Landslide accumulation prediction method based on generative adversarial network
By combining generative adversarial networks with discrete element models, the problems of high computational cost and data scarcity in landslide accumulation prediction are solved, enabling fast and accurate landslide accumulation prediction and supporting the prediction of major engineering disasters and emergency response.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-03
- Publication Date
- 2026-03-24
AI Technical Summary
Existing technologies are computationally expensive and inefficient in landslide accumulation prediction, making it difficult to meet the needs of rapid response and dynamic assessment, and they also face the challenge of data scarcity.
A landslide accumulation prediction method based on generative adversarial networks is adopted. By constructing a landslide database, generative adversarial networks and discrete element models are combined to generate images before and after the landslide, thereby transforming the image generation problem and improving computational efficiency.
Significantly shorten the prediction response time, improve calculation efficiency, meet the timeliness requirements of geological disaster emergency response, and achieve accurate response to specific situations after landslides.
Smart Images
Figure CN121723802A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application relates to a landslide accumulation prediction method based on a generative adversarial network. BACKGROUND
[0002] Mountain landslide is a common geological disaster, which often causes casualties and property losses, and threatens the life safety of mountain residents and the stability of regional infrastructure. In particular, in the mountainous area with complex terrain and poor transportation, it is the key to disaster prevention and reduction to quickly and accurately obtain landslide accumulation information.
[0003] At present, landslide accumulation prediction mainly relies on numerical simulation method, which comprehensively considers the geometric structure of landslide body, rock-soil physical parameters and hydrogeological conditions, and realizes quantitative simulation of the movement process.
[0004] The discrete element method is widely used in landslide accumulation analysis because it can effectively handle non-continuous medium, block rupture, large deformation and particle interaction. Compared with the continuum method, the discrete element method can more truly reflect the particle flow and accumulation evolution, and is particularly suitable for the study of sliding mechanism at the particle scale, and has strong physical reality and particle behavior expression advantages. However, the discrete element method also has limitations, and its calculation and storage cost is high, especially in large-scale simulation, it is easy to encounter resource and convergence bottlenecks.
[0005] Landslide prediction also faces the dual challenges of data scarcity and real-time demand, and the traditional simulation method is low in efficiency when dealing with high nonlinear dynamic process, and it is difficult to meet the requirements of rapid response and dynamic evaluation. SUMMARY
[0006] The technical problem to be solved by the application is to provide a landslide accumulation prediction method based on a generative adversarial network.
[0007] The technical solution adopted by the application is: a landslide accumulation prediction method based on a generative adversarial network, comprising: Obtaining a two-dimensional section of a landslide body to be predicted, and based on the rock-soil material of each layer of geological unit in the landslide body, comparing the pre-set rock-soil material and image feature correspondence relationship, filling the corresponding image features for each layer in the section, and generating a pre-landslide image corresponding to the landslide body. Inputting the pre-landslide image into a trained landslide accumulation prediction model constructed based on a generative adversarial network to obtain a post-landslide image, the post-landslide image including a two-dimensional section of the landslide body after the landslide, and each layer in the section retaining the image features of each layer before the landslide.
[0008] The landslide accumulation prediction model is trained based on a landslide database, and at least part of the data in the landslide database is constructed by using a landslide data construction method.
[0009] The landslide data construction method comprises: constructing a two-dimensional profile model of the landslide mass, the model being composed of multiple layers of geological units, each layer having an adjusted elevation and thickness according to a random parameter; associating a rock-soil material with each layer of the geological units in the two-dimensional profile model, and based on the associated rock-soil material, filling the image features corresponding to the material into each layer of the geological units in combination with a preset corresponding relationship between the rock-soil material and the image features, to form a pre-landslide image; based on the pre-landslide image, simulating through a discrete element model in combination with the rock-soil mechanics parameters of the rock-soil material associated with each layer of the geological units in the image, to obtain a post-landslide image.
[0010] The method for simulating through a discrete element model in combination with the rock-soil mechanics parameters of the rock-soil material associated with each layer of the geological units in the image, to obtain a post-landslide image, comprises: filling, for each layer of the geological units, a particle unit representing the associated rock-soil material by using a discrete element model MatDEM; simulating a natural accumulation process under the action of a gravity field, to reconstruct the motion and deposition behavior of the real landslide mass material.
[0011] The method for simulating a natural accumulation process under the action of a gravity field, to reconstruct the motion and deposition behavior of the real landslide mass material, comprises: under the action of a gravity load, triggering the instability of the landslide mass, the discrete element model MatDEM simulating the interaction between particles by using a spring network, calculating the normal force and shear force, and judging the failure condition according to the Mohr-Coulomb criterion, if the shear force exceeds the limit value or the normal displacement reaches the fracture threshold, then the spring is broken and the particles are separated, forming a sliding process.
[0012] The image features comprise color features.
[0013] The generative adversarial network is based on a Pix2Pix framework, and a perception loss and a semantic loss mechanism are introduced, wherein the perception loss extracts high-level features through a pre-trained VGG network, and the semantic loss guides the image structure consistency optimization by means of an image segmentation network.
[0014] A landslide accumulation prediction device based on a generative adversarial network, comprising: a preprocessing module configured to acquire a two-dimensional profile of a to-be-predicted landslide mass, and based on the rock-soil materials of each layer of the geological units in the landslide mass, fill the corresponding image features into each layer of the profile in comparison with a preset corresponding relationship between the rock-soil material and the image features, to generate a pre-landslide image corresponding to the landslide mass; The model prediction module is used to input the pre-landslide image into a trained landslide accumulation prediction model based on a generative adversarial network to obtain the post-landslide image. The post-landslide image includes a two-dimensional profile of the landslide body after the landslide, and each layer in the profile retains the image features of each layer before the landslide.
[0015] A storage medium storing a computer program executable by a processor, wherein the computer program, when executed, implements the steps of the landslide accumulation prediction method based on generative adversarial networks.
[0016] A landslide deposition prediction device has a memory and a processor. The memory stores a computer program that can be executed by the processor. When the computer program is executed, it implements the steps of the landslide deposition prediction method based on generative adversarial networks.
[0017] The beneficial effects of this invention are as follows: This invention utilizes a large number of paired pre-landslide and post-landslide images from a landslide database to train a landslide accumulation prediction model. The geological units in each layer of the pre-landslide and post-landslide images are filled with corresponding image features according to the preset correspondence between soil and rock materials and image features, thereby transforming the numerical simulation problem into an image generation problem, improving computational efficiency, greatly shortening the prediction response time, and helping to promote its application in scenarios such as major engineering disaster prediction and emergency response decision support.
[0018] This invention generates a two-dimensional profile model, associates soil and rock materials with each geological unit in the model, and fills each layer with corresponding image features based on the soil and rock materials to form a pre-landslide image. Then, based on the pre-landslide image, a discrete element model is used to simulate the post-landslide image. The post-landslide image obtained by the discrete element model can accurately reflect the specific situation after the landslide, thus overcoming the data scarcity problem that landslide prediction still faces. Attached Figure Description
[0019] Figure 1 This is a schematic diagram of the automatically generated two-dimensional landslide profile elevation in the embodiment.
[0020] Figure 2 This is a schematic diagram of particle modeling and gravity accumulation simulation in the embodiment.
[0021] Figure 3 This is a schematic diagram of color mapping in the embodiment.
[0022] Figure 4 The diagram shows the accumulation of debris before and after a landslide under different working conditions, generated by discrete element simulation in the example.
[0023] Figure 5 This is a comparison image of a landslide before and after accumulation.
[0024] Figure 6 A schematic diagram of the accumulation before a landslide is generated for discrete element simulation.
[0025] Figure 7 This is a schematic diagram of the landslide accumulation predicted by an adversarial network. Detailed Implementation
[0026] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.
[0027] In the description of this invention, "multiple" means two or more. The use of "first" and "second" is for distinguishing technical features only and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or the order of the indicated technical features. Furthermore, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art.
[0028] Example 1: This example is a landslide accumulation prediction method based on generative adversarial networks, specifically including the following steps: S100. Obtain a two-dimensional profile of the landslide body to be predicted, and based on the soil and rock materials of each geological unit in the landslide body, fill in the corresponding image features of each layer in the profile according to the preset correspondence between soil and rock materials and image features, and generate the pre-landslide image corresponding to the landslide body.
[0029] S200. Input the pre-landslide image obtained in step S100 into the trained landslide accumulation prediction model to obtain the post-landslide image. The post-landslide image includes a two-dimensional profile of the landslide body after the landslide, and each layer in the profile retains the image features of each layer before the landslide.
[0030] In this embodiment, the landslide accumulation prediction model is built based on an improved generative adversarial network and trained using a landslide database, which contains a large number of mutually matching pre-landslide and post-landslide images.
[0031] In this embodiment, the improved generative adversarial network is based on the Pix2Pix framework and introduces perceptual loss and semantic loss mechanisms. The perceptual loss extracts high-level features through a pre-trained VGG network, enhancing the ability to express image details; the semantic loss guides the optimization of image structure consistency through an image segmentation network, thereby improving the semantic rationality and spatial accuracy of the model in landslide image prediction.
[0032] In this example, at least part of the data in the landslide database was constructed using a landslide data construction method, which includes the following steps: ① Using an automated script written in Python, a two-dimensional profile model of the landslide body is constructed. The model consists of three geological units, each with its elevation and thickness adjusted according to random parameters to achieve diversity in landform structure and meet the simulation needs of different landslide scenarios.
[0033] ② Associate soil and rock materials with each geological unit in the two-dimensional profile model, and based on the associated soil and rock materials, combined with the preset correspondence between soil and rock materials and image features, fill each geological unit with the corresponding image features to form an image before the landslide.
[0034] In this embodiment, based on the principle of geological sequence division, the landslide profile is divided into three parts along the elevation direction: bottom layer, middle layer, and upper layer.
[0035] In this example, four typical soil and rock materials are defined, and soil and rock mechanics parameters such as density, friction coefficient, compressive strength, and Young's modulus are assigned to them respectively. A unique image feature is also assigned to each type of material.
[0036] In this embodiment, the image features are color features, and different colors are used to identify each rock and soil material and fill the geological unit layer corresponding to each material. Blue represents the bottom material with the highest strength, and the other three types are identified by red, green and yellow respectively. The middle and upper layer materials are combined from the latter three to form six different configuration conditions to reflect the differences in geological conditions.
[0037] ③ Based on the pre-landslide image, and combined with the geotechnical parameters of the soil and rock materials associated with each geological unit in the image, the discrete element model is used to simulate and obtain the post-landslide image.
[0038] a. The discrete element model MatDEM is used to fill each geological unit with particle elements representing the associated soil and rock materials.
[0039] b. Simulate the natural deposition process under the action of gravity field to reconstruct the movement and deposition behavior of real landslide materials.
[0040] Under gravity load, landslide instability is triggered. The discrete element model uses a spring network to simulate the interaction between particles, calculates the normal force and shear force, and determines the failure condition based on the Mohr-Coulomb criterion. If the shear force exceeds the limit value, or the normal displacement reaches the fracture threshold, the spring breaks, the particles separate, and a sliding process occurs. (1) (2) (3) Among them, F n F represents the normal force. s F represents shear force. smax X represents the maximum shear force that the spring can withstand. n X represents the normal displacement. b F represents the fracture displacement. so μ represents the shear resistance between discrete units. p K represents the friction coefficient between discrete units. n K represents the normal stiffness. s X represents tangential stiffness. s This indicates tangential displacement.
[0041] The following example of a landslide further illustrates the application of the landslide accumulation prediction method based on discrete element simulation and generative adversarial networks in this invention.
[0042] A landslide profile was selected, and a geological profile of the area before the landslide was drawn, as shown below. Figure 6 As shown in the figure. Based on the on-site investigation data after the landslide and the high-resolution digital elevation model obtained from UAV imagery, the geological profile after the landslide was completed. The landslide material consists of Quaternary residual colluvial deposits (containing gravelly silt clay) and artificial fill (blocky gravelly soil with a gravel content of about 80%).
[0043] Based on the stratigraphic profile analysis, the residual colluvial material was set to account for 40% of the mass in the MatDEM numerical simulation, and the fill material was set to account for 60%. Through parameter analysis, the physical properties of the residual colluvial material best matched those of Soil 1 (marked in red), while the characteristics of the fill material were closest to those of Soil 2 (marked in green). Finally, the landslide material was constructed as a single sliding body in the MatDEM model. Figure 7 As shown.
[0044] Preprocessed images from before a landslide event are input into the landslide deposition prediction model to generate corresponding predicted images of landslide sediments. Compared with the traditional discrete element method, the model significantly reduces the prediction time for a single scene from 226 seconds to 0.11 seconds, achieving a speedup of 2054 times, thus meeting the timeliness requirements of the golden window for geological disaster emergency response.
[0045] In the landslide case study, the proposed model achieved a displacement prediction accuracy of 92.3%, and its computational efficiency supports parallel simulation of small-scale geological disaster scenarios. A comparative analysis was conducted on the images generated by the numerical simulation, the prediction results of the deep learning model, and the actual measured values of the landslide. The deep learning model demonstrates a significant computational efficiency advantage in landslide deposition prediction, as shown in Table 1. Therefore, the method described in this embodiment is effective and feasible.
[0046] Table 1:
[0047] Example 2: This example is a landslide deposition prediction device based on generative adversarial networks, specifically including: The preprocessing module is used to obtain a two-dimensional profile of the landslide body to be predicted, and based on the rock and soil materials of each geological unit in the landslide body, it fills the corresponding image features in each layer of the profile with the preset correspondence between rock and soil materials and image features, and generates the pre-landslide image corresponding to the landslide body. The model prediction module is used to input the pre-landslide image into a trained landslide accumulation prediction model based on a generative adversarial network to obtain the post-landslide image. The post-landslide image includes a two-dimensional profile of the landslide body after the landslide, and each layer in the profile retains the image features of each layer before the landslide.
[0048] Example 3: This example is a storage medium that stores a computer program that can be executed by a processor. When the computer program is executed, it implements the steps of the landslide accumulation prediction method based on generative adversarial networks described in Example 1.
[0049] Example 4: This example is a landslide accumulation prediction device, which has a memory and a processor. The memory stores a computer program that can be executed by the processor. When the computer program is executed, it implements the steps of the landslide accumulation prediction method based on generative adversarial networks described in Example 1.
[0050] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the aforementioned functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.
[0051] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0052] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.
[0053] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the aforementioned program can be printed, because the aforementioned program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.
[0054] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0055] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0056] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
[0057] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.
Claims
1. A landslide accumulation prediction method based on generative adversarial networks, characterized in that, include: A two-dimensional profile of the landslide body to be predicted is obtained. Based on the soil and rock materials of each geological unit in the landslide body, the corresponding image features are filled into each layer of the profile according to the preset correspondence between soil and rock materials and image features, and the pre-landslide image of the landslide body is generated. The pre-landslide image is input into a trained landslide accumulation prediction model based on a generative adversarial network to obtain a post-landslide image. This post-landslide image includes a two-dimensional profile of the landslide body after the landslide, and each layer in the profile retains the image features of each layer before the landslide.
2. The landslide accumulation prediction method based on generative adversarial networks according to claim 1, characterized in that, The landslide accumulation prediction model is trained based on a landslide database, at least some of which is constructed using landslide data construction methods.
3. The landslide accumulation prediction method based on generative adversarial networks according to claim 1, characterized in that, The landslide data construction method includes: A two-dimensional profile model of the landslide body was constructed. The model consists of multiple geological units, and the elevation and thickness of each layer are adjusted according to random parameters. Associating soil and rock materials with each geological unit in the two-dimensional profile model, and based on the associated soil and rock materials, combined with the preset correspondence between soil and rock materials and image features, filling each geological unit with the corresponding image features to form an image before the landslide; Based on the pre-landslide image, and combined with the geotechnical parameters of the soil and rock materials associated with each geological unit in the image, the post-landslide image is obtained through simulation using a discrete element model.
4. The landslide accumulation prediction method based on generative adversarial networks according to claim 3, characterized in that, The method involves using a discrete element model to simulate the landslide based on pre-landslide images and combining the geotechnical mechanical parameters of the soil and rock materials associated with each geological unit in the images, to obtain post-landslide images, including: The discrete element model MatDEM is used to fill each geological unit with particle elements representing the associated soil and rock materials. Simulate the natural deposition process under the action of gravity field to reconstruct the movement and deposition behavior of real landslide materials.
5. The landslide accumulation prediction method based on generative adversarial networks according to claim 4, characterized in that, The simulation of the natural deposition process under the action of gravity field, reconstructing the movement and deposition behavior of real landslide materials, includes: Under gravity load, the landslide body is triggered to become unstable. The discrete element model MatDEM uses a spring network to simulate the interaction between particles, calculates the normal force and shear force, and judges the failure condition according to the Mohr-Coulomb criterion. If the shear force exceeds the limit value or the normal displacement reaches the fracture threshold, the spring breaks, the particles separate, and the sliding process is formed.
6. The landslide accumulation prediction method based on generative adversarial networks according to claim 1, characterized in that, The image features include color features.
7. The landslide accumulation prediction method based on generative adversarial networks according to claim 1, characterized in that, The generative adversarial network is based on the Pix2Pix framework and introduces perceptual loss and semantic loss mechanisms. The perceptual loss extracts high-level features through a pre-trained VGG network, while the semantic loss guides the optimization of image structure consistency with the help of an image segmentation network.
8. A landslide accumulation prediction device based on generative adversarial networks, characterized in that, include: The preprocessing module is used to obtain a two-dimensional profile of the landslide body to be predicted, and based on the rock and soil materials of each geological unit in the landslide body, it fills the corresponding image features in each layer of the profile with the preset correspondence between rock and soil materials and image features, and generates the pre-landslide image corresponding to the landslide body. The model prediction module is used to input the pre-landslide image into a trained landslide accumulation prediction model based on a generative adversarial network to obtain the post-landslide image. The post-landslide image includes a two-dimensional profile of the landslide body after the landslide, and each layer in the profile retains the image features of each layer before the landslide.
9. A storage medium having a computer program stored thereon that can be executed by a processor, characterized in that, When the computer program is executed, it implements the steps of the landslide accumulation prediction method based on generative adversarial networks as described in any one of claims 1 to 7.
10. A landslide accumulation prediction device, comprising a memory and a processor, wherein the memory stores a computer program executable by the processor, characterized in that, When the computer program is executed, it implements the steps of the landslide accumulation prediction method based on generative adversarial networks as described in any one of claims 1 to 7.