CN103921743B - Automobile running working condition judgement system and method for discrimination thereof - Google Patents

Abstract

Automobile running working condition judgement system and method for discrimination thereof belong to intelligent vehicle running technical field of environmental perception, the method selects the speed of a motor vehicle of automobile and pedal aperture two parameters to process, extract the characteristic parameter in the speed signal and pedal signal characterizing operating mode, first relevant function method is adopted to carry out a yojan to these characteristic parameters, front 15 parameters choosing correlativity maximum carry out correlation analysis with operating mode respectively, select 10 characteristic parameters as the standard dividing operating mode from high to low according to correlativity, then core pivot element analysis is adopted to carry out secondary yojan to these 10 characteristic parameters, it is 7 by characteristic ginseng value yojan, the driving cycle of semi-supervised kernel Fuzzy C-Means Clustering analysis to automobile of last based target function is classified.Distinguishing speed of the present invention is fast and accurate, utilizes this differentiation result both can reduce fuel consumption in vehicle traveling process and exhaust emissions amount, significant to the research of the aspects such as the exploitation of all kinds of new model and car load dynamic property coupling again.

Description

Automobile running working condition judgement system and method for discrimination thereof

Technical field

The invention belongs to intelligent vehicle running technical field of environmental perception, particularly a kind of automobile running working condition judgement system and method for discrimination thereof.

Background technology

Automobile running working condition is the speed-time curve describing vehicle traveling.Automobile running working condition for determining emission from vehicles amount, fuel consumption, the exploitation, car load dynamic property coupling etc. of all kinds of new model technology provide important theoretical foundation.At present, worldwide driving cycle comprises United States of america operating mode, European driving cycle and Japanese driving cycle.China's auto emission adopts the effluent standard in Europe, but because different urban highway traffic situations is different, therefore the driving cycle of automobile also there are differences, adopt unified standard can accurately not reflect the automobile running working condition in some cities, therefore some research institutes of China and colleges and universities are economically developed for some, environmental demands is high such as ground such as Beijing, Shanghai, Tianjin, city have carried out the research of automobile running working condition, construct the automobile running working condition of the reality for concrete urban highway traffic situation.

The method of conventional structure automobile running working condition has two kinds, and a kind of is utilize the intelligence computation method that principal component analysis (PCA), cluster analysis, Markov Model about Forecasting method and fuzzy neural network, genetic algorithm etc. are complicated to build operating mode; Another kind adopts multivariate statistics method to build operating mode, as distance discrimination method, Fei Xier (Fisher) method of discrimination, Bayes (Bayes) method of discrimination etc.The first intelligence computation analysis method ubiquity computation process is complicated, calculated amount is large, require problem higher, may there is delay for the judgement of operating mode in actual vehicle traveling process to the operating rate of equipment, can not meet the rapidity that real-time working condition differentiates.The second multivariate statistics method can not calculate the probability of a certain operating mode appearance on the whole for work condition judging, and there is the situation of misjudgement, Shortcomings in the accuracy that automobile running working condition differentiates.

Core pivot element analysis method is by means of kernel function, the training sample data of the input space are transformed into feature space through Nonlinear Mapping, then project to selected multiple principal eigenvector directions, finally can obtain multiple incoherent pivot, for feature extraction and pattern-recognition provide an effective way.Core pivot element analysis can extract the information of index to greatest extent, and have feature extraction speed fast, characteristic information retains sufficient advantage.Fuzzy cluster analysis is as one of major technique without supervision machine learning, by the method for fuzzy theory to important data analysis and modeling, establish the uncertainty description of sample generic, can be reflected reality the world more objectively, it is applied in the fields such as large-scale data analysis, data mining, vector quantization, Iamge Segmentation, pattern-recognition effectively, has important theory and practice using value.In numerous fuzzy clustering algorithm, FCM (FCM) algorithm application the most extensively and more successful, fuzzy C-clustering determines the degree of membership of all cluster centres of each sample by optimization object function, thus determine the generic of sample, the automatic classification to sampled data can be realized.

Summary of the invention

In order to solve the extraction of characteristic parameter in automobile running working condition differentiation, the problem of characteristic parameter yojan; And the larger difference that the fixing operating mode adopted due to automobile control method design at present and actual condition exist causes the control method designed that vehicle can not be made to be issued to the problem of optimal fuel economy and emission performance at actual condition, the invention provides a kind of automobile running working condition judgement system and method for discrimination thereof, the method accurately can differentiate the current affiliated driving cycle of vehicle according to the data sample of characteristic parameter.

The technical scheme that technical solution problem of the present invention is taked is as follows:

Automobile running working condition judgement system, comprises embedded system ARM, clock chip, watchdog circuit, program store, data memory, touch-screen, CAN interface and LCDs, the first signal conditioning circuit, secondary signal modulate circuit and signal acquisition circuit; Signal acquisition circuit Negotiation speed sensor, pedal position sensor obtain the speed of a motor vehicle relevant to work condition judging and pedal information data, and data are transformed to digital signal needed for CAN interface by the first signal conditioning circuit; The simulate data that secondary signal modulate circuit can identify for the digital signal that CAN interface transmits being converted to car load; Data communication between CAN Interface realization car load and embedded system ARM, the data upload that the first signal conditioning circuit received transmits is carried out work condition judging to embedded system ARM, and the data passed back by embedded system ARM send car load to by secondary signal modulate circuit; Embedded system ARM is by carrying out characteristic parameter extraction to the data received, then adopt correlation analysis and core pivot element analysis to carry out secondary yojan, guide cluster process to realize the differentiation of automobile running working condition by the training sample of 10% known class adding overall number of training in semi-supervised kernel Fuzzy C-Means Clustering Algorithm; Program store and data memory be embedded system ARM extend out chip, be respectively used to storage program and data; LCDs is for showing the current speed of a motor vehicle, pedal aperture, and automobile running working condition information; The order of touch-screen by various input and the operation of requirement control relevant device; Watchdog circuit is used for the monitoring of embedded system ARM circuit and reset; Clock chip is used for for embedded system ARM, program store, data memory provide the clock signal of needs.

Automobile running working condition method of discrimination, comprises the steps:

1) extraction of the speed of a motor vehicle, pedal characteristic parameter

The speed of a motor vehicle characteristic parameter extraction module in embedded system ARM is utilized to calculate the characteristic parameter relevant to operating mode with pedal characteristic parameter extraction module;

2) correlating module is utilized based on correlation analysis method to step 1) characteristic parameter of sign operating mode that obtains carries out a yojan

First characteristic parameter is represented with sequence a (n) and b (n) between two respectively, and compares the similarity degree r of these two sequences in the scope of sequential n=i ~ j according to formula (1):

$r=\frac{{\mathrm{\Σ}}_{n=i}^{j}a\left(n\right)b^{*}\left(n\right)}{{\left\{{\mathrm{\Σ}}_{n=i}^j{\left|a\left(n\right)\right|}^2{\mathrm{\Σ}}_{n=i}^j{\left|b\left(n\right)\right|}^2\right\}}^{\frac{1}{2}}}---\left(1\right)$

In formula, a (n) and b (n) is respectively n element in sequence a, b, and * is for get conjugation to b (n);

Operation result according to formula (1) sorts from big to small according to the eigenwert irrelevance of parameter, then 15 characteristic parameters come above are carried out correlation analysis with operating mode respectively, according to correlativity from high to low, finally have selected ten characteristic parameters as the standard dividing operating mode;

3) core pivot element analysis module is utilized based on core pivot element analysis method to step 2) ten characteristic parameters obtaining carry out secondary yojan

Core pivot element analysis module by kernel function the original data a of the input space _i∈ Q ^q(i=1 ..., N) and be mapped to a new high-dimensional feature space, then linear pivot analysis is carried out to the data after mapping, extract nonlinear characteristic, obtain and principal eigenvector that classify error rate little high with driving cycle sensivity;

4) utilize core Fuzzy C-Means Clustering module to set up core Fuzzy C-Means Clustering model based on core Fuzzy C-Means Clustering Algorithm, carry out the differentiation of automobile running working condition according to this model

Core Fuzzy C-Means Clustering module is by through step 3) sample set after yojan is divided into several fuzzy classes, by method of Lagrange multipliers structure objective function and Fuzzy C-Means Clustering model, and obtains cluster centre, subordinated-degree matrix and parameter; Cluster process is guided by the training sample of 10% known class adding overall number of training in Fuzzy C-Means Clustering Algorithm, simplify the iterative step of computation process, and improve the non-linear mapping capability of Fuzzy C-Means Clustering model by means of Wavelet Kernel Function.

Beneficial effect of the present invention is as follows:

(1) adopt the speed of a motor vehicle, acceleration pedal and brake pedal information to carry out the differentiation of automobile running working condition, taken into full account that driver is to the operation intention of current motoring condition and automobile future travel state, improve the accuracy that driving cycle differentiates;

(2) differentiate relevant characteristic parameter based on correlation analysis extraction to driving cycle, carry out the first time yojan inputting data;

(3) effectively characteristic parameter is extracted based on core pivot element analysis method second time, reduce the input data redundancy of work condition judging, obtain that sensivity is high and classification error rate is little and can reflect the principal eigenvector of road Real-road Driving Cycle exactly, namely achieve the second time yojan of characteristic parameter;

(4) core fuzzy C-clustering is by utilizing the known iterative process being subordinate to mean distance guiding target function between the data sample of operating mode and training sample, effectively improve the local extremum problem that random initializtion Matrix dividing may cause, improve accuracy and the rapidity of cluster result;

(5) hardware device based on CAN forms simple, and work condition judging feasible is practical;

(6) LCDs is added, for driver provides the auxiliary assistant driven.

Accompanying drawing explanation

Fig. 1 is the structured flowchart of automobile running working condition judgement system of the present invention.

Fig. 2 is the functional block diagram of automobile running working condition method of discrimination of the present invention.

Detailed description of the invention

Below in conjunction with accompanying drawing and specific implementation process, the present invention is described in further detail.

As shown in Figure 1, automobile running working condition judgement system of the present invention comprises embedded system ARM1, clock chip 2, watchdog circuit 3, program store 4, data memory 5, touch-screen 6, CAN interface 7 and LCDs 8, first signal conditioning circuit 9, secondary signal modulate circuit 10 and signal acquisition circuit 11; Signal acquisition circuit 11 Negotiation speed sensor, pedal position sensor obtain the speed of a motor vehicle relevant to work condition judging and pedal information data, and data are transformed to digital signal needed for CAN interface 7 by the first signal conditioning circuit 9; Secondary signal modulate circuit 10 is converted to for the digital signal transmitted by CAN interface 7 simulate data that car load 12 can identify; CAN interface 7 realizes the data communication between car load 12 and embedded system ARM1, the data upload that the first signal conditioning circuit 9 received transmits is carried out work condition judging to embedded system ARM1, and the data passed back by embedded system ARM1 send car load 12 to by secondary signal modulate circuit 10; Embedded system ARM1 is by carrying out characteristic parameter extraction to the data received, then correlation analysis and core pivot element analysis is adopted to carry out secondary yojan, by adding the training sample of 10% known class of overall number of training in semi-supervised kernel Fuzzy C-Means Clustering Algorithm, (10% is random selecting at least 1 training sample in each class from the training sample of known class, final formation quantity 10% known class training sample) guide cluster process, realize the differentiation of automobile running working condition; What program store 4 and data memory 5 were embedded system ARM1 extends out chip, is respectively used to storage program and data; LCDs 8 is for showing the current speed of a motor vehicle, pedal aperture, and automobile running working condition information; The order of touch-screen 6 by various input and the operation of requirement control relevant device; Watchdog circuit 3 is for the monitoring of embedded system ARM1 circuit and reset; Clock chip 2 is for providing the clock signal of needs for embedded system ARM1, program store 4, data memory 5.

Embedded system ARM1 comprises the speed of a motor vehicle characteristic parameter extraction module 13, pedal characteristic parameter extraction module 14, correlating module 15, core pivot element analysis module 16, core Fuzzy C-Means Clustering module 17 totally five modules that are realized by software programming, and embedded system ARM1 selects chip S3C2410ARM to realize.

As described in Figure 2, the implementation procedure of automobile running working condition method of discrimination of the present invention is as follows:

1) extraction of the speed of a motor vehicle, pedal characteristic parameter

Speed of a motor vehicle characteristic parameter extraction module 13 and pedal characteristic parameter extraction module 14 calculate the characteristic parameter relevant to operating mode and comprise: average operating time, average pick-up time, the mean deceleration time, the average at the uniform velocity time, average time of idle running, average range ability, maximum speed, average velociity, velocity standard deviation, peak acceleration, energy, entropy, maximum deceleration, average acceleration, at the uniform velocity time ratio, time of idle running ratio, accelerating sections average acceleration, braking section mean deceleration, acceleration standard deviation, pedal aperture rate of change, the average aperture of pedal, pedal aperture is the time equivalence of zero.

2) correlating module 15 is utilized based on correlation analysis method to step 1) characteristic parameter of sign operating mode that obtains carries out a yojan

First by 1) in the characteristic parameter of trying to achieve represent with sequence a (n) and b (n) respectively between two, compare the similarity degree r of these two sequences in the scope of sequential n=i ~ j according to formula (1):

$r=\frac{{\mathrm{\Σ}}_{n=i}^{j}a\left(n\right)b^{*}\left(n\right)}{{\left\{{\mathrm{\Σ}}_{n=i}^j{\left|a\left(n\right)\right|}^2{\mathrm{\Σ}}_{n=i}^j{\left|b\left(n\right)\right|}^2\right\}}^{\frac{1}{2}}}---\left(1\right)$

In formula, a (n) and b (n) is respectively n element in sequence a, b, and " * " is for get conjugation to b (n).

Operation result according to above formula sorts from big to small according to the irrelevance of characteristic ginseng value, then 15 parameters come above are carried out correlation analysis with operating mode respectively, finally have selected 10 characteristic parameters from high to low as the standard dividing operating mode, respectively: maximum speed, average velociity, maximum deceleration, average acceleration, at the uniform velocity time ratio, time of idle running ratio, accelerating sections average acceleration, braking section mean deceleration, the average aperture of pedal, pedal aperture are the time of zero according to correlativity.

3) core pivot element analysis module 16 is utilized based on core pivot element analysis method to step 2) ten characteristic parameters obtaining carry out secondary yojan

Core pivot element analysis by kernel function the original data a of the input space _i(i=1 ..., N) and be mapped to a new high-dimensional feature space, then linear pivot analysis is carried out to the data after mapping, thus effectively extract nonlinear characteristic, obtain and make original data have the nonlinear principal component of better separability; The yojan of realization character parameter, reduces data redundancy, obtains and the high and principal eigenvector that error rate of classifying is little of driving cycle sensivity.

Data α (a after mapping _i) covariance matrix A be:

$A=\frac{1}{N}{\mathrm{\Σ}}_{i=1}^N\mathrm{\α}\left(a_i\right)\mathrm{\α}{\left(a_i\right)}^T---\left(2\right)$

In formula, α is mapping function.N is the number of original data, α (a _i) ^trepresent α (a _i) transposition.

If the eigenwert of covariance matrix A is λ, corresponding proper vector is ω, then:

λω＝Aω(3)

For all i=1 ..., N, by each sample α (a after mapping _i) do inner product with formula (3) both sides respectively:

λ<α(a _i)．ω>＝<α(a _i)．Aω>

Therefore, when λ ≠ 0, there is factor beta _i(i=1 ..., N), by characteristic vector ω α (a _i) linear combination represent:

$\mathrm{\&omega;}={\mathrm{\&Sigma;}}_{i=1}^N{\mathrm{\&beta;}}_i\mathrm{\&alpha;}\left(a_i\right)---\left(4\right)$

Formula (2), (4) are substituted in formula (3):

$\mathrm{\&lambda;}{\mathrm{\&Sigma;}}_{i=1}^N{\mathrm{\&beta;}}_i<\mathrm{\&alpha;}\left(a_i\right)\&CenterDot;\mathrm{\&alpha;}\left(a_i\right)>=\frac{1}{N}{\mathrm{\&Sigma;}}_{j=1}^N{\mathrm{\&beta;}}_i<\mathrm{\&alpha;}\left(a_k\right)\&CenterDot;{\mathrm{\&Sigma;}}_{j=1}^{N}a\left(a_j\right)\mathrm{\&alpha;}{\left(a_j\right)}^T\mathrm{\&alpha;}\left(a_i\right)>---\left(5\right)$

The element M of definition symmetry square matrix M _ijfor kernel function:

M _ij=< α (a _i) α (a _j) in > (6) formula, j=1 ..., N.

Because M is symmetric matrix, formula (5) can be reduced to

Nλβ＝Mβ(7)

The eigenvalue λ can tried to achieve and meet the demands is solved to formula (7) _kwith characteristic vector β _k(k=1 ..., N).The method retaining a front p proper vector is taked to carry out yojan to the characteristic parameter number that system inputs.Eigenwert after matrix M diagonalization is designated as λ _k1>=λ _k2>=...>=λ _kNif p is the number of the minimum non-zero eigenwert of M, formula (6) and (7) are utilized to try to achieve:

${\mathrm{\&beta;}}_k^T{\mathrm{\&beta;}}_k=1/{\mathrm{\&lambda;}}_k---\left(8\right)$

In formula, k=1,2 ..., p.

Data after mapping are at proper vector ω _k(k=1 ..., the projection p), is the original data a tried to achieve by Nonlinear Mapping α _inonlinear principal component ρ _k:

${\mathrm{\&rho;}}_k=<{\mathrm{\&omega;}}_k\&CenterDot;\mathrm{\&alpha;}\left(a_i\right)>={\mathrm{\&Sigma;}}_{i=1}^N{\mathrm{\&beta;}}_i^k<\mathrm{\&alpha;}\left(a_i\right)\&CenterDot;\mathrm{\&alpha;}\left(a\right)>={\mathrm{\&Sigma;}}_{i=1}^N{\mathrm{\&beta;}}_i^{k}M(a_i,a)---\left(9\right)$

In formula, for a kth eigenwert of matrix K is for characteristic vector β _ki-th coefficient.

The present invention selects Gaussian radial basis function as kernel function M _ij:

$M_{\mathrm{ij}}=\exp [-{\left|\right|a_i-a_j\left|\right|}^2/\left({2\mathrm{\&sigma;}}^2\right)]---\left(10\right)$

In formula, a _iand a _jfor the i-th, j number in original data sequence, σ is standard deviation.

If mapping (enum) data is Non-zero Mean, then need to carry out equalization process to kernel function:

$\stackrel{\&OverBar;}{M_{\mathrm{ij}}}={(M-1_{N}M-{M1}_N+1_N{M1}_N)}_{\mathrm{ij}}=<\mathrm{\&alpha;}\left(a_i\right)\&CenterDot;\left(a_j\right)>---\left(11\right)$

In formula, 1 _nfor N × N rank constant square formation that element is 1/N; α (a _i) and α (a _j) be the mapping (enum) data after equalization.

4) utilize core Fuzzy C-Means Clustering module 17 to set up core Fuzzy C-Means Clustering model based on core Fuzzy C-Means Clustering Algorithm, carry out the differentiation of automobile running working condition according to this model:

Sample set after the yojan of core pivot is divided into several fuzzy classes, constructs objective function by method of Lagrange multipliers and obtain cluster centre, subordinated-degree matrix, parameter; Cluster process is guided by the training sample adding a small amount of known generic in Fuzzy C-Means Clustering Algorithm, reduce the iterative step of computation process, algorithm is simplified, and the initialization Matrix dividing avoiding random selecting can make cluster result be absorbed in the problem of local extremum.And by means of the non-linear mapping capability of Wavelet Kernel Function, data-mapping will be inputted in a high-dimensional feature space, and make input data have better separability, improve cluster speed and clustering precision.Specific implementation process is as follows:

First by the sample set Q={q after the yojan of core pivot _k, k=1,2 ..., m, q _k∈ R ^pbe divided into c fuzzy class.Note fuzzy clustering center matrix C={c _i∈ R ^{p × a}, i=1,2 ..., a, c _iit is the cluster centre of the i-th class; Fuzzy membership matrix V={ v _ik∈ R ^{a × m}, i=1,2 ..., a; K=1,2 ..., m, v _ikrepresent q _kbelong to the degree of membership of the i-th class, n>1 is FUZZY WEIGHTED index.

Design cluster objective function is:

$J(C,V)={\mathrm{\&Sigma;}}_{k=1}^m{\mathrm{\&Sigma;}}_{i=1}^c{v}_{\mathrm{ik}}^n{\left|\right|q_k-c_i\left|\right|}^2={\mathrm{\&Sigma;}}_{k=1}^m{\mathrm{\&Sigma;}}_{i=1}^c{v}_{\mathrm{ik}}^n{d}_{\mathrm{ik}}^2---\left(12\right)$

In formula, d _ikfor a kth data q _kwith the i-th class center c _ibetween Euclidean distance tolerance, for kth group data sample belongs to the coefficient of i classification, superscript n is FUZZY WEIGHTED index, n ∈ (1, ∞]

Meet following constraint condition:

${\mathrm{\&Sigma;}}_{k=1}^m{v}_{\mathrm{ik}}^n=\mathrm{1,0}\&le;v_{\mathrm{ik}}^n\&le;1$

In formula, q _k=[q _k1, q _k2..., q _kp] ^tfor sampled data;

As follows by method of Lagrange multipliers structure objective function:

$J_L(C,V,\mathrm{\&lambda;})={\mathrm{\&Sigma;}}_{k=1}^m{\mathrm{\&Sigma;}}_{i=1}^a{v}_{\mathrm{ik}}^n{d}_{\mathrm{ik}}^2+{\mathrm{\&Sigma;}}_{k=1}^m{\mathrm{\&lambda;}}_i({\mathrm{\&Sigma;}}_{i=1}^a{v}_{\mathrm{ik}}-1)---\left(13\right)$

By to J _lget minimum value and can obtain cluster centre C, subordinated-degree matrix V and parameter lambda.

Cluster result can be made to be absorbed in the problem of local extremum for avoiding the initialization Matrix dividing of random selecting, guide cluster process by the training sample of 10% known class adding overall number of training in Fuzzy C-Means Clustering Algorithm, thus improve cluster speed and clustering precision.

If with the data set q of known class label _lab, be designated as , wherein, it is a jth training sample of the i-th class; m _iit is the number of the i-th class training sample.Redefine objective function:

$J_K(C,V,\mathrm{\&lambda;},\mathrm{\&beta;})=\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{a}{\mathrm{\&Sigma;}}}{v}_{\mathrm{ik}}^n{d}_{\mathrm{ik}}^2+\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i(\underset{i=1}{\overset{a}{\mathrm{\&Sigma;}}}{v}_{\mathrm{ik}}-1)+\mathrm{\&beta;}\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{a}{\mathrm{\&Sigma;}}}{v}_{\mathrm{ik}}^n{d}_{\mathrm{ij}}^2---\left(14\right)$

In formula, β is the coefficient of weight to supervision item, d _ijfor supervision item mean distance

$d_{\mathrm{ij}}={\mathrm{\&Sigma;}}_{j=1}^{m_i}\left|\right|q_i-q_{i,j}^{*}\left|\right|/m_i$

Introduce Nonlinear Mapping Ψ: q _k→ Ψ (q _k), the sample distance in feature space is then defined as

$d_{\mathrm{ik}}^2={\left|\right|\mathrm{\&Psi;}\left(q_k\right)-\mathrm{\&Psi;}\left(c_i\right)\left|\right|}^2=M(q_k,q_k)+M(c_i,c_i)-2M(q_k,c_i)---\left(15\right)$

Wherein, M is kernel function, meets Mercer condition, because wavelet function has multiresolution analysis, can approach arbitrary function with higher precision.Therefore select Mexicanhat wavelet function to improve the non-linear mapping capability of Fuzzy c-means cluster.Renewal objective function is

$\begin{array}{c}{J}_{\mathrm{\&Psi;}}(C,V,\mathrm{\&lambda;},\mathrm{\&beta;})=-2\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{a}{\mathrm{\&Sigma;}}}{v}_{\mathrm{ik}}^n[1+M(q_k,c_i)]+\\ \underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}{q}_i(\underset{i=1}{\overset{a}{\mathrm{\&Sigma;}}}{q}_{\mathrm{ik}}-1)-2\mathrm{\&beta;}\underset{k=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{a}{\mathrm{\&Sigma;}}}{v}_{\mathrm{ik}}^n\left[\underset{j=1}{\overset{m_i}{\mathrm{\&Sigma;}}}\frac{1+M(q_k,q_j)}{m_i}\right]\end{array}---\left(16\right)$

Respectively to J _Ψabout c _i, v _ikask local derviation, obtain the cluster centre of the t time iteration and subordinated-degree matrix computing formula as follows:

$c_i^{\left(t\right)}=\frac{{\mathrm{\&Sigma;}}_{k=1}^m{\left(v_{\mathrm{ik}}^{(t-1)}\right)}^n{M}^{-1}(q_k,c_i)q_k}{{\mathrm{\&Sigma;}}_{k=1}^m{\left(v_{\mathrm{ik}}^{(t-1)}\right)}^n{M}^{-1}(q_k,c_i)}---\left(17\right)$

$v_{\mathrm{ik}}^{\left(t\right)}=\frac{{(1+M(q_k,c_i)+\mathrm{\&beta;}({\mathrm{\&Sigma;}}_{j=1}^{m_i}1+M(q_k,q_j))/m_i)}^{-1/n-1}}{{\mathrm{\&Sigma;}}_{i=1}^a{(1+M(q_k,c_i)+\mathrm{\&beta;}({\mathrm{\&Sigma;}}_{j=1}^{m_i}1+M(q_k,q_j))/m_i)}^{-1/n-1}}---\left(18\right)$

The driving cycle based on the yojan of core pivot and semi-supervised kernel Fuzzy Cluster Fusion that the present invention builds differentiates that implementation process is divided into two parts: Part I is the training of off-line driving cycle discrimination model, and Part II is that online real-time working condition differentiates.

1) training of off-line driving cycle discrimination model

Implementation step is as follows:

Step1 chooses the speed of a motor vehicle, acceleration pedal aperture and brake pedal aperture sampled data, calculates q=10 the eigenwert characterizing operating mode, and carries out normalisation to sampled data.

Step2 utilizes formula (10) to calculate nuclear matrix M, is tried to achieve the nuclear matrix of equalization by formula (11) .

Step3 utilizes formula (7) right carry out characteristic vector decomposition, standardized feature vector beta _k, make β _kmeet formula (8) condition.

Step4 utilizes formula (9) to extract core pivot.

Step5 calculates cumulative proportion in ANOVA according to the following formula

$R_C=\underset{k=1}{\overset{i}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_k/\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_k(i=\mathrm{1,2},...,p)$

Step6 selects cumulative proportion in ANOVA R _cthe proper vector of > 90% as core pivot characteristic amount, thus obtains the characteristic quantity math modeling based on the yojan of core pivot.

Step7 determines the number a of class, the power exponent n>1 of degree of membership and initial subordinated-degree matrix V ₀=(v _ik0), initial subordinated-degree matrix V ₀for [0,1] upper uniform random number is determined.T=1 is made to represent first step iteration.

Step8 will substitution formula (17) tries to achieve the cluster centre of t step

Step9 will substitution formula (18) revises subordinated-degree matrix , then according to formula (12) calculating target function J _Ψ(t).

Step10 is to given objective function termination tolerance ε _j> 0, works as max{|v _ik(t)-v _ik(t-1) | } < ε _jtime, stop iteration, try to achieve and meet objective function J _Ψt subordinated-degree matrix V that () is minimum and cluster centre C; Otherwise t=t+1, then goes to step Step8.

After the iteration of above step, the math modeling of semi-supervised kernel Fuzzy C-Means Clustering Algorithm can be obtained.Software programming is adopted to be present in the program store 4 of embedded system ARM1 on this model.

2) online real-time working condition differentiates

According to the speed of a motor vehicle and the pedal aperture data of new collection, calculated characteristics parameter, carries out yojan by the core pivot element analysis model established, and the semi-supervised kernel Fuzzy C-Means Clustering model then by building up carries out work condition judging, works as v _jk=max{v _ik, i=1,2 ..., j ..., during a, can judge that current working belongs to jth class operating mode.Utilize LCDs 8 to show the speed of a motor vehicle, pedal aperture, work condition judging result simultaneously.

To sum up, automobile running working condition method of discrimination of the present invention selects the speed of a motor vehicle of automobile and pedal aperture two parameters to process, extract the multiple characteristic parameters in the speed signal and pedal signal characterizing operating mode, first relevant function method is adopted to carry out a yojan to the characteristic parameter characterizing operating mode, front 15 parameters choosing correlativity maximum carry out correlation analysis with operating mode respectively, finally select 10 characteristic parameters as the standard dividing operating mode from high to low according to correlativity, then core pivot element analysis is adopted to carry out secondary yojan to these 10 characteristic parameters, it is 7 by characteristic ginseng value yojan, reduce complexity during classification, the driving cycle of semi-supervised kernel Fuzzy C-Means Clustering analysis to automobile of last based target function is classified.The classification accuracy that the present invention compensate for the existence of one-parameter work condition judging method is low, actual operating mode feature can not be reacted comprehensively, the deficiency that multi-parameter work condition judging method existing characteristics parameter redundancy, computing time are grown, by choosing the speed of a motor vehicle and the pedal information of accurately reflection working characteristics comprehensively, utilize twice yojan of correlation analysis and core pivot element analysis, effectively reduce the dimension of input characteristic parameter; By introducing Wavelet Kernel Function in Fuzzy C-Means Clustering Algorithm, add a small amount of training sample to guide cluster process, the judging nicety rate of automobile running working condition is high.Utilize actv. work condition judging result not only can reduce fuel consumption in vehicle traveling process and exhaust emissions amount, the research for the aspect such as exploitation, car load dynamic property coupling of all kinds of new model technology is significant equally.

Claims (4)

1. automobile running working condition judgement system, it comprises embedded system ARM (1), clock chip (2), program store (4), data memory (5), touch-screen (6), CAN interface (7) and LCDs (8); It is characterized in that, this system also comprises watchdog circuit (3), the first signal conditioning circuit (9), secondary signal modulate circuit (10) and signal acquisition circuit (11); Signal acquisition circuit (11) Negotiation speed sensor, pedal position sensor obtain the speed of a motor vehicle relevant to work condition judging and acceleration pedal and brake pedal information data, and data are passed through the first signal conditioning circuit (9) and be transformed to digital signal needed for CAN interface (7); Secondary signal modulate circuit (10) is converted to for the digital signal transmitted by CAN interface (7) simulate data that car load (12) can identify; CAN interface (7) realizes the data communication between car load (12) and embedded system ARM (1), the data upload that the first signal conditioning circuit (9) received transmits is carried out work condition judging to embedded system ARM (1), and the data passed back by embedded system ARM (1) send car load (12) to by secondary signal modulate circuit (10); Embedded system ARM (1) is by carrying out characteristic parameter extraction to the data received, then adopt correlation analysis and core pivot element analysis to carry out secondary yojan, guide cluster process to realize the differentiation of automobile running working condition by the training sample of 10% known class adding overall number of training in semi-supervised kernel Fuzzy C-Means Clustering Algorithm; Program store (4) and data memory (5) extend out chip for embedded system ARM (1), are respectively used to storage program and data; LCDs (8) is for showing the current speed of a motor vehicle, pedal aperture, and automobile running working condition information; The order of touch-screen (6) by various input and the operation of requirement control relevant device; Watchdog circuit (3) is for the monitoring of embedded system ARM (1) circuit and reset; Clock chip (2) is for providing the clock signal of needs for embedded system ARM (1), program store (4), data memory (5).

2., based on the method for discrimination of automobile running working condition judgement system described in claim 1, it is characterized in that, the method comprises the steps:

1) extraction of the speed of a motor vehicle, pedal characteristic parameter

The speed of a motor vehicle characteristic parameter extraction module (13) in embedded system ARM (1) is utilized to calculate the characteristic parameter relevant to operating mode with pedal characteristic parameter extraction module (14);

2) correlating module (15) is utilized based on correlation analysis method to step 1) characteristic parameter of sign operating mode that obtains carries out a yojan

First characteristic parameter is represented with sequence a (n) and b (n) between two respectively, and compares the similarity degree r of these two sequences in the scope of sequential n=i ~ j according to formula (1):

$r=\frac{{\mathrm{\&Sigma;}}_{n=i}^{j}a\left(n\right)b^{*}\left(n\right)}{{\left\{{\mathrm{\&Sigma;}}_{n=i}^j\right|a\left(n\right){|}^2{\mathrm{\&Sigma;}}_{n=i}^j\left|b\left(n\right){|}^2\right\}}^{\frac{1}{2}}}---\left(1\right)$

In formula, a (n) and b (n) is respectively n element in sequence a, b, and * is for get conjugation to b (n);

Operation result according to formula (1) sorts from big to small according to the eigenwert irrelevance of parameter, then 15 characteristic parameters come above are carried out correlation analysis with operating mode respectively, according to correlativity from high to low, finally have selected ten characteristic parameters as the standard dividing operating mode;

3) core pivot element analysis module (16) is utilized based on core pivot element analysis method to step 2) ten characteristic parameters obtaining carry out secondary yojan

Core pivot element analysis module (16) by kernel function the original data a of the input space _i∈ Q ^q(i=1 ..., N) and be mapped to a new high-dimensional feature space, then linear pivot analysis is carried out to the data after mapping, extract nonlinear characteristic, obtain and principal eigenvector that classify error rate little high with driving cycle sensivity;

4) utilize core Fuzzy C-Means Clustering module (17) to set up core Fuzzy C-Means Clustering model based on core Fuzzy C-Means Clustering Algorithm, carry out the differentiation of automobile running working condition according to this model

Core Fuzzy C-Means Clustering module (17) is by through step 3) sample set after yojan is divided into several fuzzy classes, by method of Lagrange multipliers structure objective function and Fuzzy C-Means Clustering model, and obtain cluster centre, subordinated-degree matrix and parameter; Cluster process is guided by the training sample of 10% known class adding overall number of training in Fuzzy C-Means Clustering Algorithm, simplify the iterative step of computation process, and improve the non-linear mapping capability of Fuzzy C-Means Clustering model by means of Wavelet Kernel Function.

3. method of discrimination as claimed in claim 2, it is characterized in that, step 1) described in the characteristic parameter relevant to operating mode comprise average operating time, average pick-up time, the mean deceleration time, the average at the uniform velocity time, average time of idle running, average range ability, maximum speed, average velociity, velocity standard deviation, peak acceleration, energy, entropy, maximum deceleration, average acceleration, at the uniform velocity time ratio, time of idle running ratio, accelerating sections average acceleration, braking section mean deceleration, acceleration standard deviation, pedal aperture rate of change, the average aperture of pedal, pedal aperture is the time of zero.

4. method of discrimination as claimed in claim 2, it is characterized in that, step 2) described ten final selected characteristic parameters are maximum speeies, average velociity, maximum deceleration, average acceleration, at the uniform velocity time ratio, time of idle running ratio, accelerating sections average acceleration, braking section mean deceleration, the average aperture of pedal, pedal aperture are the time of zero.