Debris flow danger grade prediction method and system based on GRA-PCA-RBF neural network model

Abstract

The invention discloses a debris flow risk level prediction method and system based on a GRA-PCA-RBF neural network model, and the method comprises the steps: taking a debris flow gully in a debris flow liable region as a research object, collecting and analyzing the data and data of historical debris flow disasters, determining the risk and risk evaluation indexes of the debris flow disasters, and carrying out the prediction of the risk level of the debris flow disasters. Analyzing the correlation degree of the debris flow danger and the evaluation indexes by using a grey correlation degree method, and selecting key evaluation indexes from the correlation degree; carrying out dimension reduction processing on the key evaluation index data by utilizing a principal component analysis method, removing redundant information of the index data, and establishing an evaluation index system; and finally, training a training sample by using an RBF neural network model optimized by a genetic algorithm, establishing a debris flow risk level prediction model, and performing accuracy estimation on the constructed model. According to the invention, a reasonable and effective debris flow danger method and system are established in a complex and hard mountain area, and an efficient working system is provided for disaster prevention and reduction of infrastructure construction in the mountain area.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster prediction, and in particular to a method and system for predicting the hazard level of debris flow based on a GRA-PCA-RBF neural network model. Technical Background

[0002] my country has a complex terrain, with mountainous areas accounting for 70% of the country's land area. It is also in a monsoon climate zone with concentrated rainfall, which greatly increases the possibility of debris flows. At present, debris flows in my country account for about 17% of the total number of geological disasters and are widely distributed throughout the country. Debris flows are characterized by suddenness, uncertainty, and concealment. When a debris flow occurs, it will carry a large amount of mud, sand and stones, flow quickly, have huge energy, and have strong destructive power, posing great risks to the safety of life and property of local residents and infrastructure projects.

[0003] Although a preliminary geological disaster monitoring and early warning system has been formed in China, there is still a lack of research on intelligent monitoring, early warning and automatic decision-making for debris flow disasters. Debris flow prediction is basically based on experience, and the reliability of empirical prediction is not good under some complex terrain and geographical conditions. As national infrastructure construction continues to extend to inland mountainous areas, how to build roads and interchanges more safely and reliably in complex and dangerous mountainous areas faces great practical challenges. Therefore, a reasonable and effective evaluation of the risk of debris flow is of great significance to the disaster prevention and mitigation work of infrastructure construction in mountainous areas.

[0004] GRA (Grey Relation Analysis), also known as grey relational analysis, is a multi-factor statistical analysis method that can measure the relative strength of an item affected by other factors in a grey system; PCA (Principal Components Analysis), also known as principal component analysis, is a data set dimensionality reduction method that reduces the dimension of a data set by converting a data set with many variables into fewer variables that still contain most of the information in the original data set, making the data set easier to explore and visualize, and making it easier and faster for machine learning algorithms to analyze data without processing irrelevant variables; RBF (Radial Basis Function) neural network model is a neural network structure that belongs to the forward neural network type. It can approximate any continuous function with arbitrary precision and is particularly suitable for solving classification problems. Summary of the invention

[0005] In view of the deficiencies pointed out in the above background technology, the present invention provides a method and system for predicting the debris flow hazard level based on the GRA-PCA-RBF neural network model, aiming to evaluate the debris flow hazard more reasonably and effectively.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for predicting the danger level of debris flow based on a GRA-PCA-RBF neural network model, characterized in that the method comprises the following steps:

[0008] S1. Select debris flow gullies in debris flow prone areas as research samples;

[0009] S2: obtaining evaluation indicators and data for risk level prediction based on the research samples, and preprocessing the data;

[0010] S3. Use grey relational analysis (GRA) and principal component analysis (PCA) to analyze and process the data of evaluation indicators and construct an evaluation indicator system;

[0011] S4, using the evaluation index system as the input feature of the neural network model, and randomly constructing test samples and training samples;

[0012] S5, construct RBF neural network model;

[0013] S6. Use genetic algorithm to optimize the RBF neural network model and build a systematic GRA-PCA-RBF neural network debris flow hazard level prediction model;

[0014] S7: Conduct model training and accuracy verification on the constructed GRA-PCA-RBF neural network debris flow hazard level prediction model.

[0015] Furthermore, step S2 specifically includes the following steps:

[0016] S21. Collect relevant debris flow disasters of research samples and analyze their dangers respectively;

[0017] S22. Collect historical data on vegetation changes during debris flow disasters in research samples;

[0018] S23, collecting rainfall data of the study samples through meteorological stations;

[0019] S24. Collect remote sensing data of research samples and process remote sensing images using ArcGIS geographic information system software;

[0020] S25. Comprehensively analyze the collected vegetation change data, rainfall data and remote sensing data, and determine the evaluation index of debris flow hazard;

[0021] S26. Clean the data of the evaluation indicators to ensure the quality of the data;

[0022] S27. Standardize the cleaned data to eliminate the influence of different dimensions on the analysis results.

[0023] Furthermore, in step S25, the evaluation indicators include: the maximum discharge volume of a debris flow (104m 3 ), frequency of debris flow (times / 100 years), basin area (km 2 ), main channel length (km), maximum relative height difference in the basin (km), average gradient of the main channel (-), average slope of the formation area (°), basin cutting density (km / km 2 ), main channel curvature coefficient (-), loose solid material reserves (104m 3 ), length ratio of mud and sand supply section (%), maximum rainfall in 24 hours (mm), annual average rainfall (mm), vegetation coverage (%), reclamation index (%), population density in the basin (person / km 2 ).

[0024] Furthermore, step S3 specifically includes the following steps:

[0025] S31. Use the grey correlation method GRA to analyze the correlation between the risk and evaluation index in the original data of debris flow disasters in the collected research samples, calculate the weight coefficient of each evaluation index to the risk of debris flow, and select key evaluation indicators according to the weight coefficient of each evaluation index. The specific steps are as follows:

[0026] The debris flow hazard is selected as the reference sequence, which is recorded as:

[0027] x 0 (k) = {x 0 (1),x 0 (2),…,x 0 (n)};

[0028] The evaluation index constitutes a comparison series, remember:

[0029] x i (k) = {x i (1),x i (2),…,x i (n)};

[0030] Where i = 1, 2, ..., 16; n is the number of debris flow samples;

[0031] Calculate the correlation coefficient between debris flow hazard and each evaluation index:

[0032]

[0033] Calculate the correlation between the evaluation index and the danger of debris flow:

[0034]

[0035] Calculate the weight coefficient of evaluation index and debris flow hazard:

[0036]

[0037] According to the weight coefficient, the key evaluation index is selected from the evaluation indexes.

[0038] S32, using principal component analysis PCA to perform dimensionality reduction processing on the data of key evaluation indicators, eliminating redundant information, converting the data of key evaluation indicators into principal component data through calculation processing, and obtaining an evaluation indicator system. The specific steps are as follows:

[0039] The j key evaluation indicators form a j-dimensional row vector (x 1 x 2 …x j ), collect i such j-dimensional row vector samples to form a matrix X, where j represents the number of key evaluation indicators, i = 1, 2, ..., that is:

[0040]

[0041] Standardized key indicator data:

[0042] Normalize the matrix X to get X * ,X * The element in is x * ij ,

[0043]

[0044] in, is the mean, s j is the variance;

[0045] Calculate X * The correlation coefficient matrix of

[0046]

[0047] Calculate the eigenvalues ​​and eigenvectors of matrix R:

[0048] The eigenvalue is λ 1 ,λ 2 , …, λ j , the corresponding eigenvector is: (u k1 ,u k2 ,…,u kj ), where k = 1, 2, ..., j;

[0049] Calculate the cumulative contribution rate:

[0050]

[0051] Select the principal components:

[0052] When the cumulative contribution rate of the current m principal components is not less than 85%, the m-dimensional principal component is selected;

[0053] Compute the principal components:

[0054]

[0055] Among them, Y k It is called the kth principal component, and the linear expression of the principal component is:

[0056]

[0057] Furthermore, step S4 specifically includes: randomly dividing the principal component data obtained in step S3 into training samples and test samples according to a ratio of 3:1.

[0058] Further, step S5 includes the following steps:

[0059] S51, constructing an RBF neural network model based on the principal component data obtained above, comprising an input layer, a hidden layer, and an output layer;

[0060] S52, using the principal component data based on the K-means clustering method to obtain the radial basis function center of the hidden layer,

[0061] Further solve the variance and calculate the weights between the hidden layer and the output layer, specifically:

[0062] ① Based on the K-means clustering method, the basis function center C is obtained:

[0063] Network initialization: Use the training data set obtained in step S4 as the cluster center C i ;

[0064] Group the input training data set according to the nearest neighbor rule: according to x p With center C i The Euclidean distance between p Each cluster set υ assigned to the input sample p (p=1, 2, ..., p);

[0065] Re-adjust cluster centers: calculate each cluster set υ p The average value of the training samples in the cluster is used to obtain the new cluster center C i ;

[0066] The cluster center C obtained when the cluster center does not change i is the basis function center of the RBF neural network;

[0067] ② Solve for variance σi :

[0068] The basis function of the RBF neural network is a Gaussian function, and the variance solution is:

[0069] Among them C max is the maximum distance between the selected center points;

[0070] ③Calculate the weights between the hidden layer and the output layer:

[0071] The connection weight calculation formula is:

[0072] Among them, i=1, 2,…, h; p=1, 2,….

[0073] Further, step S6 includes the following steps:

[0074] S61, genetic algorithm optimizes the RBF neural network model parameters, including the number of hidden layer neurons, weights and width of radial basis functions;

[0075] S62, training based on the optimized RBF neural network model, adjusting network parameters such as weight, center and width;

[0076] S63. Adjust model parameters according to error feedback until the error converges.

[0077] Further, step S7 includes the following steps:

[0078] S71. Use the validation set to evaluate the predictive performance of the model and avoid overfitting.

[0079] S72. Apply the model to the test set, evaluate the generalization ability of the model, and calculate the prediction accuracy and error;

[0080] S73. Output the prediction result of debris flow hazard level.

[0081] The present invention further provides a debris flow hazard level prediction system created by the debris flow hazard level prediction method based on the GRA-PCA-RBF neural network model.

[0082] Furthermore, the debris flow hazard level prediction system includes the following modules:

[0083] Data acquisition module: obtain historical data of debris flow disasters and vegetation changes in typical areas, rainfall data from meteorological stations, and remote sensing data;

[0084] Data analysis module: Analyze the obtained debris flow disaster data indicators, obtain the key indicators that affect the danger of debris flow disasters, eliminate redundant information, convert the key evaluation indicator information into main component information, and establish an evaluation indicator system;

[0085] GRA-PCA-RBF neural network prediction model building module: establish a GRA-PCA-RBF neural network prediction model to obtain the results of debris flow disaster hazard level classification;

[0086] Model training and evaluation module: trains and evaluates the prediction model to improve the accuracy of model prediction.

[0087] Compared with the shortcomings and deficiencies of the prior art, the present invention has the following beneficial effects:

[0088] The debris flow hazard level prediction method and system provided by the present invention are obtained through scientific calculation and analysis based on the grey correlation method GRA, the principal component analysis method PCA, and the RBF neural network model. The occurrence of debris flow disasters is affected by many factors, so there are complex and diverse evaluation indicators. Through the grey correlation method GRA calculation, key evaluation indicators with a high degree of correlation with the debris flow hazard can be screened out, and then the evaluation indicator data is further reduced in dimension through the principal component analysis method PCA, redundant data is eliminated, and a large number of key evaluation indicators are converted into the principal components of the comprehensive evaluation system, which reduces the amount of data but keeps the data information unchanged, making prediction and analysis more convenient; and the RBF neural network model optimized by the genetic algorithm is used to carry out a large number of training and testing on the training samples, and the accuracy is estimated, which greatly improves the reliability of the debris flow hazard level prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 It is a schematic diagram of the steps of the method for predicting the danger level of debris flow based on the GRA-PCA-RBF neural network model of the present invention;

[0090] Figure 2 It is a schematic diagram of the contribution rate change of the principal component analysis method in an embodiment of the method for predicting the danger level of debris flow based on the GRA-PCA-RBF neural network model of the present invention;

[0091] Figure 3 It is a schematic diagram of the training accuracy rate results of the debris flow hazard level prediction model in the embodiment of the debris flow hazard level prediction method based on the GRA-PCA-RBF neural network model of the present invention;

[0092] Figure 4 This is the architecture diagram of the debris flow hazard level prediction system based on the GRA-PCA-RBF neural network model of the present invention. DETAILED DESCRIPTION

[0093] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0094] The present invention proposes a method and system for predicting the danger level of debris flow based on the GRA-PCA-RBF neural network model. Figure 1 As shown, the prediction method includes the following steps:

[0095] S1. Select debris flow gullies in debris flow prone areas as research samples;

[0096] S2: Obtain evaluation indicators and data for risk level prediction based on the selected research samples and pre-process the data;

[0097] S3. Use grey correlation method and principal component analysis method to analyze and process the data of evaluation indicators, and build an evaluation indicator system;

[0098] S4, using the obtained evaluation index system as the input feature of the neural network model, and randomly constructing test samples and training samples;

[0099] S5, construct RBF neural network model;

[0100] S6. Use genetic algorithm to optimize the RBF neural network model and build a systematic GRA-PCA-RBF neural network debris flow hazard level prediction model;

[0101] S7: An implementation example of a method for predicting debris flow hazard level by performing model training and accuracy verification on the constructed GRA-PCA-RBF neural network debris flow hazard level prediction model.

[0102] Selection of research samples: By consulting a large number of literature, 37 debris flow single gullies in Beichuan County from the book "Desert Flow Hazard Assessment" which is widely cited by scholars were used as the original research samples.

[0103] Obtain evaluation indicators based on the research sample:

[0104] Collection of historical data on debris flow disasters and vegetation changes: By consulting literature and monographs, local county chronicles, local natural resources, forestry, water conservancy and other departments on debris flow disasters and vegetation changes in the study area, the (possible) maximum outflow of a debris flow (10 4 m 3 )S 1 , Debris flow frequency (times / 100 years) S 2 、Main groove bending coefficient (-) S 9, loose solid matter reserves (10 4 m 3 )S 10 , Vegetation coverage (%) S 14 , reclamation index (%) S 15 , Population density in the basin (person / km 2 )S 16 It is an indicator for evaluating the risk of debris flow;

[0105] Rainfall data collection: Beichuan County in this example has only one meteorological monitoring station, the Beichuan County Meteorological Bureau. However, since wind speed and temperature have little impact on debris flows, rainfall data of Beichuan County is collected through the meteorological station, and the maximum rainfall in 24 hours (mm) S is selected. 12 , Annual average rainfall (mm) S 13 It is an evaluation index of debris flow hazard;

[0106] Remote sensing data collection: Aster, Landsat, Resource 3 and MODIS were used to obtain remote sensing images of debris flow gullies in the sample area of ​​Yunnan Province, and ArcGIS was used to calculate the corresponding evaluation indicators: drainage area (km 2 )S 3 , Main trench length (km) S 4 , Basin maximum relative height difference (km) S 5 、Average gradient of main channel (-)S 6 , Average slope of the hillside in the formation area (°) S 7 , Basin cutting density (km / km 2 )S 8 , Mud and sand supply section length ratio (-) S 11 .

[0107] The grey correlation method GRA is used to analyze the correlation between the debris flow hazard in the original data and the evaluation index:

[0108] 1). Select debris flow hazard as the reference sequence, and record:

[0109] x 0 (k) = {x 0 (1),x 0 (2),…,x 0 (n)};

[0110] The evaluation index constitutes a comparison series, remember:

[0111] x i (k) = {x i (1),x i (2),…,x i (n)};

[0112] Where i = 1, 2, ..., 16; n is the number of debris flow samples;

[0113] 2). Calculate the correlation coefficient between debris flow hazard and each evaluation index:

[0114]

[0115] 3). Calculate the correlation between the evaluation index and the risk of debris flow:

[0116]

[0117] 4). Calculate the weight coefficient of evaluation index and debris flow hazard:

[0118]

[0119] 5) Based on experience, the evaluation indicators with correlation degree ≥ 0.8 and weight coefficient ≥ 0.070 are selected as key indicators;

[0120] Finally, after the grey relational analysis (GRA) method, it was decided to use the maximum outflow volume (10 4 m 3 )S 1 , Basin area (km 2 )S 3 , Main trench length (km) S 4 , Basin maximum relative height difference (km) S 5 、Average gradient of main channel (-)S 6 , Basin cutting density (km / km 2 )S 8 、Main groove bending coefficient (-) S 9 , Mud and sand supply section length ratio (-) S 11 , reclamation index (%) S 15 is the key evaluation indicator.

[0121] The principal component analysis method PCA is used to reduce the dimension of the selected key evaluation index data and eliminate the redundant information of the index data:

[0122] 1) Standardized key indicator data: The selected 9 key evaluation indicators form a 9-dimensional row vector (x 1 x 2 …x 9 ), collect 7 such 9-dimensional row vector samples to form the matrix X, that is:

[0123]

[0124] 2). Standardize the matrix X to get X * ,X * The elements in are:

[0125]

[0126] in, is the mean, s j is the variance;

[0127] 3) Calculate X * The correlation coefficient matrix R is:

[0128] 4). Calculate the eigenvalue and eigenvector: the eigenvalue is λ 1 ,λ 2 , …, λ 9 , the corresponding eigenvector is: (u k1 ,u k2 ,…,u kj ), where k = 1, 2, ..., 9;

[0129] 5). Calculate contribution rate and cumulative contribution rate (such as Figure 2 The cumulative contribution rate is calculated as follows:

[0130]

[0131] 6) Select the principal component: After calculating the cumulative contribution rate and based on the principle that the eigenvalue is greater than 1, the principal component is selected as a 4-dimensional principal component;

[0132] 7). Calculate the principal components:

[0133]

[0134] Among them, Y k It is called the kth principal component, and the linear expression of the principal component is:

[0135]

[0136] The principal component data are randomly divided into training samples and test samples according to a ratio of 3:1.

[0137] The training samples are trained using the RBF neural network model optimized by genetic algorithm, and a GRA-PCA-RBF neural network debris flow hazard level prediction model is established:

[0138] 1). Obtain the basis function center C based on the K-means clustering method:

[0139] ① Network initialization: Use the training data set obtained in step S4 as the cluster center C i ;

[0140] ② Group the input training data set according to the nearest neighbor rule: according to x pWith center C i The Euclidean distance between p Each cluster set υ assigned to the input sample p (p=1, 2, ..., p);

[0141] ③Re-adjust the cluster center: calculate each cluster set υ p The average value of the training samples in the cluster is used to obtain the new cluster center C i ;

[0142] ④ The cluster center C obtained when the cluster center does not change i is the basis function center of the RBF neural network.

[0143] 2). Solve for variance σ i :

[0144] The basis function of the RBF neural network is a Gaussian function, and the variance solution is:

[0145] Among them C max is the maximum distance between the selected center points;

[0146] 3). Calculate the weight between the hidden layer and the output layer. The connection weight calculation formula is:

[0147]

[0148] Wherein, i = 1, 2, ..., h; p = 1, 2, ..., P;

[0149] 4). Genetic algorithm optimizes RBF neural network weights and thresholds;

[0150] 5). The optimized RBF neural network is used to predict the danger of debris flow;

[0151] 6). Output the final training results.

[0152] Finally, if Figure 3 As shown in the figure, the error estimation of the constructed GRA-PCA-RBF neural network debris flow hazard prediction model is carried out:

[0153] 1). Use the validation set to evaluate the predictive performance of the model and avoid overfitting.

[0154] 2) Apply the model to the test set, evaluate the generalization ability of the model, and calculate the prediction accuracy and error;

[0155] 3). Output the prediction results of debris flow hazard level.

[0156] The present invention also provides a debris flow hazard level prediction system based on the GRA-PCA-RBF neural network model. Figure 4 As shown, including:

[0157] Data acquisition module, used to obtain historical data of debris flow disasters in typical areas, rainfall data from meteorological stations, and remote sensing data;

[0158] The data analysis module is used to analyze the obtained debris flow disaster data indicators, obtain key indicators that affect the danger of debris flow disasters, and eliminate redundant information;

[0159] The GRA-PCA-RBF neural network prediction model building module is used to obtain the results of debris flow disaster hazard level classification by establishing a GRA-PCA-RBF neural network prediction model;

[0160] The model training and evaluation module is used to train and evaluate the prediction model to improve the accuracy of model prediction.

[0161] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the danger level of debris flow based on the GRA-PCA-RBF neural network model, characterized by: The method comprises the following steps: S1. Select debris flow gullies in debris flow prone areas as research samples; S2: obtaining evaluation indicators and data for risk level prediction according to the research sample, and preprocessing the data; S3. Analyze and process the data using grey correlation method and principal component analysis method, and construct an evaluation index system; S4, using the evaluation index system as the input feature of the neural network model, and randomly constructing training samples and test samples; S5, construct RBF neural network model; S6, optimizing the RBF neural network model using a genetic algorithm, and constructing a systematic GRA-PCA-RBF neural network debris flow hazard level prediction model; S7. Performing model training and accuracy verification on the GRA-PCA-RBF neural network debris flow hazard level prediction model.

2. The method for predicting the danger level of debris flow according to claim 1, characterized in that: The step S2 comprises the following steps: S21, collecting relevant debris flow disasters of the research samples and analyzing their dangers respectively; S22. Collect historical data on vegetation changes during debris flow disasters in the research sample; S23, collecting rainfall data of the research sample through a meteorological station; S24, collecting remote sensing data of the research sample, and performing remote sensing image processing through ArcGIS geographic information system software; S25, comprehensively analyzing the collected historical data on vegetation changes, the rainfall data and remote sensing data, and determining the evaluation index of debris flow hazard; S26, performing data cleaning on the data of the evaluation index to ensure data quality; S27. Standardize the cleaned data to eliminate the influence of different dimensions on the analysis results.

3. The method for predicting the danger level of debris flow according to claim 1, characterized in that: The step S3 comprises the following steps: S31, using the grey correlation method GRA to analyze the correlation between the risk and the evaluation index in the original data of debris flow disasters of the research sample collected, calculate the weight coefficient of each evaluation index for the risk of debris flow, and select key evaluation indicators according to the obtained weight coefficient of each evaluation index; S32, using principal component analysis PCA to perform dimensionality reduction processing on the data of the key evaluation indicators, eliminating redundant information, converting the key evaluation indicator data into principal component data through calculation processing, and obtaining the evaluation indicator system based on the principal component data.

4. The method for predicting the danger level of debris flow according to claim 3, characterized in that: In step S31, the method for analyzing the correlation between debris flow hazard and evaluation index includes: The debris flow hazard is selected as the reference sequence, which is recorded as: x0(k)={x0(1),x0(2),…,x0(n)}; The evaluation index constitutes a comparison series, remember: x i (k)={x i (1),x i (2),…,x i (n)}; Where, i = 1, 2, ..., 16; n is the number of debris flow samples; Calculate the correlation coefficient between debris flow hazard and each evaluation index: Calculate the correlation between the evaluation index and the danger of debris flow: Calculate the weight coefficient of evaluation index and debris flow hazard: According to the weight coefficient, the key evaluation index is selected.

5. The method for predicting the danger level of debris flow according to claim 3, characterized in that: The step S4 is: randomly dividing the principal component data into the training sample and the test sample according to a ratio of 3:

1.

6. The method for predicting debris flow hazard level based on the GRA-PCA-RBF neural network model according to claim 5, characterized in that: The step S5 comprises the following steps: S51, constructing an RBF neural network model according to the principal component data, comprising an input layer, a hidden layer, and an output layer; S52: Utilize the principal component data based on the K-means clustering method to obtain the radial basis function center of the hidden layer, further solve the variance and calculate the weight between the hidden layer and the output layer.

7. The method for predicting debris flow hazard level based on the GRA-PCA-RBF neural network model according to claim 6, characterized in that: The step S6 comprises the following steps: S61, optimizing the parameters of the RBF neural network model using a genetic algorithm, including the number of neurons in the hidden layer, the weights, and the width of the radial basis function; S62, training based on the optimized RBF neural network model, adjusting network parameters such as weight, center and width; S63. Adjust model parameters according to error feedback until the error converges.

8. The method for predicting debris flow hazard level based on the GRA-PCA-RBF neural network model according to claim 1, characterized in that: The step S7 comprises the following steps: S71. Use the validation set to evaluate the predictive performance of the model and avoid overfitting. S72. Apply the model to the test set, evaluate the generalization ability of the model, and calculate the prediction accuracy and error. S73. Output the prediction result of debris flow hazard level.

9. A debris flow hazard level prediction system created by the debris flow hazard level prediction method according to any one of claims 1 to 8.

10. The debris flow hazard level prediction system according to claim 9, characterized in that: Includes the following modules: Data acquisition module: used to obtain historical data on debris flow disasters and vegetation changes in areas prone to debris flow, rainfall data and remote sensing data from meteorological stations, and determine evaluation indicators; Data analysis module: Analyze the evaluation index of the obtained historical data of debris flow disaster, obtain key evaluation indexes that affect the danger of debris flow disaster, remove redundant information, convert the key evaluation index information into principal component information, and establish an evaluation index system; GRA-PCA-RBF neural network prediction model building module: establish a GRA-PCA-RBF neural network prediction model to obtain the results of debris flow disaster hazard level classification; Model training and evaluation module: trains and evaluates the prediction model to improve the accuracy of model prediction.

Images