Dynamic landslide early warning method and device based on landslide physical model and data assimilation

Abstract

The invention provides a dynamic landslide early warning method and device based on a landslide physical model and data assimilation, belongs to the technical field of geological disaster monitoring and early warning, and can dynamically, more accurately and timely carry out landslide early warning on a monitoring area. According to the technical scheme, the data assimilation technology is adopted to correct the landslide physical model used for predicting the landslide safety coefficient and the machine learning model used for predicting the landslide displacement, so that the landslide safety coefficient threshold value and the landslide displacement rate ratio threshold value are dynamically adjusted; and carrying out multi-model fusion on the predicted landslide safety coefficient and landslide displacement of the landslide mass to generate a landslide early warning index of the monitoring area, and carrying out landslide early warning according to the landslide early warning index.

Description

Technical Field

[0001] The embodiments of the present application belong to the technical field of geological disaster monitoring and early warning, and relate to a dynamic landslide early warning method and device based on a landslide physical model and data assimilation. Background Art

[0002] The rock and soil mass will deform due to environmental changes. When the deformation reaches a certain level, a landslide may occur, which will cause huge losses to agricultural production and the safety of life and property, and even bring devastating disasters. Therefore, it is of great significance to conduct landslide early warning in advance.

[0003] The common landslide early warning method is to monitor or predict the landslide displacement or landslide safety factor of the landslide body. When the landslide displacement or landslide safety factor exceeds the corresponding threshold, the landslide early warning will be triggered. However, most of the index thresholds used for early warning (such as the threshold of landslide displacement) are static thresholds set based on experience. With the dynamic changes of the landslide body, it is impossible to accurately and real-time conduct landslide early warning based on static thresholds. Summary of the Invention

[0004] The present application proposes a dynamic landslide early warning method and device based on a landslide physical model and data assimilation. The data assimilation technology is used to correct the landslide physical model for predicting the landslide safety factor and the machine learning model (LSTM model) for predicting the landslide displacement, and then dynamically adjust the landslide safety factor threshold and the landslide displacement rate ratio threshold, and perform multi-model fusion on the predicted landslide safety factor and landslide displacement of the landslide body to generate a landslide early warning index for the monitoring area, and conduct landslide early warning according to the landslide early warning index.

[0005] The technical solution provided by the present application considers that the landslide body follows the geophysical laws, and combines the data-driven machine learning model to dynamically update the relevant thresholds, so as to calculate the comprehensive landslide early warning index of the monitoring area, and can dynamically, more accurately and timely conduct landslide early warning on the monitoring area.

[0006] In order to achieve the above object, the present application adopts the following technical solutions:

[0007] In a first aspect, the present application provides a dynamic landslide warning method based on a landslide physical model and data assimilation, including: performing data assimilation processing on the landslide physical model based on first landslide observation data to update the landslide physical model, where the landslide physical model is a three-dimensional stress-seepage coupling model for describing the interaction relationship between the stress field and the seepage field in the landslide body; the first landslide observation data includes the landslide displacement of the landslide body in the monitoring area, the inclination angle of the landslide body, and the landslide safety factor; and predicting a sequence of landslide safety factors based on the model parameters of the updated landslide physical model, and adjusting the landslide safety factor threshold using the sequence of landslide safety factors, where the sequence of landslide safety factors includes the landslide safety factors of the landslide body at multiple time points; further, performing data assimilation processing on the long short-term memory (LSTM) model based on second landslide observation data to update the LSTM model; the second landslide observation data includes the landslide displacement of the landslide body in the monitoring area and the inclination angle of the landslide body; and predicting a sequence of landslide displacements based on the updated LSTM model, and adjusting the landslide displacement rate ratio threshold using the sequence of landslide displacements; the input of the LSTM model is multi-source landslide data, and the output is the landslide displacement, where the multi-source landslide data includes the rainfall in the monitoring area, the landslide displacement of the landslide body, and the inclination angle of the landslide body; the sequence of landslide displacements includes the landslide displacements of multiple sampling points of the landslide body at multiple time points; finally, performing fusion processing on the current landslide safety factor of the landslide body and the current landslide displacement of the landslide body using the adjusted landslide displacement rate ratio threshold and the adjusted landslide safety factor threshold to generate a landslide warning index for the monitoring area; and performing landslide warning according to the landslide warning index; the current landslide safety factor is predicted based on the model parameters of the updated landslide physical model, and the current landslide displacement is predicted by the updated LSTM model.

[0008] In a possible implementation, the adjusted landslide safety factor threshold T new satisfies:

[0009]

[0010] T old represents the landslide safety factor threshold before adjustment, θ represents the threshold update factor, F S represents the landslide safety factor predicted according to the updated landslide physical model, λ represents the weight coefficient, μ w represents the mean value of N landslide safety factors within the sliding window in the sequence of landslide safety factors, σ w represents the standard deviation of N landslide safety factors, N is an integer greater than or equal to 2, and k represents the standard deviation sensitivity coefficient.

[0011] In a possible implementation, the landslide displacement rate ratio threshold is adjusted using the landslide displacement sequence, including: calculating the landslide displacement rate ratio based on the landslide displacement sequence; and determining the landslide deformation stage of the landslide body based on the landslide displacement rate ratio; the landslide deformation stage includes a constant velocity deformation stage, an initial acceleration deformation stage, a medium acceleration deformation stage, and an impending slide stage;

[0012] When the landslide body is in the constant velocity deformation stage, it is determined that the adjusted landslide displacement rate ratio threshold is: R threshold = R base (1 - γ); when the landslide body is in the initial acceleration deformation stage, it is determined that the adjusted landslide displacement rate ratio threshold is: When the landslide body is in the medium acceleration deformation stage, it is determined that the adjusted landslide displacement rate ratio threshold is: When the landslide body is in the impending slide stage, it is determined that the adjusted landslide displacement rate ratio threshold is: R threshold = R base (1 + γ); where, R threshold represents the adjusted landslide displacement rate ratio, R base represents the reference landslide displacement rate ratio, R represents the landslide displacement rate ratio, and γ represents the smoothing factor.

[0013] In a possible implementation, the landslide warning index ω of the monitoring area satisfies:

[0014]

[0015] R represents the currently predicted landslide displacement rate ratio of the landslide body, which is calculated based on the landslide displacement of the landslide body output by the updated LSTM model, F s represents the currently predicted landslide safety factor of the landslide body based on the updated landslide physical model, R threshold represents the landslide displacement rate ratio threshold, F Sthreshold represents the landslide safety factor threshold, and α and β respectively represent the weight coefficients of the landslide displacement rate ratio and the landslide safety factor.

[0016] In a possible implementation, the algorithm for data assimilation processing is the ensemble Kalman filter algorithm.

[0017] In a possible implementation, the landslide body is divided into n units, and the landslide safety factor F of the landslide body S satisfies:

[0018]

[0019] where, c i represents the cohesion of the i-th unit, η iDenote the normal stress of the i-th unit, ξ i Denote the pore water pressure of the i-th unit, φ i Denote the internal friction angle of the i-th unit, τ i Denote the shear stress of the i-th unit, A i Denote the acting area of the i-th unit; c i 、η i 、ξ i 、φ i 、τ i All are obtained based on the model parameters of the landslide physical model.

[0020] In one possible implementation, the landslide displacement includes horizontal landslide displacement and vertical landslide displacement.

[0021] In one possible implementation, the landslide displacement includes GNSS landslide displacement collected by a Global Navigation Satellite System (GNSS) device, and / or InSAR landslide displacement collected by a synthetic aperture radar satellite.

[0022] In a second aspect, the present application provides a landslide warning device, which includes corresponding functional units or modules, and each functional unit or module interacts with each other to implement the method described in the first aspect or any of its implementations.

[0023] In a third aspect, the present application provides a landslide warning device, including at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to execute the method described in the first aspect or any of its implementations as above.

[0024] In a fourth aspect, the present application provides a computer-readable storage medium, including computer program instructions, which when executed by a computer, cause the computer to execute the method described in the first aspect or any of its implementations as above.

[0025] In a fifth aspect, the present application provides a computer program product, including computer program instructions, which when run on a computer, cause the computer to execute the method described in the first aspect or any of its implementations as above.

[0026] The technical solution provided by this application introduces a physical model of the landslide body and a data-driven machine learning model, and uses data assimilation technology to update the model parameters, thereby dynamically updating the landslide safety factor threshold and the landslide displacement rate ratio threshold. The landslide safety factor threshold and the landslide displacement rate ratio threshold are used to fuse the predicted landslide safety factor and landslide displacement in the monitoring area, so as to calculate the comprehensive landslide warning index in the monitoring area, and perform landslide warning according to the landslide warning. This technical solution can dynamically, more accurately and timely perform landslide warning on the monitoring area. Description of the Drawings

[0027] Figure 1 is a schematic flow chart of a dynamic landslide warning method based on a landslide physical model and data assimilation provided by an embodiment of this application;

[0028] Figure 2 is a schematic structural diagram of a landslide warning device provided by an embodiment of this application. Detailed Embodiments

[0029] If there are terms such as "first" and "second" in the specification and claims of this application, they are used to distinguish different objects, rather than to describe a specific order of the objects.

[0030] The "and / or" in the embodiments of this application represents the relationship between objects. For example, A and / or B may represent the following three situations: A exists alone, B exists alone, and both A and B exist simultaneously.

[0031] In the embodiments of this application, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or more advantageous than other embodiments or design solutions. Exactly speaking, using words such as "exemplary" or "for example" is intended to present relevant concepts in a specific manner.

[0032] In the description of the embodiments of this application, unless otherwise specified, the meaning of "a plurality" refers to two or more.

[0033] First, some technical terms involved in the embodiments of this application are explained.

[0034] 1. Landslide Safety Factor

[0035] The landslide safety factor (denoted as F S ) is an important index to measure the stability of a landslide. The landslide safety factor can usually be calculated based on some physical parameters of the landslide body.

[0036] For example, the landslide safety factor can be the ratio of the anti-sliding force to the sliding force. Of course, the landslide safety factor can also be calculated based on the limit equilibrium method, numerical analysis methods (such as the finite element method, finite difference method, etc.) or other methods.

[0037] It should be understood that the larger the value of F S , the higher the stability of the landslide mass, and it is not easy to occur a landslide. Exemplarily, when F S > 1, it indicates that the landslide mass is in a stable state and will not occur a landslide; when F S = 1, it indicates that the landslide mass is in a limit equilibrium state and there is a possibility of a landslide; when F S < 1, it indicates that the landslide mass is in an unstable state and will occur a landslide.

[0038] 2. Landslide displacement rate ratio

[0039] The landslide displacement rate ratio (denoted as R) refers to the ratio between the landslide displacement rates of the landslide mass at different stages, different parts or under different influencing factors. Through the landslide displacement rate ratio, the development of the landslide can be judged, such as judging the landslide deformation stage (such as the initial deformation stage, constant velocity deformation stage, accelerating deformation stage and impending landslide stage) in which the landslide mass is located.

[0040] 3. Landslide physical model

[0041] The landslide physical model is a model that can reflect the physical mechanism, deformation law of the landslide and evaluate the stability of the landslide mass by simulating the key elements and processes of the landslide phenomenon.

[0042] Optionally, the landslide physical model can be a three-dimensional stress-seepage coupling model, and the three-dimensional stress-seepage coupling model is used to describe the interaction relationship between the stress field and the seepage field in the landslide mass. In the embodiments of the present application, based on the model parameters of the three-dimensional stress-seepage coupling model followed by the landslide mass, the landslide safety factor of the landslide mass can be calculated to be used for evaluating the stability of the landslide mass and for landslide early warning.

[0043] It can be understood that a three-dimensional stress-seepage coupling model that meets the requirements can be established according to the physical mechanics principle and related theories. For example, by combining the stress balance equation and Darcy's law, a coupled partial differential equation group is established as the three-dimensional stress-seepage coupling model). The specific model structure of the three-dimensional stress-seepage coupling model in the embodiments of the present application is not limited.

[0044] 4. Multi-source landslide data

[0045] Multi-source landslide data refers to different types of landslide data collected from different monitoring devices, and the multi-source landslide data includes the landslide displacement of the landslide mass, the rainfall in the monitoring area, the inclination angle of the landslide mass, etc.

[0046] Among them, the landslide displacement can be collected by a Global Navigation Satellite System (GNSS) receiver (referred to as GNSS landslide displacement), and / or by a Synthetic Aperture Radar (SAR) satellite (referred to as InSAR landslide displacement). For example, the GNSS receiver collects data once an hour, and the accuracy of the collected landslide displacement is at the millimeter level, ensuring high frequency and high accuracy of the data; the SAR satellite collects data once every six days, and the accuracy of the collected landslide displacement can be at the centimeter level or millimeter level, ensuring the continuity and integrity of the data.

[0047] Optionally, the landslide displacement of the landslide body includes horizontal landslide displacement and vertical landslide displacement.

[0048] The rainfall in the above monitoring area can be collected by a ground rain gauge installed at key positions in the landslide area. For example, the ground rain gauge collects data every 10 minutes, ensuring high frequency, high accuracy and real-time of the data. The rainfall in the monitoring area can also be obtained from meteorological satellite data, which is obtained through satellite remote sensing technology. The satellite has a wide coverage range and can provide regional rainfall information.

[0049] The inclination angle of the above landslide body can be collected by an inclination sensor installed at key positions on the landslide body. The inclination sensor measures the change in the inclination angle of the landslide body through an internal accelerometer and gyroscope. For example, the inclination sensor collects data once an hour to ensure high frequency and continuity of the data.

[0050] Optionally, after collecting multi-source landslide data, the multi-source landslide data can be preprocessed for subsequent use. The preprocessing includes at least one of the following: data cleaning, data standardization, data interpolation, and data synchronization.

[0051] Among them, the 3σ criterion in statistics can be used to clean the multi-source landslide data, identify and remove outliers in the multi-source landslide data to ensure the quality of the data. The processing principle of the 3σ criterion is: retain the data within the range of (μ - 3σ, μ + 3σ), and remove the data outside this range, where μ is the average value of the multi-source landslide data (such as landslide displacement), and σ is the standard deviation of the multi-source landslide data.

[0052] Standardize the data from different sources to make them have the same dimension and scale, eliminate the influence caused by differences in units and dimensions between different data sources, and make the data more unified.

[0053] Taking the landslide displacement as an example, the following formula is used to standardize the landslide displacement:

[0054]

[0055] where, u iis the i-th landslide displacement in the original landslide displacement, μ is the mean value of the landslide displacement, σ is the standard deviation of the landslide displacement, and u i ' is the landslide displacement after the normalization processing of u i .

[0056] For missing data, the linear interpolation method can be used for supplementation. By filling the blanks in the data, the integrity of the data is ensured, providing a more sufficient data basis for subsequent analysis and modeling. For example, the following formula can be adopted to perform linear interpolation using the data at adjacent time points (t1 and t2) to obtain the data at the target time point t:

[0057]

[0058] In the above formula, u(t) is the interpolated data, u(t1) and u(t2) are the data at adjacent time points, and t1 and t2 are the times of adjacent time points.

[0059] Data synchronization is to synchronize data with different time resolutions so that they have the same time step. For example, synchronize the GNSS landslide displacement collected every hour and the rainfall data collected every 10 minutes to the hourly time step.

[0060] Combined with the description of the background technology, with the dynamic changes of the landslide body, it is impossible to accurately and real-time conduct landslide early warning based on static thresholds (static thresholds rely on manual experience settings). Based on this, the embodiments of the present application provide a dynamic landslide early warning method and device based on a landslide physical model and data assimilation, which are particularly suitable for real-time monitoring and disaster prevention and control in complex geological environments (such as high mountains and remote areas). This technical solution introduces a geophysical model followed by the landslide body and a data-driven machine learning model, and uses data assimilation technology to update model parameters, thereby dynamically updating the landslide safety factor threshold and the landslide displacement rate ratio threshold, and using the landslide safety factor threshold and the landslide displacement rate ratio threshold to perform fusion processing on the predicted landslide safety factor and landslide displacement in the monitoring area to calculate the comprehensive landslide early warning index of the monitoring area, and conduct landslide early warning according to the landslide early warning index. This technical solution can dynamically and more accurately and timely conduct landslide early warning on the monitoring area.

[0061] The dynamic landslide early warning method based on a landslide physical model and data assimilation provided by the embodiments of the present application can be executed by an electronic device with processing functions. For example, the electronic device can be a computer, a server, etc. Taking the electronic device as a computer as an example, the hardware part of the computer can include: a processor, a memory, a network interface, a user interface, a communication bus, etc.

[0062] Among them, the processor is used to control the electronic device to execute relevant processing and computing tasks. The processor may include a central processing unit (CPU), a graphics processing unit (GPU), or other processors (such as a neural processing unit (NPU)). The processor can be single-core or multi-core. For example, the CPU of the electronic device is a multi-core CPU.

[0063] The memory is used to store computer instructions and related data. For example, the memory is used to store the computer instructions for executing the landslide warning method, and store data such as multi-source landslide data, landslide observation data, landslide physical models, machine learning models, landslide safety factors and thresholds, landslide displacements, and landslide displacement rate ratio thresholds.

[0064] The memory can be a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, or an optical memory, a magnetic disk storage medium, or any other magnetic storage device, or any other medium capable of storing program code or data that can be accessed by a computer. Optionally, the memory can be integrated within the processor, and the memory can also be independent of the processor.

[0065] The network interface is used for the computer to communicate with other devices or communication networks. The network interface can be a transceiver with sending and receiving functions. Optionally, the network interface can include a standard wired interface, a wireless interface (such as a WI-FI interface, a Bluetooth interface, a 5G interface). For example, when the processor triggers a landslide warning, the landslide warning notification can be sent to the user device through the network interface.

[0066] The communication bus is used to enable connection communication between different components. For example, the above-mentioned processor, memory, network interface, and user interface can be interconnected through the communication bus.

[0067] The user interface can include a display screen and an input unit (such as a keyboard). Optionally, the user interface can also include a standard wired interface and a wireless interface. For example, the display screen is used to visualize relevant result data, such as the landslide safety factor, landslide displacement, landslide warning indicators, and landslide warning notifications can be displayed.

[0068] Those skilled in the art can understand that the above computer may also include more or fewer components, or combine certain components, or have different component arrangements. The embodiments of the present application do not limit this.

[0069] The following will elaborate on the technical solutions provided in the embodiments of this application in conjunction with the accompanying drawings.

[0070] As Figure 1 shown, the dynamic landslide warning method based on the landslide physical model and data assimilation provided in the embodiments of this application includes steps S100 - S500.

[0071] S100. Perform data assimilation processing on the landslide physical model based on the first landslide observation data to update the landslide physical model.

[0072] The first landslide observation data includes the landslide displacement of the landslide body within the monitoring area, the inclination angle of the landslide body, and the landslide safety factor of the landslide body. The first landslide observation data is real landslide data. Among them, the landslide displacement of the landslide body and the inclination angle of the landslide body come from multi - source landslide data of the pre - collected monitoring area, and the landslide safety factor of the landslide body can be obtained based on geological surveys or existing geological databases.

[0073] It should be understood that in the embodiments of this application, the landslide displacement of the landslide body includes the landslide displacements of multiple sampling points of the landslide body at multiple time points (or multiple time steps), the inclination angle of the landslide body includes the inclination angles of the landslide body at multiple time points, and the landslide safety factor of the landslide body includes the landslide safety factors of the landslide body at multiple time points.

[0074] Optionally, the above - mentioned landslide physical model is a three - dimensional stress - seepage coupling model used to describe the interaction relationship between the stress field and the seepage field in the landslide body. In one implementation, a partial differential equation group can be constructed based on the stress equilibrium equation satisfied by the stress field of the landslide body and Darcy's law satisfied by the seepage field. This partial differential equation group is the three - dimensional stress - seepage coupling model. Regarding the stress equilibrium equation and Darcy's law, reference can be made to existing technical materials, and details are not described here.

[0075] Exemplarily, the three - dimensional stress - seepage coupling model can be the following partial differential equation group:

[0076]

[0077] where, u x represents the horizontal landslide displacement of the landslide body, u y represents the vertical landslide displacement of the landslide body, t represents time, η represents the normal stress of the landslide body, τ represents the shear stress of the landslide body, ρ represents the soil particle density of the landslide body, θ represents the inclination angle of the landslide body, q x represents the seepage flow velocity in the horizontal direction, q y represents the seepage flow velocity in the vertical direction.

[0078] It should be noted that in the above three-dimensional stress-seepage coupling model (Formula (1)), the normal stress η of the landslide body is related to the geological parameters of the landslide body, and η can be calculated based on the cohesion c and the internal friction angle φ; similarly, the shear stress τ of the landslide body can be calculated based on the cohesion c and the internal friction angle φ.

[0079] The seepage velocity q in the horizontal direction x and the seepage velocity q in the vertical direction y can be calculated based on the permeability coefficient K and the head height h of the landslide body. For example, wherein, the head height h can be obtained by measurement or calculated according to the pore water pressure ξ (such as ρ w is the density of water, g is the acceleration due to gravity, and z is the elevation).

[0080] In addition, the finite element method can be used to solve the above partial differential equations, and the parameter elastic model E is involved in the finite element method.

[0081] In summary of the logical relationships between the various quantities described above, the model parameters of the above exemplary landslide physical model (Formula (1)) include the elastic model E, the permeability coefficient K, the pore water pressure ξ, the cohesion c, the internal friction angle φ, and the soil particle density ρ.

[0082] It can be understood that the essence of using data assimilation technology to update the landslide physical model is to update the model parameters of the landslide physical model. In one implementation, the algorithm for data assimilation processing of the landslide physical model is the Ensemble Kalman Filter algorithm (EnKF algorithm). The process of updating the model parameters of the landslide physical model using the Ensemble Kalman Filter algorithm is briefly described below.

[0083] The variables involved in the Ensemble Kalman Filter algorithm include the state vector e, the observation data (i.e., the above-mentioned first landslide observation data), the state covariance matrix P, the observation noise covariance matrix R, the process noise covariance matrix Q, and the observation matrix H. During the execution of the Ensemble Kalman Filter algorithm, the above state vector e and state covariance matrix P need to be initialized and updated during the execution of the algorithm.

[0084] Among them, the state vector e is set as: e = [u x , u y , θ, F s , E, K, ξ, c, φ, ρ], and the model parameters of the landslide physical model are included in the state vector e. It can be seen that updating the state vector e can obtain the model parameters of the landslide physical model. Optionally, the initial landslide observation data (including u x , u y , θ, F s) and the collected geological parameters and terrain data are used to initialize the state vector e. The geological parameters and terrain data (E, K, ξ, c, φ, ρ) can be obtained through geological surveys and laboratory tests.

[0085] The state covariance matrix P is initialized as:

[0086]

[0087] where, is the standard deviation of the uncertainty of the horizontal landslide displacement in the observed data, is the standard deviation of the uncertainty of the vertical landslide displacement in the observed data, σ θ is the standard deviation of the uncertainty of the inclination angle of the landslide body in the observed data, is the standard deviation of the uncertainty of the landslide safety factor of the landslide body in the observed data.

[0088] The observation noise covariance matrix R is derived from the measurement errors of the observation equipment and is used to describe the noise level in the observed data. The observation noise covariance matrix R can be set as:

[0089]

[0090] It can be seen that the observation noise covariance matrix R can be the same as the initialized state covariance matrix P. Subsequently, the observation noise covariance matrix R can be adjusted according to the actual situation.

[0091] The process noise covariance matrix Q is derived from the uncertainties within the model, such as parameter estimation errors, environmental changes, etc., and is used to describe the uncertainties of the system model. The process noise covariance matrix Q can be set as:

[0092]

[0093] It can be seen that the process noise covariance matrix Q can be the same as the initialized state covariance matrix P. Subsequently, the process noise covariance matrix Q can be adjusted according to the changes in the system state. For example, when the rainfall increases, the uncertainty of the displacement may increase, so the process noise covariance matrix Q needs to be adjusted. Specifically, calculate the incremental process noise covariance and add it to the original process noise covariance matrix.

[0094] The observation matrix H is used to map the model state to the observation space. For the convenience of operation, the observation matrix H can be set as:

[0095]

[0096] During the Kalman filtering process, based on the state vector e obtained from the landslide physical model at the previous time step k|k-1, and the state vector e is updated using Equation (2). k|k :

[0097] e k|k = e k|k-1 + K k (y k - He k|k-1 ) Equation (2)

[0098] K k is the ensemble Kalman filter gain matrix, y k is the observed data, and H is the observation matrix.

[0099] The above ensemble Kalman filter gain K k is calculated by Equation (3):

[0100] K k = P k|k-1 H T (HP k|k-1 H T + R) -1 Equation (3)

[0101] where H is the observation matrix, H T is the transpose of the observation matrix, and R is the observation noise covariance matrix. The state covariance matrix P k|k-1 in the above Equation (3) is calculated by the following Equation (4):

[0102] P k|k-1 = FP k-1|k-1 F T + Q Equation (4)

[0103] where P k-1|k-1 is the updated state covariance matrix corresponding to the previous time step (k - 1), and Q is the process noise covariance matrix.

[0104] F in the above equations is:

[0105]

[0106] υ is the nonlinear coefficient, which can be estimated from the actual observed data, and Δt is the time step.

[0107] Furthermore, the state covariance matrix P k|k needs to be updated. The updated state covariance matrix P k|k is used to predict the state vector (e k+1k ) at the next time step (k + 1). The state covariance matrix is updated using the following Equation (5):

[0108] P k|k = (I - Kk H)P k｜k-1 Formula (5)

[0109] Wherein, I is the identity matrix.

[0110] In summary, the state vector is updated through the ensemble Kalman filtering algorithm, and the model parameters of the updated landslide physical model can be obtained.

[0111] S200. Predict the landslide safety factor sequence based on the model parameters of the updated landslide physical model, and adjust the landslide safety factor threshold by using the landslide safety factor sequence.

[0112] The landslide safety factor sequence includes the landslide safety factors of the landslide body at multiple time points.

[0113] It can be understood that based on the idea of the finite element method, the landslide body is divided into multiple units, and the parameters of the landslide physical model corresponding to different units can also be different. In the process of updating the landslide physical model in S100 above, each unit corresponds to a set of model parameters. For example, for the i-th unit, the model parameters are {E i ,K i ,ξ i ,c i ,φ i ,ρ i}. Calculate the landslide safety factor of the landslide body based on the model parameters corresponding to multiple units.

[0114] In one implementation, the landslide body is divided into n units, and the landslide safety factor F S satisfies the following formula (6):

[0115]

[0116] In formula (6), c i ,η i ,ξ i ,φ i ,τ i are all obtained based on the model parameters of the landslide physical model. c i represents the cohesion of the i-th unit, η i represents the normal stress of the i-th unit (which can be calculated through the cohesion c i and the internal friction angle φ i ), ξ i represents the pore water pressure of the i-th unit, φ i represents the internal friction angle of the i-th unit, τ i represents the shear stress of the i-th unit (which can be calculated through the cohesion c i and the internal friction angle φ i ), A iIt represents the active area of the $i$-th unit (which can be obtained by measurement or by analyzing the landslide geometry in the terrain data).

[0117] Through the above formula (6), the landslide safety factors corresponding to the model parameters at multiple time steps can be calculated, thereby obtaining a sequence of landslide safety factors. Further, the landslide safety factor threshold is adjusted using the sequence of landslide safety factors. The specific process includes:

[0118] First, set a sliding window with a size of $N$. Use the sliding window to obtain the landslide safety factors of the $N$ time steps closest to the current time in the sequence of landslide safety factors. $N$ is an integer greater than or equal to 2, and $N$ can be dynamically adjusted according to the deformation characteristics and early warning requirements of the landslide.

[0119] Second, calculate the mean $\mu$ of the $N$ landslide safety factors within the sliding window w and the standard deviation $\sigma$. w .

[0120] Finally, calculate the adjusted landslide safety factor threshold. The adjusted landslide safety factor threshold $T$ new satisfies formula (7):

[0121]

[0122] where $T$ old represents the landslide safety factor threshold before adjustment, $\theta$ represents the threshold update factor ($0 < \theta < 1$), $F$ S represents the landslide safety factor closest to the current time predicted by the updated landslide physical model, $\lambda$ represents the weight coefficient, $\mu$ w represents the mean of the $N$ landslide safety factors within the sliding window in the sequence of landslide safety factors, $\sigma$ w represents the standard deviation of the $N$ landslide safety factors within the sliding window, $N$ is an integer greater than or equal to 2, and $k$ represents the standard deviation sensitivity coefficient.

[0123] Exemplarily, the value range of the above threshold update factor $\theta$ is $0 < \theta < 1$, the weight coefficient $\lambda$ takes values between 0.3 and 0.7, and the standard deviation sensitivity coefficient $k$ is a quantification index of model uncertainty. $k$ can be calculated based on the state covariance matrix $P$ finally obtained by the above ensemble Kalman filter algorithm, such as $tr(P)$ represents the trace of the state covariance matrix $P$.

[0124] Based on the above description, it can be seen that the landslide safety factor threshold corrected by the data assimilation technology and determined is a dynamically updated time-varying threshold (which can also be called a differential threshold), which conforms to the physical characteristics of the nonlinear evolution of the landslide. Subsequently, dynamic landslide early warning is carried out based on the landslide safety factor threshold to achieve accurate and real-time early warning, and the short-term and imminent early warning ability can be improved.

[0125] S300. Perform data assimilation processing on the Long-Short Term Memory (LSTM) model based on the second landslide observation data to update the LSTM model.

[0126] Similarly, the second landslide observation data also comes from the above-mentioned multi-source landslide data. The second landslide observation data includes the landslide displacement of the landslide body and the inclination angle of the landslide body within the monitoring area.

[0127] In the embodiments of the present application, the input of the LSTM model is multi-source landslide data, which includes the rainfall in the monitoring area, the landslide displacement of the landslide body, and the inclination angle of the landslide body. The output of the LSTM model is the landslide displacement. That is to say, the LSTM model is used to predict the landslide displacement.

[0128] It can be understood that the LSTM model is a machine learning model and a special recurrent neural network. The LSTM model can analyze the input data using time series. Landslides are dynamically changing. Using the LSTM model can capture the long-term dependence relationships between data and is suitable for landslide displacement prediction.

[0129] In the embodiments of the present application, no limitation is imposed on the specific structure of the LSTM model. An existing LSTM model structure can be adopted. The LSTM model is trained based on the landslide data training set. For example, the Adam optimizer is used for model training, and the learning rate is dynamically adjusted to accelerate the convergence speed. Then, the gradient descent method is used to optimize the hyperparameters of the LSTM model, and the optimal hyperparameter combination is selected through cross-validation to improve the prediction accuracy of the model.

[0130] Similar to the above S100, the algorithm for data assimilation processing used to update the LSTM model can also be the ensemble Kalman filter algorithm.

[0131] Specifically, set the state vector e as: e = [u x , u y , θ, W, b]. The state vector e contains the model parameters of the LSTM model, namely W and b. W is the weight matrix of the LSTM model, and b is the bias term of the LSTM model.

[0132] The state covariance matrix P is initialized as:

[0133]

[0134] Among them, is the uncertainty standard deviation of the horizontal landslide displacement in the second landslide observation data, is the uncertainty standard deviation of the vertical landslide displacement in the second landslide observation data, σ θis the standard deviation of the uncertainty of the inclination angle of the landslide body in the second landslide observation data.

[0135] Similarly, the observation noise covariance matrix R and the process noise covariance matrix Q can also be set to be the same as the initialized state covariance matrix P and adjusted according to the actual situation.

[0136] The observation matrix H can be set as:

[0137]

[0138] After that, referring to formulas (2) to (5) of the above S100, the state vector e and the state covariance matrix P are updated, and the model parameters of the updated LSTM model can be obtained according to the state vector e.

[0139] S400. Predict the landslide displacement sequence based on the updated LSTM model, and adjust the landslide displacement rate ratio threshold by using the landslide displacement sequence.

[0140] Input the multi-source landslide data of the monitored landslide body into the updated LSTM model. The LSTM model outputs the predicted landslide displacement sequence, and the landslide displacement sequence includes the landslide displacements of multiple sampling points of the landslide body at multiple time points.

[0141] In one implementation manner, adjusting the landslide displacement rate ratio threshold by using the landslide displacement sequence specifically includes S401 - S403:

[0142] S401. Calculate the landslide displacement rate ratio according to the landslide displacement sequence.

[0143] First, determine the landslide displacement rate sequence of the landslide body according to the landslide displacement sequence. The landslide displacement rate refers to the displacement amount of the landslide body per unit time. The calculation formula of the landslide displacement rate v is:

[0144]

[0145] Δs is the displacement change amount within a preset time period. The displacement change amount is the difference between the landslide displacements of two time steps. Δt is the time step.

[0146] Secondly, set a sliding window with a size of M. Use the sliding window to obtain the landslide displacement rates of the M time steps closest to the current moment in the landslide displacement rate sequence. For example, v = {v1, v2,..., v M}. M is an integer greater than or equal to 2, and M can be dynamically adjusted according to the deformation characteristics and early warning requirements of the landslide.

[0147] The landslide displacement rate ratio refers to the ratio of the landslide displacement rates in two consecutive time periods. The calculation formula of the landslide displacement rate ratio R is:

[0148]

[0149] v n is the landslide displacement rate in the current time period, and v n-1 is the landslide displacement rate in the previous time period.

[0150] According to the above landslide displacement rate sequence, a landslide displacement rate ratio sequence can be calculated, and the landslide displacement rate ratio closest to the current moment is selected from the landslide displacement rate ratio sequence.

[0151] S402. Determine the landslide deformation stage of the landslide body based on the landslide displacement rate ratio of the landslide body. The landslide deformation stage includes a constant velocity deformation stage, an initial acceleration deformation stage, a medium acceleration deformation stage, and an impending slide stage.

[0152] In one implementation, when the landslide displacement rate ratio is greater than the first value and less than or equal to the second value, the landslide deformation stage is the constant velocity deformation stage; when the landslide displacement rate ratio is greater than the second value and less than or equal to the third value, the landslide deformation stage is the initial acceleration deformation stage; when the landslide displacement rate ratio is greater than the third value and less than or equal to the fourth value, the landslide deformation stage is the medium acceleration deformation stage; when the landslide displacement rate ratio is greater than the fourth value, the landslide deformation stage is the impending slide stage.

[0153] Exemplarily, the above first value is 0, the second value is 2, the third value is 6, and the fourth value is 8. If 0 < R ≤ 2, the monitoring area is in the constant velocity deformation stage; if 2 < R ≤ 6, the monitoring area is in the initial acceleration deformation stage; if 6 < R ≤ 8, the monitoring area is in the medium acceleration deformation stage; if 6 < R ≤ 8, the monitoring area is in the acceleration deformation stage; if R > 8, the monitoring area is in the impending slide stage.

[0154] S403. When the landslide body is in the constant velocity deformation stage, determine that the adjusted landslide displacement rate ratio threshold is: R threshold = R base (1 - γ); when the landslide body is in the initial acceleration deformation stage, determine that the adjusted landslide displacement rate ratio threshold is: When the landslide body is in the medium acceleration deformation stage, determine that the adjusted landslide displacement rate ratio threshold is: When the landslide body is in the impending slide stage, determine that the adjusted landslide displacement rate ratio threshold is: R threshold = R base (1 + γ).

[0155] Among them, R threshold represents the adjusted landslide displacement rate ratio, R base represents the reference landslide displacement rate ratio, R represents the landslide displacement rate ratio closest to the current moment, and γ represents the smoothing factor.

[0156] Optionally, the smoothing factor γ can be calculated based on the state covariance matrix P finally obtained by the above ensemble Kalman filtering algorithm, as tr(P) represents the trace of the state covariance matrix P.

[0157] Based on the above description, it can be known that the landslide displacement rate ratio threshold corrected by the data assimilation technology and determined for the model is also a dynamically updated time-varying threshold (which can also be called a differential threshold). Subsequently, dynamic landslide early warning can be carried out based on the landslide displacement rate ratio threshold, realizing accurate and real-time early warning, and improving the short-term and impending early warning ability.

[0158] S500. Using the adjusted landslide displacement rate ratio threshold and the adjusted landslide safety factor threshold, fuse the current landslide safety factor of the landslide body in the monitoring area and the current landslide displacement of the landslide body to generate a landslide early warning index for the monitoring area, and perform landslide early warning according to the landslide early warning index.

[0159] The current landslide safety factor of the landslide body is predicted based on the model parameters of the updated landslide physical model. Specifically, substituting the latest model parameters of the above updated landslide physical model into the above formula (6) can predict the current landslide safety factor of the landslide body.

[0160] The current landslide displacement of the landslide body is predicted by the updated LSTM model. Specifically, inputting the latest multi-source landslide data of the landslide body into the updated LSTM model can predict the current landslide displacement of the landslide body.

[0161] In the embodiment of the present application, the landslide early warning index ω of the monitoring area satisfies formula (8):

[0162]

[0163] R represents the current landslide displacement rate ratio of the landslide body predicted, and R is calculated based on the landslide displacement output by the updated LSTM model. F s represents the current landslide safety factor of the landslide body predicted based on the updated landslide physical model. R threshold represents the landslide displacement rate ratio threshold. F Sthreshold represents the landslide safety factor threshold, and α and β respectively represent the weight coefficients of the landslide displacement rate ratio and the landslide safety factor.

[0164] The landslide early warning index calculated according to formula (8) is the result of multi-model fusion of the landslide physical model and the machine learning model, and is a comprehensive landslide early warning index, which can improve the model interpretability and early warning accuracy, and accurate and effective landslide early warning can be carried out based on this landslide early warning index.

[0165] In one implementation, a warning index threshold is set. When the landslide warning index is greater than the warning index threshold, a landslide warning is triggered. Exemplarily, the above warning index threshold can be set to 0.5. Of course, α and β can also be dynamically adjusted according to the actual landslide situation and warning effect to optimize the warning decision.

[0166] In another implementation, multiple levels of warning index thresholds can be set to determine the landslide risk level and trigger a landslide warning. For example, a basic index threshold, a medium-risk index threshold, and a high-risk threshold are set. When the above comprehensive landslide warning index exceeds the basic index threshold, a warning is triggered.

[0167] Furthermore, the warning level can be determined. Specifically:

[0168] When the landslide warning index exceeds the basic index threshold but does not reach the medium-risk index threshold, the warning level is a low-risk warning; when the landslide warning index exceeds the medium-risk index threshold but does not reach the high-risk index threshold, the warning level is a medium-risk warning; when the landslide warning index exceeds the high-risk index threshold, the warning level is a high-risk warning. In the case of a high-risk warning, emergency measures need to be taken immediately.

[0169] In summary, the dynamic landslide warning method based on a landslide physical model and data assimilation provided in the embodiments of the present application uses data assimilation technology to correct the landslide physical model for predicting the landslide safety factor and the machine learning model (LSTM model) for predicting the landslide displacement. Furthermore, the landslide safety factor threshold and the landslide displacement rate ratio threshold are dynamically adjusted, and the landslide safety factor and landslide displacement of the predicted landslide body are fused by multiple models to generate a landslide warning index for the monitoring area, and a landslide warning is carried out based on the landslide warning index. This method takes into account that the landslide body follows geophysical laws and combines data-driven machine learning models to dynamically update relevant thresholds, thereby calculating a comprehensive landslide warning index for the monitoring area, and can dynamically, more accurately and timely conduct landslide warnings for the monitoring area.

[0170] Furthermore, the embodiments of the present application also provide a landslide warning device, as Figure 2As shown in the figure, the landslide warning device includes a data collection and preprocessing module 201, a model construction module 202, a data assimilation module 203, and a warning decision-making module 204. Among them, the data collection and preprocessing module 201 includes a data acquisition unit for collecting multi-source landslide data in the above embodiments, and a data preprocessing unit for preprocessing the multi-source landslide data. The model construction module 202 includes a physical model unit for constructing a landslide physical model based on the principles of unsaturated soil mechanics, and a data-driven model unit for constructing an LSTM model based on data-driven. The data assimilation module 203 is used to update the landslide physical model and the LSTM model by using the ensemble Kalman filter algorithm. The warning decision-making module 204 includes a threshold calculation unit for adjusting the landslide safety factor threshold and the landslide displacement rate ratio threshold, and a multi-model collaborative decision-making unit for calculating the landslide warning index of the monitoring area.

[0171] Each module and unit of the above landslide warning device is used to execute the corresponding steps in the above method embodiments. All relevant contents involved in the above method embodiments can be cited in the function descriptions of the corresponding functional modules and will not be repeated here.

[0172] The embodiment of the present application also provides a landslide warning device, which includes: a processor and a memory coupled to the processor; the memory is used to store computer instructions. When the landslide warning device runs, the processor executes the computer instructions stored in the memory so that the landslide warning device executes the method described in the above embodiments.

[0173] The embodiment of the present application also provides a computer-readable storage medium, which includes a computer program. When the computer program runs on a computer, it is used to execute the method described in the above embodiments.

[0174] The embodiment of the present application also provides a computer program product, which includes computer program instructions. When the computer program instructions run on a computer, it is used to execute the method described in the above embodiments.

[0175] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the differences between each embodiment and other embodiments are emphasized.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A dynamic landslide early warning method based on a landslide physical model and data assimilation, characterized in that, Including: Performing data assimilation processing on a landslide physical model based on first landslide observation data to update the landslide physical model; wherein, the landslide physical model is a three-dimensional stress-seepage coupling model used to describe the interaction relationship between the stress field and the seepage field in a landslide body; the first landslide observation data includes the landslide displacement of the landslide body in the monitoring area, the inclination angle of the landslide body, and the landslide safety factor; Predicting a landslide safety factor sequence based on the model parameters of the updated landslide physical model, and adjusting the landslide safety factor threshold using the landslide safety factor sequence; the landslide safety factor sequence includes the landslide safety factors of the landslide body at multiple time points; Performing data assimilation processing on a long short-term memory (LSTM) model based on second landslide observation data to update the LSTM model; the second landslide observation data includes the landslide displacement of the landslide body in the monitoring area and the inclination angle of the landslide body; Predicting a landslide displacement sequence based on the updated LSTM model, and adjusting the landslide displacement rate ratio threshold using the landslide displacement sequence; the input of the LSTM model is multi-source landslide data, and the output is the landslide displacement. The multi-source landslide data includes the rainfall in the monitoring area, the landslide displacement of the landslide body, and the inclination angle of the landslide body; the landslide displacement sequence includes the landslide displacements of multiple sampling points of the landslide body at multiple time points; Performing fusion processing on the current landslide safety factor of the landslide body and the current landslide displacement of the landslide body using the adjusted landslide displacement rate ratio threshold and the adjusted landslide safety factor threshold to generate a landslide warning index for the monitoring area; and performing landslide warning based on the landslide warning index; the current landslide safety factor is predicted based on the model parameters of the updated landslide physical model, and the current landslide displacement is predicted by the updated LSTM model.

2. The method according to claim 1, wherein Adjusted landslide safety factor threshold T new Satisfy: T old represents the threshold of the landslide safety factor before adjustment, θ represents the threshold update factor, F S represents the landslide safety factor predicted according to the updated landslide physical model, λ represents the weight coefficient, μ w represents the mean of N landslide safety factors within the sliding window in the landslide safety factor sequence, σ w represents the standard deviation of the N landslide safety factors, N is an integer greater than or equal to 2, and k represents the standard deviation sensitivity coefficient.

3. The method according to claim 1, wherein The adjusting the landslide displacement rate ratio threshold using the landslide displacement sequence includes: Calculating a landslide displacement rate ratio according to the landslide displacement sequence; Determining the landslide deformation stage in which the landslide body is located based on the landslide displacement rate ratio; the landslide deformation stage includes a constant velocity deformation stage, an initial acceleration deformation stage, a medium acceleration deformation stage, and an impending slide stage; When the landslide body is in the stage of constant-rate deformation, determine that the adjusted landslide displacement rate ratio threshold is: R threshold = R base (1 - γ); When the landslide body is in the initial acceleration deformation stage, determining the adjusted landslide displacement rate ratio threshold as: When the landslide body is in the medium acceleration deformation stage, determining the adjusted landslide displacement rate ratio threshold as: When the landslide body is in the stage of impending sliding, determine that the adjusted landslide displacement rate ratio threshold is: R threshold = R base (1 + γ); Among them, R threshold represents the adjusted landslide displacement rate ratio, R base represents the reference landslide displacement rate ratio, R represents the landslide displacement rate ratio, and γ represents the smoothing factor.

4. The method according to claim 1, wherein The landslide warning index ω of the monitoring area satisfies: R represents the current landslide displacement rate ratio of the predicted landslide body, and R is calculated based on the landslide displacement of the landslide body output by the updated LSTM model, F s represents the current landslide safety factor of the landslide body predicted based on the updated landslide physical model, R threshold represents the landslide displacement rate ratio threshold, F Sthreshold represents the landslide safety factor threshold, and α and β respectively represent the weight coefficients of the landslide displacement rate ratio and the landslide safety factor.

5. The method according to claim 1, wherein The algorithm used for the data assimilation processing is the ensemble Kalman filter algorithm.

6. The method according to claim 1, wherein Divide the landslide body into n units, and the landslide safety factor F S Satisfies: Among them, c i represents the cohesion of the i-th unit, η i represents the normal stress of the i-th unit, ξ i represents the pore water pressure of the i-th unit, φ i represents the internal friction angle of the i-th unit, τ i represents the shear stress of the i-th unit, A i represents the acting area of the i-th unit; c i , η i , ξ i , φ i , τ i are all obtained based on the model parameters of the landslide physical model.

7. The method according to any one of claims 1 to 4, wherein The landslide displacement includes a horizontal landslide displacement and a vertical landslide displacement.

8. The method according to claim 7, wherein The landslide displacement includes GNSS landslide displacement collected by a Global Navigation Satellite System (GNSS) device, and / or InSAR landslide displacement collected by a synthetic aperture radar satellite.

9. A landslide warning device, characterized in that, Comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to perform the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, Comprising computer program instructions which, when executed by a computer, perform the method according to any one of claims 1 to 8.

Images