CN106781502B - A kind of fuel-engined vehicle road conditions recognition methods based on vector quantization training pattern - Google Patents
A kind of fuel-engined vehicle road conditions recognition methods based on vector quantization training pattern Download PDFInfo
- Publication number
- CN106781502B CN106781502B CN201710026267.1A CN201710026267A CN106781502B CN 106781502 B CN106781502 B CN 106781502B CN 201710026267 A CN201710026267 A CN 201710026267A CN 106781502 B CN106781502 B CN 106781502B
- Authority
- CN
- China
- Prior art keywords
- road conditions
- vector
- typical
- training pattern
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
Abstract
The invention discloses a kind of fuel-engined vehicle road conditions recognition methods based on vector quantization training pattern, it is characterized in that choosing typical road conditions class and carrying out piecemeal to every kind of typical road conditions class, choose the characteristic parameter that can characterize each typical road conditions block feature and be standardized calculating；Build vector quantization training pattern and vector quantization training pattern is trained, until it meets systematic error requirement, the typical road conditions class that characterization model can be approximated to random road conditions is correctly recognized；Continuously gather the random road conditions block message in isometric time interval and calculate its corresponding standard feature parameter value, the vector quantization training pattern completed with training sequentially find with its most approximate typical road conditions class, road conditions identification is carried out to it.The present invention can identify the typical road conditions approximate with the random road conditions residing for fuel-engined vehicle online, realize that road conditions are recognized, be that on-line implement of the fuel-engined vehicle electricity system energy management strategies under random road conditions lays the foundation.
Description
Technical field
The present invention relates to a kind of vehicle technology, it is more particularly to a kind of by characteristic parameter approximately come with typical road conditions identification with
The road conditions recognition methods of machine road conditions, specifically a kind of fuel-engined vehicle road conditions identification side based on vector quantization training pattern
Method.
Background technology
During vehicle actual travel, road conditions change at random, it is impossible to be equal to any typical road conditions.So
And, for the correlative study of fuel-engined vehicle electricity system energy management strategies, it usually needs maintain vehicle in certain known allusion quotation
Under type road conditions.Therefore extremely it is necessary to carry out road conditions identification to the random road conditions during fuel-engined vehicle actual travel, is energy
The implementation of management strategy lays the foundation.
Up to now, some scholars carry out correlative study for the road conditions identification of traveling road conditions.The Qin is equal greatly
The degree of pressing close to of sample to be identified and master sample is represented by Euclid's note progress, road conditions identification is carried out；Liu Yong just etc.
23 typical state of cyclic operations are chosen, five classes are divided into clustering method, road conditions identification is carried out；Jansen etc. is by heredity
Optimization method and K mean cluster algorithm are combined, and realize that road conditions are recognized on the basis of parameter optimization.Above road conditions recognition methods
All there is deficiency to a certain extent, such as very few influence accuracy of identification of typical road conditions sample, identification model its own system error mistake
Greatly, the key parameter of model is indefinite and then the problems such as being unfavorable for practical application, thus by these approach applications to fuel-engined vehicle
Road conditions identification on when, recognition effect is poor, and accuracy of identification is relatively low.
Therefore, in the case where fuel-engined vehicle traffic information is unknown in advance, for complicated and random actual travel road conditions,
The typical road conditions approximate with the random road conditions residing for fuel-engined vehicle how are identified online, are realized that road conditions are recognized, are fuel oil vapour
Effective implementation of vehicle electric system capacity management strategy lays the foundation, and has become a urgent problem.
The content of the invention
The present invention is in order to overcome the above-mentioned deficiencies of the prior art, it is proposed that a kind of fuel oil based on vector quantization training pattern
Automobile road conditions recognition methods, to which the typical road conditions approximate with the random road conditions residing for fuel-engined vehicle can be identified online,
Effective identification of road conditions is realized, is that on-line implement of the fuel-engined vehicle electricity system energy management strategies under random road conditions establishes base
Plinth.
To achieve the above object, present invention employs following technical scheme：
A kind of the characteristics of fuel-engined vehicle road conditions recognition methods based on vector quantization training pattern of the present invention is by following step
It is rapid to carry out：
Step 1, the typical road conditions class of selection；
According to traffic and the difference of traveling region, the typical road conditions class of m kinds is chosen as identification sample, so as to constitute
Typical road conditions class set DC={ DC1,DC2,...,DCr,...,DCm, DCrRepresent the typical road conditions class of r kinds, 1≤r≤m；
Step 2, piecemeal is carried out to every kind of typical road conditions class；
Road conditions class DC typical to r kindsrPiecemeal is carried out, M typical road conditions block is obtained and constitutes set DCr={ dcr,1,
dcr,2,...,dcr,p,...,dcr,M, dcr,pRepresent corresponding p-th of the typical road conditions block of the typical road conditions class of r kinds, 1≤p≤M；
So as to obtain mM typical road conditions block and constitute set { dc1,1,...,dc1,m,...,dcr,p,...,dcm,1,...,dcm,M, note
For { dc1,dc2,...,dci,...,dcmM, dciRepresent i-th of typical road conditions block, 1≤i≤mM；
Step 3, choose and standardized calculation relevant feature parameters characterize the feature of each typical road conditions block；
To i-th of typical road conditions block dciChoose R characteristic parameter and be normalized, obtain i-th of standard feature
Parameter vector Ui=[ui,1,ui,2,...,ui,j,...,ui,R]T, ui,jRepresent i-th of typical road conditions block dciCorresponding j-th of mark
Quasi- characteristic parameter, so as to obtain mM standard feature parameter vector and constitute set U={ U1,U2,…,Ui,…,UmM, 1≤j≤
R；
Step 4, structure vector quantization training pattern；
The vector quantization training pattern is defined to be made up of input layer, competition layer and output layer；
The input vector for defining the vector quantization training pattern is X=[x1,x2,...,xj,...,xR]T；
Define the vector quantization training pattern input layer and compete interlayer connection matrix be
The input vector for defining the competition layer of the vector quantization training pattern is d=[d1,d2,...,di,...,dmM]T,
diRepresent i-th of standard feature parameter vector UiThe distance between with the input vector X of the model；
The output vector for defining the competition layer of the vector quantization training pattern is a=[a1,a2,...,ai,...,amM]T,
aiRepresent i-th of typical road conditions block dciState value, only ai=1 represents i-th of typical road conditions block dciAs described vector quantization
The typical road conditions block that the corresponding road conditions blocks of input vector X of training pattern are identified；
The connection matrix defined between the competition layer and output layer of the vector quantization training pattern isWherein u*=[1,1 ..., 1]M×1；
The output vector for defining the vector quantization training pattern is Y=[y1,y2,...,yr,...,ym]T, yrRepresent r
Plant typical road conditions class DCrState value, only yr=1 represents the typical road conditions class DC of r kindsrAs described vector quantization training pattern
The typical road conditions class that is identified of the corresponding road conditions blocks of input vector X；
Step 5, the input vector by the use of each standard feature parameter vector as vector quantization training pattern to it is described to
Amount quantifies training pattern and is trained, until it meets systematic error requirement, so as to obtain the vector quantization instruction available for identification
Practice model；
Step 5.1, defined variable b are frequency of training；Initialize b=1；
Step 5.2, initialization i=1；
Step 5.3, by the b times training when i-th of standard feature parameter vector Ui(b) as the input vector X of model
(i,b)；
Step 5.4, defined variable k, and 1≤k≤mM；Initialize k=1；
Step 5.5, judge whether k=i sets up, if so, step 5.6 is then performed, otherwise, step 5.7 is performed；
Step 5.6, k+1 is assigned to after k, judges whether k ＞ mM set up, if so, step 5.8 is then performed, otherwise, is returned
Return step 5.5；
K-th when step 5.7, the input vector X (i, b) for calculating the vector quantization training pattern are with the b times training
Standard feature parameter vector UkThe distance between (b) dkAfter (i, b), return to step 5.6；
Standard when step 5.8, the input vector X (i, b) for choosing the vector quantization training pattern are with the b times training is special
The minimum value in the distance between each standard feature parameter vector in parameter vector set U (b) is levied, d is designated asmin(i,b)；Will
The minimum value dminTypical road conditions block corresponding to (i, b) is designated as dcbest(i,b)；By the typical road conditions block dcbest(i,b)
Corresponding state value is designated as " 1 ", and the state value corresponding to remaining mM-1 typical road conditions block is designated as " 0 ", so as to obtain b
Correspond to the output vector a (i, b) of mode input vector X (i, b) competition layer during secondary training；
Step 5.9, using formula (1) obtain the b time training when corresponding to mode input vector X (i, b) model export to
Measure Y (i, b)：
Y (i, b)=U*·a(i,b) (1)
Step 5.10, find from the output vector Y (i, b) of the model typical road conditions class state value be " 1 " allusion quotation
Type road conditions class, is designated as DCbest(i,b)；
Step 5.11, defined variable δi(b) i-th of standard feature parameter vector U when for the b times trainingi(b) corresponding to
Recognition factor；Judge DCbestI-th of standard feature parameter vector U when (i, b) is with the b times trainingi(b) the typical road conditions corresponding to
Whether class is identical, if identical, then it represents that the vector quantization training pattern can correctly recognize that i-th of standard is special during the b times training
Levy parameter vector Ui(b) the typical road conditions class corresponding to, and remember Ui(b) the recognition factor δ corresponding toi(b) after=1, step is performed
5.12；Otherwise, U is rememberedi(b) the recognition factor δ corresponding toi(b) after=0, step 5.13 is performed；
Step 5.12, formula (2) is utilized to obtain i-th of typical case road conditions block dc after being adjusted in the b time trainingi
J standard feature parameter u 'i,j(b)；
u′i,j(b)=ui,j(b)+φ(xj-ui,j(b)) (2)
In formula (2), φ is learning rate, and φ be on the occasion of；
Step 5.13, formula (3) is utilized to obtain i-th of typical case road conditions block dc after being adjusted in the b time trainingi
J standard feature parameter u 'i,j(b)；
u′i,j(b)=ui,j(b)-φ(xj-ui,j(b)) (3)
Step 5.14, i+1 is assigned to after i, judges whether i ＞ mM set up, if so, then perform step 5.15；Otherwise,
Return to step 5.3；
Step 5.15, systematic error e (b)=δ (the b)/mM for calculating model after the b times training, wherein
Step 5.16, initialization system error threshold are e*；As e (b)<e*When, export connection matrix U (b), the vector quantity
Change training pattern training to complete；As e (b) >=e*When, by u 'i,j(b) i-th of typical road conditions block dc when being trained as the b+1 timesi
J-th of standard feature parameter ui,j(b+1) after, step 5.17 is performed；
Step 5.17, b+1 is assigned to b, repeat step 5.2 to step 5.16；
Step 6, to random road conditions carry out road conditions identification；
Random road conditions block in step 6.1, collection continuous time section, and R characteristic parameter of random road conditions block is carried out
Normalized, obtains the corresponding standard feature parameter vector of random road conditions block successively；
Step 6.2, it regard the corresponding standard feature parameter vector of the random road conditions block as what the training was completed successively
The input vector of vector quantization training pattern is identified, so as to obtain the identification of the random road conditions block in the continuous time section
As a result.
Compared with the prior art, beneficial effects of the present invention exist：
1st, the present invention devises a kind of fuel-engined vehicle road conditions recognition methods based on vector quantization training pattern, can be online
Ground identifies the typical road conditions approximate with the random road conditions residing for fuel-engined vehicle, realizes that road conditions are recognized.For common road conditions class
Type chooses the typical road conditions class of identical quantity and is divided into the typical road conditions block of identical quantity, while choosing identical characteristic parameter
Its feature is characterized, the fairness of road conditions identification is fundamentally embodied, it is ensured that accuracy of identification.Self instruction can be carried out by devising
Experienced vector quantization training pattern, by constantly model training, has gradually reduced the systematic error of model, until its satisfaction essence
Degree is required.The random road conditions block message in isometric time interval is continuously while gathered, with vector quantization training pattern successively
Road conditions identification is carried out to it, it is ensured that this method can embody the road in vehicle actual travel for the recognition result of random road conditions
Condition changes.
2nd, the present invention have chosen the typical road conditions class of m kinds altogether, while every kind of typical road conditions class is carried out into piecemeal, obtain identical number
The typical road conditions block of M of amount, the characteristic parameter of road conditions feature can fully be characterized by choosing identical R, and this is done so that every
Planting road conditions has the identification sample of identical quantity, embodies the fairness of road conditions identification, it is ensured that accuracy of identification.Meanwhile,
Calculating is standardized to the characteristic parameter of each typical road conditions block using the method for normalized, different characteristic ginseng is eliminated
Unit, numerical value and excursion recognize the interference caused to road conditions between number so that different characteristic parameters is known for road conditions
Not Ju You identical influence, further improve road conditions identification accuracy.
3rd, due to parameter calculation error, road conditions phantom error, and apart from the presence of approximate error, the present invention is with typical case
Road conditions block message calculates obtained initial connection matrix U and tended not to so that model has preferable accuracy of identification.Therefore, in structure
Build after vector quantization training pattern, the learning training function that performance model itself possesses, with known typical road conditions block conduct
Whether input, examine the typical road conditions class of output corresponding identical with input.By defining learning rate ф so that what it is in model is
System error is met before required precision, and model constantly carries out the training optimization using ф as speed, until the system of model is missed
Difference, which is met, to be required.The systematic error of model is reduced by as above designing, accuracy of identification is further increased.
4th, because the road conditions during vehicle actual travel are completely randoms, its own includes many unknown
Road condition change.If carrying out single road conditions for the actual road conditions in whole vehicle travel process recognizes that its result can not
Embody the road condition change process in vehicle travel process so that road conditions identification loses meaning.Therefore road conditions involved in the present invention
Recognition methods continuously gathers the random road conditions block message in isometric time interval, and the vector quantization completed with training trains mould
Type carries out road conditions identification to it successively, it is ensured that this method can embody vehicle actual travel for the recognition result of random road conditions
In road condition change process.
Brief description of the drawings
Fig. 1 is road conditions recognition methods basic flow sheet of the present invention；
Fig. 2 is the typical road conditions piecemeal schematic diagram of the present invention；
Fig. 3 is vector quantization training pattern structure chart of the present invention；
Fig. 4 is vector quantization training pattern model training flow chart of the present invention；
Fig. 5 is vector quantization training pattern road conditions identification process figure of the present invention.
Embodiment
In the present embodiment, a kind of fuel-engined vehicle road conditions recognition methods based on vector quantization training pattern is：Choose typical case
Road conditions class simultaneously carries out piecemeal；The feature of each typical road conditions block of selected characteristic parameter characterization is simultaneously standardized calculating；Build
Vector quantization training pattern is simultaneously trained to vector quantization training pattern, until it can correctly identify that random road conditions are approximated
Typical road conditions class；Gather random road conditions block message and calculate its corresponding standard feature parameter value, with training complete to
Amount quantify training pattern find with random road conditions block the most approximate typical road conditions class, realize that road conditions are recognized.This road conditions identification side
The basic procedure of method is as shown in Figure 1.
The sample preparatory stage, typical road conditions class is chosen, piecemeal is carried out to every kind of typical road conditions class, chosen and standardized calculation
Relevant feature parameters characterize the feature of each typical road conditions block.
Step 1, the typical road conditions class of selection.
According to traffic and the difference of traveling region, the typical road conditions class of m kinds is chosen as identification sample, so as to constitute
Typical road conditions class set DC={ DC1,DC2,...,DCr,...,DCm, DCrRepresent the typical road conditions class of r kinds, 1≤r≤m.Can be with
The Velocity-time of typical road conditions class needed for being obtained by the ADVISOR automobiles simulation software based on MATLAB/Simulink is bent
Line, and then obtain corresponding typical road conditions category information.
In the present embodiment, according to traffic and the difference of traveling region, traveling road conditions are divided into urban road, suburb
Road, 3 kinds of fundamental types of highway.For every kind of traveling road conditions, two kinds of typical road conditions classes are respectively have chosen as representative, it is every kind of
The corresponding typical road conditions of road conditions type have similar speed feature.Therefore, the present embodiment have chosen 6 kinds of typical road conditions altogether,
Be respectively NYCC and MANHATTAN (representing urban road), CSHVR and WVUSUB (representing rural road), US06_HWY and
HWFET (represents highway), i.e. m=6.
Step 2, piecemeal is carried out to every kind of typical road conditions class.
Vector quantization training pattern can not ensure preferable road conditions accuracy of identification when due to sample size being 6, therefore need pair
The speed-time curve of every kind of typical road conditions class carries out piecemeal, further exptended sample quantity on time dimension.To r kinds
Typical road conditions class DCrPiecemeal is carried out, M typical road conditions block is obtained, constitutes set DCr={ dcr,1,dcr,2,...,dcr,p,...,
dcr,M, dcr,pRepresent corresponding p-th of the typical road conditions block of the typical road conditions class of r kinds, 1≤p≤M；So as to obtain mM typical road
Condition block simultaneously constitutes set { dc1,1,...,dc1,m,...,dcr,p,...,dcm,1,...,dcm,M, it is designated as { dc1,dc2,...,
dci,...,dcmM, dciRepresent i-th of typical road conditions block, 1≤i≤mM.
In the present embodiment, piecemeal is carried out to typical road conditions class using " compound equisection method ", i.e., by every kind of typical road conditions class
Speed-time curve carries out decile, obtains corresponding typical road conditions block, then using the road conditions block between adjacent road conditions block midpoint as
New road conditions block, obtains corresponding M typical road conditions block, is used as the identification sample for representing the typical road conditions class altogether.Take this
Kind of method carry out piecemeal can eliminate the speed-time curves of different typical road conditions classes between total duration it is inconsistent to typical road conditions
The influence that piecemeal and calculation of characteristic parameters are brought.In the present embodiment, M=7 is taken, i.e., every kind of typical road conditions class is divided into 7
Typical road conditions block.The piecemeal signal of typical road conditions is as shown in Figure 2.
Step 3, choose and standardized calculation relevant feature parameters characterize the feature of each typical road conditions block.
To i-th of typical road conditions block dciChoose R characteristic parameter and be normalized, obtain i-th of standard feature
Parameter vector Ui=[ui,1,ui,2,...,ui,j,...,ui,R]T, ui,jRepresent i-th of typical road conditions block dciCorresponding j-th of mark
Quasi- characteristic parameter, so as to obtain mM standard feature parameter vector and constitute set U={ U1,U2,…,Ui,…,UmM, 1≤j≤
R。
In the present embodiment, each typical road of characteristic parameter sign that 10 important parameters are recognized as road conditions is have chosen altogether
The feature (i.e. R=10) of condition block, mainly includes：1. average speed2. max. speed vmax；3. average acceleration4. it is maximum
Acceleration amax；5. average retardation rate6. maximum deceleration dmax；7. at the uniform velocity time scale rc；8. acceleration time ratio ra；9.
Deceleration time ratio rd；10. dead time ratio ri.Meanwhile, using the method for normalized to the spy of each typical road conditions block
Levy parameter and be standardized calculating, and then unit, numerical value and excursion are recognized to road conditions between eliminating different characteristic parameter
The interference caused so that different characteristic parameters has identical influence for road conditions identification, further improves road conditions identification
Accuracy.
Its transforming function transformation function is as follows：
Wherein：C is characterized parameter raw data；cmimFor the minimum value of character pair parameter；cmaxFor character pair parameter
Maximum.
Step 4, structure vector quantization training pattern；
Model construction stage, the present embodiment constructs initial vector quantization training pattern, is that base is laid in the training of model
Plinth；
Definition vector quantifies training pattern and is made up of input layer, competition layer and output layer, and the input vector of model is sent into
Input layer, after competition layer, linear convergent rate layer obtains the output vector of corresponding model；
The input vector that definition vector quantifies training pattern is X=[x1,x2,...,xj,...,xR]T；
The input layer of definition vector quantization training pattern and the connection matrix of competition interlayer areWherein Ui TFor the corresponding standard feature parameter of i-th of typical case's road conditions block
Vector；ui,jFor the value of corresponding j-th of standard feature parameter of i-th of typical case's road conditions block；
The input vector that definition vector quantifies the competition layer of training pattern is d=[d1,d2,...,di,...,dmM]T, diTable
Show i-th of standard feature parameter vector UiThe distance between with the input vector X of the model, the size characterization model of distance
The corresponding road conditions blocks of input vector X and the fast dc of typical road conditionsiBetween degree of approximation, apart from it is smaller sign degree of approximation it is higher；
The output vector that definition vector quantifies the competition layer of training pattern is a=[a1,a2,...,ai,...,amM]T, aiTable
Show i-th of typical road conditions block dciState value；It is 1 to only have an element value in vectorial a, and remaining element value is 0, only ai=1
Represent i-th of typical road conditions block dciThe typical road conditions block that the corresponding road conditions blocks of input vector X of as model are identified；
Definition vector quantify training pattern competition layer and output layer between connection matrix beWherein u*=[1,1 ..., 1]M×1；Matrix U*Each column only have one 1, remaining is 0；Often
One u*Correspond to a kind of typical road conditions class, therefore shared m such u*。
The output vector that definition vector quantifies training pattern is Y=[y1,y2,...,yr,...ym]T, yrRepresent r kind allusion quotations
Type road conditions class DCrState value；It is 1 to only have an element value in vectorial Y, and remaining element value is 0, only yr=1 represents r kinds
Typical road conditions class DCrThe typical road conditions class that the corresponding road conditions blocks of input vector X of as model are identified.
In the present embodiment, m=6, M=7, R=10 are taken respectively, construct initial vector quantization training pattern.Vector quantity
The model structure for changing training pattern is as shown in Figure 3.
Step 5, model training stage, input vector of the present embodiment by the use of each standard feature parameter vector as model
Vector quantization training pattern is trained, until it meets systematic error requirement, so as to obtain the vector quantity available for identification
Change training pattern；
Step 5.1, defined variable b are frequency of training；Initialize b=1；
Step 5.2, initialization i=1；
Step 5.3, by the b times training when i-th of standard feature parameter vector Ui(b) as the input vector X of model
(i,b)；
Step 5.4, defined variable k, and 1≤k≤mM；Initialize k=1；
Step 5.5, judge whether k=i sets up, if so, step 5.6 is then performed, otherwise, step 5.7 is performed；
Step 5.6, k+1 is assigned to after k, judges whether k ＞ mM set up, if so, step 5.8 is then performed, otherwise, is returned
Return step 5.5；
K-th of standard feature parameter vector when step 5.7, the input vector X (i, b) of computation model are with the b times training
UkThe distance between (b) dkAfter (i, b), return to step 5.6；
Standard feature parameter vector set U when step 5.8, the input vector X (i, b) of Selection Model are with the b times training
(b) minimum value in the distance between each standard feature parameter vector in, is designated as dmin(i,b)；By minimum value dmin(i,b)
Corresponding typical road conditions block is designated as dcbest(i,b)；By typical road conditions block dcbestState value corresponding to (i, b) is designated as " 1 ",
State value corresponding to remaining mM-1 typical road conditions block is designated as " 0 ", so as to correspond to mode input when obtaining the b times training
The output vector a (i, b) of the competition layer of vectorial X (i, b)；
Step 5.9, using formula (1) obtain the b time training when corresponding to mode input vector X (i, b) model export to
Measure Y (i, b)：
Y (i, b)=U*·a(i,b) (1)
Step 5.10, find from the output vector Y (i, b) of model typical road conditions class state value be " 1 " typical road
Condition class, is designated as DCbest(i,b)；
Step 5.11, defined variable δi(b) i-th of standard feature parameter vector U when for the b times trainingi(b) corresponding to
Recognition factor；Judge DCbestI-th of standard feature parameter vector U when (i, b) is with the b times trainingi(b) the typical road conditions corresponding to
Whether class is identical, if identical, then it represents that vector quantization training pattern can correctly recognize i-th of standard feature ginseng during the b times training
Number vector Ui(b) the typical road conditions class corresponding to, and remember Ui(b) the recognition factor δ corresponding toi(b) after=1, step is performed
5.12；Otherwise, U is rememberedi(b) the recognition factor δ corresponding toi(b) after=0, step 5.13 is performed；
Step 5.12, formula (2) is utilized to obtain i-th of typical case road conditions block dc after being adjusted in the b time trainingi
J standard feature parameter u 'i,j(b)；
u′i,j(b)=ui,j(b)+φ(xj-ui,j(b)) (2)
In formula (2), φ is learning rate, and φ is on the occasion of learning rate characterizes ui,jRegulate the speed, i.e., model training speed
Degree, that is, model is trained optimization by speed of φ；
Step 5.13, formula (3) is utilized to obtain i-th of typical case road conditions block dc after being adjusted in the b time trainingi
J standard feature parameter u 'i,j(b)；
u′i,j(b)=ui,j(b)-φ(xj-ui,j(b)) (3)
Step 5.14, i+1 is assigned to after i, judges whether i ＞ mM set up, if so, then perform step 5.15；Otherwise,
Return to step 5.3；
Step 5.15, systematic error e (b)=δ (the b)/mM for calculating model after the b times training, wherein
The systematic error of model is that the ratio shared by correct typical road conditions block is recognized after model is once trained, with model training
Continuous progress, systematic error gradually reduces；
Step 5.16, initialization system error threshold are e*；As e (b)<e*When, connection matrix U (b) is exported, shows vector quantity
Change training pattern training to complete；As e (b) >=e*When, by u 'i,j(b) i-th of typical road conditions block dc when being trained as the b+1 timesi
J-th of standard feature parameter ui,j(b+1) step 5.17, is performed；
Step 5.17, b+1 is assigned to b, repeat step 5.2 to step 5.16.
In the present embodiment, δ=0.02, e is taken respectively*=0.01, initial vector quantization training pattern is trained,
The systematic error of model is met before requirement, and model constantly carries out the training optimization using φ as speed, and what it is until model is
Error of uniting, which is met, to be required, shows that model training is completed.The model training process of vector quantization training pattern is as shown in Figure 4.
Step 6, to random road conditions carry out road conditions identification；
Road conditions cognitive phase, the vector quantization training pattern completed using training carries out road conditions identification to random road conditions.
Random road conditions block in step 6.1, collection continuous time section, and R characteristic parameter of random road conditions block is carried out
Normalized, obtains the corresponding standard feature parameter vector of random road conditions block successively；
Step 6.2, successively using the corresponding standard feature parameter vector of random road conditions block as trained completion vector
The input vector for quantifying training pattern is identified, so as to obtain the recognition result of the random road conditions block in continuous time section.
In the present embodiment, note fixed time period is that note recognition cycle is T, takes T=120s, i.e., a length of T when continuously gathering
Random road conditions in=120s are fast, and the vector quantization training pattern for passing sequentially through training completion carries out road conditions identification to it.Using
The flow that vector quantization training pattern carries out random road conditions identification is as shown in Figure 5.
Claims (1)
1. a kind of fuel-engined vehicle road conditions recognition methods based on vector quantization training pattern, it is characterized in that carrying out as follows：
Step 1, the typical road conditions class of selection；
According to traffic and the difference of traveling region, the typical road conditions class of m kinds is chosen as identification sample, so as to constitute typical case
Road conditions class set DC={ DC1,DC2,...,DCr,...,DCm, DCrRepresent the typical road conditions class of r kinds, 1≤r≤m；
Step 2, piecemeal is carried out to every kind of typical road conditions class；
Road conditions class DC typical to r kindsrPiecemeal is carried out, M typical road conditions block is obtained and constitutes set DCr={ dcr,1,
dcr,2,...,dcr,p,...,dcr,M, dcr,pRepresent corresponding p-th of the typical road conditions block of the typical road conditions class of r kinds, 1≤p≤M；
So as to obtain mM typical road conditions block and constitute set { dc1,1,...,dc1,m,...,dcr,p,...,dcm,1,...,dcm,M, note
For { dc1,dc2,...,dci,...,dcmM, dciRepresent i-th of typical road conditions block, 1≤i≤mM；
Step 3, choose and standardized calculation relevant feature parameters characterize the feature of each typical road conditions block；
To i-th of typical road conditions block dciChoose R characteristic parameter and be normalized, obtain i-th of standard feature parameter
Vectorial Ui=[ui,1,ui,2,...,ui,j,...,ui,R]T, ui,jRepresent i-th of typical road conditions block dciCorresponding j-th of standard is special
Parameter is levied, so as to obtain mM standard feature parameter vector and constitute set U={ U1,U2,…,Ui,…,UmM, 1≤j≤R；
Step 4, structure vector quantization training pattern；
The vector quantization training pattern is defined to be made up of input layer, competition layer and output layer；
The input vector for defining the vector quantization training pattern is X=[x1,x2,...,xj,...,xR]T；
Define the vector quantization training pattern input layer and compete interlayer connection matrix be
The input vector for defining the competition layer of the vector quantization training pattern is d=[d1,d2,...,di,...,dmM]T, diTable
Show i-th of standard feature parameter vector UiThe distance between with the input vector X of the model；
The output vector for defining the competition layer of the vector quantization training pattern is a=[a1,a2,...,ai,...,amM]T, aiTable
Show i-th of typical road conditions block dciState value, only ai=1 represents i-th of typical road conditions block dciAs described vector quantization training
The typical road conditions block that the corresponding road conditions blocks of input vector X of model are identified；
The connection matrix defined between the competition layer and output layer of the vector quantization training pattern isWherein u*=[1,1 ..., 1]M×1；
The output vector for defining the vector quantization training pattern is Y=[y1,y2,...,yr,...,ym]T, yrRepresent r kind allusion quotations
Type road conditions class DCrState value, only yr=1 represents the typical road conditions class DC of r kindsrAs described vector quantization training pattern it is defeated
The typical road conditions class that the corresponding road conditions blocks of incoming vector X are identified；
Step 5, the input vector by the use of each standard feature parameter vector as vector quantization training pattern are to the vector quantity
Change training pattern to be trained, until it meets systematic error requirement, so as to obtain the vector quantization training mould available for identification
Type；
Step 5.1, defined variable b are frequency of training；Initialize b=1；
Step 5.2, initialization i=1；
Step 5.3, by the b times training when i-th of standard feature parameter vector Ui(b) as the input vector X (i, b) of model；
Step 5.4, defined variable k, and 1≤k≤mM；Initialize k=1；
Step 5.5, judge whether k=i sets up, if so, step 5.6 is then performed, otherwise, step 5.7 is performed；
Step 5.6, k+1 is assigned to after k, judges whether k ＞ mM set up, if so, step 5.8 is then performed, otherwise, step is returned to
Rapid 5.5；
K-th of standard when step 5.7, the input vector X (i, b) for calculating the vector quantization training pattern are with the b times training
Characteristic parameter vector UkThe distance between (b) dkAfter (i, b), return to step 5.6；
Standard feature when step 5.8, the input vector X (i, b) for choosing the vector quantization training pattern are with the b times training is joined
The minimum value in the distance between each standard feature parameter vector in number vector set U (b), is designated as dmin(i,b)；Will be described
Minimum value dminTypical road conditions block corresponding to (i, b) is designated as dcbest(i,b)；By the typical road conditions block dcbest(i, b) institute is right
The state value answered is designated as " 1 ", and the state value corresponding to remaining mM-1 typical road conditions block is designated as " 0 ", so as to obtain the b times instruction
Correspond to the output vector a (i, b) of mode input vector X (i, b) competition layer when practicing；
Step 5.9, utilize formula (1) obtain the b time training when corresponding to mode input vector X (i, b) model output vector Y
(i,b)：
Y (i, b)=U*·a(i,b) (1)
Step 5.10, find from the output vector Y (i, b) of the model typical road conditions class state value be " 1 " typical road
Condition class, is designated as DCbest(i,b)；
Step 5.11, defined variable δi(b) i-th of standard feature parameter vector U when for the b times trainingi(b) identification corresponding to
The factor；Judge DCbestI-th of standard feature parameter vector U when (i, b) is with the b times trainingi(b) the typical road conditions class corresponding to is
It is no identical, if identical, then it represents that the vector quantization training pattern can correctly recognize i-th of standard feature ginseng during the b times training
Number vector Ui(b) the typical road conditions class corresponding to, and remember Ui(b) the recognition factor δ corresponding toi(b) after=1, step is performed
5.12；Otherwise, U is rememberedi(b) the recognition factor δ corresponding toi(b) after=0, step 5.13 is performed；
Step 5.12, formula (2) is utilized to obtain i-th of typical case road conditions block dc after being adjusted in the b time trainingiJ-th mark
Quasi- characteristic parameter u 'i,j(b)；
u′i,j(b)=ui,j(b)+φ(xj-ui,j(b)) (2)
In formula (2), φ is learning rate, and φ be on the occasion of；
Step 5.13, formula (3) is utilized to obtain i-th of typical case road conditions block dc after being adjusted in the b time trainingiJ-th mark
Quasi- characteristic parameter u 'i,j(b)；
u′i,j(b)=ui,j(b)-φ(xj-ui,j(b)) (3)
In formula (3), φ is learning rate, and φ be on the occasion of；
Step 5.14, i+1 is assigned to after i, judges whether i ＞ mM set up, if so, then perform step 5.15；Otherwise, return
Step 5.3；
Step 5.15, systematic error e (b)=δ (the b)/mM for calculating model after the b times training, wherein
Step 5.16, initialization system error threshold are e*；As e (b)<e*When, export connection matrix U (b), the vector quantization instruction
Practice model training to complete；As e (b) >=e*When, by u 'i,j(b) i-th of typical road conditions block dc when being trained as the b+1 timesi
J standard feature parameter ui,j(b+1) after, step 5.17 is performed；
Step 5.17, b+1 is assigned to b, repeat step 5.2 to step 5.16；
Step 6, to random road conditions carry out road conditions identification；
Random road conditions block in step 6.1, collection continuous time section, and normalizing is carried out to R characteristic parameter of random road conditions block
Change is handled, and the corresponding standard feature parameter vector of random road conditions block is obtained successively；
Step 6.2, the corresponding standard feature parameter vector of the random road conditions block is used as to the vector that the training is completed successively
The input vector for quantifying training pattern is identified, so as to obtain the identification knot of the random road conditions block in the continuous time section
Really.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710026267.1A CN106781502B (en) | 2017-01-13 | 2017-01-13 | A kind of fuel-engined vehicle road conditions recognition methods based on vector quantization training pattern |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710026267.1A CN106781502B (en) | 2017-01-13 | 2017-01-13 | A kind of fuel-engined vehicle road conditions recognition methods based on vector quantization training pattern |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106781502A CN106781502A (en) | 2017-05-31 |
| CN106781502B true CN106781502B (en) | 2017-09-29 |
Family
ID=58945577
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710026267.1A Active CN106781502B (en) | 2017-01-13 | 2017-01-13 | A kind of fuel-engined vehicle road conditions recognition methods based on vector quantization training pattern |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106781502B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109714335A (en) * | 2018-12-26 | 2019-05-03 | 北京天融信网络安全技术有限公司 | A kind of information detecting method and information detector |
| CN112529104A (en) * | 2020-12-23 | 2021-03-19 | 东软睿驰汽车技术（沈阳）有限公司 | Vehicle fault prediction model generation method, fault prediction method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090295917A1 (en) * | 2008-04-24 | 2009-12-03 | Gm Global Technology Operations, Inc. | Pixel-based texture-less clear path detection |
| CN102915453A (en) * | 2012-08-30 | 2013-02-06 | 华南理工大学 | Real-time feedback and update vehicle detection method |
| CN105644548A (en) * | 2015-12-28 | 2016-06-08 | 中国科学院深圳先进技术研究院 | Energy control method and device for hybrid electric vehicle |
-
2017
- 2017-01-13 CN CN201710026267.1A patent/CN106781502B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090295917A1 (en) * | 2008-04-24 | 2009-12-03 | Gm Global Technology Operations, Inc. | Pixel-based texture-less clear path detection |
| CN102915453A (en) * | 2012-08-30 | 2013-02-06 | 华南理工大学 | Real-time feedback and update vehicle detection method |
| CN105644548A (en) * | 2015-12-28 | 2016-06-08 | 中国科学院深圳先进技术研究院 | Energy control method and device for hybrid electric vehicle |
Non-Patent Citations (1)
| Title |
|---|
| 基于数字图像和支持向量机的交通路况检测;翟玉婷等;《计算机工程与设计》;20131231;第34卷(第12期);第4273-4277页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106781502A (en) | 2017-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103353923B (en) | Adaptive space interpolation method and system thereof based on space characteristics analysis | |
| CN106781502B (en) | A kind of fuel-engined vehicle road conditions recognition methods based on vector quantization training pattern | |
| CN112101684B (en) | Plug-in hybrid electric vehicle real-time energy management method and system | |
| CN109871622A (en) | A kind of low-voltage platform area line loss calculation method and system based on deep learning | |
| CN104318482A (en) | Comprehensive assessment system and method of smart distribution network | |
| CN107085763A (en) | A kind of driving motor for electric automobile system performance evaluation method | |
| CN107784380A (en) | The optimization method and optimization system of a kind of inspection shortest path | |
| CN107705556A (en) | A kind of traffic flow forecasting method combined based on SVMs and BP neural network | |
| CN106354962A (en) | Lithium-iron-phosphate-battery fractional-order equivalent circuit model establishing method based on frequency demultiplication representation | |
| CN111951430B (en) | Vehicle drivability evaluation method and system | |
| CN106408939A (en) | Traffic flow sequence classification method based on density peak value clustering | |
| CN108875812B (en) | Driver behavior classification method based on branch convolutional neural network | |
| CN105938578A (en) | Large-scale photovoltaic power station equivalent modeling method based on clustering analysis | |
| CN102567782A (en) | Neural-network-based automobile engine torque estimation method | |
| CN105046714A (en) | Unsupervised image segmentation method based on super pixels and target discovering mechanism | |
| CN110399880A (en) | Recognition methods, device and the equipment of a kind of characters on license plate and license plate classification | |
| CN106503846A (en) | Route calculation algorithm patrolled and examined by a kind of power equipment | |
| CN105301508A (en) | Prediction method for electric automobile endurance mileage through redial basis function neural network | |
| CN112633658A (en) | Low-voltage distribution area topological relation identification method based on CNN-LSTM | |
| CN106401757A (en) | Cylinder shutting down mode implementing method and system of engine, and vehicle | |
| CN113715629B (en) | Residual driving range prediction method based on improved symbolic regression and XGBoost algorithm | |
| CN113297685B (en) | Vehicle operation condition mode identification method | |
| CN104462797A (en) | Increment integration algorithm used for procedure parameter online testing | |
| CN109147322A (en) | Multi-source data method for self-adaption amalgamation in a kind of processing of urban transportation big data | |
| CN116596129A (en) | Electric vehicle charging station short-term load prediction model construction method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |