Landslide displacement prediction method based on CNN-BiLSTM-XGB

Abstract

The invention relates to the technical field of geological displacement prediction, in particular to a CNNBiLSTM-XGB-based landslide displacement prediction method, which comprises the following steps of: firstly, decomposing a landslide displacement sequence into a periodic term, a random term and a trend term by adopting an SSA-VMD decomposition method, and selecting a proper influence factor as an input variable by adopting a grey relational degree; a combined model based on CNNBiLSTM-XGB is provided to predict displacement of a periodic item and a random item, an ensemble learning method is adopted to be combined with CNNBiLSTM and XGB, and SSA is utilized to optimize parameters of the combined model, so that the parameter adjustment process is simplified, the influence of subjective factors is avoided, and the model prediction efficiency is improved. A quadratic index average method is adopted to predict trend term displacement, and finally a time sequence theoretical model is adopted to obtain a landslide displacement prediction sequence. The combined model can more fully extract the features of the landslide displacement data, effectively solves the problem of insufficient prediction performance and generalization ability of a single model in a deep learning model, and improves the prediction precision and generalization ability of the model while ensuring the prediction efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of geological displacement prediction, and in particular to a landslide displacement prediction method based on CNN_BiLSTM-XGB. Background Art

[0002] Landslide geological disaster is one of the most destructive geological disasters in the world. It is caused by the sliding or collapse of geological bodies such as soil or rock on hillsides or mountains due to gravity. It not only poses a serious threat to human survival and safety, but also has serious negative impacts on resources, environment and property. Therefore, the prediction of landslide displacement has become an effective means to reduce the losses caused by landslide disasters. The change of displacement can be used as an important indicator to predict the landslide damage process. Accurate prediction is crucial to improve the early warning effect and disaster emergency response capabilities, and can effectively reduce the various losses caused by landslide disasters.

[0003] In the method of landslide displacement prediction, deep learning models such as eXtreme Gradient Boosting, Convolutional Neural Networks (CNN), and Bi-directional Long Short-Term Memory (BiLSTM) have good prediction effects on landslide displacement prediction models, but these models have weak data generalization ability, complex structure, and difficulty in parameter selection, resulting in low prediction accuracy and efficiency. Therefore, the present invention proposes a CNN_BiLSTM-XGB combined prediction model, which combines CNN_BiLSTM and XGB models through an integrated learning method, and uses a Sparrow Search Algorithm (SSA) to optimize the parameters of the combined model, find the optimal parameters and apply them to landslide displacement prediction. Summary of the invention

[0004] The present invention relates to a landslide displacement prediction method based on CNN_BiLSTM-XGB, which aims to improve the prediction efficiency and accuracy of the landslide displacement prediction method and enhance the generalization ability of the model.

[0005] The present invention is implemented by the following technical solution, comprising the following steps:

[0006] Step 1: Decompose the accumulated displacement of the landslide into three mode components using the SSA-VMD decomposition method; divide it into periodic term, random term and trend term according to the time series theory and the characteristics of the displacement components;

[0007] Step 2: Divide the periodic term and random term displacement component data sets into training set, validation set and test set, and introduce the influencing factors of periodic term and random term displacement prediction respectively;

[0008] Step 3: Use the CNN_BiLSTM-XGB combined model to predict each component and train a prediction model suitable for periodic items and random item displacement components;

[0009] Step 4: Use the quadratic exponential smoothing method to predict the displacement component of the trend item;

[0010] Step 5: Superimpose the predicted values ​​of each component to form the cumulative displacement prediction value and calculate the model performance index.

[0011] The steps of decomposing the accumulated displacement of landslide using SSA-VMD are as follows:

[0012] Step 1.1: Set the range of VMD parameters k and α, k range is [3,5], α range is [0,2000];

[0013] Step 1.2: Initialize parameters, use VMD to decompose the landslide displacement, and set the envelope entropy as fitness;

[0014] Step 1.3: Update the positions of the discoverer, joiner, and aware dangerous individuals in the SSA algorithm;

[0015] Step 1.4: Determine whether the current fitness is greater than the previous fitness. If so, update the fitness; if not, keep it unchanged.

[0016] Step 1.5: Determine whether the maximum number of iterations has been reached. If not, return to step 2. If so, output the optimal parameters and fitness.

[0016] Step 1.6: Input the optimal parameters into VMD to obtain the decomposition components of landslide displacement.

[0017] The specific implementation process of step 2 includes the following steps:

[0018] Step 2.1: Divide the periodic term and random term displacement components obtained by SSA-VMD decomposition into training set, validation set and test set according to 8:1:1;

[0019] Step 2.2: Use the grey correlation method to screen out appropriate influencing factors.

[0020] For step 3, the process of using the CNN_BiLSTM-XGB combined model to predict each component includes the following steps:

[0021] Step 3.1: Construct the CNN_BiLSTM-XGB combined model through the integrated learning method;

[0022] Step 3.3: Initialize the parameters and train the training set CNN_BiLSTM-XGB combined model;

[0023] Step 3.3: Use the SSA optimization algorithm to find the best parameter combination for each model;

[0024] Step 3.4: Let the true value of the validation set be y v After the validation set, the model prediction values ​​can be obtained as The errors between them can be obtained Record the error of the model on the validation set.

[0025] Step 3.5: Judge the errors of the two models on the validation set. If the error difference is large, select the model with smaller error as the final model. If the error between the two models is less than a certain threshold, weight the two models together, and the final prediction result is equal to the weighted sum of the prediction values ​​of the two models.

[0026] Step 3.6: After the test set, the predicted values ​​on the test set can be obtained as follows: Then the prediction result obtained after model integration is y T .

[0027] The specific implementation process of step 5 includes the following steps:

[0028] Step 5.1: Decompose Y according to time series theory t =T t +P t +R t , Y t is the cumulative displacement sequence of the landslide, T t is the displacement trend term predicting displacement, P t is the periodic term predicting displacement, R t The random term predicts the displacement. The final Y t Predict results;

[0029] Step 5.2: Calculate the squared sum of the deviations between the predicted values ​​and the true values ​​and the average RMSE:

[0030] Step 5.3: Calculate the average MAE of the deviation between the predicted value and the true value:

[0031] Step 5.4: Calculate the correlation R between the predicted value and the true value 2 :

[0032] Step 5.5: Verify the validity of the model based on the above calculation results.

[0033] The present invention provides a landslide displacement prediction method based on CNN_BiLSTM-XGB. Firstly, SSA-VMD is used to decompose the landslide displacement into periodic term, random term and trend term displacement with obvious characteristics. The data set is divided and the grey correlation degree is used to select the influencing factors with high correlation as the input components, so as to effectively improve the model accuracy. The CNN_BiLSTM-XGB combined model is constructed by the ensemble learning method to predict the periodic term and random term displacement. The parameters of the model are optimized by SSA to solve the influence of subjective factors in the artificial selection of parameters, so as to ensure the prediction efficiency while improving the prediction accuracy and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0035] Figure 1 It is a schematic diagram of the flow chart of the landslide displacement combined prediction model of the CNN_BiLSTM-XGB of the present invention.

[0036] Figure 2 It is a flow chart of the SSA-VMD algorithm of a specific embodiment of the present invention.

[0037] Figure 3 It is a flow chart of the CNN_BiLSTM-XGB algorithm of a specific embodiment of the present invention.

[0038] Figure 4 It is a schematic diagram of the decomposition result of the accumulated landslide displacement according to a specific embodiment of the present invention.

[0039] Figure 5 It is a schematic diagram of the landslide trend item prediction result of a specific embodiment of the present invention.

[0040] Figure 6 It is a schematic diagram of landslide period item prediction results according to a specific embodiment of the present invention.

[0041] Figure 7 It is a schematic diagram of the landslide random item prediction result of a specific embodiment of the present invention.

[0042] Figure 8 It is a schematic diagram of the prediction result of the accumulated displacement of a landslide according to a specific embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following is an explanation of some of the English abbreviations in the present invention to avoid ambiguity in the following descriptions:

[0044] Sparrow Search Algorithm (SSA);

[0045] Variational Mode Decomposition (VMD);

[0046] Convolutional Neural Networks (CNN)

[0047] Bi-direction Long Short-Term Memory (BiLSTM)

[0048] Extreme Gradient Boosting (XGB)

[0049] The present invention provides a landslide displacement prediction method based on CNN_BiLSTM-XGB, comprising the following steps:

[0050] S1: SSA-VMD is used to decompose the accumulated displacement of the landslide into periodic terms, random terms and trend terms;

[0051] S2: The periodic item and random item component data sets are divided into training set, validation set and test set, and the influencing factors are screened by the grey correlation method.

[0052] S3: Use the CNN_BiLSTM-XGB combined model to predict each component and train a prediction model suitable for each component;

[0053] S4: Use the quadratic exponential smoothing method to predict the trend item;

[0054] S5: Superimpose each prediction value to form a cumulative displacement prediction value, and complete the landslide displacement prediction.

[0055] For related algorithm flow and model flow, please refer to Figures 1 to 3 , the following combines the execution steps and specific embodiments.

[0056] In step S1, the SSA-VMD method is used to decompose the landslide displacement, which effectively avoids the influence of subjective factors. This method can find the best decomposition parameters according to different sequences, which can decompose the landslide displacement more thoroughly and improve the decomposition efficiency.

[0057] In step S2, the data sets of periodic term and random term displacement components are divided in a ratio of 8:1:1. Grey correlation is used to introduce monthly rainfall change, periodic term landslide displacement of the previous month, and cumulative periodic term landslide displacement of two months as influencing factors for periodic term displacement prediction. Based on the characteristics of the random displacement sequence itself, the random term landslide displacement of the previous month, the random term landslide change of the previous month, and the cumulative random term landslide displacement of two months are selected as influencing factors for random term displacement prediction.

[0058] The calculation method of grey relational degree is as follows:

[0059] The core of this method is to measure the correlation between two vectors. Through normalization, the influencing factor vector and the landslide component are adjusted to the same dimensional space. The distance between them is calculated, and the result range is limited to [0,1] by adjusting the minimum and maximum values. The three influencing factors with the highest correlation among the candidate influencing factors are selected as input variables.

[0060] Step S3, use the CNN_BiLSTM-XGB combined model to predict each component, train a suitable prediction model, combine the CNN_BiLSTM and XGB models through an integrated learning method, use SSA to optimize the combined model parameters, and the periodic item prediction displacement obtained by the combined model is as follows: Figure 6 As shown, the random term predicts the displacement as Figure 7 shown.

[0061] Ensemble learning combines multiple weak learners, performs sample weighting and learner weighting, and obtains a strong learner with more significant generalization performance than a single learner. It requires individuals to have certain accuracy and certain differences. The basic idea is to construct different base learners through different sampling or preprocessing methods for training data, and combine them into an integrated model to improve the overall prediction performance and generalization ability.

[0062] (1) First, the entire data set is divided into training set, validation set, and test set. The optimization methods of CNN_BiLSTM and XGBoost are set respectively, the hyperparameters are initialized, the upper and lower limits of the optimization parameters are optimized, and the number of iterations is optimized. The training set is trained on the CNN_BiLSTM model and the XGBoost model respectively.

[0063] (2) Use the SSA optimization algorithm to optimize the number of feature maps of the convolutional layer of the CNN_BiLSTM-XGB combination model, the number of neural network neurons, the step size, the maximum number of iterations, the maximum depth, and the minimum number of nodes.

[0064] (3) Let the true value of the validation set be y vAfter the validation set, the model prediction values ​​can be obtained as The errors between them can be obtained Record the model's error on the validation set. The error calculation formula is as follows:

[0065] (4) The errors between the two models on the validation set are judged. If the error difference is large, the model with the smaller error is selected as the final model. If the error between the two models is less than a certain threshold, the two models are weighted and combined, and the final prediction result is equal to the weighted sum of the prediction values ​​of the two models.

[0066] (5) If but like but

[0067] (6) After the test set, the predicted values ​​on the test set can be obtained as follows: Then the prediction result obtained after model integration is y T :

[0068] Step S4, using the quadratic exponential smoothing method to predict the trend item. The quadratic exponential smoothing method is a method for processing time series with a certain trend. The past observations and the most recent observations are weighted combined, and the trend of the time series is updated using exponential smoothing.

[0069] Trend item prediction Figure 5 shown.

[0070] Step S5, each prediction value is superimposed to form a cumulative displacement prediction value, the landslide displacement prediction is completed, and compared with the original displacement. The results show that the landslide displacement prediction performance indicators of the CNN_BiLSTM-XGB combined model are MAE 2.62mm, RMSE 3.41mm, R 2 is 0.996, which has a good prediction effect, proving the effectiveness of the prediction model of this invention. The cumulative displacement prediction results are as follows Figure 8 shown.

[0071] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A landslide displacement prediction method based on CNN_BiLSTM-XGB, characterized in that: The following steps are involved: Step 1: Use SSA-VMD to decompose the accumulated displacement of the landslide into periodic terms, random terms and trend terms; Step 2: Divide the periodic term and random term displacement component data sets into training set, validation set and test set, and introduce influencing factors through grey correlation screening; Step 3: Use the CNN_BiLSTM-XGB combined model to predict the displacement components of the periodic term and the random term, and train a suitable prediction model; Step 4: Use the quadratic exponential smoothing method to predict the trend item; Step 5: Superimpose each prediction value to form a cumulative displacement prediction value to complete the landslide displacement prediction.

2. The landslide displacement prediction method based on CNN_BiLSTM-XGB according to claim 1 is characterized in that: The process of decomposing the accumulated displacement of the landslide using the SSA-VMD decomposition method includes the following steps: Step 1.1: Set the range of VMD parameters k and α, k range is [3,5], α range is [0,2000]; Step 1.2: Initialize parameters, use VMD to decompose the landslide displacement, and set the envelope entropy as fitness; Step 1.3: Update the positions of the discoverer, joiner, and aware dangerous individuals in the SSA algorithm; Step 1.4: Determine whether the current fitness is greater than the previous fitness. If so, update the fitness; if not, keep it unchanged. Step 1.5: Determine whether the maximum number of iterations has been reached. If not, return to step 2. If so, output the optimal parameters and fitness. Step 1.6: Input the optimal parameters into VMD to obtain the decomposition components of landslide displacement.

3. The landslide displacement prediction method based on CNN_BiLSTM-XGB according to claim 1 is characterized in that: The specific implementation process of step 2 includes the following steps: Step 2.1: Divide the periodic term and random term displacement components obtained by SSA-VMD decomposition into training set, validation set and test set according to 8:1:1; Step 2.2: Use the grey correlation method to screen out appropriate influencing factors.

4. The landslide displacement prediction method based on CNN_BiLSTM-XGB according to claim 1, characterized in that: The process of predicting each component using the CNN_BiLSTM-XGB combined model includes the following steps: Step 3.1: Construct the CNN_BiLSTM-XGB combined model through ensemble learning method; Step 3.2: Initialize the parameters and train the training set CNN_BiLSTM-XGB combined model; Step 3.3: Use the SSA optimization algorithm to find the best parameter combination for each model; Step 3.4: Let the true value of the validation set be y v After the validation set, the model prediction values ​​can be obtained as The errors between them can be obtained Record the error of the model on the validation set. Step 3.5: Judge the errors of the two models on the validation set. If the error difference is large, select the model with smaller error as the final model. If the error between the two models is less than a certain threshold, weight the two models together, and the final prediction result is equal to the weighted sum of the prediction values ​​of the two models. Step 3.6: After the test set, the predicted values ​​on the test set can be obtained as follows: Then the prediction result obtained after model integration is y T .

Images