A landslide displacement dynamic prediction method based on MACSA-BiLSTM

Abstract

The application provides a landslide displacement dynamic prediction method based on MACSA-BiLSTM, and belongs to the field of landslide prediction; the method comprises the following steps: obtaining landslide displacement, setting relevant parameters and preprocessing; the displacement after preprocessing is decomposed through ensemble empirical mode decomposition (EEMD) to obtain a periodic term and a trend term; for the trend term, the optimal order of polynomial fitting of the trend term is calculated, the polynomial is fitted with the optimal order, and the trend term is predicted; for the periodic term, a new multi-head attention crow search algorithm (MACSA) is used to optimize the hyperparameters of a bidirectional long short-term memory neural network (BiLSTM), and the hyperparameters after optimization are used to establish the bidirectional long short-term memory neural network to predict the periodic term; the predicted total displacement is obtained by combining the predicted trend term and the periodic term; and the residual rate required for the next period prediction is calculated through the predicted total displacement and the next prediction period displacement after preprocessing.

Description

Technical Field

[0001] This invention relates to the field of landslide prediction, and more specifically, to a dynamic prediction method for landslide displacement based on MACSA-BiLSTM. Background Technology

[0002] Landslides are the most common type of geological hazard. Landslide displacement is the most important factor in landslide prevention and early warning systems, making landslide displacement prediction crucial.

[0003] With the development of big data and artificial intelligence technologies, deep learning methods are widely used in various fields. Among them, Bidirectional Long Short-Term Memory (BiLSTM) neural networks are widely used due to their advantages in handling time-series problems. However, BiLSTM network models have multiple hyperparameters, and improper settings can directly affect prediction results. Therefore, establishing appropriate optimization algorithms to optimize its hyperparameters and balancing the relationship between accuracy and computational time cost is of great practical significance for improving prediction accuracy and response speed, and increasing emergency response time after early warning. Summary of the Invention

[0004] The purpose of this invention is to address the problems of insufficient accuracy and low efficiency in existing landslide displacement prediction models by providing a dynamic landslide displacement prediction method based on MACSA-BiLSTM. This method can predict landslide displacement in different future periods based on historical landslide displacement monitoring data, thereby providing a scientific basis for landslide prevention and early warning.

[0005] This invention is implemented as follows:

[0006] A dynamic landslide displacement prediction method based on MACSA-BiLSTM includes the following steps:

[0007] Step 1: Obtain landslide displacement data sequence and set relevant parameters for landslide displacement prediction;

[0008] Step 2: Perform data preprocessing on the landslide displacement data series, calculate the residual rate of the previous prediction period, and perform ensemble empirical mode decomposition (EEMD) on the preprocessed data to decompose the landslide displacement data series into trend displacement series and periodic displacement series.

[0009] Step 3: Obtain the optimal order for fitting the trend term using a polynomial, perform polynomial fitting on the trend term displacement sequence using the optimal order, and predict the trend term displacement for the next period.

[0010] Step 4: Combine the periodic term shift sequence and optimize the hyperparameters of the bidirectional long short-term memory (BiLSTM) neural network using the multi-head attention crow search (MACSA) optimization algorithm;

[0011] Step 5: Based on the optimal hyperparameters obtained from the optimization, establish a BiLSTM prediction network to predict the periodic term displacement of the next period;

[0012] Step 6: Add the predicted trend term displacement to the predicted period term displacement to obtain the predicted total displacement;

[0013] Step 7: Add the next monitoring data and repeat steps 2 to 6. Each time the cycle is repeated, the total displacement for the next prediction period can be predicted.

[0014] Furthermore, in step one, the relevant parameters for landslide displacement prediction include prediction period L, maximum order of polynomial fitting D, total number of MACSA initial crows N, number of MACSA tracked crows M, MACSA-BiLSTM start-up residual rate R, and crow warning value AP.

[0015] Furthermore, in step two, the data preprocessing starts with the latest monitoring data, calculates the sum of displacements for every L monitoring data points, and transforms the original monitoring displacement length sequence into 1 / L of the original length. If the original length is not divisible by L, the data that is not divisible by L and is furthest from the current monitoring time is deleted.

[0016] Furthermore, in step two, the residual rate is calculated using the total displacement predicted in the previous cycle and the actual displacement data after preprocessing in the current cycle, i.e.:

[0017] C xh =(V xhp - V xh ) / V xh

[0018] in C xh The residual rate of the previous cycle. V xhp This represents the total displacement predicted in the previous cycle. V xh This is the actual displacement data after preprocessing in the current loop. The residual rate is initially set to 0 in the first loop.

[0019] Furthermore, in step three, the polynomial fitting is performed by calculating the root mean square error (RMSE) of the second to the Dth order fittings based on the maximum order D of the polynomial fitting, and automatically selecting the order with the lowest RMSE as the final order of the polynomial fitting.

[0020] Furthermore, in step four, the MACSA optimization algorithm includes an initialization phase, a food-hiding phase, a target identification phase, a tracking phase, and a boundary determination phase, as follows:

[0021] During the initialization phase, the population positions are initialized using a logistic-tent mixed chaotic mapping, the mathematical expression of which is:

[0022] X(i,j) =lb( j )+ X 0 (i,j) × μ ×(1- X 0 (i,j) ) ×(ub( j )- lb( j )) when i ≤N / 2

[0023] X(i,j) =lb( j )+ X 0 (i,j) / a ×(ub( j )- lb( j )) when i >N / 2 and X 0 (i,j) <a

[0024] X(i,j) =lb( j )+(1- X 0 (i,j) ) / (1- a )×(ub( j )- lb( j )) when i >N / 2 and X 0 (i,j) ≥a

[0025] Where N is the total number of crows initialized by MACSA. i The serial number of the crow. j To find the index of the dimension, X 0 (i,j) It is the first i Only a crow in the dimension j The random value is between 0 and 1. μ The range ∈ [0,4] is called the Logistic parameter.a Let ub( be the chaos coefficient) j ) as a dimension j The upper bound of the solution domain on lb( j ) as a dimension j Lower bound of the solution domain , X(i,j) For the first i Only crows in the first j Logistic-tent chaotic mapping values ​​in each dimension;

[0026] During the food storage phase, the number of crows is calculated in each cycle. i Fitness value, preserve crow i The historical best fitness value is used, and the corresponding bit value of this best fitness is used as the crow's position. i The location of the food storage;

[0027] In the phase of determining the tracking targets, the historical best fitness values ​​of each crow are compared, and the M crows with the best historical best fitness values ​​are selected as the tracking targets. The higher the fitness value ranking, the greater the probability of being tracked. The M tracking crows are arranged in descending order of fitness, and their indexes are defined as M. w (M w =1,2,…,M), then the first i Only the crows tracked the Mth w The probability of a crow P iw for:

[0028] P iw =(1 / 2) Mw / [(1 / 2)+(1 / 2) 2 +(1 / 2) 3 +…+(1 / 2) M ]

[0029] During the tracking phase, each crow i According to its selected tracked crow M w The tracking crows will try to locate the feeding grounds of other crows while simultaneously considering their own feeding locations. The tracked crows will then detect that they are being followed and begin flying randomly to confuse the other crows. i Then each crow i The location was updated after tracking:

[0030] X t+1 (i,j) = X t (i,j) +r×(c1× rand × X t mem(i,j) +c2× rand × X t mem (w,j) ) when A >AP

[0031] X t+1 (i,j) = lb( j )+ rand ×(ub( j )- lb( j )) when A ≤AP

[0032] in, X t+1 (i,j) For crows i After the location is updated, that is t+1 In Dimensions j Location, X t (i,j) Before updating the position of crow i, i.e. t In Dimensions j The position is given by r, the flight stride is given by r, and c1 and c2 are attention coefficients, with c1 + c2 = 1. rand A random number between (0,1) X t mem (i,j) for t Time Crow i In dimensions j Food storage location, X t mem (w,j) for t Crows that are constantly being tracked w In dimensions j Food storage location A It is a random value within (0,1), and AP is the crow M. w The alert value is the value of the crow that discovers other crows are tracking it. The value is between (0,1), where a smaller value represents a lower alert level and a larger value represents a higher alert level.

[0033] During the boundary determination phase, the judgment is made as follows: X t+1 (i,j) In dimensions j Is the location within the search boundary? If it is within the boundary, then... X t+1 (i,j) Perform dimensionalj The position is updated, and if it exceeds the search boundary, it is updated in the dimension. j Keep X t (i,j) The position remains unchanged.

[0034] Furthermore, in step four, the hyperparameters of the Bidirectional Long Short-Term Memory (BiLSTM) neural network are optimized by selecting one or more of the following: number of hidden layer nodes, initial learning rate, learning rate reduction factor, regularization coefficient, random inactivation rate, and batch size. The MACSA method is only used to optimize the BiLSTM prediction network when the absolute value of the residual rate of the previous cycle is greater than the MACSA-BiLSTM start-up residual rate R.

[0035] Furthermore, in step five, the BiLSTM prediction network includes an input layer, a BiLSTM layer, a random deactivation layer, an activation layer, a fully connected layer, and an output layer.

[0036] The beneficial effects of this invention are as follows: Compared with existing landslide displacement prediction methods, this invention adopts a dynamic landslide displacement prediction method based on MACSA-BiLSTM to predict landslide displacement. This method has higher prediction accuracy and higher computational efficiency than traditional dynamic prediction methods, and can predict landslide displacements of different periods, thereby providing more comprehensive and reliable information for the prediction and early warning of landslide geological disasters. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0038] Figure 1 This is a flowchart of the calculation process of the present invention provided by an embodiment of the present invention;

[0039] Figure 2 This is a flowchart of the MACSA-BiLSTM calculation provided in the embodiments of the present invention;

[0040] Figure 3 This is a comparison chart of the predicted displacement curve of the present invention and the predicted displacement curves of different methods, provided by the embodiments of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the implementation examples and accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to represent selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] This embodiment provides a dynamic prediction method for landslide displacement based on MACSA-BiLSTM. The method of the present invention will be described below using a specific landslide case as an example. Example 1

[0043] Taking a landslide displacement measurement case as an example, 800 sets of displacement monitoring data were selected, with each set representing the displacement within one hour. The specific implementation process is as follows:

[0044] (1) Obtain the landslide displacement data sequence and set the relevant parameters for landslide displacement prediction, including prediction period L=5, maximum order of polynomial fitting D=5, total number of MACSA initial crows N=20, number of MACSA tracked crows M=4, MACSA-BiLSTM start-up residual rate R=0.1, and crow warning value AP=0.1;

[0045] (2) Perform data preprocessing on the landslide displacement data sequence, calculate the residual rate required for this cycle, and perform ensemble empirical mode decomposition (EEMD) on the preprocessed data to decompose the landslide displacement data sequence into a trend term displacement sequence and a period term displacement sequence.

[0046] (3) Obtain the optimal order for fitting the trend term using a polynomial, fit the trend term displacement sequence using the optimal order using a polynomial, and predict the trend term displacement in the next period.

[0047] (4) Combining the periodic term shift sequence, the hyperparameters of the Bidirectional Long Short-Term Memory (BiLSTM) neural network are optimized using the Multi-Head Attention Raven Search (MACSA) optimization algorithm. The selected hyperparameters of the BiLSTM neural network include the number of hidden layer nodes, the initial learning rate, and the regularization coefficient. The MACSA-BiLSTM model optimization process is as follows: Figure 2 As shown;

[0048] The MACSA optimization algorithm includes an initialization phase, a food-hiding phase, a target identification phase, a tracking phase, and a boundary determination phase. The process is as follows:

[0049] During the initialization phase, the population positions are initialized using a logistic-tent mixed chaotic mapping, the mathematical expression of which is:

[0050] X(i,j) =lb( j )+ X 0 (i,j) × μ ×(1- X 0 (i,j) ) ×(ub( j )- lb( j )) when i ≤N / 2

[0051] X(i,j) =lb( j )+ X 0 (i,j) / a ×(ub( j )- lb( j )) when i >N / 2 and X 0 (i,j) <a

[0052] X(i,j) =lb( j )+(1- X 0 (i,j) ) / (1- a )×(ub( j )- lb( j )) when i >N / 2 and X 0 (i,j) ≥a

[0053] Where N=20, i The serial number of the crow. j In this implementation case, to solve for the dimension index, j=1,2,3 , X 0 (i,j) It is the first i Only a crow in the dimension j The random value is between 0 and 1. μ The range ∈ [0,4] is called the Logistic parameter. aLet ub( be the chaos coefficient) j ) as a dimension j The upper bound of the solution domain is defined in this implementation case, where the upper bounds for the number of hidden layer nodes, the initial learning rate, and the regularization coefficient are 150, 0.3, and 0.01, respectively. ， lb( j ) as a dimension j The lower bound of the solution domain is given above. In this implementation, the lower bounds for the number of hidden layer nodes, the initial learning rate, and the regularization coefficient are 70, 0.01, and 1e-10, respectively. , X(i,j) For the first i Only crows in the first j Logistic-tent chaotic mapping values ​​in each dimension;

[0054] During the food storage phase, the number of crows is calculated in each cycle. i Fitness value, preserve crow i The historical best fitness value is used, and the corresponding bit value of this best fitness is used as the crow's position. i The location of the food storage;

[0055] In the phase of determining the tracking targets, the historical best fitness values ​​of each crow are compared, and the M=4 crows with the best historical best fitness values ​​are selected as the tracking targets. The higher the fitness value ranking, the greater the probability of being tracked. Let M be the index of the M=4 crows to be tracked. w (M w =1,2,…,M), then the first i Only the crows tracked the Mth w The probability of a crow P iw for:

[0056] P iw =(1 / 2) Mw / [(1 / 2)+(1 / 2) 2 +(1 / 2) 3 +…+(1 / 2) M ]

[0057] During the tracking phase, each crow i According to its selected tracked crow M w The tracking crows will try to locate the feeding grounds of other crows while simultaneously considering their own feeding locations. The tracked crows will then detect that they are being followed and begin flying randomly to confuse the other crows. i Then each crow i The location was updated after tracking:

[0058] X t+1 (i,j) = Xt (i,j) +r×(c1× rand × X t mem (i,j) +c2× rand × X t mem (w,j) ) when A >AP

[0059] X t+1 (i,j) = lb( j )+ rand ×(ub( j )- lb( j )) when A ≤AP

[0060] in, X t+1 (i,j) For crows i After the location is updated, that is t+1 In Dimensions j Location, X t (i,j) Before updating the position of crow i, i.e. t In Dimensions j The position is given by r, the flight stride is given by r, and c1 and c2 are attention coefficients, with c1 + c2 = 1. rand A random number between (0,1) X t mem (i,j) for t Time Crow i In dimensions j Food storage location X t mem (w,j) for t Crows that are constantly being tracked w In dimensions j Food storage location A It is a random value within (0,1), and AP is the crow M. w The alert value is the value of the crow that discovers other crows are tracking it. The value is between (0,1), where a smaller value represents a lower alert level and a larger value represents a higher alert level.

[0061] During the boundary determination phase, the judgment is made as follows: X t+1 (i,j)In dimensions j Is the location within the search boundary? If it is within the boundary, then... X t+1 (i,j) Perform dimensional j The position is updated, and if it exceeds the search boundary, it is updated in the dimension. j Keep X t (i,j) The position remains unchanged.

[0062] (5) Based on the optimal hyperparameters obtained from the optimization, a BiLSTM prediction network is established to predict the periodic term displacement of the next period.

[0063] (6) Add the predicted trend term displacement to the predicted period term displacement to obtain the predicted total displacement;

[0064] (7) Add the next monitoring data and repeat steps two to six. Each time the cycle is repeated, the total displacement of the next prediction cycle can be predicted.

[0065] Through the above steps, the 800 hours of landslide displacement data, after preprocessing, became 160 sets of periodic data. Each set of data represents the displacement over a 5-hour period. This implementation case predicts the periodic data from set 121 to 160, repeating the cycle 40 times. To demonstrate the advantages of this invention, the prediction results were compared with those of BiLSTM networks and CSA-BiLSTM networks, as follows: Figure 3 As shown. Simultaneously, the root mean square error (RMSE), absolute percentage error (MAPE), mean absolute error (MAE), and coefficient of determination (R²) were used. 2 Four indicators are used to evaluate the accuracy of the prediction model. MAE is the magnitude of the absolute error; MAPE is the average deviation between the predicted and measured values ​​(the closer MAPE is to 0, the better the performance); RMSE is the deviation between the predicted and actual values ​​(the closer RMSE is to 0, the closer the predicted values ​​are to the actual values). R 2 The closer the value is to 1, the more accurate the prediction. The evaluation results are shown in Table 1:

[0066]

[0067] It is evident that the prediction accuracy of this invention is significantly superior to that of traditional BiLSTM and CSA-BiLSTM networks. The results of this invention are of great practical significance for improving the accuracy and response speed of landslide geological hazard prediction and increasing emergency response time after early warning.

Claims

1. A method for dynamic prediction of landslide displacement based on MACSA-BiLSTM, characterized in that, Includes the following steps: Step 1: Obtain landslide displacement data sequence and set relevant parameters for landslide displacement prediction; Step 2: Perform data preprocessing on the landslide displacement data sequence, calculate the residual rate of the previous prediction period, and perform ensemble empirical mode decomposition on the preprocessed data to decompose the landslide displacement data sequence into trend term displacement sequence and period term displacement sequence. Step 3: Obtain the optimal order for fitting the trend term using a polynomial, perform polynomial fitting on the trend term displacement sequence using the optimal order, and predict the trend term displacement for the next period. Step 4: Combine the periodic term shift sequence and optimize the hyperparameters of the bidirectional long short-term memory neural network using the multi-head attention crow search optimization algorithm; Step 5: Based on the optimal hyperparameters obtained from the optimization, establish a bidirectional long short-term memory prediction network to predict the periodic term shift for the next period; Step 6: Add the predicted trend term displacement to the predicted period term displacement to obtain the predicted total displacement; Step 7: Add the next monitoring data and repeat steps 2 to 6. Each time the cycle is repeated, the total displacement for the next prediction period can be predicted. The optimization process of the multi-head attention crow search algorithm is as follows: During the initialization phase, a logistic-tent hybrid chaotic mapping is used to initialize the population positions; During the foraging phase, the fitness value of each crow i is calculated, the historical best fitness value of crow i is saved, and the position corresponding to the best fitness value is used as the foraging position of crow i. In the phase of determining the tracking targets, the M crows with the best historical fitness values ​​are selected as the tracking targets. During the tracking phase, the position of each crow i is updated after tracking as follows: X t+1 (i,j) = X t (i,j) + r × (c1 × rand × X t mem (i,j) + c2 × rand × X t mem (w,j)) when A > AP X t+1 (i,j) = lb(j) + rand × (ub(j) - lb(j)) when A ≤ AP Among them, X t+1 (i,j) and X t (i,j) represent the positions of crow i in dimension j before and after the update, respectively; r is the flight stride; c1 and c2 are the attention coefficients, and c1+c2=1; rand is a random number between (0,1); X t mem (i,j) and X t mem (w,j) represent the food storage locations of crow i and the tracked crow w at time t in dimension j, respectively. A is a random value within (0,1), and AP is the alert value for the crow to discover that other crows are tracking it, with a value between (0,1). In the boundary determination stage, determine X t+1 (i,j) Determine if the position at dimension j is within the search boundary. If it is, press X. t+1 Update the position of dimension j for (i,j); otherwise, keep the position of X in dimension j. t The positions (i,j) remain unchanged.

2. The landslide displacement dynamic prediction method based on MACSA-BiLSTM as described in claim 1, characterized in that, In step one, the relevant parameters for landslide displacement prediction include prediction period L, maximum order of polynomial fitting D, total number of MACSA initial crows N, number of MACSA tracked crows M, MACSA-BiLSTM start-up residual rate R, and crow warning value AP.

3. The landslide displacement dynamic prediction method based on MACSA-BiLSTM as described in claim 1, characterized in that, In step two, the data preprocessing starts with the latest monitoring data, calculates the sum of displacements for every L monitoring data points, and transforms the original monitoring displacement length sequence into 1 / L of the original length. If the original length is not divisible by L, the data that is not divisible by L and is furthest from the current monitoring time is deleted.

4. The landslide displacement dynamic prediction method based on MACSA-BiLSTM as described in claim 1, characterized in that, In step two, the residual rate is calculated using the total displacement predicted in the previous cycle and the actual displacement data after preprocessing in the current cycle, i.e.: C xh =(V xhp - V xh ) / V xh Where C xh V represents the residual rate from the previous cycle. xhp V represents the total displacement predicted in the previous cycle. xh This is the actual displacement data after preprocessing in the current loop. The residual rate is initially set to 0 in the first loop.

5. The landslide displacement dynamic prediction method based on MACSA-BiLSTM as described in claim 1, characterized in that, In step three, the polynomial fitting is performed by calculating the root mean square error of the second to the Dth order of the polynomial fitting based on the maximum order D of the polynomial fitting, and automatically selecting the order with the lowest root mean square error as the final order of the polynomial fitting.

6. The landslide displacement dynamic prediction method based on MACSA-BiLSTM as described in claim 1, characterized in that, The mathematical expression for the initialization of the logistic-tent mixed chaotic mapping is: X(i,j)=lb(j)+X0(i,j)×μ×(1- X0(i,j)) ×(ub(j)- lb(j)) when i≤N / 2 X(i,j)=lb(j)+X0(i,j) / a ×(ub(j)- lb(j)) When i>N / 2 and X0(i,j)<a X(i,j) = lb(j) + (1 - X0(i,j)) / (1 - a) × (ub(j) - lb(j)) when i > N / 2 and X0(i,j) ≥ a Where N is the total number of crows initialized by MACSA, X0(i,j) is the random value of the i-th crow in dimension j, with a size between (0,1), μ∈[0,4] is called the Logistic parameter, a is the chaos coefficient, ub(j) and lb(j) are the upper and lower bounds of the solution domain in dimension j, respectively, and X(i,j) is the logistic-tent chaotic mapping value of the i-th crow in the j-th dimension.

7. The landslide displacement dynamic prediction method based on MACSA-BiLSTM as described in claim 1, characterized in that, In step four, the hyperparameters of the bidirectional long short-term memory neural network are optimized by selecting one or more parameters from the following: number of hidden layer nodes, initial learning rate, learning rate reduction factor, regularization coefficient, random inactivation rate, and batch size. The MACSA method is only used to optimize the long short-term memory neural network when the absolute value of the residual rate of the previous cycle is greater than the MACSA-BiLSTM startup residual rate R.

8. The landslide displacement dynamic prediction method based on MACSA-BiLSTM as described in claim 1, characterized in that, In step five, the long short-term memory neural network includes an input layer, a long short-term memory layer, a random deactivation layer, an activation layer, a fully connected layer, and an output layer.

Images