A CNN-LSTM-based method for predicting deep displacement of landslides under the influence of groundwater migration

Abstract

The present invention discloses a method for predicting deep displacement of landslide under the influence of groundwater migration based on CNN-LSTM, comprising the following steps: Step 1: Collect deep displacement and groundwater level data, and perform data interpolation and process outliers. Step 2: Merge the processed data into a data matrix in chronological order, and decompose it into a training set and a test set. Step 3: Input the training set into CNN and output feature data. Step 4: Input the feature data into LSTM, and the output data obtained is a prediction result. The loss function is calculated according to the prediction result, and the parameters of CNN and LSTM are optimized. Through multiple iterations, the optimal parameters are found. Step 5: Use the optimal parameters as the final parameters of the model, input the test set into CNN-LSTM, and the obtained prediction results are the final prediction results of CNN-LSTM. The present invention adopts a method combining CNN and LSTM, takes into account the influence of groundwater level on deep displacement, and makes full use of data, so that the stability and accuracy of the prediction results are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of landslide deep displacement prediction, and in particular to a method for predicting landslide deep displacement under the influence of groundwater migration based on CNN-LSTM. Background Art

[0002] Deep displacement can reflect the changes in the internal structure of the landslide. The rise and fall of the groundwater level will form pore water pressure on the structural surface of the landslide, change the mechanical state of the landslide, and the structural surface of the landslide will be affected by the thrust of water, which will reduce the stability of the landslide and produce deep displacement. The deep displacement is affected by the rise and fall of the groundwater level and shows nonlinear characteristics in long-term monitoring. Therefore, the prediction research on the deep displacement of landslide needs to consider the influence of the rise and fall of the groundwater level on the basis of long-term monitoring.

[0003] Among the existing deep displacement prediction models, the traditional regression model does not take into account the impact of groundwater migration, and the prediction accuracy is limited. The machine learning model can explore the potential relationship between the data, but the prediction effect of long-term nonlinear data is poor, and data decomposition is required, which increases the prediction error. Summary of the invention

[0004] To solve the above problems, the present invention provides a method for predicting deep displacement of landslide under the influence of groundwater migration based on CNN-LSTM, which is used to solve the problems that the existing deep displacement prediction model needs to perform data decomposition when predicting long-term data and produces errors, and cannot accurately extract the potential relationship between deep displacement and groundwater level.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A method for predicting deep displacement of landslides under the influence of groundwater migration based on CNN-LSTM specifically includes the following steps:

[0007] Step 1: Collect landslide data, including deep displacement data and groundwater level data. Due to external interference, the original landslide data often has data missing and data anomalies. Data interpolation and outlier processing are performed before data prediction.

[0008] Step 2: Construct training data. Merge the deep displacement data and groundwater level data processed in step 1 into a data matrix in chronological order. Use the first 80% of the data as the training set and the last 20% of the data as the test set.

[0009] Step 3: Use the training set in step 2 as CNN input data, and use the CNN method to extract and fuse the data features of deep displacement data and groundwater level data. The CNN process is as follows:

[0010] Y(i,j)=δ([w]*[Xi,j ]+b)

[0011]

[0012] In the formula, Y(i,j) is the feature data output by CNN, δ is the tanh activation function, * is the convolution symbol, [w] is the two-dimensional convolution kernel, [X i,j ] is the deep displacement of the landslide and the groundwater level in the convolution area, m is the number of rows where the data is located, n is the number of columns where the data is located, l is the convolution kernel moving step, and b is the bias parameter;

[0013] Step 4: Use the feature data output by CNN in step 3 as LSTM input data. The output data is the prediction result of the training set. The loss function is calculated based on the prediction result. The parameters of CNN and LSTM are optimized. The optimal parameters of the CNN-LSTM method are determined by iteration. The LSTM process is as follows:

[0014] f t =σ(W fh ·h t-1 +W fx ·x t +b f )

[0015] i t =σ(W ih ·h t-1 +W ix ·x t +b i )

[0016]

[0017]

[0018] O t =σ(W Oh ·h t-1 +W Ox ·x t +b O )

[0019] h t =O t tanh(C t )

[0020] In the formula, f is the forget gate, i is the input gate, O is the output gate, σ is the Sigmoid function, and h is t-1 is the output at time t-1, x t is the deep displacement characteristic data of the landslide input at time t, W is the weight of each data, b is each bias term, C tis the cell state at time t, C t-1 is the cell state at time t-1, h t is the output at time t;

[0021] Step 5: Use the optimal parameters of the CNN-LSTM method in step 4 as the final parameters of the model, and input the test set into the CNN-LSTM method. The prediction result obtained is the final prediction result of CNN-LSTM.

[0022] Furthermore, the data interpolation and outlier processing in step 1 specifically include the following steps:

[0023] Step 1-1: Determine the time period with missing data based on the chronological order;

[0024] Step 1-2: By constructing a multivariate nonlinear equation, the missing data is assumed to be the equation data, the number of terms in the equation and the parameters of each term are determined through time sequence, and the data values ​​are supplemented. The equation is as follows:

[0025] y=a n x n +a n-1 x n-1 +…+a 1 x+b

[0026] In the formula, y is the missing data to be predicted, x is the time data, n is the number of terms in the equation, and a n For x n Parameters;

[0027] Step 1-3: Process outliers through discrete linear convolution. The process is as follows:

[0028] g(i)=[a]*[v i ]

[0029]

[0030] In the formula, g(i) is the replaced data, [a] is the one-dimensional convolution kernel, * is the convolution symbol, [v i ] is the original data in the convolution area, and m is the length of the convolution kernel;

[0031] Furthermore, the loss function in step 4 is the mean square loss function, and the calculation formula is as follows:

[0032] loss(x i ,y i )=(x i -y i ) 2

[0033] In the formula, x iOutput data value for the LSTM method, y i is the actual data value;

[0034] Furthermore, the parameter optimization method in step 4 is the Adam method, and the optimized parameters include the two-dimensional convolution kernel [w] of CNN in step 3 and the data weights W in LSTM in step 4.

[0035] Furthermore, the number of iterations of the CNN-LSTM method in step 4 can be adjusted according to different input data.

[0036] The beneficial effects of the present invention are:

[0037] (1) In step 1, the original data is interpolated and outliers are processed. The multivariate nonlinear interpolation method and discrete linear convolution method used require less data computing power and have a simple structure. They can effectively supplement missing values ​​in the data and modify outliers in the data, thereby improving data quality and indirectly improving prediction accuracy.

[0038] (2) In step 3, the CNN method is used to extract different types of features using one or more convolution kernels through convolution operations, which can effectively explore the potential relationship between the deep displacement of the landslide and the groundwater level.

[0039] (3) In step 4, the LSTM method is used. LSTM has long short-term memory and can well predict long-term time series. Compared with traditional deep displacement prediction, it avoids the error caused by data decomposition and improves data utilization. In addition, the model structure is simple, the modeling time is short, and the modeling cost is low.

[0040] (4) In steps 1-5, when using the CNN-LSTM method to predict the deep displacement of the landslide, the influence of the groundwater level is taken into account. The rise and fall of the groundwater level is closely related to the deep displacement of the landslide. Predicting the deep displacement under the premise of considering the groundwater level is conducive to improving the prediction accuracy. In addition, only adding groundwater level data will not excessively increase the calculation cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of the present invention;

[0042] Figure 2 This is the relationship diagram between groundwater level and deep displacement of landslide in Example 1;

[0043] Figure 3 This is the relationship diagram between groundwater level and deep displacement of landslide in Example 2;

[0044] Figure 4 This is a data interpolation effect diagram of the present invention;

[0045] Figure 5This is an effect diagram of the present invention processing abnormal values;

[0046] Figure 6 It is the LSTM structure diagram;

[0047] Figure 7 A comparison diagram of the total displacement predicted by the method of the present invention in Example 1 and other prediction models;

[0048] Figure 8 This is a comparison chart of the total displacement predicted by the method of the present invention in Example 2 and other prediction models. DETAILED DESCRIPTION

[0049] In order to enable those skilled in the art to better understand the solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0050] A method for predicting deep displacement of landslides under the influence of groundwater migration based on CNN-LSTM specifically includes the following steps:

[0051] Step 1: Collect landslide data, including deep displacement data and groundwater level data. Due to external interference, the original landslide data often has data missing and data anomalies. Data interpolation and outlier processing are performed before data prediction.

[0052] Step 2: Construct training data. Merge the deep displacement data and groundwater level data processed in step 1 into a data matrix in chronological order. Use the first 80% of the data as the training set and the last 20% of the data as the test set.

[0053] Step 3: Use the training set in step 2 as CNN input data, and use the CNN method to extract and fuse the data features of deep displacement data and groundwater level data. The CNN process is as follows:

[0054] Y(i,j)=δ([w]*[X i,j ]+b)

[0055]

[0056] In the formula, Y(i,j) is the feature data output by CNN, δ is the tanh activation function, * is the convolution symbol, [w] is the two-dimensional convolution kernel, [X i,j ] is the deep displacement of the landslide and the groundwater level in the convolution area, m is the number of rows where the data is located, n is the number of columns where the data is located, l is the convolution kernel moving step, and b is the bias parameter;

[0057] Step 4: Use the feature data output by CNN in step 3 as LSTM input data. The output data is the prediction result of the training set. The loss function is calculated based on the prediction result. The parameters of CNN and LSTM are optimized. The optimal parameters of the CNN-LSTM method are determined by iteration. The LSTM process is as follows:

[0058] f t =σ(W fh ·h t-1 +W fx ·x t +b f )

[0059] i t =σ(W ih ·h t-1 +W ix ·x t +b i )

[0060]

[0061]

[0062] O t =σ(W Oh ·h t-1 +W Ox ·x t +b O )

[0063] h t =O t tanh(C t )

[0064] In the formula, f is the forget gate, i is the input gate, O is the output gate, σ is the Sigmoid function, and h is t-1 is the output at time t-1, x t is the deep displacement characteristic data of the landslide input at time t, W is the weight of each data, b is each bias term, C t is the cell state at time t, C t-1 is the cell state at time t-1, h t is the output at time t;

[0065] Step 5: Use the optimal parameters of the CNN-LSTM method in step 4 as the final parameters of the model, and input the test set into the CNN-LSTM method. The prediction result obtained is the final prediction result of CNN-LSTM.

[0066] Furthermore, the data interpolation and outlier processing in step 1 specifically include the following steps:

[0067] Step 1-1: Determine the time period with missing data based on the chronological order;

[0068] Step 1-2: By constructing a multivariate nonlinear equation, the missing data is assumed to be the equation data, the number of terms in the equation and the parameters of each term are determined through time sequence, and the data values ​​are supplemented. The equation is as follows:

[0069] y=a n x n +a n-1 x n-1 +…+a 1 x+b

[0070] In the formula, y is the missing data to be predicted, x is the time data, n is the number of terms in the equation, and a n For x n Parameters;

[0071] Step 1-3: Process outliers through discrete linear convolution. The process is as follows:

[0072] g(i)=[a]*[v i ]

[0073]

[0074] In the formula, g(i) is the replaced data, [a] is the one-dimensional convolution kernel, * is the convolution symbol, [v i ] is the original data in the convolution area, and m is the length of the convolution kernel;

[0075] Furthermore, the loss function in step 4 is the mean square loss function, and the calculation formula is as follows:

[0076] loss(x i ,y i )=(x i -y i ) 2

[0077] In the formula, x i Output data value for the LSTM method, y i is the actual data value;

[0078] Furthermore, the parameter optimization method in step 4 is the Adam method, and the optimized parameters include the two-dimensional convolution kernel [w] of CNN in step 3 and the data weights W in LSTM in step 4.

[0079] Furthermore, the number of iterations of the CNN-LSTM method in step 4 can be adjusted according to different input data.

[0080] like Figure 1As shown, the method for predicting deep displacement of landslide under the influence of groundwater migration based on CNN-LSTM in this embodiment specifically includes the following steps:

[0081] Step 1: Collect landslide data, including deep displacement data and groundwater level data. The data relationship is as follows: Figure 2 , Figure 3 As shown in the figure, due to external interference, the original landslide data often have missing data and data anomalies. Before data prediction, data interpolation and outlier processing are performed. The time period of data missing is determined according to the time sequence. By constructing a multivariate nonlinear equation, the missing data is assumed to be the data of the equation. The number of terms and parameters of each term in the equation are determined according to the time sequence, and the data values ​​are supplemented. The outliers are processed by discrete linear convolution. The effects of data interpolation and outlier processing are shown in the figure. Figure 4 , Figure 5 shown.

[0082] Step 2: Construct training data. Merge the deep displacement data and groundwater level data processed in step 1 into a data matrix in chronological order. Use the first 80% of the data as the training set and the last 20% of the data as the test set.

[0083] Step 3: Use the training set in step 2 as CNN input data, and use the CNN method to extract and fuse the data features of deep displacement data and groundwater level data.

[0084] Step 4: Use the feature data output by CNN in step 3 as LSTM input data. The LSTM structure is as follows: Figure 6 As shown in the figure, the output data is the prediction result of the training set, and the loss function is calculated based on the prediction result. The parameters of CNN and LSTM are optimized, and the optimal parameters of the CNN-LSTM method are determined by iteration. The loss function is the mean square loss function, and the parameter optimization method is the Adam method. The number of iterations can be adjusted according to the different input data.

[0085] Step 5: Use the optimal parameters of the CNN-LSTM method in step 4 as the final parameters of the model, and input the test set into the CNN-LSTM method. The prediction result obtained is the final prediction result of CNN-LSTM. The prediction result is compared with other prediction models. Figure 7 , Figure 8 shown.

[0086] In summary, the present invention performs data interpolation and outlier processing on the original data. The multivariate nonlinear interpolation method and discrete linear convolution method used require less data computing power and have a simple structure. They can effectively supplement missing values ​​in the data and modify outliers in the data, thereby improving data quality and indirectly improving prediction accuracy.

[0087] The present invention adopts the CNN method, through convolution operation, using one or more convolution kernels to extract different types of features, which can well explore the potential relationship between the deep displacement of the landslide and the groundwater level.

[0088] The present invention adopts the LSTM method. LSTM has long short-term memory and can well predict long-term time series. Compared with traditional deep displacement prediction, it avoids the errors caused by data decomposition and improves data utilization. The model structure is simple, the modeling time is short, and the modeling cost is low.

[0089] When the CNN-LSTM method is used to predict the deep displacement of landslides, the influence of the groundwater level is taken into account. The rise and fall of the groundwater level is closely related to the deep displacement of the landslide. Predicting the deep displacement under the premise of considering the groundwater level is conducive to improving the prediction accuracy, and only adding groundwater level data will not excessively increase the calculation cost.

[0090] The description presented in the above exemplary embodiments is only used to illustrate the technical solution of the present invention, and is not intended to be exhaustive, nor is it intended to limit the present invention to the precise form described. Obviously, it is possible for a person of ordinary skill in the art to make many changes and variations based on the above teachings. The exemplary embodiments are selected and described to explain the specific principles of the present invention and its practical application, so that other technicians in the field can easily understand, implement and use the various exemplary embodiments of the present invention and its various selected forms and modified forms. The scope of protection of the present invention is intended to be defined by the attached claims and their equivalent forms.

Claims

1. A method for predicting deep displacement of landslides under the influence of groundwater migration based on CNN-LSTM, characterized in that: The steps include: Step 1: Collect landslide data, including deep displacement data and groundwater level data. Due to external interference, the original landslide data often has data missing and data anomalies. Data interpolation and outlier processing are performed before data prediction. Step 2: construct training data, merge the deep displacement data and groundwater level data processed in step 1 into a data matrix in chronological order, use the first 80% of the data as a training set, and the last 20% of the data as a test set; Step 3: Use the training set in step 2 as CNN input data, and use the CNN method to extract and fuse the data features of deep displacement data and groundwater level data. The CNN process is as follows: Y(i,j)=δ([w]*[X i,j ]+b) In the formula, Y(i,j) is the feature data output by CNN, δ is the tanh activation function, * is the convolution symbol, [w] is the two-dimensional convolution kernel, [X i,j ] is the deep displacement of the landslide and the groundwater level in the convolution area, m is the number of rows where the data is located, n is the number of columns where the data is located, l is the convolution kernel moving step, and b is the bias parameter; Step 4: Use the feature data output by CNN in step 3 as LSTM input data, and the output data is the prediction result of the training set. The loss function is calculated based on the prediction result, and the parameters of CNN and LSTM are optimized. The optimal parameters of the CNN-LSTM method are determined by iteration. The LSTM process is as follows: f t =σ(W fh ·h t-1 +W fx ·x t +b f ) i t =σ(W ih ·h t-1 +W ix ·x t +b i ) O t =σ(W Oh ·h t-1 +W Ox ·x t +b O ) h t =O t ·tanh(C t ) In the formula, f is the forget gate, i is the input gate, O is the output gate, σ is the Sigmoid function, and h is t-1 is the output at time t-1, x t is the deep displacement characteristic data of the landslide input at time t, W is the weight of each data, b is each bias term, C t is the cell state at time t, C t-1 is the cell state at time t-1, h t is the output at time t; Step 5: The optimal parameters of the CNN-LSTM method in step 4 are used as the final parameters of the model, and the test set is input into the CNN-LSTM method. The obtained prediction result is the final prediction result of CNN-LSTM.

2. The method for predicting deep displacement of landslide under the influence of groundwater migration based on CNN-LSTM as claimed in claim 1, characterized in that: The data interpolation and outlier processing described in step 1 specifically include the following steps: Step 1-1: Determine the time period with missing data based on the chronological order; Step 1-2: By constructing a multivariate nonlinear equation, the missing data is assumed to be the equation data, the number of terms in the equation and the parameters of each term are determined through time sequence, and the data values ​​are supplemented. The equation is as follows: y=a n x n +a n-1 x n-1 +…+a1x+b In the formula, y is the missing data to be predicted, x is the time data, n is the number of terms in the equation, and a n For x n Parameters; Step 1-3: Process outliers through discrete linear convolution. The process is as follows: g(i)=[a]*[v i ] In the formula, g(i) is the replaced data, [a] is the one-dimensional convolution kernel, * is the convolution symbol, [v i ] is the original data in the convolution area, and m is the length of the convolution kernel.

3. The method for predicting deep displacement of landslide under the influence of groundwater migration based on CNN-LSTM as claimed in claim 1, characterized in that: The loss function in step 4 is a mean square loss function, and the calculation formula is as follows: loss(x i ,y i )=(x i -y i ) 2 In the formula, x i Output data value for the LSTM method, y i is the actual data value.

4. The method for predicting deep displacement of landslide under the influence of groundwater migration based on CNN-LSTM as claimed in claim 1, characterized in that: The parameter optimization method in step 4 is the Adam method, and the optimized parameters include the two-dimensional convolution kernel [w] of CNN in step 3 and the data weights W in LSTM in step 4.

5. The method for predicting deep displacement of landslide under the influence of groundwater migration based on CNN-LSTM as claimed in claim 1, characterized in that: The number of iterations of the CNN-LSTM method in step 4 can be adjusted according to different input data.

Images