A landslide dominant flow channel identification method and system based on a width learning algorithm
By using a width learning algorithm to mesh the slope and superimpose Gaussian distribution sources, the flow source current density vector is simulated, which solves the error problem caused by the simplification of flow current sources in conventional methods. This enables efficient identification and accurate inversion of the dominant flow channels in landslides, meeting the timeliness requirements of landslide monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA UNIV OF GEOSCIENCES (WUHAN)
- Filing Date
- 2026-03-04
- Publication Date
- 2026-06-16
AI Technical Summary
In landslide monitoring, conventional methods for constructing flowing current sources simplify the process to anomalies with clear boundaries and uniform energy distribution, neglecting the spatial continuity of the groundwater flow field. This leads to significant errors in identifying dominant flow channels in landslides, and deep learning methods suffer from long training times.
A width learning algorithm is adopted. By meshing the slope and superimposing multiple Gaussian distribution sources to simulate the current density vector of the flow source, and combining it with a two-dimensional Gaussian function, a width learning model is constructed to identify the dominant flow channel of the landslide. The standard deviation of the Gaussian distribution is used to characterize the spatial distribution range and intensity of the flow source. The enhancement layer extracts features and connects them to the output layer through a weight matrix to realize the inversion of the current density of the flow source.
It improves the accuracy and efficiency of identifying dominant flow channels in landslides, meets the timeliness requirements of landslide monitoring, enhances the inversion accuracy of key parameters (location, range, direction, and intensity), reduces training time, and improves the consistency of flow source current density distribution.
Smart Images
Figure CN121787610B_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of geophysical inversion technology, and more specifically, relates to a method and system for identifying landslide dominant flow channels based on a width learning algorithm. Background Technology
[0002] Deterministic inversion methods are frequently used in processing spontaneous potential (SP) data, but they rely to some extent on the initial model and regularization parameters, and are often computationally time-consuming, making it difficult to meet the timeliness requirements of monitoring geological hazards such as landslides. Deep learning methods applied to SP inversion can significantly improve computational efficiency, but the presence of numerous hyperparameters in the network increases training time and may lead to network overfitting. When generating samples, conventional methods for constructing flowing current sources often simplify them into well-defined, uniformly distributed anomalies, ignoring their spatial continuity, which contradicts the physical characteristics of real groundwater flow fields. Summary of the Invention
[0003] To address the shortcomings of existing technologies, this application aims to provide a method and system for identifying landslide dominant flow channels based on a width learning algorithm. This method addresses the problem that conventional methods for constructing flowing current sources simplify them into anomalous bodies with clear boundaries and uniform energy distribution, neglecting their spatial continuity. This contradicts the physical characteristics of real groundwater flow fields, leading to significant errors in the identification of landslide dominant flow channels obtained through intelligent learning.
[0004] The first aspect of this application relates to a method for identifying landslide dominant flow channels based on a width learning algorithm, comprising the following steps:
[0005] The natural potential data is input into the width learning model, and the flow source current density is output to identify the dominant flow channels of the landslide.
[0006] The training method for the width learning model is as follows:
[0007] The slope is gridded to generate several grids. Multiple Gaussian distribution sources are superimposed to simulate the current density vector of the flow source corresponding to each grid. The flow source current density is forward modeled to obtain the natural potential response data of the slope.
[0008] The parameters of the wide learning model are set by using the natural potential response data as input and the flow source current density as output, and then the wide learning model is trained.
[0009] In some implementations, the method for obtaining the flow source current density vector by superimposing multiple Gaussian distribution sources is as follows:
[0010] From coordinates With coordinates The center location of the flow source is characterized by the standard deviation of a Gaussian distribution, the spatial distribution range is characterized by the amplitude parameter, and the intensity of the flow source is characterized by the amplitude parameter. Combined with a two-dimensional Gaussian function, arbitrary coordinates within the slope are obtained. x , y The current density of the flow source at the location.
[0011] In some implementations, the amplitude parameter is:
[0012] ;
[0013] ;
[0014] Among them, superscript i Indicates the index of the source, subscript x and y Represent x and y Directional components; sign parameter Its value is randomly generated with equal probability; amplitude adjustment parameter and The number of effective flow sources follows a uniform distribution on the interval [-5, -2]. From the set Randomly selected from the discrete uniform distribution defined above.
[0015] In some implementations, the two-dimensional Gaussian function is:
[0016] ;
[0017] The expressions for the two directional components of the current density of the flow source are:
[0018] ;
[0019] ;
[0020] in, and Gaussian distribution x direction and y Standard deviation of direction.
[0021] In some implementations, the input layer of the width learning model is used to extract features from the surface natural potential data, the enhancement layer is used to enhance the features extracted from the mapped feature group through enhancement nodes, and the output layer is connected to the final layer, which is composed of the enhancement layer integrating the input features and enhancement nodes, through a weight matrix.
[0022] In some implementations, the function of the input layer is: ;
[0023] Where X is a of size A 3D matrix, representing the surface natural potential data, serves as the input sample set; Indicates the first k Each mapping feature matrix; and represent the random weight matrix and the bias term, respectively, whose initial values are generated by a standard uniform distribution in the range [-1, 1]. and The dimensions are respectively and ; For each mapping feature group k The number of feature nodes; The number of mapping features; For mapping functions; N The number of samples contained in the input sample set; L The number of potential measurement points corresponding to each input sample;
[0024] The function of the enhancement layer is: ;
[0025] in, Indicates the first j A matrix of augmented nodes; For all mapping features of the input sample set X; Representing the j A random weight matrix; Representing the j 1 random bias vector; nonlinear activation function The hyperbolic tangent sigmoid transfer function; M To increase the number of nodes;
[0026] The function for the output layer is: ;
[0027] in, To enhance the node set; Here is the weight matrix; Y is a matrix of size . The dimensional matrix represents the set of current density models of flow sources within the slope, serving as the output sample set; m The number of grid cells corresponding to each output sample.
[0028] In some implementations, the number of feature nodes, the number of feature layer windows, and the number of enhancement nodes are set to 36, 11, and 160, respectively.
[0029] The second aspect of this application relates to a landslide dominant flow channel identification system based on a width learning algorithm, comprising:
[0030] The identification module is used to input natural potential data into the width learning model and output the flow source current density to identify the dominant flow channels of the landslide.
[0031] The training set acquisition module is used to generate several grids by meshing the slope. By superimposing multiple Gaussian distribution sources, the current density vector of the flow source corresponding to each grid is simulated. The forward modeling of the flow source current density is performed to obtain the potential response data of the slope.
[0032] The training module is used to train the width learning model by taking potential response data as input, current density of the flow source as output, setting the parameters of the width learning model, and training the width learning model.
[0033] In some implementations, the training set acquisition module includes: a current density acquisition unit, used to obtain data from coordinates... With coordinates The center location of the flow source is characterized by the standard deviation of a Gaussian distribution, the spatial distribution range is characterized by the amplitude parameter, and the intensity of the flow source is characterized by the amplitude parameter. Combined with a two-dimensional Gaussian function, arbitrary coordinates within the slope are obtained. x , y The current density of the flow source at the location.
[0034] In some implementations, the amplitude parameter in the current density acquisition unit is:
[0035] ;
[0036] ;
[0037] Among them, superscript i Indicates the index of the source, subscript x and y Represent x and y Directional components; sign parameter Its value is randomly generated with equal probability; amplitude adjustment parameter and The number of effective flow sources follows a uniform distribution on the interval [-5, -2]. From the set Randomly selected from the discrete uniform distribution defined above;
[0038] The two-dimensional Gaussian function is:
[0039] ;
[0040] Current density of the flow source x Directional components and y The expression for the directional component is:
[0041] ;
[0042] ;
[0043] in, and They are Gaussian distributions respectively x direction and y Standard deviation of direction.
[0044] In some implementations, the input layer of the width learning model is used to extract features from the surface spontaneous potential data, the enhancement layer is used to enhance the features extracted from the mapped feature group through enhancement nodes, and the output layer is connected to the final layer composed of the enhancement layer integrating the input features and enhancement nodes through a weight matrix.
[0045] The function of the input layer is: ;
[0046] Where X is a of size A 3D matrix, representing the surface natural potential data, serves as the input sample set; Indicates the first k Each mapping feature matrix; and represent the random weight matrix and the bias term, respectively, whose initial values are generated by a standard uniform distribution in the range [-1, 1]. and The dimensions are respectively and ; For each mapping feature group k The number of feature nodes; The number of mapping features; For mapping functions; N The number of samples contained in the input sample set; L The number of potential measurement points corresponding to each input sample;
[0047] The function of the enhancement layer is: ;
[0048] in, Indicates the first j A matrix of augmented nodes; For all mapping features of the input sample set X; Representing the j A random weight matrix; Representing the j 1 random bias vector; nonlinear activation function The hyperbolic tangent sigmoid transfer function; M To increase the number of nodes;
[0049] The function for the output layer is: ;
[0050] in, To enhance the node set; Here is the weight matrix; Y is a matrix of size . The dimensional matrix represents the set of current density models of flow sources within the slope, serving as the output sample set; m The number of grid cells corresponding to each output sample.
[0051] Thirdly, this application provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method described in the first aspect or any possible implementation thereof.
[0052] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.
[0053] Fifthly, this application provides a computer program product that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.
[0054] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.
[0055] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art:
[0056] Conventional methods for constructing flow current sources have significantly limited accuracy in parameter inversion. Only the reconstruction of a few parameters basically matches theoretical expectations, while key flow field characteristics are severely distorted. In contrast, this application considers the continuity and spatial correlation of the distribution of underground flow current sources. By superimposing multiple Gaussian distribution sources to simulate the flow current density vector, the current density distribution of the constructed flow source not only conforms to the current density distribution characteristics generated by the dominant flow source on the slope, but also improves the inversion accuracy of key parameters (location, range, direction, and intensity), thereby effectively identifying the dominant flow channels of the landslide.
[0057] This application provides a method for identifying dominant flow channels in landslides based on a width learning algorithm. Due to its simpler network architecture and shorter training time, the width learning method can achieve rapid inversion of natural potential data, thereby better meeting the timeliness requirements of landslide monitoring. Attached Figure Description
[0058] Figure 1 This is a flowchart illustrating the process of retrieving natural potential data using the BLS method, as provided in an embodiment of this application.
[0059] Figure 2 This is a natural potential data curve of the land surface (slope) observed in the embodiments of this application.
[0060] Figure 3 The present application provides a width learning framework for natural potential data inversion, which consists of, from top to bottom: an input layer (surface natural potential data), a mapping feature layer, an enhancement layer, and an output layer (underground flow source current density).
[0061] Figure 4(a) shows the components of the underground flow source current density provided in the embodiments of this application. Schematic diagram.
[0062] Figure 4(b) shows the components of the underground flow source current density provided in the embodiments of this application. Schematic diagram.
[0063] Figure 5(a) shows the components obtained by inversion from a well-trained BLS network model provided in the embodiments of this application. Prediction results chart.
[0064] Figure 5(b) shows the components provided in the embodiments of this application. The actual value.
[0065] Figure 5(c) shows the components obtained by inversion from a well-trained BLS network model provided in the embodiments of this application. Prediction results chart.
[0066] Figure 5(d) shows the components provided in the embodiments of this application. The actual value.
[0067] Figure 5(e) is a comparison between the natural potential response curve obtained by forward modeling the prediction result obtained by the BLS network training model provided in the embodiment of this application and the actual natural potential response curve. Detailed Implementation
[0068] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0069] In this application, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A existing alone, A and B existing simultaneously, and B existing alone. In this application, the symbol " / " indicates that the related objects are in an "or" relationship, for example, A / B means A or B.
[0070] In this application, the terms “first” and “second” are used to distinguish different objects, rather than to describe a specific order of objects.
[0071] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.
[0072] In the description of the embodiments in this application, unless otherwise stated, "multiple" means two or more.
[0073] The embodiments of this application are described below with reference to the accompanying drawings.
[0074] Example 1
[0075] like Figure 1 As shown, this application provides a method for identifying landslide dominant flow channels based on a width learning algorithm, including the following steps:
[0076] Step S1: Mesh the slope to generate... m The algorithm iterates through grids to obtain the flow source current density at each grid, performs forward modeling on the flow source current density, and obtains the slope surface current density. L The natural potential response data at each measuring point, such as Figure 2 As shown, 10,000 samples are provided as the training dataset for the width learning model; where the input sample set X of the width learning model is the natural potential data observed on the land surface (slope), and the output sample set Y is the current density of the flow source. ;
[0077] It should be noted that the width learning model is a supervised learning model that relies on a dataset to learn mapping relationships. It requires the generation of a large number of samples. The size of the training dataset will affect the overall performance and computational cost of the width learning model. Choosing an appropriate number of samples is important. In order to minimize the computational overhead associated with the generation and training of large-scale samples, this application selects 10,000 samples as the training dataset for the width learning model.
[0078] Step S2: Using the natural potential response data as input and the flow source current density as output, set the parameters of the width learning model. The parameters of the width learning model include the number of feature nodes ( P ), number of feature layer windows ( Q ) and the number of enhanced nodes ( M Different parameter combinations can be selected to improve the accuracy and robustness of natural potential data inversion.
[0079] In some implementations, the training dataset includes natural potential response data and flow source current density generated through forward modeling;
[0080] In real-world physical scenarios, real flowing current sources typically do not possess idealized characteristics of sharp boundaries and uniform energy. To better reflect actual physical phenomena and conform to relevant physical laws, multiple Gaussian distribution sources are superimposed to simulate the flowing current density vector, thereby achieving a probabilistic characterization of the geometric shape, flow direction, and relative intensity of the flowing current source.
[0081] In some specific embodiments, the method for obtaining the current density of the flow source is as follows:
[0082] Step S1.1: Parameterize the flow source as follows: The center position of the source is determined by coordinates... With coordinates The spatial distribution range is described by the standard deviation of a Gaussian distribution. and The intensity of the source is represented by the amplitude parameter. and The two parameters are described as corresponding to the horizontal and vertical components of the flow current density, respectively.
[0083] Regarding the first i Individual source ( i =1, 2, 3, 4), define the amplitude parameters:
[0084] (1);
[0085] (2);
[0086] Among them, superscript i Indicates the index of the source, subscript x and y Represent x and y Directional components; sign parameter Its value is randomly generated with equal probability; amplitude adjustment parameter and The flow follows a uniform distribution on the interval [-5, -2]; the number of effective flow sources From the set Randomly selected from the discrete uniform distribution defined above;
[0087] The two-dimensional Gaussian function is defined as follows:
[0088] (3);
[0089] Step S1.2: Using the slope as the geological background, 15,000 flow source current densities are randomly constructed in the area above the bedrock surface based on the Gaussian distribution (formula (3)); for each model, the source parameters are randomly assigned, and the number of effective flow current sources included in each implementation is determined by the parameters. The control parameter is randomly set between 2 and 4; as shown in Figures 4(a) and 4(b), the two directional components of the current density are... and The expression is as follows:
[0090] (4);
[0091] (5);
[0092] The process ultimately generated a total of 15,000 paired samples, including the flow current density distribution and its corresponding natural potential response.
[0093] In some implementations, such as Figure 3 As shown, step S3 is the process of setting different parameter configurations for the width learning model, specifically as follows:
[0094] The BLS (Breadth-of-Layer Learning) model approximates the underlying relationship by fitting a function mapping between the input and output datasets. In this application, it is assumed that the number of samples used for network training is... N The number of potential measurement points corresponding to each input sample is L The input sample set X consists of surface natural potential data, and the output sample set Y consists of underground flow source current density. Based on the BLS network architecture, the features of the surface natural potential data need to be extracted as the input layer first.
[0095] (6);
[0096] Where X is a of size A 3D matrix, representing the surface natural potential data, serves as the input sample set; Indicates the first k Each mapping feature matrix; and Let these represent the random weight matrix and the bias term, respectively, whose initial values are generated by a standard uniform distribution in the range [-1, 1]; where the matrix... and The dimensions are respectively and ; Each mapping feature group k The number of feature nodes; It is the number of mapping features; For mapping functions; sparse autoencoders adapt... To achieve input data compression and feature extraction;
[0097] The input data features extracted from the mapped feature group are enhanced by enhancing nodes, as follows:
[0098] (7)
[0099] in, Indicates the first j A matrix of augmented nodes; For all mapping features of the input sample set X; Representing the j A random weight matrix; Representing the j 1 random bias vector; nonlinear activation function The hyperbolic tangent sigmoid transfer function; M To increase the total number of nodes;
[0100] The output layer Y is connected to the final layer, which consists of an enhancement layer integrating input features and enhancement nodes, via a weight matrix:
[0101] (8);
[0102] in, To enhance the node set; This is the weight matrix;
[0103] It is the only unknown matrix in the width learning model, and can be solved by ridge regression combined with the pseudo-inverse method:
[0104] (9);
[0105] Therefore, a well-trained weight matrix with optimal weights can be used. The BLS model is used to invert surface natural potential data. When new surface natural potential data is obtained, the input data X can be replaced in equation (6) to achieve efficient inversion. Based on this, the mapping feature matrix and enhancement matrix generated by following the same calculation process as the training phase are multiplied with the determined weight matrix to obtain the underground flow source current density (equation (8)).
[0106] Set the width learning model, for example P , Q and M The parameters were set to 36, 11, and 160 respectively. Based on this parameter combination, the model was retrained to obtain the final prediction result.
[0107] Figures 5(a) and 5(c) show the components obtained from a well-trained model using the width learning model, respectively. and The prediction results are shown in Figures 5(b) and 5(d), which are the component prediction results respectively. and The true value; as shown in Figures 5(a) to 5(d), this method recovers multiple current sources; by comparing the components and The predicted results and actual values show that this method can reconstruct the location, range, direction, and intensity of the current source in flow source inversion. Further forward simulation of the predicted results yields natural potential response curves. Comparison with actual data reveals a high degree of fit between the two curves, with a coefficient of determination (COP) of [missing value]. The value is 0.9956, as shown in Figure 5(e).
[0108] Constructing a flowing current source using conventional methods significantly limits the accuracy of parameter inversion. Only the reconstruction of a few parameters basically matches the theoretical expectations, while the key flow field characteristics are severely distorted. In contrast, this application considers the continuity and spatial correlation of the distribution of underground flowing current sources. By superimposing multiple Gaussian distribution sources to simulate the flowing current density vector, it not only makes the current density distribution of the flowing source conform to the current density distribution characteristics generated by the dominant flowing source on the slope, but also effectively improves the inversion accuracy of key parameters (location, range, direction, and intensity).
[0109] In the synthetic model experiments, the same training and validation sets as BLS were used, and for CNN (Convolutional Neural Networks)... The BLS (Convolutional Neural Network) method uses a test set containing 1000 samples. The CNN network structure used in this application includes an input layer, three convolutional blocks, a fully connected layer module, and an output layer. Prediction results show that, under the same training and test set conditions, both methods can reconstruct the spatial distribution characteristics (including location, range, direction, and intensity) of underground flowing current sources. However, the training time for the BLS method with 10,000 samples is approximately 1 / 400 of that of the CNN method. The BLS method also shows advantages in prediction accuracy (RMSE reduction of approximately 69.4%) and inversion efficiency (speed increase of approximately 39.9%). Comparing the potential response curves obtained by forward modeling the prediction results of the two methods, it is found that the simulated curve of the BLS method shows a higher degree of matching with the real curve in terms of morphological features and amplitude changes. Statistical analysis of the RMSE of the 1000 test samples of the two methods shows that the RMSE of the two methods on the test samples exhibits the same statistical characteristics. Analysis of the correlation coefficient shows that the BLS method has a higher proportion of samples with higher correlation between the predicted results and the real values.
[0110] Example 2
[0111] The second aspect of this application relates to a landslide dominant flow channel identification system based on a width learning algorithm, comprising:
[0112] The identification module is used to input natural potential data into the width learning model and output the flow source current density to identify the dominant flow channels of the landslide.
[0113] The training set acquisition module is used to generate a grid of the slope. m The flow source current density vector corresponding to each grid is simulated by superimposing multiple Gaussian distribution sources. The flow source current density is then forward modeled to obtain the natural potential response data of the slope.
[0114] The training module is used to train the wide learning model by taking the natural potential response data as input, the flow source current density as output, setting the parameters of the wide learning model, and training the wide learning model.
[0115] In some implementations, the training set acquisition module includes: a current density acquisition unit, used to obtain data from coordinates... With coordinates The center location of the flow source is characterized by the standard deviation of a Gaussian distribution, the spatial distribution range is characterized by the amplitude parameter, and the intensity of the flow source is characterized by the amplitude parameter. Combined with a two-dimensional Gaussian function, arbitrary coordinates within the slope are obtained. x , yThe current density of the flow source at the location.
[0116] In some implementations, the amplitude parameter in the current density acquisition unit is:
[0117] ;
[0118] ;
[0119] Among them, superscript i The index represents the source, and the subscript represents... x and y Directional components; sign parameter Its value is randomly generated with equal probability; amplitude adjustment parameter and The number of effective sources follows a uniform distribution on the interval [-5, -2]. From the set Randomly selected from the discrete uniform distribution defined above;
[0120] The two-dimensional Gaussian function is:
[0121] ;
[0122] Current density of the flow source x Directional components and y The expression for the directional component is:
[0123] ;
[0124] ;
[0125] in, and They are Gaussian distributions respectively x direction and y Standard deviation of direction.
[0126] In some implementations, the input layer of the width learning model is used to extract features from the surface spontaneous potential data, the enhancement layer is used to enhance the features extracted from the mapped feature group through enhancement nodes, and the output layer is connected to the final layer composed of the enhancement layer integrating the input features and enhancement nodes through a weight matrix.
[0127] The function of the input layer is: ;
[0128] Where X is a of size A 3D matrix, representing the surface natural potential data, serves as the input sample set; Indicates the first k Each mapping feature matrix; and represent the random weight matrix and the bias term, respectively, whose initial values are generated by a standard uniform distribution in the range [-1, 1]. and The dimensions are respectively and ; For each mapping feature group k The number of feature nodes; The number of mapping features; For mapping functions; N The number of samples contained in the input sample set; L The number of potential measurement points corresponding to each input sample;
[0129] The function of the enhancement layer is: ;
[0130] in, Indicates the first j A matrix of augmented nodes; For all mapping features of the input sample set X; Representing the j A random weight matrix; Representing the j 1 random bias vector; nonlinear activation function The hyperbolic tangent sigmoid transfer function; M To increase the number of nodes;
[0131] The function for the output layer is: ;
[0132] in, To enhance the node set; Here is the weight matrix; Y is a matrix of size . The dimensional matrix represents the set of current density models of flow sources within the slope, serving as the output sample set; m The number of grid cells corresponding to each output sample.
[0133] It is understood that the detailed functional implementation of each of the above units / modules can be found in the description in the aforementioned method embodiments, and will not be repeated here.
[0134] It should be understood that the above system is used to execute the methods in the above embodiments. The corresponding program modules in the device are similar in implementation principle and technical effect to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.
[0135] Based on the methods in the above embodiments, this application provides an electronic device that may include a processor, a communications interface, a memory, and a communication bus, wherein the processor, communications interface, and memory communicate with each other via the communication bus. The processor may invoke logical instructions stored in the memory to execute the methods in the above embodiments.
[0136] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, 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 application.
[0137] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.
[0138] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.
[0139] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.
[0140] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.
[0141] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).
[0142] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.
[0143] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.
Claims
1. A method for identifying dominant flow channels in landslides based on a width learning algorithm, characterized in that, Includes the following steps: The natural potential data is input into the width learning model, and the flow source current density is output to identify the dominant flow channels of the landslide. The training method for the width learning model is as follows: The slope is gridded to generate several grids. Multiple Gaussian distribution sources are superimposed to simulate the current density vector of the flow source corresponding to each grid. The flow source current density is forward modeled to obtain the natural potential response data of the slope. Using natural potential response data as input to the width learning model and flow source current density as output, the parameters of the width learning model are set, and the width learning model is trained. The method for obtaining the current density vector of a flow source by superimposing multiple Gaussian distribution sources is as follows: From coordinates With coordinates The center location of the flow source is characterized by the standard deviation of a Gaussian distribution, the spatial distribution range is characterized by the amplitude parameter, and the intensity of the flow source is characterized by the amplitude parameter. Combined with a two-dimensional Gaussian function, arbitrary coordinates within the slope are obtained. x , y Current density of the flow source at the location; The amplitude parameter is: ； ； Among them, superscript i Indicates the source index, subscript x and y Represent x and y Directional components; sign parameter Its value is randomly generated with equal probability; amplitude adjustment parameter and The number of effective flow sources follows a uniform distribution on the interval [-5, -2]. From the set Randomly selected from the discrete uniform distribution defined above; The two-dimensional Gaussian function is: ； Current density of the flow source x Directional components and y The expression for the directional component is: ； ； in, and They are Gaussian distributions respectively x direction and y Standard deviation of direction.
2. The method for identifying dominant flow channels in landslides according to claim 1, characterized in that, The input layer of the width learning model is used to extract features from the surface natural potential data, the enhancement layer is used to enhance the features extracted by the mapped feature group through enhancement nodes, and the output layer is connected to the final layer composed of the enhancement layer integrating the input features and enhancement nodes through a weight matrix.
3. The method for identifying dominant flow channels in landslides according to claim 2, characterized in that, The function of the input layer is: ; Where X is a of size A 3D matrix, representing the surface natural potential data, serves as the input sample set; Indicates the first k Each mapping feature matrix; and represent the random weight matrix and the bias term, respectively, whose initial values are generated by a standard uniform distribution in the range [-1, 1]. and The dimensions are respectively and ; For each mapping feature group k The number of feature nodes; The number of mapping features; For mapping functions; N The number of samples contained in the input sample set; L The number of potential measurement points corresponding to each input sample; The function of the enhancement layer is: ; in, Indicates the first j A matrix of augmented nodes; For all mapping features of the input sample set X; Representing the j A random weight matrix; Representing the j 1 random bias vector; nonlinear activation function The hyperbolic tangent sigmoid transfer function; M To increase the number of nodes; The function for the output layer is: ; in, To enhance the node set; Here is the weight matrix; Y is a matrix of size . The dimensional matrix represents the set of current density models of flow sources within the slope, serving as the output sample set; m The number of grid cells corresponding to each output sample.
4. A landslide dominant flow channel identification system based on a width learning algorithm, characterized in that, include: The identification module is used to input natural potential data into the width learning model and output the flow source current density to identify the dominant flow channels of the landslide. The training set acquisition module is used to generate several grids by meshing the slope. By superimposing multiple Gaussian distribution sources, the current density vector of the flow source corresponding to each grid is simulated. The forward modeling of the flow source current density is performed to obtain the natural potential response data of the slope. The training module is used to set the parameters of the wide learning model by taking the natural potential response data as the input of the wide learning model and the current density of the flow source as the output of the wide learning model. The training set acquisition module includes: a current density acquisition unit, used to obtain the current density from coordinates. With coordinates The center location of the flow source is characterized by the standard deviation of a Gaussian distribution, the spatial distribution range is characterized by the amplitude parameter, and the intensity of the flow source is characterized by the amplitude parameter. Combined with a two-dimensional Gaussian function, arbitrary coordinates within the slope are obtained. x , y Current density of the flow source at the location; The amplitude parameter in the current density acquisition unit is: ； ； Among them, superscript i Indicates the source index, subscript x and y Represent x and y Directional components; sign parameter Its value is randomly generated with equal probability; amplitude adjustment parameter and The number of effective flow sources follows a uniform distribution on the interval [-5, -2]. From the set Randomly selected from the discrete uniform distribution defined above; The two-dimensional Gaussian function is: ； Current density of the flow source x Directional components and y The expression for the directional component is: ； ； in, and They are Gaussian distributions respectively x direction and y Standard deviation of direction.
5. The landslide dominant flow channel identification system according to claim 4, characterized in that, The input layer of the width learning model is used to extract features from the surface natural potential data, the enhancement layer is used to enhance the features extracted by the mapped feature group through enhancement nodes, and the output layer is connected to the final layer composed of the enhancement layer integrating the input features and enhancement nodes through a weight matrix. The function of the input layer is: ; Where X is a of size A 3D matrix, representing the surface natural potential data, serves as the input sample set; Indicates the first k Each mapping feature matrix; and represent the random weight matrix and the bias term, respectively, whose initial values are generated by a standard uniform distribution in the range [-1, 1]. and The dimensions are respectively and ; For each mapping feature group k The number of feature nodes; The number of mapping features; For mapping functions; N The number of samples contained in the input sample set; L The number of potential measurement points corresponding to each input sample; The function of the enhancement layer is: ; in, Indicates the first j A matrix of augmented nodes; For all mapping features of the input sample set X; Representing the j A random weight matrix; Representing the j 1 random bias vector; nonlinear activation function The hyperbolic tangent sigmoid transfer function; M To increase the number of nodes; The function for the output layer is: ; in, To enhance the node set; Here is the weight matrix; Y is a matrix of size . The dimensional matrix represents the set of current density models of flow sources within the slope, serving as the output sample set; m The number of grid cells corresponding to each output sample.