Ground surface settlement prediction method and system based on variational mode snake optimization neural network

Through the variational modal snake optimization neural network combined with SBAS-InSAR and multi-head attention mechanism CNN-LSTM network, the complexity and nonlinear processing in surface settlement prediction in hilly and mountainous areas is solved, and high-precision settlement prediction is achieved. Monitor and forecast hilly arable land.

CN119047630APending Publication Date: 2024-11-29江西省自然资源测绘与监测院 +1

Patent Information

Application Number
CN202411137571.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

Under the complex terrain conditions of hilly and mountainous areas, traditional surface settlement prediction methods such as methods based on convolutional neural networks and long and short-term memory networks are difficult to obtain the expected prediction effect, and lack adaptability and accuracy.

Method used

The surface settlement prediction method based on the variational modal snake optimization neural network is adopted. The surface settlement information is processed by obtaining InSAR images, using SBAS-InSAR method and variational modal decomposition technology, and the CNN-LSTM network and snake are combined with the multi-head attention mechanism. Optimize algorithms to optimize hyperparameters to improve prediction accuracy.

Benefits of technology

It significantly improves the accuracy and reliability of surface settlement prediction under complex terrain conditions, overcomes the limitations of traditional methods in nonlinear time series processing, provides a more accurate prediction solution, and reduces operating costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119047630A_ABST
    Figure CN119047630A_ABST
Patent Text Reader

Abstract

The invention discloses a surface subsidence prediction method based on a variational mode snake optimization neural network, and the method comprises the steps: obtaining an InSAR image of a target surface, and carrying out the preprocessing of the InSAR image; processing the preprocessed target earth surface InSAR image by adopting an SBAS-InSAR method to obtain earth surface settlement information of the target earth surface; processing the ground surface settlement information by adopting a variational mode decomposition method to obtain characteristic data of the ground surface settlement information; dividing the feature data of the ground surface settlement information into a training set, a verification set and a test set, training, verifying and testing a pre-constructed CNN-LSTM network based on a multi-head attention mechanism, and optimizing hyper-parameters of a network model by adopting a snake optimization algorithm; and using the trained CNN-LSTM network based on the multi-head attention mechanism to predict the ground surface settlement information of the target ground surface. According to the method, the accuracy and reliability of ground surface settlement prediction under the complex terrain condition can be remarkably improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of settlement prediction, and more specifically, to a method and system for predicting ground settlement based on a variational modal snake optimization neural network. Background Art

[0002] The complex and changeable terrain in hilly and mountainous areas makes surface change monitoring a challenging task. In particular, in terms of subtle surface subsidence monitoring, traditional ground monitoring methods such as leveling, global navigation satellite system (GNSS) and accelerometers are subject to multiple limitations, including low spatial resolution, difficulty in deploying monitoring points and high operating costs. Although interferometric synthetic aperture radar (InSAR) technology has performed well in surface subsidence monitoring due to its wide spatial coverage and high-resolution imaging capabilities, accuracy and model selection are still challenging when dealing with complex terrain in hilly and mountainous areas. Existing surface subsidence prediction models, such as those based on convolutional neural networks and long short-term memory networks (CNN-LSTM), often have difficulty in obtaining the expected prediction results when dealing with surface subsidence data in hilly and mountainous areas with complex instability and nonlinearity. These models lack the necessary adaptability and accuracy when solving prediction problems under specific geographical and environmental conditions. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for predicting surface subsidence based on a variational modal snake optimization neural network, which can significantly improve the accuracy and reliability of surface subsidence prediction under complex terrain conditions.

[0004] In order to achieve these purposes and other advantages according to the present invention, a surface settlement prediction method based on variational modal snake optimization neural network is provided, comprising:

[0005] Acquire InSAR images of the target surface and perform preprocessing;

[0006] The SBAS-InSAR method is used to process the pre-processed target surface InSAR images to obtain the surface subsidence information of the target surface;

[0007] The surface settlement information of the target surface is processed by using the variational mode decomposition method to obtain the characteristic data of the surface settlement information;

[0008] The characteristic data of surface subsidence information are divided into training set, validation set and test set, and the pre-built CNN-LSTM network based on multi-head attention mechanism is trained, validated and tested. At the same time, the snake optimization algorithm is used to optimize the hyperparameters of the CNN-LSTM network based on multi-head attention mechanism.

[0009] The trained CNN-LSTM network based on the multi-head attention mechanism is used to predict the surface subsidence information of the target surface.

[0010] Preferably, the process of InSAR image preprocessing includes: using precise orbit files and SRTM data to remove orbit errors and terrain phase errors of the InSAR images.

[0011] Preferably, the characteristic data of the surface subsidence information includes: statistical characteristics and frequency domain characteristics of each mode.

[0012] Preferably, the CNN-LSTM network based on the multi-head attention mechanism includes: an input layer, a CNN layer, an LSTM layer, a multi-head attention mechanism layer, a fully connected layer and an output layer.

[0013] Preferably, the hyperparameters that need to be optimized for the CNN-LSTM network based on the multi-head attention mechanism include: learning rate, the number of neurons in the CNN layer, the number of neurons in the LSTM layer, and the number of iterations.

[0014] Preferably, the CNN-LSTM network based on the multi-head attention mechanism uses mean square error, root mean square error, mean absolute error and mean absolute percentage error as evaluation indicators of prediction accuracy.

[0015] The present invention also provides a surface settlement prediction system based on a variational modal snake optimization neural network, comprising:

[0016] A data acquisition module, which is used to acquire InSAR images of the target surface and perform preprocessing;

[0017] An SBAS-InSAR processing module is used to process the pre-processed target surface InSAR image using the SBAS-InSAR method to obtain the surface subsidence information of the target surface;

[0018] A variational mode decomposition processing module, which is used to process the surface settlement information of the target surface using a variational mode decomposition method to obtain characteristic data of the surface settlement information;

[0019] An optimization module is used to divide the characteristic data of surface subsidence information into training sets, validation sets and test sets, train, validate and test the pre-built CNN-LSTM network based on the multi-head attention mechanism, and use the snake optimization algorithm to optimize the hyperparameters of the CNN-LSTM network based on the multi-head attention mechanism;

[0020] The prediction module is used to predict the surface settlement information of the target surface using a trained CNN-LSTM network based on a multi-head attention mechanism.

[0021] The present invention also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the above-mentioned surface subsidence prediction method based on variational modal snake optimization neural network.

[0022] The present invention also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned surface settlement prediction method based on variational modal snake optimization neural network is implemented.

[0023] The present invention has at least the following beneficial effects:

[0024] The LSTM model has long-term memory capability, which enables it to predict seasonal and cyclical data well. Although the LSTM network has excellent training and prediction performance for time series data, its prediction accuracy and generalization ability are still affected by parameters such as learning rate, number of iterations, and number of neurons in the hidden layer. Therefore, the CNN network is introduced in the present invention. The design of its convolutional layer and pooling layer enables it to automatically learn and extract local features in the input data. This model solves the problem that the Convolutional Neural Networks (CNN) is difficult to remember early time series information, and also overcomes the problem that the LSTM model is not accurate enough for predicting data.

[0025] Snake Optimization (SO) is a heuristic search algorithm. Due to its unique search mechanism and high efficiency, it is suitable for solving complex optimization problems. It has the advantages of fast convergence speed, stable performance, good robustness and accurate results.

[0026] As an advanced signal processing method, variational mode decomposition (VMD) can split complex time series data into several single modal components, each of which represents an independent frequency in the data. This technology enables the model to more effectively distinguish valuable signals from noise through decomposition, thereby enhancing the accuracy of prediction. It overcomes the limitations of traditional statistical methods in simulating complex nonlinear time series of surface subsidence and provides a more accurate solution;

[0027] Monitoring and prediction of land subsidence in hilly cultivated areas presents specific challenges, mainly due to the inhomogeneity of the terrain and the diversity of land use. Variational mode decomposition (VMD) has significant advantages in this environment due to its ability to effectively handle nonlinear and non-stationary signals. VMD works by decomposing complex time series data into multiple single-frequency modal components, each of which focuses on capturing a specific fluctuation pattern.

[0028] In the present invention, the proposed surface settlement prediction method integrating variational mode decomposition (VMD) and snake optimization algorithm (SO) is designed for monitoring and prediction of cultivated land settlement in hilly and mountainous areas. The variational mode decomposition technology can effectively decompose complex settlement data into multiple modal components, and then the convolutional neural network and long short-term memory neural network optimized by the snake optimization algorithm and adding multi-head attention mechanism are used to analyze and predict these components with high precision, thereby significantly improving the prediction accuracy and reliability of the neural network under complex terrain conditions. The model of the present invention can also help local governments and agricultural departments to formulate more scientific land use strategies and countermeasures, improve irrigation systems, strengthen land protection, and thus support regional sustainable development. The present invention not only improves the prediction accuracy, but also has a low operating cost (INSAR images are obtained by free download and the software is self-developed), is easy to implement, and has a significant innovative significance in the field of surface settlement monitoring technology.

[0029] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of the surface settlement prediction method based on variational modal snake optimization neural network according to an embodiment of the present invention;

[0031] Figure 2 This is a workflow diagram of the SBAS-InSAR technology for processing InSAR images according to an embodiment of the present invention;

[0032] Figure 3 This is an architecture diagram of a CNN-LSTM network based on a multi-head attention mechanism according to an embodiment of the present invention;

[0033] Figure 4 Flow chart of the optimization process of the snake optimization (SO) algorithm according to an embodiment of the present invention;

[0034] Figure 5a This is the average sedimentation rate map calculated by SBAS-InSAR technology from 2022 to 2024 in Qianshan County, Jiangxi Province;

[0035] Figure 5b This is the overall settlement map of Qianshan County, Jiangxi Province from 2022 to 2024;

[0036] Figure 6 This is the LOS average sedimentation rate map of three cultivated fields A, B, and C in Qianshan County, Jiangxi Province;

[0037] Figure 7aIt is the modal component diagram of the sample point settlement data of cultivated land A after variational modal decomposition;

[0038] Figure 7b The curve diagram of each modal component at different times after the variational modal decomposition of the settlement data of the sample point of cultivated land A;

[0039] Figure 8a This is a comparison chart of the prediction accuracy of VMD-SO-CNN-LSTM-MATT and VMD-CNN-LSTM-MATT models;

[0040] Figure 8b Comparison chart of prediction accuracy between VMD-SO-CNN-LSTM-MATT and LSTM model

[0041] Figure 8c This is a comparison chart of the prediction accuracy of VMD-SO-CNN-LSTM-MATT and VMD-SSA-LSTM models;

[0042] Fig. 9 It is a structural schematic diagram of the surface subsidence prediction system based on the variational modal snake optimization neural network described in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0044] It should be noted that the experimental methods described in the following embodiments are conventional methods unless otherwise specified, and the reagents and materials are commercially available unless otherwise specified; in the description of the present invention, the terms "lateral", "longitudinal", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0045] See also Figure 1 As shown, an embodiment of the present invention provides a surface settlement prediction method based on a variational modal snake optimization neural network, comprising:

[0046] S1. Obtain the InSAR image of the target surface and perform preprocessing;

[0047] Specifically, the InSAR image of the target surface may be derived from an earth observation satellite with interferometric synthetic aperture radar (InSAR), such as the Sentinel-1A satellite;

[0048] The process of preprocessing InSAR images includes: using precise orbit files and SRTM data to remove the orbit error and terrain phase error of InSAR images;

[0049] SRTM data is jointly measured by the National Aeronautics and Space Administration (NASA) and the National Mapping and Surveying Agency (NIMA) of the Department of Defense. It provides a high-precision digital elevation model (DEM) that can show the undulating features of the terrain, such as mountains and valleys.

[0050] The precise orbit file provides the precise position information of the satellite in orbit, which is crucial for correcting the geometric distortion in InSAR images. The precise orbit file can be downloaded online.

[0051] Combining SRTM data with precise orbit files, InSAR images can be geometrically corrected and topographically corrected, thereby removing or reducing systematic errors in the images and improving the quality and accuracy of the images.

[0052] S2, using the SBAS-InSAR method to process the pre-processed target surface InSAR image to obtain the surface subsidence information of the target surface;

[0053] Specifically, SBAS-InSAR (Small Baseline Subset Interferometric Synthetic Aperture Radar) technology is an advanced radar remote sensing technology used to monitor small changes on the surface, such as ground subsidence, landslides and other geological activities. This technology is a specific application of InSAR (Interferometric Synthetic Aperture Radar), and is particularly suitable for processing scenes with good interferometric phase continuity due to small baselines in time and space.

[0054] The solution principle of SBAS-InSAR technology is as follows:

[0055] Assuming that N+1 SAR images of the target surface are obtained based on time series sorting, the images are paired in pairs based on the pre-set time and space thresholds to form interference pairs, and many small sets containing interference pairs will be obtained. The baselines of the interference pairs within the small sets are relatively small, while the baselines of the interference pairs between sets will be relatively long. The two SAR images in the interference pair are subjected to differential interference processing to obtain differential interference graphs. The number of differential interference graphs L finally obtained is in the range of:

[0056]

[0057] use Indicates that any pixel (x, y) in the SAR image is mThe phase of the moment relative to the initial moment t0, then at t i ,t j In the differential interference diagram at two moments, the differential interference phase can be expressed as

[0058]

[0059] Among them, the value range of l is [1, L], d(t i ,x,r) and d(t j ,x,r) represent t i ,t j The accumulated settlement in the radar line of sight at two moments relative to the initial moment t0.

[0060] According to the above formula, if there are L differential interference patterns, we can list L equations and replace them in matrix form:

[0061]

[0062] In the equation, A is an approximate incidence matrix, which can be estimated by the least squares method.

[0063] For the above formula, if L>1, then A T A is a singular matrix with a rank of N-L+1 and a coefficient matrix rank deficiency. At this time, the system of equations will have infinite solutions. To obtain a unique solution, it is necessary to merge many baseline sets together to solve it, introduce the singular value decomposition method, and finally obtain the least squares solution with the minimum norm. The deformation rate in the study area can be calculated, and the deformation rate is integrated over time to obtain the cumulative settlement during the observation period.

[0064] The workflow of SBAS-InSAR technology to process InSAR images is as follows: Figure 2 As shown, including:

[0065] (a) Generate a connection map: determine the spatial and temporal baseline thresholds, divide the input multiple SAR images into different subsets, and generate a connection map based on the SAR images in the subsets;

[0066] (b) Interference workflow: In this step, all connection graphs are interfered. Including coherence generation, flattening, filtering and phase unfolding, all data pairs are matched with the super master image to prepare data for the next step of orbit refinement and re-flattening and SBAS inversion. The Goldstein filter is used to improve the accuracy of phase unfolding and measurement accuracy to maximize the signal-to-noise ratio of the interferogram. External DEM data is used to remove the terrain phase. In the present invention, the consistency threshold is selected as 0.3, and the minimum cost flow (MCF) is used for unfolding.

[0067] (c) Re-leveling and orbit refinement: Ground control points (GCPs) are used to remove residual terrain phase and satellite orbit errors. These ground control points should be located in the best unfolding phase region and away from deformation areas. Ground control points are critical to compensating for errors due to residual terrain effects and inaccuracies in satellite orbit determination. By referencing known stable points on the ground, GCPs can significantly refine the phase unwrapping process and correct systematic biases in satellite data. In the present invention, GCPs are selected on the terrace plains. Flat terraces are likely to exhibit minimal undetected movement, ensuring that the GCPs provide a solid, static reference for the entire study period.

[0068] (d) First inversion: The first inversion step represents the core of the SBAS-InSAR inversion, and the second deconvolution is used to optimize the input interferogram and to give a first estimate of the deformation rate and residual topography before the next processing step.

[0069] (e) Second inversion: The second inversion step involves calculating the displacement of the time series based on the results of the first inversion. At the same time, based on the deformation rate obtained by the first inversion, atmospheric filtering estimation and removal of atmospheric phase are performed to obtain the final settlement displacement result. It is necessary to set the threshold of atmospheric filtering. In the present invention, the value of spatial filtering is selected as 1600, and the value of temporal filtering is 365. Other programs are defaulted to estimate and remove the atmospheric phase. In addition, the atmospheric effect can also be estimated so as to eliminate it.

[0070] (f) Geocoding: The geocoding step mainly involves geocoding the processed results obtained using the SBAS-InSAR technology. The results need to be displayed in color by selecting the rainbow process in the color table to show the differences in surface subsidence in different colors. It should be noted that after this processing, the surface subsidence areas will be visualized, and the darker the color, the greater the subsidence.

[0071] From the above workflow of SBAS-InSAR technology to process InSAR images, it is not difficult to see that SBAS-InSAR technology has the following characteristics:

[0072] (1) Small baseline selection: SBAS technology generates interferograms by selecting SAR (synthetic aperture radar) image pairs with small spatial and temporal baselines. This method is very suitable for monitoring small surface deformations. By carefully selecting image pairs with short spatial and temporal distances, the influence of atmospheric interference and decorative noise can be significantly reduced, thereby improving the accuracy and reliability of interferometric measurement.

[0073] (2) Multi-time series analysis: This technology makes full use of SAR (synthetic aperture radar) image data acquired at multiple time points to monitor and analyze the dynamic changes of the surface by constructing a detailed time series interferogram. This method enables researchers to draw detailed changes in the surface over different time periods. By comparing the surface state at different time points, it can identify and quantify those slow and continuous changes, providing important data for long-term environmental monitoring and geospatial research;

[0074] (3) Phase unwrapping and orbit error correction: SBAS-InSAR processing includes phase unwrapping and orbit error correction to ensure the accuracy of the measurement results. Phase unwrapping is a key process that involves converting the continuous phase difference in the interferogram into an absolute distance change, while orbit error correction is used to eliminate the interferometric phase error caused by small deviations in the satellite orbit. Due to the imperfect precision of the satellite orbit, these errors, if not corrected, may cause systematic deviations in the measurement results, thus affecting the final interpretation of the surface deformation. Orbit error correction ensures the spatial geometric consistency of images at different time points by accurately simulating the actual orbital position of the satellite at different time points, which has a great impact on the subsequent generation of accurate multi-time series interferometric results;

[0075] (4) Correction of terrain and atmospheric errors: Use external elevation data such as SRTM (Shuttle Radar Topography Mission) to remove the phase effect caused by terrain. At the same time, SBAS-InSAR also needs to consider and correct the errors caused by atmospheric delay to improve measurement accuracy;

[0076] (5) Spatial coverage and resolution: SBAS-InSAR can handle large areas while maintaining high spatial resolution, making it an effective tool for monitoring land surface changes over wide areas.

[0077] S3, using a variational mode decomposition method to process the surface settlement information of the target surface to obtain characteristic data of the surface settlement information;

[0078] Specifically, variational mode decomposition is an adaptive decomposition algorithm for non-stationary signals, which can control the bandwidth of each modal component, avoid modal aliasing during the decomposition process, and effectively separate the modal components. Its main principle is to use the central frequency bandwidth of each mode. When the sum of the modal component bandwidths takes the minimum value, the original signal is decomposed into a series of intrinsic mode components with specific bandwidths.

[0079] The solution principle of variational mode decomposition is as follows:

[0080] The original signal f is decomposed into k finite bandwidth modal functions, and each modal function u is transformed using the Hilbert transformk (t) Perform signal analysis to obtain a single-sided spectrum. Construct a variational model with constraints as follows:

[0081]

[0082] In the formula, {u k} and {ω k} are the mode set and the center rate set respectively; is the partial derivative operator, * is the convolution operator, and δ(t) is the unit pulse function. In order to make the IMF component frequency band reach the target frequency range, the following steps can be taken: (1) For each IMF component, the single-sided spectrum is obtained by Hilbert transform; (2) The central frequency band of each IMF component is adjusted to the baseband using an exponential function; (3) The bandwidth of each IMF component is estimated by calculating the gradient norm of the demodulated signal.

[0083] The surface settlement information of the target surface is processed by the variational mode decomposition method, and then the key features are extracted from the decomposed modes, including the statistical characteristics and frequency domain characteristics of each mode, as the input data of the following neural network.

[0084] S4, dividing the characteristic data of surface subsidence information into training set, validation set and test set, training, validation and testing the pre-built CNN-LSTM network based on multi-head attention mechanism, and optimizing the hyperparameters of the CNN-LSTM network based on multi-head attention mechanism by using snake optimization algorithm;

[0085] Specifically, the Snake Optimizer (SO) is a new meta-heuristic optimization algorithm that demonstrates its superiority on multiple unrestricted benchmark functions, especially in terms of exploration and exploitation balance and convergence speed. The core of the Snake Optimizer is to imitate the reproductive behavior of snakes, especially the mating strategy of snakes when there is sufficient food and the ambient temperature is suitable. Each "snake" in the algorithm represents a potential solution, and the behavior of the snake is simulated through a mathematical model, including two main processes: foraging and reproduction.

[0086] The solution principle of the snake optimization algorithm is as follows:

[0087] First, a uniformly distributed random population must be generated to start the optimization algorithm process. The initial population can be obtained by the following formula:

[0088] X i =X min +r×(X max -X min )

[0089] Among them, X iis the position of the ith individual, r is a random number between 0 and 1, X min and X max They are the minimum and maximum values ​​of the living range of the snake group, which can be understood as the lower and upper limits of the solution.

[0090] In the snake optimization algorithm, the population is divided into two groups, males and females. The individuals (solutions) in each group interact according to their fitness (efficiency of problem solving). The mating behavior between males and females is used to generate new solutions, while the competition behavior is used to eliminate inefficient solutions.

[0091] The exploration and exploitation phases of snake optimization are mainly affected by temperature (Temp) and food (Q), and the formula is as follows:

[0092] Temp=exp(-t / T)

[0093] Q = c1 × exp[(tT) / T]

[0094] Where t represents the current number of iterations, T represents the maximum number of iterations, and c1 = 0.5;

[0095] If Q < 0.25, the snakes search for food by choosing any random location and update their location (exploration phase). Male snake location X i,m renew:

[0096] X i,m (t+1)=X rand,m (t)±c2×A m ×((X max -X min )×rand+X min

[0097] Where, X rand,m refers to the random position of male snakes, c2 = 0.05, rand is a random number between 0 and 1, A m The ability to find food for males is given by the following formula:

[0098] A m =exp(-f rand,m / f i,m )

[0099] In the formula, f rand,m Yes X rand,m The fitness, f i,m For X i,m The fitness of .

[0100] Female snake position X i,f renew:

[0101] X i,f=X rand,f (t+1)±c2×A f ×((X max -X min )×rand+X min

[0102] Where, X rand,f refers to the random position of the female snake, c2 = 0.05, rand is a random number between 0 and 1, A f The ability to find food for females is given by the following formula:

[0103] A f =exp(-f rand,f / f i,f )

[0104] In the formula, f rand,f Yes X rand,f The fitness, f i,f For X i,f The fitness of .

[0105] If Q>0.25 and Temp>0.6, the snakes will only move towards the food and update their position X i,j (Development phase):

[0106] X i,j (t+1)=X food ±c3×Temp×rand×(X food -X i,j (t))

[0107] Where, X food is the food position, c3=2.

[0108] Depending on the phase and the conditions, each snake updates its position in the search space. This update may be influenced by interactions with other snakes, finding food, or following a mating ritual. The process is repeated for a set number of iterations or until convergence criteria are met. The optimal solution (the snake with the highest fitness) is considered the best solution to the problem.

[0109] Specifically, the architecture of the CNN-LSTM network based on the multi-head attention mechanism is as follows: Figure 3As shown in the figure, the convolutional neural network is a feedforward neural network with convolution processing and deep structure, while the long short-term memory neural network is good at memorizing long-term dependencies and can capture the long-term trends and periodic laws of sedimentation data. The convolutional neural network reduces the number of parameters and computational complexity through pooling in the feature extraction stage, thereby helping to reduce the risk of overfitting of the model. The gating mechanism of the long short-term memory neural network (such as the forget gate) can regulate the flow of information and further control the complexity and overfitting in the model learning process. The attention mechanism is introduced on the basis of the convolutional neural network and the long short-term memory neural network prediction model. Its network architecture mainly includes: input layer, CNN layer, LSTM layer, multi-head attention mechanism layer, fully connected layer and output layer. The CNN layer is the convolutional neural network, including the convolution layer and the pooling layer, and the LSTM layer is the long short-term memory neural network.

[0110] CNN is responsible for extracting spatiotemporal features, while the LSTM layer processes the long-term dependencies of time series data.

[0111] The characteristic data of surface subsidence information are divided into training set, validation set and test set. The pre-built CNN-LSTM network based on multi-head attention mechanism is trained, validated and tested by using snake optimization algorithm to optimize the hyperparameters of CNN-LSTM network based on multi-head attention mechanism.

[0112] The training set is used for model training, and the validation set is used for model tuning and to prevent overfitting. During the entire training process, the snake optimization (SO) algorithm continuously optimizes the model parameters to find the best network structure and parameter configuration, mainly including: learning rate, number of neurons in the CNN layer, number of neurons in the LSTM layer, number of iterations, and the optimization process of the snake optimization (SO) algorithm. Figure 4 shown.

[0113] The CNN-LSTM network based on the multi-head attention mechanism uses mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) as evaluation indicators of prediction accuracy;

[0114] RMSE is the square root of the average of the sum of squares of the deviations between the observed values ​​and the true values. It is a common method to measure the size of the prediction error. The smaller the RMSE value, the better the performance of the prediction model and the smaller the error. MAE is the average of the absolute errors, which is the absolute value of the difference between the predicted value and the actual value. Compared with RMSE, MAE is less sensitive to large errors because it does not involve squaring. Lower MAE values ​​indicate better model performance. MAPE represents the percentage of error, which is calculated by dividing the absolute value of each error by the absolute value of the actual value, then taking the average and multiplying by 100%. It provides a percentage representation of the error size, which facilitates the comparison of data of different magnitudes. Lower MAPE values ​​indicate higher prediction accuracy. The expressions of the various error functions are as follows:

[0115]

[0116] In the formula, m is the number of samples; y i is the actual value of the settlement deformation of sample i; Predicted value of settlement of sample i. In settlement prediction, if the values ​​of the above four indicators are smaller, it means that the difference between the predicted value and the actual settlement value is smaller, and the prediction accuracy is higher, otherwise, it is lower.

[0117] S5. Use the trained CNN-LSTM network based on the multi-head attention mechanism to predict the surface settlement information of the target surface.

[0118] Qianshan County, Jiangxi Province is selected as the study area to further explain the above-mentioned surface subsidence prediction method based on variational modal snake optimization neural network.

[0119] Qianshan County, Jiangxi Province is located in Shangrao City, Jiangxi Province. It is located in the Wuyi Mountains. There are a few plains along the Xinjiang River in the north, many hills in the middle, and mountains in the south. The basic landform features are high in the south and low in the north.

[0120] The Sentinel-1A satellite was used to obtain high-resolution InSAR data of cultivated land in hilly and mountainous areas. The monitoring period was selected from October 2022 to May 2024, and a total of 20 images were collected. The imaging mode of these images was IW, and the spatial resolution was 5m×20m. VV polarization was used, that is, vertical (V) polarization was used for transmitting and receiving signals.

[0121] Each scene of InSAR data is processed using SRTM data and precise orbit files to eliminate orbit errors and terrain phase effects, ensuring the accuracy and reliability of the data obtained.

[0122] Based on the SBAS-InSAR technology, 20 images were processed, and finally the overall subsidence of the county from 2022 to 2024 and the subsidence rate map calculated by SBAS-InSAR technology were obtained, as shown in the figure below: Figure 5a-5b As shown in the figure, the experimental results show that the average annual deformation rate of Qianshan County along the LOS direction is -402.7-63.8mm / y, and the cumulative deformation is -622.1-88.2mm. The area where the county has subsided accounts for 74% of the total area, and the subsidence rate in most areas is concentrated between -34.9mm / y and -17.9mm / y, accounting for 26% of the county area and 35% of the area of ​​the subsidence area. The maximum subsidence rate can reach 402.7mm / y. As can be seen from the figure, the areas where the subsidence in Qianshan County are mainly concentrated in the valleys, and a small part of the flat area has subsided.

[0123] Based on the SBAS-InSAR technology, 20 images were processed and the average LOS settlement rate of the three cultivated fields A, B and C was finally obtained, as shown in Figure 6 shown.

[0124] from Figure 6 It can be seen that during the monitoring period, the three cultivated lands showed obvious subsidence, and the maximum annual subsidence rates of the three cultivated lands were -52.2 mm / y, -52.5 mm / y, and -30.6 mm / y, respectively.

[0125] The SBAS-InSAR monitoring results of cultivated land A were selected, and 600 high coherence points with large cumulative settlement values ​​were extracted as learning samples for surface settlement prediction. The samples were sorted according to the latitude and longitude coordinates of the high coherence points, and divided into training set (the first 540 sample points) and test set (the last 60 sample points) in a ratio of 9:1.

[0126] The present invention selects representative cultivated land for surface settlement prediction and uses the variational mode decomposition method to decompose the linear sequence of cumulative settlement. Before applying the variational mode decomposition algorithm, the modal number k of the variational mode decomposition must be determined first, because the selection of k will directly affect the decomposition effect of the variational mode decomposition algorithm. The specific value of the modal number k is determined by observing the center frequency, and the k value is gradually increased from k = 2, and the change of the center frequency is analyzed, as shown in Table 1 and Figure 7a-7b As shown:

[0127] Table 1 Center frequency observation method

[0128]

[0129] As can be seen from Table 1, when the number of modes exceeds 4, the center frequency of the last mode component is always about 0.183. If the k value continues to increase, additional noise components may be introduced. Therefore, the optimal mode number k for variational mode decomposition in this study is determined to be 4, which can effectively reduce the non-stationarity of the data. The results of the sample point settlement data of cultivated land A after variational mode decomposition are shown in the figure. Figure 7a-7b shown.

[0130] The training set in the present invention accounts for 90% of the data set. After building the prediction model, the training parameters of the model of the present invention are shown in Table 2:

[0131] Table 2 Parameter Description

[0132]

[0133] In order to further study the accuracy of different deep learning prediction methods in surface subsidence prediction, this application also uses a variational modal neural network without snake optimization (VMD-CNN-LSTM-MATT), a single long short-term memory neural network (LSTM), and a variational modal sparrow search neural network without attention mechanism (VMD-SSA-CNN-LSTM) for comparison. The results are shown in Table 3 and Figures 8a to 8c As shown:

[0134] Table 3 Model error index values

[0135]

[0136] As can be seen from Table 3, the variational modal snake optimization neural network prediction model (VMD-SO-CNN-LSTM-MATT) shows obvious superiority in key performance indicators. Its root mean square error (RMSE) is 4.3813mm, which is 34.82% and 60.14% lower than 6.7246mm of VMD-CNN-LSTM-MATT and 10.9966mm of the single LSTM model, respectively, showing a significant error reduction effect. The mean absolute error (MAE) is 3.7236mm, which is 29.85% and 56.97% lower than 5.3089mm of VMD-CNN-LSTM-MATT and 8.6513mm of the single LSTM, respectively. In addition, its mean absolute percentage error (MAPE) is 10.52%, which is much better than VMD-CNN-LSTM-MATT's 15.85% and LSTM's 34.94%. Compared with these two models, MAPE is reduced by 33.69% and 69.89%, respectively. The mean square error (MSE) is 19.1955, which is also significantly lower than other models. Compared with the VMD-SSA-CNN-LSTM model, the variational modal snake optimization neural network prediction model reduces RMSE and MAE by 8.42% and 17.55%, respectively, and reduces MAPE by 59.76%, reflecting its comprehensive advantages in prediction accuracy and stability.

[0137] Through the surface subsidence prediction method provided by the present invention, large-scale monitoring and prediction analysis of cultivated land in the study area were achieved, and the following conclusions were drawn:

[0138] (1) The subsidence monitoring of cultivated land in the hilly and mountainous areas of Qianshan County using SBAS-InSAR technology revealed the deformation rate and cumulative deformation in the area from 2022 to 2024. In particular, the subsidence phenomenon is more concentrated in valleys and other areas prone to subsidence, which provides key geographic information basis for subsequent preventive measures and governance work.

[0139] (2) The cumulative settlement data were processed by the variational mode decomposition (VMD) method, which effectively reduced the non-stationarity of the data and optimized the input data quality of the prediction model. The optimal number of modes was determined to be 4, which laid the foundation for high-precision surface settlement prediction.

[0140] The variational modal snake optimization neural network prediction model shows obvious superiority in key performance indicators. Compared with VMD-CNN-LSTM-MATT, a single LSTM model, and VMD-SSA-CNN-LSTM model, this model shows lower error rates in key indicators such as root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The RMSE and MAE of this model are compared with

[0141] VMD-SSA-CNN-LSTM decreased by 8.42% and 17.55% respectively, while MAPE decreased by 59.76%. This significant improvement demonstrates the comprehensive advantages of the variational modal snake optimization neural network prediction model in improving prediction accuracy and stability.

[0142] Based on the same inventive concept, the present invention also provides a surface subsidence prediction system based on a variational modal snake optimization neural network. The surface subsidence prediction system based on a variational modal snake optimization neural network can be a personal computer, a server, or other systems that implement the aforementioned surface subsidence prediction method based on a variational modal snake optimization neural network.

[0143] See also Fig. 9 As shown, the surface settlement prediction system based on variational modal snake optimization neural network provided by the embodiment of the present invention includes:

[0144] A data acquisition module, which is used to acquire InSAR images of the target surface and perform preprocessing;

[0145] An SBAS-InSAR processing module is used to process the pre-processed target surface InSAR image using the SBAS-InSAR method to obtain the surface subsidence information of the target surface;

[0146] A variational mode decomposition processing module, which is used to process the surface settlement information of the target surface using a variational mode decomposition method to obtain characteristic data of the surface settlement information;

[0147] An optimization module is used to divide the characteristic data of surface subsidence information into training sets, validation sets and test sets, train, validate and test the pre-built CNN-LSTM network based on the multi-head attention mechanism, and use the snake optimization algorithm to optimize the hyperparameters of the CNN-LSTM network based on the multi-head attention mechanism;

[0148] The prediction module is used to predict the surface settlement information of the target surface using a trained CNN-LSTM network based on a multi-head attention mechanism.

[0149] All relevant contents of each step involved in the aforementioned embodiment of the surface subsidence prediction method based on variational modal snake optimization neural network can be referred to the functional description of the functional modules corresponding to the surface subsidence prediction system based on variational modal snake optimization neural network in the embodiment of the present application, and will not be repeated here.

[0150] The division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present invention may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0151] In the drawings of the system embodiment provided by the present invention, the connection relationship between modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines.

[0152] The present invention also provides an electronic device, comprising: at least one processor, and a memory connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the above-mentioned surface settlement prediction method based on variational modal snake optimization neural network. The electronic device can be any terminal device including a mobile phone, a laptop computer, a desktop computer, a tablet computer, a PDA (Personal Digital Assistant), a POS (Point of Sales), a car computer, etc.

[0153] The present invention also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned surface settlement prediction method based on variational modal snake optimization neural network is implemented.

[0154] Through the description of the above implementation mode, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present invention, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0155] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A surface settlement prediction method based on variational mode snake optimization neural network, characterized in that: include: Acquire InSAR images of the target surface and perform preprocessing; The SBAS-InSAR method is used to process the pre-processed target surface InSAR images to obtain the surface subsidence information of the target surface; The surface settlement information of the target surface is processed by using the variational mode decomposition method to obtain the characteristic data of the surface settlement information; The characteristic data of surface subsidence information are divided into training set, validation set and test set, and the pre-built CNN-LSTM network based on multi-head attention mechanism is trained, validated and tested. At the same time, the snake optimization algorithm is used to optimize the hyperparameters of the CNN-LSTM network based on multi-head attention mechanism. The trained CNN-LSTM network based on the multi-head attention mechanism is used to predict the surface subsidence information of the target surface.

2. The surface settlement prediction method based on variational modal snake optimization neural network according to claim 1 is characterized in that: The process of InSAR image preprocessing includes: using precise orbit files and SRTM data to remove the orbit error and terrain phase error of InSAR images.

3. The surface settlement prediction method based on variational modal snake optimization neural network according to claim 1 is characterized in that: The characteristic data of surface subsidence information include: statistical characteristics and frequency domain characteristics of each mode.

4. The surface settlement prediction method based on variational modal snake optimization neural network according to claim 1 is characterized in that: The CNN-LSTM network based on the multi-head attention mechanism includes: input layer, CNN layer, LSTM layer, multi-head attention mechanism layer, fully connected layer and output layer.

5. The surface settlement prediction method based on variational modal snake optimization neural network according to claim 1 is characterized in that: The hyperparameters that need to be optimized for the CNN-LSTM network based on the multi-head attention mechanism include: learning rate, number of neurons in the CNN layer, number of neurons in the LSTM layer, and number of iterations.

6. The surface settlement prediction method based on variational modal snake optimization neural network according to claim 1 is characterized in that: The CNN-LSTM network based on the multi-head attention mechanism uses mean square error, root mean square error, mean absolute error and mean absolute percentage error as evaluation indicators of prediction accuracy.

7. A surface settlement prediction system based on variational mode snake optimization neural network, characterized in that: include: A data acquisition module, which is used to acquire InSAR images of the target surface and perform preprocessing; An SBAS-InSAR processing module is used to process the pre-processed target surface InSAR image using the SBAS-InSAR method to obtain the surface subsidence information of the target surface; A variational mode decomposition processing module, which is used to process the surface settlement information of the target surface using a variational mode decomposition method to obtain characteristic data of the surface settlement information; An optimization module is used to divide the characteristic data of surface subsidence information into training sets, validation sets and test sets, train, validate and test the pre-built CNN-LSTM network based on the multi-head attention mechanism, and use the snake optimization algorithm to optimize the hyperparameters of the CNN-LSTM network based on the multi-head attention mechanism; The prediction module is used to predict the surface settlement information of the target surface using a trained CNN-LSTM network based on a multi-head attention mechanism.

8. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the surface subsidence prediction method based on the variational modal snake optimization neural network as described in any one of claims 1 to 6.

9. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the surface settlement prediction method based on the variational modal snake optimization neural network described in any one of claims 1 to 6 is implemented.

Citation Information

Patent Citations

  • Subway tunnel benching method construction ground surface settlement prediction method and system

    CN114742325A

  • Ground subsidence trend prediction method and system based on machine learning

    CN115077474A

  • Charging load prediction method

    CN116629442A

  • SBAS-InSAR ground subsidence prediction method considering ground load

    CN117113032A

  • River runoff prediction method and device based on mixed deep learning

    CN117114190A

Cited By

  • Shield tunnel construction ground surface settlement prediction method and system, electronic equipment and storage medium

    CN121117443A

  • Frozen soil highway subgrade settlement trend prediction method based on VMD parameter optimization

    CN121959200A