A method and system for landslide failure prediction

Abstract

This application provides a landslide failure prediction method and system. Because this application can perform multi-dimensional feature extraction on the landslide location description data matrix, obtaining a large amount of information such as key indicator local landslide features and key indicator all landslide features, this application can further extract features from the reference theme results corresponding to the landslide location description data based on the key indicator local landslide features and key indicator all landslide features to obtain important features. Furthermore, by combining the key indicator local landslide features, key indicator all landslide features, and important features, a target theme feature with a deeper content representation between the landslide location description data and the reference theme results can be obtained. Therefore, based on the target theme feature, the landslide failure prediction result of the landslide location description data can be accurately predicted.

Description

Technical Field

[0001] This application relates to the field of data simulation and processing technology, and more specifically, to a landslide failure simulation method and system. Background Technology

[0002] A landslide is a natural phenomenon in which soil or rock masses on a slope slide downhill, either as a whole or in parts, under the influence of gravity, due to factors such as river erosion, groundwater activity, rainwater soaking, earthquakes, and artificial slope cutting. The moving rock (soil) mass is called a displaced body or sliding body, while the underlying rock (soil) mass that has not moved is called a sliding bed.

[0003] Currently, landslide simulations are needed in areas where landslides may occur to predict their destructive power and take preventative measures in advance to minimize losses. Therefore, it is necessary to predict the hazards caused by landslides in advance. However, current landslide damage simulations are very inaccurate because improving the accuracy of these simulations is a technical problem that is difficult to solve at present. Summary of the Invention

[0004] To address the technical problems existing in related technologies, this application provides a landslide failure prediction method and system.

[0005] The first dimension provides a method for predicting landslide failure, including:

[0006] Obtain the landslide location description data to be processed and the reference topic results corresponding to the landslide location description data, and cluster the landslide location description data to obtain at least one landslide location description data matrix corresponding to the landslide location description data;

[0007] Multi-dimensional feature extraction is performed on the landslide location description data matrix to obtain the key indicators of local landslide features for each landslide location description data and the key indicators of all landslide features for the landslide location description data.

[0008] Based on the key indicators of local landslide characteristics and all landslide characteristics, feature extraction is performed on the reference theme results to obtain important features. The important features characterize the matching relationship between the landslide location description data and the reference theme results.

[0009] By integrating the key indicators of local landslide characteristics, all landslide characteristics, and important characteristics, the target theme characteristics of the landslide location description data are obtained.

[0010] Based on the target theme characteristics and the reference theme results, landslide failure simulation results corresponding to the landslide location description data are generated.

[0011] In this application, the step of performing multi-dimensional feature extraction on the landslide location description data matrix to obtain the key indicators of local landslide features for each landslide location description data and the key indicators of all landslide features for the landslide location description data includes:

[0012] The regional location data of each landslide location description data matrix is ​​parsed from the landslide location description data, and the regional location data is extracted using a landslide location description data parsing network to obtain the regional location features of each landslide location description data matrix.

[0013] The landslide location description data parsing network is used to extract features from the landslide location description data matrix to obtain the initial key index local landslide features and the initial key index all landslide features of the landslide location description data for each landslide location description data matrix.

[0014] The regional location features are integrated with the initial key indicator local landslide features to obtain the key indicator local landslide features of each landslide location description data matrix;

[0015] By integrating the regional location features and the initial key indicators of all landslide features, the key indicators of all landslide features of the landslide location description data are obtained.

[0016] In this application, before extracting features from the landslide location description data matrix using the landslide location description data parsing network, the method further includes:

[0017] Obtain a training dataset, which includes multiple example landslide location description data and abnormal landslide examples corresponding to the example landslide location description data;

[0018] Cluster the exemplary landslide location description data and the abnormal landslide examples respectively to obtain at least one landslide location description data region corresponding to the exemplary landslide location description data and at least one topic set corresponding to the abnormal landslide;

[0019] A specified landslide location description data parsing network is used to extract features from the landslide location description data region and the topic set, respectively, to obtain key indicator features of at least one key indicator character corresponding to the example landslide location description data, and topic features of at least one topic character corresponding to the abnormal landslide example.

[0020] The key indicator features and the topic features are distinguished by at least one distinguishing character to obtain the feature distinction result corresponding to each distinguishing character. The feature distinction result represents the distinguishability between the key indicator features and the topic features.

[0021] Based on the feature identification results, key indicator features, and theme features, the parsing network for the specified landslide location description data is converged to obtain the landslide location description data parsing network.

[0022] In this application, the step of performing feature discrimination on at least one distinguishing character to obtain the feature discrimination result corresponding to each distinguishing character, based on the key indicator features and the topic features, includes:

[0023] Select the target key indicator character corresponding to each identification character from the key indicator characters, and determine the target topic character corresponding to each identification character from the topic characters;

[0024] Select the target key indicator feature corresponding to the target key indicator character from the key indicator features, and extract the target theme feature corresponding to the target theme character from the theme features;

[0025] The target key indicator features and the target topic features are identified to obtain the feature identification result corresponding to each identified character.

[0026] In this application, the step of distinguishing the target key indicator features and the target topic features to obtain the feature discrimination result corresponding to each distinguished character includes:

[0027] At least one target character is identified from the identified characters; candidate key indicator features corresponding to the target character are selected from the target key indicator features, and candidate topic features corresponding to the target character are extracted from the target topic features;

[0028] The candidate key indicator features and the candidate topic features are used to perform feature discrimination on the target identification character to obtain the target feature discrimination result corresponding to the target identification character;

[0029] Return to the step of identifying the target character among the identified characters, until every identified character is the target character, and obtain the feature identification result corresponding to each identified character.

[0030] In this application, the candidate key indicator features include all abnormal features corresponding to the landslide location description data of the example, and the candidate topic features include all topic features corresponding to the abnormal landslide examples; the step of performing feature discrimination on the target identification character using the candidate key indicator features and the candidate topic features to obtain the target feature discrimination result corresponding to the target identification character includes:

[0031] When the target identification character is all identification characters, the similarity coefficient between the example landslide location description data and the abnormal landslide example is calculated based on all abnormal features and all topic features, where all identification characters are the identification characters between the example landslide location description data and the abnormal landslide example.

[0032] The similarity coefficients of the examples are integrated to generate all landslide feature identification results corresponding to all the identification characters, and the all landslide feature identification results are used as the target feature identification results.

[0033] In this application, the candidate key indicator features further include regional abnormal features corresponding to the landslide location description data area, and the candidate topic features further include set features corresponding to the topic set; the step of performing feature discrimination on the target identification character using the candidate key indicator features and the candidate topic features to obtain the target feature discrimination result corresponding to the target identification character includes:

[0034] When the target identification character is a region identification character, a target landslide location description data region is selected from the landslide location description data region, and a target topic set is selected from the topic set. The region identification character is the identification character between the target landslide location description data region and the target topic set.

[0035] Based on the abnormal characteristics of the region and the characteristics of the set, calculate the confidence coefficient between the target landslide location description data region and the target topic set;

[0036] Based on the confidence coefficient, the regional feature recognition result corresponding to the target character is determined, and the regional feature recognition result is used as the target feature recognition result.

[0037] In this application, determining the region feature discrimination result corresponding to the target character based on the confidence coefficient includes:

[0038] Based on the magnitude of the confidence coefficient, the regional confidence coefficient corresponding to the target landslide location description data region is selected from the confidence coefficients. The regional confidence coefficient represents the confidence coefficient between the target landslide location description data region and the candidate topic set corresponding to the target landslide location description data region in the target topic set.

[0039] Based on the magnitude of the confidence coefficient, the set confidence coefficient corresponding to the target topic set is parsed from the confidence coefficient. The set confidence coefficient represents the confidence coefficient between the target topic set and the candidate landslide location description data area corresponding to the target topic set in the target landslide location description data area.

[0040] The region confidence coefficient and the set confidence coefficient are integrated to obtain the integrated confidence coefficient, and the integrated confidence coefficient is used as the region feature recognition result corresponding to the region recognition character.

[0041] In this application, selecting a target landslide location description data region from the landslide location description data region includes: clustering the abnormal features of the region to obtain a region category corresponding to each abnormal region, and determining the target abnormal region corresponding to the region category in the landslide location description data region; selecting a target topic set from the topic set includes: clustering the set features to obtain a set category corresponding to each topic set, and determining the target topic set in the topic set.

[0042] In this application, calculating the confidence coefficient between the target landslide location description data region and the target topic set based on the abnormal characteristics of the region and the set characteristics includes:

[0043] The abnormal features of the same area category are integrated to obtain the integrated area features corresponding to the target landslide location description data area. The integrated area features and the abnormal features of the area are then integrated to obtain the target area abnormal features corresponding to the target landslide location description data area.

[0044] The set features corresponding to the same unit category are integrated to obtain the integrated theme features corresponding to the target theme set. The integrated theme features and the unit theme features corresponding to the target theme set are then integrated to obtain the target theme features corresponding to the target theme set.

[0045] Based on the abnormal characteristics of the target area and the characteristics of the target set, the confidence coefficient between the target landslide location description data area and the target subject set is calculated.

[0046] In this application, the target key indicator features include all abnormal features corresponding to the landslide location description data of the example, and the target theme features include all theme features corresponding to the abnormal landslide examples; the step of performing feature discrimination on the target identification character using the candidate key indicator features and the candidate theme features to obtain the target feature discrimination result corresponding to the target identification character includes:

[0047] When the target identification character is a potential identification character, the potential landslide hazard data corresponding to the example landslide location description data is obtained, and the potential landslide hazard data is feature extracted to obtain potential landslide hazard features. The potential identification character is the identification character between the potential landslide hazard data, the example landslide location description data, and the abnormal landslide example.

[0048] The potential landslide hazard characteristics, all abnormal characteristics, and all theme characteristics are used to identify the potential distinguishing characters to obtain the potential feature identification results corresponding to the potential distinguishing characters, and the potential feature identification results are used as the target feature identification results.

[0049] In this application, the step of performing feature identification on the potential landslide hazard characteristics, all abnormal characteristics, and all thematic characteristics on the potential identification character to obtain the potential feature identification result corresponding to the potential identification character includes:

[0050] Based on the characteristics of the potential landslide hazards and all the abnormal characteristics, calculate the degree of abnormal correlation between the potential landslide hazard data and the example landslide location description data;

[0051] Based on the characteristics of the potential landslide hazards and all thematic characteristics, the thematic correlation degree between the potential landslide hazard data and the abnormal landslide themes is calculated;

[0052] Based on the abnormal correlation degree and the topic correlation degree, the potential feature identification result corresponding to the potential identification character is generated.

[0053] In this application, the distinguishing characters include all distinguishing characters and region distinguishing characters; the step of converging the landslide location description data parsing network based on the feature identification results, key indicator features, and theme features to obtain the landslide location description data parsing network includes:

[0054] Select all landslide feature identification results corresponding to all identification characters and regional feature identification results corresponding to the regional identification characters from the feature identification results respectively;

[0055] Based on all the landslide feature identification results, the example inference performance evaluation between the example landslide location description data and the abnormal landslide examples is calculated. The example inference performance evaluation characterizes the shared inference performance evaluation between the example landslide location description data and the abnormal landslide examples.

[0056] Based on the regional feature identification results, as well as the regional abnormal features and set features corresponding to the regional identification characters, the regional matching inference performance evaluation corresponding to the landslide location description data region is determined. The regional matching inference performance evaluation characterizes the matching inference performance evaluation between the landslide location description data region and the theme set.

[0057] Based on the key indicator features and theme features, the landslide location description data of the example is predicted to obtain the theme prediction probability, and the theme prediction probability is projected to obtain the prediction inference performance evaluation data set corresponding to the landslide location description data of the example.

[0058] Based on the example extrapolation performance evaluation, the regional matching extrapolation performance evaluation, and the prediction extrapolation performance evaluation data set, the specified landslide location description data parsing network is converged to obtain the landslide location description data parsing network.

[0059] In this application, the step of determining the regional matching and inference performance evaluation corresponding to the landslide location description data region based on the regional feature identification results, as well as the regional abnormal features and set features corresponding to the regional identification characters, includes:

[0060] Based on the region feature discrimination results, determine the region training label corresponding to the region discrimination character;

[0061] Based on the abnormal characteristics and set characteristics of the region, calculate the candidate confidence coefficient between the landslide location description data region and the topic set;

[0062] Calculate the target inference performance evaluation between the region training label and the candidate confidence coefficient, and use the target inference performance evaluation as the region matching inference performance evaluation corresponding to the landslide location description data region.

[0063] In this application, the regional feature identification result includes the confidence coefficient between the target landslide location description data region of the example landslide location description data and the target topic set of abnormal landslide examples; the step of determining the regional training label corresponding to the regional identification character based on the regional feature identification result includes:

[0064] Select a reference abnormal region from the abnormal region, and select a reference topic set from the topic set;

[0065] The reference abnormal region and the reference topic set are combined to obtain at least one reference data set, the reference data set including at least one target reference abnormal region and at least one target reference topic set corresponding to the target reference abnormal region;

[0066] The target confidence coefficient corresponding to the reference data set is selected from the regional feature identification results, and the target confidence coefficient is integrated to obtain the target integrated confidence coefficient.

[0067] Based on the target integration credibility coefficient, determine the region training label corresponding to the region identification character; the step of calculating the candidate credibility coefficient between the landslide location description data region and the topic set based on the region abnormality features and set features includes: extracting the current landslide location description data region and the current topic set from the target data set, and calculating the candidate credibility coefficient between the current landslide location description data region and the current topic set based on the region abnormality features corresponding to the current abnormal region and the set features corresponding to the current topic set.

[0068] In this application, the distinguishing character also includes a potential distinguishing character; the step of converging the landslide location description data parsing network based on the example inference performance evaluation, the region matching inference performance evaluation, and the prediction inference performance evaluation data set to obtain the landslide location description data parsing network includes:

[0069] The potential feature identification results corresponding to the potential identification characters are selected from the feature identification results. The potential feature identification results include the abnormal correlation between the potential landslide hazard data and the example landslide location description data corresponding to the potential identification characters, and the theme correlation between the potential landslide hazard data and the abnormal landslide theme.

[0070] Based on the latent feature discrimination results, the latent training labels corresponding to the latent discriminant characters are determined;

[0071] Based on the potential training labels, the potential landslide hazard characteristics corresponding to the potential landslide hazard data, the key indicator characteristics, and the theme characteristics, the potential inference performance evaluation corresponding to the example landslide location description data is calculated. The potential inference performance evaluation characterizes the correlation inference performance evaluation between the potential landslide hazard data, the example landslide location description data, and the abnormal landslide examples.

[0072] Based on the datasets of the potential extrapolation performance evaluation, the example extrapolation performance evaluation, the regional matching extrapolation performance evaluation, and the prediction extrapolation performance evaluation, the parsing network for the specified landslide location description data is converged to obtain the parsing network for the landslide location description data.

[0073] In this application, the key features obtained by extracting features from the reference topic results based on the key indicators of local landslide characteristics and all key indicators of landslide characteristics include:

[0074] Feature extraction is performed on the reference topic results to obtain initial reference topic features;

[0075] The key features are obtained by integrating the local landslide characteristics of the key indicators, the total landslide characteristics of the key indicators, and the initial reference theme characteristics.

[0076] In this application, generating landslide failure projection results corresponding to the landslide location description data based on the target theme features and the reference theme results includes:

[0077] Based on the target theme characteristics, the landslide location description data is used to predict the theme, thereby obtaining the target theme;

[0078] The reference topic results and target topics are integrated to obtain an integrated topic, and the integrated topic is used as the reference topic result. The process of extracting features from the reference topic results based on the local landslide features and all landslide features of the key indicators is repeated until the prediction end requirement is met, resulting in multiple target topics.

[0079] The multiple target themes are integrated to generate the landslide failure simulation results.

[0080] The second dimension provides a landslide failure simulation system, including a processor and a memory that communicate with each other, wherein the processor is used to read a computer program from the memory and execute it to implement the above-described method.

[0081] The landslide failure prediction method and system provided in this application involve acquiring landslide location description data and corresponding reference theme results, clustering the landslide location description data to obtain at least one landslide location description data matrix, extracting multi-dimensional features from the landslide location description data matrix to obtain key indicators (local landslide features and all landslide features) for each landslide location description data, extracting features from the reference theme results based on the key indicators (local landslide features and all landslide features), and obtaining important features that characterize the matching relationship between the landslide location description data and the reference theme results. The key indicators (local landslide features, all landslide features, and important features) are then integrated to obtain the target theme features of the landslide location description data. Based on the target theme features and the reference theme results, a landslide failure prediction result corresponding to the landslide location description data is generated. Because this application can perform multi-dimensional feature extraction on the landslide location description data matrix, obtaining a large amount of information such as key indicator local landslide features and key indicator all landslide features, this application can further extract features from the reference theme results corresponding to the landslide location description data based on the key indicator local landslide features and key indicator all landslide features to obtain important features. Then, by combining the key indicator local landslide features, key indicator all landslide features, and important features, the application can obtain target theme features that have a deeper content representation between the landslide location description data and the reference theme results. Thus, based on the target theme features, the application can accurately predict the landslide failure prediction results of the landslide location description data. Attached Figure Description

[0082] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the area. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0083] Figure 1 A flowchart illustrating a landslide failure simulation method provided in this application embodiment. Detailed Implementation

[0084] To better understand the above technical solutions, the technical solutions of this application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this application and the specific features in the embodiments are detailed descriptions of the technical solutions of this application, rather than limitations on the technical solutions of this application. In the absence of conflict, the embodiments of this application and the technical features in the embodiments can be combined with each other.

[0085] Please see Figure 1 This paper presents a landslide failure simulation method, which may include the technical solutions described in steps S101-S105.

[0086] S101. Obtain the landslide location description data to be processed and the reference topic results corresponding to the landslide location description data, and cluster the landslide location description data to obtain at least one landslide location description data matrix corresponding to the landslide location description data.

[0087] The reference topic results may refer to the topics used to infer landslide failure results that are used to generate potential landslide location description data.

[0088] Regarding step S101, the method for "clustering the landslide location description data to obtain at least one landslide location description data matrix corresponding to the landslide location description data" can be as follows: obtain the abnormal data scale of the landslide location description data; determine the target segmentation region location of the landslide location description data based on the abnormal data scale; segment the landslide location description data according to the target segmentation region location to obtain at least one landslide location description data matrix.

[0089] The step "clustering the landslide location description data to obtain at least one landslide location description data matrix corresponding to the landslide location description data" can also be done in other ways. For details, please refer to the method "segmenting the example landslide location description data to obtain at least one landslide location description data region corresponding to the example landslide location description data" below, which will not be elaborated here.

[0090] S102. Perform multi-dimensional feature extraction on the landslide location description data matrix to obtain the key indicators of local landslide features and the key indicators of all landslide features for each landslide location description data.

[0091] The so-called key indicator "overall landslide characteristics" can refer to the features that characterize the overall content of the landslide location description data. The so-called key indicator "local landslide characteristics" can refer to the features that characterize the content within the landslide location description data matrix.

[0092] After obtaining the landslide location description data matrix, this application can perform multi-dimensional feature extraction on the landslide location description data matrix. As an example, the method for the step "performing multi-dimensional feature extraction on the landslide location description data matrix to obtain the key indicators of local landslide features and all landslide features of each landslide location description data" can be as shown in steps S1021 to S1024:

[0093] S1021. Parse the regional location data of each landslide location description data matrix from the landslide location description data, and use the landslide location description data parsing network to extract features from the regional location data to obtain the regional location features of each landslide location description data matrix.

[0094] Among them, the regional location data can be relevant information about the regional location in the landslide location description data matrix.

[0095] Regarding step S1021, the method for "parse out the regional location data of each landslide location description data matrix from the landslide location description data" can be as follows: parse out important feature points from the landslide location description data matrix, and query the target feature points that match the important feature points in the landslide location description data; determine the target regional location data of the target feature points in the landslide location description data, and use the target regional location data as the regional location data of the landslide location description data matrix.

[0096] Alternatively, the step "parse out the regional location data of each landslide location description data matrix from the landslide location description data" can be performed as follows: extract features from both the landslide location description data matrix and the landslide location description data to obtain the block features corresponding to the landslide location description data matrix and the features of the landslide location description data; calculate the similarity coefficient between the block features and the features of the landslide location description data; based on the similarity coefficient, determine the target feature corresponding to each block feature in the features of the landslide location description data; determine the candidate regional location data corresponding to the target feature in the landslide location description data, and use the candidate regional location data as the regional location data of the landslide location description data matrix.

[0097] Regarding step S1021, the method for "using a landslide location description data parsing network to extract features from regional location data and obtain regional location features for each landslide location description data matrix" can be as follows: using the target compression unit of the landslide location description data parsing network to extract features from regional location data and obtain regional location features for each landslide location description data matrix.

[0098] As an example, the initial feature extraction unit of the target compression unit can be used to extract features from the regional location data to obtain the regional location features of each landslide location description data matrix.

[0099] S1022. The landslide location description data parsing network is used to extract features from the landslide location description data matrix to obtain the initial key indicators of local landslide features and the initial key indicators of all landslide features for each landslide location description data matrix.

[0100] Regarding step S1022, the exemplary approach of "using a landslide location description data parsing network to extract features from the landslide location description data matrix to obtain the initial key indicator local landslide features and the initial key indicator all landslide features of the landslide location description data matrix" can be as follows: obtain all target indicator information corresponding to the landslide location description data; use the initial feature extraction unit of the target compression unit of the landslide location description data parsing network to extract features from the landslide location description data matrix and the target all indicator information to obtain the initial key indicator local landslide features and the initial key indicator all landslide features of the target all indicator information, and use the initial key indicator all landslide features of the target all indicator information as the initial key indicator all landslide features corresponding to the target all indicator information.

[0101] Among them, the target information can be the key indicators of all landslide features generated by the landslide location description data parsing network.

[0102] S1023. Integrate the regional location features with the initial key indicator local landslide features to obtain the key indicator local landslide features of each landslide location description data matrix.

[0103] S1024. Integrate the regional location features and initial key indicators of all landslide features to obtain the key indicators of all landslide features of landslide location description data.

[0104] Regarding step S1024, the method for "integrating the regional location features and all landslide features of the initial key indicators to obtain all landslide features of the key indicators of the landslide location description data" can be as follows: using target compression units to integrate the regional location features and all landslide features of the initial key indicators to obtain all landslide features of the key indicators of the landslide location description data.

[0105] As an example, this application may employ a first target feature extraction unit of the target compression unit to integrate regional location features and all landslide features of the initial key indicators to obtain all landslide features of the candidate key indicators; employ a second target feature extraction unit to integrate all landslide features of the candidate key indicators, all landslide features of the initial key indicators, and local landslide features of the initial key indicators to obtain all landslide features of the transitional key indicators; then, employ a third target feature extraction unit to extract features from all landslide features of the transitional key indicators to obtain all landslide features of the reference key indicators; and employ a fourth target feature extraction unit to integrate all landslide features of the reference key indicators and all landslide features of the transitional key indicators to obtain all landslide features of the key indicators for landslide location description data.

[0106] S103. Based on the local landslide characteristics and the total landslide characteristics of the key indicators, feature extraction is performed on the reference theme results to obtain important features.

[0107] Among them, key features characterize the matching relationship between landslide location description data and reference topic results.

[0108] Regarding step S103, the method for "extracting features from the reference theme results based on the key indicators of local landslide characteristics and the key indicators of all landslide characteristics to obtain important features" can be as shown in steps S1031 to S1032:

[0109] S1031. Extract features from the reference topic results to obtain initial reference topic features.

[0110] As an example, this application may use a first target derivation unit to extract features from the reference topic results to obtain initial topic features; and use a second target derivation unit to integrate the initial topic features and the reference topic results to obtain initial reference topic features.

[0111] S1032. Integrate the key indicators of local landslide characteristics, key indicators of all landslide characteristics, and initial reference theme characteristics to obtain important characteristics.

[0112] S104. Integrate the key indicators of local landslide characteristics, key indicators of all landslide characteristics, and important characteristics to obtain the target theme characteristics of the landslide location description data.

[0113] Regarding step S104, the method for "integrating the key indicator local landslide features, the key indicator all landslide features, and important features to obtain the target theme features of the landslide location description data" can be as follows: extract the important features to obtain the target important features; integrate the target important features, the key indicator local landslide features, and the key indicator all landslide features to obtain the target theme features of the landslide location description data.

[0114] The method for the step "extracting features from important features to obtain target important features" can be as follows: use the first candidate derivation unit to extract features from important features to obtain candidate important features; use the second candidate derivation unit to integrate the candidate important features and important features to obtain target important features.

[0115] S105. Based on the target theme characteristics and reference theme results, generate landslide failure simulation results corresponding to the landslide location description data.

[0116] Regarding step S105, the step "generating landslide failure prediction results corresponding to landslide location description data based on target theme features and reference theme results" can be performed as follows: Based on the target theme features, predict the theme of the landslide location description data to obtain the target theme; integrate the reference theme results and the target theme to obtain the integrated theme, and use the integrated theme as the reference theme result; return to execute the step of extracting features from the reference theme result based on the key indicator local landslide features and the key indicator all landslide features to obtain important features, until the prediction end requirement is met, to obtain multiple target themes; integrate the multiple target themes to generate landslide failure prediction results.

[0117] As an example, the method for the step "based on the target topic features, perform topic prediction on the landslide location description data to obtain the target topic" can be as follows: use a landslide location description data parsing network to perform topic prediction on the landslide location description data based on the target topic features to obtain the target topic.

[0118] As an example, the method for the step "integrating multiple target themes to generate landslide failure simulation results" can be as follows: extract the target reference theme result from the multiple reference theme results; integrate the target reference theme result with the multiple target themes to obtain the landslide failure simulation results.

[0119] In this application, the landslide location description data parsing network used can be a network obtained by converging a specified landslide location description data parsing network. For example, before the step "using the landslide location description data parsing network to extract features from the landslide location description data matrix", this application can perform network parameter convergence on the specified landslide location description data parsing network, as shown in steps S201 to S205:

[0120] S201. Obtain a training dataset containing multiple example landslide location descriptions and abnormal landslide examples corresponding to the example landslide location descriptions.

[0121] Among them, abnormal landslide examples can include normal abnormal landslide examples and abnormal abnormal landslide examples; normal abnormal landslide examples can be abnormal landslide examples with a normal diagnosis result; abnormal abnormal landslide examples can refer to abnormal landslide examples with an abnormal diagnosis result.

[0122] The example landslide location description data can include normal example landslide location description data and abnormal example landslide location description data; correspondingly, normal example landslide location description data can be the example corresponding to the abnormal example of a normal landslide; abnormal example landslide location description data can be the example corresponding to the abnormal example of an abnormal landslide.

[0123] In step S201, the method for "obtaining a training dataset containing multiple example landslide location description data and the abnormal landslide examples corresponding to the example landslide location description data" can be: obtaining a training dataset containing multiple example landslide location description data and the abnormal landslide examples corresponding to the example landslide location description data from the local database of the electronic device.

[0124] S202. Cluster the exemplary landslide location description data and the abnormal landslide examples respectively to obtain at least one landslide location description data region corresponding to the exemplary landslide location description data and at least one topic set corresponding to the abnormal landslide.

[0125] Regarding step 202, there are several ways to segment the example landslide location description data to obtain at least one landslide location description data region corresponding to the example landslide location description data: for example, the scale of key indicator data of the example landslide location description data can be obtained; based on the scale of key indicator data, the segmentation region position of the example landslide location description data can be determined; according to the segmentation region position, the example landslide location description data can be segmented to obtain at least one landslide location description data region.

[0126] An exemplary approach to "determining the segmentation location of exemplary landslide location description data based on the scale of key indicator data" can be as follows: obtain the specified number of segments; integrate the scale of key indicator data and the specified number of segments to obtain the segmentation location of exemplary landslide location description data.

[0127] The specified number of segments may include the number of landslide location description data regions obtained by segmenting the example landslide location description data.

[0128] Regarding step S202, clustering abnormal landslide examples to obtain at least one topic set corresponding to the abnormal landslides can be achieved in two ways: First, using an artificial intelligence network to cluster the abnormal landslide examples and obtain at least one topic set corresponding to the abnormal landslides. Second, using a specified dictionary to perform topic parsing on the abnormal landslide examples and obtain the parsing results; based on the parsing results, clustering the abnormal landslide examples to obtain at least one topic set corresponding to the abnormal landslides.

[0129] S203. Using a specified landslide location description data parsing network, feature extraction is performed on the landslide location description data region and the theme set respectively to obtain key indicator features of at least one key indicator character corresponding to the example landslide location description data, and theme features of at least one theme character corresponding to the abnormal landslide example.

[0130] Here, "key indicator characters" can refer to the amount of information contained in a key indicator feature. Information content can refer to the amount of information in the example landslide location description data. Exemplarily, key indicator characters can include all key indicator characters and regional key indicator characters; all key indicator characters can refer to characters in the key indicator feature that contain all the information in the example landslide location description data; regional key indicator characters can refer to characters in the key indicator feature that contain the information of the landslide location description data area.

[0131] After obtaining the landslide location description data region and thematic set, this application can extract features from the landslide location description data region and thematic set respectively. For example, regarding step 203, "using a specified landslide location description data parsing network to extract features from the landslide location description data region and thematic set respectively, to obtain key indicator features of at least one key indicator character corresponding to the example landslide location description data, and thematic features of at least one thematic character corresponding to the example landslide abnormality," the method can be as follows: Using a specified landslide location description data parsing network to extract features from the landslide location description data region, obtaining all abnormal features corresponding to all key indicator characters and regional abnormal features corresponding to regional key indicator characters, and using all abnormal features and regional abnormal features as key indicator features; using a specified landslide location description data parsing network to extract features from the thematic set, obtaining all thematic features corresponding to all thematic characters and set features corresponding to the characters in the thematic set, and using the set features and key indicator features as key indicator features.

[0132] As an example, the compression unit of the specified landslide location description data parsing network can be used to extract features from the landslide location description data area to obtain all abnormal features corresponding to all key indicator characters and regional abnormal features corresponding to regional key indicator characters; the report encoder of the specified landslide location description data parsing network can be used to extract features from the topic set to obtain all topic features corresponding to all topic characters and set features corresponding to the topic set characters.

[0133] This application can also obtain the regional location data of each landslide location description data area; and extract features from the regional location data by specifying an initial feature extraction unit to obtain the regional location features.

[0134] Next, this application can use a second feature extraction unit to integrate the abnormal features of each initial region based on the reference attention weight to obtain the candidate abnormal region features corresponding to each landslide location description data region; and use the second feature extraction unit to integrate the candidate all abnormal features, the initial region abnormal features and the initial all abnormal features to obtain the transition all abnormal features.

[0135] Next, this application can use a third feature extraction unit to extract features from the candidate abnormal region features and the transitional all abnormal features, to obtain the reference abnormal region features corresponding to the candidate abnormal region features and the reference all abnormal features corresponding to the transitional all abnormal features.

[0136] S204. Perform feature identification on at least one distinguishing character for key indicator features and theme features to obtain the feature identification result corresponding to each distinguishing character.

[0137] The feature identification results represent the discriminability between key indicator features and topic features. The discriminability can be represented by correlation coefficient, similarity coefficient, and confidence coefficient.

[0138] Here, the distinguishing character can refer to the character that identifies key indicator features and theme features. The distinguishing character can include at least one of the following: all distinguishing characters, region distinguishing characters, and potential distinguishing characters.

[0139] The term "complete identification character" refers to the identification character between the example landslide location description data and the abnormal landslide examples. The term "regional identification character" refers to the identification character between the target landslide location description data region and the target subject set. The term "potential identification character" refers to the identification character between potential landslide hazard data, example landslide location description data, and abnormal landslide examples.

[0140] After obtaining the key indicator features and theme features, this application can identify the key indicator features and theme features. For example, regarding step S204, the method of "identifying the key indicator features and theme features at at least one distinguishing character and obtaining the feature identification result corresponding to each distinguishing character" can be as shown in steps S1 to S3:

[0141] S1. Select the target key indicator character corresponding to each identification character from the key indicator characters, and determine the target topic character corresponding to each identification character from the topic characters.

[0142] Regarding step S1, the method for "selecting the target key indicator character corresponding to each identification character from the key indicator characters" can be as follows: obtain a projection relationship set, which includes the projection relationship between the specified identification character and the specified key indicator character; based on the projection relationship set, determine the target key indicator character corresponding to each identification character from the key indicator characters.

[0143] Regarding step S1, the method for "determining the target topic character corresponding to each distinguishing character in the topic characters" can be as follows: obtain a set of correspondences, which includes the correspondence between the specified distinguishing character and the specified topic character; based on the correspondences, determine the target topic character corresponding to each distinguishing character in the topic characters.

[0144] S2. Select the target key indicator features corresponding to the target key indicator characters from the key indicator features, and extract the target theme features corresponding to the target theme characters from the theme features.

[0145] Each target key indicator character has a corresponding target key indicator feature. For example, when the target key indicator character is all key indicator characters, the target key indicator feature can be all abnormal features; when the target key indicator character is a regional key indicator character, the target key indicator feature can be regional abnormal features.

[0146] Regarding step S2, the method for "selecting the target key indicator feature corresponding to the target key indicator character from the key indicator features" can be as follows: obtain the ranking of each key indicator feature; determine the target ranking corresponding to the target key indicator character in the ranking; extract the target key indicator feature corresponding to the target ranking from the key indicator features, and use the target key indicator feature corresponding to the target ranking as the target key indicator feature corresponding to the target key indicator character.

[0147] It is important to understand that the key indicator feature ranked first can be any abnormal feature corresponding to all key indicator characters; the key indicator feature not ranked first can be any abnormal feature corresponding to the regional key indicator characters.

[0148] Each target topic character has a corresponding target topic feature. For example, when the target topic character is all topic characters, the target topic feature can be all topic features; when the target topic character is a set of topic characters, the target topic feature can be a set feature.

[0149] Regarding step S2, the method of "extracting the target topic features corresponding to the target topic characters from the topic features" can be found in the method of "selecting the target key indicator features corresponding to the target key indicator characters from the key indicator features", and will not be repeated here.

[0150] S3. Identify the key indicator features and theme features of the target to obtain the feature identification results corresponding to each identified character.

[0151] After obtaining the target key indicator features and target theme features, this application can distinguish the target key indicator features and target theme features. For example, there are many ways to "distinguish the target key indicator features and target theme features to obtain the feature discrimination result corresponding to each discrimination character". For example, the target key indicator features and target theme features can be used as feature sets, and then the feature sets corresponding to different discrimination characters can be distinguished in parallel to obtain the feature discrimination result corresponding to each discrimination character.

[0152] For example, the method of "identifying the key indicator features and the theme features of the target to obtain the feature identification result corresponding to each identified character" can be as shown in steps S31 to S34:

[0153] S31. Identify at least one target character from the distinguishing characters.

[0154] For step S31, "determining the target distinguishing character from the distinguishing characters" can be done by: randomly extracting the target distinguishing character from the distinguishing characters; or, determining the target distinguishing character from the distinguishing characters according to the specified order corresponding to the distinguishing characters.

[0155] It is important to understand here that when there are at least two target characters identified in the character set, the candidate feature sets corresponding to different target characters can be identified in parallel. The candidate feature sets include candidate key indicator features and candidate topic features.

[0156] S32. Select candidate key indicator features corresponding to target identification characters from the target key indicator features, and extract candidate topic features corresponding to target identification characters from the target topic features.

[0157] After identifying the target distinguishing character, this application can select the candidate key indicator features and candidate topic features corresponding to the target distinguishing character.

[0158] It is important to understand that each distinguishing character has a corresponding target key indicator feature. This application can obtain the target correspondence between the distinguishing character and the key indicator feature. Based on this, for step S32, the method of "selecting the candidate key indicator feature corresponding to the target distinguishing character from the target key indicator features" can be: based on the target correspondence, select the candidate key indicator feature corresponding to the target distinguishing character from the target key indicator features.

[0159] Similarly, the step "extracting candidate topic features corresponding to the target distinguishing character from the target topic features" can be done by: extracting candidate topic features corresponding to the target distinguishing character from the target topic features based on the candidate correspondence between the distinguishing character and the topic features.

[0160] S33. Perform feature identification on the target identification character by identifying the candidate key indicator features and candidate topic features, and obtain the target feature identification result corresponding to the target identification character.

[0161] Regarding step S33, this application can perform feature identification on candidate key indicator features and candidate topic features in several ways, such as Case 1, Case 2, and Case 3:

[0162] (I) Case 1: When the target identification character is all identification characters, the candidate key indicator features include all abnormal features corresponding to the example landslide location description data, and the candidate theme features include all theme features corresponding to the abnormal landslide examples. Based on this, the specific implementation of step S33, "to perform feature identification on the target identification character based on the candidate key indicator features and candidate theme features to obtain the target feature identification result corresponding to the target identification character," can be as follows: Based on all abnormal features and all theme features, calculate the example similarity coefficient between the example landslide location description data and the abnormal landslide examples. All identification characters are the identification characters between the example landslide location description data and the abnormal landslide examples. Integrate the example similarity coefficients to generate all landslide feature identification results corresponding to all identification characters, and use all landslide feature identification results as the target feature identification results.

[0163] It is important to understand here that after integrating the example similarity coefficients to generate all landslide feature identification results corresponding to all identification characters, the results can include the example landslide location description data for each example and the example similarity coefficient between each abnormal landslide example.

[0164] (II) Scenario 2: When the target identification character is a region identification character, the candidate key indicator features also include the abnormal regional features corresponding to the landslide location description data region, and the candidate topic features also include the set features corresponding to the topic set. Based on this, the specific implementation of step S33, "to perform feature identification on the target identification character to obtain the target feature identification result corresponding to the target identification character," can be as follows: Select the target landslide location description data region from the landslide location description data region, and select the target topic set from the topic set. The region identification character is the identification character between the target landslide location description data region and the target topic set. Based on the abnormal regional features and the set features, calculate the confidence coefficient between the target landslide location description data region and the target topic set. Based on the confidence coefficient, determine the region feature identification result corresponding to the target identification character, and use the region feature identification result as the target feature identification result.

[0165] Specifically, each landslide location description data area can be used as the target landslide location description data area; or, a portion of the landslide location description data areas can be used as the target landslide location description data area; or, at least two landslide location description data areas can be integrated to obtain the target landslide location description data area.

[0166] It is important to understand that because there are a large number of abnormal regional features and set features, the related calculations for these features, such as calculating the performance evaluation of regional inference corresponding to the regional identification characters, are time-consuming under the regional identification characters. Therefore, the number of abnormal regional features and set features can be reduced to reduce data redundancy and thus improve computational efficiency.

[0167] To reduce the number of abnormal regional features and set features, this application can cluster the abnormal regional features and set features separately to reduce the number of landslide location description data regions and topic sets, thereby reducing the number of abnormal regional features and set features.

[0168] To address abnormal regional features, this application employs a sub-feature integration layer to cluster these features, obtaining a regional category for each abnormal region. Within the landslide location description data region, the target abnormal region corresponding to each regional category is identified. Then, the sub-feature integration layer integrates the abnormal regional features corresponding to the same regional category, resulting in integrated regional features for the target landslide location description data region. Finally, the cross-attention mechanism within the sub-feature integration layer integrates the integrated regional features and the abnormal regional features to obtain the target abnormal regional features for the target landslide location description data region.

[0169] The step "calculate the confidence coefficient between the target landslide location description data area and the target subject set based on the abnormal features of the target area and the characteristics of the target set" can be performed as follows: transpose the abnormal features of the target area to obtain the transpose of the abnormal features of the target area; calculate the target feature similarity coefficient between the transpose of the abnormal features of the target area and the characteristics of the target set to obtain the confidence coefficient between the target landslide location description data area and the target subject set.

[0170] Then, this application can normalize the target feature similarity coefficient to obtain the confidence coefficient between the target landslide location description data area and the target subject set.

[0171] For scenario two, as an example, the step "determine the regional feature identification result corresponding to the target identification character based on the confidence coefficient" can be implemented as follows: Based on the magnitude of the confidence coefficient, select the regional confidence coefficient corresponding to the target landslide location description data region from the confidence coefficient. The regional confidence coefficient represents the confidence coefficient between the target landslide location description data region and the candidate topic set corresponding to the target landslide location description data region in the target topic set. Based on the magnitude of the confidence coefficient, parse the set confidence coefficient corresponding to the target topic set from the confidence coefficient. The set confidence coefficient represents the confidence coefficient between the target topic set and the candidate landslide location description data regions corresponding to the target topic set in the target landslide location description data region. Integrate the regional confidence coefficient and the set confidence coefficient to obtain the integrated confidence coefficient, and use the integrated confidence coefficient as the regional feature identification result corresponding to the regional identification character.

[0172] It is important to understand here that, for the regional confidence coefficient, the maximum confidence coefficient of the target landslide location description data region for the target subject set can be obtained from the confidence coefficient based on the magnitude of the confidence coefficient, and the maximum confidence coefficient can be used as the regional confidence coefficient.

[0173] Similarly, for the set confidence coefficient, the maximum target confidence coefficient of the target set for the data area describing the target landslide location can be obtained from the confidence coefficient based on the magnitude of the confidence coefficient, and the maximum target confidence coefficient can be used as the set confidence coefficient.

[0174] (III) Case 3: When the target identification character is a potential identification character, the target key indicator features include all abnormal features corresponding to the example landslide location description data, and the target theme features include all theme features corresponding to the abnormal landslide examples. For step S33, "to perform feature identification on the target identification character to obtain the target feature identification result corresponding to the target identification character" can be: obtain the potential landslide hazard data corresponding to the example landslide location description data, and perform feature extraction on the potential landslide hazard data to obtain potential landslide hazard features. The potential identification character is the identification character between the potential landslide hazard data, the example landslide location description data, and the abnormal landslide examples. Perform feature identification on the potential landslide hazard features, all abnormal features, and all theme features on the potential identification character to obtain the potential feature identification result corresponding to the potential identification character, and use the potential feature identification result as the target feature identification result.

[0175] For scenario three, the method of "identifying the potential landslide hazard characteristics, all abnormal characteristics, and all thematic characteristics in the potential identification character to obtain the potential feature identification result corresponding to the potential identification character" can be as follows: Based on the potential landslide hazard characteristics and all abnormal characteristics, calculate the abnormal correlation degree between the potential landslide hazard data and the example landslide location description data; based on the potential landslide hazard characteristics and all thematic characteristics, calculate the thematic correlation degree between the potential landslide hazard data and the abnormal landslide theme; based on the abnormal correlation degree and thematic correlation degree, generate the potential feature identification result corresponding to the potential identification character.

[0176] Among them, abnormal correlation can characterize the degree of correlation between potential landslide hazard data and exemplary landslide location description data; topic correlation can characterize the degree of correlation between potential landslide hazard data and abnormal landslide topics.

[0177] Among them, abnormal correlation degree and topic correlation degree can be used as the potential feature identification results corresponding to potential identification characters.

[0178] An exemplary approach to “calculating the abnormal correlation between potential landslide hazard data and exemplary landslide location description data based on potential landslide hazard characteristics and all abnormal characteristics” can be as follows: simplify all abnormal characteristics to obtain simplified abnormal characteristics; calculate the similarity coefficient between the simplified abnormal characteristics and potential landslide hazard characteristics to obtain the abnormal correlation between potential landslide hazard data and exemplary landslide location description data.

[0179] Among them, the similarity coefficient between the simplified abnormal features and the potential landslide hazard features can be used as the abnormal correlation between the potential landslide hazard data and the example landslide location description data.

[0180] For example, the method of "calculating the subject correlation between potential landslide hazard data and abnormal landslide subject based on potential landslide hazard characteristics and all subject characteristics" can be found in the method of "calculating the abnormal correlation between potential landslide hazard data and example landslide location description data based on potential landslide hazard characteristics and all abnormal characteristics", which will not be repeated here.

[0181] S34. Return to the step of identifying the target character among the distinguished characters, until every distinguished character is the target character, and obtain the feature recognition result corresponding to each distinguished character.

[0182] S205. Based on the feature identification results, key indicator features, and theme features, the analytical network for describing the landslide location is converged to obtain the analytical network for describing the landslide location. The analytical network for describing the landslide location is then used to analyze the landslide location description data to obtain the landslide failure prediction results.

[0183] After obtaining the feature identification results, key indicator features, and theme features, this application can converge the parsing network for the specified landslide location description data. For example, the identification characters include all identification characters and region identification characters. Based on this, in step S205, the step "converging the parsing network for the specified landslide location description data based on the feature identification results, key indicator features, and theme features to obtain the landslide location description data parsing network" can be as shown in steps S2051 to S2055:

[0184] S2051. Select all landslide feature identification results corresponding to all identification characters and regional feature identification results corresponding to regional identification characters from the feature identification results.

[0185] S2052. Based on the identification results of all landslide features, calculate the typical inference performance evaluation between the typical landslide location description data and the abnormal landslide examples.

[0186] Among them, the exemplary extrapolation performance evaluation characterizes the shared extrapolation performance evaluation between the exemplary landslide location description data and the abnormal landslide examples.

[0187] For S2052, since all landslide feature identification results include the example similarity coefficient between the landslide location description data of each example and the abnormal landslide example, the method for the step "calculating the example inference performance evaluation between the example landslide location description data and the abnormal landslide example based on all landslide feature identification results" can be as follows: select the target example similarity coefficient from the example similarity coefficients included in all landslide feature identification results; project the example similarity coefficient and the target example similarity coefficient to obtain the example inference performance evaluation between the example landslide location description data and the abnormal landslide example.

[0188] For example, the target example similarity coefficient includes the first target example similarity coefficient corresponding to the example landslide location description data and the second target example similarity coefficient corresponding to the abnormal landslide example; the example similarity coefficient includes the first example similarity coefficient of the example landslide location description data for the abnormal landslide example and the second example similarity coefficient of the abnormal landslide example for the example landslide location description data. Based on this, the step "using the example inference performance evaluation function to project the example similarity coefficient and the target example similarity coefficient to obtain the example inference performance evaluation between the example landslide location description data and the abnormal landslide example" can be done as follows: integrate the first target example similarity coefficient and the first example similarity coefficient to obtain the first integrated similarity coefficient; integrate the second target example similarity coefficient and the second example similarity coefficient to obtain the second integrated similarity coefficient; obtain the inference performance evaluation reference parameters, and use the example inference performance evaluation function to project the first integrated similarity coefficient, the second integrated similarity coefficient, and the inference performance evaluation reference parameters to obtain the example inference performance evaluation between the example landslide location description data and the abnormal landslide example.

[0189] S2053. Based on the regional feature identification results, as well as the abnormal regional features and set features corresponding to the regional identification characters, determine the regional matching and inference performance evaluation of the landslide location description data region.

[0190] Among them, the regional matching inference performance evaluation characterizes the performance evaluation of matching inference between the landslide location description data region and the theme set.

[0191] Regarding step S2053, the method for "determining the regional matching and inference performance evaluation corresponding to the landslide location description data region based on the regional feature identification results, as well as the regional abnormal features and set features corresponding to the regional identification characters" can be as shown in steps S531 to S533:

[0192] S531. Based on the regional feature discrimination results, determine the regional training labels corresponding to the regional discrimination characters.

[0193] The regional feature identification results include the confidence coefficient between the target landslide location description data region of the example landslide location description data and the target topic set of the abnormal landslide examples. Based on this, the step "determine the regional training label corresponding to the regional identification character according to the regional feature identification results" can be performed as follows: select a reference abnormal region from the abnormal regions and select a reference topic set from the topic set; combine the reference abnormal regions and the reference topic set to obtain at least one reference data set, which includes at least one target reference abnormal region and at least one target reference topic set corresponding to the target reference abnormal region; select the target confidence coefficient corresponding to the reference data set from the regional feature identification results and integrate the target confidence coefficients to obtain the target integrated confidence coefficient; determine the regional training label corresponding to the regional identification character based on the target integrated confidence coefficient.

[0194] Among these, multiple reference abnormal regions can be randomly selected from the abnormal regions; multiple reference topic sets can be randomly selected from the topic set.

[0195] In this case, each reference abnormal region can be used as the target reference abnormal region; or, some reference abnormal regions can be used as the target reference abnormal region.

[0196] In this case, each set of reference topics can be used as the target set of reference topics; or, a subset of the set of reference topics can be used as the target set of reference topics.

[0197] As an example, the step "selecting the target confidence coefficient corresponding to the reference data set from the regional feature identification results" can be done as follows: extract the first confidence coefficient corresponding to each abnormal target reference region in the reference data set from the regional feature identification results; select the second confidence coefficient corresponding to each target reference topic set in the reference data set from the regional feature identification results; and use the first confidence coefficient and the second confidence coefficient as the target confidence coefficient corresponding to the reference data set.

[0198] As an example, the step "integrate the target confidence coefficients to obtain the target integrated confidence coefficients" can be done by adding up the target confidence coefficients corresponding to each target confidence coefficient in the reference dataset to obtain the target integrated confidence coefficients.

[0199] As an example, the step "determine the region training label corresponding to the region-distinguishing character based on the target integration confidence coefficient" can be performed as follows: project the confidence coefficient of each target integration to obtain the projected confidence coefficient; normalize the projected confidence coefficient to obtain the region training label corresponding to the region-distinguishing character.

[0200] S532. Based on regional abnormal characteristics and set characteristics, calculate the candidate confidence coefficient between the landslide location description data region and the theme set.

[0201] Regarding step S532, based on the aforementioned data set, the method for "calculating the candidate confidence coefficient between the landslide location description data area and the topic set based on the regional abnormal features and set features" can be as follows: extract the current landslide location description data area and the current topic set from the target data set, and calculate the candidate confidence coefficient between the current landslide location description data area and the current topic set based on the regional abnormal features corresponding to the current abnormal area and the set features corresponding to the current topic set.

[0202] Among them, a similarity coefficient function can be used to calculate the candidate confidence coefficient between the current landslide location description data area and the current topic set based on the regional abnormality characteristics corresponding to the current abnormal area and the set characteristics corresponding to the current topic set.

[0203] As an example, the abnormal features corresponding to the current abnormal area are transposed to obtain the target transpose corresponding to the abnormal features; the similarity coefficient function is used to calculate the candidate similarity coefficient between the target transpose and the set features corresponding to the current topic set; the candidate similarity coefficient is normalized to obtain the candidate confidence coefficient between the current landslide location description data area and the current topic set.

[0204] S533. Calculate the target inference performance evaluation between the regional training labels and candidate confidence coefficients, and use the target inference performance evaluation as the regional matching inference performance evaluation corresponding to the landslide location description data region.

[0205] It is important to understand here that the performance evaluation of region matching inference can bring the candidate confidence coefficient closer to the region training label, thereby strengthening the relationship between the topic set and the landslide location description data region.

[0206] S2054. Based on key indicator features and theme features, predict the landslide location description data to obtain theme prediction probabilities, and project the theme prediction probabilities to obtain the prediction and extrapolation performance evaluation data set corresponding to the landslide location description data.

[0207] For step S2054, the prediction performance evaluation data set function can be used to project the prediction probability of the theme to obtain the prediction performance evaluation data set corresponding to the landslide location description data.

[0208] S2055. Based on the datasets of example extrapolation performance evaluation, regional matching extrapolation performance evaluation, and prediction extrapolation performance evaluation, the parsing network for the specified landslide location description data is converged to obtain the parsing network for the landslide location description data.

[0209] The distinguishing characters also include potential distinguishing characters. Based on this, for step S2055, the method of "converging the parsing network for the specified landslide location description data based on the datasets of example extrapolation performance evaluation, regional matching extrapolation performance evaluation, and prediction extrapolation performance evaluation to obtain the parsing network for the landslide location description data" can be as shown in steps S51 to S54:

[0210] S51. Select the latent feature recognition result corresponding to the latent character from the feature recognition results.

[0211] Among them, the results of potential feature identification include the degree of abnormal correlation between potential landslide hazard data corresponding to potential identification characters and example landslide location description data, as well as the degree of thematic correlation between potential landslide hazard data and abnormal landslide themes.

[0212] S52. Based on the results of latent feature discrimination, determine the potential training labels corresponding to the potential discrimination characters.

[0213] Regarding step S52, the method for "determining the potential training labels corresponding to the potential identification characters based on the potential feature identification results" can be as follows: Obtain the potential features corresponding to the potential landslide hazard data; select reference potential landslide hazard data from the potential landslide hazard data; select reference example landslide location description data from the example landslide location description data; select reference abnormal landslide examples from the abnormal landslide examples; combine the reference potential landslide hazard data and the reference example landslide location description data to obtain at least one reference potential data set, which includes at least one target reference potential landslide hazard data and at least one target reference example landslide location description data corresponding to the target reference potential landslide hazard data; combine the reference potential landslide hazard data and the reference abnormal landslide examples... The process involves combining at least one candidate potential dataset, which includes at least one candidate reference potential landslide hazard data and at least one reference landslide abnormality example corresponding to the candidate reference potential landslide hazard data; selecting the target abnormality correlation degree corresponding to the reference potential dataset from the potential feature identification results, and integrating the target abnormality correlation degree to obtain the integrated abnormality correlation degree; determining the potential abnormality label corresponding to the potential identification character based on the integrated abnormality correlation degree; selecting the target topic correlation degree corresponding to the candidate potential dataset from the potential feature identification results, and integrating the target topic correlation degree to obtain the integrated topic correlation degree; determining the potential topic label corresponding to the potential identification character based on the integrated topic correlation degree; and using the potential abnormality label and the potential topic label as potential training labels.

[0214] The steps “selecting reference potential landslide hazard data from potential landslide hazard data”, “selecting reference example landslide location description data from example landslide location description data”, and “selecting reference abnormal landslide examples from abnormal landslide examples” can be found in the aforementioned step “selecting reference abnormal areas from abnormal areas”, and will not be repeated here.

[0215] The steps “determine the potential abnormal labels corresponding to the potential distinguishing characters based on the integration of abnormal correlations” and “determine the potential topic labels corresponding to the potential distinguishing characters based on the integration of topic correlations” can be found in the aforementioned step “determine the region training labels corresponding to the region distinguishing characters based on the target integration confidence coefficient”, and will not be repeated here.

[0216] S53. Based on the potential training labels, potential landslide hazard characteristics, key indicator characteristics and theme characteristics corresponding to the potential landslide hazard data, calculate the potential inference performance evaluation corresponding to the example landslide location description data.

[0217] Among them, the potential inference performance assessment characterizes the correlation inference performance assessment between potential landslide hazard data, exemplary landslide location description data, and abnormal landslide examples.

[0218] For step S53, the key indicator features include all abnormal features, the topic features include all topic features, and the potential training labels include potential abnormal labels and potential topic labels. The step "calculate the potential inference performance evaluation corresponding to the example landslide location description data based on the potential training labels, potential landslide hazard features, key indicator features, and topic features corresponding to the potential landslide hazard data" can be performed as follows: simplify all abnormal features to obtain simplified abnormal features; simplify all topic features to obtain simplified topic features; calculate the similarity coefficient between the simplified abnormal features and the potential landslide hazard features to obtain the candidate abnormal correlation degree between the potential landslide hazard data and the example landslide location description data; calculate the similarity coefficient between the simplified topic features and the potential landslide hazard features to obtain the candidate topic correlation degree between the potential landslide hazard data and the abnormal landslide examples; and calculate the potential inference performance evaluation corresponding to the example landslide location description data based on the potential training labels, candidate topic correlation degree, and candidate abnormal correlation degree.

[0219] The step "calculate the potential inference performance evaluation corresponding to the example landslide location description data based on potential training labels, candidate topic correlation, and candidate abnormal correlation" can be performed as follows: calculate the first similarity coefficient between the potential abnormal labels and the candidate topic similarity coefficients; calculate the second similarity coefficient between the potential topic labels and the candidate abnormal correlation; and integrate the first and second similarity coefficients using the potential inference performance evaluation function to obtain the potential inference performance evaluation corresponding to the example landslide location description data.

[0220] S54. Based on the datasets of potential extrapolation performance evaluation, example extrapolation performance evaluation, regional matching extrapolation performance evaluation, and prediction extrapolation performance evaluation, converge the parsing network for the specified landslide location description data to obtain the parsing network for the landslide location description data.

[0221] For step S54, the inference performance evaluation integration function can be used to integrate the datasets of potential inference performance evaluation, example inference performance evaluation, regional matching inference performance evaluation and prediction inference performance evaluation to obtain the integrated inference performance evaluation; based on the integrated inference performance evaluation, the parsing network for the specified landslide location description data is converged to obtain the parsing network for the landslide location description data.

[0222] As an example, this application can obtain target extrapolation performance evaluation parameters; based on the target extrapolation performance evaluation parameters, integrate the example extrapolation performance evaluation, potential extrapolation performance evaluation and regional matching extrapolation performance evaluation to obtain an initial integrated extrapolation performance evaluation; add the initial integrated extrapolation performance evaluation and the predicted extrapolation performance evaluation datasets to obtain the integrated extrapolation performance evaluation.

[0223] It should be understood here that, for steps S201 to S205, the embodiments of this application can identify abnormal features of the landslide location description data and the theme features of abnormal landslide examples, and obtain feature identification results of at least one identification character. Therefore, the feature identification results of at least one identification character, key indicator features, and theme features can be used to train a specified landslide location description data parsing network. This enables the landslide location description data parsing network to learn the ability to identify any landslide location description data, thereby improving the accuracy of the landslide failure prediction results generated by the landslide location description data parsing network.

[0224] This application can obtain landslide location description data to be processed and corresponding reference theme results, and cluster the landslide location description data to obtain at least one landslide location description data matrix; perform multi-dimensional feature extraction on the landslide location description data matrix to obtain key indicators of local landslide features and key indicators of all landslide features for each landslide location description data; extract features from the reference theme results based on the key indicators of local landslide features and key indicators of all landslide features to obtain important features, which characterize the matching relationship between the landslide location description data and the reference theme results; integrate the key indicators of local landslide features, key indicators of all landslide features, and important features to obtain the target theme features of the landslide location description data; and generate landslide failure prediction results corresponding to the landslide location description data based on the target theme features and the reference theme results. Because this application can perform multi-dimensional feature extraction on the landslide location description data matrix, obtaining a large amount of information such as key indicator local landslide features and key indicator all landslide features, this application can further extract features from the reference theme results corresponding to the landslide location description data based on the key indicator local landslide features and key indicator all landslide features to obtain important features. Then, by combining the key indicator local landslide features, key indicator all landslide features, and important features, the application can obtain target theme features that have a deeper content representation between the landslide location description data and the reference theme results. Thus, based on the target theme features, the application can accurately predict the landslide failure prediction results of the landslide location description data.

[0225] A landslide failure prediction method, the specific process of which is as follows: steps S501 to S508:

[0226] S501. The electronic device acquires a training dataset that includes multiple example landslide location descriptions and abnormal landslide examples corresponding to the example landslide location descriptions.

[0227] For example, in step S501, the electronic device sends a sample retrieval request to the storage server through the internal network, so that the storage server can extract sample landslide location description data and sample landslide location description data from the storage space corresponding to the storage server, and return sample landslide location description data and abnormal landslide examples to the electronic device; the electronic device receives the sample landslide location description data and abnormal landslide examples returned by the storage server, and generates a training dataset based on the sample landslide location description data and abnormal landslide examples.

[0228] S502 and the electronic equipment respectively cluster the exemplary landslide location description data and the abnormal landslide examples to obtain at least one landslide location description data region corresponding to the exemplary landslide location description data and at least one topic set corresponding to the abnormal landslide.

[0229] For step S502, the method for segmenting the example landslide location description data to obtain at least one landslide location description data region corresponding to the example landslide location description data can be as follows: obtain the key indicator data scale of the example landslide location description data; determine the segmentation region location of the example landslide location description data based on the key indicator data scale; segment the example landslide location description data according to the segmentation region location to obtain at least one landslide location description data region.

[0230] The method for "determining the segmented regions of the example landslide location description data based on the scale of key indicator data" can be as follows: obtain a specified number of segments; integrate the key indicator data scale and the specified number of segments to obtain the segmented regions of the example landslide location description data. The specified number of segments can include the number of landslide location description data regions obtained by segmenting the example landslide location description data.

[0231] Regarding step S502, the method for clustering abnormal landslide examples to obtain at least one topic set corresponding to abnormal landslides can be as follows: using a specified dictionary to perform topic parsing on the abnormal landslide examples and obtaining the parsing results; based on the parsing results, clustering the abnormal landslide examples to obtain at least one topic set corresponding to abnormal landslides.

[0232] S503. The electronic device uses a specified landslide location description data parsing network to extract features from the landslide location description data area and the topic set, respectively, to obtain key indicator features of at least one key indicator character corresponding to the example landslide location description data, and topic features of at least one topic character corresponding to the abnormal landslide example.

[0233] Regarding step S503, the method of "using a specified landslide location description data parsing network to extract features from the landslide location description data region and the theme set respectively, to obtain key indicator features of at least one key indicator character corresponding to the example landslide location description data, and theme features of at least one theme character corresponding to the abnormal landslide example" can be as follows: The electronic device uses a specified landslide location description data parsing network to extract features from the landslide location description data region, to obtain all abnormal features corresponding to all key indicator characters and regional abnormal features corresponding to regional key indicator characters, and uses all abnormal features and regional abnormal features as key indicator features; it uses a specified landslide location description data parsing network to extract features from the theme set, to obtain all theme features corresponding to all theme characters and set features corresponding to theme set characters, and uses set features and key indicator features as key indicator features.

[0234] S504. The electronic device selects the target key indicator character corresponding to each distinguishing character from the key indicator characters, and determines the target theme character corresponding to each distinguishing character from the theme characters.

[0235] S505. Electronic devices select the target key indicator features corresponding to the target key indicator characters from the key indicator features, and extract the target theme features corresponding to the target theme characters from the theme features.

[0236] S506. The electronic device identifies the key indicator features and the theme features of the target to obtain the feature identification result corresponding to each identified character.

[0237] Among them, the feature identification results represent the discriminability between key indicator features and theme features.

[0238] For example, when the target distinguishing character is all distinguishing characters, the candidate key indicator features include all abnormal features corresponding to the example landslide location description data, and the candidate topic features include all topic features corresponding to the abnormal landslide examples. Based on this, the electronic device can calculate the example similarity coefficient between the example landslide location description data and the abnormal landslide examples based on all abnormal features and all topic features. All distinguishing characters are the distinguishing characters between the example landslide location description data and the abnormal landslide examples. The example similarity coefficients are integrated to generate all landslide feature identification results corresponding to all distinguishing characters, and the all landslide feature identification results are used as the target feature identification results.

[0239] For example, when the target identification character is a region identification character, the candidate key indicator features also include the abnormal regional features corresponding to the landslide location description data region, and the candidate topic features also include the set features corresponding to the topic set. Based on this, the electronic device can select the target landslide location description data region from the landslide location description data region and select the target topic set from the topic set. The region identification character is the identification character between the target landslide location description data region and the target topic set. Based on the abnormal regional features and set features, the confidence coefficient between the target landslide location description data region and the target topic set is calculated. According to the confidence coefficient, the region feature identification result corresponding to the target identification character is determined, and the region feature identification result is used as the target feature identification result.

[0240] For example, when the target distinguishing character is a potential distinguishing character, the target key indicator features include all abnormal features corresponding to the example landslide location description data, and the target theme features include all theme features corresponding to the abnormal landslide examples. Based on this, the electronic device can acquire the potential landslide hazard data corresponding to the example landslide location description data, and perform feature extraction on the potential landslide hazard data to obtain potential landslide hazard features. The potential distinguishing character is the distinguishing character between the potential landslide hazard data, the example landslide location description data, and the abnormal landslide examples. The potential landslide hazard features, all abnormal features, and all theme features are then used to perform feature identification on the potential distinguishing character to obtain the potential feature identification result corresponding to the potential distinguishing character, and the potential feature identification result is used as the target feature identification result.

[0241] S507. Based on feature identification results, key indicator features, and theme features, the electronic device converges the parsing network for the specified landslide location description data to obtain the parsing network for the landslide location description data.

[0242] For example, regarding S507, the electronic device can select all landslide feature identification results corresponding to all identification characters and regional feature identification results corresponding to regional identification characters from the feature identification results; based on all landslide feature identification results, calculate the example inference performance evaluation between the example landslide location description data and the abnormal landslide examples, and the example inference performance evaluation characterizes the shared inference performance evaluation between the example landslide location description data and the abnormal landslide examples; based on the regional feature identification results, as well as the regional abnormal features and set features corresponding to the regional identification characters, determine the region corresponding to the landslide location description data. The regional matching and extrapolation performance evaluation is used to characterize the matching and extrapolation performance between the landslide location description data region and the theme set. Based on key indicator features and theme features, the example landslide location description data is predicted to obtain the theme prediction probability. The theme prediction probability is then projected to obtain the prediction extrapolation performance evaluation data set corresponding to the example landslide location description data. Based on the example extrapolation performance evaluation, regional matching and extrapolation performance evaluation, and prediction extrapolation performance evaluation data sets, the parsing network for the specified landslide location description data is converged to obtain the landslide location description data parsing network.

[0243] S508. The electronic equipment uses a landslide location description data parsing network to parse the landslide location description data and obtain the landslide failure prediction results.

[0244] For step S508, for example, the electronic device can acquire the landslide location description data to be processed and the reference topic results corresponding to the landslide location description data, and cluster the landslide location description data to obtain at least one landslide location description data matrix corresponding to the landslide location description data; use a landslide location description data parsing network to perform multi-dimensional feature extraction on the landslide location description data matrix to obtain the key indicator local landslide features and the key indicator all landslide features of each landslide location description data; based on the key indicator local landslide features and the key indicator all landslide features, feature extraction is performed on the reference topic results to obtain important features, which characterize the matching relationship between the landslide location description data and the reference topic results; integrate the key indicator local landslide features, the key indicator all landslide features, and the important features to obtain the target topic features of the landslide location description data; based on the target topic features and the reference topic results, generate the landslide failure prediction results corresponding to the landslide location description data.

[0245] This application can obtain landslide location description data to be processed and corresponding reference theme results, and cluster the landslide location description data to obtain at least one landslide location description data matrix; perform multi-dimensional feature extraction on the landslide location description data matrix to obtain key indicators of local landslide features and key indicators of all landslide features for each landslide location description data; extract features from the reference theme results based on the key indicators of local landslide features and key indicators of all landslide features to obtain important features, which characterize the matching relationship between the landslide location description data and the reference theme results; integrate the key indicators of local landslide features, key indicators of all landslide features, and important features to obtain the target theme features of the landslide location description data; and generate landslide failure prediction results corresponding to the landslide location description data based on the target theme features and the reference theme results. Because this application can perform multi-dimensional feature extraction on the landslide location description data matrix, obtaining a large amount of information such as key indicator local landslide features and key indicator all landslide features, this application can further extract features from the reference theme results corresponding to the landslide location description data based on the key indicator local landslide features and key indicator all landslide features to obtain important features. Then, by combining the key indicator local landslide features, key indicator all landslide features, and important features, the application can obtain target theme features that have a deeper content representation between the landslide location description data and the reference theme results. Thus, based on the target theme features, the application can accurately predict the landslide failure prediction results of the landslide location description data.

[0246] Based on the above, a landslide failure simulation device is provided, the device comprising:

[0247] The matrix acquisition module is used to obtain the landslide location description data to be processed and the reference topic results corresponding to the landslide location description data, and to cluster the landslide location description data to obtain at least one landslide location description data matrix corresponding to the landslide location description data.

[0248] The all-landslide feature extraction module is used to perform multi-dimensional feature extraction on the landslide location description data matrix to obtain the key indicators of local landslide features for each landslide location description data and the key indicators of all landslide features for the landslide location description data.

[0249] The important feature acquisition module is used to extract features from the reference theme results based on the local landslide features and the total landslide features of the key indicators, and obtain important features, which characterize the matching relationship between the landslide location description data and the reference theme results;

[0250] The theme feature integration module is used to integrate the key indicator local landslide features, key indicator all landslide features and important features to obtain the target theme features of the landslide location description data;

[0251] The result generation module is used to generate landslide failure simulation results corresponding to the landslide location description data based on the target theme features and the reference theme results.

[0252] Based on the above, a landslide failure simulation system is shown, including a processor and a memory that communicate with each other. The processor is used to read a computer program from the memory and execute it to implement the above method.

[0253] Based on the above, a computer-readable storage medium is also provided, on which a computer program stored implements the above method during runtime.

[0254] In summary, based on the above scheme, the landslide location description data to be processed and the corresponding reference theme results are obtained. The landslide location description data is then clustered to obtain at least one landslide location description data matrix. Multi-dimensional feature extraction is performed on the landslide location description data matrix to obtain key indicators (local landslide features and all landslide features) for each landslide location description data. Based on the key indicators (local and all landslide features), feature extraction is performed on the reference theme results to obtain important features, which characterize the matching relationship between the landslide location description data and the reference theme results. The key indicators (local, all, and important features) and the target theme features of the landslide location description data are integrated to obtain the target theme features of the landslide location description data. Based on the target theme features and the reference theme results, landslide failure prediction results corresponding to the landslide location description data are generated. Because this application can perform multi-dimensional feature extraction on the landslide location description data matrix, obtaining a large amount of information such as key indicator local landslide features and key indicator all landslide features, this application can further extract features from the reference theme results corresponding to the landslide location description data based on the key indicator local landslide features and key indicator all landslide features to obtain important features. Then, by combining the key indicator local landslide features, key indicator all landslide features, and important features, the application can obtain target theme features that have a deeper content representation between the landslide location description data and the reference theme results. Thus, based on the target theme features, the application can accurately predict the landslide failure prediction results of the landslide location description data.

[0255] It should be understood that the systems and modules described above can be implemented in various ways. For example, in some embodiments, the systems and modules can be implemented using hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated hardware. Those skilled in the art will understand that the methods and systems described above can be implemented using computer-executable instructions and / or included in processor control code, for example, on a media such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The systems and modules of this application can be implemented not only with hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips and transistors, or programmable hardware devices such as field-programmable gate arrays and programmable logic devices, but also with software, for example, executed by various types of processors, and can also be implemented using a combination of the aforementioned hardware circuits and software (e.g., firmware).

[0256] It should be noted that different embodiments may produce different beneficial effects. In different embodiments, the beneficial effects may be any one or a combination of the above, or any other possible beneficial effects.

Claims

1. A method for predicting landslide failure, characterized in that, include: Obtain the landslide location description data to be processed and the reference topic results corresponding to the landslide location description data, and cluster the landslide location description data to obtain at least one landslide location description data matrix corresponding to the landslide location description data; Multi-dimensional feature extraction is performed on the landslide location description data matrix to obtain the key indicators of local landslide features for each landslide location description data and the key indicators of all landslide features for the landslide location description data. Based on the key indicators of local landslide characteristics and all landslide characteristics, feature extraction is performed on the reference theme results to obtain important features. The important features characterize the matching relationship between the landslide location description data and the reference theme results. By integrating the key indicators of local landslide characteristics, all landslide characteristics, and important characteristics, the target theme characteristics of the landslide location description data are obtained. Based on the target theme characteristics and the reference theme results, landslide failure simulation results corresponding to the landslide location description data are generated.

2. The landslide failure prediction method according to claim 1, characterized in that, The process of extracting multi-dimensional features from the landslide location description data matrix yields key indicators for each landslide location description data, namely, local landslide features and all landslide features, including: The regional location data of each landslide location description data matrix is ​​parsed from the landslide location description data, and the regional location data is extracted using a landslide location description data parsing network to obtain the regional location features of each landslide location description data matrix. The landslide location description data parsing network is used to extract features from the landslide location description data matrix to obtain the initial key index local landslide features and the initial key index all landslide features of the landslide location description data for each landslide location description data matrix. The regional location features are integrated with the initial key indicator local landslide features to obtain the key indicator local landslide features of each landslide location description data matrix; By integrating the regional location features and the initial key indicators of all landslide features, the key indicators of all landslide features of the landslide location description data are obtained.

3. The landslide failure prediction method according to claim 2, characterized in that, Before using the landslide location description data parsing network to extract features from the landslide location description data matrix, the method further includes: Obtain a training dataset, which includes multiple example landslide location description data and abnormal landslide examples corresponding to the example landslide location description data; Cluster the exemplary landslide location description data and the abnormal landslide examples respectively to obtain at least one landslide location description data region corresponding to the exemplary landslide location description data and at least one topic set corresponding to the abnormal landslide; A specified landslide location description data parsing network is used to extract features from the landslide location description data region and the topic set, respectively, to obtain key indicator features of at least one key indicator character corresponding to the example landslide location description data, and topic features of at least one topic character corresponding to the abnormal landslide example. The key indicator features and the topic features are distinguished by at least one distinguishing character to obtain the feature distinguishing result corresponding to each distinguishing character. The feature distinguishing result represents the distinguishing degree between the key indicator features and the topic features. Based on the feature identification results, key indicator features, and theme features, the parsing network for the specified landslide location description data is converged to obtain the landslide location description data parsing network.

4. The landslide failure prediction method according to claim 3, characterized in that, The step of performing feature discrimination on at least one distinguishing character to obtain the feature discrimination result corresponding to each distinguishing character includes: Select the target key indicator character corresponding to each identification character from the key indicator characters, and determine the target topic character corresponding to each identification character from the topic characters; Select the target key indicator feature corresponding to the target key indicator character from the key indicator features, and extract the target theme feature corresponding to the target theme character from the theme features; The target key indicator features and the target topic features are identified to obtain the feature identification result corresponding to each identified character.

5. The landslide failure prediction method according to claim 4, characterized in that, The step of identifying the target key indicator features and the target topic features to obtain the feature identification result corresponding to each identified character includes: At least one target character is identified from the identified characters; candidate key indicator features corresponding to the target character are selected from the target key indicator features, and candidate topic features corresponding to the target character are extracted from the target topic features; The candidate key indicator features and the candidate topic features are used to perform feature discrimination on the target identification character to obtain the target feature discrimination result corresponding to the target identification character; Return to the step of identifying the target character among the identified characters, until every identified character is the target character, and obtain the feature identification result corresponding to each identified character.

6. The landslide failure prediction method according to claim 5, characterized in that, The candidate key indicator features include all abnormal features corresponding to the landslide location description data of the example, and the candidate topic features include all topic features corresponding to the abnormal landslide examples; the step of performing feature discrimination on the target identification character using the candidate key indicator features and the candidate topic features to obtain the target feature discrimination result corresponding to the target identification character includes: When the target identification character is all identification characters, the similarity coefficient between the example landslide location description data and the abnormal landslide example is calculated based on all abnormal features and all topic features, where all identification characters are the identification characters between the example landslide location description data and the abnormal landslide example. The similarity coefficients of the examples are integrated to generate all landslide feature identification results corresponding to all the identification characters, and the all landslide feature identification results are used as the target feature identification results.

7. The landslide failure prediction method according to claim 5, characterized in that, The candidate key indicator features also include regional abnormal features corresponding to the landslide location description data area, and the candidate topic features also include set features corresponding to the topic set; the step of performing feature discrimination on the target identification character using the candidate key indicator features and the candidate topic features to obtain the target feature discrimination result corresponding to the target identification character includes: When the target identification character is a region identification character, a target landslide location description data region is selected from the landslide location description data region, and a target topic set is selected from the topic set. The region identification character is the identification character between the target landslide location description data region and the target topic set. Based on the abnormal characteristics of the region and the characteristics of the set, calculate the confidence coefficient between the target landslide location description data region and the target topic set; Based on the confidence coefficient, the regional feature recognition result corresponding to the target character is determined, and the regional feature recognition result is used as the target feature recognition result.

8. The landslide failure prediction method according to claim 7, characterized in that, The step of determining the region feature discrimination result corresponding to the target character based on the confidence coefficient includes: Based on the magnitude of the confidence coefficient, the regional confidence coefficient corresponding to the target landslide location description data region is selected from the confidence coefficients. The regional confidence coefficient represents the confidence coefficient between the target landslide location description data region and the candidate topic set corresponding to the target landslide location description data region in the target topic set. Based on the magnitude of the confidence coefficient, the set confidence coefficient corresponding to the target topic set is parsed from the confidence coefficient. The set confidence coefficient represents the confidence coefficient between the target topic set and the candidate landslide location description data area corresponding to the target topic set in the target landslide location description data area. The region confidence coefficient and the set confidence coefficient are integrated to obtain the integrated confidence coefficient, and the integrated confidence coefficient is used as the region feature recognition result corresponding to the region recognition character.

9. The landslide failure prediction method according to claim 7, characterized in that, Selecting a target landslide location description data region from the landslide location description data region includes: clustering the abnormal features of the region to obtain a region category corresponding to each abnormal region, and determining the target abnormal region corresponding to the region category in the landslide location description data region; selecting a target topic set from the topic set includes: clustering the set features to obtain a set category corresponding to each topic set, and determining the target topic set in the topic set.

10. The landslide failure prediction method according to claim 9, characterized in that, The step of calculating the confidence coefficient between the target landslide location description data region and the target topic set based on the abnormal characteristics and set characteristics of the region includes: The abnormal features of the same area category are integrated to obtain the integrated area features corresponding to the target landslide location description data area. The integrated area features and the abnormal features of the area are then integrated to obtain the target area abnormal features corresponding to the target landslide location description data area. The set features corresponding to the same unit category are integrated to obtain the integrated theme features corresponding to the target theme set. The integrated theme features and the unit theme features corresponding to the target theme set are then integrated to obtain the target theme features corresponding to the target theme set. Based on the abnormal characteristics of the target area and the characteristics of the target set, the confidence coefficient between the target landslide location description data area and the target subject set is calculated.

11. The landslide failure prediction method according to claim 5, characterized in that, The target key indicator features include all abnormal features corresponding to the landslide location description data of the example, and the target theme features include all theme features corresponding to the abnormal landslide examples; the step of performing feature discrimination on the candidate key indicator features and the candidate theme features on the target identification character to obtain the target feature discrimination result corresponding to the target identification character includes: When the target identification character is a potential identification character, the potential landslide hazard data corresponding to the example landslide location description data is obtained, and the potential landslide hazard data is feature extracted to obtain potential landslide hazard features. The potential identification character is the identification character between the potential landslide hazard data, the example landslide location description data, and the abnormal landslide example. The potential landslide hazard characteristics, all abnormal characteristics, and all theme characteristics are used to identify the potential distinguishing characters to obtain the potential feature identification results corresponding to the potential distinguishing characters, and the potential feature identification results are used as the target feature identification results.

12. The landslide failure prediction method according to claim 11, characterized in that, The step of performing feature identification on the potential landslide hazard characteristics, all abnormal characteristics, and all thematic characteristics on the potential identification character to obtain the potential feature identification result corresponding to the potential identification character includes: Based on the characteristics of the potential landslide hazards and all the abnormal characteristics, calculate the degree of abnormal correlation between the potential landslide hazard data and the example landslide location description data; Based on the characteristics of the potential landslide hazards and all thematic characteristics, the thematic correlation degree between the potential landslide hazard data and the abnormal landslide themes is calculated; Based on the abnormal correlation degree and the topic correlation degree, the potential feature identification result corresponding to the potential identification character is generated.

13. The landslide failure prediction method according to claim 3, characterized in that, The distinguishing characters include all distinguishing characters and region distinguishing characters; the process of converging the landslide location description data parsing network based on the feature identification results, key indicator features, and theme features to obtain the landslide location description data parsing network includes: Select all landslide feature identification results corresponding to all identification characters and regional feature identification results corresponding to the regional identification characters from the feature identification results respectively; Based on all the landslide feature identification results, the example inference performance evaluation between the example landslide location description data and the abnormal landslide examples is calculated. The example inference performance evaluation characterizes the shared inference performance evaluation between the example landslide location description data and the abnormal landslide examples. Based on the regional feature identification results, as well as the regional abnormal features and set features corresponding to the regional identification characters, the regional matching inference performance evaluation corresponding to the landslide location description data region is determined. The regional matching inference performance evaluation characterizes the matching inference performance evaluation between the landslide location description data region and the theme set. Based on the key indicator features and theme features, the landslide location description data of the example is predicted to obtain the theme prediction probability, and the theme prediction probability is projected to obtain the prediction inference performance evaluation data set corresponding to the landslide location description data of the example. Based on the example extrapolation performance evaluation, the regional matching extrapolation performance evaluation, and the prediction extrapolation performance evaluation data set, the specified landslide location description data parsing network is converged to obtain the landslide location description data parsing network.

14. The landslide failure prediction method according to claim 13, characterized in that, The step of determining the regional matching and inference performance evaluation corresponding to the landslide location description data region based on the regional feature identification results, as well as the regional abnormal features and set features corresponding to the regional identification characters, includes: Based on the region feature discrimination results, determine the region training label corresponding to the region discrimination character; Based on the abnormal characteristics and set characteristics of the region, calculate the candidate confidence coefficient between the landslide location description data region and the topic set; Calculate the target inference performance evaluation between the region training label and the candidate confidence coefficient, and use the target inference performance evaluation as the region matching inference performance evaluation corresponding to the landslide location description data region.

15. The landslide failure prediction method according to claim 14, characterized in that, The regional feature identification result includes the confidence coefficient between the target landslide location description data region of the example landslide location description data and the target topic set of abnormal landslide examples; the step of determining the regional training label corresponding to the regional identification character based on the regional feature identification result includes: Select a reference abnormal region from the abnormal region, and select a reference topic set from the topic set; The reference abnormal region and the reference topic set are combined to obtain at least one reference data set, the reference data set including at least one target reference abnormal region and at least one target reference topic set corresponding to the target reference abnormal region; The target confidence coefficient corresponding to the reference data set is selected from the regional feature identification results, and the target confidence coefficient is integrated to obtain the target integrated confidence coefficient. Based on the target integration confidence coefficient, determine the region training label corresponding to the region identification character; the step of calculating the candidate confidence coefficient between the landslide location description data region and the topic set based on the region abnormality features and set features includes: extracting the current landslide location description data region and the current topic set from the target data set, and calculating the candidate confidence coefficient between the current landslide location description data region and the current topic set based on the region abnormality features corresponding to the current abnormal region and the set features corresponding to the current topic set.

16. The landslide failure prediction method according to claim 13, characterized in that, The distinguishing characters also include potential distinguishing characters; the convergence of the specified landslide location description data parsing network based on the exemplary inference performance evaluation, the regional matching inference performance evaluation, and the prediction inference performance evaluation data set to obtain the landslide location description data parsing network includes: The potential feature identification results corresponding to the potential identification characters are selected from the feature identification results. The potential feature identification results include the abnormal correlation between the potential landslide hazard data and the example landslide location description data corresponding to the potential identification characters, and the theme correlation between the potential landslide hazard data and the abnormal landslide theme. Based on the latent feature discrimination results, the latent training labels corresponding to the latent discriminant characters are determined; Based on the potential training labels, the potential landslide hazard characteristics corresponding to the potential landslide hazard data, the key indicator characteristics, and the theme characteristics, the potential inference performance evaluation corresponding to the example landslide location description data is calculated. The potential inference performance evaluation characterizes the correlation inference performance evaluation between the potential landslide hazard data, the example landslide location description data, and the abnormal landslide examples. Based on the datasets of the potential extrapolation performance evaluation, the example extrapolation performance evaluation, the regional matching extrapolation performance evaluation, and the prediction extrapolation performance evaluation, the parsing network for the specified landslide location description data is converged to obtain the parsing network for the landslide location description data.

17. The landslide failure prediction method according to claim 1, characterized in that, Based on the key indicators of local landslide characteristics and all landslide characteristics, feature extraction is performed on the reference theme results to obtain important features, including: Feature extraction is performed on the reference topic results to obtain initial reference topic features; The key features are obtained by integrating the local landslide characteristics of the key indicators, the total landslide characteristics of the key indicators, and the initial reference theme characteristics.

18. The landslide failure prediction method according to claim 1, characterized in that, The step of generating landslide failure projection results corresponding to the landslide location description data based on the target theme features and the reference theme results includes: Based on the target theme characteristics, the landslide location description data is used to predict the theme, thereby obtaining the target theme; The reference topic results and target topics are integrated to obtain an integrated topic, and the integrated topic is used as the reference topic result. The process of extracting features from the reference topic results based on the local landslide features and all landslide features of the key indicators is repeated until the prediction end requirement is met, resulting in multiple target topics. The multiple target themes are integrated to generate the landslide failure simulation results.

19. A landslide failure prediction system, characterized in that, The method includes a processor and a memory that communicate with each other, the processor being configured to read a computer program from the memory and execute it to implement the method of any one of claims 1-18.

Images