A landslide early warning method and system based on displacement-rainfall-geological structure multi-dimensional data fusion

Abstract

This invention provides a landslide early warning method and system based on multi-dimensional data fusion of displacement, rainfall, and geological structure. The method includes acquiring surface displacement, deep displacement, rainfall data, and geological structure parameters; using a dynamic displacement tangent angle recognition algorithm to extract the start time of the displacement acceleration phase and the current displacement tangent angle value, and calculating the effective cumulative rainfall and static safety factor; using the start time of the displacement acceleration phase as the dividing point, calculating the rainfall response lag time and sensitivity coefficient; fusing the above parameters to construct a feature vector; inputting the data into a landslide physical prediction model to output a dynamic safety factor sequence; calculating the time-varying instability probability and comprehensive early warning index; and comparing the results with a preset threshold to output the early warning level. This invention can achieve accurate identification and early warning of landslide disasters, improving the accuracy and timeliness of early warnings.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, and more specifically, to a landslide early warning method and system based on the fusion of multi-dimensional data of displacement, rainfall and geological structure. Background Technology

[0002] Landslides are a common type of geological hazard in nature, characterized by their sudden onset, destructive power, and high difficulty in prediction, seriously threatening people's lives and property and regional sustainable development. The formation and evolution of landslides are influenced by a combination of factors, with rainfall being the most significant external trigger, displacement changes being the most direct indicator of landslide instability, and geological structural parameters determining the intrinsic stability of landslides. With the development of monitoring technology, surface displacement monitoring, deep displacement monitoring, and rainfall monitoring have become the main means of landslide monitoring and early warning.

[0003] Currently, existing landslide early warning methods mainly include those based on displacement thresholds, those based on rainfall thresholds, and those based on stability calculations. Displacement threshold-based methods issue warnings by setting thresholds for displacement rate or cumulative displacement, but they fail to fully consider the dynamic influence of external triggering factors such as rainfall, and the threshold determination often relies on experience, making it difficult to adapt to the differentiated characteristics of landslides under different geological conditions. Rainfall threshold-based methods issue warnings by establishing a statistical relationship between rainfall and the probability of landslide occurrence, but they ignore the varying sensitivity of landslide displacement evolution stages to rainfall responses, resulting in significant lag in warning results. Stability calculation-based methods use limit equilibrium methods or finite element methods to calculate safety factors for early warning, but they typically use static geological parameters and fail to effectively integrate real-time monitoring data to reflect the dynamic evolution process of landslides. Summary of the Invention

[0004] This invention provides a landslide early warning method and system based on the fusion of multi-dimensional data on displacement, rainfall, and geological structure.

[0005] In a first aspect of the present invention, a landslide early warning method based on the fusion of multidimensional data of displacement, rainfall, and geological structure is provided, comprising: S1, acquire the surface displacement time series data, deep displacement time series data, rainfall time series data and geological structure parameter set of the target landslide body, the geological structure parameter set includes sliding surface dip angle, sliding surface internal friction angle, sliding surface cohesion and natural unit weight of the landslide body; S2, based on the surface displacement time series data and the deep displacement time series data, the displacement tangent angle dynamic identification algorithm is used to extract the starting time of the displacement acceleration stage and the current displacement tangent angle value of the target landslide body; based on the rainfall time series data, the effective cumulative rainfall is calculated using the previous effective rainfall accumulation model; based on the sliding surface dip angle, the sliding surface internal friction angle, the sliding surface cohesion, and the natural unit weight of the landslide body, the static safety factor is calculated using the infinite slope stability analysis method; S3, taking the start time of the displacement acceleration stage as the dividing point, input the current displacement tangent angle value and the effective cumulative rainfall into the displacement-rainfall response lag function to calculate the rainfall response lag time and rainfall response sensitivity coefficient of the target landslide body; S4, the rainfall response lag time, the rainfall response sensitivity coefficient, the current displacement tangent angle, the effective cumulative rainfall and the static safety factor are fused in multiple dimensions to construct a fused feature vector; S5, the fused feature vector is input into the landslide physical prediction model based on the dynamic safety factor evolution, and the landslide physical prediction model outputs the dynamic safety factor sequence of the target landslide body in the future prediction period; S6. Based on the dynamic safety factor sequence, the time-varying instability probability of the target landslide body is calculated using a reliability analysis method. S7. Based on the time-varying instability probability and the current displacement tangent angle, a comprehensive early warning index is calculated using a two-parameter early warning decision rule. S8. Compare the comprehensive early warning index with multiple preset early warning thresholds and output the corresponding early warning level.

[0006] Furthermore, the dynamic identification algorithm for the displacement tangent angle includes the following steps: S21, Perform a moving average filter on the time series data of surface displacement to obtain a filtered displacement sequence; S22, For each time point in the filtered displacement sequence, calculate the displacement increment of that time point relative to the previous time point, and calculate the time increment of that time point relative to the previous time point. Divide the displacement increment by the time increment to obtain the displacement rate at that time point. S23, differentiate the displacement rate again to obtain the displacement acceleration sequence; S24. Based on the deep displacement time series data, extract the displacement curve at the depth where the maximum deep displacement point is located, perform second-order difference calculation on the displacement curve, and obtain the time when the maximum deep displacement curvature occurs. S25, the moment when the displacement acceleration sequence first exceeds the preset acceleration threshold and the moment when the maximum value of the deep displacement curvature occurs are weighted and averaged, and the weighted average moment is taken as the starting moment of the displacement acceleration phase. S26, calculate the ratio of the cumulative displacement increment to the time increment at the current time point relative to the start time of the displacement acceleration phase in the filtered displacement sequence, and take the arctangent value of the ratio to obtain the current displacement tangent angle value.

[0007] Furthermore, the calculation of effective cumulative rainfall using the previous effective rainfall accumulation model includes: S27a, obtain the hourly rainfall for each day in the consecutive L days before the current time, where L is the preset number of days affected by previous rainfall; S27b, For the hourly rainfall of each day, based on the number of days that day is lag-off from the current time, the hourly rainfall of that day is multiplied by an attenuation coefficient, wherein the attenuation coefficient decreases exponentially with the increase of the number of days lag-off, and the attenuation coefficient is between 0 and 1. S27c, sum up all the hourly rainfall amounts for each day after multiplying by the attenuation coefficient to obtain the effective cumulative rainfall amount at the current moment; The calculation of the static safety factor using the infinite slope stability analysis method includes: S28a, Calculate the sliding force component of the sliding body along the sliding surface direction based on the natural weight of the sliding body, gravitational acceleration, thickness of the sliding body, and inclination angle of the sliding surface; S28b, Calculate the effective normal stress of the sliding body on the sliding surface based on the natural weight of the sliding body, gravitational acceleration, thickness of the sliding body, inclination angle of the sliding surface, and pore water pressure at the sliding surface; S28c, Calculate the anti-slip force of the sliding body along the sliding surface based on the effective normal stress, the internal friction angle of the sliding surface, and the cohesion of the sliding surface; S28d, the anti-slip force is divided by the sliding force component to obtain the static safety factor.

[0008] Furthermore, the displacement-rainfall response hysteresis function calculates the rainfall response hysteresis time and the rainfall response sensitivity coefficient through the following steps: S29a, obtain the effective cumulative rainfall at multiple historical moments before the start time of the displacement acceleration phase, and form a historical sequence of effective cumulative rainfall; S29b, Obtain the current displacement tangent angle value from the start time of the displacement acceleration phase to the current time, and form a displacement tangent angle value sequence; S29c, using the historical sequence of effective cumulative rainfall as the first input sequence and the sequence of displacement tangent angles as the second input sequence, perform cross-correlation operation on the first input sequence and the second input sequence to obtain the curve of cross-correlation function value changing with lag time; S29d, extract the lag time corresponding to the maximum value of the cross-correlation function from the curve, and use this lag time as the rainfall response lag time; S29e, the maximum value in the displacement tangent angle value sequence is divided by the maximum value in the effective cumulative rainfall historical sequence, and the resulting ratio is used as the rainfall response sensitivity coefficient.

[0009] Furthermore, the construction process of the fused feature vector is as follows: the rainfall response lag time is taken as a scalar feature, the rainfall response sensitivity coefficient is taken as a scalar feature, the current displacement tangent angle value is taken as a scalar feature, the effective cumulative rainfall is taken as a scalar feature, and the static safety factor is taken as a scalar feature. The above five scalar features are arranged in a fixed order to form a five-dimensional vector, which is the fused feature vector.

[0010] Furthermore, the landslide physical prediction model based on dynamic safety factor evolution employs an improved Sarma method, specifically including the following steps: S31a, the target landslide body is divided into N vertical strips, where N is an integer greater than or equal to 3; S31b, determine the initial security coefficient based on the static security coefficient in the fused feature vector; S31c, Calculate the pore water pressure increment sequence for each block during the future prediction period based on the rainfall response lag time and the rainfall response sensitivity coefficient; S31d, calculate the displacement coordination conditions at the bottom of each block based on the current displacement tangent angle value; S31e, Substitute the pore water pressure increment sequence and the displacement compatibility condition into the block force balance equation of the Sarma method, iteratively solve the horizontal force between blocks at each moment, and then calculate the overall safety factor at each moment to obtain the dynamic safety factor sequence.

[0011] Furthermore, the reliability analysis method employs the first-order second-moment method, specifically including the following steps: S32a, extract the minimum value from the dynamic safety factor sequence as the minimum dynamic safety factor; S32b, obtain the coefficient of variation of the internal friction angle of the sliding surface and the coefficient of variation of the cohesion of the sliding surface; S32c, calculate the standard deviation of the dynamic safety factor sequence based on the minimum dynamic safety factor, the coefficient of variation of the internal friction angle of the sliding surface, the internal friction angle of the sliding surface, the coefficient of variation of the cohesion of the sliding surface, and the cohesion of the sliding surface; S32d, calculate the standardized difference based on the minimum dynamic safety factor and the standard deviation of the dynamic safety factor sequence, input the standardized difference into the cumulative distribution function of the standard normal distribution, and use the output value of the cumulative distribution function as the time-varying instability probability.

[0012] Furthermore, the dual-parameter early warning decision rule calculates the comprehensive early warning index through the following steps: S33a, obtain the preset maximum reference probability, divide the time-varying instability probability by the maximum reference probability to obtain a normalized probability value, and if the time-varying instability probability is greater than the maximum reference probability, then set the normalized probability value to 1. S33b, obtain the preset maximum reference tangent angle, divide the current displacement tangent angle value by the maximum reference tangent angle to obtain the normalized tangent angle value, and if the current displacement tangent angle value is greater than the maximum reference tangent angle, then set the normalized tangent angle value to 1; S33c, obtain the preset probability weight and tangent angle weight, wherein the sum of the probability weight and the tangent angle weight is 1; S33d, multiply the normalized probability value by the probability weight to obtain a first product; multiply the normalized tangent angle value by the tangent angle weight to obtain a second product; add the first product and the second product to obtain the initial comprehensive early warning index; S33e, obtain the static safety factor; S33f, when the static safety factor is greater than the first static threshold, the initial comprehensive early warning index is multiplied by a preset first reduction factor to obtain the comprehensive early warning index; S33g, when the static safety factor is less than the second static threshold, the initial comprehensive early warning index is multiplied by a preset first amplification factor to obtain the comprehensive early warning index; S33h, when the static safety coefficient is between the second static threshold and the first static threshold, the initial comprehensive early warning index is directly used as the comprehensive early warning index.

[0013] Furthermore, the preset multiple warning thresholds include a blue threshold, a yellow threshold, and a red threshold, wherein the blue threshold is less than the yellow threshold, and the yellow threshold is less than the red threshold; The specific process for outputting the corresponding warning level is as follows: S34a, When the comprehensive early warning index is less than the blue threshold, output a blue early warning level; S34b, When the comprehensive warning index is greater than or equal to the blue threshold and less than the yellow threshold, output a yellow warning level; S34c, When the comprehensive warning index is greater than or equal to the yellow threshold and less than the red threshold, an orange warning level is output; S34d, when the comprehensive early warning index is greater than or equal to the red threshold, output the red early warning level; S34e, generate an early warning message containing the start time of the displacement acceleration phase and the rainfall response lag time according to the output early warning level, and send the early warning message to the preset monitoring center server.

[0014] In a second aspect of the invention, a landslide early warning system based on the fusion of displacement-rainfall-geological structure multidimensional data is provided, comprising: The data acquisition module is used to acquire time-series data of surface displacement, deep displacement, rainfall, and geological structure parameters of the target landslide body. The geological structure parameter set includes sliding surface dip angle, sliding surface internal friction angle, sliding surface cohesion, and natural unit weight of the landslide body. The feature extraction module is used to extract the starting time of the displacement acceleration stage and the current displacement tangent angle value based on the surface displacement time series data and the deep displacement time series data using a displacement tangent angle dynamic recognition algorithm; to calculate the effective cumulative rainfall based on the rainfall time series data using an early effective rainfall accumulation model; and to calculate the static safety factor based on the geological structure parameter set using an infinite slope stability analysis method. The hysteresis response calculation module is used to take the start time of the displacement acceleration stage as the dividing point, input the current displacement tangent angle value and the effective cumulative rainfall into the displacement-rainfall response hysteresis function, and calculate the rainfall response hysteresis time and rainfall response sensitivity coefficient. The feature fusion module is used to perform multi-dimensional feature fusion of the rainfall response lag time, the rainfall response sensitivity coefficient, the current displacement tangent angle, the effective cumulative rainfall, and the static safety factor to construct a fused feature vector. The dynamic safety factor prediction module is used to input the fused feature vector into the landslide physical prediction model based on the evolution of dynamic safety factor, and output the dynamic safety factor sequence for the future prediction period. The instability probability calculation module is used to calculate the time-varying instability probability based on the dynamic safety factor sequence using a reliability analysis method. The comprehensive early warning index calculation module is used to calculate the comprehensive early warning index based on the time-varying instability probability and the current displacement tangent angle value using a two-parameter early warning decision rule. The warning level output module is used to compare the comprehensive warning index with multiple preset warning thresholds and output the corresponding warning level.

[0015] The embodiments of the present invention have at least the following beneficial effects: 1. This invention accurately extracts the starting moment of the displacement acceleration stage through a dynamic identification algorithm of displacement tangent angle, and quantitatively calculates the rainfall response lag time and sensitivity coefficient by combining the displacement-rainfall response lag function. This effectively solves the problems of existing technologies being unable to identify the key time nodes of the landslide transition from the creep stage to the accelerated failure stage and the difficulty in quantifying the lag characteristics of rainfall-induced effects. This enables the early warning model to accurately capture the critical state in the landslide evolution process, significantly improving the timeliness and pertinence of early warning for rainfall-induced landslides.

[0016] 2. This invention integrates multi-dimensional features such as rainfall response lag time, rainfall response sensitivity coefficient, current displacement tangent angle, effective cumulative rainfall, and static safety factor, and inputs them into a landslide physical prediction model based on dynamic safety factor evolution. This effectively solves the problems of existing technologies that mostly use static geological parameters and single monitoring indicators and fail to establish an inherent correlation mechanism for multi-source heterogeneous data. It achieves accurate prediction of the dynamic safety factor sequence of landslides in the future prediction period, overcomes the shortcomings of traditional methods that cannot reflect the time-varying risk characteristics of landslides, and greatly improves the scientificity and reliability of early warning results.

[0017] 3. This invention calculates the time-varying instability probability using a reliability analysis method and combines the time-varying instability probability with the current displacement tangent angle value to calculate a comprehensive early warning index using a dual-parameter early warning decision rule. This effectively solves the problems of existing technologies having a single early warning decision rule, lacking a multi-parameter coupling mechanism, and being prone to missed or false alarms. At the same time, the invention further improves the rationality of the early warning level classification through a static safety factor grade correction mechanism, realizing dynamic quantitative assessment and graded early warning of landslide disaster risk, and significantly reducing the false alarm rate and missed alarm rate of the early warning system. Attached Figure Description

[0018] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein: Figure 1 This is a flowchart illustrating a landslide early warning method based on the fusion of multi-dimensional data of displacement, rainfall, and geological structure, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of a landslide early warning system based on the fusion of multi-dimensional data of displacement, rainfall, and geological structure, provided in an embodiment of the present invention. Detailed Implementation

[0019] The principles and spirit of the invention will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are provided merely to enable those skilled in the art to better understand and implement the invention, and are not intended to limit the scope of the invention in any way. Rather, these embodiments are provided to make the invention more thorough and complete, and to fully convey the scope of the invention to those skilled in the art.

[0020] It should be noted that this application is applicable to scenarios requiring real-time monitoring of landslide deformation and early warning, such as along mountain roads, reservoir areas, mine slopes, and areas prone to geological disasters. The technical solution of this application will be described in detail below with reference to specific embodiments.

[0021] A target landslide refers to a slope body with an identified potential sliding surface or that is undergoing slow deformation. Surface displacement time-series data refers to the sequence of horizontal and vertical displacements of the landslide surface over time, obtained from GNSS monitoring stations, measured in millimeters. Deep displacement time-series data refers to the sequence of horizontal displacements at different depths within the landslide body over time, obtained from borehole inclinometers, measured in millimeters. Rainfall time-series data refers to the sequence of hourly rainfall over time, obtained from rain gauges, measured in millimeters. The sliding surface dip angle is the angle between the sliding surface and the horizontal plane.

[0022] The internal friction angle and cohesion of the sliding surface are shear strength parameters of soil and rock masses. The internal friction angle is measured in degrees, and the cohesion is measured in kilopascals (kPa). The natural unit weight of the landslide mass is the weight per unit volume, measured in kilonewtons per cubic meter. Pore water pressure is the pressure exerted on the pore water at the sliding surface, measured in kilopascals (kPa). The displacement tangent angle is the angle between the tangent at a point on the cumulative displacement-time curve and the time axis, measured in degrees, and is used to characterize the deformation stage of a landslide. Effective cumulative rainfall is a comprehensive rainfall index obtained by exponentially decaying weighted summation of historical rainfall, measured in millimeters.

[0023] The static safety factor refers to the ratio of the anti-sliding force to the sliding force along the sliding surface of a landslide body, without considering dynamic changes. Rainfall response lag time refers to the time difference between the occurrence of rainfall and the start of significant displacement acceleration. The rainfall response sensitivity coefficient refers to the increment of the displacement tangent angle caused by a unit effective cumulative rainfall. The dynamic safety factor refers to the safety factor that varies with time, considering rainfall infiltration and displacement evolution. The time-varying instability probability refers to the probability of landslide instability when considering the variability of soil and rock parameters and the dynamic changes in the safety factor. The comprehensive early warning index is a comprehensive early warning indicator obtained by fusing the instability probability and the displacement tangent angle.

[0024] The following is for reference. Figure 1 , Figure 1This is a flowchart illustrating a landslide early warning method based on multi-dimensional data fusion of displacement, rainfall, and geological structure, as provided in an embodiment of the present invention. Figure 1 As shown, a landslide early warning method based on the fusion of displacement-rainfall-geological structure multidimensional data includes: S1, acquire the time series data of surface displacement, deep displacement, rainfall, and geological structure parameters of the target landslide body. The geological structure parameter set includes the sliding surface dip angle, sliding surface internal friction angle, sliding surface cohesion, and natural unit weight of the landslide body.

[0025] At least three GNSS monitoring stations were deployed on the surface of the target landslide. Each station used a BeiDou-compatible receiver with a sampling frequency of 1 Hz. Every hour, the average horizontal and vertical displacements of all sampling points from the previous hour were averaged to form a data point, which was recorded as follows: The corresponding timestamps were also recorded. A main sliding section was selected on the landslide body. Geological exploration boreholes were drilled to a depth of 5 meters below the sliding surface. Fixed borehole inclinometers were installed in the boreholes, with a tilting probe placed every 0.5 meters. Horizontal displacement at each depth was automatically collected every hour, forming a depth-displacement curve. The depth of the point with the maximum displacement on the curve was taken as the location of the sliding surface. Siphon-type automatic rain gauges were installed in an open area near the landslide body, recording the cumulative rainfall of the previous hour every hour. The sliding surface dip angle, determined by the concentration of geological structural parameters, was measured using a borehole television imaging system, and converted to the true dip angle by combining it with the borehole azimuth. The internal friction angle and cohesion of the sliding surface were obtained through direct shear tests in the field: undisturbed soil samples were taken near the sliding surface, and rapid shear tests were conducted using a portable shear tester. The shear strength under different normal stresses was recorded, and the friction angle and cohesion were obtained through linear fitting. The natural unit weight of the landslide body was obtained by sampling in the field using the ring cutter method, weighing the wet soil, and calculating the volume. All data is transmitted to the data center server in real time via a 4G wireless transmission module. The server stores the time-series data in a MySQL database table, with each record containing a timestamp, landslide number, and monitoring value.

[0026] S2, based on the surface displacement time series data and the deep displacement time series data, the displacement tangent angle dynamic identification algorithm is used to extract the starting time of the displacement acceleration stage and the current displacement tangent angle value of the target landslide body; based on the rainfall time series data, the effective cumulative rainfall is calculated using the previous effective rainfall accumulation model; based on the sliding surface dip angle, the sliding surface internal friction angle, the sliding surface cohesion, and the natural unit weight of the landslide body, the static safety factor is calculated using the infinite slope stability analysis method.

[0027] This step is executed on a computing server in the data center, running Ubuntu 20.04. It deploys a landslide early warning analysis program developed using Python 3.9, which utilizes the NumPy 1.21, SciPy 1.7, and Pandas 1.3 scientific computing libraries. The program employs an event-driven architecture, triggering the early warning calculation process whenever new monitoring data is written to the database.

[0028] The dynamic identification algorithm for displacement tangent angle first filters and differs the displacement data to identify the time point when the displacement acceleration first exceeds the threshold. Simultaneously, it combines this with the time of the maximum value of the deep displacement curvature, and uses a weighted average to obtain the starting time of the displacement acceleration phase, then calculates the displacement tangent angle value at the current time. The cumulative effective rainfall model reads the daily rainfall over the past 15 days, multiplies it by an exponential decay coefficient, and then sums them to obtain the effective cumulative rainfall. The infinite slope stability analysis method, based on the limit equilibrium principle, calculates the ratio of the sliding force to the anti-sliding force as the static safety factor.

[0029] S21, a moving average filter is applied to the surface displacement time series data to obtain a filtered displacement sequence. The Python program on the computing server calls the `signal.savgol_filter` function from the SciPy library, inputting the horizontal displacement component from the surface displacement time series data, setting the window length to 5, and the polynomial order to 2, to obtain the smoothed displacement sequence. This filter, based on local least squares fitting, can effectively suppress multipath effects and Gaussian noise in GNSS signals.

[0030] S22, for each time point in the filtered displacement sequence, calculate the displacement increment relative to the previous time point, and calculate the time increment relative to the previous time point. Divide the displacement increment by the time increment to obtain the displacement rate at that time point. The program uses the `diff` method of the Pandas library to calculate the first-order difference of the displacement sequence. For the i-th time point, the displacement increment is: , The time increment is: , The displacement rate is: .

[0031] Since the monitoring data sampling interval is 1 hour It is always 1, therefore Numerically equal to .

[0032] S23, differentiate the displacement rate again to obtain the displacement acceleration sequence. The program then calls the diff method again on the displacement rate sequence to calculate the acceleration. The time point at which a large positive value first appears in the acceleration sequence corresponds to the critical point at which the landslide transitions from uniform deformation to accelerated deformation.

[0033] S24, based on the deep displacement time series data, extract the displacement curve at the depth where the maximum deep displacement point is located, perform second-order difference calculation on the displacement curve to obtain the time when the maximum deep displacement curvature occurs. The program reads the horizontal displacement data at each depth measured by the borehole inclinometer, forms an array of displacement values ​​at each depth point, and uses numpy.argmax to find the depth with the maximum displacement. .

[0034] Then extract The displacement sequence at depth over time is used to calculate the first difference using the numpy.gradient function to obtain the displacement velocity curve. Then, the gradient of the velocity curve is calculated to obtain the displacement acceleration curve. The peak position of the acceleration curve is the moment of maximum curvature. .

[0035] S25, the moment when the displacement acceleration sequence first exceeds a preset acceleration threshold is weighted and averaged with the moment when the maximum deep displacement curvature occurs. The resulting weighted average moment is taken as the start moment of the displacement acceleration phase. The preset acceleration threshold is 0.01. The program iterates through the displacement acceleration sequence and finds the first moment greater than or equal to 0.01. If it does not exist Then take As the starting point. The weighted average formula is: .

[0036] The weights were determined based on statistics from on-site landslide cases. Changes in the curvature of deep displacement are more sensitive to damage to the sliding surface, and therefore are given higher weights.

[0037] S26, calculate the ratio of the cumulative displacement increment to the time increment at the current time point relative to the start time of the displacement acceleration phase in the filtered displacement sequence, and take the arctangent of this ratio to obtain the current displacement tangent angle value. Let the current time be... The cumulative displacement increment is: , The time increment is: ,ratio .

[0038] Call math.atan(k) to calculate the arctangent, get the radian value, and then multiply it by... Convert to degrees, get .

[0039] S27a retrieves the hourly rainfall for each of the L consecutive days preceding the current time, where L is the preset number of days affected by previous rainfall. In the program, L is set to 15. The server queries the MySQL database for the hourly rainfall records for each of the 15 days preceding the current time, with 24 data points per day, totaling 360 points.

[0040] S27b: For the hourly rainfall of each day, based on the number of days since the current time, the hourly rainfall of that day is multiplied by an attenuation coefficient. This attenuation coefficient decreases exponentially with increasing lag days and is between 0 and 1. The program first accumulates the hourly rainfall for each day to obtain the daily rainfall, denoted as... Where d is the lag days, d = 1, 2, ..., 15. Attenuation coefficient. =0.8. For each day, the weighted rainfall is calculated as follows: .

[0041] S27c, sum up all the hourly rainfall amounts for each day after multiplying by the attenuation coefficient to obtain the effective cumulative rainfall amount at the current moment. The calculation is as follows: .

[0042] S28a, based on the natural unit weight of the sliding body, gravitational acceleration, sliding body thickness, and sliding surface inclination angle, calculate the sliding force component along the sliding surface direction. Definitions: ρ is the natural unit weight of the sliding body, in kilonewtons per cubic meter; g is the gravitational acceleration, taken as 9.8 m / s²; h is the sliding body thickness, in meters, taken as the average thickness of the rock and soil mass above the sliding surface exposed by the borehole; α is the sliding surface inclination angle, in degrees. The sliding force component is: .

[0043] S28b, based on the natural weight of the sliding body, gravitational acceleration, thickness of the sliding body, inclination angle of the sliding surface, and pore water pressure at the sliding surface, calculate the effective normal stress of the sliding body on the sliding surface. Define u as the pore water pressure at the sliding surface, in kilopascals, measured by a pore water pressure gauge embedded in the borehole. The total normal stress is: Effective normal stress .

[0044] S28c, Calculate the anti-slip force of the sliding body along the sliding surface based on the normal effective stress, the internal friction angle of the sliding surface, and the cohesion of the sliding surface. Define φ as the internal friction angle of the sliding surface, in degrees; and c as the cohesion of the sliding surface, in kilopascals. The anti-slip force is: .

[0045] S28d, the anti-slip force is divided by the sliding force component to obtain the static safety factor. Static safety factor .

[0046] S3, taking the start time of the displacement acceleration stage as the dividing point, input the current displacement tangent angle value and the effective cumulative rainfall into the displacement-rainfall response hysteresis function to calculate the rainfall response hysteresis time and rainfall response sensitivity coefficient of the target landslide body.

[0047] The core of this step is to establish the response relationship between rainfall and displacement. The program starts at the beginning of the displacement acceleration phase. As the dividing point, it is considered Previously, it was mainly affected by accumulated rainfall. The displacement then begins to respond. The optimal lag time is found through cross-correlation analysis.

[0048] S29a, obtain the effective cumulative rainfall from multiple historical moments prior to the start of the displacement acceleration phase, forming a historical sequence of effective cumulative rainfall. The program queries the database. The effective cumulative rainfall over the previous 720 hours was sampled hourly to obtain a sequence, as shown in the following formula; .

[0049] S29b, obtain the current displacement tangent angle value from the start time of the displacement acceleration phase to the current time, forming a displacement tangent angle value sequence. The program queries the database. From then until the present moment The displacement tangent angle value is sampled hourly to obtain a sequence, as shown in the following formula; .

[0050] S29c, using the historical effective cumulative rainfall sequence as the first input sequence and the displacement tangent angle sequence as the second input sequence, a cross-correlation operation is performed on the first and second input sequences to obtain the curve of the cross-correlation function value changing with lag time. The program calls the `signal.correlate` function from the SciPy library, with the parameter `mode='full'`, to calculate the cross-correlation between the two sequences. Output array: .

[0051] S29d, extract the lag time corresponding to the maximum value of the cross-correlation function from the curve, and use this lag time as the rainfall response lag time. Use numpy.argmax(C) to find the maximum value index and convert it to lag time. The unit is hours.

[0052] S29e, the maximum value in the displacement tangent angle value sequence is divided by the maximum value in the effective cumulative rainfall historical sequence, and the resulting ratio is used as the rainfall response sensitivity coefficient.

[0053] S4, the rainfall response lag time, the rainfall response sensitivity coefficient, the current displacement tangent angle, the effective cumulative rainfall and the static safety factor are fused into a multi-dimensional feature vector.

[0054] During implementation, create a NumPy array of length 5 and store the data in order. To eliminate the influence of dimensions, the first four features are normalized using Min-Max: . Keep the original values. The normalized fused feature vector is denoted as... .

[0055] S5, the fused feature vector is input into the landslide physical prediction model based on the dynamic safety factor evolution, and the landslide physical prediction model outputs the dynamic safety factor sequence of the target landslide body in the future prediction period.

[0056] This step uses a modified Sarma method as the physical prediction model. The Sarma method is a commonly used slope stability analysis method in geotechnical engineering and is applicable to sliding surfaces of arbitrary shapes. The program preloads the geometric model of the landslide body, which is obtained by digitizing the geological profile.

[0057] S31a, the target landslide body is divided into N vertical strips, where N is an integer greater than or equal to 3. The program reads the geometric data of the landslide body, including the slope line, slip surface line, and groundwater level, and automatically divides the landslide into strips according to a preset strip width of 5 meters. If the horizontal projection length of the landslide body is L, then N = ceil(L / 5). The strips are numbered 1, 2, ..., N from left to right.

[0058] S31b, determine the initial security coefficient based on the static security coefficient in the fused feature vector. The program extracts F[4], i.e. As the initial safety factor .

[0059] S31c, Based on the rainfall response lag time and the rainfall response sensitivity coefficient, calculate the pore water pressure increment sequence for each block during the future forecast period. The program first sets a rainfall scenario for the next 48 hours, for example, using hourly rainfall from weather forecasts. For each block i, calculate the pore water pressure increment at time t: ,in This is the location coefficient for the water catchment area and infiltration capacity of the water strip. The program uses a one-dimensional seepage model to account for the rainfall response lag time. As a seepage delay parameter, the pore water pressure increment sequence of each block in the next 48 hours is obtained through convolution operation, with a length of 48.

[0060] S31d, based on the current displacement tangent angle value, calculate the displacement compatibility conditions at the bottom of each block. Convert θ to the shear strain rate at the bottom of the block: ,in The time step is 1 hour. The displacement compatibility condition is expressed as the difference in horizontal displacement between adjacent blocks equal to the block width multiplied by the shear strain.

[0061] S31e, the pore water pressure increment sequence and the displacement compatibility condition are substituted into the strip force balance equation of the Sarma method, and the horizontal forces between strips at each time step are iteratively solved to calculate the overall safety factor at each time step, thereby obtaining the dynamic safety factor sequence. The strip force balance equation of the Sarma method includes horizontal force balance, vertical force balance, and moment balance, totaling 3N equations.

[0062] In implementation, the `optimize.root` function from the SciPy library is used, employing the Newton-Raphson iterative method to solve the problem, with a convergence threshold set to... After obtaining the inter-block forces at each moment, the overall safety factor is calculated: ,in This is the sum of the anti-slip moments of all the blocks. This is the sum of the sliding moments of all blocks.

[0063] Calculations were performed for t=1, 2, ..., 48 hours respectively, and the results were obtained. .

[0064] S6. Based on the dynamic safety factor sequence, the time-varying instability probability of the target landslide body is calculated using a reliability analysis method.

[0065] S32a, extract the minimum value from the dynamic safety factor sequence as the minimum dynamic safety factor. The program uses... get .

[0066] S32b, obtain the coefficient of variation of the internal friction angle of the sliding surface and the coefficient of variation of the cohesion of the sliding surface. The program reads from the configuration file. and Based on extensive field test statistics, The value range is 0.1 to 0.2. The value range is 0.15 to 0.3.

[0067] S32c, calculate the standard deviation of the dynamic safety factor sequence based on the minimum dynamic safety factor, the coefficient of variation of the sliding surface internal friction angle, the sliding surface internal friction angle, the coefficient of variation of the sliding surface cohesion, and the sliding surface cohesion. The calculation formula is: .

[0068] in, This represents the standard deviation of the dynamic safety factor sequence.

[0069] S32d, based on the minimum dynamic safety factor and the standard deviation of the dynamic safety factor sequence, calculate the standardized difference, input the standardized difference into the cumulative distribution function of the standard normal distribution, and use the output value of the cumulative distribution function as the time-varying instability probability. The standardized difference is: .

[0070] Calling the SciPy library Calculate the standard normal cumulative distribution function value to obtain the instability probability. .

[0071] S7. Based on the time-varying instability probability and the current displacement tangent angle, a comprehensive early warning index is calculated using a two-parameter early warning decision rule.

[0072] S33a: Obtain the preset maximum reference probability; divide the time-varying instability probability by the maximum reference probability to obtain a normalized probability value; if the time-varying instability probability is greater than the maximum reference probability, then set the normalized probability value to 1. Maximum reference probability , .

[0073] S33b: Obtain the preset maximum reference tangent angle; divide the current displacement tangent angle value by the maximum reference tangent angle to obtain a normalized tangent angle value; if the current displacement tangent angle value is greater than the maximum reference tangent angle, then set the normalized tangent angle value to 1. Maximum reference tangent angle .

[0074] S33c, obtain the preset probability weight and tangent angle weight, wherein the sum of the probability weight and the tangent angle weight is 1. .

[0075] S33d, multiply the normalized probability value by the probability weight to obtain a first product; multiply the normalized tangent angle value by the tangent angle weight to obtain a second product; add the first product and the second product to obtain the initial comprehensive early warning index. .

[0076] S33e, obtain the static safety factor. The program extracts F[4], i.e. .

[0077] S33f, when the static safety factor is greater than the first static threshold, the initial comprehensive early warning index is multiplied by a preset first reduction factor to obtain the comprehensive early warning index. The first static threshold is 1.2, and the first reduction factor is 0.5. If... If E > 1.2, then E = ×0.5.

[0078] S33g, when the static safety factor is less than the second static threshold, the initial comprehensive early warning index is multiplied by a preset first amplification factor to obtain the comprehensive early warning index. The second static threshold is 0.8, and the first amplification factor is 1.5. If If E = <0.8, then E = ×1.5.

[0079] S33h, when the static safety factor is between the second static threshold and the first static threshold, the initial comprehensive early warning index is directly used as the comprehensive early warning index. That is, if 0.8 ≤ If ≤1.2, then E= .

[0080] S8. Compare the comprehensive early warning index with multiple preset early warning thresholds and output the corresponding early warning level.

[0081] S34a, When the comprehensive early warning index is less than the blue threshold, output a blue early warning level. Blue threshold =0.3. If E<0.3, output a blue warning.

[0082] S34b: When the comprehensive warning index is greater than or equal to the blue threshold and less than the yellow threshold, output a yellow warning level. Yellow threshold =0.6. If 0.3≤E<0.6, output a yellow warning.

[0083] S34c, When the comprehensive warning index is greater than or equal to the yellow threshold and less than the red threshold, an orange warning level is output. (Red threshold) =0.8. If 0.6≤E<0.8, output an orange warning.

[0084] S34d: When the comprehensive early warning index is greater than or equal to the red threshold, output a red early warning level. If E ≥ 0.8, output a red early warning.

[0085] S34e: Generate an early warning message containing the start time of the displacement acceleration phase and the rainfall response lag time based on the output early warning level, and send the early warning message to a preset monitoring center server. In implementation, this can be achieved by constructing a JSON message, for example: {"landslide_id":"LS-001","warning_level":"orange","start_time":"2025-03-15T14:00:00Z","lag_time":48,"warning_index":0.72,"timestamp":"2025-03-20T10:00:00Z"}.

[0086] The requests library is used to send an HTTP POST request to the early warning interface of the monitoring center server, while simultaneously calling the SMS gateway and email server to send notifications.

[0087] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

[0088] like Figure 2 As shown in some embodiments, a landslide early warning system based on multi-dimensional data fusion of displacement, rainfall, and geological structure is provided. The system includes: Data acquisition module 201 is used to acquire time series data of surface displacement, time series data of deep displacement, time series data of rainfall, and a set of geological structure parameters of the target landslide body. The set of geological structure parameters includes sliding surface dip angle, sliding surface internal friction angle, sliding surface cohesion, and natural unit weight of the landslide body. Feature extraction module 202 is used to extract the starting time of the displacement acceleration stage and the current displacement tangent angle value based on the surface displacement time series data and the deep displacement time series data using a displacement tangent angle dynamic recognition algorithm; calculate the effective cumulative rainfall based on the rainfall time series data using an early effective rainfall accumulation model; and calculate the static safety factor based on the geological structure parameter set using an infinite slope stability analysis method. The hysteresis response calculation module 203 is used to take the start time of the displacement acceleration stage as the dividing point, input the current displacement tangent angle value and the effective cumulative rainfall into the displacement-rainfall response hysteresis function, and calculate the rainfall response hysteresis time and rainfall response sensitivity coefficient. Feature fusion module 204 is used to perform multi-dimensional feature fusion of the rainfall response lag time, the rainfall response sensitivity coefficient, the current displacement tangent angle, the effective cumulative rainfall and the static safety factor to construct a fused feature vector; The dynamic safety factor prediction module 205 is used to input the fused feature vector into the landslide physical prediction model based on the evolution of dynamic safety factor, and output the dynamic safety factor sequence in the future prediction period. The instability probability calculation module 206 is used to calculate the time-varying instability probability based on the dynamic safety factor sequence using a reliability analysis method. The comprehensive early warning index calculation module 207 is used to calculate the comprehensive early warning index based on the time-varying instability probability and the current displacement tangent angle value using a two-parameter early warning decision rule. The warning level output module 208 is used to compare the comprehensive warning index with multiple preset warning thresholds and output the corresponding warning level.

[0089] It is understandable that the modules recorded in this landslide early warning system based on the fusion of displacement-rainfall-geological structure multidimensional data are similar to those in the reference system. Figure 1 The steps described correspond to those in the landslide early warning method based on the fusion of displacement-rainfall-geological structure multidimensional data. Therefore, the operations, characteristics, and beneficial effects described above for the landslide early warning method based on the fusion of displacement-rainfall-geological structure multidimensional data are also applicable to the landslide early warning system based on the fusion of displacement-rainfall-geological structure multidimensional data and its constituent modules, and will not be repeated here.

[0090] Furthermore, the storage medium in the embodiments of this application stores program instructions capable of implementing all the above methods. These program instructions can be stored in the storage medium in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor 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, or terminal devices such as computers, servers, mobile phones, and tablets.

[0091] The above description is merely an explanation of some preferred embodiments of the present invention and the technical principles employed. Those skilled in the art should understand that the scope of the invention as described in the embodiments of the present invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. A landslide early warning method based on displacement-rainfall-geological structure multi-dimensional data fusion, characterized in that, Includes the following steps: S1, acquire the surface displacement time series data, deep displacement time series data, rainfall time series data and geological structure parameter set of the target landslide body, the geological structure parameter set includes sliding surface dip angle, sliding surface internal friction angle, sliding surface cohesion and natural unit weight of the landslide body; S2, based on the surface displacement time series data and the deep displacement time series data, the displacement tangent angle dynamic identification algorithm is used to extract the starting time of the displacement acceleration stage and the current displacement tangent angle value of the target landslide body; based on the rainfall time series data, the effective cumulative rainfall is calculated using the previous effective rainfall accumulation model; based on the sliding surface dip angle, the sliding surface internal friction angle, the sliding surface cohesion, and the natural unit weight of the landslide body, the static safety factor is calculated using the infinite slope stability analysis method; S3, taking the start time of the displacement acceleration stage as the dividing point, input the current displacement tangent angle value and the effective cumulative rainfall into the displacement-rainfall response lag function to calculate the rainfall response lag time and rainfall response sensitivity coefficient of the target landslide body; S4, the rainfall response lag time, the rainfall response sensitivity coefficient, the current displacement tangent angle, the effective cumulative rainfall and the static safety factor are fused in multiple dimensions to construct a fused feature vector; S5, the fused feature vector is input into the landslide physical prediction model based on the dynamic safety factor evolution, and the landslide physical prediction model outputs the dynamic safety factor sequence of the target landslide body in the future prediction period; S6. Based on the dynamic safety factor sequence, the time-varying instability probability of the target landslide body is calculated using a reliability analysis method. S7. Based on the time-varying instability probability and the current displacement tangent angle, a comprehensive early warning index is calculated using a two-parameter early warning decision rule. S8. Compare the comprehensive early warning index with multiple preset early warning thresholds and output the corresponding early warning level.

2. The method of claim 1, wherein, The dynamic identification algorithm for displacement tangent angle includes the following steps: S21, Perform a moving average filter on the time series data of surface displacement to obtain a filtered displacement sequence; S22, For each time point in the filtered displacement sequence, calculate the displacement increment of that time point relative to the previous time point, and calculate the time increment of that time point relative to the previous time point. Divide the displacement increment by the time increment to obtain the displacement rate at that time point. S23, differentiate the displacement rate again to obtain the displacement acceleration sequence; S24. Based on the deep displacement time series data, extract the displacement curve at the depth where the maximum deep displacement point is located, perform second-order difference calculation on the displacement curve, and obtain the time when the maximum deep displacement curvature occurs. S25, the moment when the displacement acceleration sequence first exceeds the preset acceleration threshold and the moment when the maximum value of the deep displacement curvature occurs are weighted and averaged, and the weighted average moment is taken as the starting moment of the displacement acceleration phase. S26, calculate the ratio of the cumulative displacement increment to the time increment at the current time point relative to the start time of the displacement acceleration phase in the filtered displacement sequence, and take the arctangent value of the ratio to obtain the current displacement tangent angle value.

3. The method of claim 1, wherein, The calculation of effective cumulative rainfall using the previous effective rainfall accumulation model includes: S27a, obtain the hourly rainfall for each day in the consecutive L days before the current time, where L is the preset number of days affected by previous rainfall; S27b, For the hourly rainfall of each day, based on the number of days that day is lag-off from the current time, the hourly rainfall of that day is multiplied by an attenuation coefficient, wherein the attenuation coefficient decreases exponentially with the increase of the number of days lag-off, and the attenuation coefficient is between 0 and 1. S27c, sum up all the hourly rainfall amounts for each day after multiplying by the attenuation coefficient to obtain the effective cumulative rainfall amount at the current moment; The calculation of the static safety factor using the infinite slope stability analysis method includes: S28a, Calculate the sliding force component of the sliding body along the sliding surface direction based on the natural weight of the sliding body, gravitational acceleration, thickness of the sliding body, and inclination angle of the sliding surface; S28b, Calculate the effective normal stress of the sliding body on the sliding surface based on the natural weight of the sliding body, gravitational acceleration, thickness of the sliding body, inclination angle of the sliding surface, and pore water pressure at the sliding surface; S28c, Calculate the anti-slip force of the sliding body along the sliding surface based on the effective normal stress, the internal friction angle of the sliding surface, and the cohesion of the sliding surface; S28d, the anti-slip force is divided by the sliding force component to obtain the static safety factor.

4. The method of claim 1, wherein, The displacement-rainfall response hysteresis function calculates the rainfall response hysteresis time and the rainfall response sensitivity coefficient through the following steps: S29a, obtain the effective cumulative rainfall at multiple historical moments before the start time of the displacement acceleration phase, and form a historical sequence of effective cumulative rainfall; S29b, Obtain the current displacement tangent angle value from the start time of the displacement acceleration phase to the current time, and form a displacement tangent angle value sequence; S29c, using the historical sequence of effective cumulative rainfall as the first input sequence and the sequence of displacement tangent angles as the second input sequence, perform cross-correlation operation on the first input sequence and the second input sequence to obtain the curve of cross-correlation function value changing with lag time; S29d, extract the lag time corresponding to the maximum value of the cross-correlation function from the curve, and use this lag time as the rainfall response lag time; S29e, the maximum value in the displacement tangent angle value sequence is divided by the maximum value in the effective cumulative rainfall historical sequence, and the resulting ratio is used as the rainfall response sensitivity coefficient.

5. The method of claim 1, wherein, The process of constructing the fused feature vector is as follows: the rainfall response lag time is taken as a scalar feature, the rainfall response sensitivity coefficient is taken as a scalar feature, the current displacement tangent angle is taken as a scalar feature, the effective cumulative rainfall is taken as a scalar feature, and the static safety factor is taken as a scalar feature. The above five scalar features are arranged in a fixed order to form a five-dimensional vector, which is the fused feature vector.

6. The method of claim 1, wherein, The landslide physical prediction model based on dynamic safety factor evolution employs an improved Sarma method, specifically including the following steps: S31a, the target landslide body is divided into N vertical strips, where N is an integer greater than or equal to 3; S31b, determine the initial security coefficient based on the static security coefficient in the fused feature vector; S31c, Calculate the pore water pressure increment sequence for each block during the future prediction period based on the rainfall response lag time and the rainfall response sensitivity coefficient; S31d, calculate the displacement coordination conditions at the bottom of each block based on the current displacement tangent angle value; S31e, Substitute the pore water pressure increment sequence and the displacement compatibility condition into the block force balance equation of the Sarma method, iteratively solve the horizontal force between blocks at each moment, and then calculate the overall safety factor at each moment to obtain the dynamic safety factor sequence.

7. The method of claim 6, wherein, The reliability analysis method employs the first-order second-moment method and specifically includes the following steps: S32a, extract the minimum value from the dynamic safety factor sequence as the minimum dynamic safety factor; S32b, obtain the coefficient of variation of the internal friction angle of the sliding surface and the coefficient of variation of the cohesion of the sliding surface; S32c, calculate the standard deviation of the dynamic safety factor sequence based on the minimum dynamic safety factor, the coefficient of variation of the internal friction angle of the sliding surface, the internal friction angle of the sliding surface, the coefficient of variation of the cohesion of the sliding surface, and the cohesion of the sliding surface; S32d, calculate the standardized difference based on the minimum dynamic safety factor and the standard deviation of the dynamic safety factor sequence, input the standardized difference into the cumulative distribution function of the standard normal distribution, and use the output value of the cumulative distribution function as the time-varying instability probability.

8. The method of claim 1, wherein, The dual-parameter early warning decision rule calculates the comprehensive early warning index through the following steps: S33a, obtain the preset maximum reference probability, divide the time-varying instability probability by the maximum reference probability to obtain a normalized probability value, and if the time-varying instability probability is greater than the maximum reference probability, then set the normalized probability value to 1. S33b, obtain the preset maximum reference tangent angle, divide the current displacement tangent angle value by the maximum reference tangent angle to obtain the normalized tangent angle value, and if the current displacement tangent angle value is greater than the maximum reference tangent angle, then set the normalized tangent angle value to 1; S33c, obtain the preset probability weight and tangent angle weight, wherein the sum of the probability weight and the tangent angle weight is 1; S33d, multiply the normalized probability value by the probability weight to obtain a first product; multiply the normalized tangent angle value by the tangent angle weight to obtain a second product; add the first product and the second product to obtain the initial comprehensive early warning index; S33e, obtain the static safety factor; S33f, when the static safety factor is greater than the first static threshold, the initial comprehensive early warning index is multiplied by a preset first reduction factor to obtain the comprehensive early warning index; S33g, when the static safety factor is less than the second static threshold, the initial comprehensive early warning index is multiplied by a preset first amplification factor to obtain the comprehensive early warning index; S33h, when the static safety coefficient is between the second static threshold and the first static threshold, the initial comprehensive early warning index is directly used as the comprehensive early warning index.

9. The method according to claim 1, characterized in that, The preset multiple warning thresholds include a blue threshold, a yellow threshold, and a red threshold, wherein the blue threshold is less than the yellow threshold, and the yellow threshold is less than the red threshold; The specific process for outputting the corresponding warning level is as follows: S34a, When the comprehensive early warning index is less than the blue threshold, output a blue early warning level; S34b, When the comprehensive warning index is greater than or equal to the blue threshold and less than the yellow threshold, output a yellow warning level; S34c, When the comprehensive warning index is greater than or equal to the yellow threshold and less than the red threshold, an orange warning level is output; S34d, when the comprehensive early warning index is greater than or equal to the red threshold, output the red early warning level; S34e, generate an early warning message containing the start time of the displacement acceleration phase and the rainfall response lag time according to the output early warning level, and send the early warning message to the preset monitoring center server.

10. A landslide early warning system based on the fusion of multi-dimensional data on displacement, rainfall, and geological structure, characterized in that, include: The data acquisition module is used to acquire time-series data of surface displacement, deep displacement, rainfall, and geological structure parameters of the target landslide body. The geological structure parameter set includes sliding surface dip angle, sliding surface internal friction angle, sliding surface cohesion, and natural unit weight of the landslide body. The feature extraction module is used to extract the starting time of the displacement acceleration stage and the current displacement tangent angle value based on the surface displacement time series data and the deep displacement time series data using a displacement tangent angle dynamic recognition algorithm; to calculate the effective cumulative rainfall based on the rainfall time series data using an early effective rainfall accumulation model; and to calculate the static safety factor based on the geological structure parameter set using an infinite slope stability analysis method. The hysteresis response calculation module is used to take the start time of the displacement acceleration stage as the dividing point, input the current displacement tangent angle value and the effective cumulative rainfall into the displacement-rainfall response hysteresis function, and calculate the rainfall response hysteresis time and rainfall response sensitivity coefficient. The feature fusion module is used to perform multi-dimensional feature fusion of the rainfall response lag time, the rainfall response sensitivity coefficient, the current displacement tangent angle, the effective cumulative rainfall, and the static safety factor to construct a fused feature vector. The dynamic safety factor prediction module is used to input the fused feature vector into the landslide physical prediction model based on the evolution of dynamic safety factor, and output the dynamic safety factor sequence for the future prediction period. The instability probability calculation module is used to calculate the time-varying instability probability based on the dynamic safety factor sequence using a reliability analysis method. The comprehensive early warning index calculation module is used to calculate the comprehensive early warning index based on the time-varying instability probability and the current displacement tangent angle value using a two-parameter early warning decision rule. The warning level output module is used to compare the comprehensive warning index with multiple preset warning thresholds and output the corresponding warning level.

Images