CN112046489B - Driving style identification algorithm based on factor analysis and machine learning - Google Patents

Driving style identification algorithm based on factor analysis and machine learning Download PDF

Info

Publication number
CN112046489B
CN112046489B CN202010893251.2A CN202010893251A CN112046489B CN 112046489 B CN112046489 B CN 112046489B CN 202010893251 A CN202010893251 A CN 202010893251A CN 112046489 B CN112046489 B CN 112046489B
Authority
CN
China
Prior art keywords
representing
factor
driving
driving style
driver
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
Application number
CN202010893251.2A
Other languages
Chinese (zh)
Other versions
CN112046489A (en
Inventor
赵健
陈志成
朱冰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jilin University
Original Assignee
Jilin University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Jilin University filed Critical Jilin University
Priority to CN202010893251.2A priority Critical patent/CN112046489B/en
Publication of CN112046489A publication Critical patent/CN112046489A/en
Application granted granted Critical
Publication of CN112046489B publication Critical patent/CN112046489B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W40/00—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models
    • B60W40/08—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models related to drivers or passengers
    • B60W40/09—Driving style or behaviour
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B60W2050/0001—Details of the control system
    • B60W2050/0019—Control system elements or transfer functions
    • B60W2050/0028—Mathematical models, e.g. for simulation
    • B60W2050/0029—Mathematical model of the driver
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W2540/00—Input parameters relating to occupants
    • B60W2540/30—Driving style
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Abstract

The invention belongs to the technical field of automobiles, and particularly relates to a driving style identification algorithm based on factor analysis and machine learning. The driving style identification algorithm firstly selects data strongly related to driving style as driving style characteristic parameters, uses factor analysis to reduce the dimension of the characteristic parameters to obtain common factors, reduces redundancy among driving data and assigns corresponding physical significance to the common factors; using a common factor as input, and adopting a Gaussian mixture model clustering algorithm to mark corresponding driving style labels for different drivers; the driving style recognition model is then trained using a back-propagation neural network optimized by a genetic algorithm. By fusing unsupervised learning and supervised learning, the cost of identification can be effectively reduced. The initial weight of the back propagation neural network is optimized by using the genetic algorithm, so that the identification precision of the model can be effectively improved, and the blank that the driving style cannot be identified accurately in the prior art is filled.

Description

Driving style identification algorithm based on factor analysis and machine learning
Technical Field
The invention belongs to the technical field of automobiles, and particularly relates to a driving style identification algorithm based on factor analysis and machine learning.
Background
With the continuous progress of science and technology, the research of automobiles gradually develops from the traditional driver vehicle in the ring direction to the intelligent driving of unmanned vehicles. However, the intelligent vehicle still has great scientific and technological difficulties in many aspects such as traffic regulation preparation, vehicle safety technical redundancy, complex scene decision and the like, so that the driver still participates in the research and development process of the intelligent vehicle for a long time.
Drivers often exhibit different driving styles due to different factors such as gender, driving age, occupation, etc. In the driving process of the vehicle, a driver can select proper driving operation according to the driving style of the driver to obtain comfortable driving experience. Taking an urban car following working condition as an example, an aggressive driver usually follows a car at a larger braking deceleration when the car following distance is shorter, so that the following distance between a main car and a front car is ensured to be smaller, and the lane-changing overtaking is prepared at any time; while a conservative driver performs a braking operation at a distance far from the preceding vehicle, finely adjusts the following distance between the two vehicles at a small braking deceleration. For drivers with different driving styles, the driving styles of the intelligent automobiles should be fully considered when the intelligent automobiles perform decision control, motion planning and driving control, so that the vehicle performance is adjusted and the riding experience of the drivers is improved. Therefore, the method has important guiding significance for accurately identifying the driving style of the driver to the design of the intelligent automobile.
However, the driving style of the driver is highly random and non-linear, so that it is difficult for the conventional rule-based or model-based method to accurately identify the driving style from the driving data with large dimensions. Thanks to the rapid development of machine learning, supervised learning has higher recognition accuracy in the field of identification, but it requires an external system to provide accurate driving style labels for it. The driving style label obtained by the investigation method is high in cost, and the driving style can be obviously changed along with the experience process of the driver. Therefore, how to accurately identify the driving style becomes a difficult point to be solved urgently.
Disclosure of Invention
The invention provides a driving style identification algorithm based on factor analysis and machine learning, which is characterized in that firstly, data strongly related to driving style are selected as driving style characteristic parameters, the characteristic parameters are subjected to dimension reduction by using factor analysis to obtain common factors, the redundancy among driving data is reduced, and corresponding physical significance is assigned to the common factors; using a common factor as input, and adopting a Gaussian mixture model clustering algorithm to mark corresponding driving style labels for different drivers; the driving style recognition model is then trained using a back-propagation neural network optimized by a genetic algorithm. By fusing unsupervised learning and supervised learning, the cost of identification can be effectively reduced. The initial weight of the back propagation neural network is optimized by using the genetic algorithm, so that the identification precision of the model can be effectively improved, and the blank that the driving style cannot be identified accurately in the prior art is filled.
The technical scheme of the invention is described as follows by combining the attached drawings:
a driving style identification algorithm based on factor analysis and machine learning is characterized by comprising the following steps:
step one, extracting driving characteristic parameters strongly related to driving style from typical urban following working conditions;
step two, using a factor analysis algorithm including feasibility analysis, public factor determination and factor score to reduce the dimension and process the driving characteristic parameters, obtaining public factors and giving specific physical meanings to the public factors;
clustering the common factors by using a Gaussian mixture model, and accurately attaching a driving style label to each driver;
and step four, training a driving style identification model by adopting a back propagation neural network optimized by a genetic algorithm, and accurately outputting different driving styles.
The driving characteristic parameters in the first step comprise: mean value A of the master cylinder pressure1Maximum value A of master cylinder pressure2Standard deviation of master cylinder pressure A3Average value A of engine torque4Standard deviation of engine torque A5Average engine speed A6Standard deviation of engine speed A7Average value A of the distance between two vehicles8Standard deviation A of distance between two vehicles9Average value of two vehicle speed differences A10And standard deviation A of two vehicle speed differences11。
The specific method of the second step is as follows:
21) the driving data is normalized by a normalization method:
Figure GDA0002908885600000031
Figure GDA0002908885600000032
wherein N represents the number of collected driver data samples; x is the number ofki(i1,2, … 11) represents the ith driving style characteristic parameter value of the kth driver; x is the number ofkj(j ═ 1,2, … 11) represents the j-th driving style characteristic parameter value of the k-th driver;
Figure GDA0002908885600000033
representing the average value of ith driving characteristic parameters of N drivers;
Figure GDA0002908885600000034
representing the average value of j driving characteristic parameters of N drivers; x'kiAn ith normalized driving style characteristic parameter value representing a kth driver; x'kjA j-th normalized driving style characteristic parameter value representing a k-th driver;
22) the normalized feature data was subjected to correlation analysis:
Figure GDA0002908885600000035
in the formula, rijRepresenting elements in a matrix of correlation coefficients; x'kiAn ith normalized driving style characteristic parameter value representing a kth driver; n represents the number of collected driver data samples; x'kjA j-th normalized driving style characteristic parameter value representing a k-th driver;
23) solving the characteristic root and the corresponding characteristic vector of the related coefficient matrix according to the characteristic equation:
|λjI-Rij|＝0 (4)
RijUj＝λjUj (5)
in the formula, λj(j ═ 1,2, … 11) represents the correlation coefficient matrix RijA characteristic root of; i represents a unit coefficient matrix; u shapejRepresenting a feature vector matrix corresponding to the feature root; rijRepresenting a matrix of correlation coefficients;
24) selection factor F3×1As a common factor for the driving data and analyzed by a factor load matrix; due to the fact thatSub-load matrix A11×3From the characteristic value λ3×1And a feature vector U11×3Obtaining;
Figure GDA0002908885600000041
in the formula, A11×3A representation factor load matrix; lambda [ alpha ]1，λ2，λ3Respectively represent a common factor F3×1Corresponding characteristic values; u shapei,j(i-1, 2, … 11, j-1, 2,3) denotes a common factor F3×1A corresponding feature vector;
25) rotating the factor load matrix by adopting a maximum variance method, and naming the characteristic parameter A4-A7 as a driving factor; the characteristic parameter A1-A3 is named as a braking factor; the characteristic parameters A8-A11 are named as environmental factors;
26) the factor score model is:
Xm×1＝Am×3·F3×1+εm×1 (9)
in the formula, Xm×1(m ═ 1,2,3) represents a factor score model; a. them×3A presentation factor score coefficient; epsilonm×1Representing a particular factor, taking epsilonm×1Solving for zero vector; f3×1Representing selected three common factors; solving by using an Anderson-Rubin regression method to obtain a factor score coefficient, and finally obtaining a common factor expression according to an analysis tree of the factor as follows:
Figure GDA0002908885600000042
in the formula, F1Represents a first common factor, i.e. a drive factor; f2Represents a second common factor, the braking factor; f3Represents a third common factor, namely an environmental factor; a. the1Represents the average value of the master cylinder pressure; a. the2Represents the maximum value of the master cylinder pressure; a. the3Represents the standard deviation of the master cylinder pressure; a. the4An average value representing engine torque; a. the5Representing a standard deviation of engine torque; a. the6Representing the average rotational speed of the engine; a. the7Indicating a standard deviation of the engine speed; a. the8Representing the average value of the distance between two vehicles; a. the9Representing the standard deviation of the distance between two vehicles; a. the10Represents the average of the two vehicle speed differences; a. the11The standard deviation of the two vehicle speed differences is indicated.
The concrete method of the third step is as follows:
31) setting driving styles of a driver to be an aggressive type, a general type and a conservative type respectively;
32) carrying out probability calculation on the common factors by adopting a multidimensional Gaussian mixture model, and defining as follows:
Figure GDA0002908885600000051
Figure GDA0002908885600000052
in the formula, FiRepresenting a common factor obtained after dimensionality reduction of the ith driver driving habit data; p (F)i) Common factor F representing the ith driveriA characteristic probability density function of (a); k represents a cluster of a gaussian mixture model, and includes three types of driving styles, so that K is set to 3; mu.skRepresenting the average value of each cluster; sigmakRepresenting the covariance of each cluster; p (F)i|k)＝N(Fi|μk,σk) A probability density function representing a kth gaussian model; pikIs a weight of the kth Gaussian model, and
Figure GDA0002908885600000053
the goal of gaussian mixture model clustering is to find a suitable gaussian distribution such that the probability of clustering approaches maximum; this target value is set as the maximum likelihood function:
Figure GDA0002908885600000054
in the formula, pikIs the weight of the kth gaussian model; n represents the number of collected driver data samples; fiRepresenting a common factor obtained after dimensionality reduction of the ith driver driving habit data; mu.skRepresenting the average value of each cluster; sigmakRepresenting the covariance of each cluster.
Solving a maximum likelihood function by adopting an Expectation-Maximization (Expectation-Maximization) algorithm; at the desired step, we set the mean μ of the initial cluster of the Gaussian mixture modelintCovariance σintAnd Gaussian model weight value piintAnd based thereon calculating the likelihood that each common factor belongs to the kth Gaussian mixture model:
Figure GDA0002908885600000055
in the formula (I), the compound is shown in the specification,
Figure GDA0002908885600000056
representing the probability that the common driving habit factor of the ith driver belongs to the kth Gaussian mixture model; pikAnd pijAre the weights of the kth and jth Gaussian models, respectively, and
Figure GDA0002908885600000057
Figure GDA0002908885600000058
n represents the number of collected driver data samples; fiRepresenting a common factor obtained after dimensionality reduction of the ith driver driving habit data; mu.skAnd mujRespectively representing the average values of the kth clustering cluster and the jth clustering cluster; sigmakAnd σjRespectively representing the covariance of the kth clustering cluster and the jth clustering cluster;
at the maximum step, derived from the desired step
Figure GDA0002908885600000061
Solving each Gaussian mixture model clusterCluster new mean, new covariance, and new gaussian model weight values:
then, let the average value of the initial cluster of the Gaussian mixture model
Figure GDA0002908885600000062
Covariance of initial cluster of Gaussian mixture model
Figure GDA0002908885600000063
Weight value of initial cluster of Gaussian mixture model
Figure GDA0002908885600000064
Substituting into formula (11), repeating expectation-maximization algorithm until the maximum likelihood function in formula (10) converges, i.e. identifying the number of three drivers, aggressive type, general type and conservative type.
The concrete method of the fourth step is as follows:
41) the fitness function of the genetic algorithm is set as:
Figure GDA0002908885600000065
in the formula, Fit represents a fitness function of a genetic algorithm; y represents the correct driver driving style type; w1Weights representing a back-propagation neural network hidden layer; w2Weights representing the back-propagation neural network output layer; b is1A threshold value representing a back-propagation neural network hidden layer; b is2A threshold value representing a back-propagation neural network output layer; f. of1An activation function representing a back-propagation neural network hidden layer; f. of2Representing an activation function of an output layer of the back propagation neural network, wherein the hidden layer activation function is set as a tansig function, and the activation function of the output layer is set as a pureline function; p represents a common factor of the model inputs;
42) setting 50 initial populations of the genetic algorithm, and respectively encoding all weights and thresholds in the back propagation neural network into chromosome structures of the genetic algorithm; then, the chromosome with the maximum fitness function value in the current generation population is completely replaced by the worst chromosome, and the weight and the threshold contained in the current generation chromosomes are mutually exchanged to generate a new chromosome; in addition, new chromosomes are obtained by randomly changing the chromosome structures of the individual chromosomes, wherein the direction of the variation is expressed as the direction towards the optimal population fitness:
Figure GDA0002908885600000066
in the formula, FitbestRepresents the maximum value of the fitness function in the current generation chromosome; fit represents a fitness function; x is the number oftRepresents the current generation of chromosomes randomly selected for mutation; x is the number oft+1Representing a new generation of chromosomes after mutation;
and then continuously repeating the processes of heredity, hybridization and mutation until the fitness function meets the constraint condition, decoding the chromosome reduction weight and the threshold value, and setting the chromosome reduction weight and the threshold value as the initial value of the back propagation neural network.
The invention has the beneficial effects that:
1) the method guides different driving style identification scenes and selects proper characteristic parameters;
2) the invention adopts a factor analysis method to effectively reduce the correlation between the characteristic parameters of the original driving style on the basis of keeping the original information, thereby reducing the redundancy between the characteristic parameters
3) The method uses a factor analysis method to obtain the common factors, effectively endows the common factors with proper physical significance according to the classification result among the original characteristic parameters, and is beneficial to improving the understanding of the model;
4) according to the method, the public factors are used for replacing original characteristic parameters as the input of the driving style identification model, so that the accuracy and generalization capability of model identification can be effectively improved;
5) the Gaussian mixture clustering algorithm belongs to a soft clustering algorithm, and corresponding driving style labels are attached to drivers with different driving styles according to the probability. The clustering accuracy of such "soft" clustering algorithms is higher than that of "hard" clustering algorithms such as K-means;
6) the driving style identification model obtained based on the back propagation neural network training has the advantages of simple structure, high identification precision and strong nonlinear mapping capability;
7) compared with the traditional rule-based or model-based identification algorithm, the back propagation neural network is independent of engineering experience and model parameters, and has stronger practicability;
8) the invention adopts the genetic algorithm to optimize the back propagation neural network, and can effectively improve the convergence speed and the identification precision of the driving style identification model training.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings to be used in the description of the embodiments of the present invention will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the contents of the embodiments of the present invention and the drawings without creative efforts.
FIG. 1 is a general architecture diagram of a driving style identification strategy;
FIG. 2 is a diagram illustrating a clustering result of a Gaussian mixture model;
FIG. 3 is a flow chart of the algorithm for genetic algorithm optimization of back propagation neural network
FIG. 4 is a schematic diagram of a training result of a back propagation neural network based on genetic algorithm optimization.
Detailed Description
Although the preferred embodiments of the present invention have been described in detail with reference to the accompanying drawings, the scope of the present invention is not limited to the specific details of the above embodiments, and any person skilled in the art can substitute or change the technical solution of the present invention and its inventive concept within the technical scope of the present invention, and these simple modifications belong to the scope of the present invention.
Step one, extracting driving characteristic parameters strongly related to driving style from typical urban following working conditions;
the typical city following work condition comprises a plurality of driving data related to driving style; taking brake data as an example, signals such as brake pedal force, brake pedal travel and the like of a driver in the real vehicle motion process usually have large jitter, and if the signals are used for driving style identification, the identification model can hardly accurately reflect the real driving style of the driver. The master cylinder pressure signal sent by the vehicle-mounted controller is generally smooth, and the average value, the maximum value and the variance of the master cylinder pressure of a driver in a vehicle following period can be used for expressing the braking style of the driver. The average value of the master cylinder pressure represents the requirement of a driver for vehicle brake deceleration in a long time, the maximum value of the master cylinder pressure represents the grasp of the driver for vehicle motion control in a short time, and the standard deviation of the master cylinder pressure represents the operation proficiency of the driver for the whole vehicle brake system. In addition, the driving data and the interaction data between the host vehicle and the environment or the front vehicle are also of great importance in the recognition of the driving style, and we finally extract 11 driving characteristic parameters as shown in the chart 1 by a method similar to braking.
TABLE 1 Driving style characteristic parameters under typical urban following conditions
Figure GDA0002908885600000081
Figure GDA0002908885600000091
It should be noted that the driving style characteristic parameter extraction method designed for the typical vehicle following condition in this document may also provide reference for other scenes such as lane change, overtaking, and the like.
Step two, using a factor analysis algorithm including feasibility analysis, public factor determination and factor score to reduce the dimension and process the driving characteristic parameters, obtaining public factors and giving specific physical meanings to the public factors;
certain information redundancy exists among the 11 selected driving style characteristic parameters, and if the 11 selected driving style characteristic parameters are directly used for data analysis, the problems of large calculated amount, complex model, low identification efficiency and the like are caused. Therefore, the characteristic parameters are subjected to dimension reduction treatment by adopting a factor analysis method; the method comprises the following specific steps:
21) the dimension size of the original data parameters collected by the real vehicle has certain difference. In order to improve the accuracy of factor analysis, the driving data is normalized by a normalization method:
Figure GDA0002908885600000092
Figure GDA0002908885600000093
wherein N represents the number of collected driver data samples; x is the number ofki(i, j ═ 1,2, … 11) represents the i-th driving style characteristic parameter value of the k-th driver; x is the number ofkj(i, j ═ 1,2, … 11) respectively represent j-th driving style characteristic parameter values of the k-th driver;
Figure GDA0002908885600000094
representing the average value of ith driving characteristic parameters of N drivers;
Figure GDA0002908885600000095
representing the average value of j driving characteristic parameters of N drivers; x'kiAn ith normalized driving style characteristic parameter value representing a kth driver; x'kjA j-th normalized driving style characteristic parameter value representing a k-th driver.
22) The normalized feature data was subjected to correlation analysis:
A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11
A1 1.000 0.753 0.842 0.129 0.103 -0.149 -0.108 -0.360 -0.280 -0.240 -0.203
A2 0.753 1.000 0.942 0.249 0.153 -0.066 -0.025 -0.484 -0.347 -0.266 -0.203
A3 0.842 0.942 1.000 0.197 0.870 -0.127 -0.095 -0.455 -0.323 -0.263 -0.186
A4 0.129 0.249 0.197 1.000 0.595 0.280 0.491 -0.419 -0.326 -0.014 -0.150
A5 0.103 0.153 0.087 0.595 1.000 0.403 0.618 -0.404 -0.163 -0.110 -0.179
A6 -0.149 -0.066 -0.127 0.280 0.403 1.000 0.763 -0.052 0.029 0.164 -0.039
A7 -0.108 -0.025 -0.095 0.491 0.618 0.763 1.000 -0.141 0.045 0.259 0.069
A8 -0.360 -0.484 -0.455 -0.419 -0.404 -0.052 -0.141 1.000 0.820 0.705 0.633
A9 -0.280 -0.347 -0.323 -0.326 -0.163 0.029 0.045 0.820 1.000 0.767 0.761
A10 -0.240 -0.266 -0.263 -0.014 -0.110 0.164 0.259 0.705 0.767 1.000 0.825
A11 -0.203 -0.203 -0.186 -0.150 -1.79 -0.039 0.069 0.633 0.761 0.825 1.000
Figure GDA0002908885600000101
in the formula, rijRepresenting elements in a matrix of correlation coefficients; x'kiAn ith normalized driving style characteristic parameter value representing a kth driver; n represents the number of collected driver data samples; x'kjA j-th normalized driving style characteristic parameter value representing a k-th driver;
in order to better explain the dimension reduction process, 103 groups of drivers collected by the self under the typical city vehicle following condition are used in the patentFurther analysis is performed on the traveller data. Therefore, N is 103 and r is obtained according to the formula (3)ijForm the final correlation coefficient matrix RijAs shown in table 2.
TABLE 2 correlation coefficient matrix of characteristic parameters
Subsequently, the matrix of correlation coefficients was analyzed for feasibility using Kaiser-Meyer-Olkin (KMO) and Bartlett sphericity test, and the test results shown in Table 3 were obtained. As can be seen from Table 2, the driving data has a large correlation coefficient, and the KMO shown in Table 3 is close to 1, so that the chi-square statistic value of the Bartlett sphericity test is large, and the accompanying probability is far less than 0.01.
TABLE 3KMO and Bartlett sphericity test results
Figure GDA0002908885600000102
Figure GDA0002908885600000111
23) Solving the characteristic root and the corresponding characteristic vector of the related coefficient matrix according to the characteristic equation:
|λjI-Rij|＝0 (4)
RijUj＝λjUj (5)
in the formula, λj(j ═ 1,2, … 11) represents the correlation coefficient matrix RijA characteristic root of; i represents a unit coefficient matrix; u shapejRepresenting a feature vector matrix corresponding to the feature root; rijRepresenting a matrix of correlation coefficients;
and then the variance contribution ratio z according to the common factorjAnd cumulative variance contribution ratio ZjThe formula can be found:
Figure GDA0002908885600000112
Figure GDA0002908885600000113
in the formula, zjRepresenting a variance contribution rate; zjRepresenting a cumulative variance contribution rate; lambda [ alpha ]j(j ═ 1,2, … 11) represents the correlation coefficient matrix RijThe jth feature root of (1); lambda [ alpha ]nRepresenting a matrix R of correlation coefficientsijThe nth feature root of (1); lambda [ alpha ]πRepresenting a matrix R of correlation coefficientsijThe pi-th root of feature;
the results of the variance evaluation are shown in table 4. It can be seen from the table that the variance contribution ratio of the first 3 common factors has reached 80.430%, which is sufficient to characterize most of the information of the original characteristic parameters. The first three factors F in Table 4 were therefore selected for this application3×1As a common factor for driving data.
TABLE 4 variance and cumulative variance contribution ratio of common factors
Figure GDA0002908885600000114
Figure GDA0002908885600000121
Common factor F3×1Usually has certain physical meaning, and can be analyzed through a factor load matrix; factor load matrix A11×3Can be determined from the characteristic value lambda3×1And a feature vector U11×3Obtaining:
Figure GDA0002908885600000122
in the formula, A11×3A representation factor load matrix; lambda [ alpha ]1，λ2，λ3Respectively represent a common factor F3×1Corresponding characteristic values; u shapei,j(i-1, 2, … 11, j-1, 2,3) represents a common factorF3×1A corresponding feature vector;
25) however, the differences of the obtained factor load matrices are small, so the factor load matrices are further rotated by the maximum variance method, and the output results are shown in table 5. The characteristic parameters A4-A7 have strong relations and are all related to the main vehicle driving, so the characteristic parameters are named driving factors. The characteristic parameters A1-A3 have stronger relation and are all related to the braking of the main vehicle, so the characteristic parameters are named as braking factors; the characteristic parameters A8-A11 have strong relationship and have certain relationship with the information of two workshops, so the characteristic parameters are named as environmental factors.
TABLE 5 load matrix of factors and twiddle factors
Figure GDA0002908885600000123
After the common factors of the driving data are determined, the information of the 11 characteristic parameters can be better represented only by solving the factor score. The factor score model selected was:
Xm×1＝Am×3·F3×1+εm×1 (9)
in the formula, Xm×1(m ═ 1,2,3) represents a factor score model; a. them×3A presentation factor score coefficient; epsilonm×1Representing a particular factor, taking epsilonm×1Solving for zero vector; f3×1Representing selected three common factors; the application uses the Anderson-Rubin regression method to solve and obtain the factor score coefficient, as shown in Table 6. The common factor expressions obtained finally from the analysis tree of factors are:
Figure GDA0002908885600000131
in the formula, F1Represents a first common factor, i.e. a drive factor; f2Represents a second common factor, the braking factor; f3Represents a third common factor, namely an environmental factor; a. the1Represents the average value of the master cylinder pressure; a. the2Indicating master cylinder pressureThe maximum value of the force; a. the3Represents the standard deviation of the master cylinder pressure; a. the4An average value representing engine torque; a. the5Representing a standard deviation of engine torque; a. the6Representing the average rotational speed of the engine; a. the7Indicating a standard deviation of the engine speed; a. the8Representing the average value of the distance between two vehicles; a. the9Representing the standard deviation of the distance between two vehicles; a. the10Represents the average of the two vehicle speed differences; a. the11The standard deviation of the two vehicle speed differences is indicated.
TABLE 6 factor score matrix
Figure GDA0002908885600000132
Clustering the common factors by using a Gaussian mixture model, and accurately attaching a driving style label to each driver;
in order to meet the high-precision requirement of the whole vehicle on the driving style identification algorithm, the driving style is identified by adopting a supervised learning strategy. The common factors after the dimension reduction is analyzed by using the factors do not obtain the category of the driving style, and an accurate label cannot be provided for supervised learning. It is therefore necessary to obtain the driving style label using a clustering algorithm. The driving style label is obtained by using a soft clustering Gaussian mixture model.
31) Firstly, setting driving styles of a driver to be an aggressive type, a general type and a conservative type respectively;
32) carrying out probability calculation on the common factors by adopting a multidimensional Gaussian mixture model, and defining as follows:
Figure GDA0002908885600000141
Figure GDA0002908885600000142
in the formula, FiRepresenting a common factor obtained after dimensionality reduction of the ith driver driving habit data; p (F)i) Is shown asCommon factor F for i driversiA characteristic probability density function of (a); k represents a cluster of a gaussian mixture model, and includes three types of driving styles, so that K is set to 3; mu.skRepresenting the average value of each cluster; sigmakRepresenting the covariance of each cluster; p (F)i|k)＝N(Fi|μk,σk) A probability density function representing a kth gaussian model; pikIs a weight of the kth Gaussian model, and
Figure GDA0002908885600000143
the goal of gaussian mixture model clustering is to find a suitable gaussian distribution such that the probability of clustering approaches maximum; in general, this target value can be set as the maximum likelihood function:
Figure GDA0002908885600000144
in the formula, pikIs the weight of the kth gaussian model; n represents the number of collected driver data samples; fiRepresenting a common factor obtained after dimensionality reduction of the ith driver driving habit data; mu.skRepresenting the average value of each cluster; sigmakRepresenting the covariance of each cluster.
The application adopts Expectation-Maximization (Expectation-Maximization) algorithm to solve the maximum likelihood function. At the Expectation (Expectation) step, we randomly set the mean μ of the initial cluster of Gaussian mixture modelsintCovariance σintAnd Gaussian model weight value piintAnd based thereon calculating the likelihood that each common factor belongs to the kth Gaussian mixture model:
Figure GDA0002908885600000145
in the formula (I), the compound is shown in the specification,
Figure GDA0002908885600000146
indicates the ith drivingThe probability that the common factor of the driving habits of the person belongs to the kth Gaussian mixture model; pikAnd pijAre the weights of the kth and jth Gaussian models, respectively, and
Figure GDA0002908885600000147
Figure GDA0002908885600000151
n represents the number of collected driver data samples; fiRepresenting a common factor obtained after dimensionality reduction of the ith driver driving habit data; mu.skAnd mujRespectively representing the average values of the kth clustering cluster and the jth clustering cluster; sigmakAnd σjRespectively representing the covariance of the kth clustering cluster and the jth clustering cluster;
Figure GDA0002908885600000152
Figure GDA0002908885600000153
Figure GDA0002908885600000154
in the formula (I), the compound is shown in the specification,
Figure GDA0002908885600000155
representing the new average value of each Gaussian mixture model cluster; n represents the number of collected driver data samples;
Figure GDA0002908885600000156
representing the probability that the common driving habit factor of the ith driver belongs to the kth Gaussian mixture model; fiRepresenting a common factor obtained after dimensionality reduction of the ith driver driving habit data;
Figure GDA0002908885600000157
representing clusters of novelty per Gaussian mixture modelA covariance;
Figure GDA0002908885600000158
and representing the new weight value of the Gaussian model of each Gaussian mixture model cluster.
Then, let the average value of the initial cluster of the Gaussian mixture model
Figure GDA0002908885600000159
Covariance of initial cluster of Gaussian mixture model
Figure GDA00029088856000001510
Weight value of initial cluster of Gaussian mixture model
Figure GDA00029088856000001511
Substituting into formula (11), repeating expectation-maximization algorithm until the maximum likelihood function in formula (10) converges, i.e. identifying the number of three drivers, aggressive type, general type and conservative type.
The final recognition result is shown in fig. 2, in which conservative 38, generic 36, and aggressive 29 drivers are recognized. It is clear from fig. 2 that there are significant differences in the actual requirements of the driver for the driving factor, the braking factor and the environmental factor for the three different driving styles.
Step four, training a driving style identification model by adopting a back propagation neural network optimized by a genetic algorithm, and accurately outputting different driving styles;
traditional rule-based or model-based identification strategies rely on certain engineering experience or model parameters, and it is often difficult to identify the driving style with nonlinear characteristics with high precision. The back propagation neural network is used as a supervised learning algorithm, and is widely applied to pattern recognition and classification by virtue of the advantages of simple model structure, high identification precision, strong nonlinear mapping capability and the like. However, the identification precision of the back propagation neural network training result is closely related to the initial weight setting of the model. The reasonable initial weight can effectively accelerate the convergence speed of the training model and improve the identification precision of the model. Therefore, the initial weight of the back propagation neural network is optimized by adopting a genetic algorithm, and the specific flow is shown in fig. 3.
41) The fitness function of the genetic algorithm is set as:
Figure GDA0002908885600000161
in the formula, Fit represents a fitness function of a genetic algorithm; y represents the correct driver driving style type; w1Weights representing a back-propagation neural network hidden layer; w2Weights representing the back-propagation neural network output layer; b is1A threshold value representing a back-propagation neural network hidden layer; b is2A threshold value representing a back-propagation neural network output layer; f. of1An activation function representing a back-propagation neural network hidden layer; f. of2Representing an activation function of an output layer of the back propagation neural network, wherein the hidden layer activation function is set as a tansig function, and the activation function of the output layer is set as a pureline function; p represents a common factor of the model inputs.
42) And setting 50 initial populations of the genetic algorithm, and respectively encoding all weights and thresholds in the back propagation neural network into chromosome structures of the genetic algorithm. And then, the chromosome with the maximum fitness function value in the current generation population is completely replaced by the worst chromosome, and the weight and the threshold contained in the current generation chromosomes are mutually exchanged to generate a new chromosome. In addition, new chromosomes are obtained by randomly changing the chromosome structures of individual chromosomes, wherein the direction of variation towards the direction with the optimal population fitness can be expressed as:
Figure GDA0002908885600000162
in the formula, FitbestRepresents the maximum value of the fitness function in the current generation chromosome; fit represents a fitness function; x is the number oftRepresents the current generation of chromosomes randomly selected for mutation; x is the number oft+1Representing a new generation of chromosomes after mutation;
and then continuously repeating the processes of heredity, hybridization and mutation until the fitness function meets the constraint condition, decoding the chromosome reduction weight and the threshold value, and setting the chromosome reduction weight and the threshold value as the initial value of the back propagation neural network.
The application designs a back propagation neural network comprising a three-layer topological structure for identifying the driving style. The input layer comprises 3 neuron nodes which are respectively used as input layers of driving factors, braking factors and environmental factors; the hidden layer comprises 10 neuron nodes; the output layer outputs the identification result of the driving style, and only comprises 1 neuron node. The 103 sets of data obtained in step three were randomly divided into 73 training samples and 30 testing samples. The specific algorithm flow is as follows.
1. Initializing a back propagation neural network: the number of neurons in the input layer, hidden layer and output layer of the neural network is set to be n-3, m-10 and p-1. According to the optimization result of the genetic algorithm, respectively carrying out weighting on the weight omega of the hidden layer in the neural networkikAnd the weight omega of the output layerkjAnd threshold a of the hidden layeriAnd threshold b of output layerkRespectively carrying out assignment; the learning rate η of the neural network is set to 0.01.
2. Computing the output of the hidden layer and the output layer: setting the excitation function f of the hidden layer3For sigmoid function, the output of the hidden layer is:
Figure GDA0002908885600000171
in the formula (f)3An excitation function representing the hidden layer; omegaikRepresenting the weight of the hidden layer; a isiRepresenting a hidden layer threshold; x is the number ofiRepresenting a hidden layer input; hkRepresenting hidden layer output, and n represents the number of neurons in an input layer;
the excitation function of the output layer is set as a linear function, and then the output of the output layer is:
Figure GDA0002908885600000172
in the formula, ωkjRepresenting the weight of the output layer; bkA threshold value representing an output layer; o isjRepresenting the output of the output layer; m represents the number of hidden layer neurons; hkRepresenting a hidden layer output;
3. and (3) calculating an error: taking the driving style label obtained based on the Gaussian mixture model clustering algorithm in the step three as the expected output Y of the neural networkjThe error can be found as:
Figure GDA0002908885600000173
in the formula, E represents a neural network model training error; z represents the number of training samples; y isjRepresenting a neural network training label, namely a driver driving style identified based on a Gaussian mixture model; p represents the number of neurons of the output layer of the model input; o isjRepresenting the output of the output layer.
4. Updating the weight and the threshold value: according to the error obtained by calculation, the weight omega of the hidden layer in the neural network is subjected toikAnd the weight omega of the output layerkjUpdating:
Figure GDA0002908885600000181
in the formula, η represents a neural network learning rate; hkRepresenting a hidden layer output; x is the number ofiRepresenting a hidden layer input; e represents a neural network model training error; omegaikRepresenting the weight of the hidden layer; omegakjRepresenting the weight of the output layer; p represents the number of output layer neurons.
Similarly, the threshold value of the neural network is updated:
Figure GDA0002908885600000182
in the formula, aiRepresenting a hidden layer threshold; η represents the neural network learning rate; hkRepresentation implicationOutputting the layers; pi represents the number of neurons of the output layer; omegakjRepresenting the weight of the output layer; e represents a neural network model training error; bkA threshold value representing an output layer;
6. judging whether the constraint is met: if the calculation error meets the precision requirement, the training is finished, otherwise, the weight and the threshold are subjected to back propagation updating again, and the steps 2-6 are repeated.
Finally, through the four steps, a driving style identification model is obtained. Fig. 4 shows the recognition results of 30 sets of driving style test data. From the experimental curve, the fitness function value is gradually converged in the iterative process, and the identification accuracy of the three driving style styles respectively reaches 89.89%, 90.91% and 90.00%. Therefore, the driving style identification algorithm designed by the patent can accurately identify the driving styles of different drivers with higher precision.

Claims (3)

1. A driving style identification algorithm based on factor analysis and machine learning is characterized by comprising the following steps:
step one, extracting driving characteristic parameters strongly related to driving style from typical urban following working conditions;
step two, using a factor analysis algorithm including feasibility analysis, public factor determination and factor score to reduce the dimension and process the driving characteristic parameters, obtaining public factors and giving specific physical meanings to the public factors;
clustering the common factors by using a Gaussian mixture model, and accurately attaching a driving style label to each driver;
step four, training a driving style identification model by adopting a back propagation neural network optimized by a genetic algorithm, and accurately outputting different driving styles;
the driving characteristic parameters in the first step comprise: mean value A of the master cylinder pressure1Maximum value A of master cylinder pressure2Standard deviation of master cylinder pressure A3Average value A of engine torque4Standard deviation of engine torque A5Average engine speed A6Engine speedStandard deviation of A7Average value A of the distance between two vehicles8Standard deviation A of distance between two vehicles9Average value of two vehicle speed differences A10And standard deviation A of two vehicle speed differences11；
The specific method of the second step is as follows:
21) the driving data is normalized by a normalization method:
Figure FDA0002908885590000011
Figure FDA0002908885590000012
wherein N represents the number of collected driver data samples; x is the number ofkiI is 1,2, … 11 denotes the i-th driving style characteristic parameter value of the k-th driver; x is the number ofkjJ is 1,2, … 11 denotes the j-th driving style characteristic parameter value of the k-th driver;
Figure FDA0002908885590000013
representing the average value of ith driving characteristic parameters of N drivers;
Figure FDA0002908885590000014
representing the average value of j driving characteristic parameters of N drivers; x'kiAn ith normalized driving style characteristic parameter value representing a kth driver; x'kjA j-th normalized driving style characteristic parameter value representing a k-th driver;
22) the normalized feature data was subjected to correlation analysis:
Figure FDA0002908885590000021
in the formula, rijRepresenting elements in a matrix of correlation coefficients; x'kiAn ith normalized driving style characteristic parameter value representing a kth driver; n represents the number of collected driver data samples; x'kjA j-th normalized driving style characteristic parameter value representing a k-th driver;
23) solving the characteristic root and the corresponding characteristic vector of the related coefficient matrix according to the characteristic equation:
|λjI-Rij|＝0 (4)
RijUj＝λjUj (5)
in the formula, λjJ-1, 2, … 11 denotes a correlation coefficient matrix RijA characteristic root of; i represents a unit coefficient matrix; u shapejRepresenting a feature vector matrix corresponding to the feature root; rijRepresenting a matrix of correlation coefficients;
24) selection factor F3×1As a common factor for the driving data and analyzed by a factor load matrix; factor load matrix A11×3From the characteristic value λ3×1And a feature vector U11×3Obtaining;
Figure FDA0002908885590000022
in the formula, A11×3A representation factor load matrix; lambda [ alpha ]1，λ2，λ3Respectively represent a common factor F3×1Corresponding characteristic values; u shapei,jI 1,2, … 11, j 1,2,3 denotes a common factor F3×1A corresponding feature vector;
25) rotating the factor load matrix by adopting a maximum variance method, and naming the characteristic parameter A4-A7 as a driving factor; the characteristic parameter A1-A3 is named as a braking factor; the characteristic parameters A8-A11 are named as environmental factors;
26) the factor score model is:
Xm×1＝Am×3·F3×1+εm×1 (9)
in the formula, Xm×1M ═ 1,2,3 represents a factor score model; a. them×3A presentation factor score coefficient; epsilonm×1Representing a particular factor, take εm×1Solving for zero vector; f3×1Representing selected three common factors; solving by using an Anderson-Rubin regression method to obtain a factor score coefficient, and finally obtaining a common factor expression according to an analysis tree of the factor as follows:
Figure FDA0002908885590000031
in the formula, F1Represents a first common factor, i.e. a drive factor; f2Represents a second common factor, the braking factor; f3Represents a third common factor, namely an environmental factor; a. the1Represents the average value of the master cylinder pressure; a. the2Represents the maximum value of the master cylinder pressure; a. the3Represents the standard deviation of the master cylinder pressure; a. the4An average value representing engine torque; a. the5Representing a standard deviation of engine torque; a. the6Representing the average rotational speed of the engine; a. the7Indicating a standard deviation of the engine speed; a. the8Representing the average value of the distance between two vehicles; a. the9Representing the standard deviation of the distance between two vehicles; a. the10Represents the average of the two vehicle speed differences; a. the11The standard deviation of the two vehicle speed differences is indicated.
2. The driving style identification algorithm based on factor analysis and machine learning according to claim 1, wherein the specific method of the third step is as follows:
31) setting driving styles of a driver to be an aggressive type, a general type and a conservative type respectively;
32) carrying out probability calculation on the common factors by adopting a multidimensional Gaussian mixture model, and defining as follows:
Figure FDA0002908885590000032
Figure FDA0002908885590000033
in the formula, FiRepresenting a common factor obtained after dimensionality reduction of the ith driver driving habit data; p (F)i) Common factor F representing the ith driveriA characteristic probability density function of (a); k represents a cluster of a gaussian mixture model, and includes three types of driving styles, so that K is set to 3; mu.skRepresenting the average value of each cluster; sigmakRepresenting the covariance of each cluster; p (F)i|k)＝N(Fi|μk,σk) A probability density function representing a kth gaussian model; pikIs a weight of the kth Gaussian model, and
Figure FDA0002908885590000034
the goal of gaussian mixture model clustering is to find a suitable gaussian distribution such that the probability of clustering approaches maximum; this target value is set as the maximum likelihood function:
Figure FDA0002908885590000041
in the formula, pikIs the weight of the kth gaussian model; n represents the number of collected driver data samples; fiRepresenting a common factor obtained after dimensionality reduction of the ith driver driving habit data; mu.skRepresenting the average value of each cluster; sigmakRepresenting the covariance of each cluster;
solving a maximum likelihood function by adopting an Expectation-Maximization (Expectation-Maximization) algorithm; at the desired step, we set the mean μ of the initial cluster of the Gaussian mixture modelintCovariance σintAnd Gaussian model weight value piintAnd based thereon calculating the likelihood that each common factor belongs to the kth Gaussian mixture model:
Figure FDA0002908885590000042
in the formula (I), the compound is shown in the specification,
Figure FDA0002908885590000043
representing the probability that the common driving habit factor of the ith driver belongs to the kth Gaussian mixture model; pikAnd pijAre the weights of the kth and jth Gaussian models, respectively, and
Figure FDA0002908885590000044
Figure FDA0002908885590000045
n represents the number of collected driver data samples; fiRepresenting a common factor obtained after dimensionality reduction of the ith driver driving habit data; mu.skAnd mujRespectively representing the average values of the kth clustering cluster and the jth clustering cluster; sigmakAnd σjRespectively representing the covariance of the kth clustering cluster and the jth clustering cluster;
at the maximum step, derived from the desired step
Figure FDA0002908885590000046
Solving a new average value, a new covariance and a new Gaussian model weight value of each Gaussian mixture model cluster:
then, let the average value of the initial cluster of the Gaussian mixture model
Figure FDA0002908885590000047
Covariance of initial cluster of Gaussian mixture model
Figure FDA0002908885590000048
Weight value of initial cluster of Gaussian mixture model
Figure FDA0002908885590000049
Substituting into equation (11), repeating the expectation-maximization algorithm until the maximum likelihood function in equation (13)Convergence is the identification of the number of aggressive, normal and conservative drivers.
3. The driving style identification algorithm based on factor analysis and machine learning according to claim 1, wherein the specific method of the fourth step is as follows:
41) the fitness function of the genetic algorithm is set as:
Figure FDA0002908885590000051
in the formula, Fit represents a fitness function of a genetic algorithm; y represents the correct driver driving style type; w1Weights representing a back-propagation neural network hidden layer; w2Weights representing the back-propagation neural network output layer; b is1A threshold value representing a back-propagation neural network hidden layer; b is2A threshold value representing a back-propagation neural network output layer; f. of1An activation function representing a back-propagation neural network hidden layer; f. of2Representing an activation function of an output layer of the back propagation neural network, wherein the hidden layer activation function is set as a tansig function, and the activation function of the output layer is set as a pureline function; p represents a common factor of the model inputs;
42) setting 50 initial populations of the genetic algorithm, and respectively encoding all weights and thresholds in the back propagation neural network into chromosome structures of the genetic algorithm; then, the chromosome with the maximum fitness function value in the current generation population is completely replaced by the worst chromosome, and the weight and the threshold contained in the current generation chromosomes are mutually exchanged to generate a new chromosome; in addition, new chromosomes are obtained by randomly changing the chromosome structure of individual chromosomes, wherein the direction of variation is towards the direction of optimal population fitness and is represented as:
Figure FDA0002908885590000052
in the formula, FitbestRepresents the maximum value of the fitness function in the current generation chromosome; fit represents a fitness function; x is the number oftRepresents the current generation of chromosomes randomly selected for mutation; x is the number oft+1Representing a new generation of chromosomes after mutation;
and then continuously repeating the processes of heredity, hybridization and mutation until the fitness function meets the constraint condition, decoding the chromosome reduction weight and the threshold value, and setting the chromosome reduction weight and the threshold value as the initial value of the back propagation neural network.
CN202010893251.2A 2020-08-31 2020-08-31 Driving style identification algorithm based on factor analysis and machine learning Active CN112046489B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010893251.2A CN112046489B (en) 2020-08-31 2020-08-31 Driving style identification algorithm based on factor analysis and machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010893251.2A CN112046489B (en) 2020-08-31 2020-08-31 Driving style identification algorithm based on factor analysis and machine learning

Publications (2)

Publication Number Publication Date
CN112046489A CN112046489A (en) 2020-12-08
CN112046489B true CN112046489B (en) 2021-03-16

Family

ID=73607244

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010893251.2A Active CN112046489B (en) 2020-08-31 2020-08-31 Driving style identification algorithm based on factor analysis and machine learning

Country Status (1)

Country Link
CN (1) CN112046489B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112660140A (en) * 2020-12-28 2021-04-16 华北科技学院 Driving style classification method and device based on machine learning and electronic equipment
CN112651443A (en) * 2020-12-28 2021-04-13 华北科技学院 Driving style identification model evaluation method, device, medium and equipment based on machine learning
CN113022576B (en) * 2021-04-02 2022-02-22 东南大学 Driving style analysis method, system and device for different driving task stages
CN113392892A (en) * 2021-06-08 2021-09-14 重庆大学 Method and device for identifying driving habits of driver based on data fusion
CN116811894B (en) * 2023-08-30 2023-11-21 北京理工大学 Continuous driving style identification method, system and equipment

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102236783A (en) * 2010-04-29 2011-11-09 索尼公司 Method and equipment for detecting abnormal actions and method and equipment for generating detector
WO2013075005A1 (en) * 2011-11-16 2013-05-23 Flextronics Ap, Llc Configurable hardware unite for car systems
CN108216252A (en) * 2017-12-29 2018-06-29 中车工业研究院有限公司 A kind of subway driver vehicle carried driving behavior analysis method, car-mounted terminal and system
CN108269325A (en) * 2016-12-30 2018-07-10 中国移动通信有限公司研究院 A kind of analysis method and device of driving behavior oil consumption economy
CN108819948A (en) * 2018-06-25 2018-11-16 大连大学 Driving behavior modeling method based on reverse intensified learning
CN109635830A (en) * 2018-10-24 2019-04-16 吉林大学 For estimating the screening technique of the valid data of car mass
CN110321954A (en) * 2019-07-03 2019-10-11 中汽研（天津）汽车工程研究院有限公司 The driving style classification and recognition methods of suitable domestic people and system
CN111038485A (en) * 2019-12-30 2020-04-21 山东大学 Hybrid electric vehicle control method and system based on driving style recognition

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102236783A (en) * 2010-04-29 2011-11-09 索尼公司 Method and equipment for detecting abnormal actions and method and equipment for generating detector
WO2013075005A1 (en) * 2011-11-16 2013-05-23 Flextronics Ap, Llc Configurable hardware unite for car systems
CN108269325A (en) * 2016-12-30 2018-07-10 中国移动通信有限公司研究院 A kind of analysis method and device of driving behavior oil consumption economy
CN108216252A (en) * 2017-12-29 2018-06-29 中车工业研究院有限公司 A kind of subway driver vehicle carried driving behavior analysis method, car-mounted terminal and system
CN108819948A (en) * 2018-06-25 2018-11-16 大连大学 Driving behavior modeling method based on reverse intensified learning
CN109635830A (en) * 2018-10-24 2019-04-16 吉林大学 For estimating the screening technique of the valid data of car mass
CN110321954A (en) * 2019-07-03 2019-10-11 中汽研（天津）汽车工程研究院有限公司 The driving style classification and recognition methods of suitable domestic people and system
CN111038485A (en) * 2019-12-30 2020-04-21 山东大学 Hybrid electric vehicle control method and system based on driving style recognition

Also Published As

Publication number Publication date
CN112046489A (en) 2020-12-08

Similar Documents

Publication Publication Date Title
CN112046489B (en) Driving style identification algorithm based on factor analysis and machine learning
CN107958269B (en) Driving risk degree prediction method based on hidden Markov model
CN105741637B (en) Four-wheel hub motor electric car automated steering control method
CN111260118B (en) Vehicle networking traffic flow prediction method based on quantum particle swarm optimization strategy
CN111547064B (en) Driving style recognition and classification method for automobile adaptive cruise system
CN105034986B (en) A kind of driver's steering characteristic on-line identification method and device
CN110458337B (en) C-GRU-based network appointment vehicle supply and demand prediction method
CN110490275B (en) Driving behavior prediction method based on transfer learning
CN110442143B (en) Unmanned aerial vehicle situation data clustering method based on combined multi-target pigeon swarm optimization
CN111907342B (en) Working condition identification control method of pure electric vehicle
CN114170789B (en) Intelligent network link lane change decision modeling method based on space-time diagram neural network
CN111783826B (en) Driving style classification method based on pre-classification and ensemble learning
CN112149796A (en) Driving style identification method for optimizing BP neural network based on improved genetic algorithm
CN111746544B (en) Lane changing method for embodying individual behavior of driver
CN111783943A (en) Driver braking strength prediction method based on LSTM neural network
CN111907523B (en) Vehicle following optimizing control method based on fuzzy reasoning
CN111198501B (en) Method for determining fuel equivalent factor by RBF neural network
CN114912195B (en) Aerodynamic sequence optimization method for commercial vehicle
CN114030485B (en) Automatic driving automobile person lane change decision planning method considering attachment coefficient
CN115092165A (en) Clustering model-based driver style identification method under different cycle conditions
Zhang et al. Driving style classification for vehicle-following with unlabeled naturalistic driving data
Abonyi et al. Compact TS-fuzzy models through clustering and OLS plus FIS model reduction
Wang et al. A data-driven estimation of driving style using deep clustering
CN115691817A (en) LncRNA-disease association prediction method based on fusion neural network
CN114162132B (en) Driving mode identification method based on subjective and objective evaluation

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