Landslide displacement prediction method and system based on width learning, medium and equipment
Through the landslide displacement prediction method based on width learning, combined with robust empirical modal decomposition and width learning model, the problem of landslide displacement data noise and low prediction efficiency is solved, and more accurate and efficient landslide displacement prediction is achieved.
Patent Information
- Application Number
- CN202411994479.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-31
- Publication Date
- 2025-04-08
AI Technical Summary
The prior art has noise problems when collecting landslide displacement data, traditional methods are difficult to accurately predict landslide displacement, and traditional neural networks are inefficient when processing nonlinear and multidimensional data.
The landslide displacement prediction method based on width learning is adopted, and the landslide displacement data is decomposed into trend term displacement and period term displacement through robust empirical modal decomposition. A prediction model is established in combination with width learning, and the trend term displacement, period term displacement and fundamental external inducing factors are used for training and testing.
It effectively eliminates noise in landslide displacement data, improves the accuracy and efficiency of landslide displacement prediction, and allows dynamic addition of new features and nodes to improve model performance.
Smart Images
Figure CN119782749A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of landslide displacement prediction, and in particular to a landslide displacement prediction method, system, medium and equipment based on width learning. Background Art
[0002] Landslide displacement prediction is an important topic in geological engineering and natural disaster research. In the process of landslide displacement analysis and prediction, the existing technology usually needs to combine data processing and prediction models. When using traditional empirical mode decomposition (EMD) to decompose landslide displacement data, due to the fixed screening stop criterion, EMD easily mixes information of different frequencies in the same mode, resulting in the decomposition result cannot accurately reflect the true characteristics of the signal. Landslide displacement prediction usually relies on traditional machine learning methods such as neural networks and support vector machines. However, these methods have limitations when dealing with the nonlinear and multidimensional characteristics of landslide displacement data. Traditional neural networks need to be trained layer by layer. When the data feature dimension increases or external inducing factors need to be introduced, the model usually needs to be retrained, which is inefficient. During the acquisition process, landslide displacement experimental data is often affected by factors such as equipment accuracy and environmental interference. There may be noise in the data. It is difficult for traditional methods to extract key features from it, affecting the prediction performance. Summary of the invention
[0003] The purpose of the present invention is to solve the problem of noise in landslide displacement data collection, and propose a landslide displacement prediction method based on width learning, comprising the following steps:
[0004] S1, collecting and preprocessing the landslide displacement data to obtain the preprocessed landslide displacement data;
[0005] S2, using robust empirical mode decomposition to decompose the preprocessed landslide displacement data to obtain trend term displacement and period term displacement;
[0006] S3, obtaining candidate external inducing factors of landslide displacement, using MIC to calculate the correlation between the candidate external inducing factors and the periodic displacement, and taking the candidate external inducing factors with a correlation higher than a set threshold as the fundamental external inducing factors;
[0007] S4. Establish a landslide displacement prediction model in combination with width learning, divide the trend item displacement, period item displacement and fundamental external inducing factors into a training set and a test set, and use the training set and the test set to train and test the landslide displacement prediction model, wherein the trend item displacement is used to predict the landslide trend item displacement result, and the fundamental external inducing factors are used to predict the landslide period item displacement result. The predicted landslide trend item displacement result and the landslide period item displacement result are added together to obtain the landslide displacement prediction;
[0008] S5. Input the landslide displacement data to be tested into the landslide displacement prediction model after the test to obtain the predicted landslide displacement.
[0009] Furthermore, S1 is specifically:
[0010] The landslide displacement data were obtained, the 3σ criterion was used to identify the outliers in the landslide displacement data, and the outliers were replaced by linear interpolation to obtain the preprocessed landslide displacement data.
[0011] Furthermore, the preprocessed landslide displacement data are decomposed using robust empirical mode decomposition, specifically:
[0012] The preprocessed landslide displacement data is subjected to empirical mode decomposition to obtain the intrinsic mode components. In the process of empirical mode decomposition, the cost function is used as the screening stop condition. The cost function is expressed as:
[0013] f ik =RMS(m ik [t])+|EK(m ik [t])|
[0014]
[0015]
[0016] Among them, f ik represents the cost function, RMS(m ik [t]) represents m ik The root mean square of [t], m ik [t] represents the envelope mean of the ith IMF after k screenings, t is the sampling point, EK(m ik [t]) represents m ik The excess kurtosis of [t], N s Indicates the total number of signals, Indicates m ik The arithmetic mean of [t];
[0017] When RMS(m ik [t])<0.001, EK(m ik [t])<0 decomposition stops;
[0018] The decomposed intrinsic mode components are superimposed to reconstruct the preprocessed landslide displacement data, and the robustness of the reconstructed data is verified. The reconstructed data is expressed as:
[0019]
[0020] Where D represents the reconstructed data of the landslide displacement data after preprocessing, b represents the number of intrinsic mode components, and IMF a represents the ath eigenmode component;
[0021] The frequency threshold of the eigenmode components is set, and the eigenmode components with frequencies greater than the threshold are superimposed to obtain the periodic term displacement, and the remaining eigenmode components are used as the trend term displacement.
[0022] Furthermore, candidate external inducing factors include: displacement change, accumulated rainfall, and reservoir water level change.
[0023] Furthermore, taking the month when the landslide displacement is predicted as the benchmark, the month when the landslide displacement is predicted is the mth month, and the candidate external inducing factors specifically include: the displacement change in the m-1th month, the displacement change in the m-2th month, the displacement change in the m-3rd month, the accumulated rainfall in the m-1th month, the sum of the accumulated rainfall in the m-1st month and the m-2nd month, the sum of the accumulated rainfall in the m-1st month, the m-2nd month, and the m-3rd month, the reservoir water level change in the mth month, the reservoir water level change in the m-1st month, and the reservoir water level change in the m-2nd month.
[0024] Furthermore, the fundamental external inducing factors and the periodic displacement are used to predict the periodic displacement results of the landslide, wherein the periodic displacement used to predict the periodic displacement results of the landslide is obtained as follows:
[0025] Use MIC to calculate the correlation between the m-1th month periodic item displacement, the m-2th month periodic item displacement, the m-3th month periodic item displacement and the m-th month periodic item displacement, and select the periodic item displacement whose correlation degree value is greater than the set threshold from the m-1th month periodic item displacement, the m-2th month periodic item displacement and the m-3th month periodic item displacement.
[0026] Furthermore, the landslide displacement prediction model consists of three parts: input layer, hidden layer and output layer;
[0027] The input passes through the feature map of the input layer to obtain the mapping features:
[0028] Z j =φ(XW zj +β zj ), j = 1, 2, ..., e
[0029] Among them, Z j represents the jth group of mapping features, e represents the number of mapping feature groups, φ( ) is the feature mapping layer function, X represents the input of the landslide displacement prediction model, W zj and β zj Represents the weights and biases of the randomly generated j-th group of mapping features;
[0030] Mapping features through hidden layers to get enhanced nodes:
[0031] H k =ξ(ZW hk +β hk ), k = 1, 2, ..., g
[0032] Among them, H k represents the kth group of enhanced nodes, ξ( ) is a nonlinear activation function, W hk and β hk is the randomly generated weight and bias of the kth group of enhanced nodes, Z = [Z1, Z2, ..., Z e ] represents the mapping feature, g represents the number of enhanced nodes;
[0033] Based on the mapping features and enhanced nodes, the output of the landslide displacement prediction model is obtained in the output layer:
[0034] O=[Z|H]W
[0035] W=[Z|H] + O
[0036] Where O represents the output, W represents the connection weight between the mapping feature and the enhanced node and the output layer, [Z|H] + is the pseudo-inverse of [Z|H], H=[H1,H2,...,H g ] represents an enhanced node, and [Z|H] represents the addition of Z and H, that is, the feature concatenation of Z and H.
[0037] The present invention also proposes a landslide displacement prediction system based on width learning, comprising:
[0038] A data processing module is used to collect and pre-process the landslide displacement data to obtain the pre-processed landslide displacement data;
[0039] A data decomposition module is used to decompose the preprocessed landslide displacement data using robust empirical mode decomposition to obtain trend term displacement and period term displacement;
[0040] The fundamental external inducing factor acquisition module is used to obtain candidate external inducing factors of landslide displacement, use MIC to calculate the correlation between the candidate external inducing factors and the periodic displacement, and take the candidate external inducing factors with a correlation higher than a set threshold as the fundamental external inducing factors;
[0041] A model building module is used to establish a landslide displacement prediction model in combination with width learning, divide the trend item displacement, period item displacement and fundamental external inducing factors into a training set and a test set, and use the training set and the test set to train and test the landslide displacement prediction model, wherein the trend item displacement is used to predict the landslide trend item displacement result, and the fundamental external inducing factors are used to predict the landslide period item displacement result. The predicted landslide trend item displacement result and the landslide period item displacement result are added together to obtain the landslide displacement prediction;
[0042] The result acquisition module is used to input the displacement data of the landslide to be tested into the landslide displacement prediction model after the test to obtain the predicted landslide displacement.
[0043] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the landslide displacement prediction method based on width learning is implemented.
[0044] The present invention also proposes an electronic device, comprising a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the above-mentioned landslide displacement prediction method based on width learning.
[0045] The beneficial effects brought by the technical solution provided by the present invention are:
[0046] The present invention decomposes the landslide displacement experimental data through robust empirical mode decomposition. Robust empirical mode decomposition replaces the screening stop criterion in the standard EMD with a soft screening stop criterion (cost function), so that it dynamically adjusts the stop condition according to the local characteristics of the input data, thereby avoiding the modal aliasing problem that may exist in the standard EMD. The use of robust empirical mode decomposition can better meet the decomposition requirements of the landslide displacement experimental data, and can extract clearer intrinsic mode components. The present invention constructs a landslide displacement prediction model based on a width learning system, inputs data through incremental learning, and when it is necessary to add external induced factors to predict the periodic term displacement, the features of the new data can be directly integrated into the model, and enhanced nodes or mapping nodes can be dynamically added to improve the model performance. In addition, the width learning system effectively extracts key features from the landslide displacement experimental data through feature mapping and enhanced nodes, thereby eliminating the noise of the landslide displacement experimental data during collection. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 is a flow chart of a landslide displacement prediction method based on width learning according to an embodiment of the present invention;
[0048] Figure 2is a schematic diagram of data preprocessing according to an embodiment of the present invention;
[0049] Figure 3 The trend item displacement prediction result (BLS prediction value) of the embodiment of the present invention is compared with the trend item displacement prediction results of the BP neural network model and the SVR (support vector regression) model;
[0050] Figure 4 The periodic term displacement prediction results (BLS prediction values) of the embodiment of the present invention are compared with the periodic term displacement prediction results of the BP neural network model and the SVR (support vector regression) model;
[0051] Figure 5 The landslide displacement prediction result (BLS prediction value) of the embodiment of the present invention is compared with the landslide displacement prediction results of the BP neural network model and the SVR (support vector regression) model;
[0052] Figure 6 It is a block diagram of an electronic device in an exemplary embodiment of the present invention. DETAILED DESCRIPTION
[0053] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.
[0054] The flow chart of the landslide displacement prediction method based on width learning in the embodiment of the present invention is as follows: Figure 1 , specifically including the following steps:
[0055] S1. Collecting landslide displacement data and preprocessing it to obtain preprocessed landslide displacement data.
[0056] The landslide displacement data were obtained, the 3σ criterion was used to identify the outliers in the landslide displacement data, and the outliers were replaced by linear interpolation to obtain the preprocessed landslide displacement data.
[0057] refer to Figure 2 , Figure 2 3σ criterion is used to process the abnormal values in the original data of landslide displacement, and the abnormal values are replaced by linear interpolation method to obtain the landslide displacement experimental data.
[0058] S2. Use robust empirical mode decomposition to decompose the preprocessed landslide displacement data to obtain trend term displacement and period term displacement.
[0059] The preprocessed landslide displacement data is subjected to empirical mode decomposition to obtain the intrinsic mode components, which represent the frequency components in the original signal. In the process of empirical mode decomposition, the cost function is used as the screening stop condition, and the cost function is expressed as:
[0060] f ik =RMS(m ik [t])+|EK(m ik [t])|
[0061]
[0062] Among them, f ik represents the cost function, RMS(m ik [t]) represents m ik The root mean square of [t], m ik [t] represents the envelope mean of the ith IMF after k screenings, t is the sampling point, EK(m ik [t]) represents m ik The excess kurtosis of [t], N s Indicates the total number of signals, Indicates m ik [t] is the arithmetic mean. When RMS(m ik [t])<0.001, EK(m ik When [t])<0, the decomposition stops.
[0063] The decomposed intrinsic mode components are superimposed, and the preprocessed landslide displacement data is reconstructed to obtain an approximate representation of the original signal, and the robustness of the reconstructed data is verified. The reconstructed data is expressed as:
[0064]
[0065] Where D represents the reconstructed data of the landslide displacement data after preprocessing, b represents the number of intrinsic mode components, and IMG a represents the ath eigenmode component.
[0066] Set the frequency threshold of the eigenmode components, and use the eigenmode components with frequencies greater than the threshold as periodic term displacements, and the remaining eigenmode components as trend term displacements.
[0067] S3. Obtain candidate external inducing factors of landslide displacement, use MIC to calculate the correlation between the candidate external inducing factors and the periodic displacement, and take the candidate external inducing factors with a correlation degree higher than a set threshold as the fundamental external inducing factors.
[0068] Candidate external inducing factors include: displacement change, accumulated rainfall, and reservoir water level change.
[0069] Taking the month when the landslide displacement is predicted as the benchmark, the month when the landslide displacement is predicted is the mth month, and the candidate external inducing factors specifically include: the displacement change in the m-1th month, the displacement change in the m-2th month, the displacement change in the m-3rd month, the accumulated rainfall in the m-1th month, the sum of the accumulated rainfall in the m-1st month and the m-2nd month, the sum of the accumulated rainfall in the m-1st month, the m-2nd month, and the m-3rd month, the reservoir water level change in the mth month, the reservoir water level change in the m-1st month, and the reservoir water level change in the m-2nd month.
[0070] Refer to Table 1, which shows the correlation between the above candidate external inducing factors and the periodic term displacement calculated by MIC, where E1 represents the displacement change in the m-1th month, E2 represents the displacement change in the m-2th month, E3 represents the displacement change in the m-3rd month, E4 represents the cumulative rainfall in the m-1th month, E5 represents the sum of the cumulative rainfall in the m-2th month, E6 represents the sum of the cumulative rainfall in the m-3rd month, E7 represents the reservoir water level change in the current month (i.e., the mth month), E8 represents the reservoir water level change in the m-1th month, and E9 represents the reservoir water level change in the m-2th month.
[0071] Table 1
[0072]
[0073]
[0074] Refer to Table 2, which shows the correlation between the displacement of the m-1st month cycle term, the displacement of the m-2nd month cycle term, the displacement of the m-3rd month cycle term and the displacement of the mth month cycle term calculated by MIC. 10 It represents the correlation degree between the displacement of the periodic term in the m-1th month and the displacement of the periodic term in the mth month, E 11 It represents the correlation degree between the displacement of the periodic term of the m-2th month and the displacement of the periodic term of the mth month, E 12 It indicates the correlation degree between the displacement of the periodic term in the m-3th month and the displacement of the periodic term in the mth month.
[0075] Table 2
[0076] Periodic term displacement MIC <![CDATA[E 10 ]]> 0.7555870885930 <![CDATA[E 11 ]]> 0.3717245191112 <![CDATA[E 12 ]]> 0.2627462777974
[0077] The periodic term of landslide displacement is mainly affected by external periodic factors, such as rainfall, reservoir water level and other external inducing factors. The present invention calculates the MIC of periodic term displacement and external inducing factors, as well as the correlation between the periodic term displacement in the m-1th month, the periodic term displacement in the m-2th month, the periodic term displacement in the m-3th month and the periodic term displacement in the mth month, and screens out external inducing factors and periodic term displacement with a correlation degree higher than 0.3 as the input of the periodic term displacement prediction model.
[0078] The correlation threshold is set to 0.3, and the cumulative rainfall in the m-1th month with a correlation degree less than the threshold is eliminated to obtain the fundamental external inducing factor. The periodic item displacement in the m-3th month with a correlation degree less than the threshold is eliminated, and the periodic item displacement in the m-1st and m-2th months is retained, and together with the fundamental external inducing factor, it is used as the input of the periodic item displacement prediction model.
[0079] S4. Combine width learning to establish a landslide displacement prediction model, divide the trend item displacement, period item displacement and fundamental external inducing factors into training set and test set, use the training set and test set to train and test the landslide displacement prediction model, where the trend item displacement is used to predict the landslide trend item displacement result, and the fundamental external inducing factors are used to predict the landslide period item displacement result. The predicted landslide trend item displacement result and the landslide period item displacement result are added together to get the landslide displacement prediction.
[0080] Specifically, the landslide displacement data from 2006 to 2011 were selected and divided into trend item displacement training set and period item displacement training set and corresponding labels. The data from 2012 were used as the test set, which were divided into trend item displacement test set and period item displacement test set and corresponding labels as the verification of the results.
[0081] The landslide displacement prediction model consists of three parts: input layer, hidden layer and output layer.
[0082] The input passes through the feature map of the input layer to obtain the mapping features:
[0083] Z j =φ(XW zj +β zj ), j = 1, 2, ..., e
[0084] Among them, Z j represents the jth group of mapping features, e represents the number of mapping feature groups, φ() is the feature mapping layer function, X represents the input of the landslide displacement prediction model, and W zj and β zj Represents the weights and biases of the randomly generated j-th group of mapping features.
[0085] Mapping features through hidden layers to get enhanced nodes:
[0086] H k =ξ(ZW hk +β hk ), k = 1, 2, ..., g
[0087] Among them, H k represents the kth group of enhanced nodes, ξ( ) is a nonlinear activation function, W hk and β hk is the randomly generated weight and bias of the kth group of enhanced nodes, Z = [Z1, Z2, ..., Ze ] represents the mapping feature, and g represents the number of enhanced nodes.
[0088] Based on the mapping features and enhanced nodes, the output of the landslide displacement prediction model is obtained in the output layer:
[0089] O=[Z|H]W
[0090] W=[Z|H] + O
[0091] Where O represents the output, W represents the connection weight between the mapping feature and the enhanced node and the output layer, [Z|H] + is the pseudo-inverse of [Z|H], H=[H1,H2,...,H g ] represents an enhanced node, and [Z|H] represents the addition of Z and H, that is, the feature concatenation of Z and H.
[0092] When testing the model, the trend item displacement test set and the period item displacement test set are input, and the trend item displacement and the period item displacement are predicted respectively to obtain the prediction results, and the accuracy of the prediction results is verified according to the corresponding labels. The above trend item displacement prediction results and period item displacement prediction results are superimposed to obtain the overall landslide displacement prediction results.
[0093] S5. Input the landslide displacement data to be tested into the landslide displacement prediction model after the test to obtain the predicted landslide displacement.
[0094] The trend item displacement prediction results (BLS prediction values) of the present invention are compared with the prediction results of the BP neural network model and the SVR (support vector regression) model. Figure 3 The periodic displacement prediction results (BLS prediction values) of the present invention are compared with the prediction results of the BP neural network model and the SVR model. Figure 4 shown.
[0095] The above trend term displacement prediction results and period term displacement prediction results are superimposed to obtain the overall landslide displacement prediction results, which are expressed in terms of RMSE (root mean square error), MAE (mean absolute error) and R 2 (Determination coefficient) is used as the performance index to evaluate the prediction performance of the trend term displacement prediction results, the period term displacement prediction results and the overall landslide displacement prediction results, and compared with the BP neural network prediction model and the SVR prediction model. Table 3 shows the performance indicators of different prediction models. Figure 5 The above diagrams show the overall prediction results of different prediction models. It can be concluded that the prediction model of the present invention has improved performance indicators in landslide displacement prediction.
[0096] Table 3
[0097] RMSE MAE <![CDATA[R 2 ]]> BLS trend term displacement prediction 1.6682425571 1.2878201667 0.9981336861 BP trend item displacement prediction 7.8473175622 7.0798559341 0.9587039281 SVR trend term displacement prediction 7.1038640057 5.5136798335 0.9661580355 BLS period term displacement prediction 8.0698190521 6.3245126375 0.9924344427 BP periodic term displacement prediction 16.7640170533 13.1934250874 0.9673510272 SVR periodic term displacement prediction 17.2672917222 13.6207510697 0.9653612839 BLS landslide overall displacement prediction 8.10437829223 6.77103866667 0.9956976655 Prediction of overall displacement of BP landslide 15.2622343368 13.7345060833 0.9847418810 Prediction of overall displacement of SVR landslide 14.5717919123 10.40119741667 0.9860911674
[0098] In an exemplary embodiment, a landslide displacement prediction system based on width learning is provided, comprising:
[0099] A data processing module is used to collect and pre-process the landslide displacement data to obtain the pre-processed landslide displacement data;
[0100] A data decomposition module is used to decompose the preprocessed landslide displacement data using robust empirical mode decomposition to obtain trend term displacement and period term displacement;
[0101] The fundamental external inducing factor acquisition module is used to obtain candidate external inducing factors of landslide displacement, use MIC to calculate the correlation between the candidate external inducing factors and the periodic displacement, and take the candidate external inducing factors with a correlation higher than a set threshold as the fundamental external inducing factors;
[0102] A model building module is used to establish a landslide displacement prediction model in combination with width learning, divide the trend item displacement, period item displacement and fundamental external inducing factors into a training set and a test set, and use the training set and the test set to train and test the landslide displacement prediction model, wherein the trend item displacement is used to predict the landslide trend item displacement result, and the fundamental external inducing factors are used to predict the landslide period item displacement result. The predicted landslide trend item displacement result and the landslide period item displacement result are added together to obtain the landslide displacement prediction;
[0103] The result acquisition module is used to input the displacement data of the landslide to be tested into the landslide displacement prediction model after the test to obtain the predicted landslide displacement.
[0104] In an exemplary embodiment, a computer-readable storage medium is included, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned landslide displacement prediction method based on width learning is implemented.
[0105] See also Figure 6 In an exemplary embodiment, an electronic device is also included, including at least one processor, at least one memory, and at least one communication bus.
[0106] Wherein, a computer program is stored in the memory, and the computer program includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through the communication bus to execute the above-mentioned landslide displacement prediction method based on width learning.
[0107] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A landslide displacement prediction method based on width learning, characterized in that: The following steps are involved: S1, collecting and preprocessing the landslide displacement data to obtain the preprocessed landslide displacement data; S2, using robust empirical mode decomposition to decompose the preprocessed landslide displacement data to obtain trend term displacement and period term displacement; S3, obtaining candidate external inducing factors of landslide displacement, using MIC to calculate the correlation between the candidate external inducing factors and the periodic displacement, and taking the candidate external inducing factors with a correlation higher than a set threshold as the fundamental external inducing factors; S4. Establish a landslide displacement prediction model in combination with width learning, divide the trend item displacement, period item displacement and fundamental external inducing factors into a training set and a test set, and use the training set and the test set to train and test the landslide displacement prediction model, wherein the trend item displacement is used to predict the landslide trend item displacement result, the fundamental external inducing factors and the period item displacement are used to predict the landslide period item displacement result, and the predicted landslide trend item displacement result and the landslide period item displacement result are added together to obtain the landslide displacement prediction; S5. Input the landslide displacement data to be tested into the landslide displacement prediction model after the test to obtain the predicted landslide displacement.
2. A landslide displacement prediction method based on width learning according to claim 1, characterized in that: S1 is specifically: The landslide displacement data were obtained, the 3σ criterion was used to identify the outliers in the landslide displacement data, and the outliers were replaced by linear interpolation to obtain the preprocessed landslide displacement data.
3. The landslide displacement prediction method based on width learning according to claim 1 is characterized in that: The preprocessed landslide displacement data are decomposed using robust empirical mode decomposition, specifically: The preprocessed landslide displacement data is subjected to empirical mode decomposition to obtain the intrinsic mode components. In the process of empirical mode decomposition, the cost function is used as the screening stop condition. The cost function is expressed as: f ik =RMS(m ik [t])+|EK(m ik [t])| Among them, f ik represents the cost function, RMS(m ik [t]) represents m ik The root mean square of [t], m ik [t] represents the envelope mean of the ith IMF after k screenings, t is the sampling point, EK(m ik [t]) represents m ik The excess kurtosis of [t], N s Indicates the total number of signals, Indicates m ik The arithmetic mean of [t]; When RMS(m IK [t])<0.001, EK(m IK [t])<0, decomposition stops; The decomposed intrinsic mode components are superimposed to reconstruct the preprocessed landslide displacement data, and the robustness of the reconstructed data is verified. The reconstructed data is expressed as: Where D represents the reconstructed data of the landslide displacement data after preprocessing, b represents the number of intrinsic mode components, and IMF a represents the ath eigenmode component; The frequency threshold of the eigenmode components is set, and the eigenmode components with frequencies greater than the threshold are superimposed to obtain the periodic term displacement, and the remaining eigenmode components are used as the trend term displacement.
4. The landslide displacement prediction method based on width learning according to claim 1 is characterized in that: Candidate external inducing factors include: displacement change, accumulated rainfall, and reservoir water level change.
5. The landslide displacement prediction method based on width learning according to claim 4 is characterized in that: Taking the month when the landslide displacement is predicted as the benchmark, the month when the landslide displacement is predicted is the mth month, and the candidate external inducing factors specifically include: the displacement change in the m-1th month, the displacement change in the m-2th month, the displacement change in the m-3rd month, the accumulated rainfall in the m-1th month, the sum of the accumulated rainfall in the m-1st month and the m-2nd month, the sum of the accumulated rainfall in the m-1st month, the m-2nd month, and the m-3rd month, the reservoir water level change in the mth month, the reservoir water level change in the m-1st month, and the reservoir water level change in the m-2nd month.
6. A landslide displacement prediction method based on width learning according to claim 5, characterized in that: The fundamental external inducing factors and the periodic displacement are used to predict the periodic displacement results of the landslide. The periodic displacement used to predict the periodic displacement results of the landslide is obtained as follows: Use MIC to calculate the correlation between the m-1th month periodic item displacement, the m-2th month periodic item displacement, the m-3th month periodic item displacement and the m-th month periodic item displacement, and select the periodic item displacement whose correlation degree value is greater than the set threshold from the m-1th month periodic item displacement, the m-2th month periodic item displacement and the m-3th month periodic item displacement.
7. The landslide displacement prediction method based on width learning according to claim 1 is characterized in that: The landslide displacement prediction model consists of three parts: input layer, hidden layer and output layer; The input passes through the feature map of the input layer to obtain the mapping features: Z j =φ(XW zj +β zj ),j=1,2,...,e Among them, Z j represents the jth group of mapping features, e represents the number of mapping feature groups, v() is the feature mapping layer function, X represents the input of the landslide displacement prediction model, and W zj and β zj Represents the weights and biases of the randomly generated j-th group of mapping features; Mapping features through hidden layers to get enhanced nodes: H k =ξ(ZW hk +b hk ),k=1,2,...,g Among them, H k represents the kth group of enhanced nodes, ξ() is a nonlinear activation function, W hk and β hk is the randomly generated weight and bias of the kth group of enhanced nodes, Z = [Z1, Z2, ..., Z e ] represents the mapping feature, g represents the number of enhanced nodes; Based on the mapping features and enhanced nodes, the output of the landslide displacement prediction model is obtained in the output layer: O=[Z|H]W W=[Z|H] + ABOUT Where O represents the output, W represents the connection weight between the mapping feature and the enhanced node and the output layer, [Z|H] + is the pseudo-inverse of [Z|H], H=[H1,H2,...,H g ] indicates an enhanced node, and [Z|H] indicates the addition of Z and H.
8. A landslide displacement prediction system based on width learning, characterized in that: include: A data processing module is used to collect and pre-process the landslide displacement data to obtain the pre-processed landslide displacement data; A data decomposition module is used to decompose the preprocessed landslide displacement data using robust empirical mode decomposition to obtain trend term displacement and period term displacement; The fundamental external inducing factor acquisition module is used to obtain candidate external inducing factors of landslide displacement, use MIC to calculate the correlation between the candidate external inducing factors and the periodic displacement, and take the candidate external inducing factors with a correlation higher than a set threshold as the fundamental external inducing factors; A model building module is used to establish a landslide displacement prediction model in combination with width learning, divide the trend item displacement, period item displacement and fundamental external inducing factors into a training set and a test set, and use the training set and the test set to train and test the landslide displacement prediction model, wherein the trend item displacement is used to predict the landslide trend item displacement result, and the fundamental external inducing factors are used to predict the landslide period item displacement result. The predicted landslide trend item displacement result and the landslide period item displacement result are added together to obtain the landslide displacement prediction; The result acquisition module is used to input the displacement data of the landslide to be tested into the landslide displacement prediction model after the test to obtain the predicted landslide displacement.
9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.
10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the method according to any one of claims 1 to 7.