CN-121831724-B - Landslide deformation prediction method and device

CN121831724BCN 121831724 BCN121831724 BCN 121831724BCN-121831724-B

Abstract

The invention discloses a landslide deformation prediction method and device, and relates to the field of landslide deformation detection, wherein the method comprises the steps of acquiring InSAR (interferometric synthetic aperture radar) landslide deformation data of a region to be predicted and at least one adjacent region thereof along a sight line; the method comprises the steps of inputting deformation data of at least one adjacent area into a first layer LSTM of a double-layer LSTM landslide deformation prediction model, extracting feature vectors of the at least one adjacent area, multiplying the deformation data of the area to be predicted with an influence coefficient matrix of the area to be predicted to obtain a first matrix, multiplying the feature vectors of the at least one adjacent area with the influence coefficient matrix of the at least one adjacent area to obtain a second matrix, splicing the first matrix and the second matrix to obtain a spliced matrix, and inputting the spliced matrix into a second layer LSTM of the double-layer LSTM model to obtain a final landslide deformation prediction result. The landslide deformation can be accurately predicted by combining the deformation correlation of the region to be predicted and at least one adjacent region.

Inventors

HUANG SHAOJIN
DUAN PING
LI JIA
ZHENG ZHIPENG
LI PEIHUI
YU JIAWEN
Yang Kebu

Assignees

云南师范大学

Dates

Publication Date: 20260508
Application Date: 20260309

Claims (10)

1. A method for predicting landslide deformation, comprising: Acquiring InSAR deformation data of a region to be predicted along a sight line in a landslide interferometry manner, and InSAR deformation data of at least one adjacent region of the region to be predicted along the sight line in the landslide manner; Inputting InSAR of at least one adjacent area of the area to be predicted into a first layer LSTM of a double-layer long and short memory neural network LSTM landslide deformation prediction model along a sight line, and extracting to obtain a feature vector of the at least one adjacent area; multiplying InSAR deformation data of a region to be predicted along a sight line to a landslide to obtain a first matrix, multiplying a feature vector of at least one adjacent region by an influence coefficient matrix of at least one adjacent region to obtain a second matrix, and splicing the first matrix and the second matrix to obtain a splicing matrix of the region to be predicted and at least one adjacent region; Inputting the splicing matrix into a second layer LSTM of the double-layer LSTM landslide deformation prediction model to obtain a final prediction result of landslide deformation of the area to be predicted, wherein the second layer LSTM is generated according to historical relation sample data prediction training of the splicing matrix of the target area and at least one adjacent area and the final landslide deformation prediction result of the target area.
2. The method of claim 1, further comprising pre-training to generate a two-layer LSTM landslide deformation prediction model according to the following method: Acquiring InSAR deformation data of a target area and at least one adjacent area thereof along a sight line to a landslide; Performing three-dimensional decomposition on InSAR of the target area and at least one adjacent area thereof along the sight line to landslide deformation data to obtain an InSAR along-slope deformation data set of the target area and at least one adjacent area thereof; Constructing a double-layer LSTM landslide deformation prediction network, wherein the double-layer LSTM landslide deformation prediction network comprises a first layer LSTM and a second layer LSTM; Forming InSAR landslide deformation data of at least one adjacent area of the target area along the sight line and historical relation sample data of feature vectors of the at least one adjacent area into an InSAR landslide deformation data set of a first layer LSTM; Processing the InSAR gradient deformation data set of the target area and at least one adjacent area thereof into a history relation sample data of a splicing matrix of the target area and at least one adjacent area thereof and a final landslide deformation prediction result of the target area to form an InSAR gradient deformation data set of a second layer LSTM; Training the first layer LSTM along a slope deformation data set by utilizing the InSAR of the first layer LSTM, and training the second layer LSTM along the slope deformation data set by utilizing the InSAR of the second layer LSTM to obtain the double-layer LSTM landslide deformation prediction model.
3. The method of claim 1, further comprising determining the region-to-be-predicted influence coefficients according to the following method: Creating a tuple (T1, W1) for constraining the region-to-be-predicted influence coefficients and a container: When the corresponding container of the area to be predicted is not full, updating the area influence coefficient W1 to be predicted in the tuple according to monotonically decreasing, inserting a new tuple into the container after each round of training is completed to obtain an updated container, sorting the tuples in the updated container according to the size of a T1 value in the tuple, wherein the T1 value is large at the top of the container, and the T1 value is small at the bottom of the container; When the area to be predicted is full corresponding to the container, eliminating the preset number of tuples meeting the preset eliminating condition at the top of the container, reducing the upper limit of the container by the preset number, calculating the maximum value W1max of W1 and the minimum value W1min of W1 in the remaining tuples in the container, updating W1 in the tuples in a sine oscillation mode in the [ W1max, W1min ] interval to obtain updated tuples; And training and removing the preset number of tuples at the top of the container of the area to be predicted every preset first round number, reducing the upper limit of the size of the container of the area to be predicted by the preset number, and when the upper limit of the container is less than or equal to twice the preset number, not reducing the container of the area to be predicted any more, so as to obtain the optimal influence coefficient of the area to be predicted.
4. The method of claim 3, further comprising processing other constraints on the influence coefficients of the region to be predicted by adding non-negative constraints to the influence coefficients of the region to be predicted and not less than adjacent region influence coefficient constraints.
5. The method of claim 1, further comprising determining an influence coefficient of at least one neighboring region according to the following method: creating a tuple (T2, W2) and a container for constraining the neighboring region influence coefficients; When the corresponding container of the adjacent region is not full, the influence coefficient W2 of the adjacent region in the tuple is updated along with the back propagation in the training process, a new tuple is inserted into the container after each round of training is completed, an updated container is obtained, the tuples in the updated container are ordered according to the size of the T2 value in the tuple, the T2 value is large at the top of the container, and the T2 value is small at the bottom of the container; When the adjacent area is full corresponding to the container, eliminating the preset number of tuples meeting the preset eliminating condition at the top of the container, reducing the upper limit of the container by the preset number, calculating the maximum value W2max of W2 and the minimum value W2min of W2 in the residual tuples in the container, and updating the W2 in the tuples in a sine oscillation mode in the [ W2max, W2min ] interval without strict; And training and eliminating the preset number of tuples at the top of the container corresponding to the adjacent region every preset second round, reducing the upper limit of the size of the container corresponding to the adjacent region by the preset number, and when the upper limit of the container is less than or equal to five times of the preset number, not reducing the container corresponding to the adjacent region, so as to obtain the optimal influence coefficient of the adjacent region.
6. The method of claim 5, wherein the updating of W2 in the tuple will not strictly update in a sinusoidal oscillation over the [ W2max, W2min ] interval, comprising: w2 adds a random offset when carrying out sine oscillation updating; And not training the update interval of each round [ W2max, W2min ], but performing delay update on W2max, W2 min.
7. A landslide deformation prediction apparatus, comprising: The system comprises an acquisition unit, a prediction unit and a prediction unit, wherein the acquisition unit is used for acquiring synthetic aperture radar interferometry InSAR deformation data of a region to be predicted along a sight line and InSAR deformation data of at least one adjacent region of the region to be predicted along the sight line; The device comprises a first layer LSTM processing unit, a second layer LSTM processing unit, a first layer LSTM processing unit and a second layer LSTM processing unit, wherein the first layer LSTM processing unit is used for inputting InSAR of at least one adjacent area of a region to be predicted into landslide deformation data along a sight line into a first layer LSTM of a landslide deformation prediction model of a double-layer long and short memory neural network, and extracting to obtain a characteristic vector of the at least one adjacent area; The system comprises a processing unit, a first matrix, a second matrix, a splicing matrix, a model training process and a control unit, wherein the processing unit is used for multiplying InSAR of a region to be predicted with an influence coefficient matrix of the region to be predicted along a sight line to obtain the first matrix, multiplying a feature vector of at least one adjacent region with the influence coefficient matrix of at least one adjacent region to obtain the second matrix; The second-layer LSTM processing unit is used for inputting the splicing matrix into a second-layer LSTM of the double-layer LSTM landslide deformation prediction model to obtain a final prediction result of landslide deformation of the area to be predicted, and the second-layer LSTM is generated according to historical relation sample data prediction training of the splicing matrix of the target area and at least one adjacent area and the final landslide deformation prediction result of the target area.
8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of any of claims 1 to 6 when executing the computer program.
9. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program which, when executed by a processor, implements the method of any of claims 1 to 6.
10. A computer program product, characterized in that the computer program product comprises a computer program which, when executed by a processor, implements the method of any of claims 1 to 6.

Description

Landslide deformation prediction method and device Technical Field The invention relates to the technical field of landslide deformation detection, in particular to a landslide deformation prediction method and device. Background This section is intended to provide a background or context to the invention that is set forth herein. The description herein is not admitted to be prior art by inclusion in this section. Landslide deformation prediction is a method for continuously analyzing and pre-judging displacement changes, deformation trends and the like of a landslide body by utilizing advanced technical means such as remote sensing monitoring, ground observation equipment, machine learning, deep learning and the like. The landslide deformation is predicted, so that important basis can be provided for disaster early warning, emergency preparation and medium-long term prevention and control planning. The accurate landslide deformation prediction not only can early warn in advance before landslide, strives for precious evacuation and protection time for the area with power, but also can help decision makers to formulate a targeted disaster prevention and reduction strategy, thereby maximally reducing casualties and economic losses, and having profound significance for improving the comprehensive prevention and control capability of regional geological disasters. At present, inSAR data is combined with LSTM (long short memory neural network) to predict landslide deformation, only time series data of a single area is usually focused, namely, only historical data of a target area is used for predicting future landslide deformation, the method ignores information of areas around the target area, in fact, landslide is a geological phenomenon related to a certain space range, deformation of the target node area is often correlated with deformation of surrounding neighborhood node areas, and more accurate landslide deformation prediction research is performed by combining the target area and neighborhood data thereof. Therefore, the existing landslide deformation prediction accuracy is low. Disclosure of Invention The invention provides a landslide deformation prediction method, which is used for accurately predicting landslide deformation through the deformation correlation of a region to be predicted and at least one adjacent region, and comprises the following steps: Acquiring InSAR deformation data of a region to be predicted along a sight line in a landslide interferometry manner, and InSAR deformation data of at least one adjacent region of the region to be predicted along the sight line in the landslide manner; Inputting InSAR of at least one adjacent area of the area to be predicted into a first layer LSTM of a double-layer long and short memory neural network LSTM landslide deformation prediction model along a sight line, and extracting to obtain a feature vector of the at least one adjacent area; multiplying InSAR deformation data of a region to be predicted along a sight line to a landslide to obtain a first matrix, multiplying a feature vector of at least one adjacent region by an influence coefficient matrix of at least one adjacent region to obtain a second matrix, and splicing the first matrix and the second matrix to obtain a splicing matrix of the region to be predicted and at least one adjacent region; Inputting the splicing matrix into a second layer LSTM of the double-layer LSTM landslide deformation prediction model to obtain a final prediction result of landslide deformation of the area to be predicted, wherein the second layer LSTM is generated according to historical relation sample data prediction training of the splicing matrix of the target area and at least one adjacent area and the final landslide deformation prediction result of the target area. The invention also provides a landslide deformation prediction device, which is used for accurately predicting landslide deformation through the deformation correlation of the target node region and the surrounding neighborhood node region, and comprises the following steps: The system comprises an acquisition unit, a prediction unit and a prediction unit, wherein the acquisition unit is used for acquiring synthetic aperture radar interferometry InSAR deformation data of a region to be predicted along a sight line and InSAR deformation data of at least one adjacent region of the region to be predicted along the sight line; The device comprises a first layer LSTM processing unit, a second layer LSTM processing unit, a first layer LSTM processing unit and a second layer LSTM processing unit, wherein the first layer LSTM processing unit is used for inputting InSAR of at least one adjacent area of a region to be predicted into landslide deformation data along a sight line into a first layer LSTM of a landslide deformation prediction model of a double-layer long and short memory neural network, and extracting to obtain a characteristic vector of the