A dual-criteria joint landslide early warning method based on multi-model integrated rolling prediction

Abstract

The present invention discloses a dual-criteria joint landslide early warning method based on multi-mode integrated rolling prediction, belonging to the field of disaster early warning. For the prediction part: first, empirical mode decomposition and complete set empirical mode decomposition with adaptive noise are used to perform a secondary decomposition of the landslide displacement. According to the characteristics of different components, a multi-mode integrated prediction mechanism based on long-short-term memory network, gated recurrent unit, and spline model is designed. The update criteria of each model are designed to achieve multi-step rolling prediction. For the early warning part: first, the noise of the predicted value is removed by decomposition. The data is input into the Laida criterion + Saito model and the local outlier factor criterion + velocity inverse method for judgment, and the warning level is determined using the velocity threshold method. Using the method disclosed by the present invention, it is possible to analyze the monitoring data before the landslide damage occurs, thereby predicting the time of landslide occurrence and determining the warning level, thus achieving accurate early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of landslide disaster prediction and survey, in particular to a dual-criteria combined landslide early warning method based on multi-mode integrated rolling prediction. Background Art

[0002] Landslides are the most common geological disaster worldwide, posing a serious threat to human life and property. For example, according to the International Disaster Database, approximately 60,000 people have died worldwide from landslides since the early 20th century, with direct economic losses totaling nearly $10 billion. Furthermore, indirect economic losses from landslides are incalculable. Landslide prevention and control is a global challenge, a research hotspot and a challenging topic in international academia. However, existing landslide early warning models mostly use landslide displacement as a warning, making accurate prediction of landslide displacement crucial for subsequent early warning.

[0003] However, due to the complex causes of landslide geological hazards, which are typical of highly complex nonlinear systems, they are somewhat unpredictable and difficult to accurately predict in a timely and targeted manner. Although my country has installed various landslide monitoring instruments and equipment, such as GNSS, total stations, space-borne InSAR, and ground-based synthetic aperture radar, with some success, these instruments can only measure current landslide displacement and cannot provide future displacement estimates for the region, rendering them ineffective for early prediction. Therefore, in-depth research on how to analyze and predict landslide geological hazards using landslide displacement and fully exploit the inherent advantages of data-driven machine learning technology has important research and practical significance for disaster prevention and mitigation.

[0004] Moreover, many of the theoretically feasible landslide warning and forecast analysis methods currently available are post-analysis. Before the landslide causes damage and before the key deformation stage data is obtained, how to analyze the time of landslide occurrence as early and accurately as possible based on the existing monitoring data is the key to solving the actual application needs of slope engineering. Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a dual-criteria combined landslide early warning method based on multi-mode integrated rolling prediction.

[0006] The purpose of the present invention is achieved through the following technical solution: a dual-criteria combined landslide early warning method based on multi-mode integrated rolling prediction, comprising the following steps:

[0007] S1: Reducing the volatility of the original landslide sequence based on EMD and CEEMDAN secondary decomposition;

[0008] S2: Reconstruct the components after secondary decomposition based on the sample entropy value, select the corresponding prediction model according to the characteristics of different components, and increase the prediction length through the rolling prediction mechanism;

[0009] S3: Establish a model update mechanism based on slope change and prediction step size;

[0010] S4: Using EMD decomposition to reduce the volatility of the forecast series;

[0011] S5: Based on the rolling prediction results, the Laida criterion + Saito model and the LOF criterion + velocity inverse method model are used to jointly determine whether a landslide warning is needed;

[0012] S6: Determine the speed corresponding to each warning level in the four-level warning according to the normalized tangent angle method and give the warning level.

[0013] Preferably, in step S1, performing secondary decomposition on the original data further includes the following steps:

[0014] S11: For the original landslide displacement sequence x(t) = {x1, x2, ..., x n}Perform EMD decomposition once:

[0015]

[0016] Among them, q i (t) is the eigenmode component, r(t) is the residual component;

[0017] S12: Reconstruct the eigenmode components into trend term displacement g(t) and period term displacement u(t) according to the number of component maxima;

[0018] S13: Perform CEEMDAN secondary decomposition on the periodic term displacement:

[0019]

[0020] Among them, CIMF is the modal component and R(t) is the residual component.

[0021] Preferably, step S2 further includes the following steps:

[0022] S21: Calculate the sample entropy value of each CIMF: Let the pth modal component be h p (t), set the similarity tolerance r and embedding dimension m, and set h p (t) Reconstructed into n groups of m-dimensional vectors H p (t) = {h p (t),h p (t+1),......,h p(t+m-1)}(t=1,2,......,n), calculate two different vectors H p (i) and H p (j) the distance between

[0023]

[0024] d m (H p (i),H p (j))>rThe ratio of the number of nm is taken as B i (m,r)

[0025]

[0026] Among them, num{d m (H p (i),H p (j))>r} means d m (H p (i),H p (j)) is greater than the number of r, all B i The average value of (m,r) is recorded as B(m,r); similarly, B(m+1,r) can be obtained, then h p The sample entropy of (t) is:

[0027]

[0028] S22: Construct a reconstruction formula based on sample entropy:

[0029]

[0030] Among them, s(t) is the strong fluctuation term, k(t) is the medium fluctuation term, and w(t) is the weak fluctuation term;

[0031] S23: Strong fluctuation terms are predicted by the LSTM model, medium fluctuation terms are predicted by the GRU network, and weak fluctuation terms are predicted by the spline model;

[0032] S24: Establish a rolling forecast mechanism:

[0033]

[0034] in, All are predicted values.

[0035] Preferably, step S3 further includes the following steps:

[0036] S31: Establish a trend item prediction and update mechanism:

[0037] For the prediction of the trend term \(g(t)=\{g_1,g_2,\cdots,g\}\), calculate the slope \(k\) before and after the predicted value of the spline model. n Among them, \(g\)

[0038]

[0039] is the displacement value of the trend term corresponding to the \(n\)th moment, \(t\) is the time corresponding to the displacement. When the slope does not satisfy \(z < k < o\), retain the predicted value before the change moment and wait for new data to update. n

[0040] S32: Establish a reconstruction prediction update mechanism: Predict \(e\) steps at a time, and the total number of steps after the second prediction is \(2e\). When the total prediction steps reach \(2e\), update the spline-LSTM-GRU.

[0041] Preferably, in step S4, perform filtering and noise reduction processing through EMD decomposition.

[0042] Preferably, step S5 further includes the following steps:

[0043] S51: Find the landslide starting point through a data anomaly detection method:

[0044] Find the data exceeding \(3\sigma\) through the Laplace criterion, which is the landslide starting point.

[0045]

[0046] The distribution that the variable follows is called the normal distribution, denoted as \(x\sim N(\mu,\sigma\) 2 ), \(\sigma\) 2 is the standard deviation.

[0047] Adopt the LOF criterion to judge the starting point: The training set \(X = [x_1,x_2,\cdots,x\) N \(\in R\) D×N , the \(k\) nearest neighbor of \(x\) a is denoted as \(KNN(x\) a ). Calculate the \(k\)-distance of \(x\) a from its \(k\) nearest neighbor, and denote it as \(k\) a (x\) d ). The reachable distance between \(x\) a and another object \(x\) a is defined as: b

[0048] d r (x\) a ,x\) b ) = max{k d (x\) b ),d(x\) a ,x\)b )};

[0049] x a The inverse of the average reachable density of the k nearest neighbors is the local reachable density,

[0050]

[0051] By calculating x a and x b The local reachable density of LOF is obtained,

[0052]

[0053] S52: Calculate the landslide time of the landslide starting point found by the Laida criterion using the Saito model.

[0054]

[0055] Among them, t1, t2 and t3 are three moments with equal deformation rates, t r -t is the remaining destruction time;

[0056] S53: Calculate the landslide time of the landslide starting point found by the LOF criterion by the velocity inverse method.

[0057]

[0058] Where V is the velocity, t is the time, and t f is the landslide time, and A and α are constants related to the creep acceleration of the material.

[0059] Preferably, step S6 further includes the following steps:

[0060] S61: Calculate the normalized tangent angle,

[0061]

[0062] Among them, α i is the improved tangent angle, t i is the monitoring moment, Δt and ΔS correspond to the unit time period, and ΔT is the change of T(i) within the unit time period;

[0063] S62: Set blue, yellow, orange and red warning levels.

[0064] Preferably, in step S62, the tangent angles corresponding to the blue, yellow, orange and red warning levels are 44°, 45°, 79° and 81°, and the corresponding speeds are v1 to v4.

[0065] The present invention has the following advantages: by adopting the method disclosed by the present invention, it is possible to analyze the monitoring data before the landslide occurs, thereby predicting the time of occurrence of the landslide, giving an early warning level, and achieving accurate early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a structural diagram of the process of the dual-criteria joint landslide early warning method based on multi-model integrated rolling prediction. DETAILED DESCRIPTION

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0068] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are also within the scope of protection of the present invention.

[0069] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other.

[0070] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not require further definition or explanation in subsequent drawings.

[0071] In this embodiment, if Figure 1 As shown, a dual-criteria joint landslide early warning method based on multi-model integrated rolling prediction includes the following steps:

[0072] S1: Reducing the volatility of the original landslide sequence based on EMD and CEEMDAN secondary decomposition;

[0073] S2: Reconstruct the components after secondary decomposition based on the sample entropy value, select the corresponding prediction model according to the characteristics of different components, and increase the prediction length through the rolling prediction mechanism;

[0074] S3: Establish a model update mechanism based on slope change and prediction step size;

[0075] S4: Using EMD decomposition to reduce the volatility of the forecast series;

[0076] S5: Based on the rolling prediction results, the Laida criterion + Saito model and the LOF criterion + velocity inverse method model are used to jointly determine whether a landslide warning is needed;

[0077] S6: Determine the speed corresponding to each of the four warning levels using the normalized tangent angle method and assign a warning level. By employing the method disclosed in this invention, it is possible to analyze monitoring data before a landslide causes damage, thereby predicting the time of its occurrence and assigning a warning level, achieving accurate early warning.

[0078] Furthermore, in step S1, performing secondary decomposition on the original data further includes the following steps:

[0079] S11: For the original landslide displacement sequence x(t) = {x1, x2, ..., x n}Perform EMD decomposition once:

[0080]

[0081] Among them, q i (t) is the eigenmode component, r(t) is the residual component;

[0082] S12: Reconstruct the eigenmode components into trend term displacement g(t) and period term displacement u(t) according to the number of component maxima;

[0083] S13: Perform CEEMDAN secondary decomposition on the periodic term displacement:

[0084]

[0085] Among them, CIMF is the modal component and R(t) is the residual component.

[0086] The existing model predicts each IMF component and the residual term separately, and finally adds up the prediction results to get the total displacement. However, due to the large number of components, if each component is predicted separately, it will cause waste of prediction model and greatly increase the time required for prediction. Therefore, the present invention reconstructs the similar components, first removes the residual component as a gross error, and then calculates the sample entropy value S of each component. p (m, r), constructing a reconstruction relationship. Specifically, step S2 further includes the following steps:

[0087] S21: Calculate the sample entropy value of each CIMF: Let the pth modal component be h p (t), set the similarity tolerance r and embedding dimension m, and set h p (t) Reconstructed into n groups of m-dimensional vectors H p (t) = {h p (t),hp (t+1),......,h p (t+m-1)}(t=1,2,......,n), calculate two different vectors H p (i) and H p (j) the distance between

[0088]

[0089] d m (H p (i),H p (j))>rThe ratio of the number of nm is taken as B i (m,r)

[0090]

[0091] Among them, num{d m (H p (i),H p (j))>r} means d m (H p (i),H p (j)) is greater than the number of r, all B i The average value of (m,r) is recorded as B(m,r); similarly, B(m+1,r) can be obtained, then h p The sample entropy of (t) is:

[0092]

[0093] S22: Construct a reconstruction formula based on sample entropy:

[0094]

[0095] Among them, s(t) is the strong fluctuation term, k(t) is the medium fluctuation term, and w(t) is the weak fluctuation term;

[0096] S23: Strong fluctuation terms are predicted by the LSTM model, medium fluctuation terms are predicted by the GRU network, and weak fluctuation terms are predicted by the spline model;

[0097] S24: Establish a rolling forecast mechanism:

[0098]

[0099] in, All are predicted values.

[0100] In this embodiment, step S3 further includes the following steps:

[0101] S31: Establish a trend item prediction and update mechanism:

[0102] For the prediction of the trend term \(g(t)=\{g_1,g_2,\cdots,g n \}\), calculate the slope \(k\) before and after the predicted value of the spline model.

[0103]

[0104] where \(g n is the displacement value of the trend term corresponding to the \(n\)th moment, \(t\) is the time corresponding to the displacement. When the slope does not satisfy \(z < k < o\), retain the predicted value before the change moment and wait for new data to update;

[0105] S32: Establish a reconstruction prediction update mechanism for each item: predict \(e\) steps at a time, and the total step length after the second prediction is \(2e\). When the total prediction step length reaches \(2e\), update the spline - LSTM - GRU.

[0106] Furthermore, in step S4, filtering and noise reduction processing are performed through EMD decomposition.

[0107] [[ID=2​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​a ), by calculating x a Get x from the k nearest neighbors a The k distance is denoted as k d (x a ), x a With another object x b The reachable distance is defined as:

[0114] d r (x a ,x b )=max{k d (x b ),d(x a ,x b )};

[0115] x a The inverse of the average reachable density of the k nearest neighbors is the local reachable density,

[0116]

[0117] By calculating x a and x b The local reachable density of LOF is obtained,

[0118]

[0119] Specifically, for the LOF criterion, first find out a k nearest neighbors of (a=1,2,...,N) x a The k nearest neighbors of x are denoted as KNN(x a ), the final LOF(x a ) is obviously greater than 1, then x a is an outlier, i.e. the starting point of the landslide.

[0120] S52: Calculate the landslide time of the landslide starting point found by the Laida criterion using the Saito model.

[0121]

[0122] Among them, t1, t2 and t3 are three moments with equal deformation rates, t r -t is the remaining destruction time;

[0123] S53: Calculate the landslide time of the landslide starting point found by the LOF criterion by the velocity inverse method.

[0124]

[0125] Where V is the velocity, t is the time, and t fis the landslide time, and A and α are constants related to the creep acceleration of the material.

[0126] In this embodiment, step S6 further includes the following steps:

[0127] S61: Calculate the normalized tangent angle,

[0128]

[0129] Among them, α i is the improved tangent angle, t i is the monitoring moment, Δt and ΔS correspond to the unit time period, and ΔT is the change of T(i) within the unit time period;

[0130] S62: Set blue, yellow, orange and red warning levels. Further, in step S62, the tangent angles corresponding to the blue, yellow, orange and red warning levels are 44°, 45°, 79° and 81°, and the corresponding speeds are v1~v4. Specifically, blue warning: represents that the slope displacement deformation is in the initial or uniform deformation stage, and the slope will hardly become unstable and damaged within a year; yellow warning: represents that the slope deformation is in the initial accelerated deformation stage. During this stage, the slope is more likely to become unstable and damaged within a few months or a year, and this should be taken seriously; orange warning: represents that the slope deformation is in the medium accelerated deformation stage. During this stage, the slope is more likely to become unstable and damaged within a few days or months, and early warning messages should be issued from time to time to remind local residents to prepare for danger in advance; red warning: represents that the slope deformation is in the pre-slip stage. The slope will become unstable within a few hours or a few days, and early warning messages should be issued in real time to warn local residents to take immediate action.

[0131] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A dual-criteria joint landslide early warning method based on multi-mode integrated rolling prediction, characterized by: The following steps are involved: S1: Reducing the volatility of the original landslide sequence based on EMD and CEEMDAN secondary decomposition; S2: Reconstruct the components after secondary decomposition based on the sample entropy value, select the corresponding prediction model according to the characteristics of different components, and increase the prediction length through the rolling prediction mechanism; S3: Establish a model update mechanism based on slope change and prediction step size; S4: Using EMD decomposition to reduce the volatility of the forecast series; S5: Based on the rolling prediction results, the Laida criterion + Saito model and the LOF criterion + velocity inverse method model are used to jointly determine whether a landslide warning is needed; S6: Determine the speed corresponding to each warning level in the four-level warning according to the normalized tangent angle method and give the warning level.

2. The dual-criteria combined landslide early warning method based on multi-mode integrated rolling prediction according to claim 1 is characterized by: In step S1, performing secondary decomposition on the original data further includes the following steps: S11: For the original landslide displacement sequence x(t) = {x1, x2, ..., x n }Perform EMD decomposition once: Among them, q i (t) is the eigenmode component, r(t) is the residual component; S12: Reconstruct the eigenmode components into trend term displacement g(t) and period term displacement u(t) according to the number of component maxima; S13: Perform CEEMDAN secondary decomposition on the periodic term displacement: Among them, CIMF is the modal component and R(t) is the residual component.

3. The dual-criteria combined landslide early warning method based on multi-mode integrated rolling prediction according to claim 1 is characterized by: The step S2 further includes the following steps: S21: Calculate the sample entropy value of each CIMF: Let the pth modal component be h p (t), set the similarity tolerance r and embedding dimension m, and set h p (t) Reconstructed into n groups of m-dimensional vectors H p (t) = {h p (t),h p (t+1),......,h p (t+m-1)}(t=1,2,......,n), calculate two different vectors H p (i) and H p (j) the distance between d m (H p (i),H p (j))>rThe ratio of the number of nm is taken as B i (m,r) Among them, num{d m (H p (i),H p (j))>r} means d m (H p (i),H p (j)) is greater than the number of r, all B i The average value of (m,r) is recorded as B(m,r); similarly, B(m+1,r) can be obtained, then h p The sample entropy of (t) is: S22: Construct a reconstruction formula based on sample entropy: Among them, s(t) is the strong fluctuation term, k(t) is the medium fluctuation term, and w(t) is the weak fluctuation term; S23: Strong fluctuation terms are predicted by the LSTM model, medium fluctuation terms are predicted by the GRU network, and weak fluctuation terms are predicted by the spline model; S24: Establish a rolling forecast mechanism: in, All are predicted values.

4. The dual-criteria combined landslide early warning method based on multi-mode integrated rolling prediction according to claim 1 is characterized by: The step S3 further includes the following steps: S31: Establish a trend item prediction and update mechanism: For the trend term g(t) = {g1, g2, ..., g n }Prediction, calculate the slope k before and after the spline model prediction value, where g n is the displacement value of the trend term corresponding to the nth moment, t is the time corresponding to the displacement. When the slope does not satisfy z < k < o, the predicted value before the change moment is retained and waiting for new data to update; S32: Establish a mechanism to reconstruct various prediction updates: the first prediction step is e, and the total step after the second prediction is 2e. When the total prediction step reaches 2e, the spline-LSTM-GRU is updated.

5. The dual-criteria combined landslide early warning method based on multi-mode integrated rolling prediction according to claim 1 is characterized by: In step S4, filtering and noise reduction processing is performed through EMD decomposition.

6. The dual-criteria combined landslide early warning method based on multi-mode integrated rolling prediction according to claim 1 is characterized by: The step S5 further includes the following steps: S51: Find the starting point of the landslide by detecting data anomalies: The data exceeding 3σ is found by the Laida criterion, which is the starting point of the landslide. The distribution that the variable obeys is called normal distribution, which is denoted as x~N(μ,σ 2 ), σ 2 is the standard deviation; Use LOF criterion to determine the starting point: training set X=[x1,x2,...,x N ]∈R D×N , x a The k nearest neighbors of x are denoted as KNN(x a ), by calculating x a Get x from the k nearest neighbors a The k distance is denoted as k d (x a ), x a With another object x b The reachable distance is defined as: d r (x a ,x b )=max{k d (x b ),d(x a ,x b )}; x a The inverse of the average reachable density of the k nearest neighbors is the local reachable density, By calculating x a and x b The local reachable density of LOF is obtained, S52: Calculate the landslide time of the landslide starting point found by the Laida criterion using the Saito model. Among them, t1, t2 and t3 are three moments with equal deformation rates, t r -t is the remaining destruction time; S53: Calculate the landslide time of the landslide starting point found by the LOF criterion by the velocity inverse method. Where V is the velocity, t is the time, and t f is the landslide time, and A and α are constants related to the creep acceleration of the material.

7. The dual-criteria combined landslide early warning method based on multi-mode integrated rolling prediction according to claim 1 is characterized by: The step S6 further includes the following steps: S61: Calculate the normalized tangent angle, Among them, α i is the improved tangent angle, t i is the monitoring moment, Δt and ΔS correspond to the unit time period, and ΔT is the change of T(i) within the unit time period; S62: Set blue, yellow, orange and red warning levels.

8. The dual-criteria combined landslide early warning method based on multi-mode integrated rolling prediction according to claim 7 is characterized by: In step S62, the tangent angles corresponding to the blue, yellow, orange and red warning levels are 44°, 45°, 79° and 81°, and the corresponding speeds are v1 to v4.

Images