A landslide deformation prediction method, device, terminal device and storage medium based on variational modal decomposition

Abstract

The present invention discloses a landslide deformation prediction method, apparatus, terminal device, and storage medium based on variational modal decomposition. The method comprises: first, acquiring landslide deformation data and normalizing it to obtain target landslide deformation data; then, performing variational modal decomposition on the target landslide deformation data to obtain deformation modal components and center frequency values; then, clustering all center frequency values ​​to determine periodic deformation data; then, inputting the target landslide deformation data into a preset trend deformation prediction model to obtain a trend deformation prediction value; then, inputting the periodic deformation data and landslide deformation trend influencing factor data into a preset periodic deformation prediction model to obtain a periodic deformation prediction value; and finally, calculating the sum of the trend deformation prediction value and the periodic deformation prediction value to obtain a landslide deformation prediction value. By implementing the present invention, data noise in the original landslide deformation data can be reduced, clarity can be enhanced, and the accuracy of deformation prediction results can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of landslide deformation prediction, and in particular to a landslide deformation prediction method, apparatus, terminal equipment and storage medium based on variational modal decomposition. Background Art

[0002] Landslides, as a type of geological hazard, are devastating and pose a serious threat to human production and livelihoods. Therefore, strengthening the research and development of landslide risk prediction methods is particularly important. In recent years, deep learning technology has demonstrated significant potential in landslide risk prediction. Through the learning capabilities of deep neural networks, complex patterns and trends in landslide monitoring data can be better captured and analyzed, providing strong support for accurate landslide risk prediction.

[0003] Landslide deformation data often contains a large amount of noise and uncertainty, such as data anomalies caused by factors such as weather changes, equipment failures, or changes in ground cover. Therefore, existing models are prone to misjudgment or omission when faced with this complex data, making it difficult to accurately identify and extract key spatial variation characteristics in landslide deformation time series, thereby reducing the accuracy of deformation prediction results. Summary of the Invention

[0004] The present invention provides a landslide deformation prediction method, apparatus, terminal device and storage medium based on variational modal decomposition, which can reduce data noise in original landslide deformation data, enhance the clarity of original landslide deformation data, and then obtain the valid data portion of the original landslide deformation data, so that the model can more accurately identify and extract key spatial variation feature information in the original landslide deformation data based on the valid data portion, ultimately improving the accuracy of deformation prediction results.

[0005] An embodiment of the present invention provides a landslide deformation prediction method based on variational mode decomposition, comprising:

[0006] Acquire current landslide deformation data and normalize the landslide deformation data to obtain target landslide deformation data; wherein the landslide deformation data includes: landslide cumulative deformation data and landslide deformation trend influencing factor data;

[0007] Performing Hilbert transform on the target landslide deformation data to obtain a landslide deformation source signal, and performing variational modal decomposition on the landslide deformation source signal to obtain a preset number of deformation modal components and a center frequency value corresponding to each deformation modal component;

[0008] All center frequency values ​​are clustered, and the deformation modal components corresponding to each center frequency value in the center frequency data with the highest center frequency value are taken as periodic deformation data;

[0009] Inputting the target landslide deformation data into a preset trend deformation prediction model, so that the preset trend deformation prediction model obtains a trend deformation prediction value based on the target landslide deformation data;

[0010] Inputting the above-mentioned periodic deformation data and the above-mentioned landslide deformation trend influencing factor data into a preset periodic deformation prediction model, so that the above-mentioned preset periodic deformation prediction model obtains a periodic deformation prediction amount based on the above-mentioned periodic deformation data and the above-mentioned landslide deformation trend influencing factor data;

[0011] The sum of the trend deformation prediction amount and the periodic deformation prediction amount is calculated to obtain the landslide deformation prediction value.

[0012] Furthermore, the above-mentioned acquisition of current landslide deformation data and normalization of the above-mentioned landslide deformation data to obtain target landslide deformation data include:

[0013] Extracting minimum landslide deformation data and maximum landslide deformation data from the above landslide deformation data;

[0014] According to the above minimum landslide deformation data, the above maximum landslide deformation data and the above landslide deformation data, the above target landslide deformation data is calculated by the following formula:

[0015]

[0016] Where y represents the target landslide deformation data, x represents the landslide deformation data, min represents the minimum landslide deformation data, and max represents the maximum landslide deformation data.

[0017] Furthermore, the landslide deformation source signal is subjected to variational modal decomposition to obtain a preset number of deformation modal components and the center frequency value corresponding to each deformation modal component, including:

[0018] The augmented Lagrangian function is constructed based on the above landslide deformation source signal;

[0019] Iteratively solving the augmented Lagrangian function to obtain a preset number of first deformation modal components;

[0020] After each first deformation modal component is obtained, determining whether the first deformation modal component satisfies a preset iteration stop condition; if so, determining the current first deformation modal component as the deformation modal component, and determining the first center frequency value corresponding to each current first deformation modal component as the center frequency value;

[0021] Among them, the above preset iteration stopping condition is:

[0022]

[0023] Where, represents the kth first deformation modal component obtained after the nth iteration, represents the kth first deformation modal component obtained after the n+1th iteration, and ε represents the preset accuracy.

[0024] Furthermore, the training of the preset trend deformation prediction model includes:

[0025] Acquire a number of sample landslide deformation data, and normalize the sample landslide deformation data to obtain target sample landslide deformation data; wherein the sample landslide deformation data includes: sample landslide cumulative deformation data and sample landslide deformation trend influencing factor data;

[0026] Performing Hilbert transform on the target sample landslide deformation data to obtain a sample landslide deformation source signal, and performing variational modal decomposition on the sample landslide deformation source signal to obtain a preset number of sample deformation modal components and a sample center frequency value corresponding to each sample deformation modal component;

[0027] Clustering all the sample center frequency values ​​to obtain the first sample center frequency with the lowest center frequency value, and taking the sample deformation modal components corresponding to each center frequency value within the first sample center frequency as the sample trend item deformation data;

[0028] Obtaining a first true label corresponding to the sample trend item deformation data, and inputting the sample trend item deformation data with the first true label into the trend deformation prediction model to be trained, so that the trend deformation model to be trained predicts a first trend deformation prediction amount based on the sample trend item deformation data; wherein the first true label is the actual trend deformation amount of the sample trend item deformation data;

[0029] Calculate a first loss function value based on the first trend deformation prediction amount and the actual trend deformation amount;

[0030] Determine whether the value of the first loss function converges; if so, the training of the trend deformation prediction model to be trained is completed, and the preset trend deformation prediction model is obtained; if not, optimize the hyperparameters in the trend deformation prediction model according to the first particle swarm optimization algorithm, and then continue to train the trend deformation prediction model to be trained.

[0031] Furthermore, the hyperparameters in the trend deformation prediction model are optimized according to the first particle swarm optimization algorithm, including:

[0032] Repeating the first parameter optimization operation until the difference between the current first fitness and the previous first fitness is less than a preset threshold, thereby obtaining the optimized hyperparameters of the trend deformation prediction model;

[0033] The first parameter optimization operation includes:

[0034] Obtain the current first hyperparameter to be optimized, the current first hyperparameter position, the current first hyperparameter speed, the current first hyperparameter individual optimal position, and the current first hyperparameter overall optimal position of the trend deformation prediction model; wherein, the initial first hyperparameter to be optimized is the preset initial first hyperparameter, the initial first hyperparameter position and the first hyperparameter speed are the random hyperparameter position and the random hyperparameter speed, the initial first hyperparameter individual optimal position is the preset first hyperparameter individual optimal position, and the initial first hyperparameter overall optimal position is the preset first hyperparameter overall optimal position;

[0035] Under the current first hyperparameter speed and the first hyperparameter position, according to the current first trend deformation prediction amount and the current actual trend deformation amount, a first fitness of the current first hyperparameter to be optimized is calculated;

[0036] Calculate the first fitness difference between the current first fitness and the previous first fitness. If the current first fitness difference is not less than a preset threshold, for each first hyperparameter to be optimized, calculate the updated first hyperparameter position and first hyperparameter speed based on the current first hyperparameter position, the current first hyperparameter speed, the current first hyperparameter individual optimal position, and the current first hyperparameter overall optimal position;

[0037] Determine the difference between the current first fitness and the previous first fitness;

[0038] If the current first fitness is less than the previous first fitness, the updated first hyperparameter position is used as the updated first hyperparameter individual optimal position, and then the first hyperparameter overall optimal position is generated based on the updated first hyperparameter individual optimal positions of all first hyperparameters to be optimized; otherwise, the current first hyperparameter individual optimal position is used as the next first hyperparameter individual optimal position, and the current first hyperparameter overall optimal position is used as the next first hyperparameter overall optimal position.

[0039] Furthermore, the training of the preset periodic deformation prediction model includes:

[0040] Clustering all the sample center frequency values ​​to obtain the second sample center frequency with the highest center frequency value, and taking the sample deformation modal components corresponding to each center frequency value within the second sample center frequency as the sample periodic item deformation data;

[0041] Obtaining a second true label corresponding to the sample periodic item deformation data, and inputting the sample periodic item deformation data with the second true label into the periodic deformation prediction model to be trained, so that the periodic deformation model to be trained predicts a first periodic deformation prediction amount based on the sample periodic item deformation data; wherein the second true label is the actual periodic deformation amount of the sample periodic item deformation data;

[0042] Calculating a second loss function value based on the first periodic deformation prediction amount and the actual periodic deformation amount;

[0043] Determine whether the value of the second loss function converges; if so, the training of the periodic deformation prediction model to be trained is completed, and the preset periodic deformation prediction model is obtained; if not, optimize the hyperparameters in the periodic deformation prediction model according to the second particle swarm optimization algorithm, and then continue to train the periodic deformation prediction model to be trained.

[0044] Furthermore, the hyperparameters in the above-mentioned periodic deformation prediction model are optimized according to the second particle swarm optimization algorithm, including:

[0045] Repeat the second parameter optimization operation until the difference between the current second fitness and the previous second fitness is less than the preset threshold, thereby obtaining the optimized hyperparameters of the periodic deformation prediction model;

[0046] The second parameter optimization operation includes:

[0047] Obtain the current second hyperparameter to be optimized, the current second hyperparameter position, the current second hyperparameter speed, the current second hyperparameter individual optimal position, and the current second hyperparameter overall optimal position of the periodic deformation prediction model; wherein, the initial second hyperparameter to be optimized is the preset initial second hyperparameter, the initial second hyperparameter position and the second hyperparameter speed are the random hyperparameter position and the random hyperparameter speed, the initial first hyperparameter individual optimal position is the preset first hyperparameter individual optimal position, and the initial first hyperparameter overall optimal position is the preset first hyperparameter overall optimal position;

[0048] Under the current second hyperparameter speed and the second hyperparameter position, according to the current second periodic deformation prediction amount and the current actual periodic deformation amount, a second fitness of the current second hyperparameter to be optimized is calculated;

[0049] Calculate the second fitness difference between the current second fitness and the previous second fitness. If the current second fitness difference is not less than the preset threshold, for each second hyperparameter to be optimized, calculate the updated second hyperparameter position and second hyperparameter speed based on the current second hyperparameter position, the current second hyperparameter speed, the current second hyperparameter individual optimal position, and the current second hyperparameter overall optimal position.

[0050] Determine the difference between the current second fitness and the previous second fitness;

[0051] If the current second fitness is less than the previous second fitness, the updated second hyperparameter position will be used as the updated second hyperparameter individual optimal position, and then the second hyperparameter overall optimal position will be generated based on the updated second hyperparameter individual optimal positions of all second hyperparameters to be optimized; otherwise, the current second hyperparameter individual optimal position will be used as the next second hyperparameter individual optimal position, and the current second hyperparameter overall optimal position will be used as the next second hyperparameter overall optimal position.

[0052] Based on the above method embodiment, the present invention provides a corresponding device embodiment;

[0053] The present invention provides a landslide deformation prediction device based on variational modal decomposition, comprising:

[0054] Landslide deformation data acquisition module, deformation modal component and center frequency value determination module, periodic item deformation data acquisition module, trend deformation prediction module, periodic deformation prediction module and landslide deformation value determination module;

[0055] The landslide deformation data acquisition module is used to acquire current landslide deformation data and normalize the landslide deformation data to obtain target landslide deformation data; wherein the landslide deformation data includes: landslide cumulative deformation data and landslide deformation trend influencing factor data;

[0056] The above-mentioned deformation modal component and center frequency value determination module is used to perform Hilbert transform on the above-mentioned target landslide deformation data to obtain a landslide deformation source signal, and perform variational modal decomposition on the above-mentioned landslide deformation source signal to obtain a preset number of deformation modal components and the center frequency value corresponding to each deformation modal component;

[0057] The periodic deformation data acquisition module is used to cluster all center frequency values ​​and use the deformation modal components corresponding to each center frequency value in the center frequency data with the highest center frequency value as the periodic deformation data;

[0058] The trend deformation prediction module is used to input the target landslide deformation data into a preset trend deformation prediction model, so that the preset trend deformation prediction model obtains a trend deformation prediction value based on the target landslide deformation data;

[0059] The cyclic deformation prediction module is used to input the cyclic deformation data and the landslide deformation trend influencing factor data into a preset cyclic deformation prediction model, so that the preset cyclic deformation prediction model obtains a cyclic deformation prediction amount based on the cyclic deformation data and the landslide deformation trend influencing factor data;

[0060] The landslide deformation value determination module is used to calculate the sum of the trend deformation prediction value and the periodic deformation prediction value to obtain the landslide deformation prediction value.

[0061] Based on the above method embodiment, the present invention provides a corresponding terminal device embodiment;

[0062] The present invention provides a terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the landslide deformation prediction method based on variational modal decomposition according to any embodiment of the present invention is implemented.

[0063] Based on the above method embodiment, the present invention provides a storage medium embodiment;

[0064] The present invention provides a storage medium comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the landslide deformation prediction method based on variational modal decomposition according to any embodiment of the present invention is implemented.

[0065] The embodiments of the present invention have the following beneficial effects:

[0066] The present invention provides a landslide deformation prediction method based on variational modal decomposition. The above method first obtains the current landslide deformation data, and normalizes the above landslide deformation data to obtain target landslide deformation data; wherein the above landslide deformation data includes: landslide cumulative deformation data and landslide deformation trend influencing factor data; then the above target landslide deformation data is subjected to Hilbert transformation to obtain a landslide deformation source signal, and the above landslide deformation source signal is subjected to variational modal decomposition to obtain a preset number of deformation modal components and the center frequency value corresponding to each deformation modal component; then all the center frequency values ​​are clustered, and the shape modal components corresponding to each center frequency value in the center frequency data with the highest center frequency value are clustered. The present invention performs variational modal decomposition on the landslide deformation data, and the obtained deformation modal components reduce the data noise in the original landslide deformation data, enhance the clarity of the original landslide deformation data, and then obtain the effective data part in the original landslide deformation data, so that the model can more accurately identify and extract the key spatial variation feature information in the original landslide deformation data based on the effective data part. Then, based on the periodic deformation data obtained by variational modal decomposition, the periodic deformation of the landslide is predicted in a preset periodic deformation prediction model. Finally, the final landslide deformation prediction value is obtained based on the sum of the periodic deformation prediction amount and the trend deformation prediction amount obtained according to another preset trend deformation prediction model, thereby improving the accuracy of the landslide deformation prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 The figure is a flow chart of a landslide deformation prediction method based on variational modal decomposition provided by one embodiment of the present invention.

[0068] Figure 2 2 is a schematic diagram of the training process of the trend deformation prediction model provided by one embodiment of the present invention.

[0069] Figure 3 2 is a schematic diagram of the training process of the periodic deformation prediction model provided by one embodiment of the present invention.

[0070] Figure 4 The figure is a schematic structural diagram of a landslide deformation prediction device based on variational modal decomposition provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0071] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0072] like Figure 1 As shown, an embodiment of the present invention provides a landslide deformation prediction method based on variational mode decomposition, comprising:

[0073] Step S101: obtaining current landslide deformation data and normalizing the landslide deformation data to obtain target landslide deformation data; wherein the landslide deformation data includes: landslide cumulative deformation data and landslide deformation trend influencing factor data;

[0074] Specifically, the above landslide deformation data is a time series data set.

[0075] In a preferred embodiment, the above-mentioned steps of obtaining current landslide deformation data and normalizing the above-mentioned landslide deformation data to obtain target landslide deformation data include:

[0076] Extracting minimum landslide deformation data and maximum landslide deformation data from the above landslide deformation data;

[0077] According to the above minimum landslide deformation data, the above maximum landslide deformation data and the above landslide deformation data, the above target landslide deformation data is calculated by the following formula:

[0078]

[0079] Where y represents the target landslide deformation data, x represents the landslide deformation data, min represents the minimum landslide deformation data, and max represents the maximum landslide deformation data.

[0080] Specifically, after normalization, the original landslide deformation data can be mapped to the range of 0 to 1 to facilitate normalization processing.

[0081] Preferably, the normalization method used here is the maximum-minimum normalization method, which can avoid the subsequent slowdown of the model's learning speed as a large amount of data is input, and promote the convergence of the model in the deep learning process.

[0082] In this preferred embodiment, the acquired landslide deformation data is normalized to obtain target landslide deformation data.

[0083] Step S102: performing Hilbert transform on the target landslide deformation data to obtain a landslide deformation source signal, and performing variational modal decomposition on the landslide deformation source signal to obtain a preset number of deformation modal components and a center frequency value corresponding to each deformation modal component;

[0084] Specifically, variational mode decomposition (VMD) first determines the number of modal decompositions (i.e., the preset number) for the target landslide deformation data based on actual conditions. It then searches for and solves for the optimal center frequency and finite bandwidth of each deformation modal component, effectively separating the modal components and obtaining the effective deformation modal components of the target landslide deformation data, ultimately achieving the optimal solution to the variational problem.

[0085] In a preferred embodiment, variational modal decomposition is performed on the landslide deformation source signal to obtain a preset number of deformation modal components and the center frequency value corresponding to each deformation modal component, including:

[0086] The augmented Lagrangian function is constructed based on the above landslide deformation source signal;

[0087] Iteratively solving the augmented Lagrangian function to obtain a preset number of first deformation modal components;

[0088] After each first deformation modal component is obtained, determining whether the first deformation modal component satisfies a preset iteration stop condition; if so, determining the current first deformation modal component as the deformation modal component, and determining the first center frequency value corresponding to each current first deformation modal component as the center frequency value;

[0089] Among them, the above preset iteration stopping condition is:

[0090]

[0091] Where, represents the kth first deformation modal component obtained after the nth iteration, represents the kth first deformation modal component obtained after the n+1th iteration, and ε represents the preset accuracy.

[0092] Specifically, searching and solving the optimal center frequency and limited bandwidth of the deformation modal component is achieved by constructing and solving the variational problem. The constructed constrained variational problem is specifically described as: finding the modal function that minimizes the sum of the estimated bandwidth of each deformation modal component. The constraint condition is that the sum of each deformation modal component is equal to the input signal, which is specifically expressed by the following formula:

[0093]

[0094] Where u krepresents the kth deformation modal component, ω k represents the center frequency value of the kth deformation mode component, δ(t) represents the Diktala function at time t, u k (t) represents the kth deformation modal component at time t, K represents the number of deformation modal components, f represents the above input signal, (δ(t)+j / πt)*u k (t) represents the analytical signal for solving the landslide deformation source signal.

[0095] Specifically, by introducing the Lagrangian penalty operator and penalty coefficient, the above constrained variational problem is transformed into an unconstrained variational problem, and the augmented Lagrangian function expression is obtained:

[0096]

[0097] Wherein, λ represents the penalty operator, α represents the penalty coefficient, f(t) represents the function of the above input signal, and λ(t) represents the function of the penalty operator.

[0098] Specifically, the goal of variational modal decomposition is to separate the deformation modal components from the landslide deformation source signal. The specific method is to use the multiplier alternating inverse method to solve the augmented Lagrangian function. The center frequency value and bandwidth of each deformation modal component are continuously updated iteratively until the iteration stop condition is met. The deformation modal component obtained in each iterative update is shown in the following formula:

[0099]

[0100] Where, represents the kth deformation modal component of the n+1th iteration after Fourier transform at random frequency, represents the signal function after Fourier transform at random frequency, represents the i-th deformation modal component function after Fourier transform at random frequency, represents the penalty operator function after Fourier transform at random frequency, ω represents the random frequency, ω k Represents the center frequency value of the kth deformation mode component.

[0101] Specifically, the center frequency value of the deformation modal component obtained by each iterative update is shown in the following formula:

[0102]

[0103] Where, represents the center frequency value of the kth deformation modal component after the n+1th iteration, represents the deformation mode function after Fourier transform at random frequencies.

[0104] In this preferred embodiment, variational modal decomposition is performed on the landslide deformation source signal to determine the deformation modal components and the center frequency values ​​corresponding to the deformation modal components.

[0105] Step S103: clustering all center frequency values, and taking the deformation modal components corresponding to each center frequency value in the center frequency data with the highest center frequency value as periodic deformation data;

[0106] Specifically, after clustering all center frequency values, two groups of center frequency data are obtained, and the deformation modal components corresponding to each center frequency value in the group of center frequency data with the highest center frequency value in the two groups are taken as periodic deformation data.

[0107] Preferably, a clear frequency interval can be intuitively obtained by plotting a histogram of all center frequency values, and then the deformation modal component corresponding to the center frequency on the right side of the interval is used as the periodic deformation data.

[0108] Step S104: inputting the target landslide deformation data into a preset trend deformation prediction model, so that the preset trend deformation prediction model obtains a trend deformation prediction value based on the target landslide deformation data;

[0109] Specifically, the preset trend deformation prediction model is a VMD-LSTM-Attention network model. This network model includes an input layer, an LSTM layer, an Attention layer, a Dropout layer, a fully connected layer, and an output layer.

[0110] In a preferred embodiment, the training of the preset trend deformation prediction model includes:

[0111] Acquire a number of sample landslide deformation data, and normalize the sample landslide deformation data to obtain target sample landslide deformation data; wherein the sample landslide deformation data includes: sample landslide cumulative deformation data and sample landslide deformation trend influencing factor data;

[0112] Performing Hilbert transform on the target sample landslide deformation data to obtain a sample landslide deformation source signal, and performing variational modal decomposition on the sample landslide deformation source signal to obtain a preset number of sample deformation modal components and a sample center frequency value corresponding to each sample deformation modal component;

[0113] Clustering all the sample center frequency values ​​to obtain the first sample center frequency with the lowest center frequency value, and taking the sample deformation modal components corresponding to each center frequency value within the first sample center frequency as the sample trend item deformation data;

[0114] Obtaining a first true label corresponding to the sample trend item deformation data, and inputting the sample trend item deformation data with the first true label into the trend deformation prediction model to be trained, so that the trend deformation model to be trained predicts a first trend deformation prediction amount based on the sample trend item deformation data; wherein the first true label is the actual trend deformation amount of the sample trend item deformation data;

[0115] Calculate a first loss function value based on the first trend deformation prediction amount and the actual trend deformation amount;

[0116] Determine whether the value of the first loss function converges; if so, the training of the trend deformation prediction model to be trained is completed, and the preset trend deformation prediction model is obtained; if not, optimize the hyperparameters in the trend deformation prediction model according to the first particle swarm optimization algorithm, and then continue to train the trend deformation prediction model to be trained.

[0117] Specifically, the RMSE (root mean square error), MSE (mean square error), MAPE (mean absolute percentage error) and SMAPE (symmetric mean absolute percentage error) are used to calculate the value of the first loss function to evaluate the performance of the trend deformation prediction model. If the evaluation result cannot meet the accuracy requirements under actual conditions (that is, the first loss function does not converge), the hyperparameters in the model are optimized according to the first particle swarm optimization algorithm.

[0118] Specifically, during the initial training, through experiments and investigations, a reasonable value range is set for each hyperparameter of the trend deformation prediction model to be trained, and then training is carried out.

[0119] Specifically, the hyperparameters that need to be optimized in the above trend deformation prediction model are the number of LSTM memory units, optimizer type, batch size, number of iterations, and learning rate.

[0120] In this preferred embodiment, the trend deformation prediction model is trained until the calculated first loss function value converges, thereby obtaining a trained trend deformation prediction model.

[0121] In another preferred embodiment, the hyperparameters in the trend deformation prediction model are optimized according to a first particle swarm optimization algorithm, including:

[0122] Repeating the first parameter optimization operation until the difference between the current first fitness and the previous first fitness is less than a preset threshold, thereby obtaining the optimized hyperparameters of the trend deformation prediction model;

[0123] The first parameter optimization operation includes:

[0124] Obtain the current first hyperparameter to be optimized, the current first hyperparameter position, the current first hyperparameter speed, the current first hyperparameter individual optimal position, and the current first hyperparameter overall optimal position of the trend deformation prediction model; wherein, the initial first hyperparameter to be optimized is the preset initial first hyperparameter, the initial first hyperparameter position and the first hyperparameter speed are the random hyperparameter position and the random hyperparameter speed, the initial first hyperparameter individual optimal position is the preset first hyperparameter individual optimal position, and the initial first hyperparameter overall optimal position is the preset first hyperparameter overall optimal position;

[0125] Under the current first hyperparameter speed and the first hyperparameter position, according to the current first trend deformation prediction amount and the current actual trend deformation amount, a first fitness of the current first hyperparameter to be optimized is calculated;

[0126] Calculate the first fitness difference between the current first fitness and the previous first fitness. If the current first fitness difference is not less than a preset threshold, for each first hyperparameter to be optimized, calculate the updated first hyperparameter position and first hyperparameter speed based on the current first hyperparameter position, the current first hyperparameter speed, the current first hyperparameter individual optimal position, and the current first hyperparameter overall optimal position;

[0127] Determine the difference between the current first fitness and the previous first fitness;

[0128] If the current first fitness is less than the previous first fitness, the updated first hyperparameter position is used as the updated first hyperparameter individual optimal position, and then the first hyperparameter overall optimal position is generated based on the updated first hyperparameter individual optimal positions of all first hyperparameters to be optimized; otherwise, the current first hyperparameter individual optimal position is used as the next first hyperparameter individual optimal position, and the current first hyperparameter overall optimal position is used as the next first hyperparameter overall optimal position.

[0129] Schematically, the entire training process of the trend deformation prediction model is as follows Figure 2 shown.

[0130] Specifically, the root mean square error (RMSE) is used as the fitness function, and the first fitness value of each hyperparameter is obtained. The smaller the RMSE value, the higher the prediction accuracy of the model. The first fitness is calculated according to the following formula:

[0131]

[0132] Where y1 represents the actual trend deformation variable corresponding to the deformation data of the trend item of the i-th sample, represents the first trend deformation prediction amount corresponding to the i-th sample trend item deformation data, and N represents the number of sample trend item deformation data.

[0133] Specifically, the updated first hyperparameter position and the updated first hyperparameter speed are calculated according to the following formula:

[0134] v t+1 =ω1v t +c1r1(x pbest -x t )+c2r2(x pbest -x t )

[0135] x t+1 =x t +v t+1

[0136] Where, v t+1 represents the first hyperparameter velocity at time t+1 after update, ω1 represents the inertia weight factor, and v t represents the first hyperparameter speed at the current time t, r1 and r2 represent two random numbers uniformly distributed in the range [0,1], x pbest Indicates the current optimal position of the first hyperparameter individual, x t Indicates the first hyperparameter position at the current time t, c1 and c2 both represent acceleration coefficients, x t+1 Represents the first hyperparameter position at time t+1 after update.

[0137] Specifically, when the current first fitness is less than the previous first fitness, the updated first hyperparameter position is used as the updated first hyperparameter individual optimal position, and then the updated first hyperparameter overall optimal position is generated based on all first hyperparameter individual optimal positions; when the current first fitness is not less than the previous first fitness, the current first hyperparameter individual optimal position and the current first hyperparameter overall optimal position are kept unchanged and used for the next iterative calculation.

[0138] Preferably, the above preset threshold is 0.001.

[0139] Preferably, by introducing the first particle swarm optimization algorithm to repeatedly iterate the hyperparameters in the trend deformation prediction model, the trend deformation prediction model can learn relevant data features.

[0140] In this preferred embodiment, the hyperparameters in the trend deformation prediction model are optimized according to the first particle swarm optimization algorithm.

[0141] Step S105: inputting the periodic deformation data and the landslide deformation trend influencing factor data into a preset periodic deformation prediction model, so that the preset periodic deformation prediction model obtains a periodic deformation prediction amount based on the periodic deformation data and the landslide deformation trend influencing factor data;

[0142] Specifically, the preset periodic deformation prediction model is a VMD-LSTM-Attention network model.

[0143] In a preferred embodiment, the training of the preset periodic deformation prediction model includes:

[0144] Clustering all the sample center frequency values ​​to obtain the second sample center frequency with the highest center frequency value, and taking the sample deformation modal components corresponding to each center frequency value within the second sample center frequency as the sample periodic item deformation data;

[0145] Obtaining a second true label corresponding to the sample periodic item deformation data, and inputting the sample periodic item deformation data with the second true label into the periodic deformation prediction model to be trained, so that the periodic deformation model to be trained predicts a first periodic deformation prediction amount based on the sample periodic item deformation data; wherein the second true label is the actual periodic deformation amount of the sample periodic item deformation data;

[0146] Calculating a second loss function value based on the first periodic deformation prediction amount and the actual periodic deformation amount;

[0147] Determine whether the value of the second loss function converges; if so, the training of the periodic deformation prediction model to be trained is completed, and the preset periodic deformation prediction model is obtained; if not, optimize the hyperparameters in the periodic deformation prediction model according to the second particle swarm optimization algorithm, and then continue to train the periodic deformation prediction model to be trained.

[0148] Specifically, the second loss function value is calculated using RMSE (root mean square error), MSE (mean square error), MAPE (mean absolute percentage error) and SMAPE (symmetric mean absolute percentage error) to evaluate the performance of the periodic deformation prediction model. If the evaluation result cannot meet the accuracy requirements under actual conditions (i.e., the second loss function value does not converge), the hyperparameters in the model are optimized according to the second particle swarm optimization algorithm.

[0149] Specifically, during the initial training, through experiments and investigations, a reasonable value range is set for each hyperparameter of the periodic deformation prediction model to be trained, and then training is performed.

[0150] Specifically, similar to the trend deformation prediction model, the hyperparameters that need to be optimized in the above period deformation prediction model are the number of LSTM memory units, optimizer type, batch size, number of iterations, and learning rate.

[0151] In this preferred embodiment, the trained periodic deformation prediction model is obtained by training the periodic deformation prediction model until the calculated second loss function value converges.

[0152] In another preferred embodiment, the hyperparameters in the above-mentioned periodic deformation prediction model are optimized according to a second particle swarm optimization algorithm, including:

[0153] Repeat the second parameter optimization operation until the difference between the current second fitness and the previous second fitness is less than the preset threshold, thereby obtaining the optimized hyperparameters of the periodic deformation prediction model;

[0154] The second parameter optimization operation includes:

[0155] Obtain the current second hyperparameter to be optimized, the current second hyperparameter position, the current second hyperparameter speed, the current second hyperparameter individual optimal position, and the current second hyperparameter overall optimal position of the periodic deformation prediction model; wherein, the initial second hyperparameter to be optimized is the preset initial second hyperparameter, the initial second hyperparameter position and the second hyperparameter speed are the random hyperparameter position and the random hyperparameter speed, the initial first hyperparameter individual optimal position is the preset first hyperparameter individual optimal position, and the initial first hyperparameter overall optimal position is the preset first hyperparameter overall optimal position;

[0156] Under the current second hyperparameter speed and the second hyperparameter position, according to the current second periodic deformation prediction amount and the current actual periodic deformation amount, a second fitness of the current second hyperparameter to be optimized is calculated;

[0157] Calculate the second fitness difference between the current second fitness and the previous second fitness. If the current second fitness difference is not less than the preset threshold, for each second hyperparameter to be optimized, calculate the updated second hyperparameter position and second hyperparameter speed based on the current second hyperparameter position, the current second hyperparameter speed, the current second hyperparameter individual optimal position, and the current second hyperparameter overall optimal position.

[0158] Determine the difference between the current second fitness and the previous second fitness;

[0159] If the current second fitness is less than the previous second fitness, the updated second hyperparameter position will be used as the updated second hyperparameter individual optimal position, and then the second hyperparameter overall optimal position will be generated based on the updated second hyperparameter individual optimal positions of all second hyperparameters to be optimized; otherwise, the current second hyperparameter individual optimal position will be used as the next second hyperparameter individual optimal position, and the current second hyperparameter overall optimal position will be used as the next second hyperparameter overall optimal position.

[0160] Schematically, the entire training process of the periodic deformation prediction model is as follows Figure 3 shown.

[0161] Specifically, similar to the first particle swarm optimization algorithm, in the second particle swarm optimization algorithm, the root mean square error (RMSE) is used as the fitness function, and the updated second hyperparameter position and second hyperparameter speed are calculated at the same time. The calculation formula is similar to the formula for calculating the updated first hyperparameter position and first hyperparameter speed, except that in the second particle swarm optimization algorithm, the data substituted are the corresponding current second hyperparameter position, current second hyperparameter speed, current second hyperparameter individual optimal position, and current second hyperparameter overall optimal position.

[0162] Specifically, when the current second fitness is less than the previous second fitness, the updated second hyperparameter position is used as the updated second hyperparameter individual optimal position, and then the updated second hyperparameter overall optimal position is generated based on all second hyperparameter individual optimal positions; when the current second fitness is not less than the previous second fitness, the current second hyperparameter individual optimal position and the current second hyperparameter overall optimal position are kept unchanged and used for the calculation of the next iteration.

[0163] Preferably, by introducing the second particle swarm optimization algorithm to repeatedly iterate the hyperparameters in the periodic deformation prediction model, the periodic deformation prediction model can learn relevant data features.

[0164] In this preferred embodiment, the hyperparameters in the periodic deformation prediction model are optimized according to the second particle swarm optimization algorithm.

[0165] Step S106: Calculate the sum of the trend deformation prediction value and the periodic deformation prediction value to obtain a landslide deformation prediction value.

[0166] Specifically, after respectively predicting the trend deformation prediction amount and the periodic deformation prediction amount using the preset trend deformation prediction model and the preset periodic deformation prediction model, the sum of the two prediction amounts is calculated to obtain the landslide deformation prediction value.

[0167] In a preferred embodiment, in order to fully verify the effectiveness and practicality of the present invention, a series of experiments were designed and implemented. The VMD-LSTM-Attention network model algorithm optimized by particle swarm optimization was designed using Matlab programming language and the deformation monitoring data of the Faer landslide was used for specific implementation.

[0168] Specifically, the Faer landslide is characterized by high southeast and low northwest sides, belonging to the low-medium mountain to medium-low mountain landform formed by tectonic erosion, with the central geographic coordinates of 104°44′11″E, 25°18′20″N. Two benchmark points were set in the stable area of ​​the landslide, and 11 monitoring points were set in the deformation area. The cumulative deformation data of the landslide on 2022-01-01-05-20, one of the GNSS monitoring points in the deformation area, was selected. The data on the influencing factors of the landslide deformation trend was obtained from the local meteorological station. The data included soil moisture, daily precipitation, cumulative precipitation, etc. A total of 141 days of monitoring data were selected for landslide deformation prediction. The number of LSTM network layers was set to 1, the number of hidden layer nodes was 20, the batch size was 512, the number of model iterations was 150, and the initial learning rate was set to 0.0001.

[0169] Specifically, in the experimental design, three representative prediction models were selected for comparison: a single long short-term memory (LSTM) model, a support vector regression (SVR) model, and a VMD combined with an extreme learning machine (ELM) model. These models have achieved certain results in their respective fields, but their performance is often limited by various factors when faced with the complex task of landslide deformation prediction.

[0170] Specifically, while a single LSTM model can process time series data, it still needs improvement in capturing long-term dependencies and spatial features. The SVR model performs well in handling nonlinear relationships, but may face challenges in computational efficiency and generalization when dealing with high-dimensional data and complex time series. And while the VMD-ELM model combines the advantages of signal decomposition and rapid learning, it still has certain limitations in capturing subtle spatial variations in time series. To comprehensively evaluate the performance of each model, the root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) were used as evaluation metrics. These metrics intuitively reflect the degree of discrepancy between the model's predictions and the actual observations, thereby accurately assessing the model's prediction accuracy and reliability.

[0171] Specifically, experimental results demonstrate that the landslide deformation prediction method based on variational mode decomposition (VMD) provided by the present invention performs well in landslide deformation risk prediction. The model's RMSE, MAE, and MAPE are significantly lower than those of other comparison models, indicating a smaller error between the predicted results and the actual observed values ​​and higher prediction accuracy. Compared with the traditional single LSTM model, this method achieves significant improvements in prediction accuracy, primarily due to the introduction of the VMD method and the application of the attention mechanism. The VMD method effectively extracts key feature information from landslide deformation sequence data, while the attention mechanism dynamically adjusts the contribution of different time steps and spatial locations to the prediction results, thereby more accurately capturing the spatial variation feature information in the time series. Furthermore, the optimization effect of the particle swarm algorithm (PSO) cannot be ignored. As a global optimization algorithm, the PSO can efficiently search for the optimal network hyperparameter configuration, further improving the model's generalization and prediction performance. By continuously iterating and updating the positions and velocities of particles, the PSO enables the trend deformation prediction model and the periodic deformation prediction model in this method to better adapt to the complexity and nonlinear characteristics of landslide deformation data. In summary, the comparative experimental results show that applying this method to landslide deformation risk prediction not only increases reliability but also significantly improves prediction accuracy, providing new ideas and methods for solving the difficult problems of landslide deformation prediction. This research result is of great significance for the early warning and prevention of landslide disasters and is expected to provide useful reference and inspiration for research and practice in related fields.

[0172] Based on the above method embodiments, the present invention provides corresponding device embodiments.

[0173] like Figure 4 As shown, an embodiment of the present invention provides a landslide deformation prediction device based on variational modal decomposition, comprising:

[0174] Landslide deformation data acquisition module, deformation modal component and center frequency value determination module, periodic item deformation data acquisition module, trend deformation prediction module, periodic deformation prediction module and landslide deformation value determination module;

[0175] The landslide deformation data acquisition module is used to acquire current landslide deformation data and normalize the landslide deformation data to obtain target landslide deformation data; wherein the landslide deformation data includes: landslide cumulative deformation data and landslide deformation trend influencing factor data;

[0176] The above-mentioned deformation modal component and center frequency value determination module is used to perform Hilbert transform on the above-mentioned target landslide deformation data to obtain a landslide deformation source signal, and perform variational modal decomposition on the above-mentioned landslide deformation source signal to obtain a preset number of deformation modal components and the center frequency value corresponding to each deformation modal component;

[0177] The periodic deformation data acquisition module is used to cluster all center frequency values ​​and use the deformation modal components corresponding to each center frequency value in the center frequency data with the highest center frequency value as the periodic deformation data;

[0178] The trend deformation prediction module is used to input the target landslide deformation data into a preset trend deformation prediction model, so that the preset trend deformation prediction model obtains a trend deformation prediction value based on the target landslide deformation data;

[0179] The cyclic deformation prediction module is used to input the cyclic deformation data and the landslide deformation trend influencing factor data into a preset cyclic deformation prediction model, so that the preset cyclic deformation prediction model obtains a cyclic deformation prediction amount based on the cyclic deformation data and the landslide deformation trend influencing factor data;

[0180] The landslide deformation value determination module is used to calculate the sum of the trend deformation prediction value and the periodic deformation prediction value to obtain the landslide deformation prediction value.

[0181] It should be noted that the device embodiments described above are merely illustrative, wherein the modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected to achieve the purpose of the present embodiment according to actual needs. Furthermore, in the drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which may be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement the present invention without inventive effort. The schematic diagram above is merely an example of a landslide deformation prediction device based on variational modal decomposition and does not constitute a limitation on a landslide deformation prediction device based on variational modal decomposition. The present invention may include more or fewer components than shown, or combine certain components, or have different components.

[0182] Based on the above method embodiment, the present invention provides a corresponding terminal device embodiment.

[0183] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the landslide deformation prediction method based on variational modal decomposition described in any embodiment of the present invention is implemented.

[0184] For example, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution process of the computer program in the device.

[0185] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, or a cloud server. The device may include, but is not limited to, a processor and a memory;

[0186] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the device, connecting the various parts of the device using various interfaces and lines.

[0187] The above-mentioned memory can be used to store the above-mentioned computer programs and / or modules. The above-mentioned processor realizes various functions of the above-mentioned device by running or executing the computer programs and / or modules stored in the above-mentioned memory, and calling the data stored in the memory. The above-mentioned memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; in addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0188] Based on the above method embodiment, the present invention provides a corresponding storage medium embodiment.

[0189] Another embodiment of the present invention provides a storage medium, which includes a stored computer program. When the computer program is executed, the device where the storage medium is located is controlled to execute the landslide deformation prediction method based on variational modal decomposition according to any embodiment of the present invention.

[0190] In this embodiment, the storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in source code form, object code form, an executable file, or some intermediate form. The computer-readable medium may include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunications signal, and a software distribution medium.

[0191] Compared with the prior art, by implementing the above-mentioned embodiments of the present invention, the data noise in the original landslide deformation data can be reduced, the clarity of the original landslide deformation data can be enhanced, and the valid data part of the original landslide deformation data can be obtained, so that the model can more accurately identify and extract the key spatial change feature information in the original landslide deformation data based on the valid data part, thereby ultimately improving the accuracy of the deformation prediction results.

[0192] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A landslide deformation prediction method based on variational modal decomposition, characterized in that: include: Acquire current landslide deformation data and normalize the landslide deformation data to obtain target landslide deformation data; wherein the landslide deformation data includes: landslide cumulative deformation data and landslide deformation trend influencing factor data; Performing Hilbert transform on the target landslide deformation data to obtain a landslide deformation source signal, and performing variational modal decomposition on the landslide deformation source signal to obtain a preset number of deformation modal components and a center frequency value corresponding to each deformation modal component; All center frequency values ​​are clustered, and the deformation modal components corresponding to each center frequency value in the center frequency data with the highest center frequency value are taken as periodic deformation data; Inputting the target landslide deformation data into a preset trend deformation prediction model, so that the preset trend deformation prediction model obtains a trend deformation prediction amount according to the target landslide deformation data; Inputting the periodic deformation data and the landslide deformation trend influencing factor data into a preset periodic deformation prediction model, so that the preset periodic deformation prediction model obtains a periodic deformation prediction amount according to the periodic deformation data and the landslide deformation trend influencing factor data; The training of the preset periodic deformation prediction model includes: Clustering all the sample center frequency values ​​to obtain a second sample center frequency with the highest center frequency value, and using the sample deformation modal components corresponding to each center frequency value within the second sample center frequency as sample periodic item deformation data; Obtaining a second true label corresponding to the sample periodic deformation data, and inputting the sample periodic deformation data with the second true label into a periodic deformation prediction model to be trained, so that the periodic deformation model to be trained predicts a first periodic deformation prediction amount based on the sample periodic deformation data; wherein the second true label is the actual periodic deformation amount of the sample periodic deformation data; Calculate a second loss function value based on the first periodic deformation prediction amount and the actual periodic deformation amount; determine whether the second loss function value converges; if so, the training of the periodic deformation prediction model to be trained is completed, and the preset periodic deformation prediction model is obtained; if not, optimize the hyperparameters in the periodic deformation prediction model according to the second particle swarm optimization algorithm, and then continue to train the periodic deformation prediction model to be trained; The sum of the trend deformation prediction amount and the periodic deformation prediction amount is calculated to obtain a landslide deformation prediction value.

2. The landslide deformation prediction method based on variational modal decomposition according to claim 1 is characterized in that: The acquiring of current landslide deformation data and normalizing the landslide deformation data to obtain target landslide deformation data includes: Extracting minimum landslide deformation data and maximum landslide deformation data from the landslide deformation data; The target landslide deformation data is calculated according to the minimum landslide deformation data, the maximum landslide deformation data and the landslide deformation data using the following formula: Where y represents the target landslide deformation data, x represents the landslide deformation data, min represents the minimum landslide deformation data, and max represents the maximum landslide deformation data.

3. The landslide deformation prediction method based on variational modal decomposition according to claim 2 is characterized in that: Performing variational modal decomposition on the landslide deformation source signal to obtain a preset number of deformation modal components and a center frequency value corresponding to each deformation modal component, including: An augmented Lagrangian function is constructed according to the landslide deformation source signal; Iteratively solving the augmented Lagrangian function to obtain a preset number of first deformation modal components; After each first deformation modal component is obtained, determining whether the first deformation modal component satisfies a preset iteration stop condition; if so, determining the current first deformation modal component as the deformation modal component, and determining the first center frequency value corresponding to each current first deformation modal component as the center frequency value; Wherein, the preset iteration stopping condition is: Where, represents the kth first deformation modal component obtained after the nth iteration, represents the kth first deformation modal component obtained after the n+1th iteration, Indicates the preset precision.

4. The landslide deformation prediction method based on variational modal decomposition according to claim 3 is characterized in that: The training of the preset trend deformation prediction model includes: Acquire a number of sample landslide deformation data, and normalize the sample landslide deformation data to obtain target sample landslide deformation data; wherein the sample landslide deformation data includes: sample landslide cumulative deformation data and sample landslide deformation trend influencing factor data; Performing Hilbert transform on the target sample landslide deformation data to obtain a sample landslide deformation source signal, and performing variational modal decomposition on the sample landslide deformation source signal to obtain a preset number of sample deformation modal components and a sample center frequency value corresponding to each sample deformation modal component; Clustering all sample center frequency values ​​to obtain a first sample center frequency with the lowest center frequency value, and using the sample deformation modal components corresponding to each center frequency value within the first sample center frequency as sample trend item deformation data; Obtaining a first true label corresponding to the sample trend item deformation data, and inputting the sample trend item deformation data with the first true label into a trend deformation prediction model to be trained, so that the trend deformation model to be trained predicts a first trend deformation prediction amount based on the sample trend item deformation data; wherein the first true label is an actual trend deformation amount of the sample trend item deformation data; Calculating a first loss function value based on the first trend deformation prediction amount and the actual trend deformation amount; Determine whether the first loss function value converges; if so, the training of the trend deformation prediction model to be trained is completed, and the preset trend deformation prediction model is obtained; if not, optimize the hyperparameters in the trend deformation prediction model according to the first particle swarm optimization algorithm, and continue to train the trend deformation prediction model to be trained.

5. The landslide deformation prediction method based on variational modal decomposition according to claim 4 is characterized in that: Optimizing hyperparameters in the trend deformation prediction model according to a first particle swarm optimization algorithm includes: Repeating the first parameter optimization operation until the difference between the current first fitness and the previous first fitness is less than a preset threshold, thereby obtaining the optimized hyperparameters of the trend deformation prediction model; The first parameter optimization operation includes: Obtain the current first hyperparameter to be optimized, the current first hyperparameter position, the current first hyperparameter speed, the current first hyperparameter individual optimal position, and the current first hyperparameter overall optimal position of the trend deformation prediction model; wherein, the initial first hyperparameter to be optimized is the preset initial first hyperparameter, the initial first hyperparameter position and the first hyperparameter speed are the random hyperparameter position and the random hyperparameter speed, the initial first hyperparameter individual optimal position is the preset first hyperparameter individual optimal position, and the initial first hyperparameter overall optimal position is the preset first hyperparameter overall optimal position; Under the current first hyperparameter speed and the first hyperparameter position, according to the current first trend deformation prediction amount and the current actual trend deformation amount, a first fitness of the current first hyperparameter to be optimized is calculated; Calculate the first fitness difference between the current first fitness and the previous first fitness. If the current first fitness difference is not less than a preset threshold, for each first hyperparameter to be optimized, calculate the updated first hyperparameter position and first hyperparameter speed based on the current first hyperparameter position, the current first hyperparameter speed, the current first hyperparameter individual optimal position, and the current first hyperparameter overall optimal position; Determine the difference between the current first fitness and the previous first fitness; If the current first fitness is less than the previous first fitness, the updated first hyperparameter position is used as the updated first hyperparameter individual optimal position, and then the first hyperparameter overall optimal position is generated based on the updated first hyperparameter individual optimal positions of all first hyperparameters to be optimized; otherwise, the current first hyperparameter individual optimal position is used as the next first hyperparameter individual optimal position, and the current first hyperparameter overall optimal position is used as the next first hyperparameter overall optimal position.

6. The landslide deformation prediction method based on variational modal decomposition according to claim 5 is characterized in that: Optimizing the hyperparameters in the periodic deformation prediction model according to the second particle swarm optimization algorithm includes: Repeating the second parameter optimization operation until the second fitness difference between the current second fitness and the previous second fitness is less than the preset threshold, thereby obtaining the optimized hyperparameters of the periodic deformation prediction model; The second parameter optimization operation includes: Obtain the current second hyperparameter to be optimized, the current second hyperparameter position, the current second hyperparameter speed, the current second hyperparameter individual optimal position, and the current second hyperparameter overall optimal position of the periodic deformation prediction model; wherein, the initial second hyperparameter to be optimized is the preset initial second hyperparameter, the initial second hyperparameter position and the second hyperparameter speed are the random hyperparameter position and the random hyperparameter speed, the initial first hyperparameter individual optimal position is the preset first hyperparameter individual optimal position, and the initial first hyperparameter overall optimal position is the preset first hyperparameter overall optimal position; Under the current second hyperparameter speed and the second hyperparameter position, according to the current second periodic deformation prediction amount and the current actual periodic deformation amount, a second fitness of the current second hyperparameter to be optimized is calculated; Calculating a second fitness difference between the current second fitness and the previous second fitness; if the current second fitness difference is not less than the preset threshold, for each second hyperparameter to be optimized, calculating an updated second hyperparameter position and second hyperparameter speed based on the current second hyperparameter position, the current second hyperparameter speed, the current second hyperparameter individual optimal position, and the current second hyperparameter overall optimal position; Determine the difference between the current second fitness and the previous second fitness; If the current second fitness is less than the previous second fitness, the updated second hyperparameter position will be used as the updated second hyperparameter individual optimal position, and then the second hyperparameter overall optimal position will be generated based on the updated second hyperparameter individual optimal positions of all second hyperparameters to be optimized; otherwise, the current second hyperparameter individual optimal position will be used as the next second hyperparameter individual optimal position, and the current second hyperparameter overall optimal position will be used as the next second hyperparameter overall optimal position.

7. A landslide deformation prediction device based on variational modal decomposition, characterized in that: include: Landslide deformation data acquisition module, deformation modal component and center frequency value determination module, periodic item deformation data acquisition module, trend deformation prediction module, periodic deformation prediction module and landslide deformation value determination module; The landslide deformation data acquisition module is used to acquire current landslide deformation data and normalize the landslide deformation data to obtain target landslide deformation data; wherein the landslide deformation data includes: landslide cumulative deformation data and landslide deformation trend influencing factor data; The deformation modal component and center frequency value determination module is used to perform Hilbert transform on the target landslide deformation data to obtain a landslide deformation source signal, and perform variational modal decomposition on the landslide deformation source signal to obtain a preset number of deformation modal components and a center frequency value corresponding to each deformation modal component; The periodic deformation data acquisition module is used to cluster all center frequency values ​​and take the deformation modal components corresponding to each center frequency value in the center frequency data with the highest center frequency value as the periodic deformation data; The trend deformation prediction module is used to input the target landslide deformation data into a preset trend deformation prediction model, so that the preset trend deformation prediction model obtains a trend deformation prediction value according to the target landslide deformation data; The periodic deformation prediction module is configured to input the periodic deformation data and the landslide deformation trend influencing factor data into a preset periodic deformation prediction model, so that the preset periodic deformation prediction model obtains a periodic deformation prediction value based on the periodic deformation data and the landslide deformation trend influencing factor data; wherein the training of the preset periodic deformation prediction model includes: Clustering all the sample center frequency values ​​to obtain a second sample center frequency with the highest center frequency value, and using the sample deformation modal components corresponding to each center frequency value within the second sample center frequency as sample periodic item deformation data; Obtaining a second true label corresponding to the sample periodic deformation data, and inputting the sample periodic deformation data with the second true label into a periodic deformation prediction model to be trained, so that the periodic deformation model to be trained predicts a first periodic deformation prediction amount based on the sample periodic deformation data; wherein the second true label is the actual periodic deformation amount of the sample periodic deformation data; Calculate a second loss function value based on the first periodic deformation prediction amount and the actual periodic deformation amount; determine whether the second loss function value converges; if so, the training of the periodic deformation prediction model to be trained is completed, and the preset periodic deformation prediction model is obtained; if not, optimize the hyperparameters in the periodic deformation prediction model according to the second particle swarm optimization algorithm, and then continue to train the periodic deformation prediction model to be trained; The landslide deformation value determination module is used to calculate the sum of the trend deformation prediction value and the periodic deformation prediction value to obtain the landslide deformation prediction value.

8. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for predicting landslide deformation based on variational modal decomposition according to any one of claims 1 to 6 is implemented.

9. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is executed, the device where the storage medium is located is controlled to execute the landslide deformation prediction method based on variational modal decomposition according to any one of claims 1 to 6.

Images