A regional landslide disaster meteorological risk early warning method, device and storage medium

Abstract

The application discloses a regional landslide disaster meteorological risk early warning method and device and a storage medium, and belongs to the technical field of geological disaster early warning. The regional landslide disaster meteorological risk early warning method comprises the following steps: obtaining the disaster gestation sensitivity grade of each evaluation unit in a target region, and the steps are as follows: S1, according to the target region, a reference region is selected, and the target region is located in the reference region, or the spatial similarity of the target region and the reference region is higher than a first threshold value, the landslide sample of the reference region is greater than a second threshold value, and the rainfall record of the reference region is greater than a third threshold value; S2, a disaster gestation sensitivity evaluation model of the target region is constructed based on a disaster gestation sensitivity evaluation model of the reference region; and S3, the disaster gestation sensitivity evaluation model of the target region is used to divide the disaster gestation sensitivity grade of each evaluation unit in the target region. In the case that the region lacks historical landslide disaster event records, the application can still realize fine regional landslide disaster meteorological risk early warning and prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological disaster early warning, and in particular to a regional landslide disaster meteorological risk early warning method, device and storage medium. BACKGROUND

[0002] The occurrence of landslide geological disasters is related to special topography and geological conditions, and rainfall is an important factor inducing disasters. Regional landslide geological disaster meteorological risk early warning is an important technical means for changing from passive disaster relief to active defense. The most core is to construct a refined regional landslide geological disaster meteorological risk early warning model according to the regional geological environment background and disaster-pregnant characteristics.

[0003] In the past, landslide disaster meteorological risk early warning models were constructed for larger regions such as provinces and cities. With the improvement of disaster prevention and reduction requirements, it is necessary to construct refined early warning and forecasting models for smaller scales such as districts, towns and the like. With the refinement of regional scale, due to the reasons that a certain region may not have carried out large-scale field geological survey, the number of historical landslide samples is small, the time of historical landslide occurrence is unknown, and the like, the disaster event samples that can be used to construct the model in a certain region are scarce, and then it is impossible to construct a refined regional landslide disaster meteorological risk early warning model, and it is even more impossible to wait until the landslide records accumulate to a certain degree to carry out related landslide disaster early warning and forecasting work. SUMMARY

[0004] The present application aims to provide a regional landslide disaster meteorological risk early warning method, device and storage medium, which can still achieve refined regional landslide disaster meteorological risk early warning and forecasting in the case that a region lacks historical landslide disaster event records.

[0005] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0006] A regional landslide disaster meteorological risk early warning method comprises the following steps:

[0007] A three-dimensional criterion matrix about disaster-pregnant sensitivity level, rainfall disaster warning level and short-term predicted rainfall is constructed in a three-dimensional coordinate system;

[0008] The disaster-pregnant sensitivity level, rainfall disaster warning level and short-term predicted rainfall of each evaluation unit in the target region are obtained, and the landslide disaster meteorological risk early warning level of each evaluation unit is obtained according to the constructed three-dimensional criterion matrix;

[0009] The landslide disaster meteorological risk early warning levels of each evaluation unit in the target region are combined to obtain the landslide disaster meteorological risk early warning level of the target region;

[0010] The steps of obtaining the disaster gestation sensitivity level of each evaluation unit in the target region are as follows:

[0011] S1, according to the target region, a reference region is selected, and the target region is located in the reference region, or the spatial similarity of the target region and the reference region is higher than a first threshold value, the landslide sample of the reference region is greater than a second threshold value, and the rainfall record of the reference region is greater than a third threshold value;

[0012] S2, based on the disaster gestation sensitivity evaluation model of the reference region, a disaster gestation sensitivity evaluation model of the target region is constructed, and the steps are as follows:

[0013] S2.1, taking each evaluation unit in the reference region as an evaluation object, selecting historical disaster points and non-historical disaster points as positive samples and negative samples respectively, and selecting a plurality of basic environmental factors, an initial disaster gestation sensitivity evaluation model is constructed using a deep neural network, the initial disaster gestation sensitivity evaluation model is trained through the basic environmental factors and the positive and negative sample sets, the network model structure and network parameters after training are saved, and the disaster gestation sensitivity evaluation model of the reference region is obtained;

[0014] S2.2, obtaining the disaster gestation sensitivity evaluation model of the reference region, adding a geological environment regional feature adjuster before the output layer, and one of the input data of the geological environment regional feature adjuster is the spatial similarity of the target region and the reference region, taking each evaluation unit in the target region as an evaluation object, selecting historical disaster points and non-historical disaster points as positive samples and negative samples respectively, and selecting the same basic environmental factors as S2.1, the adjusted disaster gestation sensitivity evaluation model of the reference region is trained again, and during the training, the network parameters saved after training in S2.1 are frozen, and only the geological environment regional feature adjuster is trained, to obtain the disaster gestation sensitivity evaluation model of the target region;

[0015] S3, using the disaster gestation sensitivity evaluation model of the target region to divide the disaster gestation sensitivity level of each evaluation unit in the target region.

[0016] Further, the steps of obtaining the rainfall disaster warning level of each evaluation unit in the target region are as follows:

[0017] Collecting existing rainfall events and landslide disaster data in the target region, filtering rainfall events related to landslides, i.e. disaster-causing rainfall events, and the remaining are rainfall events unrelated to landslides, i.e. non-disaster-causing rainfall events;

[0018] The cumulative rainfall and rainfall duration of each non-disaster-causing rainfall event are counted and plotted in a logarithmic coordinate system, and a linear warning function of different warning levels is fitted according to the distribution of different non-disaster-causing rainfall events, and the fitting formula of the warning function is as follows:

[0019] E=αDβ +C

[0020] Wherein, E is the cumulative rainfall, D is the rainfall duration, and a, b, C are constants.

[0021] The cumulative rainfall and rainfall duration of each evaluation unit in the target area are entered into a logarithmic coordinate system, and the rainfall disaster warning level of each evaluation unit in the target area is obtained according to a linear warning function of different warning levels.

[0022] Further, the evaluation unit is a grid unit or a landslide unit.

[0023] Further, the calculation method of the spatial similarity of the target area and the reference area is as follows:

[0024] In the target area and the reference area, clustering analysis is performed to divide each area into multiple clusters corresponding to different spatial types.

[0025] The centroid vector of each cluster is obtained to form a geological environment feature matrix representing different areas.

[0026] The cosine similarity between the geological environment feature matrix of the target area and the geological environment feature matrix of the reference area is calculated as the spatial similarity therebetween.

[0027] Further, when a non-historical disaster point is selected as a negative sample, a random selection or a semi-supervised method is used.

[0028] Further, the basic environmental factors include elevation, slope, slope direction, plan curvature, profile curvature, terrain relief degree, rock-soil type, normalized vegetation index, normalized building index, total gully density, terrain humidity index, modified normalized difference water index, and highway density.

[0029] Further, before training the landslide disaster sensitivity evaluation model by using the basic environmental factors and the positive and negative sample set, one of the two basic environmental factors with a correlation greater than a fourth threshold value is removed.

[0030] Further, before training the landslide disaster sensitivity evaluation model by using the basic environmental factors and the positive and negative sample set, the basic environmental factors are quantified by using a frequency ratio or an evidence weight method.

[0031] A regional landslide disaster meteorological risk warning device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the steps of the above-mentioned regional landslide disaster meteorological risk warning method.

[0032] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the regional landslide disaster meteorological risk early warning method.

[0033] The present application has the following advantages:

[0034] The present application discloses a regional landslide disaster meteorological risk early warning method, device and storage medium. In the case of lack of historical landslide disaster event records in a region, a disaster-pregnant sensitivity evaluation model of a target region is constructed based on a reference region through an intelligent algorithm, and fine regional landslide disaster meteorological risk early warning and prediction can still be realized, thereby guaranteeing the safety of people's lives and property and providing effective support for geological disaster prevention and mitigation. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 A flowchart for obtaining the disaster-pregnant sensitivity level of each evaluation unit in the target region in the regional landslide disaster meteorological risk early warning method of the present application;

[0036] Figure 2 A schematic diagram for constructing a disaster-pregnant sensitivity evaluation model of a target region based on a reference region in the regional landslide disaster meteorological risk early warning method of the present application;

[0037] Figure 3 A flowchart for the regional landslide disaster meteorological risk early warning method of the present application;

[0038] Figure 4 A schematic diagram of a three-dimensional criterion matrix in the regional landslide disaster meteorological risk early warning method of the present application;

[0039] Figure 5 A flowchart for obtaining the rainfall disaster-causing early warning level of each evaluation unit in the target region in the regional landslide disaster meteorological risk early warning method of the present application;

[0040] Figure 6 A logarithmic coordinate system (without filtering rainfall events related to landslide disasters) for obtaining the rainfall disaster-causing early warning level of each evaluation unit in the target region in the regional landslide disaster meteorological risk early warning method of the present application;

[0041] Figure 7 A logarithmic coordinate system (with filtering rainfall events related to landslide disasters) for obtaining the rainfall disaster-causing early warning level of each evaluation unit in the target region in the regional landslide disaster meteorological risk early warning method of the present application. DETAILED DESCRIPTION

[0042] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0043] Referring to Figures 3-4 As shown in the figure, a regional landslide disaster meteorological risk early warning method comprises the following steps:

[0044] A three-dimensional criterion matrix is constructed in a three-dimensional coordinate system with respect to three dimensions of disaster-pregnant sensitivity level, rainfall disaster-causing early warning level and short-term predicted rainfall;

[0045] The disaster-pregnant sensitivity level, rainfall disaster-causing early warning level and short-term predicted rainfall of each evaluation unit in the target region are obtained, and the landslide disaster meteorological risk early warning level of each evaluation unit is obtained according to the constructed three-dimensional criterion matrix;

[0046] The landslide disaster meteorological risk early warning levels of each evaluation unit in the target region are combined to obtain the landslide disaster meteorological risk early warning level of the target region.

[0047] In this embodiment, the evaluation unit is exemplarily a grid unit or a landslide unit.

[0048] Referring to Figures 1-2 In the above technical solution, specifically, the disaster-pregnant sensitivity level of each evaluation unit in the target region is obtained as follows:

[0049] S1, according to the target region, a reference region is selected, and the target region is located in the reference region, or the spatial similarity of the target region and the reference region is higher than a first threshold value, the landslide sample of the reference region is greater than a second threshold value, and the rainfall record of the reference region is greater than a third threshold value, that is, one of the above two conditions is met;

[0050] It should be noted that the reference region can be a single region or a collection of multiple similar regions, and the first threshold value, the second threshold value and the third threshold value are set according to experience;

[0051] S2, based on the disaster-pregnant sensitivity evaluation model of the reference region, a disaster-pregnant sensitivity evaluation model of the target region is constructed, and the steps are as follows:

[0052] S2.1, select historical disaster points and non-historical disaster points as positive samples and negative samples respectively, and select several basic environmental factors, use a deep neural network to build an initial disaster-pregnant sensitivity evaluation model, the network structure can be RNN, CNN, MLP, etc., and attention mechanism can also be added according to the sample size, the number of network layers is not less than 3, train the initial disaster-pregnant sensitivity evaluation model through the basic environmental factors and the positive and negative sample sets, save the trained network model structure and network parameters, and obtain the disaster-pregnant sensitivity evaluation model of the reference area;

[0053] S2.2, S2.2, obtain the disaster-pregnant sensitivity evaluation model of the reference area, add a geological environment regional feature adjuster before the output layer, and one of the input data of the geological environment regional feature adjuster is the spatial similarity between the target area and the reference area, take each evaluation unit in the target area as the evaluation object, select historical disaster points and non-historical disaster points as positive samples and negative samples respectively, and select the same basic environmental factors as S2.1, retrain the adjusted disaster-pregnant sensitivity evaluation model of the reference area, and during training, freeze the trained network parameters saved by S2.1, and only train the geological environment regional feature adjuster, to obtain the disaster-pregnant sensitivity evaluation model of the target area;

[0054] Figure 2 In the formula, h1, h2, h3, h4, and h5 are input data of the geological environment regional feature adjuster, s1, s2, and s3 are intermediate neurons of the geological environment regional feature adjuster, and u1, u2, u3, and u4 are output data of the geological environment regional feature adjuster.

[0055] S3, use the disaster-pregnant sensitivity evaluation model of the target area to divide the disaster-pregnant sensitivity level of each evaluation unit in the target area.

[0056] The calculation method of the spatial similarity between the target area and the reference area is as follows:

[0057] In the target area and the reference area, clustering analysis is performed, each area is divided into multiple clusters, corresponding to different spatial types, for example, the first area is divided into three spatial types, and the second area is divided into four spatial types;

[0058] The centroid vector of each cluster is obtained to form a geological environment feature matrix representing different areas, for example. If 6 basic environmental factors are selected, the target area has three clusters, and the reference area has four clusters, then the feature matrix of the target area is a 3·6-dimensional feature matrix A, and the feature matrix of the reference area is a 4·6-dimensional feature matrix B.

[0059] The cosine similarity between the geological environment feature matrix A of the target area and the geological environment feature matrix B of the reference area is calculated as the spatial similarity between the two.

[0060] The basic environment factors include elevation, slope, slope direction, plan curvature, profile curvature, terrain relief degree, rock-soil type, normalized vegetation index, normalized building index, total gully density, terrain humidity index, modified normalized difference water index, and highway density.

[0061] More specifically, before training the landslide disaster-breed sensitivity evaluation model based on the basic environment factors and the positive and negative sample set, one of the two basic environment factors with a correlation greater than the fourth threshold is removed.

[0062] In other words, the correlation of the selected basic environment factors should be detected, and the basic environment factors with high correlation should be removed to avoid the mutual interference and repetition of information in the basic environment factors affecting the prediction accuracy of the disaster-breed sensitivity model. For example, when the correlation between two basic environment factors is greater than 0.5, only one is retained, and the one with lower correlation with other factors is generally retained.

[0063] More specifically, before training the landslide disaster-breed sensitivity evaluation model based on the basic environment factors and the positive and negative sample set, the basic environment factors are quantified by the frequency ratio or evidence weight method.

[0064] In this embodiment, the non-historical disaster points are selected as negative samples by random selection or semi-supervised method.

[0065] Please refer to Figures 5-7 In the above technical solution, specifically, the rainfall disaster warning level of each evaluation unit in the target area is obtained, and the steps are as follows:

[0066] Collect the existing rainfall events and landslide disaster data in the target area, filter the rainfall events related to landslide disasters, i.e. disaster-causing rainfall events, and the remaining ones are rainfall events unrelated to landslides, i.e. non-disaster-causing rainfall events;

[0067] The cumulative rainfall and rainfall duration of each non-disaster-causing rainfall event are counted and plotted in a logarithmic coordinate system, and linear warning functions of different warning levels are fitted according to the distribution of different non-disaster-causing rainfall events. The fitting formula of the warning function is as follows:

[0068] E = aD β + C

[0069] Where E is the cumulative rainfall, D is the rainfall duration, and a, b, and C are constants.

[0070] The accumulated rainfall and rainfall age of each evaluation unit of the target area are entered into a logarithmic coordinate system, and a rainfall disaster warning level of each evaluation unit in the target area is obtained according to a linear warning function of different warning levels.

[0071] It should be noted that at present, the rainfall threshold value for inducing landslides is generally determined by using a statistical method, historical landslide disasters and rainfall data at that time are collected, the limit of the rainfall value of the landslide disaster is counted, and thus the rainfall threshold value for inducing landslides is determined, and this kind of method needs sufficient landslide disaster records as support, but in the area where landslide disaster records are accumulated poorly or the area where systematic rainfall monitoring time is short, the quality and quantity of historical data are affected, and it is impossible to carry out related landslide disaster warning and prediction work until the landslide records accumulate to a certain degree. In view of this, the rainfall disaster warning level of each evaluation unit in the target area is also improved in the present application.

[0072] As an application, a regional landslide disaster meteorological risk warning device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above-mentioned regional landslide disaster meteorological risk warning method when executing the computer program.

[0073] As an application, a computer readable storage medium stores a computer program, and the computer program implements the steps of the above-mentioned regional landslide disaster meteorological risk warning method when executed by a processor.

[0074] In summary, in the case that the region lacks historical landslide disaster event records, the present application can still realize fine regional landslide disaster meteorological risk warning and prediction by intelligent algorithm, based on the reference region to construct the disaster-pregnant sensitivity evaluation model of the target region, which guarantees the safety of people's life and property and provides effective support for geological disaster prevention and mitigation work.

[0075] The present application is not limited to the above-mentioned specific embodiments, and those skilled in the art can use other various specific embodiments to implement the present application according to the disclosed content of the present application, therefore, any design using the design structure and idea of the present application, making some simple changes or modifications, falls within the protection scope of the present application.

Claims

1. A regional landslide disaster meteorological risk early warning method, characterized in that: The method comprises the following steps: a three-dimensional criterion matrix is constructed in a three-dimensional coordinate system with respect to three dimensions of disaster-pregnant sensitivity level, rainfall disaster warning level and short-term predicted rainfall; the disaster-pregnant sensitivity level, the rainfall disaster warning level and the short-term predicted rainfall of each evaluation unit in the target region are obtained, and the landslide disaster meteorological risk warning level of each evaluation unit is obtained according to the constructed three-dimensional criterion matrix; the landslide disaster meteorological risk warning levels of each evaluation unit in the target region are combined to obtain the landslide disaster meteorological risk warning level of the target region; the disaster-pregnant sensitivity level of each evaluation unit in the target region is obtained, and the steps are as follows: S1, according to the target region, a reference region is selected, and the target region is located in the reference region, or the spatial similarity of the target region and the reference region is higher than a first threshold value, the landslide sample of the reference region is greater than a second threshold value, and the rainfall record of the reference region is greater than a third threshold value; S2, a disaster-pregnant sensitivity evaluation model of the target region is constructed based on the disaster-pregnant sensitivity evaluation model of the reference region, and the steps are as follows: S2.1, taking each evaluation unit in the reference region as an evaluation object, historical disaster points and non-historical disaster points are selected as positive samples and negative samples respectively, and a plurality of basic environmental factors are selected, an initial disaster-pregnant sensitivity evaluation model is constructed using a deep neural network, the initial disaster-pregnant sensitivity evaluation model is trained through the basic environmental factors and the positive and negative sample sets, the network model structure and network parameters after training are saved, and the disaster-pregnant sensitivity evaluation model of the reference region is obtained; S2.2, the disaster-pregnant sensitivity evaluation model of the reference region is obtained, a geological environment regional feature adjuster is added before the output layer, one of the input data of the geological environment regional feature adjuster is the spatial similarity of the target region and the reference region, each evaluation unit in the target region is taken as an evaluation object, historical disaster points and non-historical disaster points are selected as positive samples and negative samples respectively, and the same basic environmental factors as S2.1 are selected, the adjusted disaster-pregnant sensitivity evaluation model of the reference region is trained again, and during the training, the network parameters saved after training in S2.1 are frozen, and only the geological environment regional feature adjuster is trained, thereby obtaining the disaster-pregnant sensitivity evaluation model of the target region; S3, the disaster-pregnant sensitivity evaluation model of the target region is used to divide the disaster-pregnant sensitivity level of each evaluation unit in the target region. 2.The regional landslide disaster meteorological risk early warning method according to claim 1, characterized in that: The steps for obtaining the rainfall disaster warning level of each evaluation unit in the target region are as follows: collecting existing rainfall events and landslide disaster data in the target region, filtering disaster-causing rainfall events, and the remaining are non-disaster-causing rainfall events; the cumulative rainfall and rainfall duration of each non-disaster-causing rainfall event are counted and plotted in a logarithmic coordinate system, and linear warning functions of different warning levels are fitted according to the distribution of different non-disaster-causing rainfall events, and the fitting formula of the warning function is as follows: E = aD β +C wherein, E is the cumulative rainfall, D is the rainfall duration, and α, β and C are constants; the cumulative rainfall and rainfall duration of each evaluation unit in the target region are entered into the logarithmic coordinate system, and the rainfall disaster warning level of each evaluation unit in the target region is obtained according to the linear warning function of different warning levels. 3.The regional landslide disaster meteorological risk early warning method according to claim 1, characterized in that: The evaluation unit is a grid unit or a landslide unit.

4. The regional landslide disaster meteorological risk early warning method according to claim 1, characterized in that: The calculation method of the spatial similarity of the target region and the reference region is as follows: In the target region and the reference region, clustering analysis is performed to divide each region into multiple clusters corresponding to different spatial types. The centroid vector of each cluster is obtained to form a geological environment feature matrix representing different regions. The cosine similarity between the geological environment feature matrix of the target region and the geological environment feature matrix of the reference region is calculated as the spatial similarity between the two.

5. The regional landslide disaster meteorological risk early warning method according to claim 1, characterized in that: When non-historical disaster points are selected as negative samples, random selection or semi-supervised method is used. 6.The regional landslide disaster meteorological risk warning method according to claim 1, characterized in that: The basic environmental factors include elevation, slope, slope direction, plan curvature, profile curvature, terrain relief, rock-soil type, normalized vegetation index, normalized building index, total gully density, terrain humidity index, modified normalized difference water index, and highway density.

7. The regional landslide disaster meteorological risk early warning method according to claim 1 or 6, characterized in that: Before training the landslide disaster sensitivity evaluation model based on the basic environmental factors and the positive and negative sample set, one of the two basic environmental factors with a correlation greater than the fourth threshold is removed. 8.The regional landslide disaster meteorological risk early warning method according to claim 1 or 6, characterized in that: Before training the landslide disaster sensitivity evaluation model based on the basic environmental factors and the positive and negative sample set, the basic environmental factors are quantified by frequency ratio or evidence weight. 9.A regional landslide disaster meteorological risk early warning device, characterized in that: The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the steps of the regional landslide disaster meteorological risk early warning method of any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the steps of the regional landslide disaster meteorological risk early warning method of any one of claims 1 to 8.

Images