Landslide disaster susceptibility intelligent evaluation method

Abstract

The invention discloses a landslide disaster susceptibility intelligent evaluation method, and relates to the technical field of geological disaster evaluation. The method has high-precision prediction capability, by fusing XGBoost (XGB), a support vector machine (SVM) and a residual error long short-term memory network (ResLSTM) model, the advantages of multi-source data and multiple models can be effectively integrated, the prediction precision of the landslide disaster susceptibility is improved, experimental results show that the AUC value can reach 0.8560, and the AUC value is remarkably superior to that of a single model or a traditional method; the method has strong feature extraction and fusion capabilities, key features can be extracted from multi-source data, and a richer and more comprehensive feature data set can be constructed by deeply fusing the features of the XGB and SVM models, so that higher-quality input is provided for a subsequent deep learning model.

Description

Technical Field

[0001] This invention relates to the field of geological hazard assessment technology, specifically to an intelligent assessment method for landslide hazard susceptibility. Background Technology

[0002] Landslides, a common form of geological hazard, pose a serious threat to human society and the natural environment. Accurately assessing landslide susceptibility is crucial for developing effective disaster prevention and mitigation strategies, protecting people's lives and property, and promoting regional sustainable development. Traditional landslide susceptibility assessment methods primarily rely on the experience of geological experts and qualitative analysis. These methods are often limited by the multi-source and complexity of data, making it difficult to comprehensively capture the influencing factors and inherent patterns of landslide occurrence.

[0003] With the development of information technology, Geographic Information Systems (GIS) and remote sensing technologies have provided new data sources and analytical methods for landslide susceptibility assessment. However, extracting useful information and identifying key influencing factors from massive amounts of geospatial data remains a challenge. In recent years, machine learning technology has made significant progress in fields such as image recognition and natural language processing, and its application in geological hazard prediction has attracted increasing attention. In particular, the development of deep learning technology has provided new solutions for processing high-dimensional data and complex nonlinear relationships.

[0004] Although existing research has attempted to apply machine learning techniques to landslide susceptibility assessment, most studies focus on the application of single models, lacking the integration of the advantages of multiple models and in-depth mining of feature information. Furthermore, current methods often neglect the spatial correlation and temporal dynamics of data during feature extraction and model training, resulting in room for improvement in the accuracy and reliability of prediction results.

[0005] Therefore, developing a landslide hazard susceptibility assessment method that can comprehensively utilize multi-source data, integrate the advantages of multiple machine learning models, and consider the spatial correlation of data is of great scientific significance and application value for improving prediction accuracy and practicality. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent evaluation method for landslide hazard susceptibility to solve the technical problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent assessment method for landslide hazard susceptibility, comprising at least the following steps:

[0008] S1: Data acquisition and preprocessing, collecting multi-source data related to landslide disasters, and completing data standardization and segmentation operations;

[0009] S2: Primary feature extraction, using the XGBoost model and the support vector machine model to extract features from the preprocessed data, and obtaining the feature sets output by each model. The XGBoost model is the XGB model, and the support vector machine model is the SVM model.

[0010] S3: Feature fusion, which deeply fuses the feature sets extracted by the XGB model and the SVM model to construct a fused feature dataset, in order to integrate key information from multiple algorithm perspectives;

[0011] S4: Advanced feature learning, which uses a fused feature dataset to train a Residual Long Short-Term Memory (ResLSTM) network model and leverages its excellent non-linear feature learning capabilities to uncover deep-seated relationships between fused features;

[0012] S5: Model optimization and validation. The performance of the trained ResLSTM model is validated by calculating the area under the receiver operating feature curve (AUC) to evaluate the predictive performance of the model.

[0013] S6: Susceptibility Prediction. The trained ResLSTM model is used to predict the susceptibility of landslide disasters, output the prediction results, and generate a susceptibility evaluation map to provide a basis for disaster prevention and control decisions.

[0014] Furthermore, S1 includes at least the following steps:

[0015] S101: Collect multi-source data related to landslide disasters, including at least elevation, slope, aspect, curvature, road distance, water system distance, vegetation index, and fault lithology;

[0016] S102: Standardize the collected data by scaling it to the range [0, 1] using the min-max normalization method, as shown in the following formula:

[0017] x′=

[0018] Where x is the original data value; and These are the minimum and maximum values ​​of the data, respectively; x′ is the normalized data value.

[0019] S103: Divide the data into training and testing sets in a 7:3 ratio to prepare for subsequent model training and validation.

[0020] Furthermore, S2 includes at least the following steps:

[0021] S201: The XGB model is used to extract features from the training set data. The XGB model constructs multiple decision trees and uses voting to arrive at the final result, effectively reducing overfitting and extracting important features from the data. The formula for calculating the predicted value of the XGB model is as follows:

[0022]

[0023] in, This represents the predicted value from the XGB model. This represents the prediction result of the k-th decision tree, where K represents the total number of decision trees;

[0024] S202: Feature extraction is performed on the training set data using the SVM model. The SVM model distinguishes different categories of data by finding a hyperplane that maximizes the boundary. The decision function of the SVM model is:

[0025] )

[0026] in: They are Lagrange multipliers; These are the sample labels; K is the kernel function. b is the support vector; b is the bias term.

[0027] Furthermore, the output calculation formula of the ResLSTM model is as follows:

[0028] y ResLSTM =σ(W o ⋅h t +b o )

[0029] Among them, y ResLSTM This represents the prediction result of the ResLSTM model; h t The hidden state of the last layer; W o and b o σ represents the weights and biases of the output layer, respectively; σ is the activation function.

[0030] Furthermore, S5 includes at least the following steps:

[0031] The trained ResLSTM model was validated using a test set.

[0032] The predictive performance of the model is quantitatively evaluated by calculating the area under the receiver operating characteristic curve (AUC).

[0033] The expression for the area under the curve is:

[0034]

[0035] in, Represents the area under the curve. This represents the ROC curve.

[0036] Furthermore, S6 includes at least the following steps:

[0037] S601: Use the trained ResLSTM model to predict new landslide data and generate landslide hazard susceptibility prediction results. The prediction results are output in the form of probability values, representing the probability of each sample experiencing a landslide.

[0038] S602: Based on the prediction results, the study area is divided into different susceptibility levels using the natural breakpoint method, which are usually divided into five levels: low susceptibility, lower susceptibility, medium susceptibility, higher susceptibility, and high susceptibility.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] 1. This invention has high-precision prediction capability. By integrating XGBoost (XGB), Support Vector Machine (SVM) and Residual Long Short-Term Memory Network (ResLSTM) models, this invention can effectively integrate the advantages of multi-source data and multiple models, improve the prediction accuracy of landslide hazard susceptibility. Experimental results show that the AUC value can reach 0.8560, which is significantly better than a single model or traditional methods.

[0041] 2. This invention has powerful feature extraction and fusion capabilities. It can not only extract key features from multi-source data, but also construct a richer and more comprehensive feature dataset by deeply fusing features from XGB and SVM models, thereby providing higher quality input for subsequent deep learning models.

[0042] 3. This invention considers the spatial correlation model design. The introduction of the ResLSTM model enables this method to handle the spatial correlation of data. By learning the spatial and temporal relationships between features, it enhances the model's understanding of landslide disaster occurrence patterns and improves the accuracy and reliability of prediction.

[0043] 4. This invention optimizes model performance. Through model optimization and verification steps, this method can verify the performance of the trained ResLSTM model, ensuring the effectiveness and robustness of the model in practical applications. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is the overall flowchart of the present invention;

[0046] Figure 2 This is a schematic diagram of the ROC curve of the present invention;

[0047] Figure 3 This is a comparison chart of multiple indicators of the present invention. Detailed Implementation

[0048] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0049] Example 1:

[0050] Please see Figure 1 and Figure 3 A method for intelligent assessment of landslide susceptibility includes at least the following steps:

[0051] S1: Data acquisition and preprocessing, collecting multi-source data related to landslide disasters, and completing data standardization and segmentation operations;

[0052] S2: Primary feature extraction, using the XGBoost model and the support vector machine model to extract features from the preprocessed data, and obtaining the feature sets output by each model. The XGBoost model is the XGB model, and the support vector machine model is the SVM model.

[0053] S3: Feature Fusion. This involves deep fusing the feature sets extracted by the XGB and SVM models to construct a fused feature dataset. This integrates key information from multiple algorithmic perspectives, concatenating the predicted probabilities output by each model column-wise to form a comprehensive feature matrix. This fusion method can combine the advantages of different models and capture the multi-dimensional features of the data.

[0054] S4: Advanced feature learning, which uses a fused feature dataset to train a Residual Long Short-Term Memory (ResLSTM) network model and leverages its excellent non-linear feature learning capabilities to uncover deep-seated relationships between fused features;

[0055] S5: Model optimization and validation. The performance of the trained ResLSTM model is validated by calculating the area under the receiver operating feature curve (AUC) to evaluate the predictive performance of the model.

[0056] S6: Susceptibility Prediction. The trained ResLSTM model is used to predict the susceptibility of landslide disasters, output the prediction results, and generate a susceptibility evaluation map to provide a basis for disaster prevention and control decisions.

[0057] S1 includes at least the following steps:

[0058] S101: Collect multi-source data related to landslide disasters. The multi-source data should include at least elevation, slope, aspect, curvature, road distance, water system distance, vegetation index, and fault lithology.

[0059] S102: Standardize the collected data by scaling it to the range [0, 1] using the min-max normalization method, as shown in the following formula:

[0060] x′=

[0061] Where x is the original data value; and These are the minimum and maximum values ​​of the data, respectively; x′ is the normalized data value.

[0062] S103: Divide the data into training and testing sets in a 7:3 ratio to prepare for subsequent model training and validation.

[0063] S2 includes at least the following steps:

[0064] S201: The XGB model is used to extract features from the training set data. The XGB model constructs multiple decision trees and uses voting to arrive at the final result, effectively reducing overfitting and extracting important features from the data. The formula for calculating the predicted value of the XGB model is as follows:

[0065]

[0066] in, This represents the predicted value from the XGB model. This represents the prediction result of the k-th decision tree, where K represents the total number of decision trees;

[0067] S202: Feature extraction is performed on the training set data using the SVM model. The SVM model distinguishes different categories of data by finding a hyperplane that maximizes the boundary. The decision function of the SVM model is:

[0068] )

[0069] in: They are Lagrange multipliers; These are the sample labels; K is the kernel function. b is the support vector; b is the bias term.

[0070] The formula for calculating the output of the ResLSTM model is:

[0071] y ResLSTM =σ(W o ⋅h t +b o)

[0072] Among them, y ResLSTM This represents the prediction result of the ResLSTM model; h t The hidden state of the last layer; W o and b o These represent the weights and biases of the output layer, respectively; σ is the activation function.

[0073] The ResLSTM model, through its unique structure, can further explore the spatial and temporal relationships between features, enhancing the model's nonlinear expressive power.

[0074] S5 includes at least the following steps:

[0075] The trained ResLSTM model was validated using a test set.

[0076] The predictive performance of the model is quantitatively evaluated by calculating the area under the receiver operating characteristic curve (AUC).

[0077] The expression for the area under the curve is:

[0078]

[0079] in, Represents the area under the curve. This represents the ROC curve.

[0080] S6 includes at least the following steps:

[0081] S601: Use the trained ResLSTM model to predict new landslide data and generate landslide hazard susceptibility prediction results. The prediction results are output in the form of probability values, representing the probability of each sample experiencing a landslide.

[0082] S602: Based on the prediction results, the study area is divided into different susceptibility levels using the natural discontinuity method, typically into five levels: low susceptibility, relatively low susceptibility, moderate susceptibility, relatively high susceptibility, and high susceptibility. This classification method can intuitively display the landslide risk level of different areas, providing clear guidance for disaster prevention and control.

[0083] Example 2:

[0084] This embodiment, based on Embodiment 1 above, details the entire process of applying the intelligent assessment method for landslide susceptibility of the present invention in Zogang County, Tibet Autonomous Region. Zogang County, located in the southeastern part of the Qinghai-Tibet Plateau, has complex geological conditions and a variable climate, making it a high-incidence area for landslides. This embodiment aims to provide a scientific basis and decision support for landslide prevention and control in Zogang County through the method of the present invention.

[0085] S1: Data Acquisition and Preprocessing

[0086] S101: Collect multi-source data for Zogang County, including elevation, slope, aspect, curvature, road distance, waterway distance, vegetation index, and fault lithology. Data sources include satellite remote sensing imagery, geographic information system (GIS) databases, and field survey data.

[0087] S102: Standardize the collected data by scaling it to the range [0,1] using the min-max normalization method, as shown in the following formula:

[0088] x′=

[0089] Where x is the original data value, x min and x max These are the minimum and maximum values ​​of the data, respectively, and x′ is the normalized data value.

[0090] S103: Divide the data into training and testing sets in a 7:3 ratio to prepare for subsequent model training and validation.

[0091] S2: Primary Feature Extraction

[0092] S201: The XGBoost (XGB) model is used to extract features from the training set data. The XGB model constructs multiple decision trees and uses voting to arrive at the final result, effectively reducing overfitting and extracting important features from the data. The formula for calculating the predicted value of the XGB model is:

[0093]

[0094] in, This represents the predicted value from the XGB model. This represents the prediction result of the k-th decision tree, where K represents the total number of decision trees.

[0095] S202: Feature extraction is performed on the training set data using a Support Vector Machine (SVM) model. The SVM model distinguishes different categories of data by finding a hyperplane that maximizes the boundary. The decision function of the SVM model is:

[0096] )

[0097] in It is a Lagrange multiplier. These are the sample labels, and K is the kernel function. is the support vector, and b is the bias term.

[0098] S3: Feature Fusion. This method combines the features extracted by the XGB and SVM models to form a new feature dataset. Specifically, it concatenates the predicted probabilities output by each model column-wise to create a comprehensive feature matrix. This fusion approach integrates the strengths of different models and captures the multi-dimensional features of the data.

[0099] S4: Advanced Feature Learning, using the fused feature dataset to train a Residual Long Short-Term Memory (ResLSTM) model. The ResLSTM model, through its unique structure, can further explore the spatial and temporal relationships between features, enhancing the model's non-linear expressive power. The output calculation formula of the ResLSTM model is as follows:

[0100] y ResLSTM =σ(W o ⋅h t +b o )

[0101] Among them, y ResLSTM h represents the prediction result of the ResLSTM model. t For the hidden state of the last layer, W o and b o σ represents the weights and biases of the output layer, respectively, and σ is the activation function.

[0102] S5: Model Optimization and Validation. The trained ResLSTM model is validated using a test set. The predictive performance is quantitatively evaluated by calculating metrics such as the area under the receiver operating characteristic curve (AUC). In some embodiments, the expression for the area under the curve can be:

[0103]

[0104] in, Represents the area under the curve. This represents the ROC curve.

[0105] S6: Susceptibility Prediction

[0106] S601: Using a trained ResLSTM model, landslide data in Zogang County is predicted to generate landslide susceptibility prediction results. The prediction results are output as probability values, representing the likelihood of a landslide occurring for each sample.

[0107] S602: Based on the prediction results, Zogang County is divided into different susceptibility levels using the natural discontinuity method, typically categorized into five levels: low susceptibility, relatively low susceptibility, moderate susceptibility, relatively high susceptibility, and high susceptibility. This classification method can intuitively display the landslide risk level in different areas, providing clear guidance for disaster prevention and control.

[0108] Example 3:

[0109] This embodiment proposes an electronic device, a computer-readable storage medium storing computer-executable instructions, and one or more processors coupled to the computer-readable storage medium and configured to execute the computer-executable instructions, such that the device performs an intelligent assessment method for landslide hazard susceptibility.

[0110] Example 4:

[0111] This embodiment proposes a readable storage medium storing computer-executable instructions, which, when executed by a processor, configure the processor to perform an intelligent assessment method for landslide hazard susceptibility.

[0112] The circuits, electronic components, and modules involved are all existing technologies, which can be fully implemented by those skilled in the art, and need not be elaborated upon. The scope of protection of this invention does not involve any improvement to the software and methods.

[0113] In summary:

[0114] The proposed method combines XGBoost (XGB), Support Vector Machine (SVM), and Residual Long Short-Term Memory (ResLSTM) models to achieve efficient and accurate prediction of landslide hazard susceptibility.

[0115] Specifically, this method first uses XGB and SVM models to extract features from landslide data, then fuses the extracted features, and finally uses the fused features to train a ResLSTM model.

[0116] Experimental results show that the present invention has high accuracy in predicting landslide susceptibility, with an AUC value of 0.8560, and can provide a reliable scientific basis for disaster prevention and control.

[0117] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for intelligent evaluation of landslide hazard susceptibility, characterized in that: At least the following steps are included: S1: Data acquisition and preprocessing, collecting multi-source data related to landslide disasters, and completing data standardization and segmentation operations; S2: Primary feature extraction, using the XGBoost model and the support vector machine model to extract features from the preprocessed data, and obtaining the feature sets output by each model. The XGBoost model is the XGB model, and the support vector machine model is the SVM model. S3: Feature fusion, which deeply fuses the feature sets extracted by the XGB model and the SVM model to construct a fused feature dataset, in order to integrate key information from multiple algorithm perspectives; S4: Advanced feature learning, which uses a fused feature dataset to train a residual long short-term memory network model and leverages its excellent non-linear feature learning capabilities to uncover deep-seated relationships between fused features. S5: Model optimization and validation. The performance of the trained ResLSTM model is validated by calculating the area under the receiver operating feature curve to evaluate the predictive performance of the model. S6: Susceptibility Prediction. The trained ResLSTM model is used to predict the susceptibility of landslide disasters, output the prediction results, and generate a susceptibility evaluation map to provide a basis for disaster prevention and control decisions.

2. The intelligent evaluation method for landslide hazard susceptibility according to claim 1, characterized in that: S1 includes at least the following steps: S101: Collect multi-source data related to landslide disasters, including at least elevation, slope, aspect, curvature, road distance, water system distance, vegetation index, and fault lithology; S102: Standardize the collected data by scaling it to the range [0, 1] using the min-max normalization method, as shown in the following formula: x′= Where x is the original data value; and These are the minimum and maximum values ​​of the data, respectively; x′ is the normalized data value. S103: Divide the data into training and testing sets in a 7:3 ratio to prepare for subsequent model training and validation.

3. The intelligent evaluation method for landslide hazard susceptibility according to claim 2, characterized in that: S2 includes at least the following steps: S201: The XGB model is used to extract features from the training set data. The XGB model constructs multiple decision trees and uses voting to arrive at the final result, effectively reducing overfitting and extracting important features from the data. The formula for calculating the predicted value of the XGB model is as follows: in, This represents the predicted value from the XGB model. This represents the prediction result of the k-th decision tree, where K represents the total number of decision trees; S202: Feature extraction is performed on the training set data using the SVM model. The SVM model distinguishes different categories of data by finding a hyperplane that maximizes the boundary. The decision function of the SVM model is: ) in: They are Lagrange multipliers; These are the sample labels; K is the kernel function. b is the support vector; b is the bias term.

4. The intelligent evaluation method for landslide hazard susceptibility according to claim 3, characterized in that: The output calculation formula of the ResLSTM model is as follows: y ResLSTM =σ(W o ⋅h t +b o ) Among them, y ResLSTM This represents the prediction result of the ResLSTM model; h t The hidden state of the last layer; W o and b o σ represents the weights and biases of the output layer, respectively; σ is the activation function.

5. The intelligent evaluation method for landslide hazard susceptibility according to claim 4, characterized in that: The S5 includes at least the following steps: The trained ResLSTM model was validated using a test set. The predictive performance of the model is quantitatively evaluated by calculating the area under the receiver operating characteristic curve. The expression for the area under the curve is: in, Represents the area under the curve. This represents the ROC curve.

6. The intelligent evaluation method for landslide hazard susceptibility according to claim 5, characterized in that: S6 includes at least the following steps: S601: Use the trained ResLSTM model to predict new landslide data and generate landslide hazard susceptibility prediction results. The prediction results are output in the form of probability values, representing the probability of each sample experiencing a landslide. S602: Based on the prediction results, the study area is divided into different susceptibility levels using the natural breakpoint method, which are usually divided into five levels: low susceptibility, lower susceptibility, medium susceptibility, higher susceptibility, and high susceptibility.

Images