Summary of the invention
It is an object of the invention to for the deficiencies in the prior art, it is provided that a kind of high based on PCA, accuracy rate
Urgency deceleration Activity recognition method based on vehicle operation data.
The present invention solves the technical scheme of problem: urgency deceleration Activity recognition method based on vehicle operation data, including
Following steps:
(1) collection vehicle running data: described vehicle operation data includes instantaneous oil consumption, instantaneous acceleration, instantaneous velocity
Variable quantity per second, engine speed;
(2) the speed change factor is obtained by PCA: by described instantaneous oil consumption, instantaneous acceleration, instantaneous velocity
Variable quantity per second, engine speed are as 4 original index, by the main constituent of PCA synthesis equal number, so
After choose first three main constituent that accumulative variance contribution ratio is more than 85%, then first three main constituent described in choosing is with each
Variance contribution ratio proportion in the variance contribution ratio of all selected main constituents carry out linear combination as weight, formed and become
The speed factor;
(3) the speed change factor is taken threshold value;
(4) by comparing the speed change factor identifies in the described moment to be whether anxious in the score in certain moment and the size of threshold value
Deceleration behavior: when the score of the speed change factor is less than or equal to threshold value, the described moment is anxious deceleration the moment.
Further, in described step (2), the step being obtained the speed change factor by PCA is included:
(2.1) according to the vehicle operation data gathered, data matrix X '=(x ' is set upij)n×p, wherein, n is record number, p
For index number, xi′jRepresent the data of the i-th row jth row, and i≤n, j≤p;By instantaneous oil consumption, instantaneous acceleration, instantaneous velocity
These 4 original index of variable quantity per second, engine speed, as the original variable of principal component analysis, take p=4;
(2.2) each achievement data is standardized, in order to eliminate indices in dimension and the difference of the order of magnitude, standard
The method changed is by each data xi′jFirst deduct the average (i.e. the column mean of data matrix) of jth index, then divided by jth
The standard deviation (i.e. the row standard deviation of data matrix) of index, obtains data x after standardizationij, and then obtain standardized data square
Battle array;The average of each index is 0, and variance is 1;
(2.3) set up covariance matrix R, covariance matrix R according to standardized data matrix can reflect between each index
Dependency, each element R of covariance matrix RijRepresenting the covariance of i, j variable, computing formula is:
Wherein k is integer, represents the kth value of i, j variable;
(2.4) eigenvalue and the characteristic vector of covariance matrix R are solved: obtain p by solving characteristic equation | λ E-R |=0
Eigenvalue λi, i=1,2 ... p, wherein E is unit matrix, eigenvalue λiThe variance of size each original main constituent just, energy
The reflection acceleration information amount that comprised of main constituent number, and obtain respectively corresponding to eigenvalue λiCharacteristic vector, solve
Process is to be decomposed by covariance matrix R, and formula is:
Wherein, λiIt is the eigenvalue of covariance matrix R, eiIt is the characteristic vector of a length of p, ei TIt is eiTransposed vector;
Again by eigenvalue λiArrange according to order from big to small, obtain:
λ1>λ2>…>λp;
(2.5) variance contribution ratio of each main constituent and accumulative variance contribution ratio are calculated, and true according to accumulative variance contribution ratio
Fixed final selected main constituent number:
The computing formula of variance contribution ratio is:
The computing formula of accumulative variance contribution ratio isI.e. ask accumulative for the variance contribution ratio of i before ranking
With;
Choose front m the main constituent that accumulative variance contribution ratio is more than 85%, take m=3;
(2.6) main constituent calculating formula is write out by loading matrix:
Loading matrix is the matrix representing main constituent with original variable linear transformation relation, and the coefficient of loading matrix is the most every
The value of individual main constituent characteristic of correspondence vector, writes out i-th main constituent f accordinglyiComputing formula:
fi=e1i*x1+e2i*x2+…+epi*xp,
Wherein, epiIt is the i-th component of pth characteristic vector, xpIt it is pth index;It is a n-dimensional vector, therefore
The main constituent obtained also is a n-dimensional vector;
(2.7) score of the speed change factor of every record, the change of i-th record is calculated according to m the main constituent selected
Speed factor siThe computing formula of score be:
1≤k≤m;
Wherein any one fkiRepresent the i-th component of kth main constituent, coefficientComputing formula be:
Further, in described step (3), the threshold value of the described speed change factor is-4.
Further, in described step (2.5), choose first three main constituent more than 85% of accumulative variance contribution ratio.
Further, in described step (2.5), the variance contribution ratio of first three main constituent selected is respectively
53.5%, 21.8%, 14.5%, the coefficient of first three main constituent described then calculated in step (2.7) is respectively
0.596,0.243,0.161, therefore speed change factor s of i-th recordiThe computing formula of score be:
si=0.596*f1i+0.243*f2i+0.161*f3i。
Further, described in the accumulation contribution rate of first three main constituent that selects can reach more than 89.8%, this is described
Three main constituents can represent the velocity variations information of 89.8%.
The invention have the benefit that the present invention utilizes the dimensionality reduction thought of PCA, multi objective is converted into comprehensive
Close index, reduce the dimension of observation space, obtain topmost information, by by the several variablees main one-tenth relevant to anxious deceleration
Point analytic process carries out comprehensively, significantly improves the anxious accuracy of deceleration Activity recognition, effectiveness.
Detailed description of the invention
Below in conjunction with the accompanying drawings and specific embodiment, the present invention is further illustrated.
As it is shown in figure 1, urgency deceleration Activity recognition method based on vehicle operation data, comprise the steps:
(1) collection vehicle running data: described vehicle operation data includes instantaneous oil consumption, instantaneous acceleration, instantaneous velocity
Variable quantity per second, engine speed;
(2) the speed change factor is obtained by PCA: by described instantaneous oil consumption, instantaneous acceleration, instantaneous velocity
Variable quantity per second, engine speed are as 4 original index, by the main constituent of PCA synthesis equal number, so
After choose first three main constituent that accumulative variance contribution ratio is more than 85%, then first three main constituent described in choosing is with each
Variance contribution ratio proportion in the variance contribution ratio of all selected main constituents carry out linear combination as weight, formed and become
The speed factor;
(3) the speed change factor is taken threshold value;
(4) by comparing the speed change factor identifies in the described moment to be whether anxious in the score in certain moment and the size of threshold value
Deceleration behavior: when the score of the speed change factor is less than or equal to threshold value, the described moment is anxious deceleration the moment.
In described step (2), the step being obtained the speed change factor by PCA is included:
(2.1) according to the vehicle operation data gathered, data matrix X '=(x is set upi′j)n×p, wherein, n is record number, p
For index number, x 'ijRepresent the data of the i-th row jth row, and i≤n, j≤p;By instantaneous oil consumption, instantaneous acceleration, instantaneous velocity
These 4 original index of variable quantity per second, engine speed, as the original variable of principal component analysis, take p=4;
(2.2) each achievement data is standardized, in order to eliminate indices in dimension and the difference of the order of magnitude, standard
The method changed is by each data xi′jFirst deduct the average (i.e. the column mean of data matrix) of jth index, then divided by jth
The standard deviation (i.e. the row standard deviation of data matrix) of index, obtains data x after standardizationij, and then obtain standardized data square
Battle array;The average of each index is 0, and variance is 1;
(2.3) set up covariance matrix R, covariance matrix R according to standardized data matrix can reflect between each index
Dependency, each element R of covariance matrix RijRepresenting the covariance of i, j variable, computing formula is:
Wherein k is integer, represents the kth value of i, j variable;
(2.4) eigenvalue and the characteristic vector of covariance matrix R are solved: obtain p by solving characteristic equation | λ E-R |=0
Eigenvalue λi, i=1,2 ... p, wherein E is unit matrix, eigenvalue λiThe variance of size each original main constituent just, energy
The reflection acceleration information amount that comprised of main constituent number, and obtain respectively corresponding to eigenvalue λiCharacteristic vector, solve
Process is to be decomposed by covariance matrix R, and formula is:
Wherein, λiIt is the eigenvalue of covariance matrix R, eiIt is the characteristic vector of a length of p, ei TIt is eiTransposed vector;
Again by eigenvalue λiArrange according to order from big to small, obtain:
λ1>λ2>…>λp;
(2.5) variance contribution ratio of each main constituent and accumulative variance contribution ratio are calculated, and true according to accumulative variance contribution ratio
Fixed final selected main constituent number:
The computing formula of variance contribution ratio is:
The computing formula of accumulative variance contribution ratio isI.e. ask accumulative for the variance contribution ratio of i before ranking
With;
Choose front m the main constituent that accumulative variance contribution ratio is more than 85%, take m=3;
(2.6) main constituent calculating formula is write out by loading matrix:
Loading matrix is the matrix representing main constituent with original variable linear transformation relation, and the coefficient of loading matrix is the most every
The value of individual main constituent characteristic of correspondence vector, writes out i-th main constituent f accordinglyiComputing formula:
fi=e1i*x1+e2i*x2+…+epi*xp,
Wherein, epiIt is the i-th component of pth characteristic vector, xpIt it is pth index;It is a n-dimensional vector, therefore
The main constituent obtained also is a n-dimensional vector;
(2.7) score of the speed change factor of every record, the change of i-th record is calculated according to m the main constituent selected
Speed factor siThe computing formula of score be:
1≤k≤m;
Wherein any one fkiRepresent the i-th component of kth main constituent, coefficientComputing formula be:
In described step (3), the threshold value of the described speed change factor is-4.
In described step (2.5), choose first three main constituent more than 85% of accumulative variance contribution ratio.
In described step (2.5), the variance contribution ratio of first three main constituent selected is respectively 53.5%, and 21.8%,
14.5%, the coefficient of first three main constituent described then calculated in step (2.7) is respectively 0.596, and 0.243,
0.161, therefore speed change factor s of i-th recordiThe computing formula of score be:
si=0.596*f1i+0.243*f2i+0.161*f3i。
The accumulation contribution rate of described first three selected main constituent can reach more than 89.8%, and these three main constituent is described
The velocity variations information of 89.8% can be represented.
Using the method for the invention to test, specific experiment process is:
Automobile after start-up, first slowly accelerates a period of time, brings to a halt when reaching higher speed per hour, makes automobile anxious suddenly
Slow down, the most again speed is slowly promoted, when speed reaches high value, unclamp throttle, allow car speed slowly reduce, until
Stop.The speed data of experimental record as in figure 2 it is shown, fine rule represents the change of the instantaneous velocity of traveling, thick line represent speed change because of
The fluctuation of sub-score, abscissa express time, the vertical coordinate on the left side represents instantaneous velocity, the vertical coordinate on the right represent speed change because of
Son.Present invention determine that and identify that the anxious standard slowing down the moment is: when certain moment speed change factor score≤-4, this moment is anxious subtracting
The speed moment.
As in figure 2 it is shown, the method for the invention is identifying anxious deceleration the moment, distinguish slow moderating process aspect the most effective.
With the data instance in Fig. 2, at 43s, speed change factor score is-7.64, less than threshold value-4, when being therefore identified as anxious deceleration
Carving, just can see in conjunction with instantaneous velocity change curve, after this moment of 43s, in 4s, instantaneous velocity is reduced to by 57.75km/h
0.3km/h, the anxious decelerating phase in the most corresponding experiment, therefore this identification visible is effective, and equally, this method of discrimination is to slowly
Moderating process also have differentiation, as after 80s, automobile just reaches about 57km/h in the range of decrease of more than 13s speed, just corresponding
The slow decelerating phase in experiment, and the speed change factor is the highest this moment, and it is not recognized as anxious deceleration, therefore with practical situation phase
Symbol.By carrying out the experiment of other groups, it is also demonstrated that the effectiveness of anxious deceleration Activity recognition method of the present invention.
The ultimate principle of the present invention is:
The present invention is by means of PCA, and PCA is the statistical method of a kind of Data Dimensionality Reduction, it by
In an orthogonal transformation, become to summarize a few index of original most information by original multiple index comprehensives, do not damage
On the premise of losing important information, reduce the latitude of observation space.
In practical problem is studied, for comprehensively and systematically problem analysis, it is necessary to consider numerous influence factor.These relate to
And factor be commonly referred to as index or variable.Because each index reflects some studied a question to varying degrees
There is certain dependency between information, and index each other, thus the information of the statistical data reflection of gained is to a certain extent
There is overlap.When with study of statistical methods Multivariable, variable can increase amount of calculation too much and increase the complexity of problem analysis
Property, it is desirable to during carrying out quantitative analysis, the variable related to is less, and the quantity of information obtained is more.Principal component analysis
Adapt to what this requirement produced just, be the ideal tools of such issues that solve.
On the whole, principal component analysis is intended to utilize the thought of dimensionality reduction, and multi objective is converted into a few aggregative indicator,
Reduce the dimension of observation space, to obtain topmost information.Assume there be p index, the most at most have p aggregative indicator (main
Composition).Not increasing due to population variance and do not subtract, the variance of front several aggregative indicatores is relatively big, and the variance of the most several aggregative indicatores is less.
Strictly, only front several aggregative indicatores just deserve to be called " leading " composition, rear several aggregative indicatores actually " secondary " composition.In practice always
It is that reservation is front several, several after ignoring.Retain how many main constituents and depend on that the cumulative variance of member-retaining portion is in variance summation
Percentage.
Four original index are formed four main constituents through combination by the present invention, then choose accumulation contribution rate be 85% with
On first three main constituent, then first three main constituent is carried out linear combination using its variance contribution ratio ratio as weight, finally
Synthesizing an aggregative indicator, i.e. the speed change factor, then having judged the urgency deceleration behavior of automobile by the speed change factor being taken threshold value.
The present invention is not limited to above-mentioned embodiment, in the case of without departing substantially from flesh and blood of the present invention, and art technology
Personnel it is contemplated that any deformation, improve, replace and each fall within protection scope of the present invention.