CN116933091A - Landslide vulnerability prediction method and device - Google Patents
Landslide vulnerability prediction method and device Download PDFInfo
- Publication number
- CN116933091A CN116933091A CN202311203712.9A CN202311203712A CN116933091A CN 116933091 A CN116933091 A CN 116933091A CN 202311203712 A CN202311203712 A CN 202311203712A CN 116933091 A CN116933091 A CN 116933091A
- Authority
- CN
- China
- Prior art keywords
- landslide
- prediction
- score
- similarity
- obtaining
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000012937 correction Methods 0.000 claims abstract description 60
- 238000005259 measurement Methods 0.000 claims abstract description 40
- 238000013528 artificial neural network Methods 0.000 claims abstract description 32
- 238000011156 evaluation Methods 0.000 claims abstract description 10
- 230000007613 environmental effect Effects 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 44
- 238000011524 similarity measure Methods 0.000 claims description 18
- 238000010606 normalization Methods 0.000 claims description 15
- 238000012935 Averaging Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 7
- 238000012795 verification Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 5
- 238000010200 validation analysis Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000002790 cross-validation Methods 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims 1
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 14
- 230000006399 behavior Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005336 cracking Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 244000144972 livestock Species 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
- G06Q50/265—Personal security, identity or safety
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Tourism & Hospitality (AREA)
- Biomedical Technology (AREA)
- Educational Administration (AREA)
- Development Economics (AREA)
- Computer Security & Cryptography (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a landslide vulnerability prediction method and device. The method comprises the following steps: obtaining a disaster causing factor vector or a grid file in an environmental factor, and obtaining a standard grid unit set according to the disaster causing factor vector or the grid file; respectively constructing a PU-bagging landslide susceptibility prediction network and a similarity measurement loss landslide susceptibility prediction network according to the standard grid unit set; according to the PU-bagging landslide susceptibility prediction network, a pre-correction landslide susceptibility prediction probability score based on bagging is obtained; predicting a neural network according to the similarity measurement loss landslide susceptibility to obtain a correction prediction score; obtaining a final prediction probability score according to the pre-correction landslide susceptibility prediction probability score and the correction prediction score; and obtaining landslide susceptibility prediction evaluation results from the final prediction probability score by using a natural breakpoint method. The beneficial effects of this technical scheme are: the accuracy of landslide susceptibility prediction is improved.
Description
Technical Field
The invention relates to the technical field of geological disaster prediction, in particular to a landslide vulnerability prediction method and device.
Background
Landslide can carry a large amount of incomplete slope lamination fragments to slide downwards, wash out and cover up farmland, not only can reduce the arable land, can also destroy natural ecological environment, constitutes the threat to road base, bridge and culvert, even endangers people and livestock life and property safety. Landslide also can slide to the river course, blocks up the river, makes the river course narrow, and rivers impact highway one side, induce the new side slope collapse of highway, threaten railway safety constantly. Therefore, in order to prevent possible disasters, the landslide susceptibility is predicted and pre-warned.
In the prior art, regarding to building a landslide susceptibility prediction model, a landslide machine learning network is built by adopting negative samples with the same quantity as a positive sample of a landslide by random sampling, and finally, the trained network is utilized to obtain the landslide probability distribution and predict the susceptibility. However, since the field investigation is not exhaustive, the lack of investigation data and the landslide occurring in the future are not known, the spatial position of the random sampling negative sample cannot represent the position where the landslide does not occur truly or at a low probability. In addition, since the number of collected landslide points is far smaller than the landslide position to be predicted in space, it is difficult to make a good estimate of the landslide susceptibility of a large number of whole samples by using only a small number of negative samples equal to the positive sample of the landslide. Thus, the problem of inaccurate landslide susceptibility prediction is caused.
Disclosure of Invention
The invention solves the problem of how to improve the accuracy of landslide susceptibility prediction.
In order to solve the above problems, the present invention provides a landslide susceptibility prediction method, including:
obtaining a disaster causing factor vector or a grid file in an environmental factor, and obtaining a standard grid unit set according to the disaster causing factor vector or the grid file;
respectively constructing a PU-bagging landslide susceptibility prediction network and a similarity measurement loss landslide susceptibility prediction network according to the standard grid unit set;
according to the PU-bagging landslide susceptibility prediction network, a pre-correction landslide susceptibility prediction probability score based on bagging is obtained;
predicting a neural network according to the similarity measurement loss landslide susceptibility to obtain a correction prediction score;
obtaining a final prediction probability score according to the pre-correction landslide susceptibility prediction probability score and the correction prediction score;
and obtaining landslide susceptibility prediction evaluation results from the final prediction probability score by using a natural breakpoint method.
The beneficial effects of the invention are as follows: according to the method, the PU-bagging landslide susceptibility prediction network and the similarity measurement loss landslide susceptibility prediction network are respectively constructed, the PU-bagging landslide susceptibility prediction network is utilized to obtain the pre-correction landslide susceptibility prediction probability score, and the correction prediction score is obtained according to the similarity measurement loss landslide susceptibility prediction neural network. Obtaining a final prediction probability score according to the pre-correction landslide susceptibility prediction probability score and the correction prediction score; and obtaining landslide susceptibility prediction evaluation results from the final prediction probability score by using a natural breakpoint method. Therefore, the prediction probability score obtained based on the PU-bagging landslide susceptibility prediction network is corrected through correction of the prediction score, the possible inaccuracy problem of the PU-bagging landslide susceptibility prediction network on the prediction probability score is solved, and the accuracy of landslide susceptibility prediction is improved.
Optionally, the obtaining a canonical grid cell set according to the disaster causing factor vector or the grid file includes:
and rasterizing the disaster causing factor vector or the raster file in the same size and the same spatial position according to the spatial resolution of the elevation raster file and the positions of each raster unit to obtain a standard raster unit set.
Optionally, the constructing the PU-bagging landslide vulnerability prediction network according to the canonical grid cell set includes:
extracting known historical landslide unit data and residual landslide unit data from the standard grid unit set, obtaining a positive sample data set according to the known historical landslide unit data, and obtaining an unlabeled sample data set according to the residual landslide unit data;
taking the same amount of data as the positive sample data set from the unlabeled sample data set by a random sampling method to be a negative sample, and constructing a training data set according to the negative sample and the positive sample data set;
repeatedly training T rounds of base learners according to the training data set to obtain T base learners;
constructing the PU-bagging landslide susceptibility prediction network by utilizing the base learner;
the step of obtaining the pre-correction landslide susceptibility prediction probability score based on bagging according to the PU-bagging landslide susceptibility prediction network comprises the following steps:
predicting data sets except the training data set in the standard grid unit set by using T base learners to obtain T pre-prediction probability scores;
and averaging the T pre-prediction probability scores to obtain a pre-correction landslide susceptibility prediction probability score based on bagging.
Optionally, the similarity measure loss landslide susceptibility prediction neural network comprises a landslide coding network and a similarity measure loss function; the landslide coding network is used for extracting landslide characteristics; the similarity measure loss function is used to minimize the similarity measure loss through iterative feedback.
Optionally, the landslide coding network comprises a neural network composed of three linear layers, two activation layers and one normalization layer, the neural network comprises a first formula, wherein the first formula is:
wherein H (y) is a landslide characteristic result output by the landslide coding network, L is the linear layer and is used for realizing linear combination or linear transformation of data of each previous layer; relu is the active layer that is used to drive network behavior towards nonlinearity; normolize is the normalization layer that is used to map variables between 0 and 1.
Optionally, the similarity metric loss function includes a second formula, wherein the second formula includes:
wherein, p and the anchor are both training sample coding results in the landslide positive sample, N is the batch number of the anchor, unlabel is unlabel sample coding results, UN is the batch number of the unlabel, and T is the transposed mark of a matrix.
Optionally, obtaining a corrected prediction score according to the similarity measure loss landslide susceptibility prediction neural network, and further including: obtaining an optimal model parameter result by adopting a ten-fold cross validation method, wherein the optimal model parameter result comprises a model parameter result with the most similar validation data coding result and positive sample coding result; determining similarity of the verification data encoding result and the positive sample encoding result using a similarity metric function, wherein the similarity metric function comprises a third formula comprising:
wherein Valid is the verification data encoding result, positive is the positive sample encoding result, T is the transposed label of a matrix, and similarity A is the similarity measurement result.
Optionally, obtaining a correction prediction score according to the similarity measurement loss landslide susceptibility prediction neural network, and further comprising comparing and averaging the similarity measurement of each grid unit coding result and the landslide positive sample coding result according to the optimal model parameter result to obtain a similarity measurement score result average value; the average value of the similarity measurement results is calculated by a fourth formula, and the fourth formula comprises:
wherein similarity B is the average value of the similarity measurement score results, n is the number of positive samples, S i And (5) measuring the similarity score of the single grid unit coding result and the single grid unit landslide positive sample coding result.
Optionally, obtaining a correction prediction score according to the similarity measurement loss landslide susceptibility prediction neural network, and further includes normalizing the average value of the similarity measurement score result to obtain the correction prediction score, where a normalization function used by the normalization process is represented by a sixth formula, where the sixth formula includes:
wherein similarity c is the corrected prediction score and similarity b is the average of the similarity metric score results.
The invention also provides a landslide susceptibility prediction device, which comprises:
the building unit is used for obtaining a disaster causing factor vector or a grid file in the environmental factors and obtaining a standard grid unit set according to the disaster causing factor vector or the grid file; respectively constructing a PU-bagging landslide susceptibility prediction network and a similarity measurement loss landslide susceptibility prediction network according to the standard grid unit set;
the first training unit is used for obtaining a pre-correction landslide susceptibility prediction probability score based on bagging according to the PU-bagging landslide susceptibility prediction network;
the second training unit is used for predicting the neural network according to the similarity measurement loss landslide susceptibility to obtain a correction prediction score;
the integrated unit is used for obtaining a final prediction probability score according to the pre-correction landslide susceptibility prediction probability score and the correction prediction score;
and the prediction unit is used for obtaining landslide susceptibility prediction evaluation results from the final prediction probability score by using a natural breakpoint method.
The landslide susceptibility prediction device and the landslide susceptibility prediction method have the same advantages as those of the prior art, and are not described in detail herein.
Drawings
FIG. 1 is a schematic flow chart of a landslide vulnerability prediction method according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a landslide coding network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a landslide hazard prediction apparatus according to an embodiment of the present invention;
fig. 4 is a schematic diagram of a landslide susceptibility prediction apparatus according to an embodiment of the present invention.
Detailed Description
In order that the above objects, features and advantages of the invention will be readily understood, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings.
It is noted that the terms "first," "second," and the like in the description and claims of the invention and in the foregoing figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein.
In the description of the present specification, reference to the terms "embodiment," "some embodiments," and "alternative embodiments," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or implementation is included in at least one embodiment or illustrated implementation of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same examples or implementations. Furthermore, the particular features, structures, materials, or characteristics may be combined in any suitable manner in any one or more embodiments or implementations.
Referring to fig. 1, an embodiment of the present invention provides a landslide susceptibility prediction method, including:
step 101, obtaining a disaster causing factor vector or a grid file in an environmental factor, and obtaining a standard grid unit set according to the disaster causing factor vector or the grid file;
in some embodiments, the disaster-causing factors may include, but are not limited to, factors such as elevation of landslide points, distance from roads, distance from rivers, distance from faults, lithology, land use type, grade, slope direction, and the like. In the practical application process, factors related to landslide susceptibility can be selected as much as possible according to practical requirements, and landslide susceptibility distribution is predicted through multiple factors.
In some embodiments, obtaining a canonical grid cell set from the disaster causing factor vector or the grid file includes: the raw elevation data may be de-noised, i.e., averaged over a 3 x 3 window. And then, carrying out gradient, curvature, slope direction and other factors related to the elevation on the processed elevation data to obtain the raster image. The complete image is sliced into square grids, and the spatial resolution is 30m×30m based on the original elevation. And carrying out data normalization on the formed raster data to form final training data.
Step 102, respectively constructing a PU-bagging landslide susceptibility prediction network and a similarity measurement loss landslide susceptibility prediction network according to the standard grid unit set;
step 103, obtaining a pre-correction landslide susceptibility prediction probability score based on bagging according to the PU-bagging landslide susceptibility prediction network;
and repeating training for T rounds, adding the result score value of each round of prediction, and finally averaging based on the predicted times of each unit. The process of each round is as follows: and randomly taking the same amount of data as the positive sample from the unlabeled sample data set as the negative sample, constructing a training data set together with the positive sample data, and using the training data set for training a basic learner of the SVM, and predicting all other data which does not contain the training data constructed in each round after training is completed, wherein the other data is used as a prediction result of each unit of each round of prediction.
Step 104, predicting a neural network according to the similarity measurement loss landslide susceptibility to obtain a correction prediction score;
the training data set is input into a similarity measurement loss landslide susceptibility prediction neural network to obtain a correction prediction score which is used for correcting the pre-correction landslide susceptibility prediction probability score based on bagging.
Step 105, obtaining a final prediction probability score according to the pre-correction landslide susceptibility prediction probability score and the correction prediction score;
in some embodiments, the pre-correction landslide vulnerability prediction probability score and the correction prediction score are averaged to obtain a final prediction probability score.
And 106, obtaining a landslide susceptibility prediction evaluation result from the final prediction probability score by using a natural breakpoint method.
In one example, the ground disaster easily-transmitted area is divided into 5 grades by using a natural breakpoint method according to the prediction probability score result, and the 5 grades are respectively a high easily-transmitted area, a higher easily-transmitted area, a medium easily-transmitted area, a lower easily-transmitted area and a low easily-transmitted area, so that the landslide easily-transmitted prediction evaluation result is finally obtained. The natural breakpoint method is a statistical method for grading and classifying according to numerical statistical distribution rules, and can maximize the difference between classes. Any statistical series has some natural turning points and characteristic points, and the objects under study can be divided into groups with similar properties by using the points, so that the cracking points are good boundaries of grading. The statistical data are made into a frequency histogram, a gradient graph and an accumulated frequency histogram, which are all helpful for finding out the natural cracking point of the data.
In the embodiment, a PU-bagging landslide susceptibility prediction network and a similarity measurement loss landslide susceptibility prediction network are respectively constructed, the PU-bagging landslide susceptibility prediction network is utilized to obtain a pre-correction landslide susceptibility prediction probability score, and a correction prediction score is obtained according to the similarity measurement loss landslide susceptibility prediction neural network. Obtaining a final prediction probability score according to the pre-correction landslide susceptibility prediction probability score and the correction prediction score; and obtaining landslide susceptibility prediction evaluation results from the final prediction probability score by using a natural breakpoint method. Therefore, the prediction probability score obtained based on the PU-bagging landslide susceptibility prediction network is corrected by correcting the prediction score, the possible inaccuracy problem of the PU-bagging landslide susceptibility prediction network on the prediction probability score is solved, and the accuracy of landslide susceptibility prediction is improved.
Optionally, the obtaining a canonical grid cell set according to the disaster causing factor vector or the grid file includes:
and rasterizing the disaster causing factor vector or the raster file in the same size and the same spatial position according to the spatial resolution of the elevation raster file and the positions of each raster unit to obtain a standard raster unit set.
Specifically, in this embodiment, a disaster causing factor vector or a raster file is obtained, and the factors are rasterized in the same size and in the same spatial position according to the spatial resolution of the elevation raster file and the positions of the raster units. In some embodiments, all the rasterized factor data may be normalized to yield final training data.
Optionally, the constructing the PU-bagging landslide vulnerability prediction network according to the canonical grid cell set includes:
extracting known historical landslide unit data and residual landslide unit data from the standard grid unit set, obtaining a positive sample data set according to the known historical landslide unit data, and obtaining an unlabeled sample data set according to the residual landslide unit data;
taking the same amount of data as the positive sample data set from the unlabeled sample data set by a random sampling method to be a negative sample, and constructing a training data set according to the negative sample and the positive sample data set;
repeatedly training T rounds of base learners according to the training data set to obtain T base learners;
constructing the PU-bagging landslide susceptibility prediction network by utilizing the base learner;
the step of obtaining the pre-correction landslide susceptibility prediction probability score based on bagging according to the PU-bagging landslide susceptibility prediction network comprises the following steps:
predicting data sets except the training data set in the standard grid unit set by using T base learners to obtain T pre-prediction probability scores;
and averaging the T pre-prediction probability scores to obtain a pre-correction landslide susceptibility prediction probability score based on bagging.
In this embodiment, the remaining landslide unit data includes unknown landslide-occurred or historical non-landslide unit data.
In this embodiment, the data equivalent to the positive sample data set is taken from the unlabeled sample data set by a random sampling method and taken as a negative sample, and a training data set is constructed according to the negative sample and the positive sample data set; and repeatedly training the T-turn basis learner according to the training data set to obtain T basis learners, obtaining T pre-prediction probability scores through the T basis learners, and averaging the T pre-prediction probability scores to obtain a pre-correction landslide susceptibility prediction probability score based on bagging. Through the PU-bagging landslide susceptibility prediction network, the distinguishable strength of the susceptibility of each unit is reserved, namely the relative ordering of the probability of landslide occurrence of each grid unit is reserved. The method can realize the coarse-precision prediction of landslide susceptibility. However, the unlabeled sample data set may be contaminated by the positive sample, so that the predicted score obtained by the positive sample may be low, and the predicted score obtained by the negative sample may be high, and a certain method is required to be used for correction.
Optionally, the similarity measure loss landslide susceptibility prediction neural network comprises a landslide coding network and a similarity measure loss function; the landslide coding network is used for extracting landslide characteristics; the similarity measure loss function is used to minimize the similarity measure loss through iterative feedback.
Optionally, as shown in connection with fig. 2, the landslide coding network includes a neural network composed of three linear layers, two activation layers and one normalization layer, the neural network includes a first formula, where the first formula is:
wherein H (y) is a landslide characteristic result output by the landslide coding network, L is the linear layer and is used for realizing linear combination or linear transformation of data of each previous layer; relu is the active layer that is used to drive network behavior towards nonlinearity; normolize is the normalization layer that is used to map variables between 0 and 1.
In this embodiment, the landslide feature result includes a coding result for a training sample in a positive landslide sample. Training data is input into a landslide coding network in the network, and the training data firstly passes through three linear layers, a neural network of two nonlinear layers and a standardized layer to obtain landslide feature codes. Through a linear layer and a nonlinear layer in the deep learning network, the data can be subjected to nonlinear transformation to extract landslide susceptibility feature codes more accurately than traditional machine learning.
Optionally, the similarity metric loss function includes a second formula, wherein the second formula includes:
wherein, p and the anchor are both training sample coding results in the landslide positive sample, N is the batch number of the anchor, unlabel is unlabel sample coding results, UN is the batch number of the unlabel, and T is the transposed mark of a matrix.
In the present embodiment, batch is a term specific to deep learning, and refers to an input of the amount of data for model calculation at a time. For better training results, the more and better the number of latches that are p, anchor, unlabel, the more positive samples can be trained for each region by p and Anchor.
In this embodiment, the similarity metric loss in the landslide susceptibility prediction neural network based on the similarity metric loss makes the learned positive sample encoding result of the landslide more similar in space, and makes the encoding result at the negative sample grid unit with low landslide occurrence probability in the learned positive sample encoding result of the landslide more distant from the encoding result at the negative sample grid unit with low landslide occurrence probability in other unlabeled samples, thereby being beneficial to distinguishing the positive sample from the negative sample.
Optionally, obtaining a corrected prediction score according to the similarity measure loss landslide susceptibility prediction neural network includes: obtaining an optimal model parameter result by adopting a ten-fold cross validation method, wherein the optimal model parameter result comprises a model parameter result with the most similar validation data coding result and positive sample coding result; determining similarity of the verification data encoding result and the positive sample encoding result using a similarity metric function, wherein the similarity metric function comprises a third formula comprising:
wherein Valid is the verification data encoding result, positive is the positive sample encoding result, T is the transposed label of a matrix, and similarity A is the similarity measurement result.
In this embodiment, the ten-fold cross-validation method divides the positive sample into ten halves, wherein nine halves are used as training data, and the remaining halves are used as validation data for selecting the optimal model parameters, i.e. the result that the validation data code is most similar to the positive sample code is obtained.
Optionally, obtaining a correction prediction score according to the similarity measurement loss landslide susceptibility prediction neural network, and further comprising comparing and averaging the similarity measurement of each grid unit coding result and the landslide positive sample coding result according to the optimal model parameter result to obtain a similarity measurement score result average value; the average value of the similarity measurement results is calculated by a fourth formula, and the fourth formula comprises:
wherein similarity B is the average value of the similarity measurement score results, n is the number of positive samples, S i And (5) measuring the similarity score of the single grid unit coding result and the single grid unit landslide positive sample coding result.
In some embodiments, the similarity measure score is calculated by a fifth formula comprising:
wherein X is the single grid unit coding result, P i Encoding the result for the single grid cell landslide positive sample.
Optionally, obtaining a correction prediction score according to the similarity measurement loss landslide susceptibility prediction neural network, and further includes normalizing the average value of the similarity measurement score result to obtain the correction prediction score, where a normalization function used by the normalization process is represented by a sixth formula, where the sixth formula includes:
wherein similarity c is the corrected prediction score and similarity b is the average of the similarity metric score results.
In this embodiment, since the range of the similarity score result average value similarity b is [ -1,1], the range thereof is adjusted to [0,1] using the sixth formula. Thus, the pre-correction landslide susceptibility prediction probability score based on bagging is corrected through the correction prediction score.
With reference to fig. 3, an embodiment of the present invention further provides a landslide susceptibility prediction apparatus, including:
a building unit 501, configured to obtain a disaster causing factor vector or a raster file in an environmental factor, and obtain a canonical raster unit set according to the disaster causing factor vector or the raster file; respectively constructing a PU-bagging landslide susceptibility prediction network and a similarity measurement loss landslide susceptibility prediction network according to the standard grid unit set;
the first training unit 502 is configured to obtain a bagging-based pre-correction landslide susceptibility prediction probability score according to the PU-bagging landslide susceptibility prediction network;
a second training unit 503, configured to obtain a correction prediction score according to the similarity measure loss landslide susceptibility prediction neural network;
an integration unit 504, configured to obtain a final prediction probability score according to the pre-correction landslide susceptibility prediction probability score and the correction prediction score;
and the prediction unit 505 is used for obtaining a landslide susceptibility prediction evaluation result from the final prediction probability score by using a natural breakpoint method.
As shown in connection with fig. 4, in some embodiments the landslide susceptibility prediction apparatus further includes:
a preprocessing unit 506, configured to rasterize the disaster factor vector or the raster file in the same size and in the same spatial position according to the spatial resolution of the elevation raster file and the positions of the raster units, so as to obtain a canonical raster unit set;
the PU-bagging prediction unit 507 is configured to extract known historical landslide unit data and remaining landslide unit data from the canonical grid unit set, obtain a positive sample data set according to the known historical landslide unit data, and obtain an unlabeled sample data set according to the remaining landslide unit data;
the method is also used for taking the same amount of data as the positive sample data set from the unlabeled sample data set by a random sampling method to be a negative sample, and constructing a training data set according to the negative sample and the positive sample data set;
the training data set is used for training the T-turn basis learner repeatedly according to the training data set to obtain T basis learners;
the PU-bagging landslide susceptibility prediction network is also constructed by utilizing the base learner;
the step of obtaining the pre-correction landslide susceptibility prediction probability score based on bagging according to the PU-bagging landslide susceptibility prediction network comprises the following steps:
predicting data sets except the training data set in the standard grid unit set by using T base learners to obtain T pre-prediction probability scores;
and averaging the T pre-prediction probability scores to obtain a pre-correction landslide susceptibility prediction probability score based on bagging.
A landslide code extraction unit 508, configured to construct a landslide code network, where the landslide code network includes a neural network composed of three linear layers, two activation layers, and one normalization layer, and the neural network includes a first formula, where the first formula is:
wherein H (y) is a landslide characteristic result output by the landslide coding network, L is the linear layer and is used for realizing linear combination or linear transformation of data of each previous layer; relu is the active layer that is used to drive network behavior towards nonlinearity; normolize is the normalization layer that is used to map variables between 0 and 1.
A similarity measure loss processing unit 509, configured to construct a similarity measure loss function, where the similarity measure loss function includes a second formula, and the second formula includes:
wherein, p and the anchor are both training sample coding results in the landslide positive sample, N is the batch number of the anchor, unlabel is unlabel sample coding results, UN is the batch number of the unlabel, and T is the transposed mark of a matrix.
The landslide susceptibility prediction device and the landslide susceptibility prediction method have the same advantages as those of the prior art, and are not described in detail herein.
Although the present disclosure is described above, the scope of protection of the present disclosure is not limited thereto. Various changes and modifications may be made by one skilled in the art without departing from the spirit and scope of the disclosure, and these changes and modifications will fall within the scope of the disclosure.
Claims (10)
1. A landslide vulnerability prediction method, comprising:
obtaining a disaster causing factor vector or a grid file in an environmental factor, and obtaining a standard grid unit set according to the disaster causing factor vector or the grid file;
respectively constructing a PU-bagging landslide susceptibility prediction network and a similarity measurement loss landslide susceptibility prediction network according to the standard grid unit set;
according to the PU-bagging landslide susceptibility prediction network, a pre-correction landslide susceptibility prediction probability score based on bagging is obtained;
predicting a neural network according to the similarity measurement loss landslide susceptibility to obtain a correction prediction score;
obtaining a final prediction probability score according to the pre-correction landslide susceptibility prediction probability score and the correction prediction score;
and obtaining landslide susceptibility prediction evaluation results from the final prediction probability score by using a natural breakpoint method.
2. The landslide vulnerability prediction method of claim 1, wherein obtaining a canonical grid cell set from the disaster causing factor vector or the grid file comprises:
and rasterizing the disaster causing factor vector or the raster file in the same size and the same spatial position according to the spatial resolution of the elevation raster file and the positions of each raster unit to obtain a standard raster unit set.
3. A landslide vulnerability prediction method as set forth in claim 1, wherein,
the construction of the PU-bagging landslide susceptibility prediction network according to the standard grid unit set comprises the following steps:
extracting known historical landslide unit data and residual landslide unit data from the standard grid unit set, obtaining a positive sample data set according to the known historical landslide unit data, obtaining an unlabeled sample data set according to the residual landslide unit data,
taking the same amount of data as the positive sample data set from the unlabeled sample data set by a random sampling method as a negative sample, constructing a training data set according to the negative sample and the positive sample data set,
repeatedly training T rounds of base learners according to the training data set to obtain T base learners,
constructing the PU-bagging landslide susceptibility prediction network by utilizing the base learner;
the step of obtaining the pre-correction landslide susceptibility prediction probability score based on bagging according to the PU-bagging landslide susceptibility prediction network comprises the following steps:
predicting data sets other than the training data set in the canonical grid cell set using T of the base learners to obtain T pre-prediction probability scores,
and averaging the T pre-prediction probability scores to obtain a pre-correction landslide susceptibility prediction probability score based on bagging.
4. The landslide vulnerability prediction method of claim 1 wherein the similarity metric loss landslide vulnerability prediction neural network comprises a landslide coding network and a similarity metric loss function; the landslide coding network is used for extracting landslide characteristics; the similarity measure loss function is used to minimize the similarity measure loss through iterative feedback.
5. The landslide vulnerability prediction method of claim 4 wherein the landslide coding network comprises a neural network of three linear layers, two activation layers and one normalization layer, the neural network comprising a first formula, wherein the first formula is:
wherein H (y) is a landslide characteristic result output by the landslide coding network, L is the linear layer and is used for realizing linear combination or linear transformation of data of each previous layer; relu is the active layer for causing the network behavior to tend to be nonlinear; normolize is the normalization layer used to map variables between 0 and 1.
6. The landslide vulnerability prediction method of claim 4, wherein the similarity metric loss function comprises a second formula comprising:
wherein, p and the anchor are both training sample coding results in the landslide positive sample, N is the batch number of the anchor, unlabel is unlabel sample coding results, UN is the batch number of the unlabel, and T is the transposed mark of a matrix.
7. The landslide vulnerability prediction method of claim 4 wherein losing the landslide vulnerability prediction neural network based on the similarity measure results in a corrected prediction score comprising: obtaining an optimal model parameter result by adopting a ten-fold cross validation method, wherein the optimal model parameter result comprises a model parameter result with the most similar validation data coding result and positive sample coding result; determining similarity of the verification data encoding result and the positive sample encoding result using a similarity metric function, wherein the similarity metric function comprises a third formula comprising:
wherein Valid is the verification data encoding result, positive is the positive sample encoding result, T is the transposed label of a matrix, and similarity A is the similarity measurement result.
8. The landslide vulnerability prediction method of claim 7, wherein the corrected prediction score is derived from the similarity metric losing a landslide vulnerability prediction neural network, further comprising: according to the optimal model parameter result, comparing and averaging the similarity measurement of each grid unit coding result and the landslide positive sample coding result to obtain a similarity measurement score result average value; the average value of the similarity measurement results is calculated by a fourth formula, and the fourth formula comprises:
wherein similarity B is the average value of the similarity measurement score results, n is the number of positive samples, S i And (5) measuring the similarity score of the single grid unit coding result and the single grid unit landslide positive sample coding result.
9. The landslide vulnerability prediction method of claim 8, wherein the corrected prediction score is derived from the similarity metric losing a landslide vulnerability prediction neural network, further comprising: normalizing the average value of the similarity measurement score results to obtain the correction prediction score; the normalization function used by the normalization process is represented by a sixth formula including:
wherein similarity c is the corrected prediction score and similarity b is the average of the similarity metric score results.
10. A landslide susceptibility prediction apparatus, comprising:
the building unit is used for obtaining a disaster causing factor vector or a grid file in the environmental factors and obtaining a standard grid unit set according to the disaster causing factor vector or the grid file; respectively constructing a PU-bagging landslide susceptibility prediction network and a similarity measurement loss landslide susceptibility prediction network according to the standard grid unit set;
the first training unit is used for obtaining a pre-correction landslide susceptibility prediction probability score based on bagging according to the PU-bagging landslide susceptibility prediction network;
the second training unit is used for predicting the neural network according to the similarity measurement loss landslide susceptibility to obtain a correction prediction score;
the integrated unit is used for obtaining a final prediction probability score according to the pre-correction landslide susceptibility prediction probability score and the correction prediction score;
and the prediction unit is used for obtaining landslide susceptibility prediction evaluation results from the final prediction probability score by using a natural breakpoint method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311203712.9A CN116933091B (en) | 2023-09-19 | 2023-09-19 | Landslide vulnerability prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311203712.9A CN116933091B (en) | 2023-09-19 | 2023-09-19 | Landslide vulnerability prediction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116933091A true CN116933091A (en) | 2023-10-24 |
| CN116933091B CN116933091B (en) | 2024-02-27 |
Family
ID=88384737
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311203712.9A Active CN116933091B (en) | 2023-09-19 | 2023-09-19 | Landslide vulnerability prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116933091B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100982447B1 (en) * | 2010-03-03 | 2010-09-16 | 한국지질자원연구원 | Landslide occurrence prediction system and predicting method using the same |
| US20210220739A1 (en) * | 2020-01-22 | 2021-07-22 | Electronic Arts Inc. | Character controllers using motion variational autoencoders (mvaes) |
| CN114298436A (en) * | 2021-12-31 | 2022-04-08 | 西安交通大学 | Landslide susceptibility evaluation method based on Bagging strategy |
| US20220138425A1 (en) * | 2020-11-05 | 2022-05-05 | Adobe Inc. | Acronym definition network |
| CN115495555A (en) * | 2022-09-26 | 2022-12-20 | 中国科学院深圳先进技术研究院 | Document retrieval method and system based on deep learning |
| CN116664956A (en) * | 2023-07-08 | 2023-08-29 | 南京信息工程大学 | Image recognition method and system based on multi-task automatic encoder |
-
2023
- 2023-09-19 CN CN202311203712.9A patent/CN116933091B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100982447B1 (en) * | 2010-03-03 | 2010-09-16 | 한국지질자원연구원 | Landslide occurrence prediction system and predicting method using the same |
| US20210220739A1 (en) * | 2020-01-22 | 2021-07-22 | Electronic Arts Inc. | Character controllers using motion variational autoencoders (mvaes) |
| US20220138425A1 (en) * | 2020-11-05 | 2022-05-05 | Adobe Inc. | Acronym definition network |
| CN114298436A (en) * | 2021-12-31 | 2022-04-08 | 西安交通大学 | Landslide susceptibility evaluation method based on Bagging strategy |
| CN115495555A (en) * | 2022-09-26 | 2022-12-20 | 中国科学院深圳先进技术研究院 | Document retrieval method and system based on deep learning |
| CN116664956A (en) * | 2023-07-08 | 2023-08-29 | 南京信息工程大学 | Image recognition method and system based on multi-task automatic encoder |
Non-Patent Citations (2)
| Title |
|---|
| LILI CHANG等: "Landslide susceptibility evaluation and interpretability analysis of typical loess areas based on deep learning", NATURAL HAZARDS RESEARCH, vol. 3, no. 2, pages 155 - 169 * |
| 王功健: "基于稀疏自编码和深度神经网络的多维信号预测模型研究与应用", 中国优秀硕士学位论文全文数据库基础科学辑, vol. 03, pages 13 - 15 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116933091B (en) | 2024-02-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hong et al. | Landslide susceptibility assessment at the Wuning area, China: A comparison between multi-criteria decision making, bivariate statistical and machine learning methods | |
| Dou et al. | Assessment of advanced random forest and decision tree algorithms for modeling rainfall-induced landslide susceptibility in the Izu-Oshima Volcanic Island, Japan | |
| Du et al. | Landslide susceptibility mapping using an integrated model of information value method and logistic regression in the Bailongjiang watershed, Gansu Province, China | |
| Loche et al. | Landslide susceptibility maps of Italy: Lesson learnt from dealing with multiple landslide types and the uneven spatial distribution of the national inventory | |
| Li et al. | GIS‐based landslide susceptibility mapping using information, frequency ratio, and artificial neural network methods in Qinghai Province, northwestern China | |
| CN115688404B (en) | Rainfall landslide early warning method based on SVM-RF model | |
| CN113343563B (en) | Landslide susceptibility evaluation method based on automatic sample selection and surface deformation rate | |
| Alqadhi et al. | Selecting optimal conditioning parameters for landslide susceptibility: an experimental research on Aqabat Al-Sulbat, Saudi Arabia | |
| CN110991497A (en) | Urban land use change cellular automata simulation method based on BSVC (binary coded VC) method | |
| Maucha et al. | European validation of GMES FTS soil sealing enhancement data | |
| CN104809724A (en) | Automatic precise registration method for multiband remote sensing images | |
| CN105225227A (en) | The method and system that remote sensing image change detects | |
| CN111539904B (en) | Disaster vulnerability prediction method based on rainfall | |
| Fanos et al. | A spatial ensemble model for rockfall source identification from high resolution LiDAR data and GIS | |
| CN107392122A (en) | Polarization SAR silhouette target detection method based on multipolarization feature and FCN CRF UNEs | |
| CN114036841A (en) | Landslide incidence prediction method and system based on semi-supervised support vector machine model | |
| CN103163064A (en) | Fuzzy prediction method of corrosion rate of transformer substation grounding grid | |
| KR20200052398A (en) | Method and apparatus for landslide susceptibility mapping using machine-learning architecture | |
| CN116933091B (en) | Landslide vulnerability prediction method and device | |
| CN115619275B (en) | Ecological restoration decision method and device for railway engineering | |
| CN116384627A (en) | Geological disaster evaluation method based on machine learning | |
| Abujayyab et al. | A new framework for geospatial site selection using artificial neural networks as decision rules: a case study on landfill sites | |
| CN116091850A (en) | Mining area land coverage classification model establishment and classification method | |
| Wang et al. | Modeling errors in remote sensing image classification | |
| CN114997666A (en) | Method for evaluating easiness of region debris flow |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |