Improved white whale optimized BiLSTM-Attention loess landslide displacement prediction method and system
Through the improved white whale optimization algorithm and BiLSTM-Attention model, combined with exponential smoothing method and fuzzy set strategy, the hyperparameters of the loess landslide displacement prediction model are optimized, and the problems of poor prediction effect and complex hyperparameter settings in the existing technology are solved, and the problem of poor prediction effect and complex hyperparameter settings are achieved. High-precision landslide displacement prediction and early warning.
Patent Information
- Application Number
- CN202411019415.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-29
AI Technical Summary
The existing loess landslide displacement prediction model is difficult to capture space-time features when facing complex nonlinear systems, and the hyperparameter settings are complex and easy to fall into local optimality, affecting the prediction effect and efficiency.
The improved beluga whale optimization algorithm (FBWO) combined with the BiLSTM-Attention model is used to decompose landslide displacement through exponential smoothing method, and the hyperparameters are optimized using a multi-strategy enhanced beluga whale algorithm based on fuzzy sets to improve the prediction accuracy and convergence speed of the model.
It significantly improves the accuracy and efficiency of loess landslide displacement prediction, realizes accurate prediction of landslide trends and periodic terms, and enhances the decision-making ability of the early warning system.
Smart Images

Figure CN119047506A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to but is not limited to the technical field of landslide prediction, and in particular relates to an improved Beluga optimized BiLSTM-Attention loess landslide displacement prediction method and system. Background Art
[0002] Loess landslide is a common and serious geological disaster, and has become a highly concerned geological environmental problem in the loess region. Compared with other rock and soil landslides, loess landslides have no obvious sliding surface. In addition, their high frequency of occurrence and complex disaster mechanism make the prediction of loess landslides more difficult.
[0003] The evolution of landslides is affected by a variety of internal and external factors, such as geology, rainfall, and human activities, which makes landslides a complex nonlinear system. The nonlinearity and uncertainty of landslide deformation monitoring make landslide prediction complicated, and loess landslides are no exception. Models for landslide displacement prediction can be divided into two categories: data-driven and physical-driven. Physical-driven models predict landslide events through mechanical principles and geotechnical parameters. Such models are too complex to implement and the prediction effect is not ideal. Data-driven models are more attractive because of their accurate prediction and simple implementation. Deep neural networks have strong nonlinear fitting capabilities and can achieve excellent performance in data-driven modeling. At present, the prediction research on landslide displacement usually uses time series analysis to decompose the landslide displacement time series into different signal components, and then model and predict these signal components separately. For example, the ensemble empirical mode decomposition (EEMD) method is used to decompose the landslide displacement into a fluctuation term displacement containing periodic factors and random factors and a trend term displacement evolved by the internal structure, and then Prophet and LSTM are used to predict the trend term and fluctuation term respectively. However, in deep learning, hyperparameters control the performance and efficiency of model classification and its prediction results need to rely on complex feature engineering. Therefore, many scholars have turned their attention to swarm intelligence optimization algorithms, and use swarm intelligence algorithms to tune model parameters to achieve better landslide prediction results.
[0004] In view of the above analysis, the technical problems that need to be solved urgently in the prior art are:
[0005] 1. At present, most loess landslide displacement prediction models use deep neural networks with relatively simple structures. When faced with a complex nonlinear evolution system such as loess landslide, it is difficult to capture the spatiotemporal characteristics of landslide displacement changes, resulting in poor prediction results. At the same time, the hyperparameter setting of the neural network requires a lot of experiments and finding the optimal value based on manual experience. If the setting is improper, it will greatly affect the prediction effect of the network and increase the computational cost.
[0006] 2. When dealing with complex multimodal problems, swarm intelligence optimization algorithms are prone to fall into local optimality and their convergence speed will slow down, thus affecting the algorithm's solution accuracy and efficiency. Summary of the invention
[0007] In view of the problems existing in the prior art, the present invention provides a method and system for predicting the displacement of loess landslides by improving the Beluga Whale Optimization (BWO) algorithm with a simple structure and easy application, and obtains a multi-strategy enhanced Beluga Whale Optimization (FBWO) algorithm based on fuzzy sets to improve its convergence speed and accuracy when facing complex optimization problems. At the same time, the attention mechanism (Attention) is added to the bidirectional long short-term memory neural network (Bi-directional Long Short-Term Memory, BiLSTM) and the improved Beluga Whale Optimization algorithm is used to adaptively optimize the number of hidden nodes, the maximum number of iterations, the initial learning rate and the L2 regularization coefficient of the network to improve the prediction accuracy of the model.
[0008] The present invention is implemented as follows: an improved Beluga optimized BiLSTM-Attention loess landslide displacement prediction method, comprising:
[0009] S1, using exponential smoothing method to decompose the cumulative displacement of loess landslide into trend term displacement and period term displacement, and divide it into training set and test set;
[0010] S2, establish a BiLSTM-Attention network model, design a multi-strategy enhanced Beluga algorithm (FBWO) based on fuzzy sets, use FBWO to optimize the hyperparameters in the BiLSTM-Attention model, realize the adaptive optimization of the four hyperparameters of the number of hidden layer nodes, initial learning rate, maximum number of iterations and regularization coefficient in the BiLSTM-Attention network, and finally obtain the FBWO-BiLSTM-Attention landslide displacement prediction model to predict the periodic displacement;
[0011] S3, input the trend item displacement data into the partial least squares regression model to obtain the predicted trend item displacement; input the period item displacement data into the FBWO-BiLSTM-Attention prediction model to obtain the predicted value of the period item displacement;
[0012] S4, adding the predicted trend term displacement and period term displacement to obtain the final landslide cumulative displacement prediction result.
[0013] Furthermore, in S1, the exponential smoothing method is used to decompose the accumulated displacement of the landslide into trend term displacement and period term displacement. The calculation formula of the exponential smoothing method is as follows:
[0014]
[0015] Among them, y(i) represents the cumulative displacement data point, y t (i) represents the decomposed trend term displacement data point. The exponential smoothing method is applied three times to obtain the final trend term displacement. The period term displacement is obtained by subtracting the trend term displacement from the cumulative displacement, that is, y p =yy t ,y p represents the periodic displacement, y represents the cumulative displacement, and y t Represents the trend term displacement.
[0016] Furthermore, in S2, the fuzzy set-based multi-strategy enhanced Beluga algorithm (FBWO) is used to optimize the four hyperparameters of the BiLSTM-Attention model, namely, the number of hidden layer nodes, the maximum number of training times, the initial learning rate, and the regularization coefficient. The population size and the maximum number of iterations in the FBWO algorithm are set to 10, and the solution space of the algorithm is set according to the four hyperparameters of BiLSTM-Attention. The upper bound of the solution space is ub = {300, 500, 0.01, 0.01}, and the lower bound is lb = {10, 30, 0.0001, 0.0001}. Each dimension of ub and lb corresponds to the upper and lower bounds of the number of hidden layer nodes, the maximum number of training times, the initial learning rate, and the regularization coefficient of the BiLSTM-Attention model, respectively. The individual position of the Beluga whale is interpreted as a hyperparameter of the BiLSTM-Attention model, and the mean square error (MSE) obtained on the periodic item displacement training set is used as the fitness value of the individual Beluga whale.
[0017] Furthermore, the optimization process of FBWO is as follows:
[0018] (1) Use the Sobol sequence to initialize the beluga population. The calculation formula is as follows:
[0019] X n =L b +S n ×(U b -L b )
[0020] Among them, X n represents a beluga whale individual, L b and U b are the lower and upper bounds of the search space, respectively, S nIt is a random number between [0,1] generated by the Sobol sequence. The initial population generated by the Sobol sequence can be more evenly distributed in the search space, which makes the initial population of the algorithm have higher diversity. After obtaining the initial population, the fitness value of each individual is calculated;
[0021] (2) Constructing a fuzzy population. The beluga population is considered to be a fuzzy set, and the entire population is divided into the optimal set B S and the worst set W S , each beluga individual X i The membership of (i=1,2,…,n) can be obtained from its own fitness. S It is expressed as follows:
[0022]
[0023] in, Represents X i For B S The degree of membership.
[0024] The following will introduce For the maximization problem, if F(X i )=+ve, then If F(X i )=-ve, then For the minimization problem, if F(X i )=+ve, then If F(X i )=-ve, then Among them, F i Represents X i The fitness, F max and F min Respectively represent the maximum fitness and minimum fitness in the entire population, and it can be seen from the above content that
[0025] Likewise, W S The representation is as follows:
[0026]
[0027] in, After obtaining the individual membership, the fuzzy set centroid will be constructed for each fuzzy set. The specific formula is as follows:
[0028]
[0029] X CB and X CW The fuzzy sets B S and WS The center of mass.
[0030] (3) FBWO algorithm uses the balance factor B f To achieve the transition from the exploration phase to the development phase, the calculation formula is as follows:
[0031] B f =B0(1-T / T max )
[0032] Where T is the current iteration number, T max is the maximum number of iterations, and B0 is a random number that changes between (0,1) in each iteration. f >0.5 is the exploration stage, B f When ≤0.5, it is the development stage. f The fluctuation range of will decrease from (0,1) to (0,0.5) as the number of iterations T increases.
[0033] In the exploration phase, the position update formula of the beluga whale is as follows: but
[0034]
[0035] otherwise,
[0036]
[0037] in represents the position of the i-th individual in the S1 dimension, represents the position of the S1th dimension of the global optimal individual, Represents the centroid X CB At the position of the S1 dimension, S2 is the same.
[0038] During the development phase, if but otherwise C1 is the random jump intensity coefficient that controls the Levy flight intensity, L F is the Levy flight function, which is calculated in the same way as the original White Whale optimization algorithm.
[0039] (4) After the position of the beluga individual is updated during the exploration or development phase, it will perform a Cauchy mutation to explore its own neighborhood, thereby improving its local escape ability. The formula for Cauchy mutation is as follows:
[0040]
[0041] Where cauchy is a random number generated by a Cauchy distribution that follows X~C(0,0.5), X rand1 and Xrand2 are two beluga whales randomly selected from the population. is the position before the Cauchy transformation, is the position after Cauchy transformation.
[0042] (5) After the Cauchy mutation, the beluga whale has a certain probability of updating its position during the whale fall phase, which simulates the process of beluga whales falling due to various reasons in nature. The probability of a whale falling is W f , its calculation formula is the same as the original White Whale algorithm. When W f ≥B f The present invention introduces the golden sine strategy into the whale fall stage. The individual position update formula of the white whale in this stage is as follows:
[0043]
[0044] R1 is a random number in the range [0,2π], R2 is a random number in the range [0,π], is the position of the global optimal individual in the jth dimension in the Tth iteration. x1 = a·τ+b·(1-τ), x2 = b·τ+a·(1-τ), where τ is the golden section number The initial values of a and b are -π and π respectively.
[0045] (6) After the position update is completed, the fitness value of each beluga individual is calculated and the optimal individual is found.
[0046] (7) Determine whether the maximum number of iterations has been reached. If not, return to step (2). If the maximum number of iterations has been reached, output the optimal individual of the current population. The position of the optimal individual in each dimension is the optimal setting value of each hyperparameter of BiLSTM-Attention.
[0047] Furthermore, in S3, the prediction result of the trend term displacement is obtained by using partial least squares regression, and the BiLSTM-Attention model optimized by FBWO in S2 is used to predict the periodic term to obtain the prediction result of the periodic term displacement.
[0048] Another object of the present invention is to provide a landslide displacement prediction system for realizing the BiLSTM-Attention loess landslide displacement prediction method based on improved Beluga optimization, comprising:
[0049] The training set and test set division module uses the exponential smoothing method to decompose the cumulative displacement of the loess landslide into trend term displacement and period term displacement, and divides the training set and the test set;
[0050] The loess landslide displacement prediction model establishment module establishes the BiLSTM-Attention network model, designs the multi-strategy enhanced Beluga algorithm (FBWO) based on fuzzy sets, uses FBWO to optimize the hyperparameters in the BiLSTM-Attention model, and realizes the adaptive optimization of the four hyperparameters of the number of hidden layer nodes, initial learning rate, maximum number of iterations and regularization coefficient in the BiLSTM-Attention network. Finally, the FBWO-BiLSTM-Attention loess landslide displacement prediction model is obtained to predict the periodic displacement.
[0051] The module for obtaining the predicted value of periodic item displacement inputs the trend item displacement data into the partial least squares regression model to obtain the predicted trend item displacement; the module for obtaining the predicted value of periodic item displacement inputs the periodic item displacement data into the FBWO-BiLSTM-Attention prediction model;
[0052] The module for obtaining the cumulative displacement prediction result of loess landslide adds the predicted trend item displacement and period item displacement to obtain the final cumulative displacement prediction result of the landslide.
[0053] Another object of the present invention is to provide a computer device, the computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the BiLSTM-Attention loess landslide displacement prediction method based on improved Beluga optimization.
[0054] Another object of the present invention is to provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the BiLSTM-Attention loess landslide displacement prediction method based on the improved Beluga algorithm.
[0055] Another object of the present invention is to provide an information data processing terminal, which includes the landslide displacement prediction system based on fuzzy sets.
[0056] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:
[0057] First, facing the complex nonlinear loess landslide system, the present invention uses the exponential smoothing method to decompose the cumulative displacement of the loess landslide into trend term displacement and period term displacement, accurately capturing the overall trend and change law of the landslide.
[0058] In order to improve the prediction accuracy of the BiLSTM-Attention loess landslide displacement model, the present invention combines the swarm intelligence optimization algorithm with the deep neural network, and uses the swarm intelligence algorithm to adaptively optimize the hyperparameters of the deep neural network, which greatly improves the prediction accuracy of the network.
[0059] In order to overcome the problem that traditional algorithms are prone to fall into local optimality and low convergence efficiency, the present invention proposes a multi-strategy improved enhanced Beluga optimization algorithm (FBWO) based on fuzzy sets. First, in this algorithm, the Sobol sequence with low sampling error is used to improve the initial quality of the initialized population and the uniformity of the solution space distribution. A fuzzy centroid collaborative guidance strategy is constructed, and the corresponding membership is obtained through the fitness of the Beluga individual. Then, the population is divided into the best and worst fuzzy sets and all member information is considered to generate two fuzzy centroids. Through the joint guidance of different centroids and the current optimal position, the global optimal and local optimal are learned, and the population diversity is enhanced while the convergence speed is accelerated. At the same time, the mutation operator of the Cauchy distribution is used to perform mutation operations on the updated two types of individual positions to improve the local escape ability. In addition, the golden sine strategy is introduced, and the golden section coefficient is used to continuously narrow the search range, explore high-quality areas, and improve the convergence speed. The alternating update and collaborative cooperation of the four strategies balance the diversity and convergence of the algorithm. Figure 5 This is a comparison of the convergence curves of FBWO and other enhanced Beluga algorithms on the CEC2017 test function. It can be seen from the figure that FBWO converges faster and more accurately than other algorithms.
[0060] The mean absolute error (MAE) and root mean square error (RMSE) of the prediction results of the BiLSTM-Attention model after FBWO optimization are 2.26 and 3.06, respectively, and the coefficient of determination (R 2 ) is 99.29%, which is a significant improvement compared to before optimization.
[0061] Second, the expected benefits and commercial value of the technical solution of the present invention after transformation are as follows: The present invention provides a multi-strategy enhanced Beluga algorithm optimization algorithm based on fuzzy sets to adaptively optimize the hyperparameters of BiLSTM-Attention, and successfully applies it to the displacement prediction of loess landslides, greatly improving the optimization efficiency and optimization accuracy of the Beluga optimization algorithm and the prediction accuracy of the landslide displacement prediction model. The final model can accurately predict the displacement trend of the landslide, can help the loess landslide early warning system to make decisions, and has important reference significance for the prediction and early warning of loess landslides.
[0062] The technical solution of the present invention fills the technical gap in the industry at home and abroad: the present invention introduces the fuzzy set strategy into the White Whale optimization algorithm for the first time, balances the diversity and convergence of the algorithm, and improves the optimization efficiency and accuracy of the algorithm.
[0063] Third, the present invention proposes a BiLSTM-Attention loess landslide displacement prediction method and system based on the improved Beluga algorithm. First, in step S1, the cumulative displacement of the landslide is decomposed into trend term displacement and period term displacement using the exponential smoothing method, and the training set and the test set are divided. The final trend term displacement is calculated by the triple exponential smoothing method, and the period term displacement is obtained by subtracting the trend term displacement from the cumulative displacement.
[0064] In step S2, a BiLSTM-Attention network model is established, and a multi-strategy enhanced White Whale algorithm (FBWO) based on fuzzy sets is designed to adaptively optimize the hyperparameters of the BiLSTM-Attention model. The FBWO algorithm sets the population size and the maximum number of iterations, and uses the Sobol sequence to initialize the population, constructs a fuzzy population, and calculates the membership degree and constructs the fuzzy set centroid to achieve a balance between the exploration and development phases of the population, thereby optimizing the number of hidden layer nodes, initial learning rate, maximum number of iterations, and regularization coefficient of the BiLSTM-Attention model.
[0065] In step S3, the trend item displacement data is input into the partial least squares regression model to obtain the predicted trend item displacement; the period item displacement data is input into the FBWO-BiLSTM-Attention prediction model to obtain the predicted value of the period item displacement.
[0066] In step S4, the predicted trend term displacement and period term displacement are added to obtain the final landslide cumulative displacement prediction result. This method improves the prediction accuracy by optimizing the BiLSTM-Attention model, solves the technical problems existing in the prior art, and has achieved significant technical progress in industrial applications.
[0067] Fourth, the present invention mainly solves several key technical problems in the displacement prediction of loess landslide. First, the exponential smoothing method is used to decompose the cumulative displacement of loess landslide into trend term displacement and period term displacement, and the cubic exponential smoothing method is used to accurately extract the trend term displacement, thereby effectively reducing the noise and uncertainty in the original data and ensuring the stability and accuracy of the subsequent prediction model.
[0068] Secondly, by building a BiLSTM-Attention network model and designing a fuzzy set-based multi-strategy enhanced White Whale algorithm (FBWO) to adaptively optimize the model's hyperparameters. The FBWO algorithm constructs a fuzzy population, calculates membership, and balances the exploration and development stages. It ultimately optimizes the four hyperparameters of the number of hidden layer nodes, initial learning rate, maximum number of iterations, and regularization coefficient in the BiLSTM-Attention model, thereby improving the prediction accuracy and robustness of the model.
[0069] In terms of mathematical models, the present invention adopts a combination of partial least squares regression model and FBWO-BiLSTM-Attention prediction model. The partial least squares regression model is used to predict the displacement of trend term, while the BiLSTM-Attention model optimized by FBWO uses nine input features to accurately predict the displacement of period term. This combined model effectively combines the advantages of linear and nonlinear prediction methods, significantly improving the overall effect of landslide displacement prediction.
[0070] The present invention has made significant technical progress. By optimizing the BiLSTM-Attention model and the innovative FBWO algorithm, the accuracy and stability of loess landslide displacement prediction are improved, overcoming the problems of low prediction accuracy and difficulty in optimizing model parameters in the prior art. In addition, the present invention has broad prospects in industrial applications and can be used in landslide monitoring and early warning systems to effectively reduce economic losses and casualties caused by landslide disasters. BRIEF DESCRIPTION OF THE DRAWINGS
[0071] Figure 1 It is a flow chart of a BiLSTM-Attention loess landslide displacement prediction method based on an improved Beluga algorithm provided in an embodiment of the present invention;
[0072] Figure 2 It is a comparison curve diagram of the accumulated displacement of the landslide provided by the embodiment of the present invention and the displacement of the trend term and the displacement of the period term obtained by the smoothing index method;
[0073] Figure 3 is a trend term displacement fitting curve diagram obtained by the partial least squares method provided in an embodiment of the present invention;
[0074] Figure 4 is a period term displacement fitting curve diagram of the BiLSTM-Attention model after FBWO optimization provided by an embodiment of the present invention;
[0075] Figure 5 It is a comparison diagram of the convergence curves of the present invention and other enhanced Beluga algorithms on the CEC2017 test function provided by an embodiment of the present invention;
[0076] Figure 6 is a structural diagram of a landslide displacement prediction system based on fuzzy sets provided by an embodiment of the present invention;
[0077] Figure 7 It is a histogram of landslide displacement prediction errors of BiLSTM-Attention, BiLSTM-Attention optimized by FBWO, and BiLSTM-Attention optimized by other swarm intelligence algorithms provided in the embodiments of the present invention;
[0078] Figure 8 It is a fitting curve diagram of the BiLSTM-Attention provided by the embodiments of the present invention, the BiLSTM-Attention optimized by FBWO, and the BiLSTM-Attention optimized by other swarm intelligence algorithms. DETAILED DESCRIPTION
[0079] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0080] The specific implementation method of the present invention is as follows:
[0081] Step 1: Use the exponential smoothing method to decompose the cumulative displacement of loess landslide into trend term displacement and period term displacement, and divide them into training set and test set.
[0082] Step 2: Establish a BiLSTM-Attention network model, design a fuzzy set-based multi-strategy enhanced Beluga algorithm (FBWO), use FBWO to optimize the hyperparameters in the BiLSTM-Attention model, and realize the adaptive optimization of the four hyperparameters of the number of hidden layer nodes, initial learning rate, maximum number of iterations, and regularization coefficient in the BiLSTM-Attention network. Finally, the FBWO-BiLSTM-Attention landslide displacement prediction model is obtained to predict periodic displacement.
[0083] Step three, input the trend item displacement data into the partial least squares regression model to obtain the predicted trend item displacement; input the period item displacement data into the FBWO-BiLSTM-Attention prediction model to obtain the predicted value of the period item displacement.
[0084] Step 4: Add the predicted trend term displacement and period term displacement to obtain the final prediction result of the cumulative displacement of the loess landslide.
[0085] The entire prediction process is as follows Figure 1 shown.
[0086] The following describes in detail the techniques used in each step:
[0087] In step 1, the exponential smoothing method is used to decompose the accumulated displacement of the landslide into trend term displacement and period term displacement. The calculation formula of the exponential smoothing method is as follows:
[0088]
[0089] Among them, y(i) represents the cumulative displacement data point, y t (i) represents the decomposed trend term displacement data point. The exponential smoothing method is applied three times to obtain the final trend term displacement. The period term displacement is obtained by subtracting the trend term displacement from the cumulative displacement, that is, y p =yy t ,y p represents the periodic displacement, y represents the cumulative displacement, and y t represents the displacement of the trend term. Finally, the displacement of the trend term and the displacement of the period term are obtained as follows: Figure 2 shown.
[0090] In step 2, the fuzzy set-based multi-strategy enhanced Beluga algorithm (FBWO) is used to optimize the four hyperparameters of the BiLSTM-Attention model, namely, the number of hidden layer nodes, the maximum number of training times, the initial learning rate, and the regularization coefficient. The population size and the maximum number of iterations in the FBWO algorithm are set to 10, and the algorithm is set to solve the problem according to the four hyperparameters of BiLSTM-Attention. The upper bound of the solution space is ub = {300, 500, 0.01, 0.01}, and the lower bound is lb = {10, 30, 0.0001, 0.0001}. Each dimension of ub and lb corresponds to the upper and lower bounds of the number of hidden layer nodes, the maximum number of training times, the initial learning rate, and the regularization coefficient of the BiLSTM-Attention model. The position of the individual Beluga whale is interpreted as the hyperparameter of the BiLSTM-Attention model, and the mean square error (MSE) obtained on the periodic item displacement training set is used as the fitness value of the individual Beluga whale. The optimization process of FBWO is as follows:
[0091] (1) Use the Sobol sequence to initialize the beluga population. The calculation formula is as follows:
[0092] X n =L b +S n ×(U b -L b )
[0093] Among them, X n represents a beluga whale individual, L b and U b are the lower and upper bounds of the search space, respectively, Sn It is a random number between [0,1] generated by the Sobol sequence. The initial population generated by the Sobol sequence can be more evenly distributed in the search space, which makes the initial population of the algorithm have higher diversity. After obtaining the initial population, the fitness value of each individual is calculated.
[0094] (2) Constructing a fuzzy population. The beluga population is considered to be a fuzzy set, and the entire population is divided into the optimal set B S and the worst set W S , each beluga individual X i The membership of (i=1,2,…,n) can be obtained from its own fitness. Set B S It is expressed as follows:
[0095]
[0096] in, Represents X i For B S The degree of membership.
[0097] The following will introduce For the maximization problem, if F(X i )=+ve, then If F(X i )=-ve, then For the minimization problem, if F(X i )=+ve, then If F(X i )=-ve, then Among them, F i Represents X i The fitness, F max and F min Respectively represent the maximum fitness and minimum fitness in the entire population, and it can be seen from the above content that
[0098] Likewise, W S The representation is as follows:
[0099]
[0100] in, After obtaining the individual membership, the fuzzy set centroid will be constructed for each fuzzy set. The specific formula is as follows:
[0101]
[0102] X CB and X CW The fuzzy sets B Sand W S The center of mass.
[0103] (3) FBWO algorithm uses the balance factor B f To achieve the transition from the exploration phase to the development phase, the calculation formula is as follows:
[0104] B f =B0(1-T / T max )
[0105] Where T is the current iteration number, T max is the maximum number of iterations, and B0 is a random number that changes between (0,1) in each iteration. f >0.5 is the exploration stage, B f When ≤0.5, it is the development stage. f The fluctuation range of will decrease from (0,1) to (0,0.5) as the number of iterations T increases.
[0106] In the exploration phase, the position update formula of the beluga whale is as follows: but
[0107]
[0108] otherwise,
[0109]
[0110] in represents the position of the i-th individual in the S1 dimension, represents the position of the S1th dimension of the global optimal individual, Represents the centroid X CB At the position of the S1 dimension, S2 is the same.
[0111] During the development phase, if but otherwise C1 is the random jump intensity coefficient that controls the Levy flight intensity, L F is the Levy flight function, which is calculated in the same way as the original White Whale optimization algorithm.
[0112] (4) After the position of the beluga individual is updated during the exploration or development phase, it will perform a Cauchy mutation to explore its own neighborhood, thereby improving its local escape ability. The formula for Cauchy mutation is as follows:
[0113]
[0114] Where acuchy is a random number generated by a Cauchy distribution that follows X~C(0,0.5), X rand1 and Xrand2 are two beluga whales randomly selected from the population. is the position before the Cauchy transformation, is the position after Cauchy transformation.
[0115] (5) After the Cauchy mutation, the beluga whale has a certain probability of updating its position in the whale fall stage, which simulates the process of beluga whales falling due to various reasons in nature. The probability of a whale falling is W f , its calculation formula is the same as the original White Whale algorithm. When W f ≥B f The present invention introduces the golden sine strategy into the whale fall stage. The individual position update formula of the white whale in this stage is as follows:
[0116]
[0117] R1 is a random number in the range [0,2π], R2 is a random number in the range [0,π], is the position of the global optimal individual in the jth dimension in the Tth iteration. x1 = a·τ+b·(1-τ), x2 = b·τ+a·(1-τ), where τ is the golden section number The initial values of a and b are -π and π respectively.
[0118] (6) After the position update is completed, the fitness value of each beluga individual is calculated and the optimal individual is found.
[0119] (7) Determine whether the maximum number of iterations has been reached. If not, return to step (2). If the maximum number of iterations has been reached, output the optimal individual of the current population. The position of the optimal individual in each dimension is the optimal setting value of each hyperparameter of BiLSTM-Attention.
[0120] In step 3, the prediction results of trend term displacement are obtained by using partial least squares regression as follows: Figure 3 As shown in the figure, the BiLSTM-Attention model optimized by FBWO in step 2 is used to predict the periodic item. The final prediction result is as follows Figure 4 shown.
[0121] like Figure 6 As shown, the landslide displacement prediction system based on fuzzy sets provided by the embodiment of the present invention includes:
[0122] The training set and test set division module uses the exponential smoothing method to decompose the accumulated displacement of the landslide into trend term displacement and period term displacement, and divides the training set and the test set;
[0123] The loess landslide displacement prediction model establishment module establishes the BiLSTM-Attention network model, designs the multi-strategy enhanced Beluga algorithm (FBWO) based on fuzzy sets, uses FBWO to optimize the hyperparameters in the BiLSTM-Attention model, and realizes the adaptive optimization of the four hyperparameters of the number of hidden layer nodes, initial learning rate, maximum number of iterations and regularization coefficient in the BiLSTM-Attention network. Finally, the FBWO-BiLSTM-Attention landslide displacement prediction model is obtained to predict the periodic displacement.
[0124] The module for obtaining the predicted value of periodic item displacement inputs the trend item displacement data into the partial least squares regression model to obtain the predicted trend item displacement; the module for obtaining the predicted value of periodic item displacement inputs the periodic item displacement data into the FBWO-BiLSTM-Attention prediction model;
[0125] The module for obtaining the cumulative displacement prediction result of loess landslide adds the predicted trend item displacement and period item displacement to obtain the final cumulative displacement prediction result of the landslide.
[0126] An application embodiment of the present invention provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of a BiLSTM-Attention loess landslide displacement prediction method based on an improved Beluga algorithm.
[0127] The application embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of a BiLSTM-Attention loess landslide displacement prediction method based on an improved Beluga algorithm.
[0128] An application embodiment of the present invention provides an information data processing terminal, which includes a landslide displacement prediction system based on fuzzy sets.
[0129] Application Example 1: Loess Landslide Monitoring and Early Warning System
[0130] In the loess region, landslides occur frequently and pose a serious threat to the lives and property of local residents. In order to improve the accuracy of landslide warning, researchers applied the loess landslide displacement prediction method of the present invention. They installed multiple displacement monitoring sensors in the landslide area to collect the landslide cumulative displacement data in real time.
[0131] 1) Data processing: The collected cumulative displacement data of loess landslides are decomposed using the exponential smoothing method to obtain trend term displacement and period term displacement, and the data are divided into a training set and a test set.
[0132] 2) Model training: A BiLSTM-Attention network model was established, and the fuzzy set-based multi-strategy enhanced Beluga algorithm (FBWO) was applied to optimize the model hyperparameters. After training, a high-precision FBWO-BiLSTM-Attention landslide displacement prediction model was obtained.
[0133] 3) Prediction and early warning: The decomposed trend item displacement data is input into the partial least squares regression model to obtain the trend item displacement prediction result; the period item displacement data is input into the optimized BiLSTM-Attention model to obtain the period item displacement prediction value. Finally, the two parts of the prediction results are added together to obtain the predicted value of the landslide cumulative displacement.
[0134] 4) Warning output: According to the prediction results, when the landslide displacement reaches the warning threshold, the system automatically issues a warning notification to remind relevant departments and residents to take countermeasures. This application effectively improves the accuracy and timeliness of landslide warnings and reduces the losses caused by landslide disasters.
[0135] Application Example 2: Dam Safety Monitoring System
[0136] The landslide displacement method has strong robustness and can be used not only for landslide displacement prediction in loess areas, but also for other types of landslide prediction. In a large reservoir dam, landslide risk has always been the focus of safety management. In order to ensure the stability and safety of the dam, the management introduced the landslide displacement prediction technology of the present invention.
[0137] 1) Monitoring data collection: A series of high-precision displacement sensors are arranged around the dam to monitor the landslide displacement around the dam in real time and transmit the data to the central control system.
[0138] 2) Data decomposition and analysis: The monitored cumulative displacement data is decomposed using the exponential smoothing method to obtain the trend item displacement and period item displacement data, and then divided into training set and test set.
[0139] 3) Model application: Establish and optimize the BiLSTM-Attention network model, optimize the model hyperparameters through the FBWO algorithm, and train an FBWO-BiLSTM-Attention landslide displacement prediction model suitable for the dam environment.
[0140] 4) Prediction and decision-making: The latest monitored trend item displacement data is input into the partial least squares regression model to obtain the trend item displacement prediction result; the period item displacement data is input into the BiLSTM-Attention model optimized by FBWO to obtain the period item displacement prediction value. Finally, the two parts of the prediction results are added together to obtain the predicted value of the landslide cumulative displacement.
[0141] 5) Safety assessment and early warning: Based on the predicted landslide displacement results, dam safety assessment is carried out. When the predicted displacement reaches or approaches the safety threshold, the system automatically generates an early warning report to notify relevant managers to take reinforcement measures in time to ensure the safe operation of the dam. This application improves the accuracy and early warning capabilities of dam landslide monitoring and provides important technical support for dam safety management.
[0142] The fuzzy set-based multi-strategy enhanced Beluga optimization algorithm (FBWO) proposed in this paper was compared with other enhanced Beluga algorithms on the CEC2017 and CEC2022 test sets. Some of the results are shown below:
[0143] Table 1 shows the statistical results of FBWO and other enhanced BWOs executed 30 times on the CEC202-20Dim test set. The table also gives the statistical results of the Wilcoxon rank sum test. The symbols "+", "≈" and "-" indicate that the comparison algorithm has a significant advantage, similar results and worse results than FBWO. It can be seen from the table that on the latest CEC2022 test set, FBWO surpassed IBWO, EBWO, MBWO, mBWO and HBWO in 12, 12, 12, 11 and 9 problems respectively, and achieved the optimal value of 11 functions out of 12 test functions, and achieved the best results in an average of 91.7% of the global optimization functions.
[0144] Figure 5 The following graph shows some convergence curves of FBWO and other enhanced Beluga algorithms on the CEC2017 test set. It can be seen from the figure that compared with other competitors, FBWO converges faster on most functions and obtains the smallest optimal solution value. Specifically, on F1, F3, F9 of CEC2017 and F1, F5, F7 of CEC2022, other algorithms stop earlier, while FBWO still has a trend of finding better solutions. Moreover, on most functions of F8, F15, F16, F23, F24, F25 of CEC2017 and F3, F4, F5, F6, F11 of CEC2022, FBWO only needs 20% of the iterations (100 times) to find a better solution than other algorithms, which reflects its excellent convergence speed.
[0145] Figure 7 and Figure 8The landslide displacement prediction error bar graphs and fitting curves of BiLSTM-Attention, BiLSTM-Attention optimized by FBWO, and BiLSTM-Attention optimized by other swarm intelligence algorithms. It can be seen from the figure that the prediction error of the network after FBWO optimization has dropped significantly, and compared with other swarm intelligence algorithms, FBWO has the best optimization effect, and the final fitting curve is also closer to the true value.
[0146] Table 1. Statistical results of FBWO and five enhanced BWO on the CEC2022 test function of 20Dim
[0147]
[0148]
[0149] In summary, FBWO has the best comprehensive performance in terms of convergence speed, convergence accuracy and stability, and the network optimized by FBWO has a significant improvement in the effect of landslide displacement prediction. The above results show that the landslide displacement prediction method proposed in this invention can achieve high-precision prediction results, which has important reference significance for landslide prediction and early warning.
[0150] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.
[0151] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.
Claims
1. An improved Beluga optimized BiLSTM-Attention loess landslide displacement prediction method, characterized in that: include: S1, using exponential smoothing method to decompose the accumulated displacement of landslide into trend term displacement and period term displacement, and divide it into training set and test set; S2, establish the BiLSTM-Attention network model, design the multi-strategy enhanced Beluga algorithm based on fuzzy sets, use FBWO to optimize the hyperparameters in the BiLSTM-Attention model, realize the adaptive optimization of the four hyperparameters of the number of hidden layer nodes, initial learning rate, maximum number of iterations and regularization coefficient in the BiLSTM-Attention network, and finally obtain the FBWO-BiLSTM-Attention landslide displacement prediction model to predict the periodic displacement; S3, input the trend item displacement data into the partial least squares regression model to obtain the predicted trend item displacement; input the period item displacement data into the FBWO-BiLSTM-Attention prediction model to obtain the predicted value of the period item displacement; S4, adding the predicted trend term displacement and period term displacement to obtain the final landslide cumulative displacement prediction result.
2. The BiLSTM-Attention loess landslide displacement prediction method based on the improved Beluga algorithm according to claim 1 is characterized in that: In S1, the exponential smoothing method is used to decompose the accumulated displacement of the landslide into trend term displacement and period term displacement. The calculation formula of the exponential smoothing method is as follows: Among them, y(i) represents the cumulative displacement data point, y t (i) represents the decomposed trend term displacement data point. The exponential smoothing method is applied three times to obtain the final trend term displacement. The period term displacement is obtained by subtracting the trend term displacement from the cumulative displacement, that is, y p =yy t ,y p represents the periodic displacement, y represents the cumulative displacement, and y t Represents the trend term displacement.
3. The BiLSTM-Attention loess landslide displacement prediction method based on the improved Beluga algorithm according to claim 1 is characterized in that: In S2, the fuzzy set-based multi-strategy enhanced Beluga algorithm (FBWO) is used to optimize the four hyperparameters of the BiLSTM-Attention model, namely, the number of hidden layer nodes, the maximum number of training times, the initial learning rate, and the regularization coefficient. The population size and the maximum number of iterations in the FBWO algorithm are set to 10, and the solution space of the algorithm is set according to the four hyperparameters of BiLSTM-Attention. The upper bound of the solution space is ub = {300, 500, 0.01, 0.01}, and the lower bound is lb = {10, 30, 0.0001, 0.0001}. Each dimension of ub and lb corresponds to the upper and lower bounds of the number of hidden layer nodes, the maximum number of training times, the initial learning rate, and the regularization coefficient of the BiLSTM-Attention model, respectively. The individual position of the Beluga whale is interpreted as the hyperparameter of the BiLSTM-Attention model, and the mean square error obtained on the periodic item displacement training set is used as the fitness value of the individual Beluga whale.
4. The BiLSTM-Attention loess landslide displacement prediction method based on the improved Beluga algorithm according to claim 1 is characterized in that: The optimization process of FBWO is as follows: (1) Use the Sobol sequence to initialize the beluga population. The calculation formula is as follows: X n =L b +S n ×(U b -L b ) Among them, X n represents a beluga whale individual, L b and U b are the lower and upper bounds of the search space, respectively, S n is a random number between [0,1] generated by the Sobol sequence; the initial population generated by the Sobol sequence can be more evenly distributed in the search space, which makes the initial population of the algorithm have higher diversity; after obtaining the initial population, the fitness value of each individual is calculated; (2) Constructing a fuzzy population; the beluga population is considered to be a fuzzy set, and the entire population is divided into the optimal set B S and the worst set W S , each beluga individual X i The membership of (i=1,2,…,n) can be obtained from its own fitness. S It is expressed as follows: in, Represents X i For B S The degree of membership; The following will introduce The calculation process; for the maximization problem, if F(X i )=+ve, then If F(X i )=-ve, then For the minimization problem, if F(X i )=+ve, then If F(X i )=-ve, then Among them, F i Represents X i The fitness, F max and F min Respectively represent the maximum fitness and minimum fitness in the entire population, and it can be seen from the above content that Likewise, W S The representation is as follows: in, After obtaining the individual membership, the fuzzy set centroid will be constructed for each fuzzy set. The specific formula is as follows: X CB and X CW The fuzzy sets B S and W S The centroid of (3) FBWO algorithm uses the balance factor B f To achieve the transition from the exploration phase to the development phase, the calculation formula is as follows: B f =B0(1-T / T max ) Where T is the current iteration number, T max is the maximum number of iterations, B0 is a random number that changes between (0,1) in each iteration; when the balance factor B f >0.5 is the exploration stage, B f ≤0.5 is the development stage; B f The fluctuation range of will decrease from (0,1) to (0,0.5) as the number of iterations T increases; In the exploration phase, the position update formula of the beluga whale is as follows: but otherwise, in represents the position of the i-th individual in the S1 dimension, represents the position of the S1th dimension of the global optimal individual, Represents the centroid X CB In the position of the S1th dimension, S2 is the same; During the development phase, if but otherwise C1 is the random jump intensity coefficient that controls the Levy flight intensity, L F is the Levy flight function, which is calculated in the same way as the original White Whale optimization algorithm; (4) After the position of the beluga individual is updated during the exploration or development phase, it will perform a Cauchy mutation to explore its own neighborhood, thereby improving its local escape ability; the formula for Cauchy mutation is as follows: Where cauchy is a random number generated by a Cauchy distribution that follows X~C(0,0.5), X rand1 and X rand2 are two beluga whales randomly selected from the population. is the position before the Cauchy transformation, is the position after Cauchy mutation; (5) After the Cauchy mutation, the beluga whale has a certain probability of updating its position during the whale fall stage, which simulates the process of beluga whales falling due to various reasons in nature. The probability of a whale falling is W f , its calculation formula is the same as the original White Whale algorithm. When W f ≥B f Then it will enter the whale fall stage; in this invention, the golden sine strategy is introduced into the whale fall stage, and the individual position update formula of the white whale in this stage is as follows: R1 is a random number in the range [0,2π], R2 is a random number in the range [0,π], is the position of the global optimal individual in the jth dimension in the Tth iteration; x1 = a·τ+b·(1-τ), x2 = b·τ+a·(1-τ), where τ is the golden section number The initial values of a and b are -π and π respectively; (6) After the position update is completed, the fitness value of each beluga individual is calculated and the optimal individual is found; (7) Determine whether the maximum number of iterations has been reached. If not, return to step (2). If the maximum number of iterations has been reached, output the optimal individual of the current population. The position of the optimal individual in each dimension is the optimal setting value of each hyperparameter of BiLSTM-Attention.
5. The BiLSTM-Attention loess landslide displacement prediction method based on the improved Beluga algorithm according to claim 1, characterized in that: In S3, partial least squares regression is used to predict the displacement of the trend term to obtain the prediction result, and the BiLSTM-Attention model optimized by FBWO in S2 is used to predict the periodic term, and finally the periodic term displacement prediction result is obtained.
6. A loess landslide displacement prediction system that implements the BiLSTM-Attention loess landslide displacement prediction method based on the improved Beluga optimization algorithm as described in any one of claims 1 to 5, characterized in that: include: The training set and test set division module uses the exponential smoothing method to decompose the accumulated displacement of the landslide into trend term displacement and period term displacement, and divides the training set and the test set; The landslide displacement prediction model establishment module establishes the BiLSTM-Attention network model, designs the multi-strategy enhanced Beluga algorithm based on fuzzy sets, uses FBWO to optimize the hyperparameters in the BiLSTM-Attention model, and realizes the adaptive optimization of the four hyperparameters of the number of hidden layer nodes, initial learning rate, maximum number of iterations and regularization coefficient in the BiLSTM-Attention network. Finally, the FBWO-BiLSTM-Attention landslide displacement prediction model is obtained to predict the periodic displacement. The module for obtaining the predicted value of periodic item displacement inputs the trend item displacement data into the partial least squares regression model to obtain the predicted trend item displacement; the module for obtaining the predicted value of periodic item displacement inputs the periodic item displacement data into the FBWO-BiLSTM-Attention prediction model; The landslide cumulative displacement prediction result acquisition module adds the predicted trend item displacement and period item displacement to obtain the final landslide cumulative displacement prediction result.
7. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the BiLSTM-Attention loess landslide displacement prediction method based on improved Beluga optimization as described in any one of claims 1 to 5.
8. A computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the BiLSTM-Attention loess landslide displacement prediction method based on improved Beluga optimization as described in any one of claims 1 to 5.
9. An information data processing terminal, comprising the landslide displacement prediction system based on fuzzy set and improved Beluga optimization as claimed in claim 6.
Citation Information
Cited By
Cloud computing roadbed construction slope displacement monitoring system
CN120337684A
Multi-branch network and cross attention multi-source data fusion slope displacement prediction method
CN121071345A
