Landslide three-dimensional deformation field monitoring method and system based on physical information neural network model

Abstract

The invention discloses a landslide three-dimensional deformation field monitoring method and system based on a physical information neural network model. Comprising the following steps: acquiring multi-period point cloud data of a landslide mass by utilizing three-dimensional laser scanning, and acquiring interior and exterior time sequence deformation data of the landslide mass in combination with limited inclinometry and GNSS monitoring data; resolving the point cloud data by adopting an improved ICP algorithm to obtain a surface deformation field of the landslide mass; a physical information neural network model is constructed, training is performed according to monitoring data, and displacement in the slope direction and the vertical slope direction is predicted by optimizing model parameters; and deducing a landslide three-dimensional overall time sequence deformation field according to the trained model to realize real-time monitoring. The system has the advantages that high-precision landslide deformation monitoring can be provided, real-time performance and accuracy are achieved, and landslide disasters can be effectively early warned.

Description

Technical Field

[0001] The present invention relates to the technical field of geological engineering, and particularly relates to a landslide three-dimensional deformation field monitoring method and system based on a physics-informed neural network model. Background Art

[0002] In recent years, global landslide disasters have occurred frequently, causing huge economic losses and casualties. The occurrence of landslide disasters not only threatens people's lives and property safety, but also affects the normal operation of infrastructure. Therefore, landslide monitoring and early warning have become a very important part of landslide disaster prevention and mitigation work. However, traditional landslide monitoring methods still face some challenges and limitations, and there is an urgent need for a more efficient and accurate technical system to improve the early warning ability of landslide monitoring.

[0003] Existing landslide monitoring methods are mainly divided into point-type deformation monitoring methods and surface deformation monitoring methods based on remote sensing technology. Point-type deformation monitoring methods usually monitor the local deformation of a landslide body by deploying devices such as GNSS (Global Navigation Satellite System), multi-point displacement gauges, and inclinometers on the landslide body. Although this method can provide certain monitoring data, its limitations are also very obvious: First, point-type monitoring can only provide deformation information of a local area and cannot comprehensively reflect the deformation of the entire landslide body; Second, this method relies on the deployment and maintenance of a large number of devices and is easily affected by factors such as equipment failure, local collapse, and corrosion, resulting in data anomalies and false alarms. In addition, point-type monitoring methods are generally applicable to gravity traction-type landslides and are not applicable to rainfall-type landslides or landslides with stepped deformation characteristics.

[0004] Surface deformation monitoring methods based on remote sensing technology, such as InSAR (Interferometric Synthetic Aperture Radar) and three-dimensional laser scanning technology, can obtain deformation information on the surface of a landslide in a large area. InSAR technology uses the phase difference of electromagnetic wave signals to calculate the deformation amount on the surface of a landslide, and three-dimensional laser scanning obtains deformation data by analyzing multi-period point cloud data. However, these methods mainly provide surface deformation information and cannot penetrate deep into the landslide body, lacking a comprehensive understanding of the internal deformation and instability process of the landslide body. In addition, the InSAR calculation method may lead to large errors when interfered by factors such as surrounding vegetation and terrain. The three-dimensional laser scanning method has problems such as large computational volume and discontinuous data when processing a large amount of point cloud data, making it difficult to achieve real-time monitoring of landslide disasters.

[0005] There are also some landslide monitoring methods that set thresholds based on geotechnical parameters (such as acoustic emission, rainfall, deformation rate, etc.). Although they can determine landslide instability and issue disaster warnings to a certain extent, due to the strong spatial variability of geotechnical parameters, there are significant differences between different landslide cases, and the thresholds based on historical experience and statistical data are often not universal. Therefore, these methods are difficult to be widely applied to all types of landslide disasters.

[0006] In summary, the existing technologies are still unable to provide a comprehensive and accurate landslide monitoring and warning system, especially lacking real-time monitoring and prediction methods for the three-dimensional deformation of landslide bodies. Summary of the Invention

[0007] In view of the deficiencies of the existing technologies, the present invention provides a method and system for monitoring the three-dimensional deformation field of landslides based on a physics-informed neural network model.

[0008] In order to achieve the above invention purposes, the technical solutions adopted by the present invention are as follows:

[0009] A method for monitoring the three-dimensional deformation field of landslides based on a physics-informed neural network model, comprising the following steps:

[0010] Step 1: Use three-dimensional laser scanning to obtain multi-period point cloud data of the landslide body, and combine finite inclinometer and GNSS monitoring to obtain finite internal and external time-series deformation monitoring data of the landslide body;

[0011] Step 2: Use an improved ICP algorithm to solve the multi-period point cloud data to obtain the surface deformation field of the landslide body; the improved ICP algorithm is used to identify the corresponding point pairs between the source point cloud and the target point cloud, and construct rotation and translation matrices based on these point pairs to ensure the minimization of the error function.

[0012] Step 3: Construct a physics-informed neural network model, train the model according to the finite internal and external time-series deformation monitoring data of the landslide to obtain the optimal parameters. The input of the model is (x, y, z, t), where (x, y, z) are the three-dimensional spatial coordinates of the monitoring points, and t is the monitoring time. The output is the displacement u along the slope direction and the displacement v perpendicular to the slope direction; the loss of the physics-informed neural network model includes traditional loss and physical information loss. The traditional loss is the mean square error loss, and the physical information loss is used to express the physical laws of the internal and external mechanical deformations of the landslide body.

[0013] Step 4: According to the optimized network parameters, use the trained physics-informed neural network model to deduce the three-dimensional overall time-series deformation field of the landslide to achieve real-time monitoring of the three-dimensional deformation field of the landslide.

[0014] Furthermore, the improved ICP algorithm described in Step 2 includes the following steps:

[0015] Identify the corresponding point pairs between the source point cloud and the target point cloud, and construct a rotation matrix and a spatial translation vector based on these point pairs. The specific formulas are as follows:

[0016] Q = MP + t

[0017]

[0018] where P is the source point cloud, Q is the target point cloud, M is the spatial rotation matrix, and t is the spatial translation vector; t y represents the translation amount along the Y-axis, t z represents the translation amount along the Z-axis, r ij where i, j = 1, 2, 3 represent a component of the rotation matrix, which is used to rotate the coordinates of the point cloud.

[0019] Use the rotation matrix M and the spatial translation vector t to estimate the error function, and minimize the error function by iterative optimization of the transformation. The expression of the error function J is:

[0020] J = Σ||q i -(M * p i + t)|| 2

[0021] where q i is the i-th point in the target point cloud, and p i is the i-th point in the source point cloud;

[0022] In each iteration, calculate the new rotation matrix M n and the translation vector t n , so that the sum of the error function is minimized. The specific calculation formula is:

[0023]

[0024] where p i n-1 is the point cloud of the previous iteration, q i is the point of the target point cloud, and k is the total number of points in the point cloud;

[0025] When the maximum number of iterations n max or the error threshold δ is satisfied, stop the iteration and complete the calculation.

[0026] Furthermore, for the n-th iteration, the n-th point cloud P n is expressed as:

[0027] P n = M n P n-1 + t n

[0028] Corresponding to the point cloud P nThe average distance between and Q is expressed as follows:

[0029]

[0030] When n > n max or d ≤ δ, the calculation iteration stops, is the position of the i-th point in the source point cloud after the n-th iteration.

[0031] Furthermore, the architecture of the physics-informed neural network model is any one of an artificial neural network, a recurrent neural network, a long short-term memory neural network, or a Transformer attention mechanism.

[0032] Furthermore, the traditional loss is the mean square error loss, and the formula is:

[0033]

[0034] where u i and v i are the measured deformations along the slope direction and perpendicular to the slope direction, and are the deformations along the slope direction and perpendicular to the slope direction predicted by the model;

[0035] Furthermore, the physics-informed neural network model uses the ReLU activation function, and the expression of the ReLU activation function is:

[0036]

[0037] Furthermore, the boundary conditions of the physics-informed neural network model include the landslide deformation boundary, and at the monitoring time t = 0, the landslide body has no deformation; at the monitoring time t = ∞, the landslide deformation converges to the maximum values U max and V max .

[0038] Furthermore, the landslide body is segmented into several strips by the slice method, and the angle between the strip and the horizontal plane is θ i , and the relationships between the deformation u along the slope direction and the deformation v perpendicular to the slope direction and the monitoring time t are as follows:

[0039]

[0040] Furthermore, the total loss L of the physics-informed neural network model is composed of the superposition of the traditional loss L data and the physics-information loss L physics , and the expression is:

[0041] L = L data + L physics .

[0042] The present invention also discloses a landslide three-dimensional deformation field monitoring system, which can be used to implement the above-mentioned landslide three-dimensional deformation field monitoring method. Specifically, it includes:

[0043] A three-dimensional laser scanning device for acquiring multi-period point cloud data of the landslide body;

[0044] A limited inclinometer device for acquiring internal and external time-series deformation monitoring data of the landslide body;

[0045] A GNSS monitoring device for acquiring time-series deformation data of the landslide body;

[0046] A data processing module for solving multi-period point cloud data through an improved ICP algorithm to obtain the surface deformation field of the landslide body. The improved ICP algorithm is used to identify corresponding point pairs between the source point cloud and the target point cloud, and based on these point pairs, a rotation and translation matrix is constructed to ensure the minimization of the error function;

[0047] A physical information neural network model construction module for training the physical information neural network model according to the limited internal and external time-series deformation monitoring data of the landslide to obtain optimal parameters. The input of the physical information neural network model is (x, y, z, t), where (x, y, z) are the three-dimensional spatial coordinates of the monitoring point and t is the monitoring time, and the output is the displacement u along the slope direction and the displacement v perpendicular to the slope direction. The loss of the physical information neural network model includes a traditional loss and a physical information loss. The traditional loss is the mean square error loss, and the physical information loss is used to express the physical laws of the internal and external mechanical deformations of the landslide body;

[0048] A real-time monitoring module for deducing the three-dimensional overall time-series deformation field of the landslide according to the optimized network parameters and using the trained physical information neural network model to realize the real-time monitoring of the landslide three-dimensional deformation field.

[0049] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the above-mentioned landslide three-dimensional deformation field monitoring method is implemented.

[0050] The present invention also discloses a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the above-mentioned landslide three-dimensional deformation field monitoring method is implemented.

[0051] Compared with the prior art, the advantages of the present invention are:

[0052] 1. The present invention adopts a landslide three-dimensional deformation field monitoring technology based on the physics-informed neural network, breaking through the limitations of traditional point-type deformation monitoring methods and upgrading from local deformation monitoring to the overall three-dimensional deformation field monitoring of the landslide. This dimensionality-upgrading method can comprehensively reflect the deformation of the entire landslide body, including surface deformation and deep deformation, providing more accurate and comprehensive landslide deformation information than traditional methods.

[0053] 2. The present invention realizes the real-time monitoring and prediction of landslide deformation by using various monitoring technologies such as three-dimensional laser scanning, finite inclinometry, and GNSS, combined with a physics-informed neural network model. This enables more timely and accurate early warning of landslide disasters and can provide strong support for disaster prevention and control and emergency decision-making.

[0054] 3. Traditional landslide monitoring methods often cannot adapt to different types of landslides due to their reliance on experience and thresholds. In contrast, the present invention can make adaptive adjustments according to the different deformation characteristics of landslides through the training of the physics-informed neural network. This method is not only applicable to gravity-driven landslides but can also be widely used in the monitoring of other different types of landslide disasters such as rainfall-induced landslides and reservoir area landslides.

[0055] 4. The present invention combines the advantages of physical mechanics mechanisms and neural networks. Through the physical information loss in the loss function, it ensures that the displacement prediction output by the model conforms to the physical laws of the landslide body. By continuously optimizing the parameters of the neural network model, the predicted displacement values are made more consistent with the measured data, thereby effectively reducing errors and improving prediction accuracy.

[0056] 5. Adopting a variety of neural network architectures (such as artificial neural networks, long short-term memory neural networks, and Transformer attention mechanisms) and selecting the optimal network structure according to actual needs helps to improve the learning ability and computational efficiency of the model. In addition, through the setting of appropriate activation functions and boundary conditions, the computational process of the model is further optimized, ensuring efficient and accurate results.

[0057] 6. Compared with traditional point-type monitoring methods, the present invention reduces the dependence on a large number of sensors and monitoring devices, reducing the risk of data anomalies and false alarms caused by equipment failures or local collapses. At the same time, by comprehensively using three-dimensional laser scanning and GNSS data, the dependence on a single device can be further reduced, improving the robustness and reliability of the monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a schematic diagram of the joint monitoring of a slope realized by the present invention embodiment based on three-dimensional laser scanning, finite inclinometry, and GNSS;

[0059] Figure 2It is the architecture diagram of the physical information neural network in the embodiment of the present invention; in the figure, (a - b) is the improved ICP algorithm for solving the landslide surface deformation field; (c - e) is the schematic diagram of the physical and mechanical mechanism of landslide instability; (f - g) is the architecture of the physical information neural network; (h) is the network structure diagrams of different networks such as the long - short - term memory neural network, the recurrent neural network, and the Transformer attention mechanism;

[0060] Figure 3 It is the schematic diagram of solving the landslide surface deformation field by three - dimensional laser scanning in the embodiment of the present invention; in the figure, (a) is the deformation u along the slope direction of the landslide in the middle stage of development; (b) is the deformation u along the slope direction of the landslide in the late stage of development; (c) is the deformation v perpendicular to the slope direction of the landslide in the middle stage of development; (d) is the deformation v perpendicular to the slope direction of the landslide in the late stage of development.

[0061] Figure 4 It is the schematic diagram of constructing the landslide three - dimensional deformation field based on the physical information neural network in the embodiment of the present invention. In the figure, (a) is the model training and parameter optimization process; (b) is the construction result of the landslide three - dimensional deformation field. Specific implementation manner

[0062] To make the purpose, technical solution and advantages of the present invention clearer and more understandable, the following further elaborates on the present invention in detail according to the attached drawings and by listing embodiments.

[0063] The present invention provides a method for monitoring the three - dimensional deformation field of a landslide based on a physical information neural network model, including the following steps:

[0064] Step 1: Use three - dimensional laser scanning to obtain multi - period point cloud data of the landslide body, and use finite inclinometer and GNSS monitoring to obtain limited internal and external appearance time - series deformation monitoring data of the landslide body. The monitoring schematic diagram is as Figure 1 shown;

[0065] Step 2: Use the improved ICP algorithm to solve the multi - period point cloud data, so as to obtain the surface deformation field of the landslide body. This algorithm identifies the corresponding point pairs between the source point cloud and the target point cloud, constructs the rotation and translation matrices based on these point pairs, and uses this matrix to estimate the error function. This algorithm iteratively optimizes the transformation to ensure that the error function meets the predefined criteria, as described below:

[0066] Q = MP + t (1)

[0067]

[0068] where P is the source point cloud, Q is the target point cloud, M is the spatial rotation matrix, and t is the spatial translation vector. The representation of the total error J is as follows:

[0069] J = Σ||q i -(M * p i+t)|| 2 (3)

[0070] Among them, the minimum value of J should be calculated, and M0 and t0 are defined as the initial rotation matrix and the initial translation vector. The maximum number of iterations is defined as n max . The sum threshold of the squared Euclidean distance is set to δ. The nth iteration is calculated as follows:

[0071]

[0072] For the nth iteration, the nth point cloud P n can be expressed as:

[0073] P n = M n P n-1 + t n (5)

[0074] The average distance between the corresponding point clouds P n and Q can be expressed as follows:

[0075]

[0076] When n > n max or d ≤ δ, the calculation iteration stops. At this time, the ICP calculation process is completed. Examples of the landslide surface deformation field solution process and results are shown as Figure 2 (a - b). To ensure the calculation accuracy, the target point cloud is divided into blocks for separate calculations.

[0077] Step 3: Build a physics-informed neural network model. According to the limited internal and external time-series deformation monitoring data of the landslide, train the model to obtain the optimal parameters. The physical and mechanical mechanisms of landslide instability are shown as Figure 2 (c - e). The input of the neural network model is (x, y, z, t), where (x, y, z) are the three-dimensional spatial coordinates of the monitoring points and t is the monitoring time. The output of the neural network model is the displacement u along the slope direction and the displacement v perpendicular to the slope direction. The architecture of the physics-informed neural network model is shown as Figure 2 (f - g). According to different situations, artificial neural networks, recurrent neural networks, long short-term memory neural networks or Transformer attention mechanism architectures can be selected, as shown in Figure 2 (h). The loss of the physics-informed neural network model includes traditional loss (root mean square error loss) and physical information loss. The traditional loss is mean square error loss or other forms of traditional loss. The mean square error loss is shown in Equation (7):

[0078]

[0079] In Equation (7), ui and v i are the measured deformations along the slope direction and perpendicular to the slope direction, and are the deformations along the slope direction and perpendicular to the slope direction predicted by the model. The ReLU activation function or other activation function types are adopted, and the expression of the ReLU activation function is shown in Equation (8):

[0080]

[0081]

[0082] The boundary conditions of the model are shown in Equations (9), (10) and (11). At the monitoring time t = 0, it is the initial state and the landslide body does not deform. At the monitoring time t = ∞, as the time scale extends infinitely, the landslide deformation will not increase infinitely and converges to the maximum values U max and V max In addition, when the position (x, y, z) is outside the deformation boundary shown in Figure 2 d, the deformation value is also 0. In Figure 2 e, the slope body is divided into several strips by the slice method, and the total number of strips is i. The angle between strip i and the horizontal plane is represented by θ i , so the relationships among the deformation u along the slope direction, the deformation v perpendicular to the slope direction and the monitoring time t on the landslide sliding surface are as follows:

[0083]

[0084] The total loss L of the physics-informed neural network model is the superposition of the traditional loss and the physics-informed loss of the model, as shown in Equation (13):

[0085] L = L data + L physics (13)

[0086] Step 4: According to the optimized network parameters, use the trained physics-informed neural network model to deduce the three-dimensional overall time-series deformation field of the landslide, realize the real-time monitoring of the three-dimensional deformation field of the landslide, and upgrade from local deformation monitoring to three-dimensional overall deformation monitoring.

[0087] This embodiment is first applied to the three-dimensional overall deformation field monitoring of a small landslide model.

[0088] (1) According to the three-dimensional laser scanning, multi-period point cloud data of the slope surface are obtained, and according to the limited inclinometer, the time-series deformation monitoring data inside and outside the slope are obtained.

[0089] (2) Three-dimensional laser scanning is adopted to obtain multi-period point cloud data on the surface of the slope body. The average domain vector algorithm is used to solve the point cloud data to obtain the surface deformation field of the slope, as Figure 3 shown.

[0090] (3) According to the surface deformation field and the deformation time series data of the slope finite inclinometer monitoring, a physics-informed neural network model is trained to obtain the optimal parameters. Using the physics-informed neural network, the three-dimensional overall deformation field of the landslide is deduced, as Figure 4 shown.

[0091] In another embodiment of the present invention, a three-dimensional landslide deformation field monitoring system is provided. This system can be used to implement the above-mentioned three-dimensional landslide deformation field monitoring method. Specifically, it includes:

[0092] A three-dimensional laser scanning device for obtaining multi-period point cloud data of the landslide body;

[0093] A finite inclinometer device for obtaining the internal and external time series deformation monitoring data of the landslide body;

[0094] A GNSS monitoring device for obtaining the time series deformation data of the landslide body;

[0095] A data processing module for solving multi-period point cloud data through an improved ICP algorithm to obtain the surface deformation field of the landslide body. The improved ICP algorithm is used to identify the corresponding point pairs between the source point cloud and the target point cloud, and based on these point pairs, a rotation and translation matrix is constructed to ensure the minimization of the error function;

[0096] A physics-informed neural network model construction module for training the physics-informed neural network model according to the limited internal and external time series deformation monitoring data of the landslide to obtain the optimal parameters. The input of the physics-informed neural network model is (x, y, z, t), where (x, y, z) are the three-dimensional spatial coordinates of the monitoring point and t is the monitoring time, and the output is the displacement u along the slope direction and the displacement v perpendicular to the slope direction. The loss of the physics-informed neural network model includes traditional loss and physical information loss. The traditional loss is the mean square error loss, and the physical information loss is used to express the physical laws of the internal and external mechanical deformations of the landslide body;

[0097] A real-time monitoring module for deducing the three-dimensional overall time series deformation field of the landslide using the trained physics-informed neural network model according to the optimized network parameters to realize the real-time monitoring of the three-dimensional landslide deformation field.

[0098] In another embodiment of the present invention, a terminal device is provided. The terminal device includes a processor and a memory. The memory is used to store a computer program, and the computer program includes program instructions. The processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function. The processor described in the embodiment of the present invention can be used for the operation of the landslide three-dimensional deformation field monitoring method.

[0099] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is the memory device in the terminal device and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and, of course, the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, and the operating system of the terminal is stored in this storage space. And, one or more instructions suitable for being loaded and executed by the processor are also stored in this storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory.

[0100] One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the landslide three-dimensional deformation field monitoring method in the above embodiment; one or more instructions in the computer-readable storage medium are loaded and executed by the processor.

[0101] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0102] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices produce means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0103] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0105] Those of ordinary skill in the art will realize that the embodiments described herein are for helping readers understand the implementation methods of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations that do not depart from the essence of the present invention based on the technical revelations disclosed in the present invention, and these deformations and combinations are still within the protection scope of the present invention.

Claims

1. A method for monitoring the three-dimensional deformation field of landslides based on a physical information neural network model, characterized in that: The following steps are involved: Step 1: Use 3D laser scanning to obtain multi-period point cloud data of the landslide body, and combine limited inclinometer and GNSS monitoring to obtain limited internal and external appearance time-series deformation monitoring data of the landslide body; Step 2: Using an improved ICP algorithm to solve the multi-period point cloud data to obtain the surface deformation field of the landslide body; the improved ICP algorithm is used to identify the corresponding point pairs between the source point cloud and the target point cloud, and to construct rotation and translation matrices based on these point pairs to ensure that the error function is minimized; Step 3: construct a physical information neural network model, train the model according to the limited internal and external appearance time series deformation monitoring data of the landslide, and obtain the optimal parameters. The input of the model is (x, y, z, t), where (x, y, z) is the three-dimensional spatial coordinates of the monitoring point, t is the monitoring time, and the output is the displacement u along the slope and the displacement v perpendicular to the slope. The loss of the physical information neural network model includes traditional loss and physical information loss. The traditional loss is the mean square error loss, and the physical information loss is used to express the physical law of the internal and external mechanical deformation of the landslide body. Step 4: Based on the optimized network parameters, the trained physical information neural network model is used to deduce the three-dimensional overall time-series deformation field of the landslide to achieve real-time monitoring of the three-dimensional deformation field of the landslide.

2. The method for monitoring the three-dimensional deformation field of a landslide according to claim 1, characterized in that: The improved ICP algorithm described in step 2 includes the following steps: Identify the corresponding point pairs between the source point cloud and the target point cloud, and construct the rotation matrix and spatial translation vector based on these point pairs. The specific formula is as follows: Q=MP+t Among them, P is the source point cloud, Q is the target point cloud, M is the spatial rotation matrix, and t is the spatial translation vector; t y represents the translation along the Y axis, t z represents the translation along the Z axis, r ij Where i, j = 1, 2, 3 represents a component of the rotation matrix, which is used to rotate the coordinates of the point cloud; The error function is estimated using the rotation matrix M and the spatial translation vector t, and the error function is minimized by iteratively optimizing the transformation. The expression of the error function J is: J=Σ||q i -(M*p i +t)|| 2 Among them, q i is the i-th point in the target point cloud, p i is the i-th point in the source point cloud; In each iteration, a new rotation matrix M is calculated n and the translation vector t n , so that the sum of the error functions is minimized. The specific calculation formula is: Among them, p i n-1 is the point cloud of the previous iteration, q i is the point of the target point cloud, k is the total number of points in the point cloud; When the maximum number of iterations n is met max Or when the error threshold δ is reached, the iteration is stopped and the calculation is completed.

3. The method for monitoring the three-dimensional deformation field of a landslide according to claim 2, characterized in that: For the nth iteration, the nth point cloud P n It is expressed as: P n =M n P n-1 +t n Corresponding point cloud P n The average distance between and Q is expressed as follows: When n>n max or d≤δ, the iteration of calculation stops, is the position of the i-th point in the source point cloud after the n-th iteration.

4. The method for monitoring the three-dimensional deformation field of a landslide according to claim 1, characterized in that: The architecture of the physical information neural network model is any one of artificial neural networks, recurrent neural networks, long short-term memory neural networks or Transformer attention mechanisms.

5. The method for monitoring the three-dimensional deformation field of a landslide according to claim 1, characterized in that: The traditional loss is the mean square error loss, and the formula is: Among them, u i and v i is the measured deformation along and perpendicular to the slope, and are the deformations predicted by the model along and perpendicular to the aspect.

6. The method for monitoring the three-dimensional deformation field of a landslide according to claim 1, characterized in that: The physical information neural network model uses the ReLU activation function, and the expression of the ReLU activation function is:

7. The method for monitoring the three-dimensional deformation field of a landslide according to claim 1, characterized in that: The boundary conditions of the physical information neural network model include the landslide deformation boundary, and when the monitoring time t=0, the landslide body has no deformation; when the monitoring time t=∞, the landslide deformation converges to the maximum value U max and V max .

8. The method for monitoring the three-dimensional deformation field of a landslide according to claim 1, characterized in that: The landslide body is divided into several strips by the strip division method, and the angle between the strip and the horizontal plane is θ i , the relationship between the deformation u along the slope and the deformation v perpendicular to the slope and the monitoring time t is as follows:

9. The method for monitoring the three-dimensional deformation field of a landslide according to claim 1, characterized in that: The total loss L of the physical information neural network model is composed of the traditional loss L data and physical information loss L physics The superposition composition is expressed as: L=L data +L physics .

10. A landslide three-dimensional deformation field monitoring system, characterized by: The system can be used to implement the landslide three-dimensional deformation field monitoring method according to any one of claims 1 to 9, specifically comprising: A three-dimensional laser scanning device is used to obtain multi-period point cloud data of the landslide body; The limited inclinometer is used to obtain the time-series deformation monitoring data of the internal and external appearance of the landslide body; GNSS monitoring device, used to obtain time-series deformation data of the landslide body; A data processing module is used to solve multi-period point cloud data by using an improved ICP algorithm to obtain the surface deformation field of the landslide body. The improved ICP algorithm is used to identify corresponding point pairs between the source point cloud and the target point cloud, and to construct rotation and translation matrices based on these point pairs to ensure that the error function is minimized; A physical information neural network model building module is used to train the physical information neural network model according to the limited internal and external appearance time-series deformation monitoring data of the landslide to obtain the optimal parameters. The input of the physical information neural network model is (x, y, z, t), where (x, y, z) is the three-dimensional spatial coordinates of the monitoring point, t is the monitoring time, and the output is the displacement u along the slope and the displacement v perpendicular to the slope. The loss of the physical information neural network model includes traditional loss and physical information loss. The traditional loss is the mean square error loss, and the physical information loss is used to express the physical law of the internal and external mechanical deformation of the landslide body. The real-time monitoring module is used to deduce the three-dimensional overall time-series deformation field of the landslide based on the optimized network parameters and the trained physical information neural network model, so as to realize the real-time monitoring of the three-dimensional deformation field of the landslide.

Images