Landslide displacement prediction method based on VMD-DBO-LSTM model

Abstract

The invention discloses a landslide displacement prediction method based on a VMD-DBO-LSTM model. The method comprises the following steps: carrying out preprocessing operation on landslide displacement data; obtaining a plurality of intrinsic mode function components by using the preprocessed data subjected to variational mode decomposition; inputting the plurality of intrinsic mode function components into the trained DBO-LSTM model for prediction to obtain a prediction result of each component; and performing signal reconstruction on the prediction result of each component through component superposition to obtain a landslide displacement prediction result. The VMD method is introduced to decompose data, remove high-frequency noise, extract main feature signals, avoid modal aliasing, reduce end effect and the like, the accuracy of model prediction is improved, the DBO is introduced to optimize hyper-parameters of the LSTM model, global and local search capabilities are combined to adaptively adjust the search range and improve the optimization efficiency, and the method is suitable for large-scale popularization and application. The most suitable network parameters can be found more quickly, and the precision and stability of the model are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a landslide displacement prediction method based on a VMD-DBO-LSTM model. Background Art

[0002] Landslides are sudden natural disasters that cause damage not only in mountainous areas but also into adjacent built-up areas. Landslides and the geological hazards they trigger have severely impacted human life and safety, while also causing enormous economic losses. The surge in damage caused by landslides has drawn the attention of many researchers worldwide to the development of early landslide prediction systems. Properly developed monitoring systems can predict the likelihood of landslides in advance, facilitating the implementation of necessary measures to control and mitigate the damage caused by landslides to human life and the economy. Landslide assessment systems primarily address two key issues: the timing and location of landslides. However, accurately predicting the timing and location of impending landslides remains difficult for researchers. This is because landslides are triggered by a variety of factors, such as heavy rain, earthquakes, snowmelt, floods, volcanoes, or any other natural cause that can lead to slope instability. Furthermore, the varying topographical conditions of soil and rock also increase slope damage.

[0003] At present, the main methods of landslide prediction can be roughly divided into three categories: traditional monitoring methods, remote sensing technology and machine learning. Traditional landslide monitoring mainly relies on instruments such as total stations, theodolites, and levels. To a certain extent, they have reliable measurement accuracy and low equipment costs. However, they have obvious defects in the context of modern landslide monitoring needs. They require a lot of manual operation, are difficult to cover a large area, cannot achieve real-time monitoring, and are affected by weather and terrain conditions. In addition, the discreteness of data, poor real-time performance and high requirements for operating technology also make these traditional methods unable to meet the complex and rapidly changing landslide monitoring needs. The traditional dynamic model LSTM has a high prediction accuracy, and its prediction advantage is particularly prominent in the step deformation period. It does not rely on the analysis of the timeliness of training data (Yang Beibei, Yin Kunlong, Du Juan. Landslide displacement dynamic prediction model based on time series and long short-term memory network. Journal of Rock Mechanics and Engineering, 2018, (10): 2334-2343.). While remote sensing technology offers advantages for landslide monitoring and prevention, such as wide coverage, high efficiency, and contactlessness, it suffers from significant drawbacks in spatial and temporal resolution, weather impacts, and data processing complexity. Furthermore, because remote sensing lacks information about the internal structure of landslides and struggles to quantitatively simulate and analyze dynamic processes, remote sensing alone cannot fully and accurately prevent landslides. Machine learning-based landslide prediction methods, on the other hand, build machine learning models on time series to predict specific landslide displacements. As long as the error is kept within a certain range, this method can be proven practical and reasonable. Numerous existing machine learning-based landslide prediction methods exist, but they suffer from long data collection cycles, ranging from two to three years to decades, and untimely landslide predictions. Therefore, an algorithm with a short data collection cycle and timely prediction capabilities is urgently needed to predict landslide displacement data. Summary of the Invention

[0004] Aiming at the time series landslide displacement prediction, this paper proposes a landslide displacement prediction method based on the VMD-DBO-LSTM model. Through model improvement, it solves technical problems such as complex signal decomposition, nonlinear and non-stationary modeling, noise processing, and hyperparameter optimization in landslide displacement prediction, reduces sequence complexity, enhances the model noise reduction capability, and improves model accuracy and stability, providing a technical reference for modern landslide disaster management and early warning.

[0005] The present invention is achieved through at least one of the following technical solutions.

[0006] The landslide displacement prediction method based on the VMD-DBO-LSTM model includes the following steps:

[0007] Step 1: pre-processing the landslide displacement data;

[0008] Step 2: Using the preprocessed data using variational mode decomposition to obtain multiple intrinsic mode function components;

[0009] Step 3: Input multiple intrinsic mode function components into the trained DBO-LSTM model for prediction to obtain the prediction results of each component;

[0010] Step 4: Reconstruct the signal of each component prediction result by component superposition to obtain the landslide displacement prediction result.

[0011] Furthermore, the landslide displacement data includes GNSS monitoring data, wireline monitoring data and deep inclinometer data.

[0012] Furthermore, the preprocessing operation includes abnormal data processing and data normalization; the abnormal value processing of the original landslide displacement data includes unifying the format and unit, removing invalid data, removing unreasonable data, and removing outlier noise points.

[0013] Furthermore, the training of the DBO-LSTM model includes the following steps:

[0014] Step S1, collecting landslide displacement data, and preprocessing the data to divide the preprocessed data into a training set and a validation set;

[0015] Step S2, using variational mode decomposition (VMD) to decompose the data of the training set and the validation set into multiple intrinsic mode function components;

[0016] Step S3: Initialize the LSTM network model, set the LSTM model hyperparameter vector structure, and combine the dung beetle optimization algorithm with the LSTM network model to construct a DBO-LSTM model;

[0017] Step S4: The intrinsic mode function components obtained by decomposing the training set are fed into the DBO-LSTM model for training; the intrinsic mode function components of the validation set are fed into the trained DBO-LSTM model for model accuracy evaluation.

[0018] Furthermore, in step S3, the LST M network model is initialized, the Relu function is used as the activation function, and the Adam algorithm is used as the back propagation algorithm; three model hyperparameters are set: learning rate, number of neurons, and dropout layer dropout rate.

[0019] Furthermore, the DBO-LSTM model is optimized based on the dung beetle optimization algorithm, including the following steps:

[0020] Step (1), parameter initialization, generating initial dung beetle population information;

[0021] Step (2), inputting the initial dung beetle population information into the LSTM network model training;

[0022] Step (3) calculate the fitness f of each dung beetle based on the mean square error RMSE obtained from DBO-LSTM model training:

[0023]

[0024] Step (4), updating the position of each dung beetle. Each dung beetle population is divided into rolling dung beetles, brooding dung beetles, small dung beetles, and stealing dung beetles. The position updating method of each dung beetle is different.

[0025] Step (5) determines whether each target agent exceeds the boundary according to the constraint conditions. If it exceeds the boundary, the current value is reset to the boundary value, and the current optimal solution and fitness are updated by comparing the fitness;

[0026] After the dung beetle position is updated in step (6), return to step (2) and perform a new round of training;

[0027] Repeat steps (2) to (6) until the maximum number of iterations is reached. When the predetermined maximum number of iterations is reached, the global optimal position is output and the DBO-LSTM model optimization is completed.

[0028] Furthermore, the position update of the rolling ball dung beetle can be calculated by the following formula:

[0029]

[0030] In the formula, t represents the number of iterations, x i (t) represents the position information of the i-th ball-rolling dung beetle at the t-th iteration; k∈{0,1} divides the ball-rolling behavior into obstacle mode and barrier-free mode, k=1 represents barrier-free mode, and k=0 represents obstacle mode; W∈(0,1), β∈(0,0.2), α is a natural coefficient that takes 1 or -1, -1 represents deviation from the original direction, and 1 represents no deviation; θ is a randomly generated number uniformly distributed in the range [0,π], representing the deflection angle of the ball-rolling dung beetle; x w is the current global worst position;

[0031] The position update of the brooding dung beetle is calculated as follows:

[0032]

[0033] Where, X * is the current local optimal position; L * and U * are the lower and upper limits of the spawning area, respectively, R = 1-t / T max ,T max represents the maximum number of iterations, L and U represent the lower bound and upper bound of the optimization problem respectively; B Q(t) is the position information of the Q-th egg ball at the t-th iteration, b1 and b2 represent two independent random vectors of size 1×D, and D represents the dimension of the optimization problem.

[0034] The position update of the small dung beetle is calculated by the following formula:

[0035]

[0036] Where, X b Indicates the current global best position, L b and U b are the lower and upper bounds of the optimal foraging area respectively; p A (t) represents the position information of the Ath small dung beetle at the tth iteration, C1 represents a random number that obeys the normal distribution, and C2 represents a random vector belonging to (0,1);

[0037] The updated position of the stealing dung beetle can be calculated as follows:

[0038] C j (t+1)=X b +S×g×(|C j (t)-X * |+|C j (t)-X b |) (12);

[0039] Where C j (t) represents the position information of the j-th thieving dung beetle at the t-th iteration; g is a random vector of size 1×D dimensions that obeys the normal distribution; S represents a constant value.

[0040] Furthermore, the constraint condition is used to limit the dung beetle position update to not exceed the set upper and lower bounds:

[0041] X i (t+1)=min(max(X i (t),L b ),U b ) (13);

[0042] where X i (t) represents the location information of the i-th dung beetle population at the t-th iteration, x i (t), B Q (t), p A (t), C j (t)∈X i (t).

[0043] Furthermore, in step S4, the evaluation index of the DBO-LSTM model uses the root mean square error (RMSE) to calculate the absolute error between the predicted value and the true value:

[0044]

[0045] Where y q Represents the observation value of the qth point in the time series, and the observation value is the true value. is the predicted value output by the DBO-LSTM model, and n is the number of data points.

[0046] A computer device of the present invention comprises: a memory and a processor and a computer program stored in the memory. When the computer program is executed on the processor, the landslide displacement prediction method based on the VMD-DBO-LSTM model is implemented.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. This paper proposes for the first time a landslide displacement prediction method based on the VMD-DBO-LSTM model, combining the advantages of signal decomposition, hyperparameter optimization, and deep learning time series modeling to form an end-to-end landslide displacement prediction method. By introducing the VMD method to decompose data, remove high-frequency noise, and extract the main characteristic signals, it has obvious advantages in avoiding modal aliasing and reducing endpoint effects. It can not only reduce the complexity of the sequence, but also make the input data more physically meaningful and representative, which can significantly improve the accuracy of model prediction. By simulating the natural behavior of dung beetles for optimization solution, the most suitable parameters for the LSTM network can be found more quickly, thereby improving the accuracy and stability of the model.

[0049] 2. This method introduces DBO to optimize the hyperparameters of the LSTM model, combining global and local search capabilities to adaptively adjust the search range, improve optimization efficiency, and find the most suitable network parameters more quickly, thereby improving the accuracy and stability of the prediction model. Since the performance of the LSTM network usually depends on the settings of multiple hyperparameters, such as the number of hidden layers, number of time steps, learning rate, etc., these parameters have an important impact on the training results of the model. Traditional manual parameter adjustment or grid search methods are inefficient and difficult to find the global optimal solution. The present invention improves the LSTM network based on the DBO algorithm, automatically optimizes the hyperparameter configuration of the LSTM, thereby finding the optimal solution and improving the stability and accuracy of the prediction model. This enables the model to automatically find the optimal configuration within a limited time, solving the problem of difficult hyperparameter selection.

[0050] 3. The present invention can efficiently and accurately predict landslide displacement changes, which makes the method have important application value in landslide early warning, disaster prevention and mitigation, and engineering safety assessment, thereby bringing good social and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Flowchart of a landslide displacement prediction method based on VMD-DBO-LSTM model;

[0052] Figure 2 Schematic diagram of components after VMD decomposition in the embodiment;

[0053] Figure 3 1 is a landslide displacement prediction result diagram of an embodiment;

[0054] Figure 4 The figure is a flow chart of real-time monitoring and early warning of landslides in an embodiment. DETAILED DESCRIPTION

[0055] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] like Figure 1 As shown, a landslide displacement prediction method based on the VMD-DBO-LSTM model of this embodiment includes the following steps:

[0057] Step 1: pre-processing the landslide displacement data;

[0058] The landslide displacement data includes GNSS monitoring data, wireline monitoring data, and deep inclinometer data. Preprocessing operations include abnormal data processing and data normalization. Outlier processing of the raw landslide displacement data includes standardizing formats and units, removing invalid data, removing unreasonable data, and removing outlier noise points.

[0059] Step 2: Using the preprocessed data using variational mode decomposition to obtain multiple intrinsic mode function components;

[0060] Step 3: Input multiple intrinsic mode function components into the trained DBO-LSTM model for prediction to obtain the prediction results of each component;

[0061] The training of the DBO-LSTM model consists of the following steps:

[0062] Step S1: collect landslide displacement sequence data, perform preprocessing on the data, and divide the preprocessed data into an 80% training set and a 20% validation set.

[0063] The preprocessing steps include removing outliers and normalizing the data. Data normalization primarily scales eigenvalues, balances the influence of different features, accelerates model convergence, improves DBO-LSTM model performance, and reduces numerical errors within the model. This preprocessing method not only improves DBO-LSTM model training but also enables the model to better understand and utilize each feature, thereby enhancing the model's generalization and predictive accuracy. Furthermore, after data normalization, the collected data is brought to the same order of magnitude across all metrics, making it suitable for comprehensive comparative evaluation.

[0064] Step S2: Decompose the training set and validation set data into multiple IMF (Intrinsic Mode Functions) components using VMD (Variational Mode Decomposition).

[0065] VMD decomposition is a decomposition technique based on variational methods. Compared with traditional decomposition methods such as empirical mode decomposition (EMD), it is more robust to noise, has more stable decomposition results, can effectively suppress modal aliasing problems, and has good noise resistance. The number of components can be adjusted according to the actual data set. As an example, the data can be decomposed into 3 IMF components. The schematic diagram of each component after decomposition is as follows: Figure 2 As shown, Figure 2 (a) is the low-frequency imf0 component (trend term), Figure 2 (b) is the intermediate frequency imf1 component (periodic term), Figure 2 (c) is the high-frequency imf2 component (noise term).

[0066] Step S3: Each IMF component obtained by decomposing the training set is fed into the DBO-LSTM model for training. Then, the IMF components in the validation set are fed into the trained DBO-LSTM model to perform model accuracy test and evaluation. The DBO-LSTM model is a combination of the Dung Beetle Optimization Algorithm and the Long Short-Term Memory Network (LSTM), and is an improved model of the LSTM. DBO is an optimization algorithm. The purpose of introducing DBO is to optimize the hyperparameters of the LSTM model during training to better improve the prediction effect of the model. For example, the model in landslide displacement prediction based on ICEEMDAN decomposition and SE reconstruction and DBO-LSTM (Feng Qingqing, Li Limin, Chen Feiyang, et al. Landslide displacement prediction based on ICEEMDAN decomposition and SE reconstruction and DBO-LSTM [J]. Electronic Measurement Technology, 2024, 47(07): 80-87. DOI: 10.19651 / j.cnki.emt.2415498.).

[0067] The evaluation index of the DBO-LSTM model in this embodiment uses the root mean square error (RMSE) to calculate the absolute error between the predicted value and the true value, as shown in Formula 1:

[0068]

[0069] Where y q Represents the observed value of the qth point in the time series, which is the true value. is the predicted value output by the DBO-LSTM model, and n is the number of data points.

[0070] This embodiment uses the DBO-LSTM model optimized based on the dung beetle optimization algorithm, including the following steps:

[0071] Step (1): Initialize parameters and generate initial dung beetle population information.

[0072] As an embodiment, the parameters of this embodiment are set as follows: population size 5; the proportion of rolling dung beetles is 0.2, the proportion of breeding dung beetles is 0.4, the proportion of small dung beetles is 0.2, and the proportion of stealing dung beetles is 0.2; the search dimension is set to 2, which is used to optimize two parameters of the DBO-LSTM model: learning rate and number of neurons.

[0073] Step (2) inputs the LSTM (Long Short-Term Memory) network model; initializes the LSTM network model, uses the Relu (Rectified Linear Unit) function as the activation function, and uses the Adam (Adaptive Moment Estimation) algorithm as the backpropagation algorithm; selects three model hyperparameters: learning rate, number of neurons, and dropout layer dropout rate.

[0074] Step (3) calculates the fitness of each dung beetle based on the RMSE (mean square error) obtained from the DBO-LSTM model training. The fitness f is calculated as follows:

[0075]

[0076] Step (4): Update the position of each dung beetle. Each dung beetle population is divided into ball-rolling dung beetles, brooding dung beetles, small dung beetles, and stealing dung beetles. The position update method of each dung beetle is different. The update method adopts the existing method (Xue, Jiankai, Shen, et al. Dung beetle optimizer: A new meta-heuristic algorithm for global optimization [J]. The Journal of Supercomputing, 2023, 79 (7): 7305-7336.).

[0077] The position update of the rolling dung beetle can be calculated by the following formula:

[0078]

[0079] In the formula, t represents the number of iterations, x i (t) represents the position information of the i-th ball-rolling dung beetle at the t-th iteration; k∈{0,1} divides the ball-rolling behavior into obstacle mode and barrier-free mode, k=1 represents barrier-free mode, and k=0 represents obstacle mode; W∈(0,1) is a constant, β∈(0,0.2) is a constant, α is a natural coefficient that takes 1 or -1, -1 represents deviation from the original direction, and 1 represents no deviation; θ is a randomly generated number uniformly distributed in the range [0,π], representing the deflection angle of the ball-rolling dung beetle; x w is the current global worst position.

[0080] The position update of the brooding dung beetle can be calculated as follows:

[0081]

[0082] Where, X * is the current local optimal position; L * and U * are the lower and upper limits of the spawning area, respectively, R = 1-t / T max , T max represents the maximum number of iterations, L and U represent the lower bound and upper bound of the optimization problem respectively; B Q (t) is the position information of the Q-th egg ball at the t-th iteration, b1 and b2 represent two independent random vectors of size 1×D, and D represents the dimension of the optimization problem.

[0083] The position update of the small dung beetle can be calculated by the following formula:

[0084]

[0085] Where, X b Indicates the current global best position, L b and U b are the lower and upper bounds of the optimal foraging area respectively; p A (t) represents the position information of the Ath small dung beetle at the tth iteration, C1 represents a random number that obeys the normal distribution, and C2 represents a random vector belonging to (0,1);

[0086] The updated position of the stealing dung beetle can be calculated as follows:

[0087] C j (t+1)=X b +S×g×(|Cj (t)-X * |+|C j (t)-X b |) (20);

[0088] Where C j (t) represents the position information of the j-th thieving dung beetle at the t-th iteration; g is a random vector of size 1×D, which obeys the normal distribution; S represents a constant value, such as 0.2 or 0.5. In this embodiment, the value is 0.5.

[0089] Step (5) determines whether each target agent (dung beetle) exceeds the boundary according to the constraint conditions. If it exceeds the boundary, the current value is reset to the boundary value, and then the current optimal solution and its fitness are updated by comparing the fitness.

[0090] After the dung beetle position is updated in step (6), the model returns to step (1) and undergoes a new round of training.

[0091] The constraint condition is used to limit the dung beetle's position update to not exceed the set upper and lower bounds. The specific constraint condition can be expressed by the following formula:

[0092] X i (t+1)=min(max(X i (t),L b ),U b ) (twenty one);

[0093] where X i (t) represents the location information of the i-th dung beetle population at the t-th iteration, x i (t), B Q (t), p A (t), C j (t)∈X i (t).

[0094] Repeat the above operations (step (2)-step (6)) until the maximum number of iterations is reached. When the predetermined maximum number of iterations is reached, the global optimal position is output and the DBO-LSTM model optimization is completed.

[0095] Step 4: Reconstruct the signal by superimposing the predicted results of each component obtained in step 3, obtain the predicted result of landslide displacement and draw the result graph. After obtaining the predicted result, the tangent angle G can be calculated by the following formula:

[0096]

[0097] Where y q represents the predicted value, o qRepresents the displacement moment. If the tangent-tangent angle G is greater than 85°, an alarm will be issued. Figure 4 shown.

[0098] The results of this embodiment are shown in FIG. Figure 3 As shown in the figure, the overall trend of the dotted line (predicted value) and the solid line (true value) is consistent, indicating that the prediction method can well capture the main trend of displacement change over time. There are slight deviations at some local points, which may be caused by the error of the prediction model. Overall, the prediction model can well fit the changing trend of the cumulative displacement, with a small overall error, and has certain practicality.

[0099] This embodiment further provides a real-time landslide monitoring and early warning device for implementing the landslide displacement prediction method based on the VMD-DBO-LSTM model, comprising:

[0100] at least one processor;

[0101] at least one memory for storing at least one program;

[0102] When the at least one program is executed by at least one processor, the at least one processor implements Figure 4 The program shown.

[0103] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, numerous modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, so that those skilled in the art can better understand and utilize the present invention.

Claims

1. The landslide displacement prediction method based on VMD-DBO-LSTM model is characterized by: The following steps are involved: Step 1: pre-processing the landslide displacement data; Step 2: Using the preprocessed data using variational mode decomposition to obtain multiple intrinsic mode function components; Step 3: Input multiple intrinsic mode function components into the trained DBO-LSTM model for prediction to obtain the prediction results of each component; Step 4: Reconstruct the signal of each component prediction result by component superposition to obtain the landslide displacement prediction result.

2. The landslide displacement prediction method of the VMD-DBO-LSTM model according to claim 1 is characterized in that: The landslide displacement data includes GNSS monitoring data, wireline monitoring data and deep inclinometer data.

3. The landslide displacement prediction method based on the VMD-DBO-LSTM model according to claim 1 is characterized in that: The preprocessing operation includes abnormal data processing and data normalization; the abnormal value processing of the original landslide displacement data includes unifying the format and unit, removing invalid data, removing unreasonable data, and removing outlier noise points.

4. The landslide displacement prediction method based on the VMD-DBO-LSTM model according to claim 1 is characterized in that: The training of the DBO-LSTM model consists of the following steps: Step S1, collecting landslide displacement data, and preprocessing the data to divide the preprocessed data into a training set and a validation set; Step S2, using variational mode decomposition (VMD) to decompose the data of the training set and the validation set into multiple intrinsic mode function components; Step S3: Initialize the LSTM network model, set the LSTM model hyperparameter vector structure, and combine the dung beetle optimization algorithm with the LSTM network model to construct a DBO-LSTM model; Step S4: The intrinsic mode function components obtained by decomposing the training set are fed into the DBO-LSTM model for training; the intrinsic mode function components of the validation set are fed into the trained DBO-LSTM model for model accuracy evaluation.

5. The landslide displacement prediction method based on the VMD-DBO-LSTM model according to claim 4 is characterized in that: In step S3, the LST M network model is initialized, the Relu function is used as the activation function, and the Adam algorithm is used as the back propagation algorithm; three model hyperparameters are set: learning rate, number of neurons, and dropout layer dropout rate.

6. The landslide displacement prediction method based on the VMD-DBO-LSTM model according to claim 4 is characterized in that: Optimizing the DBO-LSTM model based on the dung beetle optimization algorithm includes the following steps: Step (1), parameter initialization, generating initial dung beetle population information; Step (2), inputting the initial dung beetle population information into the LSTM network model training; Step (3) calculate the fitness f of each dung beetle based on the mean square error RMSE obtained from DBO-LSTM model training: Step (4), updating the position of each dung beetle. Each dung beetle population is divided into rolling dung beetles, brooding dung beetles, small dung beetles, and stealing dung beetles. The position updating method of each dung beetle is different. Step (5) determines whether each target agent exceeds the boundary according to the constraint conditions. If it exceeds the boundary, the current value is reset to the boundary value, and the current optimal solution and fitness are updated by comparing the fitness; After the dung beetle position is updated in step (6), return to step (2) and perform a new round of training; Repeat steps (2) to (6) until the maximum number of iterations is reached. When the predetermined maximum number of iterations is reached, the global optimal position is output and the DBO-LSTM model optimization is completed.

7. The landslide displacement prediction method based on the VMD-DBO-LSTM model according to claim 6 is characterized in that: The position update of the rolling dung beetle can be calculated by the following formula: In the formula, t represents the number of iterations, x i (t) represents the position information of the i-th ball-rolling dung beetle at the t-th iteration; k∈{0,1} divides the ball-rolling behavior into obstacle mode and barrier-free mode, k=1 represents barrier-free mode, and k=0 represents obstacle mode; W∈(0,1), β∈(0,0.2), α is a natural coefficient that takes 1 or -1, -1 represents deviation from the original direction, and 1 represents no deviation; θ is a randomly generated number uniformly distributed in the range [0,π], representing the deflection angle of the ball-rolling dung beetle; x w is the current global worst position; The position update of the brooding dung beetle is calculated as follows: Where, X * is the current local optimal position; L * and U * are the lower and upper limits of the spawning area, respectively, R = 1-t / T max ,T max represents the maximum number of iterations, L and U represent the lower and upper bounds of the optimization problem respectively; B Q (t) is the position information of the Q-th egg ball at the t-th iteration, b1 and b2 are two independent random vectors of size 1×D, and D is the dimension of the optimization problem; The position update of the small dung beetle is calculated by the following formula: Where, X b Indicates the current global best position, L b and U b are the lower and upper bounds of the optimal foraging area respectively; p A (t) represents the position information of the Ath small dung beetle at the tth iteration, C1 represents a random number that obeys the normal distribution, and C2 represents a random vector belonging to (0,1); The updated position of the stealing dung beetle can be calculated as follows: C j (t+1)=X b +S×g×(|C j (t)-X * |+|C j (t)-X b |) (5); Where C j (t) represents the position information of the j-th thieving dung beetle at the t-th iteration; g is a random vector of size 1×D dimensions that obeys the normal distribution; S represents a constant value.

8. The landslide displacement prediction method based on the VMD-DBO-LSTM model according to claim 7 is characterized in that: The constraints are used to prevent the dung beetle from updating its position beyond the upper and lower bounds: X i (t+1)=min(max(X i (t),L b ),U b ) (6); where X i (t) represents the location information of the i-th dung beetle population at the t-th iteration, x i (t), B Q (t), p A (t), C j (t)∈X i (t).

9. The landslide displacement prediction method based on the VMD-DBO-LSTM model according to claim 4 is characterized in that: In step S4, the evaluation index of the DBO-LSTM model uses the root mean square error (RMSE) to calculate the absolute error between the predicted value and the true value: Where y q Represents the observation value of the qth point in the time series, and the observation value is the true value. is the predicted value output by the DBO-LSTM model, and n is the number of data points.

10. A computer device, characterized in that: The invention comprises: a memory and a processor and a computer program stored in the memory, and when the computer program is executed on the processor, implements the landslide displacement prediction method based on the VMD-DBO-LSTM model as described in any one of claims 1 to 9.

Images