KR101588232B1 - Landslide Prediction System using Geographic Information System and NeuroFuzzy techniques and Landslide Prediction Method using Thereof - Google Patents

Abstract

A landslide prediction system based on a geographic information system and a neuro fuzzy technique, and a method of predicting a landslide using the system.
The landslide prediction system based on the geographic information system and the neuro fuzzy technique is a geographic information system (GIS) which digitizes and stores at least one map of slope, lipid, line structure, clinical, and soil, A landslide space database 110 of the environment; A landslide spatial information database 110, a landslide spatial location database 110, a landslide spatial location database 110, and landslide location information providing units 110. The landslide spatial location database 110 stores the landslide location information, A landslide prediction analyzing unit 120 for predicting and analyzing landslide occurrence points using a neuro fuzzy technique for learning by an algorithm; A landslide prediction diagram generation unit 130 for generating a landslide prediction diagram based on the information analyzed by the landslide prediction analysis unit 120; A landslide prediction degree verifying unit 140 that quantitatively verifies the prediction degree created by the landslide prediction diagram creating unit using the existing landslide location; And a result / comparison deriving unit 150 for comparing the verification results verified by the landslide prediction degree verifying unit 140.

Description

(Landslide Prediction System using Geographic Information System and Neurofuzzy Techniques and Landslide Prediction Method using Thereof)

The present invention relates to a system and method for predicting a landslide in a specific area, and more particularly, to a landslide prediction system based on a geographic information system and a neuro fuzzy technique and a landslide prediction method using the same.

Landslides are a major risk factor for the national people 's life safety and are a national problem causing economic loss and property damage. Landslides in natural slopes are mainly caused by complex correlations of environmental factors due to heavy rainfall and earthquakes. Therefore, in order to mitigate the occurrence of landslides and prevent them in advance, it is necessary to study the causes and patterns of landslides and quantitative prediction of landslide prediction.

Predictive analysis requires accurate data and appropriate analysis methods for useful decision making. A variety of existing probabilistic, statistical, data mining analysis corporation discrimination analysis, multivariate analysis, frequency ratio, weight of evidence (WOE), logistic regression analysis, artificial neural network and fuzzy model are widely used in landslide prediction analysis such as GIS .

However, the neuro-fuzzy system is a new method that is applied to the artificial neural network using the learning algorithm and the fuzzy function together.

Accordingly, the present invention proposes a landslide prediction system based on a geographic information system and a neuro fuzzy technique capable of predicting landslides in a specific area by using a geographic information system and a neuro fuzzy technique, and a landslide prediction method using the same.

Korean Patent Publication No. 10-2009-0102311

A problem to be solved by the present invention is to provide a geographical information system capable of predicting and analyzing quantitative landslides through neuro fuzzy analysis based on landslide location information, terrain, lipid, clinical, and soil data, Prediction system and a method of predicting landslide using the same.

According to an embodiment of the present invention, there is provided a landslide prediction system based on a geographic information system and a neuro fuzzy technique, wherein the landslide occurrence location information and the indicators related to the occurrence of the landslide, inclination, A landslide space database (110) of a geographic information system (GIS) environment that provides at least one geological information; After calculating the spatial correlation between the landslide occurrence location information provided from the landslide space database 110 and the indicating factors as a probability value, an if-then fuzzy rule derived from the calculated probability value is applied to each input data, A landslide prediction analysis unit 120 for predicting and analyzing a landslide occurrence point by applying the at least one neuro fuzzy function having a weight value; A landslide prediction chart generation unit 130 for generating a landslide prediction chart based on the analyzed information by the landslide prediction analysis unit 120; A landslide prediction degree verifying unit 140 for comparing quantitatively the matching accuracy by comparing the occurrence position information of the existing landslide with the prediction degree created by the landslide prediction degree creating unit 130; And a result / comparison deriving unit (150) for comparing the verification results verified by the landslide prediction degree verifying unit (140) for each of the at least one or more neuro fuzzy functions.

According to an embodiment of the present invention, there is provided a method of predicting a landslide using a landslide prediction system based on a geographic information system and a neuro fuzzy technique. The landslide occurrence location information and the indicators related to the occurrence of the landslide, (S110) of constructing a landslide space database of a geographic information system (GIS) environment in which geographical information including at least one of soil, Calculating a spatial correlation between the landslide occurrence location information and the indicator factors as a probability value, applying an if-then fuzzy rule derived from the calculated probability value to each input data, and generating at least one neural network A landslide prediction analysis step (S120) for predicting and analyzing a landslide occurrence point by applying it to fuzzy functions; A landslide prediction diagram generation step (S130) of creating landslide prediction maps based on the predicted and analyzed information in the landslide prediction analysis step (S120); A landslide prediction degree verification step (S140) of comparing the occurrence position information of the existing landslide with the landslide prediction level created in the landslide prediction level generation unit (130) and quantitatively verifying the matching accuracy; And a comparison and analysis step (S150) of comparing the verification results verified in the landslide prediction degree verifying step (S140) by the at least one neuro fuzzy function.

The landslide prediction and analysis step S120 may include a step S121 of deriving a spatial correlation between a location where a landslide occurs and a designating factor associated with the occurrence of a landslide in an if-then fuzzy rule; And applying the if-then fuzzy rule to at least one neuro-fuzzy function having different weight values (S122) by applying the if-then fuzzy rule to each input data.

In the landslide prediction diagram generation step S130, the if-then fuzzy rule is classified into a triangular membership function, a trapezoidal membership function, a Gaussian membership function, a generalized bell membership function, A step S131 of inputting the identified if ~ then fuzzy rule to the hybrid learning algorithm after identifying the membership function using at least one of a sigmoidal membership function and a polynomial membership function, ; A step (S132) of inputting quantitative weight values of factors related to landslide occurrence to each factor; And a step (S133) of superimposing and analyzing the factors on the basis of each function to prepare a predicted value of a landslide occurrence point on landslide prediction.

In the step S140, the success rate method (SRM) is used to quantitatively calculate the accuracy of the landslide prediction calculated for each function. The X-axis of the SRM classifies the landslide occurrence prediction value into a higher percentage And the y-axis is a rank value indicating the number of landslide occurrence locations in cumulative percentage. The quantitative accuracy value of the landslide prediction using the SRM is characterized in that the predicted landslide occurrence value is expressed as a ratio value of the landslide occurrence location per area.

The comparison and analysis step S150 is to use Area Under Curves (AUC) to compare and analyze the accuracy of the landslide prediction. The AUC is calculated as a product of the X-axis and the Y-axis of the SRM graph. And the accuracy of the landslide prediction is compared and analyzed.

According to the present invention, more accurate landslide occurrence can be predicted by performing neuro-fuzzy analysis using landslides and related indicators, and performing landslide prediction analysis according to each function using various membership functions.

1 is a block diagram illustrating a landslide prediction system based on a geographic information system and a neuro fuzzy technique according to an embodiment of the present invention.
Fig. 2 shows an example of storage of all the factors related to landslides stored in the landslide space database shown in Fig.
3 is a detailed view of the landslide prediction analysis unit shown in FIG.
FIG. 4 is a flowchart for explaining a landslide prediction method using a landslide prediction system based on the geographic information system and the neuro fuzzy technique shown in FIG. 1. FIG.
FIG. 5 illustrates landslide occurrence locations for landslide prediction and landslide prediction analysis according to the present invention.
6A is an exemplary view showing the structure of a neuro-purge system.
6B shows an analysis flow of the neuro-fuzzy technique in a GIS environment.
FIG. 6C is a table illustrating the spatial correlation between the landslide occurrence and the related indicators, and each factor is applied to the neuro-fuzzy system based on the likelihood ratio value.
Figures 6d and 6e are tables showing initial Membership Functions (MFs) and variables (indicators associated with landslides) of "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal, Polynomial" And FIG. 6F is a table showing the training parameters and the structure of the neuro-fuzzy system.
FIG. 6G is a graph showing training RMSE (Root Mean Square Error) values for the membership functions "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal", and "Polynomial".
7A to 7F show landslide prediction diagrams using the membership functions "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal", and "Polynomial" of each neuro fuzzy.
Fig. 8 is a graph showing the accuracy of the landslide prediction derived from the neuro fuzzy membership functions "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal", and "Polynomial".

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a detailed description of preferred embodiments of the present invention will be given with reference to the accompanying drawings. In the following description of the present invention, detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

Embodiments in accordance with the concepts of the present invention can make various changes and have various forms, so that specific embodiments are illustrated in the drawings and described in detail in this specification or application. It is to be understood, however, that it is not intended to limit the embodiments according to the concepts of the present invention to the particular forms of disclosure, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between. Other expressions that describe the relationship between components, such as "between" and "between" or "neighboring to" and "directly adjacent to" should be interpreted as well.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises ", or" having ", or the like, specify that there is a stated feature, number, step, operation, , Steps, operations, components, parts, or combinations thereof, as a matter of principle.

The present invention provides a landslide prediction system based on a geographic information system and a neuro fuzzy technique and a method using the same. In particular, in neuro fuzzy analysis, at least one membership function among various membership functions Triangular, Trapezoidal, Gaussian, Generalized bell, Sigmoidal, And the analysis of landslide prediction is applied to the analysis.

Hereinafter, a description will be given of a landslide prediction system based on a geographic information system and a neuro fuzzy technique according to an embodiment of the present invention and a method of predicting a landslide using the same.

1 is a block diagram illustrating a landslide prediction system based on a geographic information system and a neuro fuzzy technique according to an embodiment of the present invention.

Fig. 2 shows an example of storage of all the factors related to landslides stored in the landslide space database shown in Fig.

3 is a detailed view of the landslide prediction analysis unit shown in FIG.

FIG. 4 is a flowchart for explaining a landslide prediction method using a landslide prediction system based on the geographic information system and the neuro fuzzy technique shown in FIG. 1. FIG.

FIG. 5 illustrates a topographical map of a landslide occurrence area according to the present invention and a location of occurrence of a landslide for prediction analysis of a landslide.

6A is an exemplary view showing the structure of a neuro-purge system.

6B shows an analysis flow of the neuro-fuzzy technique in a GIS environment.

FIG. 6C illustrates the spatial correlation between the occurrence of landslides and the indicators related thereto as a table of likelihood ratios, and each factor is applied to the neuro-fuzzy system based on the likelihood ratio value.

Figures 6d and 6e show the initial membership functions (MFs) and variables (indicators associated with landslides) of "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoida", "Polynomial" And FIG. 6F is a table showing the training parameters and the structure of the neuro-fuzzy system.

FIG. 6G is a graph showing training RMSE (Root Mean Square Error) values for the membership functions "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal", and "Polynomial".

7A to 7F show landslide prediction diagrams using the membership functions "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal", and "Polynomial" of each neuro fuzzy.

Fig. 8 is a graph showing the accuracy of the landslide prediction derived from the neuro fuzzy membership functions "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal", and "Polynomial".

1, a landslide prediction system 100 based on a geographic information system and a neuro fuzzy technique according to an embodiment of the present invention includes a landslide space database 110, a landslide prediction analysis unit 120, A landslide prediction degree verifying unit 140, and a result / comparison deriving unit 150. [0035]

The landslide space database 110 stores at least one of map information of inclination, geology, line structure, clinic, and soil, which is an indicator related to occurrence of a landslide, in a numerical state, and a geographic information system (GIS) .

The landslide prediction analyzer 120 analyzes the landslide occurrence location prediction information by applying the landslide occurrence location information provided from the landslide space database 110 and the indicator factors neuro fuzzy technique.

The landslide prediction diagram generation unit 130 creates a landslide prediction diagram based on the information analyzed by the landslide prediction analysis unit 120.

The landslide prediction degree verifying unit 140 performs a function of quantitatively verifying the landslide prediction degree created in the landslide prediction diagram creating unit using the existing position of the landslide.

The comparison / result deriving unit 150 performs a function of comparing the verification results verified by the landslide prediction degree verifier 140.

More specifically, the landslide space database 110 is classified into a plurality of sub-databases. The sub-database includes landslide locations identified through aerial photograph analysis and field survey of the analysis target area, Extracted slope, geology extracted from the geological map of the area to be analyzed, and data related to the line structure extracted from the IRS satellite image are stored in a script form.

In addition, it can further comprise the soil extracted from the soil map of the clinical area and the analysis area extracted from the resolution map of the analysis target area.

3, the landslide prediction analysis unit 120 includes a triangular membership calculation unit 121, a trapezoidal membership calculation unit 122, a Gaussian membership calculation unit 123, a generalized bell membership calculation unit 124, a Sigmoidal The membership calculation unit 125 and the polynomial membership calculation unit 126 are provided and the functions calculated in at least one of the enumerated calculation units 121, 122, 123, 124, 125, and 126 are applied, That is the point of analysis.

FIG. 4 is a flowchart for explaining a landslide prediction method using a landslide prediction system based on the geographic information system and the neuro fuzzy technique shown in FIG. 1. FIG.

4, a landslide prediction method S100 according to an embodiment of the present invention includes a landslide space database construction step S110, a landslide prediction analysis step S120, a landslide prediction diagram generation step S130, (S140) and a comparative analysis step (S150).

The construction step S110 of the landslide space database may be a step of digitizing and storing at least one map of an inclination, a geology, a line structure, a clinic, and a soil, which are indicators related to the occurrence of landslides provided from the outside.

In the landslide prediction analysis step S120, the spatial correlation between the landslide occurrence location information and the indicating factors is calculated as a probability value, and an if-then fuzzy rule derived from the calculated probability value is applied to each input data, Applying at least one neuro-fuzzy function having different weight values to predict and analyze the landslide occurrence point.

The landslide prediction diagram generation step S130 may be a step of creating a landslide prediction map based on the predicted and analyzed information in the landslide prediction analysis step S120.

The landslide prediction degree verification step S140 may be a step of quantitatively verifying the matching accuracy by comparing the occurrence position information of the existing landslide with the landslide prediction degree created by the landslide prediction level generation unit 130. [

The comparison and analysis step S150 may be a step of comparing and analyzing the verification results verified in the landslide prediction degree verification step S140 for each of the at least one or more neuro fuzzy functions.

More specifically, the landslide prediction analysis step (S120) includes a step of deriving a spatial correlation between a location where a landslide occurs and a designating factor associated with occurrence of a landslide to an if-then fuzzy rule, Applying to the input data at least one neuro-fuzzy function having different weight values.

In the landslide prediction diagram generation step S130, the if-then fuzzy rule is classified into a triangular membership function, a trapezoidal membership function, a Gaussian membership function, a generalized bell membership function, A fuzzy membership function and a polynomial membership function, and inputting the identified if-then fuzzy rule to the hybrid learning algorithm, Inputting a quantitative weight value of a factor related to occurrence of a landslide occurrence into each factor, and superimposing each of the factors on the basis of each function to prepare a predicted value of a landslide occurrence point on a landslide prediction road.

In the step S140, the success rate method (SRM) is used to quantitatively calculate the accuracy of the landslide prediction calculated for each function. The X-axis of the SRM classifies the landslide occurrence prediction value into a higher percentage And the y-axis is a rank value indicating the number of landslide occurrence locations in cumulative percentage. The quantitative accuracy value of the landslide prediction using the SRM may be a step of representing the landslide occurrence prediction value as a ratio value of landslide occurrence position per area.

The comparison and analysis step S150 is to use Area Under Curves (AUC) to compare and analyze the accuracy of the landslide prediction. The AUC is calculated as a product of the X-axis and the Y-axis of the SRM graph. The accuracy of the landslide prediction can be compared and analyzed.

FIG. 5 illustrates a topographic map and a landslide occurrence location of a bow region where landslides occur most frequently as an example according to the present invention, and a prediction method of the present invention will be described with reference to an analysis embodiment of the present invention. The Boeun area is located in geographical coordinates at latitude 36 ° 26 '23.80 "~ 36 ° 28' 54.74", longitude 127 ° 40 '02.06 "~ 127 ° 44" 54.05 ".

The landslide distribution, topographic map, soil map, clinical map, and geologic map were collected for the prediction analysis of landslide occurrence in Boeun area. The location of the landslide occurred using the aerial photographs, and the location of the landslide was recorded using GPS.

In the topographic map, the slope was extracted, in the soil map, in the soil map, in the resampling map, in the clinical map, in the geological maps, and in the IRS satellite images. All the factors related to landslide were constructed as spatial database in GIS environment and set to 5m x 5m grid considering the scale of input data and the number of grid in study area is 2,735,776.

In landslide prediction analysis, existing landslide location plays an important role in understanding the environment where landslide can occur in the future. Therefore, it is possible to obtain the best analytical result by precise landslide location.

Field trips to landslides can identify the most accurate landslide location. However, it is difficult or impossible to conduct a site survey in mountainous areas where it is time-consuming and expensive to find landslides in the first place, and where access is difficult. Therefore, it is impossible to survey all landslides in the vicinity of roads and accessible areas. Therefore, when neurofuzzy analysis is performed, it is impossible to input all data and it is difficult to expect accurate results. In addition, the accuracy of the mapping is lowered because the map is roughly mapped on the map without performing the actual surveying. To overcome these shortcomings, we used GPS and improved accuracy and reliability using 1: 5,000 large scale maps.

In this experiment, these field surveys were conducted primarily for aerial photographs after aerial photo analysis. Landslide location analysis through aerial photograph analysis has the advantage that the location of landslides can be accurately detected in a relatively short time.

However, in Korea, it is difficult to match with the occurrence of landslides because aerial photographs are taken every five years for the purpose of making maps. Furthermore, it is costly to conduct aerial photographs to confirm the location of landslides.

However, when a landslide occurs, the trail usually lasts for more than 10 years, so landslides that have not been restored are not a timely problem. In addition, scanned and geometrically corrected aerial photographs can be more precisely identified by landslides compared to 1: 5,000 digital topographic maps.

However, for aerial photographs, it is a bit cumbersome to scan first, difficult to correct geometry, and also to acquire aerial photographs through the National Geographic Information Center. Moreover, it is more difficult to obtain all aerial photographs before and after landslides.

In the present invention, landslides occurred in the Boeun area in 1998 were analyzed using photographs taken in 1995 and 1999. The landslide coordinates were extracted from 377 landslides that were read through these procedures and they were constructed as point - based spatial data in ArcGIS 9.0 environment.

6A and 6B show the configuration of a landslide prediction analysis unit.

FIG. 6A shows a structure of a typical neuro-fuzzy system composed of five layers, and the hybrid learning algorithm used in the system is an example of adjusting the parameters to match the results with the training data.

FIG. 6B is a flowchart illustrating a method of predicting landslide prediction using a neuro-fuzzy system analyzed by a combination of an adaptive network, a probabilistic model, and a fuzzy interface system in a GIS environment.

Especially, in the landslide prediction map, various membership functions "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal", "Polynomial" And is verified using the location of occurrence.

Then, the accuracy of the landslide prediction for each membership function can be used to determine the membership function that is most suitable for the prediction analysis.

As shown in the figure, the landslide prediction method of the present invention can be applied not only to landslide prediction analysis but also to prediction of geological phenomena occurring in a geospatial space such as various natural disasters, geological resources, environmental pollution, etc. as GIS.

Hereinafter, the neuro-fuzzy technique, the landslide prediction analysis unit, the landslide prediction diagram creation unit, the landslide prediction verification unit, and the result / comparison unit will be described in more detail.

1) Neuro fuzzy theory

The neuro-fuzzy technique is a multi-layer feed-forward network with reference to 6a, a learning algorithm for artificial neural networks, and a fuzzy inference function for multiple inputs and single results.

,

,

,

과 4개의 퍼지 If-then 규칙 집합을 갖는다(Sugeno, 1985).
This is a fuzzy inference system and is implemented in artificial neural networks. A typical neuro-fuzzy system consists of two inputs x and y and four MFs (Membership Functions)

,

,

,

And four fuzzy If-then rule sets (Sugeno, 1985).

[Formula 1]

,

및

(

)는 If-then 규칙 집합을 통해 계산된 결과의 상수들이다.here,

,

And

(

) Are constants of the result calculated through the If-then rule set.

A typical neuro-fuzzy structure consists of five layers, each of which plays a different role. The detailed description is as follows (Wang and Elhag, 2008).

First layer: All nodes in this layer are adaptive nodes and determine the degree of membership of inputs. The results of this layer are given by the following equation.

[Formula 2]

[Formula 3]

와

는 적절한 MFs(Membership Functions)에 의해 낮거나 혹은 높은 특징을 가지는 퍼지-집합(fuzzy-set)들이며, MFs는 "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal", "Polynomial" 또는 그 밖에 다른 형태의 모양으로써 표현되어진다.Where x and y are explicit inputs,

Wow

Are fuzzy-sets with low or high characteristics by appropriate MFs (Membership Functions) and MFs are "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal"Quot; Polynomial "or any other form of shape.

Second layer: The nodes in this layer are marked as fixed nodes and perform simple multiplication operations. The results of this layer are described as follows.

[Formula 4]

[Equation 4] is expressed by the starting strength for each rule, i.e., the starting strength represents the degree to which the preceding part of the rule is satisfied.

기호로 표시되며, 네트워크상에서 정규화 역할을 수행한다. 이 층에서의 결과는 다음과 같이 표현되어진다.Third layer: Nodes at this layer are also called fixed nodes

Symbol, and performs a normalization function on the network. The results at this layer are expressed as follows.

[Formula 5]

[Equation 5] is called the normalized starting strength.

Fourth layer: Each node in this layer is an adaptive node and its result is represented by a simple multiplication of the normalized starting intensity and the first order polynomial (the first Sugeno model), so the results of this layer are given as .

[Formula 6]

The variables in this layer are referred to as the resulting variables for [Equation 6].

  5th floor: A single node in this layer is represented by a symbol as a fixed node and calculates the overall result as the sum of all incoming signals.

[Equation 7]

[Equation 7] is a linear combination of the variables that result when the values of all variables are fixed. This can be observed when the neuro-purge system structure has two adaptive layers (1 and 4 layers).

과

을 갖으며, 4층은 1차 다항식에 속하는 수정할 수 있는 변수들

을 갖는다. 이러한 뉴로퍼지 시스템 구조를 위한 학습 알고리즘은 뉴로퍼지 시스템 결과를 훈련 자료와 맞추기 위해 모든 수정할 수 있는 변수들을 조정한다. 이러한 수정할 수 있는 변수들은 역전파 알고리즘을 단독으로 사용하여 조절되어지거나, 혹은 최소 제곱 형태의 방법을 사용하여 조절되며, 이러한 두 가지 방법을 조합하여 사용한 것을 "Hybrid learning algorithm"으로 알려져 있다. 그러므로 그 결과로서 일어나는 변수들은 하나의 값으로 고정되어지며, 오류 신호들을 줄이고 또한 전제된 변수들은 Gradient descent 방법에 의해 경신되어진다.
The first floor is the variable (s) associated with the input MFs (Membership Functions)

and

And the fourth layer is a variable that can be modified to belong to the first order polynomial

Respectively. The learning algorithm for this neuro-fuzzy system structure adjusts all the modifiable parameters to match the neuro-fuzzy system results with the training data. These modifiable variables are controlled by using the backpropagation algorithm alone, or by using the least squares method. The combination of these two methods is known as the "Hybrid learning algorithm". Therefore, the resulting variables are fixed to one value, reducing the error signals and also the prerequisite variables are updated by the gradient descent method.

2) Landslide Prediction Analysis - Applying the likelihood ratio and neuro fuzzy technique

The spatial correlation between the factors related to landslide occurrence was calculated using the likelihood ratio model. The likelihood ratio means the ratio of landslide occurrence area of each factor based on the conditional probability principle, and is shown in FIG. 6c. To calculate the initial value, the likelihood value for each factor was added to calculate the Landslide Susceptibility Index (LSI) value, which is shown in Equation 8.

 [Equation 8]

은 변수들이며,

은 각각의 요인들의 형태 및 범위에 대한 우도비(likelihood ratio) 값을 나타낸다.
here

Are variables,

Represents the likelihood ratio value for the type and extent of each factor.

The variables can be obtained as follows.

 [Equation 9]

 Where p represents a variable.

As a result, each variable is calculated by each calculated factor. Each of the variables is recalculated by a neuro-fuzzy system, and a conceptual diagram of the neuro-fuzzy system in a GIS environment is shown in Fig. 6B.

The system of the present invention for predicting the occurrence of landslides is designed using the relationship with the factors related to landslide occurrence.

Slope, lipid, line structure, clinic, and soil were considered as input factors for landslide related factors. In the neuro-fuzzy system, the "Hybrid learning algorithm" was applied to induce effective results. MFs (Membership Functions) with 6 forms were applied to each of 5 inputs to establish the neuro-fuzzy system, if-then 'rule was derived.

Figures 6d and 6e show the initial Memberships Functions and variables such as "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal", "Polynomial" The system structure is shown in Fig.

This model is designed to complete the most optimal learning when RMSE (Root Mean Square Error) value satisfies 0.0000001 or 500 repetition times. After the final training, MFs (Membership Functions) such as "Triangular", "Trapezoidal", "Gaussian", "Generalized bell", "Sigmoidal" and "Polynomial" and RMSE (Membership Functions) for MFs Mean Square Error) values were 0.00035, 0.0034, 0.0024, 0.0023, 0.0023 and 0.0028.

3) Landslide forecasting department

Slope, lipid, line structure, clinical and soil parameters were calculated using a neurofuzzy system evaluated to calculate the Landslide Susceptibility Index (LSI).

After this process, the landslide predictability for each of the MFs (Membership Functions) types was calculated using the landslide predictive analysis index.

Fig. 7A is a graphical MFs, Fig. 7E is a sigmoidal MFs, and Fig. 7F is a graphical example using "Polynomial MFs". In Fig. 7A, "Triangular MFs", "Trapezoidal MFs", "Gaussian MFs" And the landslide predictions derived from these maps.

7A to 7F, the landslide prediction index was created using the landslide prediction analysis index value, and the index values were classified into five classes using the equal area method.

4) Landslide Prediction Verification Department

GIS and neuro - fuzzy techniques were used to generate landslide predictions for specific areas and verified using SRM (Success Rate Method) and AUC (Area Under Curves) method. The SRM (Success Rate Method) graph describes how accurate the predicted degree is. In order to verify the landslide predictability, the landslide prediction indexes of all the pixels in the study area are sorted in descending order. All the sorted pixel values are classified into 100 grades and accumulated 1% intervals.

FIG. 8 is a summary of an SRM (Success Rate Method) and an AUC (Area Under Curve) for each MFs. As a result, in terms of accuracy, the neuro - fuzzy system technique showed the best result using "Triangular MF", "Trapezoidal MF" and "Polynomial MF" as 84.96%.

In the foregoing detailed description of the present invention, specific examples have been described. However, various modifications are possible within the scope of the present invention. The technical idea of the present invention should not be limited to the embodiments of the present invention but should be determined by the equivalents of the claims and the claims.

100: Landslide prediction system
110: Landslide spatial database
120: Landslide Prediction Analysis Unit
121: Triangular membership function calculating unit
122: Trapezoidal membership function calculating unit
123: Gaussian membership function calculating unit
124: Generalized bell membership function calculating unit
125: Sigmoidal membership function calculation unit
126: Polynomial membership function calculating unit
130: Landslide prediction chart
140: landslide predictor verification unit
150: Result / comparison derivation unit

Claims (6)

delete In landslide prediction methods based on geographic information systems and neurofuzzy techniques,
Construction of landslide spatial database to construct landslide spatial database of geographic information system (GIS) environment where geological information including landslide occurrence location information and at least one of slope, lipid, line structure, clinical, Step S110:
The landslide prediction analysis unit 120 may calculate the spatial correlation between the landslide occurrence location information and the indicator factors as a probability value and apply the derived if to then fuzzy rule to each input data having different weight values, A landslide prediction analysis step (S120) for predicting and analyzing landslide occurrence points using a fuzzy function;
A landslide prediction diagram generation step (S130) of creating a landslide prediction diagram based on information predicted and analyzed in the landslide prediction analysis step (S120);
A landslide prediction degree verification step (S140) for comparing the existing landslide occurrence location information with the landslide prediction degree created in the landslide prediction degree generation step (S130) and quantitatively verifying the matching accuracy; And
And comparing and verifying the verification results verified by the result / comparison deriving unit 150 for each of at least one or more neurofuzzy functions (S150) Based landslide prediction method.
3. The method of claim 2,
In the landslide prediction analysis step S120,
(S121) the spatial correlation between the location where the landslide occurred and the indicators related to the landslide occurrence to the if ~ then fuzzy rule; And
Applying the if-then fuzzy rule to each input data, and applying the at least one fuzzy rule to at least one neuro-fuzzy function having different weight values (S122) based on the neuro-fuzzy technique. Landslide prediction method.
3. The method of claim 2,
The landslide prediction diagram generation step (S130)
The above-described if-then fuzzy rules are classified into a triangular membership function, a trapezoidal membership function, a Gaussian membership function, a generalized bell membership function, a sigmoidal membership function, A membership function, and inputting the identified if-then fuzzy rule into a hybrid learning algorithm at step S131;
A step (S132) of inputting quantitative weight values of factors related to landslide occurrence to each factor; And
And a step (S133) of overlapping analysis of the factors for each function to form a predicted value of a landslide occurrence point at a landslide prediction (S133), and a landslide prediction method based on the neuro fuzzy technique.
3. The method of claim 2,
The landslide prediction degree verification step (S140)
The success rate method (SRM) is used to quantitatively calculate the accuracy of the landslide prediction calculated for each function. The X-axis of the SRM is a value obtained by grading the landslide occurrence prediction value as a higher percentage, and the Y- The number is a cumulative percentage of the rating value. Wherein the quantitative accuracy value of the landslide prediction using the SRM indicates a predicted value of the occurrence of the landslide as a value of the position of the occurrence of the landslide per area, and the method of predicting the landslide based on the neuro fuzzy technique.
3. The method of claim 2,
The comparison and analysis step S150 is performed using area under curves (AUC) to compare and analyze the accuracy of the landslide prediction. The AUC is calculated as a product of the X-axis and the Y-axis of the SRM graph. A method of predicting landslides based on geographic information systems and neuro fuzzy techniques, characterized by comparing and analyzing the accuracy of landslide predictability.

Images