阅读说明：本技术 基于svc自动生成地震灾害风险评估中定性图件的方法及系统 (SVC-based method and system for automatically generating qualitative graph in earthquake disaster risk assessment ) 是由 陈小芳 戚洪飞 李三凤 刘辉 黄宽 俞岗 于 2020-05-21 设计创作，主要内容包括：本发明提供了一种基于SVC自动生成地震灾害风险评估中的定性图件的方法,其包括以下成图步骤：S1：基于系统获取第一成图数据。S2：将第一成图数据做归一化处理,得到第二成图数据。S3：用第二成图数据构建矩阵点,得到第三成图数据。S4：将第二成图数据作为训练样本,对第三成图数据进行训练,得到第四成图数据。S5：用第四成图数据构建栅格。S6：栅格矢量化,生成矢量分布图。将上述步骤编程集成,即构成系统中自动成图功能模块。本发明还提供了一种基于SVC自动生成地震灾害风险评估中定性图件的系统,其采用上述方法自动生成地震灾害风险评估中的定性图件,以对地震灾害风险进行评估。本发明的方法可以对定性的离散点进行更加合理的分类,而不是简单的使用具体数值进行插值而分类。这个成图过程可以修正插值分类的缺陷,为工程人员以及其他专业人员提供准确的评估图。此外,本发明的成图方法不再使用人工勾画,提高了成图效率与精度,降低了人工成本。(The invention provides a method for automatically generating a qualitative graph in earthquake disaster risk assessment based on SVC, which comprises the following steps of mapping: s1: first imaging data is acquired based on the system. S2: and normalizing the first imaging data to obtain second imaging data. S3: and constructing matrix points by using the second imaging data to obtain third imaging data. S4: and taking the second imaging data as a training sample, and training the third imaging data to obtain fourth imaging data. S5: the grid is constructed with the fourth imaging data. S6: and carrying out grid vectorization to generate a vector distribution map. The steps are integrated in a programming mode, and an automatic mapping function module in the system is formed. The invention also provides a system for automatically generating the qualitative map in the earthquake disaster risk assessment based on the SVC, and the method is adopted to automatically generate the qualitative map in the earthquake disaster risk assessment so as to assess the earthquake disaster risk. The method of the invention can carry out more reasonable classification on qualitative discrete points, rather than simply carrying out classification by interpolation by using specific numerical values. The mapping process can correct defects of interpolation classification and provide accurate evaluation maps for engineers and other professionals. In addition, the mapping method does not use manual drawing any more, thereby improving the mapping efficiency and precision and reducing the labor cost.)

1. A method for automatically generating a qualitative graph in earthquake disaster risk assessment based on SVC comprises the following steps of mapping:

s1: acquiring first imaging data based on a system;

s2: normalizing the first imaging data to obtain second imaging data;

s3: constructing matrix points by using the second imaging data to obtain third imaging data;

s4: training the third imaging data by taking the second imaging data as a training sample to obtain fourth imaging data;

s5: constructing a grid with the fourth mapping data;

s6: carrying out grid vectorization to generate a vector distribution map;

the steps are integrated in a programming mode, and an automatic mapping function module in the system is formed.

2. The method for SVC automatic generation of qualitative charts in seismic disaster risk assessment according to claim 1, characterized in that the first chart data comprises borehole origin point locations and corresponding property information.

3. The method for SVC automatic generation of qualitative graphs in earthquake disaster risk assessment according to claim 1 is characterized in that the X, Y axis data of the original point in the first graph data is normalized.

4. The method of claim 1, wherein a third mapping data is obtained by interpolating a construction matrix with an outer rectangle of an original point of the second mapping data as a boundary and a distance set after normalization, and the third mapping data is nxm matrix points, which is an SVC training model.

5. The method for automatically generating a qualitative graph in an SVC disaster risk assessment according to claim 1, wherein said fourth graph data is an SVC predictive model.

6. The method of claim 1, wherein the normalized drill point information is used as a training sample, the third imaging data is trained to generate an SVC prediction model, and the SVC prediction model is saved in a text file.

7. The method for automatically generating a qualitative graph in an SVC risk assessment according to claim 1, wherein the kernel function K in the training of the third graph data is selected from Gaussian kernel function and gamma is 50.

8. The method for SVC based automatic generation of qualitative graphs in earthquake disaster risk assessment according to claim 4 characterized in that N2000, M2000.

9. A qualitative graph automatically generated using the method of automatically generating a qualitative graph in an SVC risk assessment based on any of claims 1-8.

10. A system for automatically generating a qualitative graph using the method in SVC automatic generation of seismic disaster risk assessment based on any of claims 1-8.

Technical Field

The invention relates to the technical field of earthquake disaster risk assessment, in particular to a method and a system for automatically generating a qualitative map in earthquake disaster risk assessment based on SVC.

Background

China is a country with serious earthquake disaster loss in the world. Once an earthquake happens, very serious casualties and huge economic losses are caused. People's correct assessment and scientific management of earthquake disaster risks become the best way to achieve disaster reduction. For risk assessment of earthquake disasters, the chart can provide visual assessment results for people. The map in the earthquake disaster risk assessment is a field soil type distribution map, a field type distribution map, a sandy soil liquefaction distribution map, a soft soil seismic subsidence distribution map, and the like, and further includes various map of urban land generated from the maps and disaster maps similar to the map. The field soil type distribution map is generated according to the type of soil drilled at each discrete point of the research area, such as rock, hard soil or soft rock, medium hard soil, medium soft soil and soft soil types. The field category distribution map is generated based on the field category of the borehole at each discrete point in the area of interest, e.g. I₀、Ⅰ₁II, III, IV. The sand liquefaction profile is generated based on whether the borehole is liquefied (i.e., liquefied, not liquefied) or slightly liquefied, moderately liquefied, or heavily liquefied at each discrete point in the study area. The distribution diagram of soft soil seismic subsidence isAnd (4) generating a seismic trap (namely seismic trap and non-seismic trap) according to whether drilling holes at each discrete point of the research area are generated. The above figures are all of the planar type.

The commonly used mapping method at present is manually sketched according to the result of discrete points of a research area, and is time-consuming and labor-consuming. When the area of a research area is large, such as thousands of square kilometers, and the discrete points are unevenly distributed, the defects are more obvious. The other method is an evaluation method which forcibly gives 0 and 1 numerical values to the discrete point attribute and then carries out interpolation gridding, namely, a grid is generated by interpolation (adopting inverse distance ratio) of the discrete points, and the grid is classified and output as a vector diagram according to the grid.

Disclosure of Invention

In order to overcome the defects of the existing analysis and evaluation method and solve the defects in the prior art, the invention provides an automatic mapping method and system based on SVC (support vector classification), which has a perfect mapping process taking SVC as a classification tool and embodies the technical advantages of less sample points and no density of SVC.

The invention provides a method for automatically generating a qualitative graph in earthquake disaster risk assessment based on SVC, which comprises the following steps of mapping:

s1: first imaging data is acquired based on the system. The first mapping data comprises original point positions of the drill hole and corresponding attribute information. Preferably, the first imaging data is saved in a text file in the format of a record X, Y, Z for each line, where the X and Y axes represent position coordinates and Z is a classification and must be an integer value (INT).

S2: and normalizing the first imaging data to obtain second imaging data. Preferably, the X, Y axis data of the original point in the first mapping data is normalized.

S3: and constructing matrix points by using the second imaging data to obtain third imaging data. Preferably, interpolation is performed by taking an outsourcing rectangle of the original point location of the second mapping data as a boundary and a set interval after normalization to construct matrix points, so as to obtain third mapping data, where the third mapping data are nxm matrix points, and are the SVC training model. The ranges of N and M need to be determined empirically, typically within 2000 to balance efficiency with resultant fidelity. M, N, the larger the value, the denser the corresponding grid, the closer the SVC prediction is to the trusted data, but the more computationally intensive it is. Preferably, N is 2000 and M is 2000.

S4: and taking the second imaging data as a training sample, and training the third imaging data to obtain fourth imaging data. Preferably, the fourth graph data is an SVC prediction model, and the SVC prediction model is saved as a text file. Preferably, the normalized drilling point information is used as a training sample, the SVC training model is trained to obtain an attribute value of the N × M matrix prediction point, the SVC prediction model is generated, and the SVC prediction model is stored in a text file. Preferably, the kernel function K in the training SVC model is a gaussian kernel function.

S5: the grid is constructed with the fourth imaging data. Preferably, the grid is constructed with X, Y, Z data of the fourth graphing data.

S6: and carrying out grid vectorization to generate a vector distribution map. Preferably, the estimation range is used for clipping, and finally, a distribution diagram in the designated range is generated.

The steps are integrated in a programming mode, and an automatic mapping function module in the system is formed.

The invention also provides a system for automatically generating the qualitative map in the earthquake disaster risk assessment based on the SVC, and the method is adopted to automatically generate the qualitative map in the earthquake disaster risk assessment so as to assess the earthquake disaster risk.

The SVC-based method for automatically generating the qualitative map in the earthquake disaster risk assessment can classify qualitative discrete points more reasonably, and does not simply use specific numerical values for interpolation for classification. The interpolation classification method depends on the density of discrete points, is very hard, easily weakens some non-prominent points, and classifies the points classified as A into B. The SVC-based method of the invention obtains an accurate evaluation graph by constructing a matrix, training an SVC model to correct attribute values of matrix points, and vectorizing a corrected matrix grid. The mapping process can correct defects of interpolation classification and provide accurate evaluation maps for engineers and other professionals. In addition, the mapping method does not use manual drawing any more, thereby improving the mapping efficiency and precision and reducing the labor cost.

Drawings

FIG. 1 is a flow chart illustrating a method for automatically generating a qualitative map in earthquake disaster risk assessment based on SVC according to the present invention;

FIG. 2 is a flow chart of a method for automatically generating a site soil type qualitative map in earthquake disaster risk assessment based on SVC in embodiment 1;

FIG. 3-1 shows a prior art generated plot of field soil type (with point location data) for the study area of example 1;

3-2 show a prior art generated plot field soil type profile (without point location data) for the study area of example 1;

3-3 show the field soil type profile (local magnification, with point location data) for the study area in example 1 generated by the prior art;

FIG. 4-1 shows the plot soil type distribution map (with point location data) of the study area in example 1 automatically generated based on SVC;

4-2 show the plot soil type profile (without point location data) for the study area in example 1 automatically generated based on SVC;

4-3 show the SVC-based automatically generated plot soil type distribution map (local magnification, with point location data) for the study area in example 1;

FIG. 5 is a view showing a program interface for automatically generating a field soil type distribution map based on SVC in example 1;

fig. 6 shows a flowchart of a method for automatically generating a soft soil seismic trapping qualitative map in earthquake disaster risk assessment based on SVC in embodiment 2;

FIG. 7-1 shows a prior art generated soft soil sag profile (with point location data) for the study area in example 2;

FIG. 7-2 shows a prior art generated soft soil sag profile (without point location data) for the study area of example 2;

7-3 show the soft soil sag profiles (local amplification with point location data) for the study area in example 2 generated by the prior art;

FIG. 8-1 shows the soft soil collapse distribution map (with point location data) of the research area in example 2 based on SVC automatic generation;

fig. 8-2 shows the soft soil collapse distribution map (without point location data) of the research area in example 2 based on SVC automatic generation;

8-3 show the soft soil sag distribution map (local magnification, with point location data) of the study area in example 2 based on SVC automatic generation;

fig. 9 shows a program interface for automatically generating a soft soil collapse distribution map based on SVC in example 2.

Detailed Description

SVM (Support Vector Machine) is a supervised learning model related to the associated learning algorithm, and can analyze data, recognize models, and use them for classification and regression analysis. Different models can be made according to different input data. When SVM is used for classification analysis, its input label is a classification value, which is also referred to as SVC at this time. SVC can be used not only for general linear classification, but also for some linearly indivisible problems, since some may be non-linearly separable. When using SVC for non-linear classification, it is often necessary to use a kernel function to assist in the classification. The kernel function can be constructed according to actual conditions, and the existing kernel function can also be selected. However, for a constructed kernel function, it is difficult to verify that it is a semi-positive definite matrix for any gram matrix in the input space, so a common choice is to use an off-the-shelf kernel function. When SVC classification analysis is carried out, the construction and selection of the kernel function can greatly simplify the complexity of calculation.

FIG. 1 shows a flow chart of the SVC-based method for automatically generating a qualitative graph in earthquake disaster risk assessment, which specifically comprises the following steps:

s1: acquiring first imaging data based on a system; the first mapping data comprises original point positions of the drill hole and corresponding attributesAnd (4) information. Preferably, the first imaging data is saved in a text file in the format of a record X, Y, Z for each line, where the X and Y axes represent position coordinates and Z is a classification and must be an integer value (INT). The Z value is obtained by the system through calculation according to a series of national standards and industrial specifications such as building earthquake-resistant design specification GB50011-2010, geotechnical engineering survey specification GB50021-2001(2009 edition) and geotechnical engineering survey specification JGJ83-2011 by utilizing actually measured information of the drilling hole, such as shear wave speed, standard penetration or stratum information. This original information is saved in a text file in the format of a record per line X, Y, Z. The classification information Z correspondingly generates attribute information of the map pieces such as a field soil type distribution map, a field type distribution map, a sandy soil liquefaction distribution map, a soft soil seismic subsidence distribution map and the like, for example, (1) the field soil type distribution map corresponds to the field soil type of each discrete point drilled hole in the research area: rock, hard or soft soil, medium hard soil, medium soft soil, soft soil; (2) the field category distribution map corresponds to the field category of each discrete point drilling in the research area: i₀、Ⅰ₁II, III, IV; (3) whether the sand liquefaction distribution diagram corresponds to each discrete point drilling hole of the research area is liquefied: liquefaction, no liquefaction, or liquefaction grade: light liquefaction, medium liquefaction, severe liquefaction; and (4) whether the drilling hole of each discrete point of the corresponding research area of the soft soil seismic subsidence distribution map is seismic subsided or not: collapse due to vibration and collapse due to vibration.

S2: and normalizing the first imaging data to obtain second imaging data. Preferably, the X, Y axis data of the original point in the first mapping data is normalized.

Because the coordinate values of the original data collected in step S1 are actual coordinates and have large values, the later SVC training model is abnormal and the classification effect is poor, so normalization is performed here, the default range is [ -1,1], and the grid spacing is more conveniently adjusted by a program.

S3: the second imaging data is used to construct matrix points. Preferably, interpolation is performed by taking an outsourcing rectangle of the original point location in the second mapping data as a boundary and a set distance after normalization to construct matrix points, so as to obtain third mapping data, where the third mapping data is N × M matrix prediction points, and is an SVC training model.

Interpolation is carried out according to the outsourcing range of the original coordinate point and the distance set after normalization in the step S2, an N multiplied by M matrix prediction point is constructed, the attribute value of the N multiplied by M matrix prediction point needs to be predicted, namely the range of the predicted value Z, N and M needs to be determined empirically, and balance can be obtained on efficiency and result fidelity within 2000 generally. N, M, the larger the mesh, the closer the SVC prediction is to the trusted data, but the more computationally intensive and time consuming.

S4: and taking the second imaging data as a training sample, and training the third imaging data to obtain fourth imaging data. Preferably, the fourth graph data is an SVC prediction model. Preferably, the normalized drilling point information is used as a training sample, the SVC training model is trained to obtain an attribute value of the N × M matrix prediction point, the SVC prediction model is generated, and the SVC prediction model is saved in a text file (txt format).

In SVC, a certain rule is used to map a sample that cannot be linearly segmented into a space with a higher latitude, and then find out a hyperplane, so that the core is to find out appropriate parameters, so that the distance between segmented hyperplanes is maximized, and data can be correctly classified. The kernel function K is K (x, y) < f (x), f (y) >, where x and y are input values for n dimensions, and f (x) and f (y) are n-dimensional to m-dimensional mappings (typically m > > n), < f (x), and f (y) > is the inner product of x and y. The kernel function includes a linear kernel function, a polynomial kernel function, a gaussian kernel function, and the like. In this step, python script code is used, which is packaged into exe by pyinstteller, invoked in a procedural manner in the. NET environment, which greatly simplifies the complexity of system integration.

In the case of dense and disordered points, a linear kernel function cannot be used, but a gaussian kernel function (kernel) is selected because training takes time and time is too expensive when a polynomial kernel function (kernel) is used. The gaussian kernel Function is also called Radial Basis Function (RBF) in SVM, and is the most dominant kernel Function of non-linear classification SVM, and the formula is

Due to K (x)⁽ⁱ⁾,x^(j))＝Ф(x⁽ⁱ⁾)^TФ(x⁽ⁱ⁾) Through proper mathematical transformation, the characteristic transformation function corresponding to the Gaussian kernel function can be obtained asThe accumulator physical meaning of the foregoing infinite polynomial is to convert the feature vector into an infinite multidimensional vector space, i.e., the gaussian function can expand the input feature into an infinite multidimensional space. The derivation of the formula would be to use taylor's formula.

Wherein K (x)⁽ⁱ⁾,x^(j)) Is a Gaussian kernel function, a_iOnly at the samples to which the support vector corresponds is not 0. It can be seen that, in predicting the linear combination of the gaussian functions with the central point at the support vector machine, the coefficient of the linear combination is a_iy⁽ⁱ⁾. Thus, the gaussian kernel function is also referred to as RBF kernel function, i.e. a linear combination of inverse bell-shaped functions.

When training an SVC with a Radial Basis (RBF) kernel, two parameters have to be considered: c penalty factor and gamma. The parameter C is used for balancing model accuracy and complexity, is in conflict with the simplicity of a decision surface, and can perform valuable conversion on the misclassification of the training sample. A smaller C makes the decision surface smoother, while a higher C aims at correctly classifying all training samples. Gamma defines how much a single training sample can play. Larger gamma will allow other samples to be affected even more. Intuitively, the gamma parameter defines how far the effect of a single training example is reached, with a low value meaning "far" and a high value meaning "close". The gamma parameter can be seen as the inverse of the radius of influence of the sample selected by the model support vector. This C-parameter corresponds the misclassification of the training examples to the simplicity of the decision surface. A low value of C smoothes the decision surface, while a high value of C aims at correctly classifying all training samples by giving the model freedom to choose more samples as support vectors.

S5: the grid is constructed with the fourth imaging data. Preferably, the grid is constructed with the X, Y, Z axis data of the fourth graphical data.

S6: and (4) carrying out grid vectorization on the grid constructed in the step (S5) to generate a vector distribution map. Preferably, the estimation range is used for clipping, and finally, a distribution diagram in the designated range is generated.

Because the matrix coordinates and the corresponding attribute value data of step S4 are trained by the SVC model, and the data is accurate, vectorizing the grid in step S5 is equivalent to including points of the entire evaluation range, thereby obtaining an evaluation map of the corresponding attribute of the entire evaluation range.

The steps are integrated in a programming mode, and an automatic mapping function module in the system is formed. When the graph is formed, the qualitative evaluation graph can be generated only by clicking the graph forming menu.

The method for automatically generating different qualitative graphs in earthquake disaster risk assessment based on SVC according to the present invention will be described in detail below.