Dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT

Abstract

The invention provides a dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT, and belongs to the field of landslide prediction.The dynamic landslide displacement real-time prediction method comprises the steps that online monitored landslide displacement data are obtained and preprocessed; the method comprises the following steps: optimizing hyper-parameters of a gating cycle unit (GRU) based on a particle fireworks algorithm (PFA), and establishing a PFA-GRU initial sub-model; secondly, each time the latest monitoring data is acquired, based on an online learning (OL) strategy, carrying out real-time training on the input sub-model and predicting the displacement of the next round; and finally, after next monitoring data is obtained, prediction precision evaluation is carried out in combination with an online learning prediction result, whether an online learning sub-model or a PFA-GRU dynamic reconstruction (MRT) sub-model is used as a sub-model of next prediction is judged, and dynamic real-time prediction of landslide displacement is realized. According to the invention, a basis can be provided for accurate dynamic landslide displacement real-time prediction and landslide prevention and control.

Description

Technical Field

[0001] The invention relates to the technical field of landslide prediction, and in particular to a dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT. Background Art

[0002] Landslide geological disasters can cause significant casualties and property losses, and landslide displacement is the most important basis for preventing and controlling landslide geological disasters. Therefore, predicting landslide displacement is of great practical significance for reducing casualties and economic losses.

[0003] The existing landslide displacement prediction mainly focuses on the prediction of long-term landslide displacement trends, while there are few predictions for short-term, especially real-time, landslide displacement trends. Real-time landslide displacement trend prediction has the following difficulties: on the one hand, the real-time dynamic characteristics of landslides determine that the static model is not accurate enough when predicting landslide displacement, especially when the landslide trend changes, it is even more difficult to effectively predict. Therefore, it is necessary to establish a dynamic model to improve the accuracy of real-time prediction; on the other hand, the dynamic model greatly increases the training cost of the model, especially when the optimization algorithm is used to optimize the model hyperparameters, a large number of evaluations are required to find the optimal solution, which further increases the cost of model training and reduces the timeliness of prediction. Therefore, how to establish a dynamic real-time prediction model suitable for landslide trend changes and minimize the computational cost while improving the accuracy is crucial for high-precision and efficient real-time prediction of landslide displacement. Summary of the invention

[0004] The purpose of the present invention is to provide a real-time prediction method for dynamic landslide displacement based on PFA-GRU-OL-MRT, which can solve the problem that traditional static prediction models are difficult to meet the accurate prediction of dynamic landslide trends, and dynamic prediction models often increase the calculation cost and bring difficulties to real-time prediction, and provide reliable technical support for improving the accuracy and efficiency of landslide displacement prediction.

[0005] The present invention is achieved in that: The technical solution to achieve the purpose of the present invention is: a dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT, characterized in that it includes the following steps: Step 1: Obtain historical landslide displacement data obtained through online monitoring and preprocess the data; Step 2: Set the hyperparameters required for the PFA-GRU-OL-MRT model; The PFA-GRU-OL-MRT model includes three sub-models, namely, a PFA-GRU initial sub-model, an online learning sub-model and a PFA-GRU dynamic reconstruction sub-model; Step 3: Based on the particle fireworks algorithm and historical landslide displacement data, the sliding window flow training method is used to optimize the hyperparameters of the gated recurrent unit; Step 4: Based on the optimized hyperparameters and combined with the historical landslide displacement data, the PFA-GRU initial sub-model is established; Step 5: Get the latest monitoring data D every time t , based on the online learning strategy, the input sub-model is trained in real time to obtain the online learning sub-model and predict the landslide displacement DP at time t+1 t+1 , and output the prediction result DP t+1 ; Step 6: Get the next monitoring data D t+1 After that, combined with DP t+1 The prediction results are evaluated for accuracy, and then D t+1 Assign to D t , that is, D t =D t+1 If the prediction deviation is less than the allowed value, the online learning sub-model is used as the input sub-model and returned to step five for the next round of prediction; if the prediction deviation is greater than the allowed value, the sliding window flow training method is adopted, and the PFA-GUR dynamic model is reconstructed, and the PFA-GRU dynamic reconstruction sub-model is used as the input sub-model and returned to step five for the next round of prediction.

[0006] Furthermore, in step 2, the hyperparameters required for the PFA-GRU-OL-MRT model include the PFA-GRU initial sub-model window size W1, the online learning sub-model window size W2, and the PFA-GRU dynamic reconstruction sub-model window reduction size W minus , the minimum window size W of the PFA-GRU dynamic reconstruction sub-model min , PFA-GRU initial sub-model training iterations initial_epochs, online learning sub-model training iterations OL_epochs, PFA-GRU dynamic reconstruction sub-model training iterations MRT_epochs, PFA-GRU initial sub-model sliding window flow training times initial_num, PFA-GRU dynamic reconstruction sub-model sliding window flow training times MRT_num, PFA total number of particles swarmsize, PFA iterations maxiter, prediction deviation allowable value Re, upper limit of hyperparameter optimization range ub and lower limit lb .

[0007] Furthermore, in step 3, the particle fireworks algorithm is an improved algorithm of the particle swarm optimization algorithm, and its steps are as follows: 1. Initialization In the particle fireworks algorithm, the initial particles are generated by random initialization: X i 0 =lb+ ( ub-lb )× rand (0,1) in, i is the particle number, X i Representative i The position of the particle, the superscript represents the number of iterations, where 0 represents the initialization stage, lb is the lower limit of the optimization range, ub is the upper limit of the optimization range, rand (0,1) represents a random number between 0 and 1 M dimensional vector, where M Represents the dimension of the problem to be solved; in the initialization phase, the particle fireworks algorithm only randomly generates 80% of the total number of particles, and the remaining 20% ​​of the particles are generated in the local enhanced search phase; The fitness of all initialized particles is evaluated, and the particle with the best fitness is defined as the fireworks particle; 2. Local enhanced search With the fireworks particle as the center, the remaining 20% ​​of particles are randomly generated within a certain range around it, and these individuals are called spark particles; the total number of spark particles N k and the position of each spark particle G bj It is expressed as: N k =0.2× N G bj = G b + 0.2×( ub - lb )×e -2t / T ×( 2 × rand (0,1)-1) in, N is the total number of particles, j is the number of the spark particle, G b is the position of the fireworks particles, t Represents the current iteration number, T Represents the maximum number of iterations; Then, the fitness values ​​of all spark particles are calculated; (III) Particle Update In the particle update stage, three situations are handled differently: (1) For particles other than fireworks particles and spark particles, the update process is: V i k+1 =( 1-t / T ) V i k / 2 + c 1 r 1 ( P i k -X i k )+ c 2 r 2 ( G b k -X i k ) X i k+1 = X i k + V i k+1 in, V Represents speed, c 1 and c 2 is the learning factor; r 1 and r 2 is a random number between 0 and 1. P i Representative particles i Search for the best position in history.

[0008] (2) For fireworks particles, in each iteration, the position of the particle with the best fitness among all fireworks particles and spark particles is directly updated, and the speed is defined as 0; (3) Delete all spark particles; 4. Elimination Mechanism Recalculate the fitness values ​​of all particles, redefine the particle with the best fitness among all particles as the fireworks particle, and update the particle i The best historical position Pi , while deleting the particle with the worst fitness and randomly generating a new particle within the solution domain; For each iteration, steps (ii) to (iv) are repeated until the convergence condition is met or the maximum number of iterations is reached, and then the position of the optimal solution is output.

[0009] In the above way, not only is the global search ability and the ability to jump out of local optima of the PFA algorithm guaranteed, but also the search attention is more focused on the vicinity of the global optimum with the best current fitness, thereby increasing the local fast convergence ability of the algorithm.

[0010] Furthermore, in step three, the sliding window transfer training method refers to using a sliding window for model training. The window means that the scale of the training data selected from the time series data is fixed, and all the data within the window is used as the input, and the next data outside the window is used as the output for training; and using a sliding window for model training means that after each window is trained, the window slides along the time series towards the direction of the latest monitored data. The sliding will delete the earliest monitored data in the window and add the next monitored data after the window to retrain. Under the sliding window transfer training method, MSE ave is used as the fitness function when the particle firework algorithm optimizes the hyperparameters of the gated recurrent unit. The calculation method of MSE ave is as follows: ① Calculate the mean square error MSE between the predicted value and the corresponding measured value obtained at each sliding window position, ② Obtain the average value of MSE at all sliding window positions, which is MSE ave .

[0011] Furthermore, in step three, the hyperparameters include the number of units, learning rate, Dropout rate, and L2 regularization strength of the gated recurrent unit model.

[0012] Furthermore, in step five, the online learning strategy means that every time a latest monitored data D t is obtained, training data is acquired with the online learning window size W2, that is, using the W2 data before D t as the training input and D t as the training output to further train the PFA-GRU model for parameter fine-tuning, where W2 < the initial sub-model window size W1 of the PFA-GRU.

[0013] Furthermore, in step six, for the dynamic model reconstruction, the model reconstruction window size W3 is dynamic. At the initial stage of model reconstruction, W3 = the online learning window size W2. If the prediction deviation after model reconstruction is still greater than the allowable value, then each time W minusThe earlier monitoring data is used to reduce the window size, and then the model is reconstructed until the minimum window size W is reached. min Stop reconstructing the model and output the sub-model with the highest prediction accuracy.

[0014] The beneficial effects of the present invention are as follows: the present invention provides a dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT. Among them, the particle fireworks algorithm (PFA) can enhance the local search capability and accelerate the convergence speed; the PFA-GRU landslide prediction model established through historical landslide displacement data can fully mine the historical landslide displacement information; through online learning (OL), the PFA-GRU model can be fine-tuned with a small computing cost, and the latest landslide displacement trend can be obtained; through model reconstruction (MRT), the latest displacement trend can be fully mined accurately when the displacement trend changes significantly, so that the model has better prediction accuracy and shorter prediction time, providing reliable technical support for real-time prediction of landslide displacement. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 It is a flow chart of a dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT provided by an embodiment of the present invention; Figure 2 It is a flowchart of optimizing GRU hyperparameters based on PFA provided by an embodiment of the present invention; Figure 3 It is a comparison chart of prediction results of PFA-GRU and GRU models provided by an embodiment of the present invention; Figure 4 It is a comparison chart of prediction results of PFA-GRU-OL and GRU-OL models provided by an embodiment of the present invention; Figure 5 It is a comparison chart of the prediction results of the PFA-GRU-OL-MRT and GRU-OL-MRT models provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in combination with the implementation cases and drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0018] The method of the present invention is described below by taking a specific landslide displacement real-time prediction case as an example.

[0019] like Figure 1 A dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT is characterized by comprising the following steps: Step 1: Obtain historical landslide displacement data obtained through online monitoring and preprocess the data; Step 2: Set the hyperparameters required for the PFA-GRU-OL-MRT model; The PFA-GRU-OL-MRT model includes three sub-models, namely, a PFA-GRU initial sub-model, an online learning sub-model and a PFA-GRU dynamic reconstruction sub-model; Step 3: Based on the particle fireworks algorithm and historical landslide displacement data, the sliding window flow training method is used to optimize the hyperparameters of the gated recurrent unit; Step 4: Based on the optimized hyperparameters and combined with the historical landslide displacement data, the PFA-GRU initial sub-model is established; Step 5: Get the latest monitoring data D every time t , based on the online learning strategy, the input sub-model is trained in real time to obtain the online learning sub-model and predict the landslide displacement DP at time t+1 t+1 , and output the prediction result DP t+1 ; Step 6: Get the next monitoring data D t+1 After that, combined with DP t+1 The prediction results are evaluated for accuracy, and then D t+1 Assign to D t , that is, D t =D t+1If the prediction deviation is less than the allowed value, the online learning sub-model is used as the input sub-model and returned to step five for the next round of prediction; if the prediction deviation is greater than the allowed value, the sliding window flow training method is adopted, and the PFA-GUR dynamic model is reconstructed, and the PFA-GRU dynamic reconstruction sub-model is used as the input sub-model and returned to step five for the next round of prediction.

[0020] Furthermore, in step 2, the hyperparameters required for the PFA-GRU-OL-MRT model include the PFA-GRU initial sub-model window size W1, the online learning sub-model window size W2, and the PFA-GRU dynamic reconstruction sub-model window reduction size W minus , the minimum window size W of the PFA-GRU dynamic reconstruction sub-model min , PFA-GRU initial sub-model training iterations initial_epochs, online learning sub-model training iterations OL_epochs, PFA-GRU dynamic reconstruction sub-model training iterations MRT_epochs, PFA-GRU initial sub-model sliding window flow training times initial_num, PFA-GRU dynamic reconstruction sub-model sliding window flow training times MRT_num, PFA total number of particles swarmsize, PFA iterations maxiter, prediction deviation allowable value Re, upper limit of hyperparameter optimization range ub and lower limit lb .

[0021] Furthermore, in step 3, the particle fireworks algorithm is an improved algorithm of the particle swarm optimization algorithm, and its steps are as follows: 1. Initialization In the particle fireworks algorithm, the initial particles are generated by random initialization: X i 0 =lb+ ( ub-lb )× rand (0,1) in, i is the particle number, X i Representative i The position of the particle, the superscript represents the number of iterations, where 0 represents the initialization stage, lb is the lower limit of the optimization range, ub is the upper limit of the optimization range, rand (0,1) represents a random number between 0 and 1 M dimensional vector, where MRepresents the dimension of the problem to be solved; in the initialization phase, the particle fireworks algorithm only randomly generates 80% of the total number of particles, and the remaining 20% ​​of the particles are generated in the local enhanced search phase; The fitness of all initialized particles is evaluated, and the particle with the best fitness is defined as the fireworks particle; 2. Local enhanced search With the fireworks particle as the center, the remaining 20% ​​of particles are randomly generated within a certain range around it, and these individuals are called spark particles; the total number of spark particles N k and the position of each spark particle G bj It is expressed as: N k =0.2× N G bj = G b + 0.2×( ub - lb )×e -2t / T ×( 2 × rand (0,1)-1) in, N is the total number of particles, j is the number of the spark particle, G b is the position of the fireworks particles, t Represents the current iteration number, T Represents the maximum number of iterations; Then, the fitness values ​​of all spark particles are calculated; (III) Particle Update In the particle update stage, three situations are handled differently: (1) For particles other than fireworks particles and spark particles, the update process is: V i k+1 =( 1-t / T ) V i k / 2 + c 1 r 1 ( P i k -X i k )+c 2 r 2 ( G b k -X i k ) X i k+1 = X i k + V i k+1 in, V Represents speed, c 1 and c 2 is the learning factor; r 1 and r 2 is a random number between 0 and 1. P i Representative particles i The best position in the search history.

[0022] (2) For fireworks particles, in each iteration, the position of the particle with the best fitness among all fireworks particles and spark particles is directly updated, and the speed is defined as 0; (3) Delete all spark particles; 4. Elimination Mechanism Recalculate the fitness values ​​of all particles, redefine the particle with the best fitness among all particles as the fireworks particle, and update the particle i The best historical position P i , while deleting the particle with the worst fitness, and randomly generating a new particle in the solution domain; For each iteration, steps (ii) to (iv) are repeated until the convergence condition is met or the maximum number of iterations is reached, and then the location of the optimal solution is output.

[0023] Through the above method, not only is the PFA algorithm guaranteed to have the ability to search globally and escape from the local optimum, but the search attention is also focused more on the vicinity of the global optimum with the current optimal fitness, thereby increasing the algorithm's local rapid convergence ability.

[0024] Further, in step three, the sliding window transfer training method refers to using a sliding window for model training. The window means that the scale of the training data selected from the time series data is fixed, and all the data within the window is used as the input, and the next data outside the window is used as the output for training. Using a sliding window for model training means that after each window is trained, the window slides along the time series towards the direction of the latest monitored data. The sliding will delete the earliest monitored data in the window and add the next data after the window for retraining. In the sliding window transfer training method, MSE ave is used as the fitness function when the particle fireworks algorithm optimizes the hyperparameters of the gated recurrent unit. ave The calculation method is as follows: ① Calculate the mean square error MSE between the predicted value and the corresponding measured value obtained at each sliding window position. ② Obtain the average value of MSE at all sliding window positions, which is the MSE. ave .

[0025] Through the above method, the historical information of the landslide can be fully mined.

[0026] Further, in step three, the hyperparameters refer to the structural parameters of the GRU model, including the number of units, learning rate, Dropout rate, and L2 regularization strength of the GRU model.

[0027] Further, in step five, the online learning strategy means that every time a latest monitored data D t is obtained, training data is acquired with the online learning window size W2, that is, using the W2 data before D t as the training input and D t as the training output to further train the PFA-GRU model for parameter fine-tuning, where W2 < the initial sub-model window size W1 of the PFA-GRU.

[0028] Through online learning, the network parameters of the PFA-GRU prediction model will be continuously accumulated and updated as the monitored data increases, improving the real-time prediction accuracy. On the other hand, since there is no need to reconstruct the model, the computational cost is minimized to the greatest extent, ensuring the prediction timeliness.

[0029] Further, in step six, for the dynamic model reconstruction, the model reconstruction window size W3 is dynamic. In the initial stage of model reconstruction, W3 = the online learning window size W2. If the prediction deviation after model reconstruction is still greater than the allowable value, then each time W minus earlier monitored data is deleted to reduce the window size, and then the model is reconstructed continuously until the minimum window size W min is reached, at which point the model reconstruction stops and the sub-model with the highest prediction accuracy is output.

[0030] Through model reconstruction, the model can be automatically adjusted and the recent landslide trend can be identified when the landslide trend changes significantly, thereby greatly improving the prediction accuracy when the landslide trend changes.

[0031] The experimental data comes from real-time monitoring data of a landslide in Sichuan Province. The landslide body experienced a significant change in landslide trend between 19:00 on August 26, 2020 and 16:00 on September 11, 2020. A total of 400 sets of measured data were obtained during this period.

[0032] The specific implementation process is as follows: like Figure 1 , a dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT, including the following steps: Step 1: Obtain historical landslide displacement data obtained through online monitoring and preprocess the data; Step 2: Set the hyperparameters required for the PFA-GRU-OL-MRT model; The PFA-GRU-OL-MRT model includes three sub-models, namely, a PFA-GRU initial sub-model, an online learning sub-model and a PFA-GRU dynamic reconstruction sub-model; Among them, the initial model window size W1=250, the online learning window size W2=50, and the model reconstruction window size reduction W minus =5, minimum window size W min =20, initial model training iterations initial_epochs=50, online learning training iterations OL_epochs=5, model reconstruction training iterations MRT_epochs=30, initial model sliding window flow training times initial_num=5, model reconstruction sliding window flow training times MRT_num=5, total number of PFA particles swarmsize=10, PFA iterations maxiter=5, prediction deviation allowable value Re=5, upper limit of the optimization range of each hyperparameter of the gated recurrent unit (GRU) ub =[100,0.1,0.5,0.01] and lower limit lb =[10,0.0001,0,01], representing the number of units, learning rate, Dropout rate, and L2 regularization strength respectively.

[0033] Step 3: Based on the particle fireworks algorithm and historical landslide displacement data, the sliding window flow training method is used to optimize the hyperparameters of the gated recurrent unit, such as Figure 2 As shown; Step 4: Based on the optimized hyperparameters and combined with the historical landslide displacement data, the PFA-GRU initial sub-model is established; Step 5: Get the latest monitoring data D every time t, based on the online learning strategy, the input sub-model is trained in real time to obtain the online learning sub-model and predict the landslide displacement DP at time t+1 t+1 , and output the prediction result DP t+1 ; Step 6: Get the next monitoring data D t+1 After that, combined with DP t+1 The prediction results are evaluated for accuracy, and then D t+1 Assign to D t , that is, D t =D t+1 If the prediction deviation is less than the allowed value, the online learning sub-model is used as the input sub-model and returned to step five for the next round of prediction; if the prediction deviation is greater than the allowed value, the sliding window flow training method is adopted, and the PFA-GUR dynamic model is reconstructed, and the PFA-GRU dynamic reconstruction sub-model is used as the input sub-model and returned to step five for the next round of prediction.

[0034] In order to verify the advantages of the method of the present invention, as well as the effects of PFA, online learning and dynamic reconstruction of the model on the prediction effect, the prediction results of the GRU static model without PFA optimization training, the static + online learning fine-tuning model (GRU-OL), the static + online learning fine-tuning + dynamic reconstruction model (GRU-OL-MRT), and the initial static model of GRU based on PFA optimization (PFA-GRU) and the static + online learning fine-tuning model (PFA-GRU-OL) were compared. Among them, when there is no PFA optimization, the number of units, learning rate, Dropout rate, and L2 regularization strength of the GRU model are 50, 0.01, 0.25, and 0.005, respectively, and the other model parameters involved are consistent with the settings of the model proposed in this paper.

[0035] Figure 3~Figure 5 The prediction results of the three models under PFA optimization and the three models without PFA optimization are compared in 150 prediction cycles. Figure 3 It can be seen that the PFA-GRU model has obvious advantages over the single GRU model, because PFA finds better GRU structure parameters and the trained GRU model is better. However, since the landslide trend has changed significantly in the later period, even the better PFA-GRU model cannot adapt to the rapidly changing landslide trend.

[0036] from Figure 4 and Figure 5It can be seen that after adding online learning and model reconstruction, the accuracy of landslide displacement prediction is significantly improved. Among them, although the predicted trends of the PFA-GRU-OL model and the GRU-OL model are consistent with the measured values, the rapid change of the landslide trend causes a large fluctuation in the predicted values, affecting the accuracy of the prediction. The prediction accuracy of the PFA-GRU-OL-MRT model and the GRU-OL-MRT model has been greatly improved due to the introduction of the reconstruction mechanism. However, when the PFA optimization is not performed, the volatility of the GRO model is significantly stronger, indicating that unreasonable hyperparameters will lead to a decrease in prediction performance.

[0037] In order to further explore the performance of each model, the mean absolute error (MAE), absolute percentage error (MAPE), root mean square error (RMSE) and goodness of fit R were used. 2 There are 4 indicators to quantitatively evaluate each model. Among them, the closer MAE, MAPE and RMSE are to 0, the better the effect is. 2 The closer it is to 1, the more accurate the prediction is. The evaluation results of each model are shown in Table 1: Table 1 Comparison of evaluation indicators of various models

[0038] As shown in Table 1, the PSO-GRU-OL-MRT model achieved 0.672, 3.08, 1.20, and 0.961 in MAE, MAPE, RMSE, and goodness of fit R2, respectively, which are the optimal values ​​in the evaluation of each model, fully reflecting the advantages of the model in predicting landslides with trend changes. The results of the present invention have very important practical significance for improving the accuracy and response speed of landslide geological disaster prediction and increasing the emergency response time after early warning.

[0039] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT, characterized in that: The following steps are involved: Step 1: Obtain historical landslide displacement data obtained through online monitoring and preprocess the data; Step 2: Set the hyperparameters required for the PFA-GRU-OL-MRT model; The PFA-GRU-OL-MRT model includes three sub-models, namely, a PFA-GRU initial sub-model, an online learning sub-model and a PFA-GRU dynamic reconstruction sub-model; Step 3: Based on the particle fireworks algorithm and historical landslide displacement data, the sliding window flow training method is used to optimize the hyperparameters of the gated recurrent unit; Step 4: Based on the optimized hyperparameters and combined with the historical landslide displacement data, the PFA-GRU initial sub-model is established; Step 5: Get the latest monitoring data D every time t , based on the online learning strategy, the input sub-model is trained in real time to obtain the online learning sub-model and predict the landslide displacement DP at time t+1 t+1 , and output the prediction result DP t+1 ; Step 6: Get the next monitoring data D t+1 After that, combined with DP t+1 The prediction results are evaluated for accuracy, and then D t+1 Assign to D t , that is, D t =D t+1 If the prediction deviation is less than the allowed value, the online learning sub-model is used as the input sub-model and returned to step five for the next round of prediction; if the prediction deviation is greater than the allowed value, the sliding window flow training method is used to reconstruct the PFA-GUR dynamic model, and the PFA-GRU dynamic reconstruction sub-model is used as the input sub-model and returned to step five for the next round of prediction.

2. A dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT as claimed in claim 1, characterized in that: In step 2, the hyperparameters required for the PFA-GRU-OL-MRT model include the PFA-GRU initial sub-model window size W1, the online learning sub-model window size W2, and the PFA-GRU dynamic reconstruction sub-model window reduction size W minus , the minimum window size W of the PFA-GRU dynamic reconstruction sub-model min , PFA-GRU initial sub-model training iteration number initial_epochs, online learning sub-model training iteration number OL_epochs, PFA-GRU dynamic reconstruction sub-model training iteration number MRT_epochs, PFA-GRU initial sub-model sliding window flow training number initial_num, PFA-GRU dynamic reconstruction sub-model sliding window flow training number MRT_num, total number of PFA particles swarmsize, PFA iteration number maxiter, prediction deviation allowable value Re, upper limit ub and lower limit lb of hyperparameter optimization range.

3. A dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT as claimed in claim 1, characterized in that: In step 3, the particle fireworks algorithm is an improved algorithm of the particle swarm optimization algorithm, and its steps are as follows:

1. Initialization In the particle fireworks algorithm, the initial particles are generated by random initialization: X i 0 =lb+ ( ub-lb )× rand (0,1) in, i is the particle number, X i Representative i The position of the particle, the superscript represents the number of iterations, where 0 represents the initialization stage, lb is the lower limit of the optimization range, ub is the upper limit of the optimization range, rand (0,1) represents a random number between 0 and 1 M dimensional vector, where M Represents the dimension of the problem to be solved; in the initialization phase, the particle fireworks algorithm only randomly generates 80% of the total number of particles, and the remaining 20% ​​of the particles are generated in the local enhanced search phase; The fitness of all initialized particles is evaluated, and the particle with the best fitness is defined as the fireworks particle; 2. Local enhanced search With the fireworks particle as the center, the remaining 20% ​​of particles are randomly generated within a certain range around it, and these individuals are called spark particles; the total number of spark particles N k and the position of each spark particle G bj It is expressed as: N k =0.2× N G bj = G b + 0.2×( ub - lb )×e -2t / T ×( 2 × rand (0,1)-1) in, N is the total number of particles, j is the number of the spark particle, G b is the position of the fireworks particles, t Represents the current iteration number, T Represents the maximum number of iterations; Then, the fitness values ​​of all spark particles are calculated; (III) Particle Update In the particle update stage, three situations are handled differently: (1) For particles other than fireworks particles and spark particles, the update process is: V i k+1 =( 1-t / T ) V i k / 2 + c 1 r 1 ( P i k -X i k )+ c 2 r 2 ( G b k -X i k ) X i k+1 = X i k + V i k+1 in, V Represents speed, c 1 and c 2 is the learning factor; r 1 and r 2 is a random number between 0 and 1. P i Representative particles i Search for the best position in history; (2) For fireworks particles, in each iteration, the position of the particle with the best fitness among all fireworks particles and spark particles is directly updated, and the speed is defined as 0; (3) Delete all spark particles; 4. Elimination Mechanism Recalculate the fitness values ​​of all particles, redefine the particle with the best fitness among all particles as the fireworks particle, and update the particle i The best historical position P i , while deleting the particle with the worst fitness, and randomly generating a new particle in the solution domain; For each iteration, steps (ii) to (iv) are repeated until the convergence condition is met or the maximum number of iterations is reached, and then the location of the optimal solution is output.

4. A dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT as claimed in claim 1, characterized in that: In step 3, the sliding window flow training method refers to using a sliding window for model training. The window refers to a fixed size of training data selected from the time series data, and all data in the window are used as input, and the next data outside the window is used as output for training. Using a sliding window for model training means that after each pair of windows is trained, the window will slide along the time series toward the latest monitoring data. The sliding will delete the earliest monitoring data in the window, and add the monitoring data after the window to retrain. In the sliding window flow training method, MSE is used ave As the fitness function of the particle fireworks algorithm to optimize the hyperparameters of the gated recurrent unit, MSE ave The calculation method is as follows: ① Calculate the mean square error (MSE) between the predicted value and the corresponding measured value at each sliding window position; ② Calculate the average value of the MSE of all sliding window positions, which is the MSE ave .

5. The method for real-time prediction of dynamic landslide displacement based on PFA-GRU-OL-MRT according to claim 1, characterized in that: In step three, the hyperparameters include the number of units of the gated recurrent unit model, the learning rate, the Dropout rate, and the L2 regularization strength.

6. A dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT as claimed in claim 1, characterized in that: In step five, the online learning strategy means that every time a latest monitoring data D is obtained t , training data is obtained with an online learning window size W2, that is, using the W2 data before D t as the training input and D t as the training output to further train the PFA-GRU model to achieve parameter fine-tuning, where W2 < the initial sub-model window size W1 of the PFA-GRU.

7. A dynamic landslide displacement real-time prediction method based on PFA-GRU-OL-MRT as claimed in claim 1, characterized in that: In step 6, the dynamic model is reconstructed, and the model reconstruction window size W3 is dynamic. In the initial stage of model reconstruction, W3 = online learning window size W2. If the prediction deviation after model reconstruction is still greater than the allowable value, W3 is deleted each time. minus The earlier monitoring data is used to reduce the window size, and then the model is reconstructed until the minimum window size W is reached. min Stop reconstructing the model and output the sub-model with the highest prediction accuracy.

Images