Multi-slope deformation prediction method based on deformation correlation

Abstract

A multi-slope deformation prediction method based on deformation correlation relates to the technical field of slope deformation detection technology and is used to solve the technical problem of deformation prediction for multiple slopes, multiple regions, and circular slope regions. The multi-slope deformation prediction method based on deformation correlation includes the following steps: constructing a three-dimensional spatial coordinate system based on geographic location; measuring the deformation correlation between monitoring points based on whether they belong to the same slope; constructing an adjacency matrix of landslide correlation relationships between monitoring points; fusing slope deformation data based on time series with the deformation correlation matrix data; and performing deformation fitting training on the data fusion results based on a bidirectional LSTM neural network, and predicting the deformation displacement coordinate deviation of each monitoring point.

Description

Technical Field

[0001] The present invention relates to the technical field of slope deformation detection, and in particular to a multi-slope deformation prediction method based on deformation correlation. Background Art

[0002] Existing methods primarily rely on time series-based single-point and multi-point single-area slope deformation prediction. Existing slope deformation prediction techniques are limited by the specific geographic environment and topography. As the name suggests, single-point deformation prediction can only predict deformation at a specific coordinate point.

[0003] Currently, the main process for single-point slope deformation prediction involves using monitoring equipment to collect displacement changes at monitoring points in a time series, fitting a single monitoring point position change curve using time series data and regression algorithms, and ultimately using the fitted model to predict the deformation of the monitoring point. Multi-point single-slope single-area deformation prediction, as the name suggests, can only predict deformation for a specific area within a slope. Deformation prediction for a single area within a multi-point single slope differs significantly from single-point deformation prediction in terms of data collection methods. Multi-point single-slope single-area deformation prediction combines the displacement data of the monitoring point itself with the distance between each monitoring point, and often uses a neural network algorithm for regression prediction of deformation at each monitoring point. Summary of the Invention

[0004] The purpose of the present invention is to provide a multi-slope deformation prediction method based on deformation correlation, which is used to solve the technical problem of deformation prediction of multiple slopes, multiple regions and circular slope regions. That is, the main purpose of the present invention is to solve the deformation prediction of multi-slope regions and valley-type circular slope regions with multiple orientations of mountains formed by natural environment or human factors.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A multi-slope deformation prediction method based on deformation correlation includes the following steps:

[0007] Construct a three-dimensional spatial coordinate system based on geographic location;

[0008] Based on whether the monitoring points belong to the same slope, the deformation correlation between the monitoring points is measured pairwise;

[0009] Construct the landslide correlation adjacency matrix between monitoring points;

[0010] Fusion of slope deformation data and deformation correlation matrix data based on time series;

[0011] Based on the bidirectional LSTM neural network, deformation fitting training is performed on the data fusion results, and the deformation displacement coordinate deviation of each monitoring point is predicted.

[0012] Among them, constructing a three-dimensional spatial coordinate system based on geographic location specifically includes:

[0013] Taking the monitoring points dispersedly arranged on two slopes A and B as an example, the coordinate values ​​of the three slope center points are calculated respectively;

[0014] In a single slope, the objective function is to minimize the sum of the Euclidean distances between each monitoring point and the center point, and clustering is used to calculate the optimal center point coordinates of the three monitoring points in the slope.

[0015] The coordinates of the center points are OA(xA,yA,zA) and OB(xB,yB,zB);

[0016] The mean horizontal coordinate value of the optimal center point of the slope is used as the horizontal coordinate value of the origin of the three-dimensional space; at the same time, the horizontal coordinate z value of the origin of the three-dimensional space is set to 0;

[0017] The due east direction is used as the positive direction of the x-axis, the due north direction is used as the positive direction of the y-axis, and the vertical upward direction is used as the positive direction of the z-axis to complete the modeling of the geographical location spatial coordinate system of the slope monitoring point.

[0018] Among them, the deformation correlation between monitoring points is measured based on whether the monitoring points belong to the same slope, including:

[0019] The deformation correlation between different monitoring points in the same slope is measured by the nonlinear transformation method of spatial Euclidean distance.

[0020] Different monitoring points located in different slopes will use the nonlinear transformation product of weight and Euclidean space distance to measure the deformation correlation;

[0021] The weight is defined as the sum of the distances from different slope monitoring points to the coordinate origin. The sigmoid function and the linear function are combined into a weight reduction function, and the value threshold is converted into the range of 0 to 1.

[0022] At this time, when the monitoring points located in different slopes are closer to the coordinate origin, the value of the weight function approaches 1. When the monitoring points located in different slopes are far away from the coordinate origin, the value of the weight function approaches 0.

[0023] The landslide correlation relationship between monitoring points located on different slopes is the product of the Euclidean spatial distance between the observation points and the weight function value.

[0024] Among them, the landslide correlation adjacency matrix between monitoring points is constructed, which specifically includes:

[0025] Assuming that there are N observation points in total, according to the deformation relationship measurement rules between monitoring points, the deformation relationship values ​​are calculated pairwise to form an NxN monitoring point deformation association relationship adjacency matrix, in which the main diagonal element values ​​are set to 0.

[0026] The fusion of slope deformation data based on time series and deformation correlation matrix data specifically includes:

[0027] Assuming that time t+1 is the next time after time t, the coordinate value of the monitoring point at time t+1 is calculated by subtracting it from the coordinate value of the monitoring point at time t to the coordinate value matrix of the monitoring point at time t to time t+1. At the same time, the difference calculation is performed on the landslide association relationship adjacency matrix of the monitoring points at time t and time t+1.

[0028] Concatenate the above two difference matrices according to the second dimension to complete the data fusion from time t to time t+1;

[0029] The data fusion method not only represents the changes in the positions of monitoring points over time, but also fully reflects the changes in the deformation correlation between monitoring points over time.

[0030] Among them, based on the bidirectional LSTM neural network, deformation fitting training is performed on the data fusion results, and the deformation displacement coordinate deviation of each monitoring point is predicted, including:

[0031] Based on the bidirectional LSTM neural network, the deformation displacement coordinate deviation of each monitoring point at different times is predicted.

[0032] Compared with the existing technology, the multi-slope deformation prediction method based on deformation correlation described in the present invention has the following advantages:

[0033] The multi-slope deformation prediction method based on deformation correlation provided by the present invention is used to measure the deformation relationship between monitoring points and construct a correlation matrix between monitoring points; the calculation method not only uses the nonlinear transformation of the distance between monitoring points as the main factor for measuring the deformation correlation relationship between monitoring points, but also introduces a correlation weight calculation method for monitoring points in different slopes to measure the deformation correlation of monitoring points located in different slopes; when the monitoring point position data is collected according to time, the monitoring point deformation correlation matrix is ​​used to fuse the deformation data as the input of the neural network fitting deformation feature; in addition, compared with the existing method, the method can simultaneously predict the position deformation of monitoring points in multiple areas of multiple slopes, solves the problem of simultaneous prediction of multi-slope deformation, and uses weighted control to control the deformation correlation degree of monitoring points on different slopes, so as to express the deformation correlation relationship of monitoring points located in different slopes in a more accurate way, quantify and fuse the spatial displacement change of the monitoring point and the change of the deformation correlation relationship between the monitoring points, and effectively increase the amount of data information. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is an auxiliary schematic diagram of the deformation correlation weight calculation process between monitoring points in different slopes in the multi-slope deformation prediction method based on deformation correlation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] For ease of understanding, the following describes in detail the multi-slope deformation prediction method based on deformation correlation provided by an embodiment of the present invention in conjunction with the accompanying drawings.

[0036] An embodiment of the present invention provides a multi-slope deformation prediction method based on deformation correlation, comprising the following steps:

[0037] Construct a three-dimensional spatial coordinate system based on geographic location;

[0038] Based on whether the monitoring points belong to the same slope, the deformation correlation between the monitoring points is measured pairwise;

[0039] Construct the landslide correlation adjacency matrix between monitoring points;

[0040] Fusion of slope deformation data and deformation correlation matrix data based on time series;

[0041] Based on the bidirectional LSTM neural network, deformation fitting training is performed on the data fusion results, and the deformation displacement coordinate deviation of each monitoring point is predicted.

[0042] Compared with the prior art, the multi-slope deformation prediction method based on deformation correlation described in the embodiment of the present invention has the following advantages:

[0043] The multi-slope deformation prediction method based on deformation correlation provided by an embodiment of the present invention is used to measure the deformation relationship between monitoring points and construct a correlation matrix between monitoring points; the calculation method not only uses the nonlinear transformation of the distance between monitoring points as the main factor for measuring the deformation correlation between monitoring points, but also introduces a correlation weight calculation method for monitoring points in different slopes to measure the deformation correlation of monitoring points located in different slopes; when the monitoring point position data is collected by time, the monitoring point deformation correlation matrix is ​​used to fuse the deformation data as the input of the neural network fitting deformation feature; in addition, compared with the existing method, the method can simultaneously predict the position deformation of monitoring points in multiple areas within multiple slopes, solves the problem of simultaneous prediction of multi-slope deformation, and uses weighted control to control the deformation correlation degree of monitoring points on different slopes, so as to express the deformation correlation of monitoring points located in different slopes in a more accurate manner, quantify and fuse the spatial displacement change of the monitoring point and the change of the deformation correlation between the monitoring points, and effectively increase the amount of data information.

[0044] Among them, constructing a three-dimensional spatial coordinate system based on geographic location specifically includes:

[0045] Taking the monitoring points dispersedly arranged on two slopes A and B as an example, the coordinate values ​​of the three slope center points are calculated respectively;

[0046] In a single slope, the objective function is to minimize the sum of the Euclidean distances between each monitoring point and the center point, and clustering is used to calculate the optimal center point coordinates of the three monitoring points in the slope.

[0047] The coordinates of the center points are OA(xA,yA,zA) and OB(xB,yB,zB);

[0048] The mean horizontal coordinate value of the optimal center point of the slope is used as the horizontal coordinate value of the origin of the three-dimensional space; at the same time, the horizontal coordinate z value of the origin of the three-dimensional space is set to 0;

[0049] The due east direction is used as the positive direction of the x-axis, the due north direction is used as the positive direction of the y-axis, and the vertical upward direction is used as the positive direction of the z-axis to complete the modeling of the geographical location spatial coordinate system of the slope monitoring point.

[0050] Among them, the deformation correlation between monitoring points is measured based on whether the monitoring points belong to the same slope, including:

[0051] The deformation correlation between different monitoring points in the same slope is measured by the nonlinear transformation method of spatial Euclidean distance.

[0052] Different monitoring points located in different slopes will use the nonlinear transformation product of weight and Euclidean space distance to measure the deformation correlation;

[0053] The weight is defined as the sum of the distances from different slope monitoring points to the coordinate origin (the value from the monitoring point to the origin is greater than 0). The sigmoid function and the linear function are combined to form a weight reduction function (i.e., weight function), and the value threshold is converted to the range of 0 to 1.

[0054] At this time, when the monitoring points located in different slopes are closer to the coordinate origin, the value of the weight function approaches 1. When the monitoring points located in different slopes are far away from the coordinate origin, the value of the weight function approaches 0.

[0055] The landslide correlation relationship between monitoring points located on different slopes is the product of the Euclidean spatial distance between the observation points and the weight function value.

[0056] Among them, the landslide correlation adjacency matrix between monitoring points is constructed, which specifically includes:

[0057] Assuming that there are N observation points in total, according to the deformation relationship measurement rules between monitoring points, the deformation relationship values ​​are calculated pairwise to form an NxN monitoring point deformation association relationship adjacency matrix, in which the main diagonal element values ​​are set to 0.

[0058] The fusion of slope deformation data based on time series and deformation correlation matrix data specifically includes:

[0059] Assuming that time t+1 is the next time after time t, perform a difference operation on the coordinate value of the monitoring point at time t+1 and time t to obtain the coordinate difference matrix of the monitoring point from time t to time t+1 (dimension is Nx3); at the same time, perform a difference calculation on the landslide association relationship adjacency matrix of the monitoring points at time t and time t+1;

[0060] Concatenate the above two difference matrices according to the second dimension to complete the data fusion from time t to time t+1;

[0061] The data fusion method not only represents the changes in the positions of monitoring points over time, but also fully reflects the changes in the deformation correlation between monitoring points over time.

[0062] Among them, based on the bidirectional LSTM neural network, deformation fitting training is performed on the data fusion results, and the deformation displacement coordinate deviation of each monitoring point is predicted, including:

[0063] Based on the bidirectional LSTM neural network, the deformation displacement coordinate deviation of each monitoring point at different times is predicted. Specific embodiments

[0064] Select multiple slopes to be monitored, ridge slopes or valley slopes; install multiple monitoring point equipment in each slope area that needs to be monitored. After the monitoring equipment is installed on the slope, the coordinate origin is found according to the method of selecting the coordinate origin position in the present invention and an origin monitoring device is installed for subsequent auxiliary calculations; after the monitoring equipment is installed, the geographical location coordinate data of each monitoring point is collected at fixed time intervals and stored in time series and slope label.

[0065] like Figure 1 The figure shows an auxiliary schematic diagram of the deformation correlation weight calculation process between monitoring points on different slopes. Formula (1) is the formula for calculating the Euclidean distance between two points in three-dimensional space. As shown in the figure, there are multiple monitoring points distributed on slopes A and B. The monitoring points on slope A are monitoring points a1 and a2, and the monitoring points on slope B are monitoring points b1 and b2. La1, a2 is the distance between monitoring points a1 and a2, La3 is the distance between monitoring point a3 and the coordinate origin, and Lb1 is the distance between monitoring point b1 and the coordinate origin.

[0066] How to calculate the weight of the deformation correlation between monitoring points on different slopes? Take monitoring points a3 and b1 as an example. As shown in the figure below, formula (2) is the sigmod nonlinear transformation function, which has a domain from negative infinity to positive infinity and a range from negative 0.5 to positive 0.5, and a monotonically increasing function. The sigmod function is transformed to obtain formula (3), which has a domain from negative infinity to positive infinity and a range from 0 to 1, and a monotonically decreasing function. Formula (3) is used to solve the function value of Lb1 and La3, which is used to represent the weight coefficient of the deformation correlation between monitoring points a3 and b1 on different slopes. At the same time, formula (3) is used to solve the function value of La3, b1. The product of the above two function values ​​is used to obtain the quantitative result of the deformation correlation between monitoring points on different slopes. The w coefficient for monitoring points on the same slope and different slopes is shown in formula (4). The quantitative matrix W of the deformation correlation relationship of all monitoring points can be obtained as shown in formula (5) below.

[0067] Formula (6) shows the spatial coordinate deformation difference matrix of N monitoring points from time t to time t+1. Assume that the coordinates of any coordinate point are p(x, y, z). At the same time, the difference matrix of the W matrix from time t to time t+1 is calculated according to time and concatenated with the above D matrix column by column to obtain an Nx(N+3) dimensional matrix as data input to the LSTM algorithm to predict the position coordinate offset of the N monitoring points caused by deformation.

[0068] (1)

[0069] (2)

[0070] (3)

[0071] (4)

[0072] (5)

[0073] (6)

[0074] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A multi-slope deformation prediction method based on deformation correlation, characterized in that: The following steps are involved: Construct a three-dimensional spatial coordinate system based on geographic location; Taking whether the monitoring points belong to the same slope as the premise, different measurement methods are selected according to the premise to measure the deformation correlation relationship between any two monitoring points; Construct deformation correlation matrix between monitoring points; Fusion of slope deformation data based on time series and deformation correlation matrix data based on time series; Based on the bidirectional LSTM neural network, deformation fitting training is performed on the data fusion results, and the deformation displacement coordinate deviation of each monitoring point is predicted; Among them, whether the monitoring points belong to the same slope is the premise, and different measurement methods are selected according to the premise to measure the deformation correlation relationship between the two monitoring points, including: The deformation correlation between different monitoring points in the same slope is measured by the nonlinear transformation method of spatial Euclidean distance. Different monitoring points located in different slopes will use the nonlinear transformation product of weight and Euclidean space distance to measure the deformation correlation; The weight is defined as the sum of the distances from different slope monitoring points to the coordinate origin. The sigmoid function and the linear function are combined into a weight reduction function, and the value threshold is converted into the range of 0 to 1. At this time, when the monitoring points located in different slopes are closer to the coordinate origin, the value of the weight function approaches 1. When the monitoring points located in different slopes are far away from the coordinate origin, the value of the weight function approaches 0. The deformation correlation relationship between monitoring points located on different slopes is the product of the nonlinear transformation of the Euclidean space distance between the observation points and the weight function value.

2. The method for predicting multi-slope deformation based on deformation correlation according to claim 1, characterized in that: Constructing a three-dimensional spatial coordinate system based on geographic location, specifically including: Taking the monitoring points dispersedly arranged on two slopes A and B as an example, the coordinate values ​​of the three slope center points are calculated respectively; In a single slope, the objective function is to minimize the sum of the Euclidean distances between each monitoring point and the center point, and clustering is used to calculate the optimal center point coordinates of the three monitoring points in the slope. The coordinates of the center points are O A (x A ,y A ,z A ), O B (x B ,y B ,z B ); The mean horizontal coordinate value of the optimal center point of the slope is used as the horizontal coordinate value of the origin of the three-dimensional space; at the same time, the horizontal coordinate z value of the origin of the three-dimensional space is set to 0; The due east direction is used as the positive direction of the x-axis, the due north direction is used as the positive direction of the y-axis, and the vertical upward direction is used as the positive direction of the z-axis to complete the modeling of the geographical location spatial coordinate system of the slope monitoring point.

3. The method for predicting multi-slope deformation based on deformation correlation according to claim 2, characterized in that: Construct the deformation correlation matrix between monitoring points, including: Assuming that there are N observation points in total, according to the deformation relationship measurement rules between monitoring points, the deformation relationship values ​​are calculated pairwise to form an NxN monitoring point deformation association relationship adjacency matrix, in which the main diagonal element values ​​are set to 0.

4. The method for predicting multi-slope deformation based on deformation correlation according to claim 3 is characterized in that: The slope deformation data based on time series is fused with the deformation correlation matrix data based on time series, specifically including: Assuming that time t+1 is the next moment after time t, perform a difference operation on the coordinate value of the monitoring point at time t+1 and time t to obtain the coordinate difference matrix of the monitoring point from time t to time t+1; at the same time, perform a difference calculation on the deformation correlation matrix of the monitoring points at time t and time t+1; Concatenate the above two difference matrices according to the second dimension to complete the data fusion from time t to time t+1; The data fusion method not only represents the changes in the positions of monitoring points over time, but also fully reflects the changes in the deformation correlation between monitoring points over time.

5. The method for predicting multi-slope deformation based on deformation correlation according to claim 4, characterized in that: Based on the bidirectional LSTM neural network, deformation fitting training is performed on the data fusion results, and the deformation displacement coordinate deviation of each monitoring point is predicted, including: Based on the bidirectional LSTM neural network, the deformation displacement coordinate deviation of each monitoring point at different times is predicted.

Images