An intelligent displacement prediction method with embedded characteristic information of wading landslide
By constructing functional blocks of trend, period and step terms, embedded feature information of wading landslides, the intelligent displacement prediction method solves the problem of insufficient accuracy in traditional methods when predicting displacement of wading landslides, and achieves higher prediction accuracy and reliability.
Patent Information
- Application Number
- CN202311863146.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-29
- Publication Date
- 2025-05-06
- Estimated Expiration
- 2043-12-29
AI Technical Summary
When traditional landslide displacement prediction methods deal with wading landslides, it is difficult to accurately capture the effects of reservoir water level changes and short-term heavy rainfall, resulting in a decrease in prediction accuracy.
Using an intelligent displacement prediction method with embedded water-shaped landslide feature information, the trend, period and step changes of landslide displacement are predicted respectively by constructing functional blocks 1 (trend term), functional block 2 (period term) and functional block 3 (step term), and the prediction results are integrated to improve prediction accuracy.
It improves the accuracy of landslide displacement prediction in the reservoir area, can more accurately capture the various impacts faced by wading landslides, and enhances the reliability of the prediction.
Smart Images
Figure CN118133902B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of landslide prediction, in particular to an intelligent displacement prediction method with embedded characteristic information of wading type landslide. Background Art
[0002] The topography along both sides of the river is complex, geological disasters occur frequently, and large-scale landslides are widely developed. Affected by periodic inducing factors, such as the rise and fall of water levels in the reservoir area and seasonal rainfall, the landslide monitoring displacement curve shows a periodic, step-like growth pattern. In view of the characteristics of the landslide displacement monitoring curve in the reservoir area, the existing technology decomposes the collected landslide displacement curve into periodic terms, trend terms, and residuals by using decomposition algorithms such as moving average method, empirical mode decomposition, variational mode decomposition, etc. And based on the decomposition results, different deep learning algorithms are used to construct the corresponding relationship between periodic terms, trend terms, residuals and time. Finally, the predicted values of periodic terms, trend terms, and residuals are combined to obtain the displacement prediction curve of future landslides.
[0003] The traditional method of combining "decomposition algorithm + deep learning algorithm" will cause data distortion of periodic and trend terms due to the existence of residual terms, and the regularity of residual terms is not obvious, which will affect the accuracy of the results of deep learning algorithms; at the same time, the traditional method only considers periodic and trend terms. When wading landslides are affected by sudden rise / fall of reservoir water level and short-term heavy rainfall, the rising part of the stepped displacement will inevitably show a steep upward trend. Due to the above factors, the traditional method of combining "decomposition algorithm + deep learning algorithm" will reduce the accuracy of landslide displacement prediction. Summary of the invention
[0004] The purpose of the present invention is to provide an intelligent displacement prediction method with embedded water-wading landslide characteristic information, comprising the following steps:
[0005] 1) Construct universal linear and nonlinear layers;
[0006] 2) Encapsulate the linear layer, the nonlinear layer and the trend term function to construct function block 1;
[0007] Encapsulate the linear layer, the nonlinear layer and the periodic term function to construct function block 2;
[0008] Encapsulate the linear layer, the nonlinear layer and the step term function to construct function block 3;
[0009] 3) Obtain the time series data x corresponding to the landslide displacement as the input data x1 of function block 1, and input it into function block 1 to obtain the output backtest value and predicted landslide displacement
[0010] 4) Compare the input data x1 with the output backtest value The residual is used as the input data x2 of function block 2 and is input into function block 2 to obtain the output backtest value and predicted landslide displacement
[0011] 5) Compare the input data x2 with the output backtest value The residual is used as the input data x2 of function block 3 and is input into function block 3 to obtain the predicted value of landslide displacement
[0012] 6) Integrate the landslide displacement predictions of different functional blocks to obtain the final landslide displacement prediction results.
[0013] Further, the nonlinear layer is used to connect the input data with the linear layer;
[0014] Further, the nonlinear layer includes n fully connected layers; n≥4;
[0015] Among them, the output of the first fully connected layer is as follows:
[0016] h L,1 =FC L,1 (x L ) (1)
[0017] Where L represents the Lth block, x L is the input data of the Lth block; FC L,1 represents the nonlinear activation function of the first fully connected layer of the L block; h L,1 represents the output of the first fully connected layer of the L block;
[0018] The output of the i-th fully connected layer is as follows:
[0019] h L,i =FC L,i (h L,i-1 ) (2)
[0020] In the formula, h L,i-1 is the output of the i-1th fully connected layer of block L; h L,i is the output of the i-th fully connected layer of block L; FC L,i represents the nonlinear activation function of the i-1th fully connected layer of block L; i = 2, 3, …n.
[0021] Furthermore, the nonlinear activation function includes but is not limited to: Sigmoid, Tanh, and ReLu.
[0022] Furthermore, when Sigmoid is used as the nonlinear activation function, the outputs of the first fully connected layer and the i-th fully connected layer of the L block are as follows:
[0023] h L,1 =Sigmoid(W L,1 x L +b L,1 ) (3)
[0024] h L,i =Sigmoid(W L,i h L,i-1 +b L,i ) (4)
[0025] Where W L,1 , W L,i is the weight; b L,1 、b L,i is the bias. L,1 is the output of the first fully connected layer of the L block.
[0026] Furthermore, the output of the linear layer is as follows:
[0027]
[0028]
[0029] In the formula, is the output bias result of the linear layer; is a linear activation function. L,n is the output of the nth fully connected layer of the L block.
[0030] Furthermore, the final landslide displacement prediction results are as follows:
[0031]
[0032] In the formula, It represents the final landslide displacement prediction result; They represent the landslide displacement prediction of function block 1, function block 2 and function block 3 respectively.
[0033] Furthermore, the landslide displacement predictions of function block 1, function block 2 and function block 3 are as follows:
[0034]
[0035]
[0036]
[0037] In the formula, vector T = [t p ,t p ,t p ,…,t p]; t is time; p = {0, 1, 2}, is a parameter that controls the speed of linear monotonically increasing or decreasing behavior. When the function block is activated, p will be randomly assigned a value; are the linear layer output results in function block 1, function block 2, and function block 3 respectively; vector H is the forward prediction window length; vector
[0038] Furthermore, the input data x1, input data x2 and input data x3 are as follows:
[0039] x1=x (11)
[0040]
[0041]
[0042] Among them, the backtest value Backtest value They are as follows:
[0043]
[0044]
[0045] In the formula, Output result of the linear layer in function block 1; V1 b Vector V1 composed of historical time data input to function block 1 b =x1; is bias; A vector of historical data input to function block 2. This is the output result of the linear layer in function block 2.
[0046] The technical effect of the present invention is unquestionable. The present invention uses block 1 (trend item), block 2 (cycle item) and block 3 (step item) to predict the trend item, cycle item and step item of the landslide displacement data, and finally merges the data of the three parts together, thereby improving the accuracy of landslide displacement prediction in the reservoir area. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 This is a schematic diagram of displacement changes of wading landslide;
[0048] Figure 2 It is a structural diagram of the intelligent prediction algorithm for displacement of wading landslide with embedded prior information;
[0049] Figure 3 It is an enlarged view of the structure of block 1 (trend item) in the intelligent prediction algorithm structure of wading landslide displacement embedded with prior information;
[0050] Figure 4 It is an enlarged view of the block 2 (periodic term) structure in the intelligent prediction algorithm structure of wading landslide displacement embedded with prior information;
[0051] Figure 5 It is an enlarged view of the block 3 (step term) structure in the intelligent prediction algorithm structure of wading landslide displacement embedded with prior information. DETAILED DESCRIPTION
[0052] The present invention is further described below in conjunction with the embodiments, but it should not be understood that the above subject matter of the present invention is limited to the following embodiments. Without departing from the above technical ideas of the present invention, various substitutions and changes are made according to the common technical knowledge and customary means in the art, which should all be included in the protection scope of the present invention.
[0053] Embodiment 1:
[0054] See also Figures 1 to 5 , an intelligent displacement prediction method with embedded characteristic information of wading landslide, comprising the following steps:
[0055] 1) Construct universal linear and nonlinear layers;
[0056] 2) Encapsulate the linear layer, nonlinear layer and trend term function to construct function block 1;
[0057] Encapsulate the linear layer, the nonlinear layer and the periodic term function to construct function block 2;
[0058] Encapsulate the linear layer, the nonlinear layer and the step term function to construct function block 3;
[0059] 3) Obtain the time series data x corresponding to the landslide displacement as the input data x1 of function block 1, and input it into function block 1 to obtain the output backtest value and predicted landslide displacement
[0060] 4) Compare the input data x1 with the output backtest value The residual is used as the input data x2 of function block 2 and is input into function block 2 to obtain the output backtest value and predicted landslide displacement
[0061] 5) Compare the input data x2 with the output backtest value The residual is used as the input data x2 of function block 3 and is input into function block 3 to obtain the predicted value of landslide displacement
[0062] 6) Integrate the landslide displacement predictions of different functional blocks to obtain the final landslide displacement prediction results.
[0063] The nonlinear layer is used to connect the input data with the linear layer;
[0064] The nonlinear layer includes n fully connected layers; n≥4;
[0065] Among them, the output of the first fully connected layer is as follows:
[0066] h L,1 =FC L,1 (x L ) (1)
[0067] Where L represents the Lth block, x L is the input data of the Lth block; FC L,1 represents the nonlinear activation function of the first fully connected layer of the L block; h L,1 represents the output of the first fully connected layer of the L block;
[0068] The output of the i-th fully connected layer is as follows:
[0069] h L,i =FC L,i (h L,i-1 ) (2)
[0070] In the formula, h L,i-1 is the output of the i-1th fully connected layer of block L; h L,i is the output of the i-th fully connected layer of block L; FC L,i represents the nonlinear activation function of the i-1th fully connected layer of block L; i = 2, 3, …n.
[0071] The nonlinear activation function includes but is not limited to: Sigmoid, Tanh, and ReLu.
[0072] When Sigmoid is used as the nonlinear activation function, the outputs of the first fully connected layer and the i-th fully connected layer of the L block are as follows:
[0073] h L,1 =Sigmoid(W L,1 x L +b L,1 ) (3)
[0074] h L,i =Sigmoid(W L,i h L,i-1 +b L,i ) (4)
[0075] Where W L,1 , W L,i is the weight; b L,1 、b L,i is the bias.L,1 is the output of the first fully connected layer of the L block.
[0076] The output of the linear layer is as follows:
[0077]
[0078]
[0079] In the formula, is the output bias result of the linear layer; is a linear activation function. L,n is the output of the nth fully connected layer of the L block.
[0080] The final landslide displacement prediction results are shown below:
[0081]
[0082] In the formula, It represents the final landslide displacement prediction result; They represent the landslide displacement prediction of function block 1, function block 2 and function block 3 respectively.
[0083] The landslide displacement predictions of function block 1, function block 2 and function block 3 are as follows:
[0084]
[0085]
[0086]
[0087] In the formula, vector T = [t p ,t p ,t p ,…,t p ]; t is time; p = {0, 1, 2}, is a parameter that controls the speed of linear monotonically increasing or decreasing behavior. When the function block is activated, p will be randomly assigned a value; are the linear layer output results in function block 1, function block 2, and function block 3 respectively; vector H is the forward prediction window length; vector
[0088] The input data x1, input data x2 and input data x3 are as follows:
[0089] x1=x (11)
[0090]
[0091]
[0092] Among them, the backtest value Backtest value They are as follows:
[0093]
[0094]
[0095] In the formula, Output result of the linear layer in function block 1; V1 b Vector V1 composed of historical time data input to function block 1 b =x1; is bias; A vector of historical data input to function block 2. This is the output result of the linear layer in function block 2.
[0096] Embodiment 2:
[0097] An intelligent displacement prediction method with embedded characteristic information of wading landslides comprises the following steps:
[0098] 1) Construct universal linear and nonlinear layers;
[0099] 2) Encapsulate the linear layer, nonlinear layer and trend term function to construct function block 1;
[0100] Encapsulate the linear layer, the nonlinear layer and the periodic term function to construct function block 2;
[0101] Encapsulate the linear layer, the nonlinear layer and the step term function to construct function block 3;
[0102] 3) Obtain the time series data x corresponding to the landslide displacement as the input data x1 of function block 1, and input it into function block 1 to obtain the output backtest value and predicted landslide displacement
[0103] 4) Compare the input data x1 with the output backtest value The residual is used as the input data x2 of function block 2 and is input into function block 2 to obtain the output backtest value and predicted landslide displacement
[0104] 5) Compare the input data x2 with the output backtest value The residual is used as the input data x2 of function block 3 and is input into function block 3 to obtain the predicted value of landslide displacement
[0105] 6) Integrate the landslide displacement predictions of different functional blocks to obtain the final landslide displacement prediction results.
[0106] Embodiment 3:
[0107] An intelligent displacement prediction method with embedded characteristic information of wading landslide, the technical content is the same as that of Example 2, further, the nonlinear layer is used to connect the input data with the linear layer; Example 4:
[0108] An intelligent displacement prediction method with embedded characteristic information of wading landslide, the technical content of which is the same as any one of Embodiments 2-3, further, the nonlinear layer includes n fully connected layers; n≥4;
[0109] Among them, the output of the first fully connected layer is as follows:
[0110] h L,1 =FC L,1 (x L ) (1)
[0111] Where L represents the Lth block, x L is the input data of the Lth block; FC L,1 represents the nonlinear activation function of the first fully connected layer of the L block; h L,1 represents the output of the first fully connected layer of the L block;
[0112] The output of the i-th fully connected layer is as follows:
[0113] h L,i =FC L,i (h L,i-1 ) (2)
[0114] In the formula, h L,i-1 is the output of the i-1th fully connected layer of block L; h L,i is the output of the i-th fully connected layer of block L; FC L,i represents the nonlinear activation function of the i-1th fully connected layer of block L; i = 2, 3, …n.
[0115] Embodiment 5:
[0116] An intelligent displacement prediction method with embedded characteristic information of wading landslide, the technical content of which is the same as any one of embodiments 2-4, and further, the nonlinear activation function includes but is not limited to: Sigmoid, Tanh, ReLu.
[0117] Embodiment 6:
[0118] An intelligent displacement prediction method with embedded characteristic information of wading landslide, the technical content is the same as any one of embodiments 2-5. Further, when the nonlinear activation function is Sigmoid, the outputs of the first fully connected layer and the i-th fully connected layer are respectively as follows:
[0119] h L,1 =Sigmoid(W L,1 x L +b L,1 ) (3)
[0120] h L,i =Sigmoid(W L,i h L,i-1 +b L,i ) (4)
[0121] Where W L,1 , W L,i is the weight; b L,1 、b L,i For bias.
[0122] Embodiment 7:
[0123] An intelligent displacement prediction method with embedded characteristic information of wading landslide, the technical content is the same as any one of embodiments 2-6, further, the output of the linear layer is as follows:
[0124]
[0125]
[0126] In the formula, is the output bias result of the linear layer; is a linear activation function.
[0127] Embodiment 8:
[0128] An intelligent displacement prediction method with embedded water-wading landslide characteristic information, the technical content is the same as any one of embodiments 2-7, and further, the final landslide displacement prediction result is as follows:
[0129]
[0130] In the formula, It represents the final landslide displacement prediction result; They represent the landslide displacement prediction of function block 1, function block 2 and function block 3 respectively.
[0131] Embodiment 9:
[0132] An intelligent displacement prediction method with embedded characteristic information of wading landslide, the technical content of which is the same as any one of embodiments 2-8, further, the landslide displacement prediction of function block 1, function block 2 and function block 3 are respectively as follows:
[0133]
[0134]
[0135]
[0136] In the formula, vector T = [t p ,t p ,t p ,…,t p ]; t is time; p = {0, 1, 2}, is a parameter that controls the speed of linear monotonically increasing or decreasing behavior. When the function block is activated, p will be randomly assigned a value; are the linear layer output results in function block 1, function block 2, and function block 3 respectively; vector H is the forward prediction window length; vector
[0137] Embodiment 10:
[0138] An intelligent displacement prediction method with embedded characteristic information of wading landslide, the technical content of which is the same as any one of Embodiments 2-9, and further, the input data x1, input data x2 and input data x3 are as follows:
[0139] x1=x (11)
[0140]
[0141]
[0142] Among them, the backtest value Backtest value They are as follows:
[0143]
[0144]
[0145] In the formula, Output result of the linear layer in function block 1; V1 b Vector V1 composed of historical time data input to function block 1 b =x1; is bias; A vector of historical data input to function block 2. This is the output result of the linear layer in function block 2.
[0146] Embodiment 11:
[0147] An intelligent displacement prediction method with embedded characteristic information of wading landslides. It should be noted in advance that, considering the displacement change of wading slopes in reservoir areas, in addition to the existence of slippage itself, uniform rainfall and reservoir water level fluctuations will cause periodic displacement changes, and short-term heavy rainfall and sudden rise / fall of reservoir water level will cause a sudden increase in landslide displacement. Therefore, this method provides block 1 with trend prediction function, block 2 with periodic prediction function, and block 3 with step prediction function to improve the accuracy of landslide displacement prediction.
[0148] The method described in this embodiment includes the following steps:
[0149] According to the periodic variation characteristics of landslide monitoring data, the sizes of the backward prediction window and the forward prediction window are set, and the backward prediction window is 2 to 5 times the forward prediction window.
[0150] Input the divided data set into the pre-built model for training; use MSE mean square error as the evaluation indicator of the model;
[0151] 2-1) The structure of the intelligent algorithm model for wading landslide displacement with embedded prior information is detailed in Figure 2 As shown, the entire algorithm model consists of a stack, which includes 3 blocks;
[0152] 2-2) Block 1 is responsible for predicting the landslide displacement trend term, and its structure is as follows: Figure 3 As shown, block 1 is connected through a nonlinear layer and a linear layer, and the forward prediction coefficients are output through the linear layer and the backward prediction coefficient The calculation formula is as follows:
[0153] h 1,1 =FC 1,1 (x1) (1)
[0154] h 1,2 =FC 1,2 (h 1,1 ) (2)
[0155] h 1,3 =FC 1,3 (h 1,2 ) (3)
[0156] h 1,4 =FC 1,4 (h 1,3 ) (4)
[0157]
[0158]
[0159] The non-linear activation functions include but are not limited to: Sigmoid, Tanh, ReLu, such as: h 1,1 =Sigmoid(W 1, 1x1+b 1,1 ), W 1,1 and b 1,1 The parameters set when building the neural network will be automatically adjusted as the network iterates;
[0160] Predicted value and backtest value It is built into the algorithm model through the following formula:
[0161]
[0162]
[0163] Among them, the vector T = [t p ,t p ,t p ,…,t p ]; t is time; p = {0, 1, 2}, is a small value of custom change, used to control the speed of linear monotonous increasing or decreasing behavior, so that the trend changes slowly; when the function block is activated, p will be randomly assigned a value;
[0164] 2-3) Block 2 is responsible for predicting the landslide displacement period term, and its structure is as follows: Figure 4 As shown, block 2 is connected through a nonlinear layer and a linear layer, and the forward prediction coefficients are output through the linear layer and the backward prediction coefficient The calculation formula is as follows:
[0165] h 2,1 =FC 2,1 (x2) (9)
[0166] h 2,2 =FC 2,2 (h 2,1 ) (10)
[0167] h 2,3 =FC 2,3 (h 2,2 ) (11)
[0168] h 2,4 =FC 2,4 (h 2,3 ) (12)
[0169]
[0170]
[0171] The non-linear activation functions include but are not limited to: Sigmoid, Tanh, ReLu, such as: h 2,1 =Sigmoid(W 2, 1x2+b 2,1 ), W 2,1 and b 2,1 The parameters set when building the neural network will be automatically adjusted as the network iterates;
[0172] The input data of block 2 is the residual backtest value remaining after training through block 1. The calculation formula is as follows:
[0173]
[0174] Predicted value and backtest value It is built into the algorithm model through the following formula:
[0175]
[0176]
[0177] in H is the forward prediction window length; A vector of historical data input for block 2;
[0178] 2-5) Block 3 is responsible for predicting the step term of landslide displacement, and its structure is as follows: Figure 5 As shown, block 3 is connected through a nonlinear layer and a linear layer, and the forward prediction coefficients are output through the linear layer The calculation formula is as follows:
[0179] It is built into the algorithm model through the following formula:
[0180] h 3,1 =FC 3,1 (x3) (10)
[0181] h 3,2 =FC 3,2 (h 3,1 ) (11)
[0182] h 3,3 =FC 3,3 (h 3,2 ) (12)
[0183] h 3,4 =FC 3,4 (h3,3 ) (13)
[0184]
[0185] Among them, FC is a nonlinear RELU activation function, such as: h 3,1 =Sigmoid(W 3,1 x3+b 3,1 ), W 3,1 and b 3,1 The parameters set when building the neural network will be automatically adjusted as the network iterates;
[0186] The input data of block 3 is the residual backtest value remaining after training through block 2. The calculation formula is as follows:
[0187]
[0188] Predicted value It is built into the model algorithm through the following formula:
[0189]
[0190] in H is the forward prediction window length;
[0191] 2-6) By Add them together to get the predicted value of landslide displacement.
[0192]
Claims
1. An intelligent displacement prediction method with embedded characteristic information of wading landslide, characterized in that: The following steps are involved: 1) Construct universal linear and nonlinear layers; 2) Encapsulate the linear layer, nonlinear layer and trend term function to construct function block 1; Encapsulate the linear layer, the nonlinear layer and the periodic term function to construct function block 2; Encapsulate the linear layer, the nonlinear layer and the step term function to construct function block 3; 3) Obtain the time series data x corresponding to the landslide displacement as the input data x1 of function block 1, and input it into function block 1 to obtain the output backtest value and predicted landslide displacement 4) Compare the input data x1 with the output backtest value The residual is used as the input data x2 of function block 2 and is input into function block 2 to obtain the output backtest value and predicted landslide displacement 5) Compare the input data x2 with the output backtest value The residual is used as the input data x2 of function block 3 and is input into function block 3 to obtain the predicted value of landslide displacement 6) Integrate the landslide displacement predictions of different functional blocks to obtain the final landslide displacement prediction results.
2. The intelligent displacement prediction method with embedded characteristic information of wading landslide according to claim 1 is characterized in that: The non-linear layer is used to connect the input data with the linear layer.
3. The intelligent displacement prediction method with embedded characteristic information of wading landslide according to claim 1 is characterized in that: The nonlinear layer includes n fully connected layers; n≥4; Among them, the output of the first fully connected layer is as follows: h L,1 =FC L,1 (x L ) (1) Where L represents the Lth block, x L is the input data of the Lth block; FC L,1 represents the nonlinear activation function of the first fully connected layer of the L block; h L,1 represents the output of the first fully connected layer of the L block; The output of the i-th fully connected layer is as follows: h L,i =FC L,i (h L,i-1 ) (2) In the formula, h L,i-1 is the output of the i-1th fully connected layer; h L,i is the output of the i-th fully connected layer; FC L,i represents the nonlinear activation function of the i-1th fully connected layer; i = 2, 3, …n.
4. The intelligent displacement prediction method with embedded characteristic information of wading landslide according to claim 3 is characterized in that: The nonlinear activation function includes but is not limited to: Sigmoid, Tanh, and ReLu.
5. The intelligent displacement prediction method with embedded characteristic information of wading landslide according to claim 4 is characterized in that: When Sigmoid is used as the nonlinear activation function, the outputs of the first fully connected layer and the i-th fully connected layer of the L block are as follows: h L,1 =Sigmoid(W L,1 x L +b L,1 ) (3) h L,i =Sigmoid(W L,i h L,i-1 +b L,i ) (4) Where W L,1 , W L,i is the weight; b L,1 , b L,i is the bias; h L,1 is the output of the first fully connected layer of the L block.
6. The intelligent displacement prediction method with embedded characteristic information of wading landslide according to claim 1 is characterized in that: The output of the linear layer is as follows: In the formula, is the output bias result of the linear layer; is a linear activation function; h L,n is the output of the nth fully connected layer of the L block.
7. The intelligent displacement prediction method with embedded characteristic information of wading landslide according to claim 1 is characterized in that: The final landslide displacement prediction results are shown below: In the formula, It represents the final landslide displacement prediction result; They represent the landslide displacement prediction of function block 1, function block 2 and function block 3 respectively.
8. The intelligent displacement prediction method with embedded characteristic information of wading landslide according to claim 7 is characterized in that: The landslide displacement predictions of function block 1, function block 2 and function block 3 are as follows: In the formula, vector T = [t p ,t p ,t p ,…,t p ]; t is time; p = {0, 1, 2}, is the parameter that controls the speed of linear monotonically increasing or decreasing behavior. When the function block is activated, p will be randomly assigned a value; are the linear layer output results in function block 1, function block 2, and function block 3 respectively; vector H is the forward prediction window length; vector 9. The intelligent displacement prediction method with embedded characteristic information of wading landslide according to claim 1 is characterized in that: The input data x1, input data x2 and input data x3 are as follows: x1=x (11) Among them, the backtest value Backtest value They are as follows: In the formula, Output result of the linear layer in function block 1; V1 b =x1 is the vector composed of historical time data input by function block 1; is bias; A vector of historical data input to function block 2. This is the output result of the linear layer in function block 2.
Citation Information
Patent Citations
Landslide displacement prediction method based on improved VMD-TCN
CN116227365A
Landslide displacement prediction method based on attention mechanism and bidirectional gating circulation unit
CN116821586A