CN108664673A - A kind of wheel rail force load identification feature-based data model method for building up and device - Google Patents
A kind of wheel rail force load identification feature-based data model method for building up and device Download PDFInfo
- Publication number
- CN108664673A CN108664673A CN201710202179.2A CN201710202179A CN108664673A CN 108664673 A CN108664673 A CN 108664673A CN 201710202179 A CN201710202179 A CN 201710202179A CN 108664673 A CN108664673 A CN 108664673A
- Authority
- CN
- China
- Prior art keywords
- matrix
- neuron
- hidden
- hidden neuron
- data
- 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
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
- G06N3/08—Learning methods
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)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Hardware Design (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Geometry (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The present invention relates to a kind of wheel rail force load identification feature-based data model method for building up and device, wherein method includes：The time-frequency characteristics of rail vehicle system detectio data are extracted using the adaptive signal decomposition algorithm of based on variable element domain and in short-term Gaussian linear frequency modulation base；Sparse Principal Component Analysis Algorithm is conjugated using multi-node collaborative to merge the time-frequency characteristics of the rail vehicle system detectio data；Using multinode Fusion Features data as sample data set, L is used1/2SparsePCA ELM neural network machine learning algorithms establish wheel rail force load identification feature-based data model.The technical program picks out wheel rail force in real time using Data Modeling Method, and then assess the safe condition of rail vehicle system interaction, it is the important supplement of the existing track quality judgment criteria detected based on track geometry status, is conducive to comprehensive analysis rail vehicle system mode and high speed railway track maintenance is instructed to work.
Description
Technical field
The present invention relates to rolling stock detection technique field, more particularly to a kind of wheel rail force load recognizes feature-based data model
Method for building up and device.
Background technology
Along with the rapid development of China express railway, due to the growth of train running speed and operating mileage, in order to protect
Train operating safety is demonstrate,proved, needs that key factor-wheel-rail interaction power that wheel rail system acts on is identified and is monitored.Mesh
Before, although instrumented wheelset can detect wheel rail force in real time, its cost is higher and inconvenient maintenance, can be only installed at minority
It is used on comprehensive detection train.Comprehensive detection train passes through vehicle dynamic response acceleration system and the advanced detection of force-measuring wheel equity
Equipment carries out dynamic detection to the state of vehicle and wheel rail system respectively, acquires a large amount of track-Vehicular system dynamic detection
Data after the time domain and the frequency domain character attribute that extract errant-Vehicular system state, use feature by analyzing these data
It is a kind of new way for recognizing wheel rail force that data, which establish wheel rail force load Identification Data model,.
Track-Vehicular system is the system with interaction of a Dynamic Coupling, rolling stock when being run on rail track,
Vibration is generated by the excitation of track irregularity；Other load generated in the gravity of rolling stock and operation are acted on by wheel
On rail, and cause the flexible deformation and track settlement of rail, to make guideway irregularity aggravate, wheel-rail interaction therein
It is presented in the form of power.The dynamic performance information that vehicle is not only contained in wheel rail force, further comprises under travel condition of vehicle
Track-Vehicular system Coupled Dynamic response message.Wheel rail force can reveal that the safety in operation of vehicle, by measure wheel track it
Between the vertical and cross force that interacts and then the indexs such as derailment coefficients, rate of wheel load reduction, wheel shaft cross force can be calculated.This
Whether a little indexs are for evaluation vehicle operational safety, research derailing mechanism etc. play very important effect.
Domestic and international researcher generally recognizes wheel-rail force by the method for establishing mechanism model parsing, that is, establishes dynamics
Simulation model, according to the dynamic exciting wheel rail force suffered by the dynamic characteristic of known system and the dynamic response inverse structure of actual measurement.
Mainly use two methods of time domain and frequency domain.
Time domain method is a kind of method that the 1980s, mid-term grew up, and is mainly based upon and is gone out in the form of convolution
The equation of motion of system is carried out modal coordinate conversion by the relationship between existing excitation and system response, and kinetic equation decouples
To non-coupled kinetic equation, the final inverse time history for solving dynamic loads.But in time domain often there is diverging in Dynamic Load Identification
Problem, or the precision identified under conditions of having measurement noise and modal parameter error are relatively low.Due to time domain Dynamic Load Identification skill
The research of art is started late, and still has many problems to need to solve.In the time domain, the input and output of system are closed at complicated convolution
System, to making troubles in Mathematical treatment.
Frequency domain technology is that the kinetics equation of system is transformed into frequency domain, and unknown dynamic load is identified with known conditions
Dynamic loading in frequency domain is finally carried out inverse fourier transform and is converted to time domain dynamic loading by lotus.Due in a frequency domain, system it is defeated
Enter and export into simple linear relationship, complicated problem is made to simplify.But the vehicle dynamic response in rail wheeling action system and wheel
There is stronger non-linear relations between rail power, and degree of freedom is more, operating mode is complicated, therefore establish the meter after mechanism model
Calculation amount is larger.The softwares such as the Rail modules of current existing business software such as many-body dynamics software MSC, ADAMS and ANSYS
Calculating speed is difficult to meet the requirement of engineer application.
Invention content
To solve problem of the prior art, the present invention proposes that a kind of wheel rail force load identification feature-based data model is established and filled
It sets, the technical program can not only effectively reduce calculation amount, and can preferably describe the pass of the complex nonlinear between wheel rail system
System.
To achieve the above object, the present invention provides a kind of wheel rail force load to recognize feature-based data model method for building up, packet
It includes：
Track-vehicle is extracted using the adaptive signal decomposition algorithm of based on variable element domain and in short-term Gaussian linear frequency modulation base
The time-frequency characteristics of system detectio data；
It is conjugated time-frequency of the sparse Principal Component Analysis Algorithm to the track-Vehicular system detection data using multi-node collaborative
Feature is merged；
Using multinode Fusion Features data as sample data set, L is used1/2- SparsePCA-ELM neural network machines
Learning algorithm establishes wheel rail force load identification feature-based data model.
Preferably, further include：
Before carrying out fusion treatment to the time-frequency characteristics of the track-Vehicular system detection data, by the track-vehicle
System detectio data are filtered by time-frequency characteristics.
Preferably, further include：
Training group and test group are divided into the time-frequency characteristics of the track-Vehicular system detection data after being filtered,
Every group of the data that output and input are normalized respectively.
Preferably, the sparse Principal Component Analysis Algorithm expression formula of the multi-node collaborative conjugation is：
Wherein, s and c is two preassigned constants；uTFor unit orthogonal eigenvectors [u1,u2…,up] transposition square
Battle array；U is unit orthogonal eigenvectors [u1,u2…,up]；ΣXFor covariance matrix, ΣX=E [(X-E (X)) (X-E (X))T], X
Indicate the attributive character of node, X=(X1,X2,…Xp), p indicates the attribute number of node.
Preferably, described to use L1/2- SparsePCA-ELM neural network machine learning algorithms are established wheel rail force load and are distinguished
Know feature-based data model the step of include：
Determine the number initial value of hidden neuron；
Input weight coefficient matrix and the hidden layer god between hidden neuron are obtained according to the number initial value of the hidden neuron
Through first threshold matrix；
Hidden neuron activation primitive is determined according to input weight coefficient matrix and hidden neuron threshold matrix；
Multinode Fusion Features data in the training group and test group are substituted into neural network input layer, pass through input
Weight coefficient matrix, hidden neuron threshold matrix and hidden neuron activation primitive obtain the output matrix of hidden layer neuron；
Using the output matrix of the hidden layer neuron, according to L1/2Regularization threshold value iterative algorithm obtains the hidden layer
The optimal solution of connection weight matrix between neuron and output layer neuron, according to the hidden neuron and output layer nerve
The number of non-zero element redefines the number of the hidden neuron in the optimal solution of connection weight matrix between member；
Using input weight coefficient matrix, hidden neuron threshold matrix, the hidden neuron and output layer neuron it
Between the optimal solution of connection weight matrix carry out cross validation, judge whether wheel rail force load identification feature-based data model full
Sufficient engineer application demand；If it is satisfied, then utilizing input weight coefficient matrix, hidden neuron threshold matrix, hidden layer nerve
Non-zero member in the optimal solution of connection weight matrix between member and output layer neuron and the optimal solution according to connection weight matrix
The number for the hidden neuron that the number of element redefines, which is established, is based on multinode L1/2The wheel of-SparsePCA-ELM neural networks
Rail power load recognizes feature-based data model；Otherwise, instruction is repartitioned to the time-frequency characteristics of the track-Vehicular system detection data
Practice group and test group, selectes input weight coefficient matrix, hidden neuron threshold matrix, the hidden neuron and output layer again
The optimal solution of connection weight matrix between neuron, until the wheel rail force load identification feature-based data model established meets engineering
Until application requirement.
To achieve the above object, the present invention also provides a kind of wheel rail force load identification feature-based data models to establish device,
Including：
Feature extraction unit, for utilizing based on the variable element domain and in short-term adaptive signal decomposition of Gaussian linear frequency modulation base
Algorithm extracts the time-frequency characteristics of track-Vehicular system detection data；
Fused data unit, for using multi-node collaborative to be conjugated sparse Principal Component Analysis Algorithm to the track-vehicle
The time-frequency characteristics of system detectio data are merged；
Model foundation unit, for using multinode Fusion Features data as sample data set, using L1/2-
SparsePCA-ELM neural network machine learning algorithms establish wheel rail force load identification feature-based data model.
Preferably, further include：
Filter unit is used for before carrying out fusion treatment to the time-frequency characteristics of the track-Vehicular system detection data,
The track-Vehicular system detection data is filtered by time-frequency characteristics.
Preferably, further include：
Normalized unit, for the time-frequency characteristics to the track-Vehicular system detection data after being filtered
It is divided into training group and test group, every group of the data that output and input is normalized respectively.
Preferably, the multi-node collaborative that the fused data unit uses is conjugated the expression formula of sparse Principal Component Analysis Algorithm
For：
Wherein, s and c is two preassigned constants；uTFor unit orthogonal eigenvectors [u1,u2…,up] transposition square
Battle array；U is unit orthogonal eigenvectors [u1,u2…,up]；ΣXFor covariance matrix, ΣX=E [(X-E (X)) (X-E (X))T], X
Indicate the attributive character of node, X=(X1,X2,…Xp), p indicates the attribute number of node.
Preferably, the model foundation unit includes：
Hidden neuron number initial value determining module, the number initial value for determining hidden neuron；
First model parameter determining module, for according between the number initial value of hidden neuron acquisition hidden neuron
Input weight coefficient matrix and hidden neuron threshold matrix；
Hidden neuron activation primitive determining module, for according to input weight coefficient matrix and hidden neuron threshold matrix
Determine hidden neuron activation primitive；
Second model parameter determining module was used for the multinode Fusion Features data generation in the training group and test group
Enter neural network input layer, is obtained by inputting weight coefficient matrix, hidden neuron threshold matrix and hidden neuron activation primitive
Obtain the output matrix of hidden layer neuron；
Third model parameter determining module, for the output matrix using the hidden layer neuron, according to L1/2Regularization
Threshold value iterative algorithm obtains the optimal solution of the connection weight matrix between the hidden neuron and output layer neuron, according to institute
The number for stating non-zero element in the optimal solution of the connection weight matrix between hidden neuron and output layer neuron redefines institute
State the number of hidden neuron；
Wheel rail force load identification feature-based data model establishes module, for utilizing input weight coefficient matrix, hidden neuron
The optimal solution of connection weight matrix between threshold matrix, the hidden neuron and output layer neuron carries out cross validation,
Judge whether the wheel rail force load identification feature-based data model meets engineer application demand；If it is satisfied, then utilizing input power
Connection weight matrix between coefficient matrix, hidden neuron threshold matrix, the hidden neuron and output layer neuron
The number for the hidden neuron that the number of non-zero element redefines in optimal solution and optimal solution according to connection weight matrix is built
Be based on multinode L1/2The wheel rail force load of-SparsePCA-ELM neural networks recognizes feature-based data model；Otherwise, to described
The time-frequency characteristics of track-Vehicular system detection data repartition training group and test group, again select input weight coefficient matrix,
The optimal solution of connection weight matrix between hidden neuron threshold matrix, the hidden neuron and output layer neuron, directly
Until the wheel rail force load of foundation identification feature-based data model meets engineer application requirement.
Above-mentioned technical proposal has the advantages that：
The technical program obtains input data, these sensor installation sides using axle box, framework, car body acceleration sensor
Just and convenient for safeguarding, typical load cell wheel can be overcome to installing, safeguarding the shortcomings such as cumbersome；And relative to using dynamometry
Wheel is relatively low to the price of direct detection mode, can be provided on more operation trains and carry out routine testing, ensure train peace
Row for the national games, has broad application prospects.In addition, the technical program picks out wheel rail force in real time using Data Modeling Method,
And then the safe condition of track-Vehicular system interaction is assessed, it is the existing track matter detected based on track geometry status
The important supplement for measuring judgment criteria is conducive to comprehensive analysis track-Vehicular system state and instructs high speed railway track maintenance dimension
Repair work.
Description of the drawings
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
Obtain other attached drawings according to these attached drawings.
Fig. 1 is the principle schematic of the technical program；
Fig. 2 is that a kind of wheel rail force load that the embodiment of the present invention proposes recognizes feature-based data model method for building up flow chart；
Fig. 3 is the gradual change rectangular window intercept signal schematic diagram of the present embodiment；
Fig. 4 is one of the vehicle dynamic response acceleration information time-frequency characteristics CPST-AGCD distribution maps of the present embodiment；
Fig. 5 is the two of the vehicle dynamic response acceleration information time-frequency characteristics CPST-AGCD distribution maps of the present embodiment；
Fig. 6 is one of the wheel track force data time-frequency characteristics CPST-AGCD distribution maps of the present embodiment；
Fig. 7 is the two of the wheel track force data time-frequency characteristics CPST-AGCD distribution maps of the present embodiment；
Fig. 8 is the multinode wheel rail force load Identification Data idea about modeling schematic diagram of the present embodiment；
Fig. 9 a are that the collaboration of the present embodiment is conjugated the 1st principal component result schematic diagram of sparse Principal Component Analysis Algorithm；
Fig. 9 b are that the collaboration of the present embodiment is conjugated the 2nd principal component result schematic diagram of sparse Principal Component Analysis Algorithm；
Fig. 9 c are that the collaboration of the present embodiment is conjugated the 3rd principal component result schematic diagram of sparse Principal Component Analysis Algorithm；
Fig. 9 d are that the collaboration of the present embodiment is conjugated the 4th principal component result schematic diagram of sparse Principal Component Analysis Algorithm；
Fig. 9 e are that the collaboration of the present embodiment is conjugated the 5th principal component result schematic diagram of sparse Principal Component Analysis Algorithm；
Fig. 9 f are that the collaboration of the present embodiment is conjugated the 6th principal component result schematic diagram of sparse Principal Component Analysis Algorithm；
Figure 10 is the multinode L of the present embodiment1/2Identification of-SPCA-ELM Neural Network Datas the model to actual measurement wheel rail force
Result schematic diagram；
Figure 11 is the multinode L of the present embodiment1/2Identification of-SPCA-ELM Neural Network Datas the model to emulation wheel rail force
Result schematic diagram；
Figure 12 be a kind of wheel rail force load provided in an embodiment of the present invention recognize feature-based data model establish device block diagram it
One；
Figure 13 be a kind of wheel rail force load provided in an embodiment of the present invention recognize feature-based data model establish device block diagram it
Two；
Figure 14 be a kind of wheel rail force load provided in an embodiment of the present invention recognize feature-based data model establish device block diagram it
Three；
Figure 15 is that the wheel rail force load identification feature-based data model of the embodiment of the present invention establishes model foundation unit in device
Functional block diagram.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation describes, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
The operation principle of the technical program is：Although having carried out a large amount of research work in terms of wheel rail force identification both at home and abroad
Make, but how using vehicle dynamic response acceleration establish data model identification wheel rail force yet there are no pertinent literature report and specially
Sharp explanation.If directly establishing data model using vehicle dynamic response acceleration information recognizes wheel rail force, two problems can be encountered：
(1) vehicle dynamic response and wheel rail force do not correspond in a frequency domain：The sensor frequency response model used due to detection device
Wide, sample frequency height is enclosed, therefore vehicle dynamic response and wheel rail force can be divided into low frequency component and high fdrequency component in a frequency domain, and
And include noise jamming, so to carry out time-frequency characteristics extraction to initial data and can just establish data after being filtered
Model；
(2) there is delay effect in vehicle dynamic response and wheel rail force：To wheel rail system effect mechanism analyzed it is found that
The vehicle dynamic response of multiple nodes of locations is related to current location node and before this for wheel rail force at a certain nodes of locations, because
This characteristic for needing introducing multinode thought fusion time frequency analysis to extract establishes data model.
For presently, there are the problem of, propose a kind of to be based on multinode L1/2The wheel rail force of-SpasePCA-ELM neural networks
Load recognizes characteristic Fusion Modeling Method, the method for opposite modelling by mechanism identification wheel rail force, and data modeling method can not only have
Effect reduces calculation amount, and can preferably describe the complex nonlinear relationship between wheel rail system, good, the extensive energy with stability
The advantages that power is strong, training speed is fast, identification precision is high.
Detailed technical solution is as shown in Figure 1, this programme includes three parts：First part calculates vehicle dynamic response and accelerates
The time-frequency distributions of degree and wheel track force data utilize based on the variable element domain and in short-term adaptive signal decomposition of Gaussian linear frequency modulation base
Algorithm extracts the time-frequency characteristics of data；Second part merges the characteristic of multinode using sparse principal component analysis on this basis
The correlation of data is eliminated according to this and reduces dimension, and training dataset is provided for Part III；After Part III is according to fusion
Multinode characteristic, uses L1/2- SpasePCA-ELM algorithms establish data model identification wheel rail force.
Based on above-mentioned analysis, as shown in Fig. 2, recognizing characteristic for a kind of wheel rail force load that the embodiment of the present invention proposes
Method for establishing model, including：
Step 201)：It is extracted using the adaptive signal decomposition algorithm of based on variable element domain and in short-term Gaussian linear frequency modulation base
The time-frequency characteristics of track-Vehicular system detection data；
In this step, using based on the variable element domain and in short-term adaptive signal decomposition algorithm pair of Gaussian linear frequency modulation base
The time-frequency characteristics attribute of track-Vehicular system detection data is analyzed and is extracted, including：Survey the cross of car body, framework, axle box
To, Vertical Acceleration and laterally, vertical wheel track force data.
The step of adaptive signal decomposition algorithm of Gaussian linear frequency modulation base can be described as follows based on variable element domain and in short-term：
(1) m=0, the signal s (t) being analysed to are assigned to Sm(t) prepare to decompose；
(2) to signal Sm(t) Short Time Fourier Transform is done, energy peak point (t is obtainedm,ωm).It may search at this time more
The equal peak point of a energy, energy peak point of the optional peak point as signal；
(3) using the peak point as time-frequency center, STFT segments are intercepted in time frequency plane with small initial rectangular window, it is contemplated that
The length of STFT Time-Frequency Analysis Method features, initial rectangular window is typically no less than 33 data collection points, is generally no greater than simultaneously
1024 data points；
(4) still widen rectangular window using the peak point as time-frequency center, intercept STFT segments, the width widened every time again
Degree can take the half of initial rectangular window width or initial rectangular window width.When rectangular window intercept signal arriving signal length boundary
When stop, as shown in Figure 3.
(5) segment being intercepted to j-th seeks its time bandwidth T with following formulajWith frequency bandwidth Bj, j=0,1,
2…；
In formula：The energy for being intercepted segment for j-th.Here to be decomposed
Signal is finite energy signal.
(6) observing time bandwidth TjVariation, when time bandwidth in some section stablize when, obtain TstableWith
Bstable.The Heisenberg boxes being made up of stable time width and frequency bandwidth calculate stable αstableAnd βstableValue is such as
Under：
In formula：G is the time bandwidth T of stabilizationstableWith time scale factor αstableBetween function, H is stable time
Bandwidth TstableWith stable frequency bandwidth BstableWith frequency modulation rate βstableBetween function；
(7) with the energy peak point (t of the secondary searchm,ωm) it is used as time-frequency center, in the α found outstableAnd βstableNear
K basic function h is generated in the range of 5%mk(αmk,tm,ωm,βmk), k ∈ 1,2 ... K calculate inner product, inner product will be made to be maximized
One group of parameter as the step decompose in best match Gaussian linear frequency modulation base hm(αm,tm,ωm,βm) one group of parameter.
(8) from signal sm(t) s is removed inm(t) in hm(t) projection obtains residue signal sm+1(t), residue signal is calculated
Energy, when residue signal energy be less than a certain setting value when stop decompose, repeat the above steps when being unsatisfactory for and continue to decompose.
As shown in figure 4, for the present embodiment vehicle dynamic response acceleration information time-frequency characteristics CPST-AGCD distribution maps it
One；As shown in figure 5, being the two of the vehicle dynamic response acceleration information time-frequency characteristics CPST-AGCD distribution maps of the present embodiment；Such as
It is one of the wheel track force data time-frequency characteristics CPST-AGCD distribution maps of the present embodiment shown in Fig. 6；As shown in fig. 7, being this implementation
The two of the wheel track force data time-frequency characteristics CPST-AGCD distribution maps of example.Vehicle vibration response acceleration and wheel track force data when
The corresponding incidence relation of existing characteristics on frequency face, thus extract character pair as the training dataset for establishing data model, can
Effectively remove noise jamming.
Step 202)：Sparse Principal Component Analysis Algorithm is conjugated using multi-node collaborative to detect the track-Vehicular system
The time-frequency characteristics of data are merged；
Before establishing wheel rail force load Identification Data model, it is desirable to acquisition coherent detection data as much as possible, to
Prestige can have wheel rail force load identification problem relatively complete understanding and assurance.Although these detection datas can more comprehensively and
The dynamic interaction situation of track-Vehicular system is accurately described, but is applied in real data modeling process, these detections
Data really not necessarily are able to play expected effect, it is also possible to dimension disaster, mass data meter can be brought to data modeling
Calculate, the correlation between variable, synteny the problems such as.Information high superposed can give system to related between variable in these data
The successful application of meter learning method brings many difficulties.Such as in regression problem, there is stronger between many explanatory variables
Correlation, i.e. height multicollinearity then can bring many difficulties to the parameter Estimation of regression equation, lead to regression equation parameter
The problems such as data model that inaccuracy even obtains is unavailable.
To solve the above-mentioned problems it is necessary to reduce the number of variable, while wishing to retain the letter in original variable as possible again
Breath.For this reason, it may be necessary to using a kind of more efficiently solution, it can about subtract the variable number for participating in data modeling, while again
It will not lead to a large amount of loss of information.Sparse principal component analysis be exactly it is such a can not only effectively reduce dimension, but also
The main information in initial data and the also Feature Data Fusion Method with interpretation can be retained.
In the conventional method of data modeling and processing regression problem, data mould is established commonly using the thought of single node
Type, the i.e. input sample of data model and output sample are one-to-one.Such as：Each group of input sample of model be all
Car body, framework, left and right axle box at a certain position after time-frequency characteristics extract be vertical and 8 dimensions of transverse acceleration composition
According to each group of output sample of model is also the vertical and lateral wheel track obtained after time-frequency characteristics extract at same position
Power, during training data model, these samples are one-to-one.But for wheel rail force load identification problem, root
There is wavelength model according to track irregularity known to the prioris such as existing wheel-rail interaction mechanism and the measuring principle of instrumented wheelset
Enclose, vehicle dynamic response also has delayed-action, so the wheel rail force measured at a certain position of instrumented wheelset be not only by
It is also close with the state of several nodes neighbouring before current location caused by the track irregularity and vehicle acceleration of current location
Cut phase is closed.Therefore the thought for introducing multinode establishes wheel rail force load Identification Data model, as shown in Figure 8.
When establishing multinode wheel rail force load Identification Data model, node where current location and neighbouring before is used
The vehicle dynamic response characteristic of wide node location is combined into one group of sample, defeated at this time as the input of data model
It is wide*8 to enter data dimension, and the output of model is still the wheel rail force of current location.Since there is correlations for higher-dimension input data
Property and multicollinearity, so need using sparse principal component analysis method carry out characteristic fusion, eliminate it is therein
The interference such as correlation and multicollinearity, while the dimension of input data is reduced, to establish wheel rail force load Identification Data model
Effective characteristic is provided.
Sparse principal component analytical method increases sparsity in Principal Component Analysis, and principal component analysis uses Data Dimensionality Reduction
Method, the generalized variable for finding out several minorities replaces original multiple variables, between the generalized variable for keeping these a small number of
It is orthogonal and the information in former variable can be retained as far as possible.In general, processing method mathematically is that former variable is made line
Property combination.Information in former variable can use variance measures, and variance is bigger, and the information of reservation is more.First linear group of note
It is Y to close obtained generalized variable1, there is maximum variance var (Y1), referred to as first principal component；If first principal component is not enough to represent
The information of p variable originally just chooses second linear combination Y again2, it is called Second principal component, and so on, principal component
The mathematical notation of analytic approach is as follows.
Real data X=(X1,X2,…Xp) in have p stochastic variable, covariance matrix ΣXIt is as follows：
ΣX=E [(X-E (X)) (X-E (X))T] (4)
In formula, X indicates the attributive character of node；If covariance matrix ΣXCharacteristic value be λ1≥λ2…≥λp>=0, it is corresponding
Unit orthogonal eigenvectors [u1,u2…,up]：
In formula：uk=(u1k,u2k…,upk)T, k=1,2 ..., p
Then corresponding k-th of the principal component of X is：Yk=ukX=u1kX1+u2kX2+…+upkXp (6)
Principal component has following two attributes：
I.e.：New Set Y1,Y2,…Yj…YkIt can fully reflect the information in original index set and mutual indepedent, drop
The low dimension of data, eliminates the interference of the correlation and synteny in initial data variable.
Based on traditional principal component analytical method, introduces rational sparsity constraints and obtain sparse load vectors, be dilute
Dredge the main problem of principal component analysis.Introduce L0After restrictive condition, the form of sparse principal component analysis is as follows：
In formula：The first two constraints indicates load vectors ukWith unit length and mutually orthogonal, the last one is about
Beam condition increases L0Sparsity limits, t0It is a constant.But the problem of above problem is a NP-Hard can not directly be asked
Solution can only use the methods of relaxed algorithm to obtain approximate solution.Therefore, it is conjugated sparse Principal Component Analysis using collaboration and passes through iteration
Search determines that conjugate variables, these conjugate variables can be such that the object function in sparse principal component analysis problem obtains utmostly
On increase, and increase according to object function the coefficient of progressive updating conjugate variables, the expression formula of the algorithm is as follows：
In formula：S and c is two preassigned constants.uTFor unit orthogonal eigenvectors [u1,u2…,up] transposition square
Battle array；U is unit orthogonal eigenvectors [u1,u2…,up]；ΣXFor covariance matrix, ΣX=E [(X-E (X)) (X-E (X))T], X
Indicate the attributive character of node, X=(X1,X2,…Xp), p indicates the attribute number of node.
Collaboration be conjugated sparse Principal Component Analysis Algorithm propose it is a kind of calculate sparse principal component analysis whole solution path can
Capable method has good application effect, syncretizing effect such as Fig. 9 a, figure for track-Vehicular system detection data Fusion Features
9b, Fig. 9 c, Fig. 9 d, Fig. 9 e, shown in Fig. 9 f.From the result that sparse principal component analysis obtains as can be seen that in track-vehicle system
In detection data of uniting, the relationship between car body vertical acceleration and axle box vertical acceleration and wheel rail force is more close.Therefore, melt
Characteristic after conjunction will also play prior effect in establishing wheel rail force load Identification Data model.
Step 203)：Using multinode Fusion Features data as sample data set, L is used1/2- SparsePCA-ELM nerves
Net machine learning algorithm establishes wheel rail force load identification feature-based data model.
It is based on variable element domain in use and Gaussian linear frequency modulation base adaptive signal decomposition algorithm is extracted track-vehicle in short-term
The time-frequency characteristics of system detectio data, after having carried out the fusion of multinode characteristic by the method for sparse principal component analysis,
It has obtained non-correlation and has contained the valid data collection of multinode characteristic complete information for establishing data model.With
Multinode Fusion Features data use L as training and test sample data set1/2- SparsePCA-ELM neural network machines
Learning algorithm establishes wheel rail force load Identification Data model.Data model is established to be as follows：
Step1：Using based on the variable element domain and in short-term adaptive signal decomposition algorithm of Gaussian linear frequency modulation base to track-
Vehicular system detection data is filtered data after carrying out time-frequency characteristics extraction and analysis, forms characteristic.
Step2：Characteristic is divided into training group and test group, normalizing is carried out to every group of the data that output and input respectively
Change is handled, and eliminates the interference come by order of magnitude different band.
Step3：The thought of multinode is introduced, according to test result, the size of wide is generally set to 6, using based on more piece
The method of the sparse principal component analysis of point carries out fusion treatment to the characteristic after normalization, reduces the dimension of multinode characteristic
Count and eliminate correlation present in data and synteny interference.
Step4：Determine the number initial value of hidden neuron；Wherein, use larger hidden neuron number as initial value, it can
It is identical as training set sample number, when training dataset sample number is very big, one can be selected to be far longer than input data dimension
Value as hidden neuron node number initial value.
Step5：According to the number initial value of the hidden neuron obtain hidden neuron between input weight coefficient matrix and
Hidden neuron threshold matrix；
Step6：Hidden neuron activation primitive is determined according to input weight coefficient matrix and hidden neuron threshold matrix；
Step7：Multinode Fusion Features data in the training group and test group are substituted into neural network input layer, are led to
Cross the output that input weight coefficient matrix, hidden neuron threshold matrix and hidden neuron activation primitive obtain hidden layer neuron
Matrix；
Step8：Using the output matrix of the hidden layer neuron, according to L1/2Regularization threshold value iterative algorithm obtains institute
The optimal solution for stating the connection weight matrix between hidden neuron and output layer neuron, according to the hidden neuron and output
The number of non-zero element redefines of the hidden neuron in the optimal solution of connection weight matrix between layer neuron
Number；
Step9：Utilize input weight coefficient matrix, hidden neuron threshold matrix, the hidden neuron and output layer god
Optimal solution through the connection weight matrix between member carries out cross validation, judges the wheel rail force load identification feature-based data model
Whether engineer application demand is met；If it is satisfied, then utilizing input weight coefficient matrix, hidden neuron threshold matrix, described hidden
The optimal solution of connection weight matrix between layer neuron and output layer neuron and the optimal solution according to connection weight matrix
In the number of hidden neuron that redefines of number of non-zero element establish and be based on multinode L1/2- SparsePCA-ELM nerves
The wheel rail force load of network recognizes feature-based data model；Otherwise, to the time-frequency characteristics weight of the track-Vehicular system detection data
It is new to divide training group and test group, input weight coefficient matrix, hidden neuron threshold matrix, the hidden neuron are selected again
The optimal solution of connection weight matrix between output layer neuron, until the wheel rail force load established recognizes feature-based data model
Until meeting engineer application requirement.
It is based on multinode L using measured data and emulation data verification1/2- SparsePCA-ELM neural network methods are established
Wheel rail force load Identification Data model, as a result as shown in Figure 10,11, this Data Modeling Method is right in whole statistical significance
Actual measurement and emulation wheel rail force have good identification effect.
In addition, the research that the wheel rail force load discrimination method based on data modeling is hybrid modeling indicates a new side
To.I.e.：To certain vehicle and certain circuit, a small amount of representative emulation data set first is obtained using modelling by mechanism method,
Data model is established by Data Modeling Method again, the wheel rail force of whole circuit can be gone out with Fast Identification, and with without making
The advantage being trained with the actual measurement wheel track force data of instrumented wheelset.
The embodiment of the present invention also provides a kind of computer-readable program, wherein when executing described program in the electronic device
When, described program makes computer execute method as described in Figure 2 in the electronic equipment.
The embodiment of the present invention also provides a kind of storage medium being stored with computer-readable program, wherein the computer can
Reader makes computer execute method as described in Figure 2 in the electronic device.
One of ordinary skill in the art will appreciate that realizing all or part of flow in above-described embodiment method, Ke Yitong
It crosses computer program and is completed to instruct relevant hardware, the program can be stored in general computer read/write memory medium
In, the program is when being executed, it may include such as the flow of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic
Dish, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random Access
Memory, RAM) etc..
It should be noted that although describing the operation of the method for the present invention with particular order in the accompanying drawings, this is not required that
Or imply and must execute these operations according to the particular order, it could the realization phase or have to carry out operation shown in whole
The result of prestige.Additionally or alternatively, it is convenient to omit multiple steps are merged into a step and executed by certain steps, and/or will
One step is decomposed into execution of multiple steps.
As shown in figure 12, it is that a kind of wheel rail force load identification feature-based data model provided in an embodiment of the present invention establishes device
One of block diagram.Including：
Feature extraction unit 1, for utilization, based on variable element domain and in short-term, the self-adapting signal of Gaussian linear frequency modulation base divides
Resolving Algorithm extracts the time-frequency characteristics of track-Vehicular system detection data；
Fused data unit 2, for using multi-node collaborative to be conjugated sparse Principal Component Analysis Algorithm to the track-vehicle
The time-frequency characteristics of system detectio data are merged；Wherein, the multi-node collaborative conjugation that the fused data unit 2 uses is sparse
The expression formula of Principal Component Analysis Algorithm is：
Wherein, s and c is two preassigned constants；uTFor unit orthogonal eigenvectors [u1,u2…,up] transposition square
Battle array；U is unit orthogonal eigenvectors [u1,u2…,up]；ΣXFor covariance matrix, ΣX=E [(X-E (X)) (X-E (X))T], X
Indicate the attributive character of node, X=(X1,X2,…Xp), p indicates the attribute number of node.
Model foundation unit 3, for using multinode Fusion Features data as sample data set, using L1/2-
SparsePCA-ELM neural network machine learning algorithms establish wheel rail force load identification feature-based data model.
As shown in figure 13, it is that a kind of wheel rail force load identification feature-based data model provided in an embodiment of the present invention establishes device
The two of block diagram.On the basis of Figure 12, further include：
Filter unit 1 ', for the time-frequency characteristics of the track-Vehicular system detection data carry out fusion treatment it
Before, the track-Vehicular system detection data is filtered by time-frequency characteristics.
As shown in figure 14, it is that a kind of wheel rail force load identification feature-based data model provided in an embodiment of the present invention establishes device
The three of block diagram.On the basis of Figure 13, further include：
Normalized unit 1 ", it is special for the time-frequency to the track-Vehicular system detection data after being filtered
Sign is divided into training group and test group, and every group of the data that output and input are normalized respectively.
As shown in figure 15, model in device is established for the wheel rail force load of embodiment of the present invention identification feature-based data model to build
Vertical Elementary Function block diagram.Including：
Hidden neuron number initial value determining module 31, the number initial value for determining hidden neuron；
First model parameter determining module 32, for obtaining hidden neuron according to the number initial value of the hidden neuron
Between input weight coefficient matrix and hidden neuron threshold matrix；
Hidden neuron activation primitive determining module 33, for according to input weight coefficient matrix and hidden neuron threshold value square
Battle array determines hidden neuron activation primitive；
Second model parameter determining module 34 is used for the multinode Fusion Features data in the training group and test group
Neural network input layer is substituted into, by inputting weight coefficient matrix, hidden neuron threshold matrix and hidden neuron activation primitive
Obtain the output matrix of hidden layer neuron；
Third model parameter determining module 35, for the output matrix using the hidden layer neuron, according to L1/2Canonical
Change the optimal solution that threshold value iterative algorithm obtains the connection weight matrix between the hidden neuron and output layer neuron, according to
The number of non-zero element redefines in the optimal solution of connection weight matrix between the hidden neuron and output layer neuron
The number of the hidden neuron；
Wheel rail force load identification feature-based data model establishes module 36, for utilizing input weight coefficient matrix, hidden layer nerve
The optimal solution of connection weight matrix between first threshold matrix, the hidden neuron and output layer neuron intersect testing
Card, judges whether the wheel rail force load identification feature-based data model meets engineer application demand；If it is satisfied, then utilizing input
Connection weight matrix between weight coefficient matrix, hidden neuron threshold matrix, the hidden neuron and output layer neuron
Optimal solution and optimal solution according to connection weight matrix in non-zero element the number of hidden neuron that redefines of number
It establishes and is based on multinode L1/2The wheel rail force load of-SparsePCA-ELM neural networks recognizes feature-based data model；Otherwise, to institute
The time-frequency characteristics for stating track-Vehicular system detection data repartition training group and test group, select input weight coefficient square again
The optimal solution of connection weight matrix between battle array, hidden neuron threshold matrix, the hidden neuron and output layer neuron,
Until the wheel rail force load identification feature-based data model established meets engineer application requirement.
Those skilled in the art will also be appreciated that the various functions that the embodiment of the present invention is listed are by hardware or soft
Part depends on the design requirement of specific application and whole system to realize.Those skilled in the art can be specific for each
Using, the function that the realization of various methods can be used described, but this realization is understood not to protect beyond the embodiment of the present invention
The range of shield.
In addition, although being referred to several units of device in above-detailed, this division is only not strong
Property processed.In fact, according to the embodiment of the present invention, the feature and function of two or more above-described units can be
It is embodied in one unit.Equally, the feature and function of an above-described unit can also be further divided by multiple
Unit embodies.
Above-described specific implementation mode has carried out further the purpose of the present invention, technical solution and advantageous effect
It is described in detail, it should be understood that the foregoing is merely the specific implementation mode of the present invention, is not intended to limit the present invention
Protection domain, all within the spirits and principles of the present invention, any modification, equivalent substitution, improvement and etc. done should all include
Within protection scope of the present invention.
Claims (10)
1. a kind of wheel rail force load recognizes feature-based data model method for building up, which is characterized in that including：
Track-Vehicular system is extracted using the adaptive signal decomposition algorithm of based on variable element domain and in short-term Gaussian linear frequency modulation base
The time-frequency characteristics of detection data；
It is conjugated time-frequency characteristics of the sparse Principal Component Analysis Algorithm to the track-Vehicular system detection data using multi-node collaborative
It is merged；
Using multinode Fusion Features data as sample data set, L is used1/2- SparsePCA-ELM neural network machines learn
Algorithm establishes wheel rail force load identification feature-based data model.
2. the method as described in claim 1, which is characterized in that further include：
Before carrying out fusion treatment to the time-frequency characteristics of the track-Vehicular system detection data, by the track-vehicle system
System detection data is filtered by time-frequency characteristics.
3. method as claimed in claim 2, which is characterized in that further include：
Training group and test group are divided into the time-frequency characteristics of the track-Vehicular system detection data after being filtered, respectively
Every group of the data that output and input are normalized.
4. the method as described in claim 1-3 any claims, which is characterized in that the multi-node collaborative is conjugated sparse master
Constituent analysis algorithm expression formula is：
Wherein, s and c is two preassigned constants；uTFor unit orthogonal eigenvectors [u1,u2…,up] transposed matrix；u
For unit orthogonal eigenvectors [u1,u2…,up]；ΣXFor covariance matrix, ΣX=E [(X-E (X)) (X-E (X))T], X is indicated
The attributive character of node, X=(X1,X2,…Xp), p indicates the attribute number of node.
5. method as claimed in claim 3, which is characterized in that described to use L1/2- SparsePCA-ELM neural network machines
Learning algorithm establish wheel rail force load identification feature-based data model the step of include：
Determine the number initial value of hidden neuron；
The input weight coefficient matrix and hidden neuron between hidden neuron are obtained according to the number initial value of the hidden neuron
Threshold matrix；
Hidden neuron activation primitive is determined according to input weight coefficient matrix and hidden neuron threshold matrix；
Multinode Fusion Features data in the training group and test group are substituted into neural network input layer, by inputting power system
Matrix number, hidden neuron threshold matrix and hidden neuron activation primitive obtain the output matrix of hidden layer neuron；
Using the output matrix of the hidden layer neuron, according to L1/2Regularization threshold value iterative algorithm obtains the hidden layer nerve
The optimal solution of connection weight matrix between member and output layer neuron, according to the hidden neuron and output layer neuron it
Between connection weight matrix optimal solution in the number of non-zero element redefine the number of the hidden neuron；
Using between input weight coefficient matrix, hidden neuron threshold matrix, the hidden neuron and output layer neuron
The optimal solution of connection weight matrix carries out cross validation, judges whether the wheel rail force load identification feature-based data model meets work
Journey application demand；If it is satisfied, then using input weight coefficient matrix, hidden neuron threshold matrix, the hidden neuron with
Non-zero element in the optimal solution of connection weight matrix between output layer neuron and the optimal solution according to connection weight matrix
The number for the hidden neuron that number redefines, which is established, is based on multinode L1/2The wheel rail force of-SparsePCA-ELM neural networks
Load recognizes feature-based data model；Otherwise, training group is repartitioned to the time-frequency characteristics of the track-Vehicular system detection data
And test group, input weight coefficient matrix, hidden neuron threshold matrix, the hidden neuron and output layer nerve are selected again
The optimal solution of connection weight matrix between member, until the wheel rail force load identification feature-based data model established meets engineer application
Until it is required that.
6. a kind of wheel rail force load identification feature-based data model establishes device, which is characterized in that including：
Feature extraction unit, for utilizing based on the variable element domain and in short-term adaptive signal decomposition algorithm of Gaussian linear frequency modulation base
Extract the time-frequency characteristics of track-Vehicular system detection data；
Fused data unit, for using multi-node collaborative to be conjugated sparse Principal Component Analysis Algorithm to the track-Vehicular system
The time-frequency characteristics of detection data are merged；
Model foundation unit, for using multinode Fusion Features data as sample data set, using L1/2-SparsePCA-ELM
Neural network machine learning algorithm establishes wheel rail force load identification feature-based data model.
7. device as claimed in claim 6, which is characterized in that further include：
Filter unit is used for before carrying out fusion treatment to the time-frequency characteristics of the track-Vehicular system detection data, by institute
Track-Vehicular system detection data is stated to be filtered by time-frequency characteristics.
8. device as claimed in claim 7, which is characterized in that further include：
Normalized unit is divided into for the time-frequency characteristics to the track-Vehicular system detection data after being filtered
Training group and test group are respectively normalized every group of the data that output and input.
9. the device as described in claim 6-8 any claims, which is characterized in that the fused data unit uses more
The expression formula that node collaboration is conjugated sparse Principal Component Analysis Algorithm is：
Wherein, s and c is two preassigned constants；uTFor unit orthogonal eigenvectors [u1,u2…,up] transposed matrix；u
For unit orthogonal eigenvectors [u1,u2…,up]；ΣXFor covariance matrix, ΣX=E [(X-E (X)) (X-E (X))T], X is indicated
The attributive character of node, X=(X1,X2,…Xp), p indicates the attribute number of node.
10. device as claimed in claim 8, which is characterized in that the model foundation unit includes：
Hidden neuron number initial value determining module, the number initial value for determining hidden neuron；
First model parameter determining module, for according to defeated between the number initial value of hidden neuron acquisition hidden neuron
Enter weight coefficient matrix and hidden neuron threshold matrix；
Hidden neuron activation primitive determining module, for being determined according to input weight coefficient matrix and hidden neuron threshold matrix
Hidden neuron activation primitive；
Second model parameter determining module, for the multinode Fusion Features data in the training group and test group to be substituted into god
It is hidden by inputting weight coefficient matrix, hidden neuron threshold matrix and the acquisition of hidden neuron activation primitive through network input layer
The output matrix of the neuron containing layer；
Third model parameter determining module, for the output matrix using the hidden layer neuron, according to L1/2Regularization threshold value
Iterative algorithm obtains the optimal solution of the connection weight matrix between the hidden neuron and output layer neuron, according to described hidden
The number of non-zero element redefines described hidden in the optimal solution of connection weight matrix between layer neuron and output layer neuron
The number of layer neuron；
Wheel rail force load identification feature-based data model establishes module, for utilizing input weight coefficient matrix, hidden neuron threshold value
The optimal solution of connection weight matrix between matrix, the hidden neuron and output layer neuron carries out cross validation, judges
Whether the wheel rail force load identification feature-based data model meets engineer application demand；If it is satisfied, then utilizing input weight coefficient
Connection weight matrix between matrix, hidden neuron threshold matrix, the hidden neuron and output layer neuron it is optimal
The number for the hidden neuron that the number of non-zero element redefines in solution and the optimal solution according to connection weight matrix establishes base
In multinode L1/2The wheel rail force load of-SparsePCA-ELM neural networks recognizes feature-based data model；Otherwise, to the rail
The time-frequency characteristics of road-Vehicular system detection data repartition training group and test group, select input weight coefficient matrix, hidden again
The optimal solution of connection weight matrix between layer neuron threshold matrix, the hidden neuron and output layer neuron, until
Until the wheel rail force load identification feature-based data model of foundation meets engineer application requirement.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710202179.2A CN108664673A (en) | 2017-03-30 | 2017-03-30 | A kind of wheel rail force load identification feature-based data model method for building up and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710202179.2A CN108664673A (en) | 2017-03-30 | 2017-03-30 | A kind of wheel rail force load identification feature-based data model method for building up and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108664673A true CN108664673A (en) | 2018-10-16 |
Family
ID=63785516
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710202179.2A Pending CN108664673A (en) | 2017-03-30 | 2017-03-30 | A kind of wheel rail force load identification feature-based data model method for building up and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108664673A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110263431A (en) * | 2019-06-10 | 2019-09-20 | 中国科学院重庆绿色智能技术研究院 | A kind of concrete 28d Prediction of compressive strength method |
| CN110443302A (en) * | 2019-08-02 | 2019-11-12 | 天津相和电气科技有限公司 | Load discrimination method and its application based on Fusion Features and deep learning |
| CN111027727A (en) * | 2019-12-27 | 2020-04-17 | 中南大学 | Cross-domain operation and maintenance key element identification method for track system |
| CN112013895A (en) * | 2019-05-28 | 2020-12-01 | 浙江德盛铁路器材股份有限公司 | Method for monitoring and evaluating quality condition of railway track basic equipment |
| CN112381027A (en) * | 2020-11-23 | 2021-02-19 | 西南交通大学 | Wheel polygon wave depth estimation method based on train axle box vertical acceleration signal |
| CN113155063A (en) * | 2021-01-26 | 2021-07-23 | 北京讯腾智慧科技股份有限公司 | Method and device for determining track bridge deformation data during train passing through bridge |
| CN114061534A (en) * | 2021-11-01 | 2022-02-18 | 中国国家铁路集团有限公司 | Track plate deformation detection method and device and electronic equipment |
| CN115060407A (en) * | 2022-06-14 | 2022-09-16 | 华东交通大学 | Flexible multi-mode sensing device for multi-dimensional wheel-rail force measurement and fusion method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101968629A (en) * | 2010-10-19 | 2011-02-09 | 天津理工大学 | PID (Proportional Integral Derivative) control method for elastic integral BP neural network based on RBF (Radial Basis Function) identification |
| CN102360454A (en) * | 2011-10-12 | 2012-02-22 | 北京交通大学 | Wheel-track force prediction method based on NARX (Nonlinear Auto-regressive with Extra Inputs) neural network |
| KR101256901B1 (en) * | 2011-07-27 | 2013-04-22 | 서울과학기술대학교 산학협력단 | Prediction methods for derailment of the wheels using the external force acted on the wheelset |
-
2017
- 2017-03-30 CN CN201710202179.2A patent/CN108664673A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101968629A (en) * | 2010-10-19 | 2011-02-09 | 天津理工大学 | PID (Proportional Integral Derivative) control method for elastic integral BP neural network based on RBF (Radial Basis Function) identification |
| KR101256901B1 (en) * | 2011-07-27 | 2013-04-22 | 서울과학기술대학교 산학협력단 | Prediction methods for derailment of the wheels using the external force acted on the wheelset |
| CN102360454A (en) * | 2011-10-12 | 2012-02-22 | 北京交通大学 | Wheel-track force prediction method based on NARX (Nonlinear Auto-regressive with Extra Inputs) neural network |
Non-Patent Citations (1)
| Title |
|---|
| 郭剑峰: "基于数据建模的轮轨力载荷辨识理论和应用研究", 《万方数据库知识服务平台》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112013895A (en) * | 2019-05-28 | 2020-12-01 | 浙江德盛铁路器材股份有限公司 | Method for monitoring and evaluating quality condition of railway track basic equipment |
| CN110263431A (en) * | 2019-06-10 | 2019-09-20 | 中国科学院重庆绿色智能技术研究院 | A kind of concrete 28d Prediction of compressive strength method |
| CN110443302A (en) * | 2019-08-02 | 2019-11-12 | 天津相和电气科技有限公司 | Load discrimination method and its application based on Fusion Features and deep learning |
| CN110443302B (en) * | 2019-08-02 | 2023-06-09 | 天津相和电气科技有限公司 | Load identification method based on feature fusion and deep learning and application thereof |
| CN111027727A (en) * | 2019-12-27 | 2020-04-17 | 中南大学 | Cross-domain operation and maintenance key element identification method for track system |
| CN111027727B (en) * | 2019-12-27 | 2023-06-09 | 中南大学 | Rail system cross-domain operation and maintenance key element identification method |
| CN112381027A (en) * | 2020-11-23 | 2021-02-19 | 西南交通大学 | Wheel polygon wave depth estimation method based on train axle box vertical acceleration signal |
| CN113155063A (en) * | 2021-01-26 | 2021-07-23 | 北京讯腾智慧科技股份有限公司 | Method and device for determining track bridge deformation data during train passing through bridge |
| CN114061534A (en) * | 2021-11-01 | 2022-02-18 | 中国国家铁路集团有限公司 | Track plate deformation detection method and device and electronic equipment |
| CN115060407A (en) * | 2022-06-14 | 2022-09-16 | 华东交通大学 | Flexible multi-mode sensing device for multi-dimensional wheel-rail force measurement and fusion method |
| CN115060407B (en) * | 2022-06-14 | 2023-12-26 | 华东交通大学 | Flexible multi-mode sensing device for measuring multi-dimensional wheel rail force and fusion method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108664673A (en) | A kind of wheel rail force load identification feature-based data model method for building up and device | |
| CN104392071B (en) | A kind of bullet train system security assessment method based on complex network | |
| CN102629298B (en) | A kind of Rail Transit System security of operation appraisal procedure | |
| CN103335617B (en) | A kind of railway track geometric deformation detection method based on vibration signal | |
| CN103852269A (en) | High-speed train operation kinetic parameter detection method | |
| CN103995237A (en) | Satellite power supply system online fault diagnosis method | |
| CN106021789A (en) | Fuzzy-intelligence-based rail car suspension system fault classification method and system | |
| CN103471865A (en) | Train suspension system failure isolation method based on LDA method | |
| CN105242544A (en) | Non-linear multi-unmanned-aerial-vehicle-system fault-tolerance formation control method with consideration of random disturbance | |
| CN103674578A (en) | Detection method for high-speed train operation dynamics performance state | |
| CN105678343A (en) | Adaptive-weighted-group-sparse-representation-based diagnosis method for noise abnormity of hydroelectric generating set | |
| CN107239857A (en) | Overhead transmission line methods of risk assessment based on LS_SVM and PCA | |
| CN104809292A (en) | On-line recognizing method for nonlinear dynamic model parameter of high-speed train | |
| CN112989660B (en) | Method for predicting corrosion of pipeline under subway stray current based on partial least square method | |
| Mazzola et al. | Effect of suspension parameter uncertainty on the dynamic behaviour of railway vehicles | |
| CN102364490B (en) | Automatic synchronization recognition method based on hierarchical analyzing model | |
| CN104407273A (en) | Electric energy quality disturbance source positioning method considering monitoring reliability | |
| CN105071380A (en) | Power system wide-area coherency identification method and device | |
| CN106959397A (en) | A kind of method for designing of small fault diagnostic system for high ferro inverter | |
| Jiang et al. | A systematic optimization design method for complex mechatronic products design and development | |
| CN102519577A (en) | Method and system for identifying road surface in a road | |
| CN114580828A (en) | Safety evaluation method for numerical control machine during transportation process | |
| CN116467570B (en) | Heavy-duty locomotive coupler swing angle quantitative identification method based on data driving | |
| CN104036118B (en) | A kind of power system parallelization trace sensitivity acquisition methods | |
| Fan et al. | Adaptive RAS/SPS System Setting for Improving Grid Reliability and Asset Utilization through Predictive Simulation and Controls. A Use Case for Transformative Remedial Action Scheme Tool (TRAST): Jim Bridger RAS Evaluation and Analysis |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: No. 2, Da Liu Shu Road, Haidian District, Beijing Applicant after: China Railway Scientific Research Institute Group Co., Ltd. Applicant after: Institute of infrastructure inspection, China Railway Research Institute Group Co., Ltd. Applicant after: Beijing IMAP Technology Co., Ltd. Applicant after: Xi'an Jiaotong University Address before: No. 2, Da Liu Shu Road, Haidian District, Beijing Applicant before: China Academy of Railway Sciences Applicant before: Infrastructure Testing Institute of China Academy of Railway Sciences Applicant before: Beijing IMAP Technology Co., Ltd. Applicant before: Xi'an Jiaotong University |
|
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181016 |