments of this application.

[0022] Figure 6 This is a cumulative energy contribution rate curve applicable to the embodiments of this application.

[0023] Figure 7 The cutoff frequency f applicable to embodiments of this application is... c The filtering result is shown in the figure for 0.0083Hz.

[0024] Figure 8 The cutoff frequency f applicable to embodiments of this application is... c The filtering result diagram is shown at 0.001Hz.

[0025] Figure 9 The cutoff frequency f applicable to embodiments of this application is... c The filtering result is shown in the figure for 0.007Hz.

[0026] Figure 10 This is a filtering result diagram applicable to the embodiments of this application with a continuous discrimination time of 10 days.

[0027] Figure 11 This is a filtering result diagram applicable to the embodiments of this application with a continuous discrimination time of 1 day.

[0028] Figure 12 This is a graph showing the trend identification and alarm results of on-site stress applicable to embodiments of this application. Detailed Implementation

[0029] Example 1

[0030] This embodiment provides an early warning method for bank collapse in tidal river channels based on the hierarchical evolution characteristics of soil stress response, such as... Figure 1 As shown, the method may include, but is not limited to, the following steps:

[0031] Step 101: Obtain the original online monitoring stress data of the soil inside the bank section at risk of collapse.

[0032] In practice, the original online monitoring stress data of the soil within the bank section at risk of collapse is obtained. This can be achieved through joint real-time acquisition using stress-strain measurement devices deployed in these sections to acquire online monitoring stress data reflecting the effective stress within the bank slope soil. In areas of river evolution and bank erosion, bank collapse risk zones are typically defined based on the bank slope ratio, the difference in elevation between the riverbank and the channel, the proximity of the main channel to the bank, and the proximity of deep channels to the bank. Monitoring devices are preferentially deployed at locations with significant erosion at the bank slope toe, large differences in elevation between the riverbank and the channel, and steep slopes. Through continuously operating stress-strain measurement devices, the stress changes within the soil can be continuously collected at a fixed sampling frequency, providing a time-series data basis for subsequent bank collapse risk assessment.

[0033] Step 102: Perform abnormal pulse cleaning and low-pass filtering smoothing on the original online monitoring stress data to obtain smooth stress data with reduced high-frequency interference.

[0034] The raw online monitoring stress data collected is typically a superposition of trend and noise terms. The noise term is mainly caused by inherent instrument errors, ship disturbances, communication anomalies, and random environmental interference. Directly using unprocessed data for bank collapse identification can easily lead to misjudgments. By configuring a pulse pre-detection mechanism to remove extreme abnormal amplitude interference and using a low-pass filter and moving average algorithm to suppress high-frequency environmental noise, this process can output cleaner, smoother stress data without disrupting the true physical evolution of low frequencies.

[0035] Step 103: Calculate and separate the tidal stress components in the smoothed stress data based on the tidal harmonic model, and extract the residual trend stress sequence after removing tidal response interference.

[0036] For riverbanks located in tidal river sections, the internal stress of the soil is strongly and periodically modulated by tidal fluctuations. If the trend is extracted directly without differentiation, the large-scale periodic fluctuations caused by tides will mask the subtle stress accumulation resulting from gradual soil failure. A harmonic model based on the main astronomical tidal frequencies is used to fit and quantify the tidal response in the smoothed stress data, and then subtract it from the overall data. This subtraction operation achieves physical decoupling between environmental periodic disturbances and the soil's mechanical evolution characteristics; the output residual trend stress sequence only represents the changes in the soil's own mechanical state.

[0037] Step 104: Based on the residual trend stress sequence, extract multidimensional trend feature parameters to characterize the evolution of slope failure within a continuous sliding window.

[0038] After obtaining the clean residual signal, a continuous sliding window is divided with a preset time length as the step size. In the local time domain, the first-order evolution characteristics that reflect the degree of stress accumulation and the second-order kinematic characteristics that reflect the accelerated evolution state of stress are calculated. The above characteristic parameters together constitute a multidimensional trend characteristic parameter set, realizing a mathematical quantitative description of each stage of the slope from slow deformation to accelerated creep.

[0039] Step 105: Use multi-dimensional trend feature parameters and preset bank collapse risk thresholds to perform a comprehensive comparison of multiple indicators, and combine the warning level triggering and cancellation logic to output the corresponding bank collapse risk classification warning result.

[0040] The extracted multidimensional trend feature parameters are fed into the logic discrimination stage, where they are matched with a set of preset thresholds representing different risk levels using logic gates. The discrimination process requires multiple feature indicators to simultaneously meet exceedance conditions to filter out random fluctuations from single factors. To enhance the system's stability under complex operating conditions, state verification is performed using trigger and de-counting logic across different time windows. The final output, including safety status, warning level, or danger level, is used to guide on-site inspections and engineering disaster prevention scheduling.

[0041] This embodiment presents the basic workflow framework for a tidal riverbank collapse early warning method based on the graded evolution characteristics of soil stress response. The specific implementation mechanisms of data preprocessing, tidal decoupling, trend parameter calculation, and multi-index early warning involved in the above workflow will be detailed in subsequent embodiments.

[0042] According to another aspect of this application, a tidal riverbank collapse early warning system based on the graded evolution characteristics of soil stress response is also provided, comprising:

[0043] At least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor to implement a tidal riverbank collapse early warning method based on the graded evolution characteristics of soil stress response proposed in this application.

[0044] Example 2

[0045] Based on the above embodiment one, the abnormal pulse cleaning and low-pass filtering smoothing process is further described in detail.

[0046] In one possible implementation, such as Figure 2 As shown, abnormal pulse cleaning of raw online monitoring stress data may include, but is not limited to, the following steps:

[0047] Step 201: Obtain the median and absolute median difference of the original online monitoring stress data within a local time window.

[0048] In practice, a fixed sampling frequency of once per minute is set. A finite-length data sequence containing the current processing point is extracted to form a local time window. The median of the numerical values ​​in this sequence is extracted as the median. For each discrete data point, the absolute value of the difference between it and the median is calculated, and then the median of all absolute values ​​is calculated to obtain the absolute median difference. This operation uses the statistical median instead of the arithmetic mean as the benchmark, reducing the disturbance of isolated high-amplitude abrupt changes on the overall baseline assessment.

[0049] Step 202: Calculate the standardized deviation of each data point based on the median and absolute median difference.

[0050] The data within the window is measured and calculated using a standardized processing formula. The specific calculation is as follows:

[0051] Z score =(X i -X median ) / MAD value ;

[0052] Among them, Z score X is the standardized deviation of the current data point. i X represents the original stress value for the current data point. median MAD is the median of the data. value This represents the absolute median difference.

[0053] For example, suppose the median of the data extracted within a local time window is 20 MPa, the absolute median deviation is 1.5 MPa, and the original stress value at a certain sampling time is 29 MPa. Calculate the difference between this and the data median, which is 9 MPa. Divide this difference by the absolute median deviation to obtain the current standardized deviation of 6.0.

[0054] Step 203: Data points with a standardized deviation greater than a preset judgment threshold are identified as pulse anomalies, and the pulse anomalies are replaced using the median of the data to obtain the cleaned stress data.

[0055] In this embodiment, the preset judgment threshold is set to 5. Those skilled in the art will understand that this threshold can be adaptively adjusted according to the noise level of the actual monitoring data.

[0056] In one possible implementation, low-pass filtering smoothing may include the following process:

[0057] Step a: The cleaned stress data is filtered using a pre-set low-pass filter to suppress high-frequency interference components, resulting in filtered stress data. The cutoff frequency of the low-pass filter is determined based on the cumulative energy contribution rate of the stress signal to retain low-frequency trend components that characterize the stress variation trend.

[0058] In engineering practice, the Nyquist frequency coverage area corresponding to the sampling frequency contains a large amount of irregular environmental clutter, making frequency domain analysis a necessary operation to separate high-frequency signals. A Fast Fourier Transform (FFT) is performed on the original signal to construct a mapping relationship between frequency and amplitude, and the integral contribution of different frequency bands to the total power spectral density is calculated. Although the high-frequency component accounts for less than 1% of the energy, its time-domain behavior is characterized by short-time oscillations, which can easily cause distortion in the filter response. In a specific configuration, the cutoff frequency of the low-pass filter can be set to 0.007Hz. In this embodiment, the frequency corresponding to a cumulative energy contribution rate of 99% is selected as the cutoff frequency, which is determined to be 0.007Hz in this embodiment. This frequency value serves as the passband boundary, blocking high-frequency interference signals above this frequency from penetrating.

[0059] Alternatively, in some implementation scenarios where time delay errors need to be avoided, the operation of filtering the cleaned stress data using a pre-set low-pass filter can be broken down into the following two execution steps:

[0060] Step a1: Use an infinite impulse response low-pass filter to perform real-time forward filtering calculations on the cleaned stress data to obtain the filtered stress data.

[0061] Real-time online systems rely on data samples from both current and historical moments. Zero-phase forward and backward filtering algorithms, due to their non-causal nature, cannot be applied to online early warning systems with strict time delay constraints. For example, a fourth-order Butterworth infinite impulse response low-pass filter can be used to process the input data sequentially along the positive time axis, achieving high-frequency noise suppression without introducing data buffering delays.

[0062] Step a2: Obtain the known group delay characteristics of the infinite impulse response low-pass filter, and use the known group delay characteristics to perform phase delay compensation on the time axis during the subsequent extraction of multidimensional trend feature parameters.

[0063] An infinite impulse response (IRFR) low-pass filter causes phase lag in each frequency component during operation, resulting in a time-axis shift in the extracted physical trend. The group delay value within the passband is analyzed based on the filter's design parameters. Subsequently, when performing sliding window segmentation and time stamp alignment on the filtered stress data, the timestamps are adjusted in the opposite direction based on this group delay value. This operation corrects the time deviation in the trend feature extraction process.

[0064] Step b: Use a sliding window with a preset time length to perform a moving average process on the filtered stress data to obtain smoothed stress data.

[0065] After low-pass filtering, a small amount of short-term random fluctuations may remain within the passband. A sliding window with a time length of 1 hour is configured, and the data is shifted forward sequentially along the time series using this window as the statistical unit. The arithmetic mean is calculated for the data points falling within the window. This second statistical averaging operation reduces the impact of local extreme values ​​on linear fitting, making the output smooth stress data curve more stable in reflecting the changes in the internal physical state of the soil.

[0066] Example 3

[0067] After obtaining smooth stress data, this embodiment further details the specific process of constructing and separating the tidal harmonic model, taking into account the periodic disturbances unique to tidal river sections.

[0068] In one possible implementation, such as Figure 3 As shown, the tidal harmonic model can be pre-built through the following steps:

[0069] Step 301: Obtain the preset astronomical frequencies of multiple major tidal constituents, and construct the corresponding periodic harmonic functions based on the astronomical frequencies.

[0070] In practice, known a priori astronomical tidal parameters are extracted. The internal stress of the soil in the tidal river section is affected by the periodic changes in tidal level, which can be equivalent to the superposition of a finite number of tidal constituent responses. The four main tidal constituents are configured as M2 primary lunar semi-diurnal tide, S2 primary solar semi-diurnal tide, K1 lunar solar declination diurnal tide, and O1 primary lunar declination diurnal tide. The corresponding astronomical frequencies are obtained, with the unit set to times / min. Independent periodic harmonic functions are constructed for each astronomical frequency using sine and cosine functions, serving as basis functions characterizing stress response fluctuations.

[0071] Step 302: Obtain a preset linear trend term to prevent long-term stress trends from being misclassified as tidal components.

[0072] Over a long time span, the cumulative damage to the slope soil itself manifests as a slow linear or nonlinear stress increase. If only periodic and constant terms are configured in the harmonic fitting, the algorithm will forcibly fit this non-periodic long-term physical growth value into the calculation of the periodic amplitude, resulting in distortion of the extracted pure trend. When constructing the mathematical model, a monotonic linear trend term containing a time variable is explicitly introduced to absorb the non-periodic baseline drift within the time window.

[0073] Step 303: Combine the various periodic harmonic functions with the preset linear trend term to construct the tidal harmonic model.

[0074] The periodic harmonic functions, constant bias terms, and preset linear trend terms of the four main tidal constituents extracted above are mathematically added together to form a model that can describe the dynamic evolution of stress during the tidal period. The mathematical relationships of the tidal harmonic model are as follows:

[0075] σ tidal (t)=σ0+Σ k=1 K [A k *cos(2πf k t)+B k *sin(2πf k t)]+λt;

[0076] Where, σ tidal (t) represents the output value of the tidal stress response model at time t, σ0 is the constant bias term, k is the traversal index number of the tidal constituent, and K is the number of major tidal constituents. In this embodiment, K is taken as 4. A k With B k The tidal stress response coefficients for the k-th tidal constituent are fk and fk respectively. k Let λ be the astronomical frequency of the k-th tidal constituent, and λ be the slope coefficient of the linear trend term.

[0077] In one possible implementation, the tidal stress component in the smoothed stress data is calculated and separated based on the tidal harmonic model, and the residual trend stress sequence after removing tidal response interference is extracted, such as... Figure 4 As shown, the process may include the following:

[0078] Step S01: Extract historical calibration window data of a preset time length from the smoothed stress data.

[0079] A continuous and valid data segment preceding the current moment on the timeline is selected as the sample set for model calibration. The preset time length needs to cover a sufficient number of complete tidal cycles to ensure that the algorithm can separate semi-diurnal and diurnal tide components with similar frequencies. As an example, this preset time length can be 15 days. The smoothed stress data from these 15 days is used to construct the historical calibration window data.

[0080] Step S02: Based on historical calibration window data and the tidal harmonic model, establish an overdetermined set of equations and calculate the tidal stress response coefficients corresponding to each major tidal constituent.

[0081] Stress values ​​at multiple discrete sampling times are extracted within the historical calibration window. The observed values ​​corresponding to the time series are used as the target vector on one side of the equation system. The periodic harmonic function values ​​calculated from time and frequency are combined with the linear time values ​​to form the design matrix. Since the number of observed data points is much larger than the number of tidal stress response coefficients to be determined, an overdetermined equation system is formed. The least squares method is used to numerically solve this overdetermined equation system, obtaining the constant bias term, the amplitude response coefficients corresponding to each tidal constituent, and the coefficients of the linear trend term.

[0082] Step S03: Substitute the current acquisition time into the tidal harmonic model containing the tidal stress response coefficient after removing the preset linear trend term, and calculate the tidal stress component at the current time.

[0083] After solving the overdetermined equations and locking the tidal stress response coefficients, the preset linear trend terms in the model are removed, leaving only the constant bias terms and periodic harmonic function terms. The latest timestamp value is then substituted into the truncated mathematical model to calculate the increase or decrease in stress caused by pure tidal fluctuations at that moment. The output is the tidal stress component.

[0084] Step S04: Subtract the tidal stress component from the smoothed stress data to obtain the residual trend stress sequence.

[0085] The smoothed stress data, acquired at the current moment and processed through low-pass filtering, is subtracted from the tidal stress components derived in the previous step. This subtraction removes the interference response caused by the physical environment's tidal level acting on the soil within the data stream. This subtraction operation is performed continuously along the time axis, resulting in a sequence of residual trend stresses free from tidal response interference. This sequence is used to characterize the true evolution of the slope soil's mechanical state.

[0086] Step S05, wherein, according to the preset sliding update cycle, the latest historical calibration window data is extracted and the tidal stress response coefficient is recalculated to realize the dynamic update of the tidal harmonic model.

[0087] The preset sliding update cycle can be set to 1 day. Every time this cycle occurs, the time frame is slid towards the current moment, discarding the oldest data segment and incorporating the latest smoothed stress data, thus re-triggering the construction and solution steps of the overdetermined equations. Through this periodic parameter overwriting mechanism, the harmonic model's ability to dynamically track changes in the physical state of the monitored medium is maintained.

[0088] After the above-mentioned tidal harmonic separation processing, the periodic tidal level interference component in the residual trend stress sequence is effectively suppressed. Subsequent feature extraction can be performed based on signals that only reflect changes in the mechanical state of the soil itself, avoiding the masking of the cumulative trend of slope damage caused by large-amplitude periodic fluctuations caused by tides.

[0089] Example 4

[0090] Based on the above-mentioned acquisition of residual trend stress sequence, this embodiment provides an implementation method for multidimensional trend feature extraction, and elaborates on the specific execution process of the above step 104.

[0091] In one possible implementation, the multidimensional trend characteristic parameters include first-order evolution characteristic parameters that reflect the amount of stress accumulation evolution, and second-order kinematic characteristic parameters that reflect the stress-accelerated creep state.

[0092] When extracting physical features to determine the structural state of riverbank slopes, single-dimensional rate indicators can easily miss precursors of nonlinear abrupt changes. This paper maps the input time-series features to two independent dimensional categories in terms of physical properties. The first category focuses on the first-order state shift of the data sequence over time, i.e., the rate and macroscopic direction of stress evolution, used to measure the basic magnitude of accumulated failure. The second category further differentiates and transforms the first-order features to extract second-order kinematic features characterizing changes in the stress growth rate, used to identify accelerated creep segments in the riverbank slope failure cycle. This decoupling of dimensions constructs a multi-dimensional feature system that comprehensively describes the mechanical state of the soil.

[0093] An alternative approach to extracting first-order evolutionary feature parameters may include, but is not limited to, the following process:

[0094] Step 401: Perform linear fitting on the residual trend stress sequence within the continuous sliding window to calculate the local fitting slope that reflects the overall trend direction and rate.

[0095] Obtain a pre-defined continuous sliding window length, which can be set to a one-day time span. Step the continuous sliding window along the time axis, moving the data span by one hour in each step. Perform a univariate linear regression using standard least squares for all residual trend stress sequence data points falling within the current continuous sliding window. Solve for the slope parameter of the regression equation using time as the independent variable and residual stress value as the dependent variable. This locally fitted slope, by comprehensively considering all data fluctuations within the window, provides information on the direction and slope of stress changes in an overall statistical sense.

[0096] Step 402: Calculate the arithmetic mean of the data segment at the beginning and the data segment at the end of the continuous sliding window, and calculate the daily average rate of change reflecting the net change in stress based on the difference between the arithmetic mean of the data segment at the end and the arithmetic mean of the data segment at the beginning.

[0097] Further, it can be implemented by calculating the arithmetic mean of the starting data segment and the ending data segment of the continuous sliding window respectively, and by dividing the difference between the arithmetic mean of the ending data segment and the arithmetic mean of the starting data segment by the time interval between the corresponding midpoints of the starting data segment and the ending data segment, to calculate the daily average rate of change reflecting the net change in stress.

[0098] To eliminate the influence of local extreme mutations on the first and last data points, a predetermined number of data points from the earlier part of the continuous sliding window are extracted as the starting data segment, and a corresponding number of data points from the later part are extracted as the ending data segment. This predetermined number can be 60 sampling points, corresponding to one hour of physical time. The arithmetic mean of all values ​​contained in the starting and ending data segments is calculated separately. A subtraction operation is then performed, and the result is divided by the time span.

[0099] The specific calculations are as follows:

[0100] v daily =(σ last -σ first ) / ΔT;

[0101] Among them, v daily σ is the daily average rate of change. last σ is the arithmetic mean of the last data segment. first ΔT is the arithmetic mean of the starting data segment, and ΔT is the time interval between the corresponding midpoints of the starting and ending data segments.

[0102] This operation uses the statistical mean at both ends of the time window as a reference to measure the magnitude of the net physical change that occurs within the time window.

[0103] Step 403: The local fitting slope and the daily average rate of change together constitute the first-order evolutionary characteristic parameters.

[0104] Under complex stress conditions, soil stress curves exhibit local troughs and peaks. When the curve shows a fluctuating characteristic of first decreasing and then rising, the daily average rate of change extracted from the beginning and end may be calculated to a higher value, but the least-squares local fitting slope of the overall data approaches zero. Defining both as first-order evolutionary characteristic parameters can provide a basis for comparison in subsequent steps, avoiding misjudging local fluctuations as long-term monotonic dangerous trends.

[0105] An alternative approach to extracting second-order kinematic feature parameters may include, but is not limited to, the following process:

[0106] Step 1a: Calculate the difference between the daily average rate of change of two adjacent consecutive sliding windows to obtain the stress change acceleration, and perform moving average smoothing on the stress change acceleration to obtain the smoothed acceleration.

[0107] In some embodiments, the difference in the daily average rate of change between two adjacent consecutive sliding windows is calculated and divided by the sliding window step size to obtain the stress change acceleration.

[0108] To identify whether there is accelerated deterioration of slope stress, a time-axis difference operation is performed on the first-order evolution parameters.

[0109] The specific calculations are as follows:

[0110] a m =(v m -v m-1 ) / Δt step ;

[0111] Among them, a m v represents the acceleration due to stress change corresponding to the current continuous sliding window. m v represents the average daily rate of change of the current continuous sliding window. m-1 Let Δt be the daily average rate of change of the immediately preceding consecutive sliding window on the timeline. step The time interval between the center moments of two adjacent consecutive sliding windows is called the sliding step size. In this embodiment, the sliding window step size is a fixed 1 hour. In actual calculations, the acceleration comparison between windows depends only on the relative magnitude of the rate difference, and this fixed divisor does not affect the logical consistency of subsequent threshold determination.

[0112] Since difference operations amplify numerical fluctuations, a smoothing operation is further performed on the acceleration values ​​over multiple days:

[0113] a avg =(a m +a m-1 +a m-2 ) / 3;

[0114] Among them, a avg To smooth the acceleration, a m a m-1 and a m-2 These represent the stress change accelerations output by three time-continuous sliding windows. This averaging operation reduces the noise sensitivity caused by second-order differentiation and outputs a smooth acceleration that stably characterizes the stress increase state.

[0115] Due to the sliding window step size Δt step As a fixed constant, this fixed divisor does not affect the logical consistency of positive / negative judgment and relative size relationship in comparison operations and threshold discrimination. Therefore, in some simplified implementations, the division operation can be omitted and the acceleration feature can be directly represented by the rate difference.

[0116] Step 3a: Use smooth acceleration as the second-order kinematic characteristic parameter.

[0117] The second-order kinematic characteristic parameters, which constitute the smooth acceleration of stress change acceleration state, not only characterize the development trend of bank slope failure, but also provide a time prediction benchmark with engineering guidance value. The first-order evolution characteristic parameters and the second-order kinematic characteristic parameters serve as the basic inputs of the early warning logic and are transmitted to the next comparison and processing stage.

[0118] By jointly extracting the first-order and second-order features, information on the state transition of the bank slope from slow cumulative damage to accelerated creep can be captured simultaneously. The dual-index design of the first-order parameters avoids the risk of misjudgment in V-shaped fluctuations caused by a single rate index.

[0119] Example 5

[0120] Based on the above extraction of multidimensional trend feature parameters, the early warning judgment and terminal linkage links are further explained in detail.

[0121] In one possible implementation, multi-dimensional trend characteristic parameters are compared with a preset bank collapse risk threshold using a comprehensive comparison of multiple indicators, such as... Figure 5 As shown, the steps may include, but are not limited to, the following:

[0122] Step 501: Divide the preset bank collapse risk thresholds into a warning level threshold set and a danger level threshold set. Each threshold set contains a corresponding rate threshold and slope threshold.

[0123] Multiple sets of numerical boundaries are pre-configured as risk assessment benchmarks. The warning-level threshold set corresponds to the evolutionary state of the bank slope in which continuous damage accumulation has not yet reached the instability limit, while the danger-level threshold set corresponds to the state where the bank slope structure is about to undergo substantial damage. Each threshold set includes two independent discrimination limits decoupled from first-order features: a slope threshold for assessing the overall direction of stress evolution, and a rate threshold for measuring the absolute net change in physical quantities. The coordinated configuration of multi-dimensional thresholds provides boundary conditions for fault-tolerant assessment in complex hydrological environments.

[0124] Step 502: When the local fitting slope exceeds the corresponding slope threshold and the daily average rate of change simultaneously exceeds the corresponding rate threshold, or when the smoothing acceleration is continuously positive within a preset period and the daily average rate of change exceeds the warning baseline, it is determined that the current sliding window meets the triggering conditions of the corresponding risk level.

[0125] Specifically, it can also be implemented as follows: when the local fitting slope exceeds the corresponding slope threshold and the daily average rate of change also exceeds the corresponding rate threshold, or when the smoothing acceleration is positive within a consecutive preset number of consecutive sliding windows and the daily average rate of change exceeds the corresponding rate threshold in the warning level threshold set, the current sliding window is determined to meet the triggering conditions of the corresponding risk level.

[0126] An alternative discrimination path based on second-order kinematic parameters is introduced. When the smooth acceleration remains positive for three consecutive adjacent sliding windows and the daily average rate of change exceeds the corresponding rate threshold in the warning level threshold set, the soil deformation is determined to have entered the accelerated creep zone, and the corresponding risk trigger state is marked in the same sliding window.

[0127] This step is configured with a dual-condition judgment threshold based on logical AND operation. If only the daily average rate of change is relied upon, when the monitored data shows a fluctuating pattern of decline followed by recovery due to non-monotonic environmental disturbances, it is easy to calculate an excessively high difference between the first and last average values, leading to false alarms. If only the local fitting slope is relied upon, when the stress is in a monotonically slow increase but the net increment does not reach the dangerous level, there is also no engineering risk.

[0128] The overall trend and net increment of the soil's mechanical response are required to simultaneously exceed the safety boundary to filter out local pseudo-fluctuations. In addition, an alternative discrimination path based on second-order kinematic parameters is introduced. When the daily average rate of change does not reach the high-level setting, but the smooth acceleration remains positive within the preset period and the daily average rate of change crosses the warning baseline, the soil deformation is determined to have entered the accelerated creep zone, and the corresponding risk trigger state is marked within this sliding window.

[0129] An alternative approach is that the alert level triggering and deactivation logic may include the following process:

[0130] Step ①: When multiple consecutive adjacent sliding windows meet the triggering conditions of the same risk level, and the number of consecutively met windows reaches the first preset number corresponding to that level, the corresponding bank collapse risk classification early warning result is triggered and output.

[0131] Parameter exceedances within a single time window may be caused by sporadic data noise. To suppress false alarms, a time-series accumulation judgment mechanism is configured along the time axis, with a first preset number set for each level. The first preset number for the warning level can be set to 3, while for the danger level, due to the smaller remaining fault tolerance margin, the first preset number can be set to 2. The calculation unit continuously tracks the number of windows in a high-risk trigger state, and only when this number is greater than or equal to the first preset number is a substantive warning level transition operation performed.

[0132] Step ②: When the corresponding risk warning has been triggered, the risk warning is lifted or reduced only when the judgment results of multiple consecutive adjacent sliding windows are all lower than the threshold corresponding to the current warning level, and the number of consecutive windows lower than the threshold reaches the second preset number; wherein, in order to prevent the warning level from changing frequently, the second preset number is greater than the first preset number.

[0133] High-frequency fluctuations in hydrological conditions can cause extracted stress characteristic parameters to repeatedly penetrate the judgment baseline, triggering frequent flashing of the system's warning level. To enhance operational stability, an asymmetric hysteresis mechanism is introduced into the warning downgrade or cancellation procedure, configuring a larger second preset number. The second preset number for canceling the warning or danger level is uniformly defined as 6. When performing state fallback judgment, the number of observation windows where data remains within the safe range is required to be greater than the number of observation windows needed to trigger the alarm, constructing a state transition path with fault tolerance depth.

[0134] The aforementioned asymmetric hysteresis control strategy ensures that the system's early warning status remains stable under complex hydrological fluctuations, reducing frequent increases and decreases in warning levels caused by occasional data fluctuations. The design, which requires fewer windows to trigger than to deactivate, ensures a rapid response to actual risk trends while preventing premature deactivation of alarms during risk mitigation.

[0135] The following typical operating scenario illustrates the working process of the above-mentioned discrimination logic. Assume that during continuous observation at a certain monitoring point, the residual trend stress sequence from day T1 to T2 is affected by non-tidal short-term water level disturbances, exhibiting a V-shaped fluctuation pattern of first decreasing and then rebounding. Within this V-shaped fluctuation range, because the stress value at the end rebounds to near the level at the beginning, the daily average rate of change is calculated as a relatively high positive value. However, the least squares fit of the full window data shows that the overall slope approaches zero. In the dual-logic-gate judgment, because the local fitting slope does not exceed the slope threshold, even if the daily average rate of change exceeds the rate threshold, it is determined that the current window does not meet the warning-level triggering condition, thus avoiding a false alarm.

[0136] Further assuming that the stress at this monitoring point enters a true monotonic upward phase starting from day T3, the local fitting slope and daily average rate of change of the first, second, and third consecutive sliding windows simultaneously exceed the warning level threshold, and the number of consecutive windows meeting the first preset number of three triggers a warning. Subsequently, on day T4, an indicator in one window falls back below the threshold, but since the number of falling windows is only one, far below the second preset number of six, the warning level remains unchanged. This asymmetric hysteresis characteristic effectively suppresses frequent jumps in warning levels caused by short-term data fluctuations.

[0137] In practical engineering applications, the method of this invention was retrospectively verified using three consecutive months of online stress data from a monitoring station in a tidal river section. The results show that, compared to conventional methods that directly determine a single indicator based on the original stress change rate and a fixed empirical threshold, the method of this invention reduces the number of false triggers caused by tidal periodic fluctuations and enhances the stability of early warning outputs under the same monitoring conditions. Those skilled in the art will understand that the specific performance improvement may vary depending on the geological conditions of the river section, the tidal intensity, and the sensor deployment scheme.

[0138] An optional approach, after outputting the corresponding bank collapse risk classification and early warning results, also includes:

[0139] Step S1: The risk classification and early warning results of bank collapse and the corresponding multi-dimensional trend characteristic parameters are simultaneously sent to the remote monitoring terminal.

[0140] The output risk status assessment conclusion, along with relevant underlying quantitative indicators such as slope values, rate values, and potential instability time, is encapsulated into a network data frame and pushed to a remote monitoring terminal via a wireless network or fiber optic channel. This step provides quantitative technical support for rear-area flood control dispatch personnel to assess the stability trend of the bank slope and guide on-site investigations of bank cracks and slope toe scour.

[0141] Step S2: When the triggering conditions of the danger level are determined to be met, the local audible and visual alarm devices deployed on the shoreline are automatically activated, and a monitoring interruption or intervention prompt command is generated.

[0142] When the conditions for triggering a dangerous level are met, warning information will be displayed and dynamically released on the remote monitoring terminal.

[0143] Example 6

[0144] Regarding the various preset risk thresholds involved in Example 5, traditional fixed empirical thresholds are prone to failure due to differences in geological conditions across different river sections. Therefore, this example provides a threshold construction method.

[0145] In one possible implementation, the preset bank collapse risk threshold can be pre-built through the following steps:

[0146] Step 601: Obtain historical online monitoring stress data during the initial stable phase of system deployment, and use it as data for the stable baseline period.

[0147] In practice, the basic stress state and fluctuation characteristics of the riverbank soil differ in different river sections. During the initial operational phase after the physical deployment of the monitoring system, data from the first 60 days after deployment are extracted as the stabilization baseline. This period typically corresponds to a safe state where the riverbank has not experienced significant displacement or accelerated deformation. Historical online monitoring stress data for this interval are obtained to establish a baseline reflecting the characteristics of the monitoring points themselves.

[0148] Step 602: Extract the corresponding historical daily rate of change sequence based on the stable baseline data, and calculate the statistical mean and statistical standard deviation of the historical daily rate of change sequence.

[0149] The obtained continuous sequence data is segmented, and the daily average rate of change within each sliding time window is calculated to form a complete historical daily average rate of change sequence. Statistical operations are performed on this one-dimensional numerical array. The statistical mean, which characterizes the central tendency of daily average changes under steady-state conditions, is obtained, along with the statistical standard deviation, a parameter characterizing the degree of dispersion of data due to environmental disturbances. The extraction of these statistical parameters establishes the basic mathematical expectation and discrete distribution characteristics for subsequent calculations of personalized dynamic thresholds.

[0150] Using the same methodology, the corresponding historical local fitting slope sequence is extracted based on the stable baseline data, and the statistical mean and statistical standard deviation of the historical local fitting slope sequence are calculated for subsequent construction of slope threshold.

[0151] Step 603: Configure corresponding preset amplification weights for different bank collapse risk levels, and combine the statistical mean with the statistical standard deviation after processing with the corresponding preset amplification weights to construct and generate preset bank collapse risk thresholds for each risk level.

[0152] Weighting coefficients with an increasing relationship are pre-assigned to warning and danger levels. In this embodiment, the preset amplification weight corresponding to the warning level threshold is set to 6, and the preset amplification weight corresponding to the danger level threshold is set to 20. Those skilled in the art can adaptively adjust the above weight values ​​according to the geological conditions and disaster prevention management needs of the river section where the monitoring point is located. The statistical standard deviation is multiplied by the configured preset amplification weight, and the resulting product is added to the statistical mean. The specific calculation is as follows:

[0153] v warn =μ0+α1*s0;

[0154] Among them, v warn α1 is the warning level threshold, μ0 is the statistical mean, α1 is the preset amplification weight assigned to the warning level, and s0 is the statistical standard deviation.

[0155] Accordingly, the formula for calculating the hazard level threshold is:

[0156] v danger =μ0+α2s0;

[0157] Among them, v danger α1 is the threshold for the danger level, and α2 is the preset amplification weight assigned to the danger level. In this embodiment, α2 is set to 20.

[0158] The slope threshold is constructed in the same way as the rate threshold. It is based on the statistical mean and statistical standard deviation of the historical local fitted slope sequence, and the slope threshold for each risk level is calculated with the corresponding preset amplification weight.

[0159] This operation makes the output preset bank collapse risk threshold no longer a fixed, rigid constant, but a dynamic boundary that adaptively floats based on the local soil background fluctuation characteristics at the monitoring point.

[0160] Example 7

[0161] During the real-time online monitoring process described in Examples 1 to 5 above, data acquisition may be interrupted due to harsh outdoor environments. This example provides a parallel conditional traffic splitting mechanism, specifically detailing the hierarchical fault-tolerant processing strategy in the event of data communication interruption.

[0162] In one possible implementation, if a data communication interruption occurs during the acquisition of raw online monitoring stress data, resulting in data loss, a graded loss handling strategy is executed.

[0163] Sensor networks deployed at engineering sites may experience data packet loss or channel outages due to power supply fluctuations or wireless signal attenuation. The system is equipped with a timestamp continuity monitoring module. When the interval between two consecutive received data timestamps exceeds the preset system sampling period, data loss is determined. At this point, the subsequent routine frequency domain filtering operations are halted, and a parallel, tiered fault-tolerant processing branch is initiated. Based on the cumulative outage duration, the processing flow is divided into three independent execution conditions.

[0164] Step 701: Under the first working condition, for short-term missing data with an interruption time less than the first duration, linear interpolation is used to fill in the missing data segments.

[0165] Obtain the current outage time span for calculation. A pre-configured first duration is set as the time threshold for short-term missing data; for example, this threshold corresponds to 5 consecutive sampling points, or 5 minutes. When the detected outage time does not exceed this first duration, it indicates that the missing data span is small and has limited impact on the overall envelope of low-frequency trend features. Extract the coordinates of the last valid data point before the missing segment occurred, and the coordinates of the first valid data point after communication was restored. Construct a linear geometric line segment using these two points, and generate interpolated values ​​along the time axis to fill the gaps. The interpolated data sequence then participates normally in subsequent filtering and trend calculations.

[0166] Step 702: Under the second working condition, for mid-time missing data with an interruption time greater than or equal to the first duration and less than the second duration, the average of the most recent valid data segment before the missing data is used to fill the missing data, and a confidence downgrade mark is added to the warning judgment result containing data of that time period.

[0167] A second duration is pre-configured as the upper limit threshold for medium-scale missing data; for example, this threshold corresponds to 60 consecutive sampling points, or 60 minutes. When the interruption time reaches or exceeds the first duration but is less than the second duration, the linear interpolation method is prone to producing spurious slopes due to the large span, leading to distortion in subsequent first-order feature calculations. Multiple consecutive valid sampling points before the missing time are extracted, for example, the most recent 30 valid points, and the arithmetic mean of this batch of data is calculated. This mean sequence is used to fill the entire interruption interval at equal intervals. Since the filled data represents a shifted historical state, identifiers are generated at the calculation nodes of subsequent consecutive sliding windows involving this time period. When the amount of data containing such low-confidence identifiers exceeds 10% of the current 1-day discrimination window, the bank collapse risk classification warning result output by this window will be appended with a confidence downgrade flag to indicate that the discrimination result in subsequent stages relies on approximate substitute data.

[0168] Step 703: In the third working condition, for long-term data loss with an interruption time greater than or equal to the second duration, trigger the data interruption state and suspend automatic early warning judgment.

[0169] When the interruption time continues to climb and exceeds the second duration (i.e., the time span corresponding to 60 consecutive sampling points), it indicates a possible physical disconnection of the communication link or a failure of the data acquisition equipment. Interpolation or extrapolation based on historical data faces the risk of failure, and continuing to execute trend fitting will output false alarms with no physical meaning. The control logic intercepts the data stream from entering the trend calculation layer, triggering the data interruption status variable. A monitoring interruption alarm is output, prompting relevant personnel to perform manual intervention and on-site repairs. After data transmission is restored, the startup phase program is re-executed to establish a new data observation window.

[0170] Example 8

[0171] Unlike Embodiments 1 to 6 described above, this embodiment provides an alternative implementation method for early warning and discrimination in inland river sections unaffected by tides. The difference lies in omitting the tidal harmonic model construction and separation processing step for periodic tidal rise and fall. Instead, the smoothed stress data output from the low-pass filter is directly input into the trend feature extraction module, and a fixed engineering experience threshold is used to perform risk discrimination. The remaining data acquisition and pre-filtering steps are the same as in Embodiment 1, and can be referred to the foregoing description; they will not be repeated here.

[0172] Smoothed stress data was obtained after processing with a low-pass filter and a moving average algorithm. A sliding time window (continuous sliding window) was constructed with one day as the continuous discrimination time (continuous sliding window length), and this time window was stepped along the data time axis. Within each time window, the smoothed stress data was fitted using the least squares method with a univariate linear fit, and the local fitting slope was calculated. At the same time, the daily average rate of change was calculated based on the changes in stress data at the beginning and end of the time window. Since there are no large-amplitude periodic physical oscillations caused by tides in inland river sections, the locally fitted slope and daily average rate of change obtained by direct fitting can effectively characterize the long-term true evolution state of the internal stress of the soil.

[0173] A pre-configured fixed empirical threshold is used as the benchmark for assessing bank collapse risk. This fixed empirical threshold is pre-set based on historical geological disaster survey statistics for the inland river section, eliminating the need for adaptive generation based on initial equipment deployment data. The comparison module extracts feature parameters within the current sliding time window. When the local fitting slope is determined to be greater than 0.03 kPa / min and the daily average rate of change is greater than 40 kPa / day, a warning-level bank collapse risk status indicator is output. This status indicator suggests that the slope soil is under unfavorable boundary conditions, the stress change rate is approaching a critical value, and a significant stress accumulation failure process has begun.

[0174] During continuous monitoring, the judgment module cyclically performs the aforementioned comparison operation within each sliding time window. When the calculated local fitting slope is greater than 0.1 kPa / min and the daily average rate of change is greater than 150 kPa / day, a hazard-level bank collapse risk status indicator is output. This status indicator indicates that the bank slope soil has transitioned from the precursor accumulation stage to the substantial accelerated instability stage.

[0175] Field measurements revealed that the stress response process could be summarized into three distinct progressive stages: a development from relative stability and continuous accumulation to accelerated instability, exhibiting strong consistency across these stages. Figure 12 As shown.

[0176] This alternative embodiment skips complex harmonic decoupling calculations and, in hydrological scenarios without tidal disturbances, achieves effective identification and classification of riverbank collapse disasters with low processor computing power consumption by comparing numerical thresholds with fixed boundaries.

[0177] According to one aspect of this application, the layout of soil stress monitoring points in bank collapse risk sections may specifically include:

[0178] Based on the slope ratio of the riverbank section, the difference in elevation between the shoal and the channel, the proximity of the main stream to the bank, the proximity of the deep channel to the bank, and the historical characteristics of bank collapse development, the risk area of ​​bank collapse is determined, and an online soil stress monitoring device is deployed in the area.

[0179] In this embodiment, the monitoring target is a section of the right bank of a sandbar in a typical branching section of a large river, which is at risk of bank collapse. This section is affected by the main current approaching the bank, the deep channel pressing against the bank, and the rise and fall of the tides. It suffers from strong near-shore erosion and poor bank slope stability, making it suitable as a typical bank collapse early warning monitoring area.

[0180] Monitoring points should be prioritized in locations with significant erosion at the toe of the bank slope, large differences in elevation between the bank and channel, steep slopes, and areas with a history of frequent landslides. For key risk areas, the number of stress monitoring points can be increased to improve the ability to detect local instability and sudden stress changes.

[0181] According to one aspect of this application, the process of frequency domain analysis and filtering / denoising of the original stress data can also include:

[0182] First, the original stress sequence was decomposed in the frequency domain using Fast Fourier Transform (FFT) to obtain the frequency-amplitude spectrum, identifying the main frequency components and energy distribution characteristics in the original stress signal. Spectral analysis revealed that the information truly reflecting the incubation, development, and failure process of the slope collapse was mainly concentrated in the low-frequency range, while the high-frequency range consisted mostly of random noise and occasional anomalous disturbances. From a signal decomposition perspective, the on-site stress time series can be represented as the superposition of a trend term and a noise term, i.e.:

[0183] ;

[0184] In the formula, S(t) is the original stress sequence; η(t) is the low-frequency trend term reflecting the stress accumulation and redistribution process of the bank slope; It consists of random noise, high-frequency disturbances, and outlier components.

[0185] like Figure 6 As shown, the cumulative energy contribution rate curve is plotted based on the FFT analysis results, and the cutoff frequency of the Butterworth low-pass filter is determined by combining the energy contribution rate. The amplitude-frequency response function of the Butterworth low-pass filter is:

[0186] H(f) = 1 / sqrt(1 + (f / f)) c ) 2N );

[0187] Where f is the signal frequency, f c Where N is the cutoff frequency and N is the filter order.

[0188] In this embodiment, the sampling frequency of the on-site stress monitoring data is 1 time / min, i.e., the sampling frequency f s =1 / 60Hz, the Nyquist frequency is 0.0083Hz, therefore the cutoff frequency f c It should be less than 0.0083Hz.

[0189] A comparative analysis of the filtering results at different cutoff frequencies was conducted, such as... Figure 7 As shown, when the cutoff frequency is f c At 0.0083Hz, the filtered trend curve still retains strong high-frequency fluctuations, indicating insufficient noise suppression. Figure 8 As shown, when the cutoff frequency is f c At 0.001Hz, the curve exhibits obvious over-smoothing, significantly weakening the rapid stress rise process in the early stage of bank collapse, making it difficult to meet the requirements for early warning identification. For example... Figure 9 As shown, for f c Analysis of the filtering results at 0.007Hz shows that this parameter effectively filters out high-frequency noise while also preserving the true trend of continuous stress rise and rapid enhancement, achieving a good balance between smoothness and fidelity. Considering the spectral structure, energy contribution rate, and trend recognition effect, the optimal cutoff frequency f is selected. c =0.007Hz, filter order N=4.

[0190] According to one aspect of this application, the calculation of the stress variation trend within the soil may further include the following process:

[0191] For the filtered and smoothed stress sequence, least-squares linear fitting is performed using a continuous sliding time window to calculate the local trend slope and the daily average stress change rate. The continuous discrimination time refers to the length of time used for trend fitting and risk discrimination within the sliding window, reflecting the statistical scale of trend rate calculation and anomaly identification. Since bank collapse processes typically exhibit an evolutionary characteristic of long-term accumulation – short-term acceleration – abrupt instability, the selection of the discrimination time needs to strike a balance between identification sensitivity and result stability.

[0192] In this embodiment, different discrimination times are analyzed, such as... Figure 10 As shown, when the continuous discrimination time is 10 days, the trend curve is too flat and cannot reflect the rapid enhancement of the response characteristics in the short term before and after the bank collapse; for example... Figure 11 As shown, when the continuous discrimination time is set to 1 day, it can not only more sensitively capture the continuous increase of stress before the collapse and the high-level fluctuation after the collapse, but also better suppress the influence of local extreme values, making the early warning effect the most reasonable. Therefore, combining the on-site stress response characteristics and the physical process of the collapse, the continuous discrimination time of 1 day was finally determined. Based on this, a field stress monitoring data processing scheme of 4th-order Butterworth low-pass filter + cutoff frequency of 0.007Hz + 60-point moving average + 1-day continuous discrimination window was finally established.

[0193] This application eliminates high-frequency physical noise by introducing statistical median anomaly cleaning and a low-pass filtering mechanism with known group delay phase compensation, while enabling effective phase alignment of the signal on the real-time online processing time axis.

[0194] By constructing a tidal harmonic model that includes multiple major tidal constituents and non-periodic linear terms, the periodic environmental excitation is physically removed from the total signal, so that the final extracted residual trend can truly reflect the accumulation of insignificant mechanical damage to the soil itself, effectively suppressing the spurious trend interference caused by the rise and fall of tides.

[0195] To address the issue of fixed early warning thresholds failing due to varying geological conditions, an adaptive dynamic threshold construction method based on stable baseline data from the initial deployment phase is established. Combined with physical resolution lower limit constraints, the early warning boundaries for each monitoring point can adaptively float according to their own background fluctuation characteristics, enhancing the system's adaptability across river sections.

[0196] This application designs a dual-logic judgment threshold that operates in parallel with the first-order slope and rate of change, supplemented by an asymmetric hysteresis control strategy that requires the number of consecutive downgrade windows to exceed the number of upgrade trigger windows. This effectively filters out instantaneous false high values ​​caused by V-shaped fluctuations, resulting in a more stable output of the early warning status and reducing the ineffective response rate of engineering disaster prevention and control scheduling. Furthermore, a multi-level data communication interruption fault-tolerant processing strategy further ensures the continuity and robustness of the online monitoring chain under extreme operating conditions.

[0197] It should be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

Claims

1. A method for early warning of bank collapse in tidal channels based on the hierarchical evolution characteristics of soil stress response, characterized in that, include: Obtain original online monitoring stress data of the soil inside the bank section at risk of bank collapse; Abnormal pulse cleaning and low-pass filtering smoothing are performed on the raw online monitoring stress data to obtain smooth stress data with reduced high-frequency interference; The tidal stress component in the smoothed stress data is calculated and separated based on the tidal harmonic model, and the residual trend stress sequence after removing tidal response interference is extracted. Based on the residual trend stress sequence, multidimensional trend feature parameters for characterizing the evolution of bank slope failure are extracted within a continuous sliding window; By comparing multi-dimensional trend feature parameters with preset bank collapse risk thresholds, and combining the warning level triggering and cancellation logic, the corresponding bank collapse risk classification warning results are output.

2. The method according to claim 1, characterized in that, Abnormal pulse cleaning is performed on the raw online monitoring stress data, including: Obtain the median and absolute median difference of the raw online monitoring stress data within a local time window; The standardized deviation of each data point is calculated based on the median and absolute median. Data points with a standardized deviation greater than a preset threshold are identified as pulse anomalies, and the pulse anomalies are replaced using the median of the data to obtain the cleaned stress data.

3. The method according to claim 2, characterized in that, Low-pass filtering smoothing process includes: The cleaned stress data is filtered using a pre-set low-pass filter to suppress high-frequency interference components, resulting in filtered stress data. The cutoff frequency of the low-pass filter is determined based on the cumulative energy contribution rate of the stress signal to retain the low-frequency trend components that characterize the stress change trend. The filtered stress data is processed by moving average using a sliding window with a preset time length to obtain smooth stress data.

4. The method according to claim 1, characterized in that, Tidal harmonic models are pre-built through the following steps: Obtain the astronomical frequencies of several pre-defined major tidal constituents, and construct the corresponding periodic harmonic functions based on the astronomical frequencies; By combining the various periodic harmonic functions with a preset linear trend term, a tidal harmonic model is constructed.

5. The method according to claim 4, characterized in that, Based on the tidal harmonic model, the tidal stress components in the smoothed stress data are calculated and separated, and the residual trend stress sequence after removing tidal response interference is extracted, including: Extract historical calibration window data of a preset time length from the smoothed stress data; Based on historical calibration window data and tidal harmonic model, an overdetermined set of equations was established, and the tidal stress response coefficients corresponding to each major tidal component were calculated. Substitute the current acquisition time into the tidal harmonic model containing the tidal stress response coefficient after removing the preset linear trend term, and calculate the tidal stress component at the current time. Subtracting the tidal stress component from the smoothed stress data yields the residual trend stress sequence; Specifically, the tidal stress response coefficient is recalculated by extracting the latest historical calibration window data according to the preset sliding update cycle, so as to realize the dynamic update of the tidal harmonic model.

6. The method according to claim 1, characterized in that, The multidimensional trend characteristic parameters include first-order evolution characteristic parameters that reflect the evolution of stress accumulation, and second-order kinematic characteristic parameters that reflect the state of stress-accelerated creep.

7. The method according to claim 6, characterized in that, Extract first-order evolutionary feature parameters, including: Linear fitting is performed on the residual trend stress sequence within a continuous sliding window to calculate the local fitting slope that reflects the overall trend direction and rate. The arithmetic mean of the data segments at the beginning and end of the continuous sliding window is calculated separately. Based on the difference between the arithmetic mean of the end data segment and the arithmetic mean of the beginning data segment, the daily average rate of change reflecting the net change in stress is calculated. The local fitting slope and the daily average rate of change together constitute the first-order evolutionary characteristic parameters.

8. The method according to claim 1, characterized in that, The original online monitoring stress data of the soil inside the bank slope at risk of collapse is obtained. Specifically, stress and strain measurement equipment deployed in the bank slope at risk of collapse is used for joint real-time acquisition to obtain online monitoring stress data that reflects the effective stress inside the bank slope soil.

9. The method according to claim 1, characterized in that, If data communication is interrupted during the acquisition of raw online monitoring stress data, resulting in data loss, a graded data loss handling strategy will be implemented: For short-term missing data with an interruption time less than the first duration, linear interpolation is used to fill in the missing data segments. For mid-time missing data with an interruption time greater than or equal to the first duration and less than the second duration, the average of the most recent valid data segment before the missing data is used to fill the missing data, and a confidence downgrade mark is added to the warning judgment result containing data of that time period. For long-term data loss with an interruption time greater than or equal to the second duration, a data interruption state is triggered and automatic early warning judgment is suspended.

10. A tidal riverbank collapse early warning system based on the hierarchical evolution characteristics of soil stress response, characterized in that, include: At least one processor; as well as, A memory communicatively connected to at least one of the processors; wherein, The memory stores instructions executable by the processor to implement the method according to any one of claims 1 to 9.

Images