CN109214046A - A kind of seismic liquefaction evaluation method and earthquake liquefaction potential model - Google Patents
A kind of seismic liquefaction evaluation method and earthquake liquefaction potential model Download PDFInfo
- Publication number
- CN109214046A CN109214046A CN201810828062.XA CN201810828062A CN109214046A CN 109214046 A CN109214046 A CN 109214046A CN 201810828062 A CN201810828062 A CN 201810828062A CN 109214046 A CN109214046 A CN 109214046A
- Authority
- CN
- China
- Prior art keywords
- liquefaction
- earthquake
- seismic
- potential model
- layer
- 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.)
- Pending
Links
- 238000011156 evaluation Methods 0.000 title claims abstract description 49
- 238000012549 training Methods 0.000 claims abstract description 69
- 238000000034 method Methods 0.000 claims abstract description 52
- 238000013135 deep learning Methods 0.000 claims abstract description 18
- 238000012360 testing method Methods 0.000 claims abstract description 15
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 11
- 230000004044 response Effects 0.000 claims description 14
- 238000005457 optimization Methods 0.000 claims description 8
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 238000004088 simulation Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 4
- 239000007788 liquid Substances 0.000 claims 2
- 230000007246 mechanism Effects 0.000 abstract description 5
- 238000009533 lab test Methods 0.000 abstract description 4
- 230000008878 coupling Effects 0.000 abstract description 3
- 238000010168 coupling process Methods 0.000 abstract description 3
- 238000005859 coupling reaction Methods 0.000 abstract description 3
- 239000013589 supplement Substances 0.000 abstract description 3
- 239000002689 soil Substances 0.000 description 9
- 210000002569 neuron Anatomy 0.000 description 6
- 230000004069 differentiation Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000007547 defect Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000010419 fine particle Substances 0.000 description 2
- 239000003673 groundwater Substances 0.000 description 2
- 239000004576 sand Substances 0.000 description 2
- 230000009897 systematic effect Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000006854 communication Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005183 dynamical system Methods 0.000 description 1
- 230000008014 freezing Effects 0.000 description 1
- 238000007710 freezing Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- PXDJXZJSCPSGGI-UHFFFAOYSA-N palmityl palmitate Chemical compound CCCCCCCCCCCCCCCCOC(=O)CCCCCCCCCCCCCCC PXDJXZJSCPSGGI-UHFFFAOYSA-N 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000010206 sensitivity analysis Methods 0.000 description 1
- 239000004575 stone Substances 0.000 description 1
- 230000003319 supportive effect Effects 0.000 description 1
- 230000002195 synergetic effect Effects 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
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
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Investigation Of Foundation Soil And Reinforcement Of Foundation Soil By Compacting Or Drainage (AREA)
Abstract
The present invention provides a kind of seismic liquefaction evaluation method and earthquake liquefaction potential model, and this method includes collecting earthquake liquefaction place data, as training set;Seismic liquefaction evaluation training is carried out using the Liquefaction index in training set, generates earthquake liquefaction potential model;Liquefaction is carried out to testing data according to seismic liquefaction potential model, obtains Liquefaction result;This method realizes the high precision nonlinear modeling of the features such as uncertainty liquefied to earthquake, randomness, multiple-factor coupling based on deep learning algorithm, differentiate that accuracy rate is more than 95%, it can be used as the supplement of the Deterministic Methods such as experience and laboratory test, provide support to explore the problem in science of earthquake liquefaction deep mechanism.
Description
Technical field
The invention belongs to seismic liquefaction evaluation technical field, in particular to a kind of seismic liquefaction evaluation method and earthquake liquefaction
Potential model.
Background technique
Nearly two during the last ten years, and disastrously earthquake centre worldwide is repeatedly with the liquefaction of large area and its caused
The serious breakoff phenomenon of ground structure and underground installation, the liquefaction of Osaka gulf area large area as caused by nineteen ninety-five Osaka-Kobe earthquake,
The series earthquake of 2010-2011 New Zealand causes seashore city Christchurch large area, and seriously liquefaction, 2011 Japan are big repeatedly
Earthquake causes Tokyo Bay area large area seriously to liquefy.Liquefaction and its earthquake are existing in China's Wenchuan Ms8.0 grades of earthquake in 2008
It is that affected area is most wide since the establishment of the nation, macroscopic appearance is the most abundant primary as significant.In view of the engineering geological condition of China's complexity
And the risk of potential complications, research liquefaction genesis mechanism, development liquefaction Judgment Method and corresponding defense technique obtain science
Boundary's attention consistent with engineering circles.
Currently, the method for assessment seismic liquefaction potential is generally based on the method for laboratory test and the semiempirical side of field test
Method.However it is too big to soil sample disturbance when using conventional method field sampling during laboratory test, it can not truly react former
Shape soil characteristics;Although freezing method takes soil that can preferably reduce the disturbance to soil sample, its original state characteristic is kept, cost is too
It is high.There is also tests itself to disturb greatly to soil layer for the semi-empirical approach of field test, and test data poor continuity, discreteness is big,
The defects of testing equipment is complicated for operation, unsuitable application.
Nearly 2 years, the artificial intelligence program AlphaGo based on depth learning technology defeated world's go champion Lee's generation in succession
Stone, Ke Jie promote the concept of artificial intelligence (AI) to move towards further flourishing.Deep learning algorithm carries out in such a way that level connects
Progressive abstract nonlinear information processing has stronger fault-tolerance, plasticity, hierarchy and systematicness, and be particularly good at solution
Complex nonlinear in nonlinear multivariable dynamical system from original input signal to desired output converts fitting problems.Deep learning
Algorithm is realized from the ability for being input to output fitting complex nonlinear.In view of skills such as image procossing, speech recognition, search engines
Art inherently belongs to the nonlinear system of height, thus in recent years deep learning algorithm all achieved in these fields it is breakthrough
Progress.
It is big to have both randomness for site liquefaction phenomenon when in view of earthquake, uncertain strong, by many factors influenced and it is each because
There are the features of nonlinearity between element, and it is urgently to be solved at present that site liquefaction, which carries out accurate differentiation, when how to earthquake
One problem.
The factor for influencing site liquefaction is numerous, is limited by objective physical condition, there is presently no a set of generally acknowledged complete systems
One Liquefaction index, it is different to there is definition in the Liquefaction index between different regions or even areal different earthquake
The phenomenon that cause, part index number shortage of data, redundancy.
Summary of the invention
In order to solve the problems in the existing technology, the present invention provides a kind of method of high-precision analog place data simultaneously
Earthquake liquefaction potential model is established, this method differentiates that accuracy rate is high.
Specific technical solution of the present invention is as follows:
The present invention provides a kind of seismic liquefaction evaluation method, and this method comprises the following steps:
Earthquake liquefaction place data are collected, as training set, earthquake liquefaction place data include Liquefaction index;
Seismic liquefaction evaluation training is carried out using the Liquefaction index in training set, generates earthquake liquefaction potential model;
Liquefaction is carried out to testing data according to seismic liquefaction potential model, obtains Liquefaction result.
It is further to improve, seismic liquefaction evaluation training is carried out using the Liquefaction index in training set, is generated
Earthquake liquefaction potential model, including seismic liquefaction evaluation training is carried out by deep learning algorithm, generate seismic liquefaction potential mould
Type, method particularly includes::
Using the Liquefaction index in training set as the data of input layer, by the processing and nonlinear transformation of hidden layer
It is transmitted to output layer;
The response obtained by output layer is compared with preset desired value, when being consistent, training stops, and generates earthquake
Liquefy potential model;
When response and desired value are not inconsistent, the error of response with desired value is calculated, and carry out instead from hidden layer to input layer
To propagation, until the response that output layer obtains is consistent with preset desired value, training stops, and generates earthquake liquefaction potential model.
Further to improve, the Liquefaction index is 12, it is preferable that the number of nodes of the input layer is 12.
Further to improve, the number of plies of the hidden layer is 4 layers, and the number of nodes of each hidden layer is respectively 7,3,4 and 3.
It is further to improve, seismic liquefaction evaluation training is carried out using the Liquefaction index in training set, generates earthquake
Liquefy potential model, including carries out seismic liquefaction evaluation training by digital simulation method, specifically includes:: the liquefaction to training set
Discriminant criterion adds the white noise of default decibel respectively, then carries out seismic liquefaction evaluation training, generates earthquake liquefaction potential model.
Further to improve, presetting decibel is 80dB.
It is realized using digital simulation method to the high-precision analog of existing Liquefaction Ground data, compensates for firsthand information data
Inconsistent etc. the defect of missing, redundancy, discriminant criterion.
It is further to improve, the method also includes the optimization to Liquefaction index, method particularly includes:
The Liquefaction index of training set is passed through rough set theory respectively to calculate and Information Entropy calculating, acquisition rough set reason
By calculated result X and Information Entropy weighted value Y, and X and Y's and sum is calculated, sum is compared with threshold value sum1, is picked out
Sum is greater than Liquefaction index corresponding to sum1, the data as input layer.
Further to improve, the node of hidden layer is respectively 9,5,10 and 8 in the earthquake liquefaction potential model.
Further to improve, the method also includes optimizing to seismic liquefaction potential model, specific method includes: to instruction
The Liquefaction index for practicing collection adds the white noise of default decibel respectively, is input in earthquake liquefaction potential model and is trained, repeatedly
For after M times, the earthquake liquefaction potential model by optimization is obtained.
Further to improve, presetting decibel is 60dB or 100dB, M=100.
Another aspect of the present invention provides a kind of earthquake liquefaction potential model that seismic liquefaction evaluation is used for based on deep learning, should
Earthquake liquefaction potential model includes 6 layers of neural network, respectively one layer of input layer, 4 layers of hidden layer and 1 layer of output layer, the input
The number of nodes of layer is 12, it is preferable that the number of nodes of each hidden layer is respectively 7,3,4 and 3.
Another aspect of the present invention also provides a kind of earthquake liquefaction potential model that seismic liquefaction evaluation is used for based on deep learning,
The earthquake liquefaction potential model includes 6 layers of neural network, respectively one layer of input layer, 4 layers of hidden layer and 1 layer of output layer, described defeated
The number of nodes for entering layer is 5, it is preferable that the number of nodes of each hidden layer is respectively 9,5,10 and 8.
The present invention provides a kind of seismic liquefaction evaluation method and earthquake liquefaction potential model, and this method is based on deep learning algorithm
The high precision nonlinear modeling for realizing the features such as uncertainty liquefied to earthquake, randomness, multiple-factor coupling, differentiates quasi-
True rate is more than 95%, can be used as the supplement of the Deterministic Methods such as experience and laboratory test, to explore earthquake liquefaction deep mechanism
Problem in science provide support.
Detailed description of the invention
Fig. 1 is a kind of flow chart of the seismic liquefaction evaluation method of embodiment 1;
Fig. 2 is the distribution that mark passes through test site data point;
Fig. 3 is the seismic liquefaction evaluation training that embodiment 2 carries out deep learning using training set, generates seismic liquefaction potential mould
The flow chart of type;
Fig. 4 is the structural schematic diagram of earthquake liquefaction potential model.
Specific embodiment
Embodiment 1
The embodiment of the present invention 1 provides a kind of seismic liquefaction evaluation method, as shown in Figure 1, this method comprises the following steps:
Earthquake liquefaction place data are collected, as training set, earthquake liquefaction place data include Liquefaction index;
The seismic liquefaction evaluation training that deep learning is carried out using the Liquefaction index in training set, generates earthquake liquefaction
Potential model;
Liquefaction is carried out to testing data according to seismic liquefaction potential model, obtains Liquefaction result.
The present invention provides a kind of seismic liquefaction evaluation method, and initial data used in the present invention includes that China and the U.S. exist
382 interior marks pass through test site data.Wherein U.S.'s Liquefaction Ground database generalization cover (1984) data such as Seed and
Cetin etc. (2000,2004) data, and introduce liquefaction/non-liquefaction data of Earthquakes in Japan, including 115 liquefaction points, 112
A non-liquefaction point.Used CHINESE REGION data are used to work out China's seismic design provision in building code soil liquefaction discrimination formula
Initial data contain Liquefaction Ground 94 including 155 of 7 secondary earthquakes liquefaction/non-Liquefaction Ground data between 1962-1976
A, non-Liquefaction Ground 61, and the distribution that mark passes through test site data point is given in conjunction with Fig. 2.This method is based on deep learning
Algorithm realizes the high precision nonlinear modeling of the features such as uncertainty liquefied to earthquake, randomness, multiple-factor coupling, sentences
Other accuracy rate significantly improves.
Embodiment 2
The embodiment of the present invention 2 provides a kind of seismic liquefaction evaluation method, and this method is essentially identical with embodiment 1, different
, as shown in figure 3, carrying out the seismic liquefaction evaluation training of deep learning using training set, generate earthquake liquefaction potential model packet
It includes:
Using the Liquefaction index in training set as the data of input layer, wherein Liquefaction index is 12, respectively
For earthquake magnitude (Mw), horizontal direction peak ground acceleration (PGA) liquefies critical depth (ds), Ground-water level (dw), above covers total
Stress (σv), above it is covered with efficacy (σv'), equivalent clean sand corrects SPT blow count ((N1)60CS), fine particle content (FC), soil layer is cut
Stress reduction factor (τd), burden pressure correction factor (Kσ), earthquake magnitude correction factor (MSF), soil layer cycle stress ratio (CSR);It will
The data of input layer are transmitted to by the processing and nonlinear transformation of hidden layer to output layer;Wherein the number of plies of hidden layer is 4 layers;
The response obtained by output layer is compared with preset desired value, when being consistent, training stops, and generates earthquake
Liquefy potential model;
When response and desired value are not inconsistent, the error of response with desired value is calculated, and carry out instead from hidden layer to input layer
To propagation, until the response that output layer obtains is consistent with preset desired value, training stops, and generates earthquake liquefaction potential model.
The earthquake liquefaction place data that the present invention is collected, as training set.It is based on deep learning algorithm, that is, deep layer herein
The earthquake liquefaction potential model of neural network algorithm selects 6 layers of neural network, and the classification that output layer outputs test data is divided into liquefaction
With two classes of non-liquefaction, the activation primitive of hidden layer neuron and output layer selects Sigmoid function, earthquake liquefaction potential model
Specific structure is as shown in Figure 4.In input layer to during hidden layer forward-propagating, input Liquefaction index sample parameter passes through
Input layer is transmitted to output layer by the layer-by-layer processing of hidden layer and nonlinear transformation.If in response and expectation that output layer obtains
Value is not inconsistent, then the error of the response and desired value is taken successively to carry out backpropagation from hidden layer to input layer direction.Communication process
Middle error distribution gives each layer, and then is updated to each layer neuron weight.By adjusting hidden layer and hidden layer, hidden layer with
The linking intensity and threshold value of neuron node between input layer, hidden layer and output layer decline error along gradient direction.Big
In the study and training process of amount repeatedly, the discrimination precision of model is correspondingly improved, and is set in advance when output layer signal reaches
When fixed desired value, training stops.Pass through the conversion that 5 layers of neuron indicates original input data before neural network, last layer
Differentiation whether neuron can be completed to liquefaction.When subsequent similar sample data inputs the neural network model saved,
Model voluntarily calculation process can obtain differentiating result accordingly.According to above-mentioned differentiation process, testing data is input to model simultaneously
After repetitive exercise 100 times, optimal accuracy rate of the model on testing data has reached 95.12%.It follows that the present invention proposes
Earthquake liquefaction potential model based on deep learning and method of discrimination can each influence factor of concentrated expression whether liquefy with place
Between non-linear relation, data recognition capability and Liquefaction accuracy rate with higher.And deep learning model buildings letter
It is single, once it is determined that model initial configuration, by sufficient training sample, can adaptively determine each layer neuron weight parameter and
Without considering the mechanism of complexity and the effect of intercoupling that may cause liquefied various factors inside site liquefaction.Therefore the mould
Type can be used as a kind of novel multiple parameter overall assessment model to carry out earthquake site liquefaction tendency it is correctly predicted.
Embodiment 3
The embodiment of the present invention 3 provides a kind of seismic liquefaction evaluation method, and this method is essentially identical with embodiment 2, different
, the number of nodes of each hidden layer is respectively 7,3,4 and 3.Number of nodes by adjusting each layer of hidden layer can reduce network
Systematic error, and can guarantee that net training time is short, be easily found optimum point in training, and reduce in training process
" over-fitting " probability occurred, effect possessed by the above node is all not achieved when being adjusted to number of nodes.
Embodiment 4
The embodiment of the present invention 4 provides a kind of seismic liquefaction evaluation method, and this method is essentially identical with embodiment 1, different
, which carries out seismic liquefaction evaluation training by digital simulation method, generates earthquake liquefaction potential model, specific to wrap
It includes: seismic liquefaction evaluation training is carried out again to the white noise that the Liquefaction index in training set adds the decibel of 80dB respectively,
Generate earthquake liquefaction potential model.The present invention adds the white noise of different decibels to training set initial data respectively, is trained, repeatedly
It generation 100 times, obtains, when signal-to-noise ratio takes 80dB, after original base expands 14 times, the error rate of model judgement is training set
4.08%, precision reaches 95.92%;When signal-to-noise ratio be more than or less than 80dB, the error rate of model entirety be held in 7.0% with
On.Show that the liquefaction data simulated using the white Gaussian noise method of specific signal-to-noise ratio have sufficiently high precision, it can be certain
The essential characteristic that Liquefaction Ground data are really represented in degree can play the foundation of current earthquake liquefaction place database
Supplement and support study dies effect, enhance the applicability and discriminating power of model;Choose the Gauss that suitable noise decibel number obtains
The high-precision analog to existing Liquefaction Ground data may be implemented in white noise, while having both easy obtaining property and representative advantage,
Inconsistent etc. the defect of firsthand information shortage of data, redundancy, discriminant criterion is compensated for a certain extent.
Embodiment 5
The embodiment of the present invention 5 provides a kind of seismic liquefaction evaluation method, and this method is essentially identical with embodiment 3, different
, the seismic liquefaction evaluation training of deep learning is carried out using training set, further includes over the ground after generation earthquake liquefaction potential model
The step of shake liquefaction potential model optimizes, specific method includes: to add 80dB's respectively to the Liquefaction index of training set
White noise is input in earthquake liquefaction potential model and is trained, and after iteration 100 times, obtains the seismic liquefaction potential mould by optimization
Type.
Optimal accuracy rate of the optimized earthquake liquefaction potential model on testing data has reached 96.75%.Surface
The earthquake liquefaction potential model that deep approach of learning and data analogy method are established, which is treated, has synergistic effect on measured data Liquefaction.
Embodiment 6
The embodiment of the present invention 6 provides a kind of seismic liquefaction evaluation method, and this method is essentially identical with embodiment 1, different
, the method also includes the optimization to Liquefaction index, method particularly includes:
The Liquefaction index of training set is passed through rough set theory respectively to calculate and Information Entropy calculating, acquisition rough set reason
By calculated result X and Information Entropy weighted value Y, and X and Y's and sum is calculated, sum is compared with threshold value sum1, is picked out
Sum is greater than Liquefaction index corresponding to sum1.
The present invention is utilized respectively rough set theory and is calculated with Information Entropy calculating to Liquefaction based on the data obtained
Index sensibility, and analyzed, analysis the results are shown in Table 1, optimize based on the analysis results to Liquefaction index.
1 rough set theory of table calculates and Information Entropy is calculated to Liquefaction index sensitivity analysis result
Citing, set sum1=0.09, sum greater than 0.09 be respectively PGA (horizontal direction peak ground acceleration), dw
(Ground-water level), (N1)60CS(equivalent clean sand amendment SPT blow count), FC (fine particle content) and CSR (soil layer pulsating stress
Than), index is distinguished to liquefied significant correlation in view of this 5 kinds liquefaction, may be characterized as the master for influencing earthquake liquefaction inherent mechanism
Want factor.Therefore optimization 5 and the highest Liquefaction index of earthquake liquefaction susceptibility are made from 12 Liquefaction indexs
For the input data of input layer.It follows that sentencing extracting 5 liquefaction most sensitive with earthquake liquefaction from initial data concentration
Other index, after the characteristic dimension for reducing input layer, accuracy rate improves 1.22% compared with archetype, is 96.34%;Thus
Out, the prediction that the intrinsic dimensionality that effectively reduction inputs Liquefaction Ground data can improve liquefaction potential model to a certain extent is accurate
Rate, the reduction of deep learning model complexity objectively also improve the arithmetic speed of subsequent algorithm;Show by characteristic index
Liquefaction potential model after optimization still has the applicability to a large amount of Liquefaction Ground data, and supportive and higher differentiation is accurate
Rate.
The present invention realizes the transmission to data by double-channel, can reduce the delay and jitter of data, improves and sends effect
Rate.
Embodiment 7
The embodiment of the present invention 7 provides a kind of seismic liquefaction evaluation method, and this method is essentially identical with embodiment 6, different
, which carries out seismic liquefaction evaluation training by digital simulation method, generates earthquake liquefaction potential model, specific to wrap
It includes: seismic liquefaction evaluation training is carried out again to the white noise that the Liquefaction index in training set adds 60dB or 100dB respectively,
Generate earthquake liquefaction potential model.
When Liquefaction index becomes 5, the earthquake liquefaction potential model generated by digital simulation method is in number to be measured
96.17% and 96.13% are reached according to upper optimal accuracy rate, 1.05% and 1.01% have been respectively increased than original model.
Embodiment 8
The embodiment of the present invention 8 provides a kind of seismic liquefaction evaluation method, and this method is essentially identical with embodiment 5, with reality
It further include that adjust the node of each hidden layer be respectively 9,5,10 and after Liquefaction index that example 5 is picked out is applied as input layer
8.Due to the change of input layer, in order to reduce the systematic error of network, guarantees that net training time is short, be easily found most in training
Advantage, so the number of nodes of the hidden layer of earthquake liquefaction potential model is accordingly adjusted to 9,5,10 and 8, output layer number of nodes is 1,
Exporting result is liquefaction or non-liquefaction.
Embodiment 9
The embodiment of the present invention 9 provides a kind of seismic liquefaction evaluation method, and this method is essentially identical with embodiment 7, described
Method further includes optimizing to seismic liquefaction potential model, and specific method includes: to add respectively to the Liquefaction index of training set
The white noise for adding 60dB or 100dB is input in earthquake liquefaction potential model and is trained, and after iteration 100 times, obtains by optimization
Earthquake liquefaction potential model.
5 Liquefaction indexs after the present invention optimizes training set add the white noise of different decibels respectively, and input
Be trained, iteration 100 times, obtain into model, when signal-to-noise ratio takes 60dB, training set original base expand 14 times after,
The error rate that model judges is 2.41%, and precision reaches 97.59%;When signal-to-noise ratio takes 100dB, training set expands in original base
After filling 14 times, model judges that precision reaches 97.18%.
After the present invention is by rejecting the redundancy feature index in Liquefaction Ground data discriminant criterion, the potential model that liquefies is to data
The efficiency and predictablity rate of processing are improved, and still maintain high-resolution in the case where mass data input.
The present invention is not limited to above-mentioned preferred forms, anyone can show that other are various under the inspiration of the present invention
The product of form, however, make any variation in its shape or structure, it is all that there is skill identical or similar to the present application
Art scheme, is within the scope of the present invention.
Claims (10)
1. a kind of seismic liquefaction evaluation method, which is characterized in that described method includes following steps:
Earthquake liquefaction place data are collected, as training set, earthquake liquefaction place data include Liquefaction index;
Seismic liquefaction evaluation training is carried out using the Liquefaction index in training set, generates earthquake liquefaction potential model;
Liquefaction is carried out to testing data according to seismic liquefaction potential model, obtains Liquefaction result.
2. seismic liquefaction evaluation method as described in claim 1, which is characterized in that utilize the Liquefaction index in training set
Seismic liquefaction evaluation training is carried out, generates earthquake liquefaction potential model, including earthquake liquefaction is carried out by deep learning algorithm
Discriminative training generates earthquake liquefaction potential model, method particularly includes:
Using the Liquefaction index in training set as the data of input layer, it is transmitted to by the processing and nonlinear transformation of hidden layer
To output layer;
The response obtained by output layer is compared with preset desired value, when being consistent, training stops, and generates earthquake liquefaction
Potential model;
When response is not inconsistent with desired value, the error of response with desired value is calculated, and is reversely passed from hidden layer to input layer
It broadcasts, until the response that output layer obtains is consistent with preset desired value, training stops, and generates earthquake liquefaction potential model.
3. seismic liquefaction evaluation method as claimed in claim 2, which is characterized in that the Liquefaction index is 12, excellent
Selection of land, the number of nodes of the input layer are 12.
4. seismic liquefaction evaluation method as claimed in claim 3, which is characterized in that the number of plies of the hidden layer is 4 layers, preferably
Ground, the number of nodes of each hidden layer are respectively 7,3,4 and 3.
5. seismic liquefaction evaluation method as claimed in claim 1 or 2, which is characterized in that the liquefaction using in training set
Discriminant criterion carries out seismic liquefaction evaluation training, generates earthquake liquefaction potential model, including carry out earthquake by digital simulation method
Liquefaction training, specifically includes: the white noise for adding default decibel respectively to the Liquefaction index in training set carries out again
Seismic liquefaction evaluation training, generates earthquake liquefaction potential model;Preferably, presetting decibel is 80dB.
6. seismic liquefaction evaluation method as claimed in claim 5, which is characterized in that the method also includes referring to Liquefaction
Mark optimizes, method particularly includes:
Liquefaction index in training set is passed through rough set theory respectively to calculate and Information Entropy calculating, acquisition rough set theory
Calculated result X and Information Entropy weighted value Y, and X and Y's and sum is calculated, sum is compared with threshold value sum1, picks out sum
Greater than Liquefaction index corresponding to sum1.
7. seismic liquefaction evaluation method as claimed in claim 6, which is characterized in that hidden layer in the earthquake liquefaction potential model
Node be respectively 9,5,10 and 8.
8. seismic liquefaction evaluation method as claimed in claim 7, which is characterized in that the method also includes to seismic liquefaction potential
Model optimizes, specific method include: to training set=in Liquefaction index add the white noise of default decibel respectively,
It is input in earthquake liquefaction potential model and is trained, after iteration M times, obtain the earthquake liquefaction potential model by optimization;Preferably,
Default decibel is 60dB or 100dB, M=100.
9. a kind of earthquake liquefaction potential model for being used for seismic liquefaction evaluation based on deep learning, which is characterized in that the earthquake liquid
Changing potential model includes 6 layers of neural network, respectively one layer of input layer, 4 layers of hidden layer and 1 layer of output layer, the section of the input layer
Points are 12, it is preferable that the number of nodes of each hidden layer is respectively 7,3,4 and 3.
10. a kind of earthquake liquefaction potential model for being used for seismic liquefaction evaluation based on deep learning, which is characterized in that the earthquake liquid
Changing potential model includes 6 layers of neural network, respectively one layer of input layer, 4 layers of hidden layer and 1 layer of output layer, the section of the input layer
Points are 5, it is preferable that the number of nodes of each hidden layer is respectively 9,5,10 and 8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810828062.XA CN109214046A (en) | 2018-07-25 | 2018-07-25 | A kind of seismic liquefaction evaluation method and earthquake liquefaction potential model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810828062.XA CN109214046A (en) | 2018-07-25 | 2018-07-25 | A kind of seismic liquefaction evaluation method and earthquake liquefaction potential model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109214046A true CN109214046A (en) | 2019-01-15 |
Family
ID=64990276
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810828062.XA Pending CN109214046A (en) | 2018-07-25 | 2018-07-25 | A kind of seismic liquefaction evaluation method and earthquake liquefaction potential model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109214046A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110142803A (en) * | 2019-05-28 | 2019-08-20 | 上海电力学院 | A kind of mobile welding robot working state of system detection method and device |
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 (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5402520A (en) * | 1992-03-06 | 1995-03-28 | Schnitta; Bonnie S. | Neural network method and apparatus for retrieving signals embedded in noise and analyzing the retrieved signals |
JPH08179047A (en) * | 1994-12-26 | 1996-07-12 | Osaka Gas Co Ltd | Judging method of liquefaction caused by earthquake with neural network |
KR100911896B1 (en) * | 2009-03-11 | 2009-08-13 | (주)대우건설 | Earthquake alarming device based on artificial structural damage learning |
-
2018
- 2018-07-25 CN CN201810828062.XA patent/CN109214046A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5402520A (en) * | 1992-03-06 | 1995-03-28 | Schnitta; Bonnie S. | Neural network method and apparatus for retrieving signals embedded in noise and analyzing the retrieved signals |
JPH08179047A (en) * | 1994-12-26 | 1996-07-12 | Osaka Gas Co Ltd | Judging method of liquefaction caused by earthquake with neural network |
KR100911896B1 (en) * | 2009-03-11 | 2009-08-13 | (주)대우건설 | Earthquake alarming device based on artificial structural damage learning |
Non-Patent Citations (7)
Title |
---|
刘勇健等: "基于粗糙集―支持向量机的震害预测模型及应用", 《地震研究》 * |
唐睿旋等: "基于粗糙集和BP神经网络的滑坡易发性评价", 《煤田地质与勘探》 * |
张红梅等: "粗糙集与熵值法在综合评价中的运用述评", 《科技信息》 * |
李方明等: "基于BP神经网络的饱和砂土液化判别方法", 《自然灾害学报》 * |
杨小凤等: "基于不同方法的泥石流危险性评价对比分析――以四川汶川七盘沟泥石流为例", 《中国地质灾害与防治学报》 * |
郝杰等: "基于模糊C-均值算法粗糙集理论的云模型在岩爆等级评价中的应用", 《岩土力学》 * |
马浩,曹君林,彭斌,穆成林: "基于组合赋权-云模型巷道围岩稳定性预测评价研究", 《煤炭技术》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110142803A (en) * | 2019-05-28 | 2019-08-20 | 上海电力学院 | A kind of mobile welding robot working state of system detection method and device |
CN111400915A (en) * | 2020-03-17 | 2020-07-10 | 桂林理工大学 | Sand liquefaction discrimination method and device based on deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109492822B (en) | Air pollutant concentration time-space domain correlation prediction method | |
Liu et al. | Random forest regression evaluation model of regional flood disaster resilience based on the whale optimization algorithm | |
CN109165664A (en) | A kind of attribute missing data collection completion and prediction technique based on generation confrontation network | |
CN107563574A (en) | A kind of earth's magnetic field Forecasting Methodology based on long memory models Recognition with Recurrent Neural Network in short-term | |
Rezania et al. | Evaluation of liquefaction potential based on CPT results using evolutionary polynomial regression | |
CN110135502A (en) | A kind of image fine granularity recognition methods based on intensified learning strategy | |
CN109375252A (en) | Consider the ground motion parameter evaluation method of different seismic structure maximum credible earthquakes | |
CN108509939A (en) | A kind of birds recognition methods based on deep learning | |
CN113009553A (en) | Interpretable seismic motion parameter probability density distribution prediction method based on NGboost and SHAP values | |
CN114330101A (en) | Slope earthquake slip prediction method and system based on artificial neural network | |
CN113393488A (en) | Behavior track sequence multi-feature simulation method based on quantum migration | |
CN111709072B (en) | Underground vibration amplitude parameter prediction method | |
CN109214046A (en) | A kind of seismic liquefaction evaluation method and earthquake liquefaction potential model | |
CN113011660A (en) | Air quality prediction method, system and storage medium | |
CN113470316A (en) | Debris flow monitoring and early warning method based on self-coding single classification model | |
Si et al. | A Novel coal-gangue recognition method for top coal caving face based on IALO-VMD and improved MobileNetV2 network | |
CN112883646B (en) | Building settlement amount extraction method, system and device combining machine learning and soil mechanics model | |
Li et al. | Soil seismic response modeling of KiK-net downhole array sites with CNN and LSTM networks | |
Chen et al. | The application of the genetic adaptive neural network in landslide disaster assessment | |
CN106844626A (en) | Using microblogging keyword and the method and system of positional information simulated air quality | |
CN116911148A (en) | Method and system for evaluating earthquake damage of sedimentary basin building group | |
Chen et al. | A prediction method of ground motion for regions without available observation data (LGB-FS) and its application to both Yangbi and Maduo earthquakes in 2021 | |
CN109933921B (en) | Rolling rock disaster risk assessment method, device and system and storage medium | |
Rudman et al. | Using ground-motion simulations within a Monte Carlo approach to assess probabilistic seismic risk | |
Wang et al. | Application of Deep Learning in Microseismic Waveform Classification: A Case Study of The Yebatan Hydropower Station Project |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190115 |