A landslide risk prediction method based on the classification of slope steepness

Abstract

The present invention provides a landslide risk prediction method based on the classification of the gentle and steep degrees of slope bodies. First, based on the digital elevation model (DEM), the slopes of the potential hazard areas are classified into gentle dip slope body areas and steep dip slope body areas; then, the genetic algorithm (GA) is used to optimize the parameters of the BP neural network to construct a landslide risk prediction model for gentle dip slope bodies, namely the GA-BP model, and the particle swarm optimization (PSO) algorithm is used to obtain the optimal parameters of the bidirectional gated recurrent unit (Bi-GRU) neural network model to construct a landslide risk prediction model for steep dip slope bodies, namely the PSO-Bi-GRU model; finally, the monitoring data of the potential hazard areas are collected and input into the landslide risk prediction model to obtain the risk prediction results. The solution of the present invention constructs different landslide risk prediction models for different types of terrain slope bodies, improving the accuracy of landslide risk prediction.

Description

Technical Field

[0001] The present invention relates to the fields of landslide risk prediction and computer deep learning, and in particular to a landslide risk prediction method based on the classification of the gentle or steep degree of a slope body. Background Art

[0002] The geological environment conditions in mountainous areas are usually relatively fragile. For example, in mountainous areas with alpine canyon landforms and complex and diverse geological structures, when the mountain body is affected by external inducing factors, geological disasters such as collapses, landslides, and debris flows are likely to occur. This will cause serious losses to the lives and property of the people in the disaster area. Among all natural disasters, landslides are a common natural disaster in mountainous areas, and it is very necessary to predict the risk of landslide disasters.

[0003] The evolution of a landslide is a dynamic process. When the external conditions change, the risk of the landslide also changes accordingly. Due to the terrain structure and complex climate conditions in mountainous areas, it is difficult for traditional prediction methods to extract hidden risk patterns from a large amount of heterogeneous data, resulting in low accuracy of risk prediction results.

[0004] The existing landslide risk prediction models can be divided into 3 categories: manual evaluation models, physical mechanics models, and data-driven models. Manual evaluation models, such as the expert scoring method, rely on the experience and knowledge structure of experts themselves, and their prediction accuracy is greatly affected by subjective factors. The analysis of physical mechanics models has the advantages of clear physical meaning and accurate analysis results, but requires a large number of geological and hydrological parameters, and is more suitable for single slope body risk prediction at specific monitoring points. Data-driven models make decisions and take actions based on data analysis, and form an automated decision-making model through training and fitting on the basis of data. This method relies on a large amount of data input to help the model learn patterns, make predictions, and make decisions. Based on the diversification of data acquisition methods, data-driven models are generally applied to the rapid prediction of regional landslide risks and can capture complex non-linear relationships. Among data-driven models, machine learning models have gradually become the most widely used landslide risk prediction models. Commonly used machine learning model methods include Support Vector Machine (SVM), Logistic Regression (LR), Artificial Neural Network (ANN), and Decision Tree (DT), etc.

[0005] As an important method in statistical analysis models, artificial neural networks can be used to analyze complex and incoherent data patterns at different scales. They have obvious advantages, especially in solving uncertainty or nonlinear problems. They learn the patterns and laws of landslide occurrence from a large amount of historical landslide data and are then used to predict landslide risks. Among them, the Back Propagation (BP) neural network is the most widely used artificial neural network. The BP neural network has a strong nonlinear mapping ability and a flexible network structure. By establishing a three-layer perceptron, it can learn and obtain the weights of the input layer, and the model weights are the key to accurate landslide prediction. However, the BP neural network also has some limitations. For example, the nonlinear relationship between influencing factors is relatively complex, it is difficult to determine the weights of each index in the model, and at the same time, its generalization ability is limited, resulting in uncertain simulation results. Therefore, directly applying the BP neural network method to landslide risk prediction, its accuracy is not high.

[0006] At present, some neural network prediction model methods can be used for landslide risk prediction of mountain slopes, but most of them only predict the landslide risks of the entire region based on a single model without considering the complexity of the regional environment, which restricts the prediction effect of the model and leads to low accuracy of existing landslide risk prediction methods in complex terrain and geomorphic monitoring areas. Summary of the Invention

[0007] Aiming at the problem of low accuracy of existing landslide risk prediction methods in complex terrain and geomorphic monitoring areas, the present invention proposes a landslide risk prediction method based on the classification of slope gentle and steep degrees, which can improve the accuracy of landslide risk prediction. The specific process of a landslide risk prediction method based on the classification of slope gentle and steep degrees provided by the present invention is as follows Figure 1 shown, and the method includes:

[0008] Step S1: Determine the potential landslide area. According to the geological condition factors of the monitoring area, select the area with relatively high landslide risk as the potential area for monitoring;

[0009] Step S2: Classify the slopes of the potential area based on the Digital Elevation Model (DEM). By processing the DEM data of the monitoring area and using the logistic regression algorithm, automatically classify the slopes of the monitoring area. According to the different gentle and steep degrees of the monitoring area, the slopes of the monitoring area are divided into two categories, one is the gently inclined slope area, and the other is the steeply inclined slope area;

[0010] Step S3: Learn the landslide risk prediction model for gently inclined slopes. Combine the BP neural network model and the Genetic Algorithm (GA). Use the Genetic Algorithm (GA) to optimize the parameters of the BP neural network, obtain the optimal number of hidden nodes, and optimize the neural network model, so as to learn the risk prediction model for the gently inclined slope area, that is, the GA-BP model;

[0011] Step S4: Learning of the landslide risk prediction model for steep dip slope bodies. The particle swarm optimization (PSO) algorithm is used to obtain the optimal parameters of the bidirectional gated recurrent unit (Bi-GRU) neural network model, thereby learning the risk prediction model for the steep dip slope body region, i.e., the PSO-Bi-GRU model;

[0012] Step S5: Collection of monitoring data for potential hazard areas. In the determined landslide potential hazard areas, data for these areas are collected, including data monitored by various sensors in the monitoring area, as well as meteorological, hydrological, geological, and geological remote sensing data for the monitoring area obtained through external systems;

[0013] Step S6: Landslide risk prediction and result visualization for slope bodies in potential hazard areas. The data monitored by various sensors collected, as well as the meteorological, hydrological, geological, and geological remote sensing data for the monitoring area, are input into the trained landslide risk prediction models for gentle dip slope bodies and steep dip slope bodies, respectively, to predict the landslide risks for gentle dip slope bodies and steep dip slope bodies, obtaining the result values of the risk prediction. Then, a map plugin is used to perform visualization processing on the result values of the risk prediction.

[0014] Furthermore, the said Step S2 includes:

[0015] S201: Obtain DEM data. High-quality digital elevation model data are obtained from publicly available professional website systems;

[0016] S202: Select the calculation resolution;

[0017] S203: Gridification. The ground surface is divided into regular grid cells, and each grid cell contains multiple DEM data points, that is, multiple elevation values;

[0018] S204: Calculate the elevation difference. For each grid cell, calculate the elevation value of the center point, and then calculate the elevation difference between other points in this grid cell and the center point;

[0019] S205: Calculate the horizontal distance. For each grid cell, calculate the horizontal distance between the center point and other points;

[0020] S206: Calculate the slope angle. The following formula is used to calculate the slope angle from the center point to other points in each grid cell: Slope angle = arctan(elevation difference / horizontal distance);

[0021] S207: Convert the calculated slope angle to degrees. The calculated slope angle is in radians. By multiplying the radian value by 180 / π, the radian value is converted to degrees;

[0022] S208: Analyze and visualize the calculated data. According to the calculated slope of the slope body, use the logistic regression algorithm to automatically classify the slope body types in the monitoring area into gentle dip slope body areas and steep dip slope body areas, and perform visualization processing on the calculated slope data to intuitively display the slope distribution of the terrain.

[0023] In a possible implementation, the slope of the gentle dip slope body is distributed between 10° and 30°, and the slope of the steep dip slope body is greater than 30°.

[0024] In a possible implementation, the Sigmoid function is used as the main calculation formula of the logistic regression binary classification model, which satisfies the following formula:

[0025]

[0026] In the formula, x is the input slope of the slope body, and g(x) is the classification result.

[0027] Furthermore, the step S3 includes:

[0028] S301: Obtain the geological condition data and meteorological data of the monitoring area from public websites and input them into the BP neural network model. Determine the weights of each parameter in the model through the connection matrix between the input layer and the hidden layer. The calculation formula is as follows:

[0029]

[0030] Among them, v jl is the element of the connection matrix, m represents the number of input sample indicators, k is the number of hidden nodes and the optimal fitness value;

[0031] S302: Use GA to optimize the parameters of BP to obtain the optimal number of hidden nodes k, specifically including:

[0032] Step (1), determine the coding range and length. The coding range is [-1, 1], and the length is 61;

[0033] Step (2), determine the coding method. Adopt the real number coding method, and the coding value is a random number within the coding range;

[0034] Step (3), determine the fitness function: In the formula, k is the coefficient, n is the number of training set samples, t i is the true value of the i-th sample, y i is the predicted value of the i-th sample;

[0035] Step (4), select individuals according to the fitness by using the tournament method;

[0036] Step (5), the selected individuals generate offspring through crossover and mutation operations;

[0037] Step (6), calculate the fitness of the offspring;

[0038] Step (7), combine the offspring and the individuals of the previous generation, and sort them according to the fitness;

[0039] Step (8), if the number of iterations is greater than 20 or the best fitness value remains unchanged after 10 iterations, end the algorithm to obtain the number of hidden nodes and the best fitness value. Otherwise, repeat steps (4) to (8).

[0040] Furthermore, the said step S4 includes:

[0041] S401: Use the Bi-GRU neural network to capture the features of the input training data from two different directions, namely the forward and reverse directions;

[0042] S402: Use the particle swarm optimization algorithm to select the optimal parameters of the Bi-GRU neural network model, specifically including:

[0043] Step (1), initialize each parameter, set the number of population particles to 3, and set the maximum number of iterations to 20;

[0044] Step (2), calculate the fitness value of each particle. The fitness function is: In the formula, m and n respectively represent the lengths of the training set and the test set, k represents the dimension of the data, and respectively represent the predicted value and the actual value in the training set, and respectively represent the predicted value and the actual value in the test set;

[0045] Step (3), calculate the particle velocity and position;

[0046] Step (4), determine whether the optimal parameters have been reached. If they have been reached, output the optimal parameters. Otherwise, repeat steps (2) to (4);

[0047] S403: Obtain the geological condition data and meteorological data of the monitoring area from public websites as the data set, divide it into training set samples and validation set samples according to the ratio of 9:1. After completing the parameter optimization of the Bi-GRU neural network model, input the training set samples into the PSO-Bi-GRU model for model training. After the training is completed, input the validation set samples into the PSO-Bi-GRU model for landslide risk prediction.

[0048] A landslide risk prediction method based on the classification of the gentle and steep slopes of the slope body provided by the present invention first classifies the slope of the hidden danger area based on the digital elevation model (DEM) into a gentle dip slope body area and a steep dip slope body area; then uses the genetic algorithm (GA) to optimize the parameters of the BP neural network to construct a landslide risk prediction model for the gentle dip slope body, that is, the GA-BP model, and uses the particle swarm optimization algorithm (PSO) to obtain the optimal parameters of the bidirectional gated recurrent unit (Bi-GRU) neural network model to construct a landslide risk prediction model for the steep dip slope body, that is, the PSO-Bi-GRU model; finally, collects the monitoring data of the hidden danger area, inputs it into the landslide risk prediction model, and obtains the risk prediction result. The solution of the present invention constructs different landslide risk prediction models for different types of terrain slopes, improving the accuracy of landslide risk prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0050] Figure 1 is a flowchart of a landslide risk prediction method based on the classification of the gentle and steep slopes of the slope body according to an embodiment of the present invention;

[0051] Figure 2 is a landslide risk prediction model for the gentle dip slope body provided by an embodiment of the present invention;

[0052] Figure 3 is a graph showing the change of the best fitness value during the iteration process when GA optimizes the parameters of BP according to an embodiment of the present invention;

[0053] Figure 4 is a landslide risk prediction model for the steep dip slope body provided by an embodiment of the present invention;

[0054] Figure 5 is a visualization graph of the landslide risk prediction result provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0056] The present invention proposes a landslide risk prediction method based on the classification of the gentle and steep degrees of slopes, which can improve the accuracy of landslide risk prediction. First, based on the classification idea, the mountain bodies in the monitoring area are divided into two categories according to the slope. The first category is the gentle dip slope bodies developed in the nearly horizontal strata, and the second category is the steep dip slope bodies with larger dips. Different models are constructed for slope bodies with different gentle and steep degrees to predict the landslide risk respectively. For the first category of gentle dip slope body areas, the overall slope is small, and the present invention uses the BP neural network optimized by the Genetic Algorithm (GA) to predict the landslide risk in this area. For the second category of steep dip slope body areas, the overall slope is large, and the present invention uses the neural network of the Bidirectional-Gated Recurrent Unit (Bi-GRU) optimized by the Particle Swarm Optimization (PSO) to predict the landslide risk in this area. As Figure 1 shown, the specific implementation steps of the present invention are as follows:

[0057] 1. Given the landslide hidden danger area

[0058] According to the geological condition factors, select the area with a relatively large landslide risk as the hidden danger area for monitoring. The present invention will subsequently conduct risk prediction for the landslide hidden danger area.

[0059] 2. Classify the slopes of the hidden danger area based on DEM

[0060] According to the different gentle and steep degrees in the monitoring area, the slope bodies are divided into gentle dip slope body areas and steep dip slope body areas. The difference between these two types of areas lies in the slope of the slope. The gentle dip slope body areas are distributed between 10° and 30°, while the steep dip slope body areas are all greater than 30°.

[0061] Using the Digital Elevation Model (DEM) to calculate the slope is a common terrain analysis method, which can help us obtain the inclination degree of the ground surface. The slope is the change rate of a certain point on the ground surface in the horizontal direction, usually expressed in degrees. The following are the basic steps for calculating the slope based on DEM:

[0062] a. Obtain DEM data: Obtain high-quality digital elevation model data from a publicly available professional website system.

[0063] b. Select the calculation resolution: Select an appropriate calculation resolution according to the application requirements. A smaller resolution will provide more detailed terrain information, but may require more computing resources. A larger resolution will result in information loss.

[0064] c. Grid meshing: Divide the ground surface into regular grid cells. Each grid cell usually contains a DEM data point, that is, an elevation value.

[0065] d. Calculate elevation differences: For each grid cell, calculate the elevation value at the center point. Then, calculate the elevation differences between other points within the cell and the center point. These elevation differences are the basis for slope calculation.

[0066] e. Calculate horizontal distances: For each calculation cell, calculate the horizontal distances between the center point and other points. This is to obtain a horizontal distance for use in slope calculation.

[0067] f. Calculate slope angles: Use the following formula to calculate the slope angle for each grid cell: Slope angle = arctan(elevation difference / horizontal distance)

[0068] g. Convert to degrees: Usually, the calculated slope angles are in radians. Convert these radian values to degrees by multiplying the radian values by 180 / π.

[0069] h. Analyze and visualize: Analyze and visualize the calculated slope data to intuitively display the slope distribution of the terrain.

[0070] After processing the digital elevation data of the monitoring area, the slope of the slope body is obtained. The logical regression algorithm is used to automatically classify the slope body types in the monitoring area. Usually, the Sigmoid function is used as the main calculation formula for the logical regression binary classification model, which satisfies the following formula:

[0071]

[0072] In the formula, x is the input slope of the slope body, and g(x) is the classification result, taking values of 0 or 1.

[0073] 3. Learning of the landslide risk prediction model for gently inclined slope bodies

[0074] For the special terrain structure and complex climate conditions in mountainous areas, traditional risk prediction methods are difficult to solve problems such as patterns, trends, and relationships hidden in multi-source and massive landslide data, resulting in low accuracy of risk prediction results in gently inclined slope body areas. The present invention proposes a new method for determining weights, combining the BP neural network model and GA to obtain the optimal number of hidden nodes and optimize the neural network model, as Figure 2 shown.

[0075] Obtain the geological condition data and meteorological data of the monitoring area from public websites and input them into the BP neural network model. Determine the weights of each parameter in the model through the connection matrix between the input layer and the hidden layer. The calculation formula is as follows:

[0076]

[0077] In the formula, v jl is an element of the connection matrix, m represents the number of input sample indexes, and k is the number of hidden nodes.

[0078] The present invention uses GA to optimize the parameters of the BP neural network model. The genetic algorithm is an optimization algorithm that simulates natural selection and gene evolution in biology. It usually includes three genetic operators: selection, crossover, and mutation. GA is a swarm intelligence optimization algorithm with good global search ability. It can optimize the parameter combinations that are difficult to select in BP, avoid the interference of subjective factors, and automatically search for the best parameter combination. The specific steps are as follows:

[0079] (1) Determine the coding range and length. The coding range is [-1, 1], and the length is 61;

[0080] (2) Determine the coding method. The genetic algorithm is random, and the real number coding method is adopted. The coding value is a random number within the coding range;

[0081] (3) Determine the fitness function: In the formula, k is a coefficient, n is the number of samples in the training set, t i is the true value of the i-th sample, and y i is the predicted value of the i-th sample;

[0082] (4) Select individuals using the tournament method according to the fitness;

[0083] (5) The selected individuals generate offspring through crossover and mutation operations;

[0084] (6) Calculate the fitness of the offspring;

[0085] (7) Combine the offspring and the individuals of the previous generation, and sort them according to the fitness;

[0086] (8) If the number of iterations is greater than 20 or the best fitness in 10 iterations remains unchanged, end the algorithm to obtain the number of hidden nodes and the best fitness value. Otherwise, repeat steps (4) to (8).

[0087] In the experiment of the present invention, the displacement data of the Baoxing landslide monitoring point in Ya'an is adopted, and the satellite remote sensing image data on December 1, 2020 is selected. The above-introduced method is used to predict the landslide risk in this interval.

[0088] In GA, the number of population individuals is set to 3, the maximum number of iterations is 20, and the optimization range of the number of hidden nodes k is [0.1, 20]. The final optimization results are shown in Table 1.

[0089] Table 1 GA Parameter Optimization Results for BP

[0090]

[0091] The variation of the best fitness value during the iteration process is as Figure 3 shown below.

[0092] The result obtained by the above optimization method is used as the number of hidden nodes of the BP neural network model to ensure the accuracy of the landslide risk prediction result and optimize the end-point effect.

[0093] 4. Learning of the landslide risk prediction model for steep dip slope bodies

[0094] The present invention uses a Bi-GRU neural network model to predict the landslide risk of steep dip slope bodies. It has the advantages of a deep depth, wide width, and recyclable iterative modeling of the deep learning network layer, and can predict the non-linear relationship between environmental factors.

[0095] The GRU network includes two gate structures, an update gate and a reset gate. The update gate determines which important historical information needs to be retained, and the reset gate determines which unimportant historical information needs to be forgotten. The defect of the traditional GRU neural network is that it can only read the periodic displacement information of the input from a single direction. However, in actual situations, the feature representations of the front and rear regions need to be fully considered, that is, the regional features in both positive and negative directions need to be taken into account. To overcome the limitations of the traditional GRU neural network, the present invention uses a Bi-GRU neural network to capture the features of the input training data from two different directions, positive and negative.

[0096] To obtain more accurate weight values and to avoid the problem of overfitting, basic parameters need to be set when establishing the Bi-GRU model. The traditional method is manual empirical parameter tuning, which lacks scientificity and accuracy. To scientifically select the optimal parameters of Bi-GRU, the present invention uses a particle swarm optimization algorithm to set the Bi-GRU parameters. The specific steps are as follows:

[0097] (1) Initialize each parameter, set the number of population particles to 3, and the maximum number of iterations to 20;

[0098] (2) Calculate the fitness value of each particle. The fitness function is: In the formula, m and n respectively represent the lengths of the training set and the test set, k represents the dimension of the data, and respectively represent the predicted value and the actual value in the training set, and respectively represent the predicted value and the actual value in the test set;

[0099] (3) Calculate the particle velocity and position;

[0100] (4) Determine whether the optimal parameters have been reached. If so, output the optimal parameters; otherwise, repeat steps (2) to (4).

[0101] Obtain the geological condition data and meteorological data of the monitoring area from public websites as a data set, and divide it into training set samples and validation set samples according to a ratio of 9:1. After the parameter optimization of Bi-GRU by PSO is completed, input the training set samples into PSO-Bi-GRU for model training. After the training is completed, input the validation set samples into the PSO-Bi-GRU model for landslide risk prediction. The PSO-Bi-GRU model is as Figure 4 shown.

[0102] 5. Collection of Monitoring Data in Hazard Areas

[0103] Collect the data of the given landslide hazard area. In addition to obtaining the data monitored by various sensors in the monitoring area, it is also necessary to crawl the regional meteorological and hydrological data, regional geological data, regional geological remote sensing data, etc. of the external system.

[0104] 6. Landslide Risk Prediction and Visualization Result Presentation in Hazard Area Slopes

[0105] After obtaining the landslide risk prediction models for gentle dip slope bodies and steep dip slope bodies, input the monitoring data of various sensors collected, as well as the regional meteorological and hydrological data, geological data, and geological remote sensing data of the monitoring area into the trained prediction models to respectively predict the landslide risks of gentle dip slope bodies and steep dip slope bodies, and obtain the result values of the risk prediction. Finally, use a map plugin to perform visualization processing on the result values of the risk prediction. The example results are as Figure 5 shown, where the left side is the area of gentle dip slope bodies and the right side is the area of steep dip slope bodies.

[0106] To obtain the technical effects of the method of the present invention, we carried out a comparative experiment on landslide disaster point risk prediction using the trained prediction model. The experimental scheme selected the Ya'an Baoxing landslide disaster point area, with a total of 300 samples as experimental data, and used traditional methods and the method proposed in the present invention for risk prediction respectively. For the gently inclined slope area, the experimental result data of the present invention show that the landslide risk prediction method using the GA-BP model has a better prediction effect than the landslide risk prediction method using the traditional BP model. The average error rate of the landslide risk prediction result of the GA-BP model is 8.3%, while the average error rate of the landslide risk prediction result of the traditional BP model is 12.1%; for the steeply inclined slope area, the experimental result data of the present invention show that the landslide risk prediction method using the PSO-Bi-GRU model has a better prediction effect than the landslide risk prediction method using the traditional GRU model. The average error rate of the landslide risk prediction result of the PSO-Bi-GRU model is 7.6%, while the average error rate of the landslide risk prediction result of the traditional GRU model is 13.5%. Through a landslide risk prediction method based on the classification of the gentle and steep degrees of slope proposed in the present invention, the accuracy of landslide risk prediction is improved.

[0107] The above are only the preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A landslide risk prediction method based on the classification of slope gentle and steep degrees, characterized in that, the method includes: Step S1: Determine the landslide hidden danger area. According to the geological condition factors of the monitoring area, select the area with greater landslide risk as the hidden danger area for monitoring; Step S2: Classify the slope of the hidden danger area based on the digital elevation model DEM. By processing the DEM data of the monitoring area, use the logistic regression algorithm to automatically classify the slope of the monitoring area. According to the different gentle and steep degrees of the monitoring area, the slope of the monitoring area is divided into two categories, one is the gentle dip slope area, and the other is the steep dip slope area; Step S3: Learn the landslide risk prediction model for the gentle dip slope. Combine the BP neural network model and the genetic algorithm GA. Use the genetic algorithm GA to optimize the parameters of the BP neural network, obtain the optimal number of hidden nodes, and optimize the neural network model, so as to learn the risk prediction model for the gentle dip slope area, that is, the GA-BP model; Step S4: Learn the landslide risk prediction model for the steep dip slope. Use the particle swarm optimization algorithm PSO to obtain the optimal parameters of the bidirectional gated recurrent unit Bi-GRU neural network model, so as to learn the risk prediction model for the steep dip slope area, that is, the PSO-Bi-GRU model; Step S5: Collect monitoring data in the hidden danger area. Collect the data of this area in the determined landslide hidden danger area, including obtaining the data monitored by various sensors in the monitoring area, as well as the meteorological and hydrological data, geological data and geological remote sensing data of the monitoring area obtained through external systems; Step S6: Predict the landslide risk of the slope in the hidden danger area and visualize the results. Input the data monitored by various sensors collected, as well as the meteorological and hydrological data, geological data and geological remote sensing data of the monitoring area into the trained landslide risk prediction models for the gentle dip slope and the steep dip slope, respectively predict the landslide risks of the gentle dip slope and the steep dip slope, obtain the result values of the risk prediction, and then use the map plugin to visualize the result values of the risk prediction.

2. The method according to claim 1, characterized in that, the step S2 further includes: S201: Obtain DEM data. Obtain high-quality digital elevation model data from a publicly available professional website system; S202: Select the calculation resolution; S203: Grid. Divide the ground surface into regular grid cells, and each grid cell contains multiple DEM data points, that is, multiple elevation values; S204: Calculate the elevation difference. For each grid cell, calculate the elevation value of the center point, and then calculate the elevation difference between other points in the grid cell and the center point; S205: Calculate the horizontal distance. For each grid cell, calculate the horizontal distance between the center point and other points; S206: Calculate the slope angle. Use the following formula to calculate the slope angle from the center point to other points in each grid cell: Slope angle = arctan(elevation difference / horizontal distance); S207: Convert the calculated slope angle to degrees. The calculated slope angle is in radians. By multiplying the radian value by 180 / π, the radian value is converted to degrees. S208: Analyze and visualize the calculated data. According to the calculated slope of the slope body, use the logistic regression algorithm to automatically classify the slope body types in the monitoring area into gentle dip slope body areas and steep dip slope body areas, and perform visualization processing on the calculated slope data to intuitively display the slope distribution of the terrain.

3. The method according to claim 2, characterized in that the slope of the gentle dip slope body is distributed between 10° and 30°, and the slope of the steep dip slope body is greater than 30°.

4. The method according to claim 2, characterized in that the Sigmoid function is used as the main calculation formula of the logistic regression binary classification model, which satisfies the following formula: where x is the input slope of the slope body, and g(x) is the classification result.

5. The method according to claim 1, characterized in that the step S3 further includes: S301: Obtain the geological condition data and meteorological data of the monitoring area from the public website and input them into the BP neural network model. Determine the weights of each parameter in the model through the connection matrix between the input layer and the hidden layer. The calculation formula is as follows: where v jl is an element of the connection matrix, m represents the number of input sample indicators, and k is the number of hidden nodes; S302: Use GA to optimize the parameters of BP to obtain the optimal number of hidden nodes k and the best fitness value. Specifically, it includes: Step (1), determine the coding range and length. The coding range is [-1, 1], and the length is 61; Step (2), determine the coding method. The real number coding method is adopted, and the coding value is a random number within the coding range; Step (3), determine the fitness function: where k is a coefficient, n is the number of training set samples, and t i is the true value of the i-th sample, and y i is the predicted value of the i-th sample; Step (4), select individuals according to the fitness using the tournament method; Step (5), the selected individuals generate offspring through crossover and mutation operations; Step (6), calculate the fitness of the offspring; Step (7), combine the offspring and the individuals of the previous generation, and sort them according to the fitness; Step (8), if the number of iterations is greater than 20 or the best fitness value remains unchanged after 10 iterations, end the algorithm to obtain the number of hidden nodes and the best fitness value. Otherwise, repeat steps (4) to (8).

6. The method according to claim 1, characterized in that the step S4 further includes: S401: Use the Bi-GRU neural network to capture the features of the input training data from two different directions, forward and backward; S402: Use the particle swarm optimization algorithm to select the optimal parameters of the Bi-GRU neural network model. Specifically, it includes: Step (1), initialize each parameter, set the number of population particles to 3, and the maximum number of iterations to 20; Step (2), calculate the fitness value of each particle, and the fitness function is: where m and n respectively represent the lengths of the training set and the test set, k represents the dimension of the data, and respectively represent the predicted value and the actual value in the training set, and respectively represent the predicted value and the actual value in the test set; Step (3), calculate the particle velocity and position; Step (4), determine whether the optimal parameters have been reached. If they have been reached, output the optimal parameters. Otherwise, repeat steps (2) to (4); S403: Obtain the geological condition data and meteorological data of the monitoring area from the public website as the data set, divide it into training set samples and validation set samples according to a ratio of 9:

1. After completing the parameter optimization of the Bi-GRU neural network model, input the training set samples into the PSO-Bi-GRU model for model training. After training is completed, input the validation set samples into the PSO-Bi-GRU model for landslide risk prediction.

Images