Regional landslide hazard assessment prediction method, device and equipment based on CNN and CA

Abstract

The present application discloses a regional landslide hazard assessment prediction method, device and equipment based on CNN and CA, and relates to the technical field of landslide hazard assessment and prediction. The method is to first propose a regional landslide hazard static assessment method based on landslide influencing factor data and convolutional neural network, then construct a cellular automaton, and based on the static evaluation result data, deeply analyze the temporal causal relationship and spatial correlation of the landslide hazard state in the landslide evolution process, and construct an evolution rule integrating temporal and spatial characteristics based on the densely connected convolutional neural network DenseNet, and finally use the evolution rule in the cellular automaton to predict the landslide hazard level value to obtain the regional landslide hazard dynamic prediction result, so as to overcome the limitations of a single evolution rule, improve the prediction accuracy, provide scientific support for landslide disaster prevention, and thus easily meet the actual needs of disaster prevention and mitigation, and have broad application prospects.

Description

Technical Field

[0001] The invention belongs to the technical field of landslide hazard assessment and prediction, and in particular relates to a method, device and equipment for regional landslide hazard assessment and prediction based on CNN and CA. Background Art

[0002] Landslide geological disasters have a huge impact on social economy and human life safety. Landslide hazard assessment is one of the important basic tasks for landslide disaster prevention. Due to the great uncertainty of the time, location and scale of landslides, the traditional single landslide hazard assessment scheme has certain limitations in preventing and reducing disaster losses. Therefore, it is particularly important to carry out regional landslide hazard assessment and prediction.

[0003] At present, most of the existing regional landslide hazard prediction schemes only consider the spatial distribution or temporal changes of landslides; however, the occurrence of landslides is affected by both time and space, and a single evolution rule has certain limitations in expressing the dynamic evolution process of the landslide system. Therefore, how to propose a new regional landslide hazard evaluation and prediction scheme that integrates the temporal and spatial evolution rules in order to accurately predict the landslide hazard is a topic that technicians in this field urgently need to study. Summary of the invention

[0004] The purpose of the present invention is to provide a regional landslide hazard assessment and prediction method, device, computer equipment, computer readable storage medium and computer program product based on CNN and CA, so as to solve the problem that the existing regional landslide hazard prediction scheme has certain limitations in expressing the dynamic evolution process of the landslide system due to the use of a single evolution rule, thereby resulting in the accuracy of the landslide hazard prediction results needs to be improved.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] Firstly, a regional landslide hazard assessment and prediction method based on CNN and CA is provided, including:

[0007] Collect the target area in the current recent continuous The landslide point data of historical unit periods and the basic data of the target area, among which, represents a positive integer greater than or equal to ten;

[0008] According to the landslide point data, the most recent continuous The historical unit periods correspond one to one and have the same number of elements. A set of landslide points and non-landslide points, among which, Indicates less than or equal to A positive integer, the landslide point and non-landslide point set contains Landslide points and Non-landslide points, and Respectively represent positive integers;

[0009] According to the basic data, a plurality of gridded influence factor distribution maps corresponding to a plurality of static factors affecting landslide hazard assessment in the target area and a plurality of gridded influence factor distribution time series map groups corresponding to a plurality of dynamic factors affecting landslide hazard assessment in the target area are processed and generated, wherein the gridded influence factor distribution time series map groups contain corresponding dynamic factors and correspond to the current most recent continuous One-to-one correspondence between historical unit periods A grid-based impact factor distribution map;

[0010] The target area is spatially divided to obtain a regular square grid, and for each grid unit in the regular square grid, according to the multiple gridded impact factor distribution graphs, the multiple gridded impact factor distribution time series graph groups and the A set of landslide points and non-landslide points is obtained based on the convolutional neural network CNN, which corresponds to the current most recent continuous One-to-one correspondence between historical unit periods a landslide hazard level value, wherein the landslide hazard level value refers to a label value corresponding to a low landslide hazard level, a medium landslide hazard level, or a high landslide hazard level;

[0011] The following cellular automaton CA is constructed according to the regular square grid: all cells corresponding to all grid cells in the regular square grid, the states and evolution rules corresponding to the cells are defined in the entire cellular space, and for each grid cell, the cellular neighbor relationship of the corresponding cell is made consistent with the corresponding grid cell neighbor relationship;

[0012] According to the landslide hazard level values, based on the state transition of the landslide hazard level values ​​of each grid unit in two adjacent historical units and the actual spatial position of each grid unit, and based on the densely connected convolutional neural network DenseNet, an evolution rule integrating time characteristics and spatial characteristics is obtained;

[0013] The cellular automaton CA is started as follows: The initial state for evolution prediction is extracted based on the landslide hazard level value corresponding to a certain pair of adjacent historical unit periods in a historical unit period, and then the state of all the cells is updated based on the evolution rule integrating the time characteristics and the space characteristics to obtain the predicted landslide hazard level value of each grid unit in the next unit period after the certain pair of adjacent historical unit periods, and the state of all the cells is continuously updated iteratively with the certain pair of adjacent historical unit periods and the next unit period as an iteration cycle until the predicted landslide hazard level value of each grid unit in the target unit period is obtained, wherein the target unit period is a unit period located after the certain pair of adjacent historical unit periods in time sequence;

[0014] The predicted landslide hazard level values ​​of the respective grid units in the target unit period are summarized to obtain a dynamic prediction result of regional landslide hazard of the target area in the target unit period.

[0015] Based on the above invention content, a new scheme for regional landslide hazard evaluation and prediction based on convolutional neural network, densely connected convolutional neural network and cellular automaton is provided, that is, firstly based on landslide influencing factor data, a regional landslide hazard static evaluation method is proposed in combination with convolutional neural network, then a cellular automaton is constructed, and based on the static evaluation result data, the temporal causal relationship and spatial correlation of the landslide hazard state in the landslide evolution process are deeply analyzed, and based on the densely connected convolutional neural network DenseNet, an evolution rule integrating temporal characteristics and spatial characteristics is constructed, and finally the evolution rule is used in the cellular automaton to predict the landslide hazard level value to obtain the regional landslide hazard dynamic prediction result, which can overcome the limitations of a single evolution rule, improve the prediction accuracy, provide scientific support for landslide disaster prevention, and then easily meet the actual needs of disaster prevention and mitigation, have broad application prospects and promotion value, and are convenient for practical application and promotion.

[0016] In a possible design, based on the landslide point data, the most recent continuous The historical unit periods correspond one to one and have the same number of elements. A set of landslide points and non-landslide points, including:

[0017] According to the landslide point data, it is determined that the target area is the most recent continuous From the historical unit period closest to the present Historical unit period to the most recent one Historical unit period Landslide points, among which represents a positive integer greater than or equal to four and less than or equal to eight, Indicates less than or equal to A positive integer, represents a positive integer;

[0018] From the around the landslide point and at least Randomly select from the area outside the point at a distance of meters Points as non-landslide points, among which represents a positive integer greater than or equal to four hundred and less than or equal to six hundred, represents a positive integer, and The sum of is a constant;

[0019] The landslide points and The non-landslide point is added to the closest point to the current The landslide points and non-landslide points corresponding to each historical unit period are included in the set.

[0020] In a possible design, the basic data contains digital elevation model data, geological map data, road network data, water system data, remote sensing image data, earthquake point data, rainfall distribution data and / or population data;

[0021] The multiple static factors include elevation, slope and / or aspect obtained based on digital elevation model data processing, stratum lithology and / or distance to faults obtained based on geological map data processing, distance to road network obtained based on road network data processing, and / or distance to water system obtained based on water system data processing;

[0022] The multiple dynamic factors include land use type and / or vegetation coverage obtained based on remote sensing image data processing, distance from the epicenter obtained based on earthquake point data processing, rainfall obtained based on rainfall distribution data processing and / or population density obtained based on population data processing.

[0023] In a possible design, for each grid unit in the regular square grid, according to the plurality of gridded impact factor distribution graphs, the plurality of gridded impact factor distribution time series graph groups and the A set of landslide points and non-landslide points is obtained based on the convolutional neural network CNN, which corresponds to the current most recent continuous One-to-one correspondence between historical unit periods The landslide hazard level values ​​include:

[0024] For each grid unit in the regular square grid, according to the position of the corresponding grid in the multiple gridded impact factor distribution maps, a plurality of static factor values ​​corresponding to the multiple static factors are extracted from the multiple gridded impact factor distribution maps, and according to the position of the corresponding grid in the multiple gridded impact factor distribution time series diagram groups, a plurality of dynamic factor value groups corresponding to the multiple dynamic factors are extracted from the multiple gridded impact factor distribution time series diagram groups, wherein the dynamic factor value groups contain the corresponding dynamic factors and correspond to the current most recent continuous One-to-one correspondence between historical unit periods dynamic factor values, and also according to the corresponding grid and the The position matching result of the landslide point and the landslide point or non-landslide point in the non-landslide point set is determined to correspond to and be consistent with the current most recent continuous One-to-one correspondence between historical unit periods Landslide tags;

[0025] For each grid cell, the corresponding and most recent continuous One-to-one correspondence between historical unit periods The first sample data will correspond to and be consistent with the current most recent continuous All dynamic factor values ​​corresponding to any historical unit period in the historical unit periods and the corresponding multiple static factor values ​​are used as first model input items, and the corresponding landslide label corresponding to any historical unit period is used as a first model output item, and then the first model input item and the first model output item are used as first sample data corresponding to any historical unit period;

[0026] Dividing all the first sample data into a training sample data set, a verification sample data set and a test sample data set, and applying the training sample data set and the verification sample data set to calibrate and verify the modeling of a first artificial intelligence model based on a convolutional neural network (CNN), to obtain a landslide probability estimation model, and then applying the test sample data set to perform model testing on the landslide probability estimation model, to obtain a landslide probability average value, a landslide probability standard deviation, a non-landslide probability average value and a non-landslide probability standard deviation, and finally using the value obtained by subtracting the landslide probability standard deviation from the landslide probability average value as a first dividing value between a high landslide risk level and a medium landslide risk level, and using the value obtained by adding the non-landslide probability standard deviation to the non-landslide probability average value as a second dividing value between a medium landslide risk level and a low landslide risk level;

[0027] For each grid unit, Each first model input item in the first sample data is respectively imported into the landslide probability estimation model, and the output is obtained by comparing the current most recent continuous One-to-one correspondence between historical unit periods Then, according to the comparison result of the landslide probability with the first demarcation value and the second demarcation value, the corresponding and most recent continuous One-to-one correspondence between historical unit periods A landslide hazard level value is a label value corresponding to a low landslide hazard level, a medium landslide hazard level or a high landslide hazard level.

[0028] In a possible design, according to the The landslide hazard level values ​​are obtained based on the densely connected convolutional neural network DenseNet, which integrates the evolution rules of time and space features, including:

[0029] For each grid cell in the regular square grid, obtain the corresponding The landslide hazard level value and the corresponding value of all adjacent grid cells Landslide hazard level value;

[0030] For each grid cell, the corresponding and most recent continuous One-to-one correspondence between historical unit periods The second sample data: first, for each cell in the corresponding cell and all adjacent grid cells of the corresponding cell, according to the corresponding and closest to the current The landslide hazard level value corresponding to the historical unit period and the corresponding and closest to the current The landslide hazard level value corresponding to the historical unit period is determined to determine the corresponding state change category value, and then according to the square matrix structure of the corresponding unit and all the adjacent grid units of the corresponding unit, the state change category values ​​of the respective units are formed into a two-dimensional array whose matrix structure is consistent with the square matrix structure, and the two-dimensional array is used as the second model input item, and the corresponding unit and the first unit closest to the current one are used as the second model input item. The landslide hazard level value corresponding to the historical unit period is used as the second model output item, and finally the second model input item and the second model output item are used as the corresponding and closest to the current The second sample data corresponding to the historical unit period;

[0031] All of the second sample data are used to perform model training on a second artificial intelligence model based on a densely connected convolutional neural network DenseNet to obtain evolution rules that integrate temporal features and spatial features.

[0032] In one possible design, the densely connected convolutional neural network DenseNet includes an input layer, a padding layer, a convolution layer, a first normalization layer, a rectified linear unit layer, a first dense block, a first transition layer, a second dense block, a second transition layer, a third dense block, a second normalization layer, a global pooling layer, a fully connected layer and an output layer connected in sequence;

[0033] The padding layer is used to widen the input size so that the padding width of all inputs is consistent;

[0034] The rectified linear unit layer is used to adopt a linear and non-saturated activation function in the network;

[0035] The first dense block, the second dense block and the third dense block each include twelve convolution groups, and each two convolution groups in the twelve convolution groups are directly connected to achieve dense connection between networks;

[0036] The first transition layer and the second transition layer both include a convolution sublayer and a pooling sublayer, and are both used to extract features output by the corresponding previous dense block and compress the features;

[0037] The global pooling layer is used to compress the features into a one-dimensional array in a pooled form, and input the one-dimensional array into the fully connected layer;

[0038] The fully connected layer is used to convert the received features into the landslide hazard level value.

[0039] In the second aspect, a regional landslide hazard assessment and prediction device based on CNN and CA is provided, which includes an original data collection unit, a landslide data processing unit, a basic data processing unit, a landslide static evaluation unit, a cellular automaton construction unit, an evolution rule derivation unit, a landslide dynamic prediction unit and a prediction result summary unit;

[0040] The raw data collection unit is used to collect the target area's most recent continuous The landslide point data of historical unit periods and the basic data of the target area, among which, represents a positive integer greater than or equal to ten;

[0041] The landslide data processing unit is communicatively connected to the raw data collection unit and is used to process the landslide point data to obtain the most recent continuous The historical unit periods correspond one to one and have the same number of elements. A set of landslide points and non-landslide points, among which, Indicates less than or equal to A positive integer, the landslide point and non-landslide point set contains Landslide points and Non-landslide points, and Respectively represent positive integers;

[0042] The basic data processing unit is communicatively connected to the original data collection unit, and is used to process and generate a plurality of gridded influence factor distribution maps corresponding to a plurality of static factors affecting landslide hazard assessment in the target area and a plurality of gridded influence factor distribution time series map groups corresponding to a plurality of dynamic factors affecting landslide hazard assessment in the target area according to the basic data, wherein the gridded influence factor distribution time series map groups contain corresponding dynamic factors and correspond to the current most recent continuous One-to-one correspondence between historical unit periods A grid-based impact factor distribution map;

[0043] The landslide static evaluation unit is respectively connected to the landslide data processing unit and the basic data processing unit for spatially segmenting the target area to obtain a regular square grid, and for each grid unit in the regular square grid, the plurality of gridded impact factor distribution graphs, the plurality of gridded impact factor distribution time series graph groups and the A set of landslide points and non-landslide points is obtained based on the convolutional neural network CNN, which corresponds to the current most recent continuous One-to-one correspondence between historical unit periods a landslide hazard level value, wherein the landslide hazard level value refers to a label value corresponding to a low landslide hazard level, a medium landslide hazard level, or a high landslide hazard level;

[0044] The cellular automaton construction unit is communicatively connected to the landslide static evaluation unit, and is used to construct the following cellular automaton CA according to the regular square grid: defining all cells corresponding to all grid cells in the regular square grid, states and evolution rules corresponding to the cells in the entire cellular space, and for each grid cell, making the cellular neighbor relationship of the corresponding cell consistent with the corresponding grid cell neighbor relationship;

[0045] The evolution rule derivation unit is communicatively connected to the landslide static evaluation unit, and is used to determine the landslide static evaluation unit according to the landslide hazard level values, based on the state transition of the landslide hazard level values ​​of each grid unit in two adjacent historical units and the actual spatial position of each grid unit, and based on the densely connected convolutional neural network DenseNet, an evolution rule integrating time characteristics and spatial characteristics is obtained;

[0046] The landslide dynamic prediction unit is respectively connected to the cellular automaton construction unit and the evolution rule derivation unit for communication, and is used to start the cellular automaton CA in the following manner: The initial state for evolution prediction is extracted based on the landslide hazard level value corresponding to a certain pair of adjacent historical unit periods in a historical unit period, and then the state of all the cells is updated based on the evolution rule integrating the time characteristics and the space characteristics to obtain the predicted landslide hazard level value of each grid unit in the next unit period after the certain pair of adjacent historical unit periods, and the state of all the cells is continuously updated iteratively with the certain pair of adjacent historical unit periods and the next unit period as an iteration cycle until the predicted landslide hazard level value of each grid unit in the target unit period is obtained, wherein the target unit period is a unit period located after the certain pair of adjacent historical unit periods in time sequence;

[0047] The prediction result summary unit is communicatively connected to the landslide dynamic prediction unit, and is used to summarize the predicted landslide hazard level values ​​of each grid unit in the target unit period to obtain the regional landslide hazard dynamic prediction result of the target area in the target unit period.

[0048] In a third aspect, the present invention provides a computer device comprising a memory, a processor and a transceiver which are communicatively connected in sequence, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the regional landslide hazard assessment and prediction method as described in the first aspect or any possible design of the first aspect.

[0049] In a fourth aspect, the present invention provides a computer-readable storage medium having instructions stored thereon, which, when executed on a computer, executes the regional landslide hazard assessment and prediction method as described in the first aspect or any possible design of the first aspect.

[0050] In a fifth aspect, the present invention provides a computer program product, including a computer program or instructions, which, when executed by a computer, implements the regional landslide hazard assessment and prediction method as described in the first aspect or any possible design of the first aspect.

[0051] Beneficial effects of the above scheme:

[0052] The present invention creatively provides a new scheme for evaluating and predicting regional landslide hazard based on convolutional neural network, densely connected convolutional neural network and cellular automaton, that is, firstly based on landslide influencing factor data, a regional landslide hazard static evaluation method is proposed in combination with convolutional neural network, then a cellular automaton is constructed, and based on the static evaluation result data, the temporal causal relationship and spatial mutual relationship of the landslide hazard state in the landslide evolution process are deeply analyzed, and based on the densely connected convolutional neural network DenseNet, an evolution rule integrating temporal characteristics and spatial characteristics is constructed, and finally the evolution rule is used in the cellular automaton to predict the landslide hazard level value to obtain the regional landslide hazard dynamic prediction result, which can overcome the limitation of a single evolution rule, improve the prediction accuracy, provide scientific support for landslide disaster prevention, and then easily meet the actual needs of disaster prevention and mitigation, have broad application prospects and promotion value, and are convenient for practical application and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 A flow chart of a regional landslide hazard assessment and prediction method based on CNN and CA provided in an embodiment of the present application.

[0055] Figure 2 This is an example diagram of the static evaluation results of landslide hazards obtained over many years using Lushan County as an example provided in the embodiment of the present application, wherein: Figure 2 (a) shows an example of the static landslide hazard assessment results in 2015. Figure 2 (b) shows an example of the static evaluation results of landslide hazard in 2016. Figure 2 (c) shows an example of the static evaluation results of landslide hazard in 2017. Figure 2 (d) in the figure shows an example diagram of the static assessment results of landslide hazard in 2018.

[0056] Figure 3 A schematic diagram of the structure of a densely connected convolutional neural network provided in an embodiment of the present application.

[0057] Figure 4 A schematic diagram of a process for constructing evolution rules integrating time features provided in an embodiment of the present application.

[0058] Figure 5 A schematic diagram of a process for constructing evolution rules integrating spatial features provided in an embodiment of the present application.

[0059] Figure 6 A schematic diagram of a process for constructing evolution rules integrating temporal and spatial features provided in an embodiment of the present application.

[0060] Figure 7 This is an example diagram of the dynamic prediction results of landslide hazards obtained over many years using Lushan County as an example provided in the embodiment of the present application, wherein: Figure 7 (a) in the figure shows an example of the dynamic prediction results of landslide hazard in 2017. Figure 7 (b) shows an example of the dynamic prediction results of landslide hazard in 2018. Figure 7 (c) in the figure shows an example of the dynamic prediction results of landslide hazard in 2022. Figure 7 (d) in the figure shows an example diagram of the dynamic prediction results of landslide hazard in 2025.

[0061] Figure 8 A schematic diagram of the structure of a regional landslide hazard assessment and prediction device based on CNN and CA provided in an embodiment of the present application.

[0062] Fig. 9 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structures of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other embodiments can be obtained based on these embodiments without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0064] It should be understood that although the terms first and second, etc. may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another object. For example, a first object can be referred to as a second object, and similarly, a second object can be referred to as a first object without departing from the scope of the exemplary embodiments of the present invention.

[0065] It should be understood that the term "and / or" that may appear in this document is merely a description of the association relationship between associated objects, indicating that there may be three relationships. For example, A and / or B can indicate three situations: A exists alone, B exists alone, or A and B exist at the same time. For another example, A, B and / or C can indicate the existence of any one of A, B and C or any combination of them. The term " / and" that may appear in this document describes another type of association object relationship, indicating that there may be two relationships. For example, A / and B can indicate two situations: A exists alone or A and B exist at the same time. In addition, the character " / " that may appear in this document generally indicates that the previous and next associated objects are in an "or" relationship.

[0066] Example

[0067] like Figure 1 As shown, the regional landslide hazard assessment and prediction method based on CNN and CA provided in the first aspect of this embodiment can be, but is not limited to, executed by a computer device with certain computing resources, such as a cloud server, a personal computer (PC, which refers to a multi-purpose computer with a size, price and performance suitable for personal use; desktops, laptops, small laptops, tablets and ultrabooks are all personal computers), smart phones, personal digital assistants (PDA) or wearable devices. Figure 1 As shown, the regional landslide hazard assessment and prediction method may include, but is not limited to, the following steps S1 to S8.

[0068] S1. Collect the target area in the current recent continuous The landslide point data of historical unit periods and the basic data of the target area, among which, Represents a positive integer greater than or equal to ten.

[0069] In step S1, the target area is a geographical object for regional landslide hazard assessment and prediction, such as Lushan County, Sichuan Province (located on the western edge of the Sichuan Basin, sandwiched between the Qinghai-Tibet Plateau and the Sichuan Basin, with a long and narrow terrain from north to south, a relative height difference of 4732 meters, and generally presents a terrain form of low south and high north; the county covers an area of ​​1166.39 square kilometers, and as of 2019, the county has a population of nearly 110,000. Because it has been in the transition zone between the tectonic active zone and the stable block for a long time, the territory is mainly composed of tectonic landforms formed by folds and faults; Besides, except for the Cambrian and Carboniferous systems, all other systems are exposed; the rainfall is abundant and concentrated, with an average annual precipitation of 1,260 mm; affected by the topography, the precipitation in the area is mostly concentrated from June to September each year, and shows a decreasing trend from south to north in space; the water system in the area is well developed, and the tributaries, main streams, and ditches are distributed in a dendritic pattern, with a large river drop, rapid water flow, and concentrated flood season; and since the 7.0 magnitude earthquake in 2013, landslide disasters have occurred frequently, so the area can be selected for landslide hazard assessment and prediction research). The aforementioned unit period can be, but not limited to, years, quarters, or months, for example, the landslide point data of the target area in the current consecutive period of at least ten years (i.e., each year is abstracted as a unit period) are collected. The landslide point data specifically but not limited to the geographical coordinates of historical landslide points. Specifically, the basic data include but are not limited to digital elevation model data, geological map data, road network data, water system data, remote sensing image data, earthquake point data, rainfall distribution data and / or population data, etc., wherein, since the remote sensing image data, the earthquake point data, the rainfall distribution data and / or the population data have time-varying characteristics, they can be in the current most recent continuous In addition, the specific collection method of the landslide point data and the basic data is the existing conventional technology, which will not be repeated here.

[0070] S2. According to the landslide point data, the most recent continuous The historical unit periods correspond one to one and have the same number of elements. A set of landslide points and non-landslide points, among which, Indicates less than or equal to A positive integer, the landslide point and non-landslide point set contains Landslide points and Non-landslide points, and Represent positive integers respectively.

[0071] In step S2, due to the The landslide points and the non-landslide points have the same number of elements, so and Specifically, according to the landslide point data, the most recent continuous The historical unit periods correspond one to one and have the same number of elements. A landslide point and a non-landslide point set, including but not limited to the following steps S21-S23.

[0072] S21. According to the landslide point data, determine that the target area is the most recent continuous From the historical unit period closest to the present Historical unit period to the most recent one Historical unit period Landslide points, among which represents a positive integer greater than or equal to four and less than or equal to eight, Indicates less than or equal to A positive integer, Represents a positive integer.

[0073] In step S21, For example, 6, if the current most recent continuous If the historical unit period is the most recent ten consecutive years, all the landslide points in the target area in a certain year and the six years between them (i.e., seven consecutive years in total) can be regarded as the landslide points corresponding to the certain year, and .

[0074] S22. From the around the landslide point and at least Randomly select from the area outside the point at a distance of meters Points as non-landslide points, among which represents a positive integer greater than or equal to four hundred and less than or equal to six hundred, represents a positive integer, and The sum of is a constant.

[0075] In step S22, An example is 500.

[0076] S23. landslide points and The non-landslide point is added to the closest point to the current The landslide points and non-landslide points corresponding to each historical unit period are included in the set.

[0077] S3. Based on the basic data, a plurality of gridded influence factor distribution maps corresponding to a plurality of static factors affecting landslide hazard assessment in the target area are processed and generated, and a plurality of gridded influence factor distribution time series map groups corresponding to a plurality of dynamic factors affecting landslide hazard assessment in the target area are processed and generated, wherein the gridded influence factor distribution time series map groups contain corresponding dynamic factors and correspond to the current most recent continuous One-to-one correspondence between historical unit periods A gridded impact factor distribution map.

[0078] In step S3, specifically, the multiple static factors include but are not limited to elevation, slope and / or aspect obtained based on digital elevation model data processing, stratum lithology and / or distance to faults obtained based on geological map data processing, distance to road network obtained based on road network data processing, and / or distance to water system obtained based on water system data processing, etc.; the multiple dynamic factors include but are not limited to land use type and / or vegetation coverage obtained based on remote sensing image data processing, distance to epicenter obtained based on earthquake point data processing, rainfall obtained based on rainfall distribution data processing, and / or population density obtained based on population data processing, etc. In addition, the specific processing and obtaining methods of the aforementioned multiple static factors and multiple dynamic factors can be conventionally derived based on existing technical means, and will not be repeated here.

[0079] S4. spatially divide the target area to obtain a regular square grid, and for each grid unit in the regular square grid, divide the target area into a regular square grid according to the plurality of gridded impact factor distribution graphs, the plurality of gridded impact factor distribution time series graphs, and the A set of landslide points and non-landslide points is obtained based on the convolutional neural network CNN, which corresponds to the current most recent continuous One-to-one correspondence between historical unit periods A landslide hazard level value is a label value corresponding to a low landslide hazard level, a medium landslide hazard level or a high landslide hazard level.

[0080] In step S4, the side length of the regular square grid can be determined according to the resolution of the multiple gridded impact factor distribution maps and the multiple gridded impact factor distribution time series map groups (generally, the pixel size with the lowest resolution in the multiple gridded impact factor distribution maps and the multiple gridded impact factor distribution time series map groups is used as the side length); specifically, the center point of the grid represents the regular square grid, and each grid is used as a landslide hazard assessment and prediction unit, that is, also as a cell in the cellular automaton (CA) described later. Specifically, for each grid unit in the regular square grid, according to the multiple gridded impact factor distribution maps, the multiple gridded impact factor distribution time series map groups and the A set of landslide points and non-landslide points is obtained based on the convolutional neural network CNN, which corresponds to the current most recent continuous One-to-one correspondence between historical unit periods A landslide hazard level value includes but is not limited to the following steps S41 to S44.

[0081] S41. For each grid unit in the regular square grid, according to the position of the corresponding grid in the multiple gridded impact factor distribution maps, a plurality of static factor values ​​corresponding to the multiple static factors are extracted from the multiple gridded impact factor distribution maps, and according to the position of the corresponding grid in the multiple gridded impact factor distribution timing diagram groups, a plurality of dynamic factor value groups corresponding to the multiple dynamic factors are extracted from the multiple gridded impact factor distribution timing diagram groups, wherein the dynamic factor value groups contain the corresponding dynamic factors and correspond to the current most recent continuous One-to-one correspondence between historical unit periods dynamic factor values, and also according to the corresponding grid and the The position matching result of the landslide point and the landslide point or non-landslide point in the non-landslide point set is determined to correspond to and be consistent with the current most recent continuous One-to-one correspondence between historical unit periods Landslide label.

[0082] In step S41, the specific method of extracting the multiple static factor values ​​and the multiple dynamic factor value groups is the prior art. In addition, for example, if the grid of a certain grid unit matches the position of a certain landslide point in a set of landslide points and non-landslide points (for example, the certain landslide point falls in the grid of the certain grid unit), then it can be determined that the certain grid unit and the current most recent continuous The landslide label corresponding to a certain historical unit period in the historical unit periods (because it corresponds to the certain landslide point and non-landslide point set) is "landslide" (the specific corresponding label value can be designed to be "1"); and if the grid of the certain grid unit matches the position of a non-landslide point in the certain landslide point and non-landslide point set (for example, the non-landslide point falls in the grid of the certain grid unit), it can be determined that the landslide label corresponding to the certain grid unit and the certain historical unit period is "non-landslide" (the specific corresponding label value can be designed to be "0").

[0083] S42. For each grid unit, the corresponding and most recent continuous One-to-one correspondence between historical unit periods The first sample data will correspond to and be consistent with the current most recent continuous All dynamic factor values ​​corresponding to any historical unit period in the historical unit periods and the corresponding multiple static factor values ​​are taken as the first model input items, and the landslide label corresponding to and corresponding to any historical unit period is taken as the first model output item, and then the first model input item and the first model output item are taken as the first sample data corresponding to and corresponding to any historical unit period.

[0084] In the step S42, for example, all dynamic factor values ​​of a certain grid unit corresponding to last year and the multiple static factor values ​​of the certain grid unit can be used as the first model input item, and the landslide label of the certain grid unit corresponding to last year can be used as the first model output item, and then the first model input item and the first model output item can be used as the first sample data of the certain grid unit corresponding to last year.

[0085] S43. All the first sample data are divided into a training sample data set, a verification sample data set and a test sample data set, and the training sample data set and the verification sample data set are used to calibrate and verify the modeling of the first artificial intelligence model based on the convolutional neural network CNN to obtain a landslide probability estimation model, and then the test sample data set is used to perform model testing on the landslide probability estimation model to obtain a landslide probability average value, a landslide probability standard deviation, a non-landslide probability average value and a non-landslide probability standard deviation, and finally a value obtained by subtracting the landslide probability standard deviation from the landslide probability average value is used as a first dividing value between a high landslide hazard level and a medium landslide hazard level, and a value obtained by adding the non-landslide probability average value to the non-landslide probability standard deviation is used as a second dividing value between a medium landslide hazard level and a low landslide hazard level.

[0086] In the step S43, the sample ratio of the training sample data set, the verification sample data set and the test sample data set is 8:1:1, for example, wherein the training sample data set and the verification sample data set are used to adjust the model parameters, and the test sample data set does not participate in the model training, but is only used to evaluate the model performance. The convolutional neural network CNN (Convolutional Neural Networks) is an existing neural network, and its specific network structure is not repeated here. The specific process of the calibration verification modeling is an existing process, including a calibration process and a verification process of the model, that is, first by comparing the model simulation results with the measured data, and then adjusting the model parameters according to the comparison results, so that the simulation results are consistent with the actual process, so the landslide probability estimation model can be obtained by conventional training. Since the landslide probability estimation model can output the landslide probability and / or non-landslide probability after inputting the first model input item, the landslide probability average value, the landslide probability standard deviation, the non-landslide probability average value and the non-landslide probability standard deviation can be obtained by conventional statistics based on multiple first sample data in the test sample data set, after comparing the actual first model output items to remove the samples with evaluation errors.

[0087] S44. For each of the grid cells, Each first model input item in the first sample data is respectively imported into the landslide probability estimation model, and the output is obtained by comparing the current most recent continuous One-to-one correspondence between historical unit periods Then, according to the comparison result of the landslide probability with the first demarcation value and the second demarcation value, the corresponding and most recent continuous One-to-one correspondence between historical unit periods A landslide hazard level value is a label value corresponding to a low landslide hazard level, a medium landslide hazard level or a high landslide hazard level.

[0088] In the step S44, for example, the value "1" can be used to represent the label value corresponding to the low landslide hazard level, the value "2" can be used to represent the label value corresponding to the medium landslide hazard level, and the value "3" can be used to represent the label value corresponding to the high landslide hazard level. In addition, by summarizing the landslide hazard level values ​​of each grid unit in different unit periods, the static evaluation results of the regional landslide hazard of the target area in different unit periods can be obtained, such as Figure 2 shown.

[0089] S5. Construct the following cellular automaton CA according to the regular square grid: define all cells corresponding to all grid cells in the regular square grid, the states and evolution rules corresponding to the cells in the entire cellular space, and for each grid cell, make the cellular neighbor relationship of the corresponding cell consistent with the corresponding grid cell neighbor relationship.

[0090] In step S5, the cellular automaton (CA) is a spatiotemporal discrete dynamic model. By abstracting and encapsulating the research object into cells on the cellular space, each cell takes a finite discrete state and changes the cellular state simultaneously according to the prescribed local evolution rules, it can achieve evolution from simple to complex and from local to global, and is suitable for simulating complex system behaviors and phenomena. In recent years, CA has been widely used in the field of geosciences, for example, applying dynamic irregular vector cellular automaton models to simulate urban growth, combining genetic algorithms to improve CA evolution rules for simulating vegetation changes, CA under multi-criteria constraint rules to achieve disaster simulation, or combining deep learning CA for dynamic spatiotemporal land use change simulation. The cellular automaton is mainly composed of six parts: cellular space, cells, evolution rules, finite state set, neighbors and discrete time set. Its operation is shown in the following formula:

[0091]

[0092] In the formula, represents the entire cellular space, Represents the spatial dimension, which can be any positive integer; represents the set of cell states, usually a finite set of integer values; Represents the total number of cell neighbors of a single cell; Represents the cell state in the next unit period; Represents the cell state in the current unit period; Represents the state of the neighbors in the current unit period, which is usually a set of values; Represents the evolution rule, which can be a constrained knowledge rule or function in various forms. It is the core of the operation of the cellular automaton and is used to indicate how the cell state changes within an evolution step. In addition, keeping the cell neighbor relationship of the corresponding cell consistent with the corresponding grid cell neighbor relationship means that for a certain grid cell, if there are multiple grid cell neighbors around it (for example, eight grid cell neighbors located in the 8 directions of the upper left and lower right), then there are multiple cell neighbors around the corresponding cell that correspond one to one with the multiple grid cell neighbors.

[0093] S6. According to the said each grid unit The landslide hazard level value is calculated based on the state transition of the landslide hazard level value of each grid unit in two adjacent historical units and the actual spatial position of each grid unit, and the evolution rules integrating time and space characteristics are obtained based on the densely connected convolutional neural network DenseNet.

[0094] In step S6, the densely connected convolutional neural network DenseNet is a deep convolutional neural network structure proposed by Gao Huang, Zhuang Liu, Laurens van der Maaten and Kilian Q. Weinberger in 2016. It introduces dense connectivity in the network so that each layer is directly connected to all previous layers. This connection method significantly improves information flow, alleviates the gradient vanishing problem, and promotes feature reuse. Specifically, Figure 3 As shown, the densely connected convolutional neural network DenseNet includes but is not limited to an input layer, a padding layer, a convolution layer, a first normalization layer, a rectified linear unit layer, a first dense block, a first transition layer, a second dense block, a second transition layer, a third dense block, a second normalization layer, a global pooling layer, a fully connected layer and an output layer connected in sequence; the padding layer is used to widen the input size so that the padding width of all inputs is consistent; the rectified linear unit layer is used to use a linear and non-saturated activation function in the network; the first dense block, the second dense block and the third dense block each contain twelve convolution groups, And every two convolution groups in the twelve convolution groups are directly connected to achieve dense connection between networks; the first transition layer and the second transition layer both include a convolution sublayer and a pooling sublayer, and are both used to extract features output by the corresponding previous dense blocks and compress the features; the global pooling layer is used to compress the features into a one-dimensional array in a pooling form, and input the one-dimensional array into the fully connected layer; the fully connected layer is used to convert the received features into the landslide hazard level value (for example, 1, 2 or 3, etc., so as to correspond to the low hazard level of the landslide, the medium hazard level of the landslide or the high hazard level of the landslide).

[0095] In step S6, the construction principle of the evolution rule integrating time features and space features can be derived based on the following (A) construction principle of the evolution rule integrating time features and (B) construction principle of the evolution rule integrating space features.

[0096] (A) Construction principle of evolutionary rules integrating time characteristics: Time is abstracted as a discrete infinite set and divided into continuous equidistant time periods in units of years, and each year is a "time point". It is believed that the dangerous state of landslide geological hazards gradually changes in adjacent time periods, and this change accumulates with the extension of time, and the speed of state transfer may be different in different development stages. In order to explore the changing characteristics of the landslide dangerous state in different time periods and pass the state change information of the landslide in the current iteration cycle to the next iteration cycle, by analyzing the change of the landslide dangerous level value from the previous moment to the next moment, each change is assigned a category, which represents the state change characteristics of the landslide prediction unit in the time period, so as to quantify the time characteristics. By analyzing the changes in the state values ​​of the central cell and the neighboring cells in two consecutive periods, the evolutionary rules based on time characteristics are extracted based on the DenseNet network constructed above. The extraction of time feature evolution rules is also realized based on sample features and sample labels. Figure 4 The red grid at time t represents a central cell, the gray grid represents its neighboring cells, and the grid value is the cell state at time t; the position of the cell at time t+1 and time t is one-to-one corresponding, that is, the red grid value at time t+1 represents the state value of the central cell at time t at time t+1, and the gray grid value is the state value of the corresponding neighboring cell at time t+1; the state values ​​of the central cell and neighboring cells at time t and time t+1 are organized into 9×1 column vectors respectively, and the corresponding relationship between each element in the vector and the actual position of the cell from top to bottom is: central cell, south, southeast, east, northeast, north, northwest, west, southwest neighbor cells; according to the state transition of the same cell at time t and time t+1, a category value is assigned to each transition, and it is reorganized into a new 9×1 column vector as a sample feature; Figure 4The red grid value at time t+2 is the state value of the above central cell at time t+2, which is used as the sample label. The cell state values ​​in the sample features are 1, 2, 3, 4, 5, 6, 7, 8 and 9, representing the types of landslide hazard transition from time t to time t+1: low->low, low->medium, low->high, medium->low, medium->medium, medium->high, high->low, high->medium and high->high, respectively. The neighbor cell state values ​​are 0, 1, 2, 3, 4, 5, 6, 7, 8 and 9, where 0 represents the grid cell outside the study area; the cell state values ​​in the sample label are 1, 2 and 3, representing low, medium and high landslide hazard at time t+2. Since the state transition category value still maintains the state of the cell at time t+1, the sample label contains the cell state at time t+2, and the extraction method of the evolution rule conforms to the CA concept. The sample set data is organized based on the above data organization method, and the model training is completed. In the process of evolution prediction, the landslide hazard state transition category values ​​at two consecutive moments are taken as the initial state. The cell state is updated once the CA is iterated. The outputs of the first two models can be organized and used as the input for the next one, thus realizing the dynamic prediction of landslide hazard based on the time characteristic evolution rules.

[0097] (B) Construction principle of evolution rules integrating spatial characteristics: The evolution process of landslide geological hazards in geological space is essentially a geological problem. Landslide evolution units have spatial characteristics. The closer the adjacent landslide evolution units are, the greater the interaction between them. For each evolution unit, the selection of neighboring units and the spatial distribution of neighboring units are important factors in determining the landslide hazard level of the evolution unit. For each cell in the landslide cellular space and its neighboring cells, according to their actual positions in the cellular space, they are organized into entities within a local spatial range, and the cell state value corresponds to the position one by one. The DenseNet network can be used to extract spatial features in the data, and then obtain evolution rules based on spatial features. The sample feature data contains the actual spatial position information of the cells. Figure 5The blue grid at time t is the central cell, the gray grid is the neighboring cell, and the grid value is the state value of the cell at time t. The central cell and the neighboring cells are organized into a 3×3 two-dimensional array according to their actual positions in the cellular space as sample features; the state value of the central cell at time t+1 is used as the sample label; similarly, the state values ​​of the central cell at time t and time t+1 are obtained from 1, 2 and 3, and the state values ​​of the neighboring cells are obtained from 0, 1, 2 and 3. By reorganizing the sample set data, sample data with spatial features are obtained, and then the evolution rules based on spatial features can be extracted based on the DenseNet network. Taking the landslide hazard level value of the target area as the initial state of the evolution prediction, all cells in the cellular space execute the evolution rule once to obtain their state at the next moment, and the output of the model last time is used as the input for the next time, so as to realize the dynamic evolution of landslide hazard based on the spatial feature evolution rule.

[0098] In step S6, specifically, according to the A landslide hazard level value is obtained based on the state transition of the landslide hazard level value of each grid unit in two adjacent historical units and the actual spatial position of each grid unit, and an evolution rule integrating time characteristics and spatial characteristics is obtained based on a densely connected convolutional neural network DenseNet, including but not limited to the following steps S61 to S63.

[0099] S61. For each grid cell in the regular square grid, obtain the corresponding The landslide hazard level value and the corresponding value of all adjacent grid cells landslide hazard level value.

[0100] In step S61, for a certain grid unit, all the corresponding adjacent grid units are, for example but not limited to, 8 adjacent grid units located around it, such as Figure 6 shown.

[0101] S62. For each grid unit, the corresponding and most recent continuous One-to-one correspondence between historical unit periods The second sample data: first, for each cell in the corresponding cell and all adjacent grid cells of the corresponding cell, according to the corresponding and closest to the current The landslide hazard level value corresponding to the historical unit period and the corresponding and closest to the current The landslide hazard level value corresponding to the historical unit period is determined to determine the corresponding state change category value, and then according to the square matrix structure of the corresponding unit and all the adjacent grid units of the corresponding unit, the state change category values ​​of the respective units are formed into a two-dimensional array whose matrix structure is consistent with the square matrix structure, and the two-dimensional array is used as the second model input item, and the corresponding unit and the first unit closest to the current one are used as the second model input item. The landslide hazard level value corresponding to the historical unit period is used as the second model output item, and finally the second model input item and the second model output item are used as the corresponding and closest to the current The second sample data corresponding to the historical unit period.

[0102] In step S62, the determination rule of the state change category value and the composition example of the two-dimensional array can be specifically as follows: Figure 6 shown.

[0103] S63. Apply all of the second sample data to perform model training on a second artificial intelligence model based on a densely connected convolutional neural network DenseNet to obtain evolution rules that integrate temporal features and spatial features.

[0104] In step S63, the specific process of the model training is also the specific process of the calibration verification modeling, which will not be described in detail. In addition, the evolution rule integrating the time characteristics and the space characteristics is specifically a landslide hazard level value prediction model obtained after the model training.

[0105] S7. Start the cellular automaton CA as follows: The initial state for evolution prediction is extracted based on the landslide hazard level values ​​corresponding to a certain pair of adjacent historical unit periods in a historical unit period, and then the states of all the cells are updated based on the evolution rule integrating time characteristics and spatial characteristics to obtain the predicted landslide hazard level value of each grid unit in the next unit period after the certain pair of adjacent historical unit periods, and the states of all the cells are continuously iterated and updated with the certain pair of adjacent historical unit periods and the next unit period as an iteration cycle until the predicted landslide hazard level value of each grid unit in the target unit period is obtained, wherein the target unit period is a unit period located after the certain pair of adjacent historical unit periods in time sequence.

[0106] In step S7, the specific process of iterative updating of all cells can be conventionally derived based on the working principle of the existing cellular automaton CA, which will not be repeated here. In addition, for example, when the two adjacent historical unit periods are 2014 and 2015 respectively, the next unit period is 2016, so the cell states of 2017, 2018 and later years can be continuously iterated and updated until the predicted landslide hazard level value of each grid unit in the target unit period (for example, 2025) is obtained.

[0107] S8. Summarize the predicted landslide hazard level values ​​of the respective grid units during the target unit period to obtain a dynamic prediction result of regional landslide hazard of the target area during the target unit period.

[0108] In step S8, the dynamic prediction result of regional landslide hazard can be, but is not limited to, Figure 7 The landslide hazard classification zoning shown is presented in the form of a map (i.e. the next unit period is 2017, and subsequent iterations can predict the landslide hazard classification zoning in 2018, 2022 and 2025).

[0109] Therefore, based on the regional landslide hazard assessment and prediction method described in the aforementioned steps S1 to S8, a new scheme for regional landslide hazard assessment and prediction based on convolutional neural networks, densely connected convolutional neural networks and cellular automata is provided, that is, firstly, based on the landslide influencing factor data, a regional landslide hazard static assessment method is proposed in combination with convolutional neural networks, and then a cellular automaton is constructed, and based on the static evaluation result data, the temporal causal relationship and spatial correlation of the landslide hazard state in the landslide evolution process are deeply analyzed, and based on the densely connected convolutional neural network DenseNet, an evolution rule integrating temporal characteristics and spatial characteristics is constructed, and finally, the evolution rule is used in the cellular automaton to predict the landslide hazard level value to obtain the regional landslide hazard dynamic prediction result, which can overcome the limitations of a single evolution rule, improve the prediction accuracy, provide scientific support for landslide disaster prevention, and thus easily meet the actual needs of disaster prevention and mitigation, have broad application prospects and promotion value, and are convenient for practical application and promotion.

[0110] like Figure 8 As shown, the second aspect of this embodiment provides a virtual device for implementing the regional landslide hazard assessment and prediction method described in the first aspect, including an original data collection unit, a landslide data processing unit, a basic data processing unit, a landslide static evaluation unit, a cellular automation construction unit, an evolution rule derivation unit, a landslide dynamic prediction unit and a prediction result summary unit;

[0111] The raw data collection unit is used to collect the target area's most recent continuous The landslide point data of historical unit periods and the basic data of the target area, among which, represents a positive integer greater than or equal to ten;

[0112] The landslide data processing unit is communicatively connected to the raw data collection unit and is used to process the landslide point data to obtain the most recent continuous The historical unit periods correspond one to one and have the same number of elements. A set of landslide points and non-landslide points, among which, Indicates less than or equal to A positive integer, the landslide point and non-landslide point set contains Landslide points and Non-landslide points, and Respectively represent positive integers;

[0113] The basic data processing unit is communicatively connected to the original data collection unit, and is used to process and generate a plurality of gridded influence factor distribution maps corresponding to a plurality of static factors affecting landslide hazard assessment in the target area and a plurality of gridded influence factor distribution time series map groups corresponding to a plurality of dynamic factors affecting landslide hazard assessment in the target area according to the basic data, wherein the gridded influence factor distribution time series map groups contain corresponding dynamic factors and correspond to the current most recent continuous One-to-one correspondence between historical unit periods A grid-based impact factor distribution map;

[0114] The landslide static evaluation unit is respectively connected to the landslide data processing unit and the basic data processing unit for spatially segmenting the target area to obtain a regular square grid, and for each grid unit in the regular square grid, the plurality of gridded impact factor distribution graphs, the plurality of gridded impact factor distribution time series graph groups and the A set of landslide points and non-landslide points is obtained based on the convolutional neural network CNN, which corresponds to the current most recent continuous One-to-one correspondence between historical unit periods a landslide hazard level value, wherein the landslide hazard level value refers to a label value corresponding to a low landslide hazard level, a medium landslide hazard level, or a high landslide hazard level;

[0115] The cellular automaton construction unit is communicatively connected to the landslide static evaluation unit, and is used to construct the following cellular automaton CA according to the regular square grid: defining all cells corresponding to all grid cells in the regular square grid, states and evolution rules corresponding to the cells in the entire cellular space, and for each grid cell, making the cellular neighbor relationship of the corresponding cell consistent with the corresponding grid cell neighbor relationship;

[0116] The evolution rule derivation unit is communicatively connected to the landslide static evaluation unit, and is used to determine the landslide static evaluation unit according to the landslide hazard level values, based on the state transition of the landslide hazard level values ​​of each grid unit in two adjacent historical units and the actual spatial position of each grid unit, and based on the densely connected convolutional neural network DenseNet, an evolution rule integrating time characteristics and spatial characteristics is obtained;

[0117] The landslide dynamic prediction unit is respectively connected to the cellular automaton construction unit and the evolution rule derivation unit for communication, and is used to start the cellular automaton CA in the following manner: The initial state for evolution prediction is extracted based on the landslide hazard level value corresponding to a certain pair of adjacent historical unit periods in a historical unit period, and then the state of all the cells is updated based on the evolution rule integrating the time characteristics and the space characteristics to obtain the predicted landslide hazard level value of each grid unit in the next unit period after the certain pair of adjacent historical unit periods, and the state of all the cells is continuously updated iteratively with the certain pair of adjacent historical unit periods and the next unit period as an iteration cycle until the predicted landslide hazard level value of each grid unit in the target unit period is obtained, wherein the target unit period is a unit period located after the certain pair of adjacent historical unit periods in time sequence;

[0118] The prediction result summary unit is communicatively connected to the landslide dynamic prediction unit, and is used to summarize the predicted landslide hazard level values ​​of each grid unit in the target unit period to obtain the regional landslide hazard dynamic prediction result of the target area in the target unit period.

[0119] In a possible design, it also includes a basic data display unit, a risk assessment display unit, a risk prediction display unit and / or a result mapping processing unit;

[0120] The basic data display unit is used to provide static factor preview and dynamic factor preview functions;

[0121] The risk assessment display unit is used to provide a preview of the risk assessment classification diagram;

[0122] The hazard prediction display unit is used to provide the functions of hazard prediction result generation, hazard prediction map generation, layer publishing and layer preview;

[0123] The result mapping processing unit is used to provide the title, resolution, latitude and longitude grid, compass, legend, scale setting of the hazard result map and the function of saving the result map.

[0124] The working process, working details and technical effects of the aforementioned device provided in the second aspect of this embodiment can be referred to the regional landslide hazard assessment and prediction method described in the first aspect, and will not be repeated here.

[0125] like Fig. 9 As shown, the third aspect of this embodiment provides a computer device for executing the regional landslide hazard assessment prediction method as described in the first aspect, including a memory, a processor and a transceiver that are sequentially connected in communication, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the regional landslide hazard assessment prediction method as described in the first aspect. For example, the memory may include, but is not limited to, a random access memory (Random-Access Memory, RAM), a read-only memory (Read-Only Memory, ROM), a flash memory (Flash Memory), a first-in-first-out memory (First Input First Output, FIFO) and / or a first-in-last-out memory (First Input Last Output, FILO), etc.; the processor may include, but is not limited to, a microprocessor of the STM32F105 series. In addition, the computer device may also include, but is not limited to, a power module, a display screen and other necessary components.

[0126] The working process, working details and technical effects of the aforementioned computer equipment provided in the third aspect of this embodiment can be found in the regional landslide hazard assessment and prediction method described in the first aspect, and will not be repeated here.

[0127] The fourth aspect of this embodiment provides a computer-readable storage medium storing instructions including the method for evaluating and predicting the hazard of regional landslides as described in the first aspect, that is, the computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the method for evaluating and predicting the hazard of regional landslides as described in the first aspect is executed. The computer-readable storage medium refers to a carrier for storing data, which may include but is not limited to computer-readable storage media such as floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or memory sticks, and the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0128] The working process, working details and technical effects of the aforementioned computer-readable storage medium provided in the fourth aspect of this embodiment can be referred to the regional landslide hazard assessment and prediction method described in the first aspect, and will not be repeated here.

[0129] The fifth aspect of this embodiment provides a computer program product, including a computer program or instructions, which, when executed by a computer, implements the regional landslide hazard assessment and prediction method as described in the first aspect. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0130] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A regional landslide hazard assessment and prediction method based on CNN and CA, characterized in that: include: Collect the target area in the current recent continuous The landslide point data of historical unit periods and the basic data of the target area, among which, represents a positive integer greater than or equal to ten; According to the landslide point data, the most recent continuous The historical unit periods correspond one to one and have the same number of elements. A set of landslide points and non-landslide points, among which, Indicates less than or equal to A positive integer, the set of landslide points and non-landslide points contains Landslide points and Non-landslide points, and Respectively represent positive integers; According to the basic data, a plurality of gridded influence factor distribution maps corresponding to a plurality of static factors affecting landslide hazard assessment in the target area and a plurality of gridded influence factor distribution time series map groups corresponding to a plurality of dynamic factors affecting landslide hazard assessment in the target area are processed and generated, wherein the gridded influence factor distribution time series map groups contain corresponding dynamic factors and correspond to the current most recent continuous One-to-one correspondence between historical unit periods A grid-based impact factor distribution map; The target area is spatially divided to obtain a regular square grid, and for each grid unit in the regular square grid, according to the multiple gridded impact factor distribution graphs, the multiple gridded impact factor distribution time series graph groups and the A set of landslide points and non-landslide points is obtained based on the convolutional neural network CNN, which corresponds to the current most recent continuous One-to-one correspondence between historical unit periods a landslide hazard level value, wherein the landslide hazard level value refers to a label value corresponding to a low landslide hazard level, a medium landslide hazard level, or a high landslide hazard level; The following cellular automaton CA is constructed according to the regular square grid: all cells corresponding to all grid cells in the regular square grid, the states and evolution rules corresponding to the cells are defined in the entire cellular space, and for each grid cell, the cellular neighbor relationship of the corresponding cell is made consistent with the corresponding grid cell neighbor relationship; According to the landslide hazard level values, based on the state transition of the landslide hazard level values ​​of each grid unit in two adjacent historical units and the actual spatial position of each grid unit, and based on the densely connected convolutional neural network DenseNet, an evolution rule integrating time characteristics and spatial characteristics is obtained; The cellular automaton CA is started as follows: The initial state for evolution prediction is extracted based on the landslide hazard level value corresponding to a certain pair of adjacent historical unit periods in a historical unit period, and then the state of all the cells is updated based on the evolution rule integrating the time characteristics and the space characteristics to obtain the predicted landslide hazard level value of each grid unit in the next unit period after the certain pair of adjacent historical unit periods, and the state of all the cells is continuously updated iteratively with the certain pair of adjacent historical unit periods and the next unit period as an iteration cycle until the predicted landslide hazard level value of each grid unit in the target unit period is obtained, wherein the target unit period is a unit period located after the certain pair of adjacent historical unit periods in time sequence; The predicted landslide hazard level values ​​of the respective grid units in the target unit period are summarized to obtain a dynamic prediction result of regional landslide hazard of the target area in the target unit period.

2. The regional landslide hazard assessment and prediction method according to claim 1 is characterized in that: According to the landslide point data, the most recent continuous The historical unit periods correspond one to one and have the same number of elements. A set of landslide points and non-landslide points, including: According to the landslide point data, it is determined that the target area is the most recent continuous From the historical unit period closest to the present Historical unit period to the most recent one Historical unit period Landslide points, among which represents a positive integer greater than or equal to four and less than or equal to eight, Indicates less than or equal to A positive integer, represents a positive integer; From the around the landslide point and at least Randomly select from the area outside the point at a distance of meters Points as non-landslide points, among which represents a positive integer greater than or equal to four hundred and less than or equal to six hundred, represents a positive integer, and The sum of is a constant; The landslide points and The non-landslide point is added to the closest point to the current The landslide points and non-landslide points corresponding to each historical unit period are included in the set.

3. The method for regional landslide hazard assessment and prediction according to claim 1, characterized in that: The basic data includes digital elevation model data, geological map data, road network data, water system data, remote sensing image data, earthquake point data, rainfall distribution data and / or population data; The multiple static factors include elevation, slope and / or aspect obtained based on digital elevation model data processing, stratum lithology and / or distance to faults obtained based on geological map data processing, distance to road network obtained based on road network data processing, and / or distance to water system obtained based on water system data processing; The multiple dynamic factors include land use type and / or vegetation coverage obtained based on remote sensing image data processing, distance from the epicenter obtained based on earthquake point data processing, rainfall obtained based on rainfall distribution data processing and / or population density obtained based on population data processing.

4. The method for regional landslide hazard assessment and prediction according to claim 1, characterized in that: For each grid unit in the regular square grid, according to the plurality of gridded impact factor distribution graphs, the plurality of gridded impact factor distribution time series graph groups and the A set of landslide points and non-landslide points is obtained based on the convolutional neural network CNN, which corresponds to the current most recent continuous One-to-one correspondence between historical unit periods The landslide hazard level values ​​include: For each grid unit in the regular square grid, according to the position of the corresponding grid in the multiple gridded impact factor distribution maps, a plurality of static factor values ​​corresponding to the multiple static factors are extracted from the multiple gridded impact factor distribution maps, and according to the position of the corresponding grid in the multiple gridded impact factor distribution time series diagram groups, a plurality of dynamic factor value groups corresponding to the multiple dynamic factors are extracted from the multiple gridded impact factor distribution time series diagram groups, wherein the dynamic factor value groups contain the corresponding dynamic factors and correspond to the current most recent continuous One-to-one correspondence between historical unit periods dynamic factor values, and also according to the corresponding grid and the The position matching result of the landslide point and the landslide point or non-landslide point in the non-landslide point set is determined to correspond to and be consistent with the current most recent continuous One-to-one correspondence between historical unit periods Landslide tags; For each grid cell, the corresponding and most recent continuous One-to-one correspondence between historical unit periods The first sample data will correspond to and be consistent with the current most recent continuous All dynamic factor values ​​corresponding to any historical unit period in the historical unit periods and the corresponding multiple static factor values ​​are used as first model input items, and the corresponding landslide label corresponding to any historical unit period is used as a first model output item, and then the first model input item and the first model output item are used as first sample data corresponding to any historical unit period; Dividing all the first sample data into a training sample data set, a verification sample data set and a test sample data set, and applying the training sample data set and the verification sample data set to calibrate and verify the modeling of a first artificial intelligence model based on a convolutional neural network (CNN), to obtain a landslide probability estimation model, and then applying the test sample data set to perform model testing on the landslide probability estimation model, to obtain a landslide probability average value, a landslide probability standard deviation, a non-landslide probability average value and a non-landslide probability standard deviation, and finally using the value obtained by subtracting the landslide probability standard deviation from the landslide probability average value as a first dividing value between a high landslide risk level and a medium landslide risk level, and using the value obtained by adding the non-landslide probability standard deviation to the non-landslide probability average value as a second dividing value between a medium landslide risk level and a low landslide risk level; For each grid unit, Each first model input item in the first sample data is respectively imported into the landslide probability estimation model, and the output is obtained by comparing the current most recent continuous One-to-one correspondence between historical unit periods Then, according to the comparison result of the landslide probability with the first demarcation value and the second demarcation value, the corresponding and most recent continuous One-to-one correspondence between historical unit periods A landslide hazard level value is a label value corresponding to a low landslide hazard level, a medium landslide hazard level or a high landslide hazard level.

5. The method for regional landslide hazard assessment and prediction according to claim 1, characterized in that: According to the The landslide hazard level values ​​are obtained based on the densely connected convolutional neural network DenseNet, which integrates the evolution rules of time and space features, including: For each grid cell in the regular square grid, obtain the corresponding The landslide hazard level value and the corresponding value of all adjacent grid cells Landslide hazard level value; For each grid cell, the corresponding and most recent continuous One-to-one correspondence between historical unit periods The second sample data: first, for each cell in the corresponding cell and all adjacent grid cells of the corresponding cell, according to the corresponding and closest to the current The landslide hazard level value corresponding to the historical unit period and the corresponding and closest to the current The landslide hazard level value corresponding to the historical unit period is determined to determine the corresponding state change category value, and then according to the square matrix structure of the corresponding unit and all the adjacent grid units of the corresponding unit, the state change category values ​​of the respective units are formed into a two-dimensional array whose matrix structure is consistent with the square matrix structure, and the two-dimensional array is used as the second model input item, and the corresponding unit and the first unit closest to the current one are used as the second model input item. The landslide hazard level value corresponding to the historical unit period is used as the second model output item, and finally the second model input item and the second model output item are used as the corresponding and closest to the current The second sample data corresponding to the historical unit period; All of the second sample data are used to perform model training on a second artificial intelligence model based on a densely connected convolutional neural network DenseNet to obtain evolution rules that integrate temporal features and spatial features.

6. The method for regional landslide hazard assessment and prediction according to claim 1, characterized in that: The densely connected convolutional neural network DenseNet includes an input layer, a padding layer, a convolution layer, a first normalization layer, a rectified linear unit layer, a first dense block, a first transition layer, a second dense block, a second transition layer, a third dense block, a second normalization layer, a global pooling layer, a fully connected layer and an output layer connected in sequence; The padding layer is used to widen the input size so that the padding width of all inputs is consistent; The rectified linear unit layer is used to adopt a linear and non-saturated activation function in the network; The first dense block, the second dense block and the third dense block each include twelve convolution groups, and each two convolution groups in the twelve convolution groups are directly connected to achieve dense connection between networks; The first transition layer and the second transition layer both include a convolution sublayer and a pooling sublayer, and are both used to extract features output by the corresponding previous dense block and compress the features; The global pooling layer is used to compress the features into a one-dimensional array in a pooled form, and input the one-dimensional array into the fully connected layer; The fully connected layer is used to convert the received features into the landslide hazard level value.

7. A regional landslide hazard assessment and prediction device based on CNN and CA, characterized in that: It includes original data collection unit, landslide data processing unit, basic data processing unit, landslide static evaluation unit, cellular automaton construction unit, evolution rule derivation unit, landslide dynamic prediction unit and prediction result summary unit; The raw data collection unit is used to collect the target area's most recent continuous The landslide point data of historical unit periods and the basic data of the target area, among which, represents a positive integer greater than or equal to ten; The landslide data processing unit is communicatively connected to the raw data collection unit and is used to process the landslide point data to obtain the most recent continuous The historical unit periods correspond one to one and have the same number of elements. A set of landslide points and non-landslide points, among which, Indicates less than or equal to A positive integer, the set of landslide points and non-landslide points contains Landslide points and Non-landslide points, and Respectively represent positive integers; The basic data processing unit is communicatively connected to the original data collection unit, and is used to process and generate a plurality of gridded influence factor distribution maps corresponding to a plurality of static factors affecting landslide hazard assessment in the target area and a plurality of gridded influence factor distribution time series map groups corresponding to a plurality of dynamic factors affecting landslide hazard assessment in the target area according to the basic data, wherein the gridded influence factor distribution time series map groups contain corresponding dynamic factors and correspond to the current most recent continuous One-to-one correspondence between historical unit periods A grid-based impact factor distribution map; The landslide static evaluation unit is respectively connected to the landslide data processing unit and the basic data processing unit for spatially segmenting the target area to obtain a regular square grid, and for each grid unit in the regular square grid, the plurality of gridded impact factor distribution graphs, the plurality of gridded impact factor distribution time series graph groups and the A set of landslide points and non-landslide points is obtained based on the convolutional neural network CNN, which corresponds to the current most recent continuous One-to-one correspondence between historical unit periods a landslide hazard level value, wherein the landslide hazard level value refers to a label value corresponding to a low landslide hazard level, a medium landslide hazard level, or a high landslide hazard level; The cellular automaton construction unit is communicatively connected to the landslide static evaluation unit, and is used to construct the following cellular automaton CA according to the regular square grid: defining all cells corresponding to all grid cells in the regular square grid, states and evolution rules corresponding to the cells in the entire cellular space, and for each grid cell, making the cellular neighbor relationship of the corresponding cell consistent with the corresponding grid cell neighbor relationship; The evolution rule derivation unit is communicatively connected to the landslide static evaluation unit, and is used to determine the landslide static evaluation unit according to the landslide hazard level values, based on the state transition of the landslide hazard level values ​​of each grid unit in two adjacent historical units and the actual spatial position of each grid unit, and based on the densely connected convolutional neural network DenseNet, an evolution rule integrating time characteristics and spatial characteristics is obtained; The landslide dynamic prediction unit is respectively connected to the cellular automaton construction unit and the evolution rule derivation unit for communication, and is used to start the cellular automaton CA in the following manner: The initial state for evolution prediction is extracted based on the landslide hazard level values ​​corresponding to a pair of adjacent historical unit periods in a historical unit period, and then the states of all the cells are updated based on the evolution rule integrating the time characteristics and the space characteristics to obtain the predicted landslide hazard level value of each grid unit in the next unit period after the pair of adjacent historical unit periods, and the state of all the cells is continuously updated iteratively with the pair of adjacent historical unit periods and the next unit period as an iteration cycle until the predicted landslide hazard level value of each grid unit in the target unit period is obtained, wherein the target unit period is a unit period located after the pair of adjacent historical unit periods in time sequence; The prediction result summary unit is communicatively connected to the landslide dynamic prediction unit, and is used to summarize the predicted landslide hazard level values ​​of each grid unit in the target unit period to obtain the regional landslide hazard dynamic prediction result of the target area in the target unit period.

8. A computer device, characterized in that: It comprises a memory, a processor and a transceiver which are communicatively connected in sequence, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the regional landslide hazard assessment and prediction method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed on the computer, the regional landslide hazard assessment and prediction method as described in any one of claims 1 to 6 is executed.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or the instruction is executed by a computer, the method for evaluating and predicting regional landslide hazard according to any one of claims 1 to 6 is implemented.

Images