CN109271734B - Site earthquake liquefaction potential evaluation method based on machine learning technology - Google Patents
Site earthquake liquefaction potential evaluation method based on machine learning technology Download PDFInfo
- Publication number
- CN109271734B CN109271734B CN201811184759.4A CN201811184759A CN109271734B CN 109271734 B CN109271734 B CN 109271734B CN 201811184759 A CN201811184759 A CN 201811184759A CN 109271734 B CN109271734 B CN 109271734B
- Authority
- CN
- China
- Prior art keywords
- length
- turning
- data
- err
- equal
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
Abstract
The invention discloses a field earthquake liquefaction potential evaluation method based on a machine learning technology, which comprises the following steps of: s1: starting; s2: inputting: a data set containing n reason attribute variables and 1 label variable, and a fold number k of cross validation; s3: defining data; s4: let i =0, go to S5; s5: when i is not more than length-F n When, let length-F i Is F i The number of the included attributes, and simultaneously, j =1, and S6 is converted; if i is greater than length-F n Then, go directly to S12; s6: let j =1, go to S7; s7: when the value of j is not more than length-F n And when i is not equal to 0, define f j Is F i J attribute of middle, let F j =F i \f j (ii) a Dividing the data set into k parts, and turning to S8; when i is equal to 0, directly dividing the data set into k parts, and turning to S8; when the value of j is more than length-F n Turning to S11; s9: repeatedly judging the sizes of s and k, and performing cross validation; the reliability of the model is improved, the generalization capability of the model is improved, and the practical application of the foundation liquefaction potential evaluation method based on the machine learning technology is further promoted.
Description
Technical Field
The invention belongs to the field of earthquake resistance, and particularly relates to a field earthquake liquefaction potential evaluation method based on a machine learning technology.
Background
It is well known to those skilled in the art that soil undergoes drastic changes in state and loss of strength and rigidity under the action of earthquakes, resulting in destruction of ground structures, a phenomenon known as liquefaction. Liquefaction is a main factor causing instability and damage of buildings in earthquakes and is also an important aspect of earthquake research and foundation earthquake-resistant design. The liquefaction potential is the excess pore water pressure generated by the saturated sandy soil or the saturated silt under the earthquake action, so that the effective shearing strength of the soil body is reduced or disappears, and the soil body is sprayed with water and blown out or slides and is unstable. The liquefaction potential is evaluated, and the possibility of liquefaction of the foundation under the earthquake action can be judged, so that a basis is provided for anti-liquefaction measures in building design.
Disclosure of Invention
The invention aims to provide a field earthquake liquefied potential evaluation method based on a machine learning technology, and aims to improve the generalization capability of a liquefied potential evaluation classifier.
The technical scheme of the invention is as follows: a field earthquake liquefied potential evaluation method based on a machine learning technology comprises the following steps:
s1: starting;
s2: inputting: a data set containing n reason attribute variables and 1 label variable, and a fold number k of cross validation;
s3: defining: f n length-F for the collective set of attributes n Is F n Number of attributes included, F i Is F n And initializing: f i =F n ;
S4: let i =0, go to S5;
s5: when i is not more than length-F n When, let length-F i Is F i The number of the included attributes, and simultaneously, j =1, and S6 is converted;
if i is greater than length-F n Then, go directly to S12;
s6: making j =1, and turning to S7;
s7: when the value of j is not more than length-F n And when i is not equal to 0, define f j Is F i J attribute of middle, let F j =F i \f j (ii) a Dividing the data set into k parts, and turning to S8;
when i is equal to 0, directly dividing the data set into k parts, and turning to S8;
when the value of j is more than length-F n Turning to S11;
s8: let S =1, turn S9;
s9: repeatedly judging the sizes of s and k, and performing cross validation;
the cross validation is completed with k error values, let err ij For the maximum of k error values, err is recorded ij While determining whether i is equal to 0, if soEqual to 0, let err i =err ij Turning to S10, if i is not equal to 0, making j = j +1, and returning to S7;
s10: let i = i +1, return to S5;
s11: order toLet F i Is err i Corresponding attribute set, determining err i Whether or not it is greater than or equal to err i-1 ,
If so, output F i-1 Turning to S12;
if not, returning to S10;
s12: and (6) ending.
Preferably, the cross-validation in step S9 is implemented as: when s is not more than k, with F j The s-th data is used for evaluating the classifier, and the other k-1 data are used for training the classifier; meanwhile, assigning the value of s +1 to a new s;
when s is greater than k, cross-validation is complete.
Preferably, data preprocessing is performed before S1 to normalize the attribute variable data, and the normalized calculation formula is as follows:
using the above formula, attribute variable data is mapped to the interval [0,1 ]]. The resulting variable y i ∈{-1,1},y i =1 denotes a non-liquefied foundation, y i =1 represents a liquefiable foundation.
The implementation process of the steps is completed under the operation of a computer memory, and manual activities or intelligent rules are not completed.
Has the advantages that: taking the maximum error of cross validation as an optimization object, deleting attributes one by one in an optimization strategy, and searching an attribute set which can minimize the maximum error of cross validation; the method can effectively reduce the fluctuation of the classification accuracy of the model, improve the reliability of the model, improve the generalization capability of the model, and further promote the practical application of the foundation liquefaction potential evaluation method based on the machine learning technology.
Drawings
FIG. 1 is a flow chart of the algorithm of the present invention;
FIG. 2 is a graph of cross-validation accuracy before and after attribute selection;
fig. 3 is a graph of classifier cross-validation AUC after attribute selection.
Detailed Description
The invention is further illustrated by the following examples in conjunction with the accompanying drawings:
as shown in figure 1 of the drawings, in which,
a field earthquake liquefied potential evaluation method based on a machine learning technology comprises the following steps:
s1: starting;
s2: inputting: a data set containing n reason attribute variables and 1 label variable, and a fold number k of cross validation;
s3: defining: f n length-F for the collective set of attributes n Is F n Number of attributes included, F i Is F n And initializing: f i =F n ;
S4: making i =0, and turning to S5;
s5: when i is not more than length-F n When, let length-F i Is F i The number of the included attributes, and simultaneously, j =1, and S6 is converted;
if i is greater than length-F n Then, go directly to S12;
s6: let j =1, go to S7;
s7: when the value of j is not more than length-F n And when i is not equal to 0, define f j Is F i Property j in, let F j =F i \f j (ii) a Dividing the data set into k parts, and turning to S8;
when i is equal to 0, directly dividing the data set into k parts, and turning to S8;
when the value of j is larger than length-F n Turning to S11;
s8: let S =1, turn S9;
s9: repeatedly judging the sizes of s and k, and performing cross validation;
the cross validation is completed with k error values, let err ij For the maximum of k error values, err is recorded ij Meanwhile, determine if i is equal to 0, if it is equal to 0, let err i =err ij Turning to S10, if i is not equal to 0, making j = j +1, and returning to S7;
s10: let i = i +1, return to S5;
s11: order toLet F i Is err i Corresponding attribute set, determining err i Whether or not it is greater than or equal to err i-1 ,
If yes, output F i-1 Turning to S12;
if not, returning to S10;
s12: and (6) ending.
Further, the cross-validation in step S9 is implemented as: when s is not more than k, with F j The s data is used for evaluating the classifier, and the other k-1 data is used for training the classifier; meanwhile, assigning the value of s +1 to a new s; the comparison of s and k can be circularly compared, and cross validation can be rapidly carried out.
When s is larger than k, the cross validation is completed;
further, data preprocessing is performed before S1, attribute variable data is normalized, and a normalized calculation formula is as follows:
using the above formula, attribute variable data is mapped to the interval [0,1 ]]. The resulting variable y i ∈{-1,1},y i =1 denotes a non-liquefied foundation, y i =1 represents a liquefiable foundation.
As shown in FIG. 2, before selecting attributes on the left side, and after selecting attributes on the right side, the support vector machine is used as a classifier. The cross validation accuracy obtained by the method proposed by the present study and the cross validation accuracy obtained for the method proposed by the present study are shown in fig. 2, and the attributes before attribute selectionIncluding Depth of soil (Depth), measuring conical resistance (q) c ) Friction to drag ratio (R) f ) Effective vertical stress (σ' v ) Total vertical stress (σ) v ) Peak ground level acceleration (α) max ) Harmonic moment magnitude (M) w ) Calculating to obtain attribute set { Depth, q c ,R f ,σ' v ,α max ,M w }。
In 10-fold cross validation, the maximum error before attribute selection is 0.1304, and the corresponding accuracy is 0.8696; the maximum error after attribute selection is 0.1053, and the corresponding accuracy is 0.8947. Meanwhile, the accuracy is 100% for 3 times before the attribute selection, and 100% for 5 times after the attribute selection. The coefficient of variation of the cross validation accuracy before attribute selection was 0.04959; the coefficient of variation for the cross-validation accuracy after attribute selection was 0.0478. After the attribute selection, the accuracy of the classifier is improved, the generalization capability is better, the discreteness of the calculation accuracy is reduced, and the calculation result is more reliable.
As shown in fig. 3, when AUC >0.9, it indicates that the classifier has better distinguishing performance, 10-fold cross validation is adopted, the classifier is trained by using different data, and 10 AUC values are obtained, as shown in fig. 3, all AUC values are greater than 0.9, which indicates that the classifier with better performance can be trained by using the selected data, and the evaluation index commonly used by the classifier in machine learning practice is AUC (Area under currve): area under Roc curve, between 0.1 and 1. The AUC can be used as a numerical value to intuitively evaluate the quality of the classifier, and the larger the value is, the better the value is.
In summary, the data set used in this application includes 226 data, of which 133 were liquefied and the other 93 were not liquefied. These data were collected from 52 out of 6 sites by cone penetration test (CPT for short), and the soil types included sandy soil, silty soil and silt soil. The method provided by the research can effectively reduce the fluctuation of the classification accuracy of the model, improve the reliability of the model, improve the generalization capability of the model and further promote the practical application of the foundation liquefaction potential evaluation method based on the machine learning technology.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and should not be taken as limiting the scope of the present invention, which is intended to cover any modifications, equivalents, improvements, etc. within the spirit and scope of the present invention.
Claims (1)
1. A field seismic liquefaction potential evaluation method based on a machine learning technology is characterized by comprising the following steps:
s1: starting;
s2: inputting: a data set containing n reason attribute variables and 1 label variable, and a fold number k of cross validation;
s3: defining: f n length-F for the collective set of attributes n Is F n Number of attributes included, F i Is F n And initializing: f i =F n ;
S4: making i =0, and turning to S5;
s5: when i is not more than length-F n When, let length-F i Is F i The number of the included attributes, and simultaneously, j =1, and S6 is converted;
if i is greater than length-F n Then, go directly to S12;
s6: making j =1, and turning to S7;
s7: when the value of j is not more than length-F n And when i is not equal to 0, define f j Is F i Property j in, let F j =F i \f j (ii) a Dividing the data set into k parts, and turning to S8;
when i is equal to 0, directly dividing the data set into k parts, and turning to S8;
when the value of j is more than length-F n Turning to S11;
s8: let S =1, turn S9;
s9: repeatedly judging the sizes of s and k, and performing cross validation;
the cross validation is completed with k error values, let err ij For the maximum of k error values, err is recorded ij And simultaneously, judging whether i is equal to 0, if so,let err i =err ij Turning to S10, if i is not equal to 0, making j = j +1, and returning to S7;
s10: let i = i +1, return to S5;
s11: order toLet F i Is err i Corresponding attribute set, determining err i Whether or not it is greater than or equal to err i-1 If yes, output F i-1 Turning to S12;
if not, returning to S10;
s12: ending;
the cross validation in step S9 is specifically implemented as: when s is not more than k, with F j The s data is used for evaluating the classifier, and the other k-1 data is used for training the classifier; meanwhile, assigning the value of s +1 to a new s;
when s is larger than k, the cross validation is completed;
data preprocessing is carried out before S1, attribute variable data are standardized, and a standardized calculation formula is as follows:
attributes before attribute selection include soil Depth (Depth), measured cone drag (q) c ) Friction to drag ratio (R) f ) Effective vertical stress (σ' v ) Total vertical stress (σ) v ) Peak ground level acceleration (α) max ) Harmonic moment magnitude (M) w ) Calculating to obtain attribute set { Depth, q c ,R f ,σ' v ,α max ,M w };
The data set includes liquefied data and non-liquefied data collected from a plurality of sites in a multi-site earthquake by cone penetration test, and the soil types include sandy soil, silty soil and silt soil.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811184759.4A CN109271734B (en) | 2018-10-11 | 2018-10-11 | Site earthquake liquefaction potential evaluation method based on machine learning technology |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811184759.4A CN109271734B (en) | 2018-10-11 | 2018-10-11 | Site earthquake liquefaction potential evaluation method based on machine learning technology |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109271734A CN109271734A (en) | 2019-01-25 |
CN109271734B true CN109271734B (en) | 2023-04-07 |
Family
ID=65195803
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811184759.4A Active CN109271734B (en) | 2018-10-11 | 2018-10-11 | Site earthquake liquefaction potential evaluation method based on machine learning technology |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109271734B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111400915A (en) * | 2020-03-17 | 2020-07-10 | 桂林理工大学 | Sand liquefaction discrimination method and device based on deep learning |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107292406A (en) * | 2016-03-30 | 2017-10-24 | 中国石油化工股份有限公司 | Seismic properties method for optimizing based on vector regression and genetic algorithm |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10909647B2 (en) * | 2015-12-09 | 2021-02-02 | One Concern, Inc. | Damage data propagation in predictor of structural damage |
-
2018
- 2018-10-11 CN CN201811184759.4A patent/CN109271734B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107292406A (en) * | 2016-03-30 | 2017-10-24 | 中国石油化工股份有限公司 | Seismic properties method for optimizing based on vector regression and genetic algorithm |
Non-Patent Citations (1)
Title |
---|
唐小微 ; 白旭 ; 胡记磊 ; .基于贝叶斯网络的自由场地震液化沉降评估.振动与冲击.2018,(18),全文. * |
Also Published As
Publication number | Publication date |
---|---|
CN109271734A (en) | 2019-01-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107562833B (en) | Contour line-based numerical model modeling method for complex terrain three-dimensional discrete unit method | |
CN110346831B (en) | Intelligent seismic fluid identification method based on random forest algorithm | |
CN113239439B (en) | Shield construction ground surface settlement prediction system and method | |
CN106897714A (en) | A kind of video actions detection method based on convolutional neural networks | |
CN104598766B (en) | Towards the space reasoning by cases method of Comprehensive Evaluation of Reservoirs | |
CN109271734B (en) | Site earthquake liquefaction potential evaluation method based on machine learning technology | |
CN107895092A (en) | A kind of interwell communication quantitative evaluation method that modeling is adopted based on complex nonlinear note | |
CN112329939A (en) | Non-sampling point rock-soil parameter evaluation method based on random forest machine learning algorithm | |
CN106441221A (en) | Method for treating surveying and mapping cross section based on secondary development of AutoCAD | |
CN104636748B (en) | A kind of method and device of number plate identification | |
CN108197328A (en) | Geographical national conditions data variation type automatic identifying method | |
KR101677159B1 (en) | Ensemble-based method for reservoir characterization using multiple kalman gains and selective use of dynamic data | |
CN111611422B (en) | SVC-based method and system for automatically generating qualitative graph in earthquake disaster risk assessment | |
CN106897554A (en) | A kind of automation computational methods that P Y profiles are carried out based on MATLAB | |
CN117312812A (en) | Landslide susceptibility evaluation model establishment method, landslide susceptibility evaluation model establishment device, landslide susceptibility evaluation model establishment equipment and storage medium | |
CN116628886A (en) | Real-time optimization method for shield tunneling machine crossing construction parameters based on limited engineering data | |
CN104834934A (en) | Nuclear body capturing method used for identifying reservoir fluid | |
CN115131571A (en) | Building local feature point identification method based on six fields of point cloud pretreatment | |
CN112907567B (en) | SAR image ordered artificial structure extraction method based on spatial reasoning method | |
CN116108392A (en) | Geological structure identification technology based on improved random forest algorithm | |
CN114086887A (en) | Downhole planning method for track of borehole to be drilled based on artificial intelligence | |
CN103955953A (en) | Terrain collaborative variable selection method for digital soil cartography | |
CN115577696B (en) | Project similarity evaluation and analysis method based on WBS tree | |
Dang et al. | Quality Evaluation of Land-Cover Classification Using Convolutional Neural Network | |
Cai et al. | A best-first multivariate decision tree method used for urban land cover classification |
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 |