CN115774942A - Driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM - Google Patents
Driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM Download PDFInfo
- Publication number
- CN115774942A CN115774942A CN202211612078.XA CN202211612078A CN115774942A CN 115774942 A CN115774942 A CN 115774942A CN 202211612078 A CN202211612078 A CN 202211612078A CN 115774942 A CN115774942 A CN 115774942A
- Authority
- CN
- China
- Prior art keywords
- data
- training
- svm
- vehicle
- driving
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM, which mainly utilizes a large amount of actual vehicle driving data of an enterprise to carry out data analysis, acquires state information of a driver when driving a vehicle from a large amount of user driving data, and seeks behavior rules expressed by the driver in the vehicle driving process so as to develop a driving behavior comprehensive evaluation system of the driver by combining technologies such as data mining, machine learning, deep learning and the like in the following process.
Description
Technical Field
The invention belongs to the technical field of intelligent networking automobile data analysis, and particularly relates to a driving style identification model modeling and statistical method based on real automobile data and SVM of a networking automobile.
Background
With the continuous development of intelligent networked automobile technologies of various domestic large automobile enterprises, the enterprise Internet of vehicles data is also growing explosively, and how to utilize the user Internet of vehicles information to dig out valuable information is of great significance for improving the updating and upgrading of enterprise vehicle products, the safe trip behavior of users and the driving experience of the users to vehicles.
The driving style refers to a relatively stable behavior characteristic exhibited when the driver manipulates the vehicle, is a relatively stable state exhibited in terms of driving behavior, is different among the driver groups, and reflects the conscious choice of the driver. Statistical analysis shows that the aggressive driving style driver has the largest standard deviation of the speed of the vehicle, more obvious fluctuation of the vehicle speed, the highest maximum values of the deceleration and the acceleration, more conservative variation of the acceleration and the deceleration, and more obvious comparison with the common type, and has the situations of rapid acceleration and rapid deceleration. If the driver drives on the road in an aggressive driving style for a long time, the probability of road traffic accidents is higher than that of the normal type and the conservative type. The vehicle enterprises and the fleet managers can regularly analyze the driving style of the driver, each vehicle manufacturer can be used as a driver reference factor when product function upgrading or internal list screening is carried out by using an over-the-air (OTA) technology, and meanwhile, the design of vehicle individual control strategies is facilitated by researching the driving style of the driver, the control strategies corresponding to different driving styles are formulated to realize the function requirement of the driver, and the conversion from a human-adapted vehicle to a vehicle-adapted vehicle is realized, for example, in the aspects of electric power steering, brake-by-wire, lane Keeping Assist (LKA), adaptive cruise (ACC), automatic lane changing, automatic emergency brake control, hybrid electric vehicle energy management, battery Soc prediction and the like, so that the driving experience and feeling of different types of drivers are met, and the driving comfort is improved.
Therefore, in the process of function development of an automobile remote monitoring management system and design of a vehicle controller control strategy, style type judgment and type statistics are carried out on driving data of a driver, and information of the driving style of the driver is introduced into the monitoring management system and the vehicle control strategy, so that the vehicle driving safety is improved, and a personalized vehicle control system is formulated. Therefore, the SVM-based method for identifying the driving style of the driver is particularly important.
The information disclosed in this background section is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person skilled in the art.
Disclosure of Invention
The invention aims to provide a driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM, so as to overcome the defects in the prior art.
In order to achieve the purpose, the invention provides a driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM, which specifically comprises the following steps:
the method comprises the following steps that S1, effective driving history data of a vehicle in a certain specific day for 24 hours are obtained from an enterprise Internet of vehicles service cloud platform database, the driving history data are data information which can be collected by hardware equipment on the existing vehicle network and comprise data collection time, vehicle speed, accumulated mileage, longitude, latitude, atmospheric pressure, engine information, motor information and battery information, and the data collection time and the vehicle driving speed are mainly required to be collected;
s2, processing the historical driving data to form a vehicle driving data sample suitable for an SVM classification algorithm model;
s3, preparing a training data set of a Support Vector Machine (SVM) classification recognition model, collecting cycle working condition data of a road in a certain area by using an existing vehicle model of IVECO, and processing to form an initial data set suitable for an unsupervised learning algorithm;
s4, after the initial data set is learned through an unsupervised learning algorithm, analyzing and determining driving style type labels of all sample data in the initial data set to form a classification recognition model training data set;
s5, training an SVM classification recognition model by using a data set with a driving style class label, adjusting the hyper-parameters of the model, and optimizing the prediction accuracy of the adjusted model;
and S6, recognizing the vehicle driving data sample by using the trained SVM driving style recognition model, and counting the driving style types of the current day.
Preferably, in the above technical solution, the step S2 of processing the historical travel data includes:
s2.1 represents the Data acquired in step (1) as Data = [ T Veh _ v =]Wherein the vehicle data acquisition time T = [ T ] 1 t 2 … t n ]Acquiring primary vehicle state information with the acquisition frequency of 1HZ; vehicle running state information: running vehicle speed Veh _ v = [ v ] 1 v 2 … v n ]；
S2.2, processing the running Data to be identified in blocks, and carrying out windowing division on the running Data at a fixed running time t from a time point when the first speed is not 0 to obtain a running Data working condition block;
s2.3, obtaining acceleration information from the speed information difference value of each working condition block;
s2.4, calculating characteristic parameter information of each working condition block, wherein the characteristic parameter information is 9, and comprises a speed standard deviation, an average acceleration, a maximum acceleration, an average deceleration, a maximum deceleration, an acceleration standard deviation, a deceleration standard deviation, an acceleration time proportion and a deceleration time proportion, and all the characteristic parameters are characteristic variables derived around the speed and the acceleration;
s2.5, the characteristic parameter information calculation formula is as follows:
standard deviation of speed:
average acceleration:
maximum acceleration: a is a max ＝max(a 1 ,a 2 ,a 3 ,a 4 ,...,a n )；
Average deceleration:
maximum deceleration rate: d is a radical of max ＝max(d 1 ,d 2 ,d 3 ,d 4 ,...,d n )；
Acceleration standard deviation:
deceleration standard deviation:
acceleration time ratio:
deceleration time proportion:
s2.6 all the condition blocks calculate 9 characteristic parameters in (2.5), and then represent:
Pre_Data＝[v m a av a max d av d max a m d m a p d p ]。
preferably, in the above technical solution, the step of preparing training data of the SVM classification model in step S3 includes:
s3.1, collecting road working condition data of a certain region as initial model training data, wherein the collection frequency of vehicle equipment is 1HZ;
s3.2, acquiring data of a certain area road, wherein the certain area road is an area with a certain topographic trend and an area where a vehicle runs for a long time in the actual production process;
s3.3, performing windowing processing on the model training Data in the step (2), further performing composite division, increasing the number of Data set samples, reducing the accident contingency and randomness of the training Data, and obtaining the training Data Training ；
S3.4, unsupervised learning, acquiring real labels of training samples, and carrying out normalization processing on the training samples, wherein the calculation formula is as follows:
s3.5, performing Data dimension reduction on the Data subjected to normalization processing in the step (3.4), selecting a proper method to perform Data reduction on the Data to reduce the study complexity of the model, and selecting a principal component analysis method to perform Data dimension reduction to acquire dimension reduction Data Dimensionality Reduction ；
S3.6, performing cluster analysis on the dimensionality-reduced Data, finding out category labels in the Data, dividing driving styles into three categories, namely three driving styles of an aggressive type, a common type and a conservative type, comprehensively considering Data samples and classification tasks, performing cluster analysis by adopting a K-Means + + clustering algorithm, wherein the initial clustering categories are three, obtaining a real label of each Data sample, and returning the Data labels to the step S3.3 of training Data Training And each sample in the training data corresponds to a driving style label.
Preferably, in the above technical solution, the step of analyzing the driving style type label in the training data set in step S4 includes:
s4.1 clustering analysis is performed in step S3.6, and the label returns to the training Data Training Classifying the returned data labels, and respectively calculating an average value of 9 characteristic parameters in each type of data to serve as a final clustering center of each type;
and S4.2, analyzing and comparing the three clustering center data, and endowing the corresponding type to each data label through analysis and comparison.
Preferably, in the above technical solution, the training of the SVM classification recognition model in step S5 includes:
s5.1 Pair Data Training Dividing a data set, wherein 80% of data samples in the data set are used as a training set, and 20% of data samples are used as a verification set;
s5.2, training and learning the SVM machine learning model by using a training set, calling a Sklearn library by using a Python3 language, training the model by using a Sklearn.svm module, and identifying by using three SVM machine learning models, wherein a pair of multi-mode (one rest) is required due to the fact that three driving styles relate to multi-classification, namely, the decisionfussion = 'ovr', and finally, a decision is made on an identification result according to the result of each SVM machine learning model, and the MSE mean square error is used as a standard in the evaluation method;
s5.3, further adjusting the super parameters of the SVM by adopting methods such as grid search, three-fold cross validation and the like, wherein the parameters to be adjusted have a penalty coefficient C of a target function, a kernel function kernel and a coefficient gamma of the kernel function;
s5.4 after parameter adjustment, the optimal model parameters are as follows: { 'C':1.793, 'kernel': 'linear' }, the model training accuracy is 97.1%, the verification set verification accuracy is 98.2%, the verification accuracy is higher than the training accuracy, and the method can be considered to have stronger generalization capability, so the method has higher accuracy and strong generalization capability.
Preferably, in the above technical solution, in step S6, the applying step of the SVM driving style recognition model includes:
s6.1, acquiring Pre _ Data in the step S2, identifying by using the trained SVM driving style identification model, and returning a Data label;
and S6.2, mapping the returned data labels to corresponding driving style types, carrying out statistical analysis on the driving style types, obtaining proportion data of various types, and showing that the driver tends to a certain driving style type in a period of time through the proportion of the types.
Compared with the prior art, the invention has the following beneficial effects:
the method analyzes the driving data of the vehicle user, periodically identifies and counts the driving style of the user, provides data as reference for realizing the safety analysis of the driving behavior of the user, the upgrading and internal measurement of the vehicle function and the personalized control strategy, fully utilizes the conditions of the existing vehicle-mounted equipment T-box, the Internet of vehicles cloud platform, a database, a vehicle-mounted controller and the like, and can identify and count the driving style of all vehicles accessed into the enterprise Internet of vehicles cloud platform through a set of identification and counting methods.
Drawings
FIG. 1 is a flow chart illustrating a SVM-based driver driving style identification and statistics method of the present invention;
FIG. 2 is a flow chart illustrating the processing of historical travel data in accordance with the present invention;
FIG. 3 is a schematic diagram of the complex partitioning of the condition blocks of the training data set according to the present invention;
FIG. 4 is a flow chart illustrating the analysis of driving style type labels of a training data set in accordance with the present invention;
FIG. 5 is a graph of principal component analysis contribution rate and cumulative contribution rate of PCA in accordance with the present invention;
FIG. 6 is a graph of SSE as a function of cluster class K, as described herein;
FIG. 7 is a graph of shading area and mean profile coefficient for K =3 according to the present invention;
FIG. 8 is a graph of the results of a numerical analysis of the cluster centers as described herein;
FIG. 9 is a flowchart illustrating SVM classification model training according to the present invention;
FIG. 10 is a flow chart illustrating the driving style recognition recording and statistics of the present invention;
FIG. 11 is a sample driving style recognition result of the present invention;
FIG. 12 illustrates sample driving style fractions for each type of driving style in accordance with an exemplary embodiment of the present invention.
Detailed Description
The following detailed description of specific embodiments of the invention is provided, but it should be understood that the scope of the invention is not limited to the specific embodiments.
Throughout the specification and claims, unless explicitly stated otherwise, the word "comprise", or variations such as "comprises" or "comprising", will be understood to imply the inclusion of a stated element or component but not the exclusion of any other element or component.
As shown in fig. 1, the method for identifying and counting the driving style of a driver based on an SVM specifically includes the following steps:
(1) Obtaining effective driving history data of a vehicle in 24 hours in a certain specific day from an enterprise Internet of vehicles service cloud platform database, wherein the driving history data comprises data acquisition time, vehicle driving speed and longitude and latitude position information;
(2) Processing the historical driving data to form a vehicle driving data sample suitable for an SVM classification algorithm model;
(3) Preparing a Support Vector Machine (SVM) classification recognition model training data set, and processing the data to form an initial data set suitable for an unsupervised learning algorithm by acquiring roads in a certain area or adopting vehicle standard driving cycle working condition data;
(4) After the initial data set is subjected to unsupervised learning algorithm learning, analyzing and determining driving style type labels of all sample data in the initial data set to form a classification recognition model training data set;
(5) Training an SVM classification recognition model by using a data set with driving style class labels, and optimizing the accuracy rate after the super-parameter adjustment;
(6) And recognizing the vehicle driving data sample by using a trained Support Vector Machine (SVM) driving style recognition model, and counting the driving style types of the current day.
As shown in fig. 2, the step (2) of processing the historical travel data includes:
(2.1) representing the Data acquired in step (1) as Data = [ T Veh _ v =]Wherein vehicle data acquisition time T = [ T ] 1 t 2 … t n ]Collecting vehicle state information once at intervals of 1 s;
vehicle running state information: running vehicle speed Veh _ v = [ v ] 1 v 2 … v n ]；
(2.2) carrying out block processing on the Data of the running Data to be identified, and carrying out windowed division on the Data of the running Data from a time point at which the first speed is not 0 and by using a fixed running time t to obtain a running Data working condition block;
(2.3) solving acceleration information for the speed information difference value of each working condition block;
(2.4) calculating characteristic parameter information of each working condition block, wherein the characteristic parameter information comprises a speed standard deviation, an average acceleration, a maximum acceleration, an average deceleration, a maximum deceleration, an acceleration standard deviation, a deceleration standard deviation, an acceleration time proportion and a deceleration time proportion;
(2.5) the characteristic parameter information calculation formula:
standard deviation of speed:
average acceleration:
maximum acceleration: a is a max ＝max(a 1 ,a 2 ,a 3 ,a 4 ,...,a n )；
Average deceleration:
maximum decelerationDegree: d is a radical of max ＝max(d 1 ,d 2 ,d 3 ,d 4 ,...,d n )；
Acceleration standard deviation:
deceleration standard deviation:
acceleration time ratio:
deceleration time proportion:
(2.6) all the condition blocks calculate the 9 characteristic parameters in (2.5), and then are expressed as:
Pre_Data＝[v m a av a max d av d max a m d m a p d p ]。
as shown in fig. 4, the step (3) of preparing training data of the SVM classification model includes:
(3.1) collecting road working condition data of a certain region as initial model training data, wherein the collection frequency of vehicle equipment is 1HZ;
(3.2) the road in a certain area is an area with a certain topographic trend and an area where the vehicle runs for a long time in the actual production process, and data are acquired;
(3.3) performing windowing processing on the model training Data in the step (2), performing composite division as shown in fig. 3, increasing the number of Data set samples, and reducing the incident contingency and randomness of the training Data to obtain the training Data Training ；
(3.4) unsupervised learning, acquiring a real label of the training sample, and carrying out normalization processing on the training sample, wherein the calculation formula is as follows:
(3.5) performing Data dimensionality reduction on the Data subjected to normalization processing in the step (3.4), selecting a proper method to perform Data dimensionality reduction on the Data, so as to reduce the research complexity of the model, and selecting a Principal Component Analysis (PCA) method to perform Data dimensionality reduction, wherein as shown in FIG. 5, the information represented by the initial Data set can be expressed by generally considering that the first n principal components with the cumulative contribution rate of more than 85%, and for this reason, selecting the first four principal components with the cumulative contribution rate of more than 85% to represent the initial training Data to obtain the dimensionality reduction Data Dimensionality Reduction ；
(3.6) performing cluster analysis on the dimension reduced Data to find out class labels in the Data, as shown in fig. 6 and 7, according to the SSE and the cluster contour coefficient curve, it can be seen that when the cluster class K =3, the SSE curve has an inflection point, K =3 is the optimal cluster class number by using the elbow rule, and the area shadow of the contour coefficient for each class exceeds the dashed line of the sample average contour coefficient, so the cluster number is reasonable, dividing the driving style into three categories, namely an aggressive driving style, a normal driving style and a conservative driving style, comprehensively considering the Data samples and the classification tasks, adopting a K-Means + + clustering algorithm to perform clustering analysis, wherein the initial clustering category is three categories, obtaining a real label of each Data sample, and returning the Data labels to the step (3.3) of training Data Training And each sample in the training data corresponds to a driving style label.
As shown in fig. 4, the step (4) of analyzing the driving style type label in the training data set includes:
(4.1) returning the label to the training Data after the clustering analysis in the step (3.7) Training In the method, returned data labels are classified, an average value of 9 characteristic parameters in each class of data is obtained to be used as a final clustering center, as shown in fig. 8, the obtained clustering center is used for mapping corresponding driving style type labels after numerical analysis of the characteristic parameters, and for the obtained clustering center, the corresponding driving style type labels are mappedThe standard deviation of the speed of the aggressive driver is 18.883 which is the largest of the other two types, and the parameters such as the maximum acceleration, the maximum deceleration, the average acceleration, the average deceleration, the maximum deceleration and the like are the largest compared with the other two types, the conservative type is opposite to the conservative type, and the normal type is between the two types;
and (4.2) analyzing and comparing the three clustering center data, and assigning a corresponding type to each data label through analysis and comparison.
As shown in fig. 9, the training of the SVM classification recognition model in step (5) includes:
(5.1) for Data Training Dividing a data set, wherein 80% of data samples in the data set are used as a training set, and 20% of data samples are used as a verification set;
(5.2) training and learning the SVM machine learning model by using a training set, calling a Sklearn library by using a Python3 language, training the model by using a Sklearn.svm module, and identifying by using three SVM machine learning models, wherein a pair of multi-mode (one rest) is required because three driving styles relate to multi-classification, namely, the decisionfundamental = 'ovr', and finally, the identification result is decided according to the result of each SVM machine learning model, and the MSE mean square error is used as a standard in the evaluation method;
and (5.3) further adjusting the super parameters of the SVM by adopting methods such as grid search, three-fold cross validation and the like, wherein the parameters to be adjusted have a penalty coefficient C of a target function, a kernel function kernel and a coefficient gamma of the kernel function.
(5.4) after parameter adjustment, the optimal model parameters are as follows: { 'C':1.793, 'kernel': 'linear' }, the model training accuracy is 97.1%, the prediction accuracy is 98.2%, and the method has high accuracy and strong generalization capability.
As shown in fig. 10, the SVM driving style recognition model applying step includes:
(6.1) acquiring Pre _ Data Data in the step (2), recognizing by using the trained SVM driving style recognition model, and returning a Data label, such as a label returning result example shown in FIG. 11;
(6.2) mapping the returned data labels to corresponding driving style types, performing statistical analysis on the driving style types, and acquiring proportion data of each type, as shown in fig. 12, showing that the driver tends to a certain driving style type for a period of time through the proportion of the types.
When the system is applied to specific production deployment, the system can be operated in a user terminal or a cloud platform server, the application deployment is flexible, the system can be applied to various service frameworks to serve as a simple and efficient analysis and identification method tool, real-time analysis and identification are carried out by calling an enterprise internal vehicle networking service cloud platform database or a vehicle terminal, and an identification result is transmitted to a vehicle networking remote monitoring service cloud platform or a vehicle controller, so that the driving style of a driver is analyzed and tracked, the driving safety of the driver is guaranteed, or an individualized control strategy is executed as key information, and the driving experience of the driver is improved.
Generally speaking, the invention provides a simple, efficient and generalized algorithm recognition model which is strong in generalization capability and easy to implement, a large amount of actual vehicle driving data of an enterprise are utilized to carry out data analysis, state information of a driver when the driver drives the vehicle is obtained from a large amount of user driving data, a behavior law shown by the driver in the vehicle driving process is searched, so that a comprehensive evaluation system for the driving behavior of the driver is developed by combining technologies such as data mining, machine learning and deep learning, risk management measures such as safety education recommendation items and corresponding UBI vehicle insurance schemes can be formulated for the identified bad and non-driving drivers, and meanwhile, an individual control system is developed for vehicle functions according to different driving style requirements to meet different driver requirements and optimize the vehicle driving experience.
The above-described embodiments are intended to illustrate rather than to limit the invention, and all such modifications and variations are within the spirit and scope of the invention.
The foregoing descriptions of specific exemplary embodiments of the present invention have been presented for purposes of illustration and description. It is not intended to limit the invention to the precise form disclosed, and obviously many modifications and variations are possible in light of the above teaching. The exemplary embodiments were chosen and described in order to explain certain principles of the invention and its practical application to enable one skilled in the art to make and use various exemplary embodiments of the invention and various alternatives and modifications as are suited to the particular use contemplated. It is intended that the scope of the invention be defined by the claims and their equivalents.
Claims (6)
1. A driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM is characterized by comprising the following steps:
the method comprises the following steps that S1, effective driving history data of a vehicle in a certain specific day for 24 hours are obtained from an enterprise Internet of vehicles service cloud platform database, the driving history data are data information which can be collected by hardware equipment on the existing vehicle network and comprise data collection time, vehicle speed, accumulated mileage, longitude, latitude, atmospheric pressure, engine information, motor information and battery information, and the data collection time and the vehicle driving speed are mainly required to be collected;
s2, processing the historical driving data to form a vehicle driving data sample suitable for an SVM classification algorithm model;
s3, preparing a training data set of a Support Vector Machine (SVM) classification recognition model, collecting cycle working condition data of a road in a certain area by using an existing vehicle model of IVECO, and processing to form an initial data set suitable for an unsupervised learning algorithm;
s4, after the initial data set is learned through an unsupervised learning algorithm, analyzing and determining driving style type labels of all sample data in the initial data set to form a classification recognition model training data set;
s5, training an SVM classification recognition model by using a data set with a driving style class label, adjusting the hyper-parameters of the model, and optimizing the prediction accuracy of the adjusted model;
and S6, recognizing the vehicle driving data sample by using the trained SVM driving style recognition model, and counting the driving style type of the current day.
2. The driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM according to claim 1, characterized in that:
the step S2 of processing the historical travel data includes:
s2.1 represents the Data acquired in step (1) as Data = [ T Veh _ v =]Wherein the vehicle data acquisition time T = [ T ] 1 t 2 … t n ]Acquiring primary vehicle state information with the acquisition frequency of 1HZ; vehicle running state information: running vehicle speed Veh _ v = [ v ] 1 v 2 … v n ]；
S2.2, performing block processing on the Data of the traveling Data to be identified, and performing windowed division on the Data of the traveling Data from a time point at which the first speed is not 0 by using a fixed traveling time t to obtain a working condition block of the traveling Data;
s2.3, obtaining acceleration information from the speed information difference value of each working condition block;
s2.4, calculating characteristic parameter information of each working condition block, wherein the characteristic parameter information is 9, and comprises a speed standard deviation, an average acceleration, a maximum acceleration, an average deceleration, a maximum deceleration, an acceleration standard deviation, a deceleration standard deviation, an acceleration time proportion and a deceleration time proportion, and all the characteristic parameters are characteristic variables derived around the speed and the acceleration;
s2.5, the characteristic parameter information calculation formula is as follows:
standard deviation of speed:
average acceleration:
maximum acceleration: a is max ＝max(a 1 ,a 2 ,a 3 ,a 4 ,...,a n )；
Average deceleration:
maximum deceleration: d max ＝max(d 1 ,d 2 ,d 3 ,d 4 ,...,d n )；
Acceleration standard deviation:
deceleration standard deviation:
acceleration time ratio:
deceleration time proportion:
s2.6 all the working condition blocks calculate 9 characteristic parameters in the (2.5), and then are expressed as:
Pre_Data＝[v m a av a max d av d max a m d m a p d p ]。
3. the driving style identification model modeling and statistical method based on Internet of vehicles data and SVM of claim 1, characterized in that: the SVM classification model training data preparation step in the step S3 comprises the following steps:
s3.1, collecting road working condition data of a certain region as initial model training data, wherein the collection frequency of vehicle equipment is 1HZ;
s3.2, acquiring data of a certain area road, wherein the certain area road is an area with a certain topographic trend and an area where a vehicle runs for a long time in the actual production process;
s3.3, performing windowing processing on the model training data in the step (2),and further performing compound division, increasing the number of Data set samples, and reducing the accident contingency and randomness of the training Data to obtain the training Data Training ；
S3.4, unsupervised learning, acquiring a real label of the training sample, and carrying out normalization processing on the training sample, wherein the calculation formula is as follows:
s3.5, performing Data dimensionality reduction on the Data subjected to normalization processing in the step (3.4), selecting a proper future-oriented method to perform Data dimensionality reduction on the Data, reducing the research complexity of the model, selecting a principal component analysis method to perform Data dimensionality reduction, and acquiring dimensionality reduction Data Dimensionality Reduction ；
S3.6, performing cluster analysis on the dimensionality-reduced Data, finding out category labels in the Data, dividing driving styles into three categories, namely three driving styles of an aggressive type, a common type and a conservative type, comprehensively considering Data samples and classification tasks, performing cluster analysis by adopting a K-Means + + clustering algorithm, wherein the initial clustering categories are three, obtaining a real label of each Data sample, and returning the Data labels to the step S3.3 of training Data Training And each sample in the training data corresponds to a driving style label.
4. The driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM according to claim 1, characterized in that: the step S4 of analyzing the driving style type labels in the training data set comprises the following steps:
s4.1 clustering analysis of step S3.6 and returning of labels to training Data Training Classifying the returned data labels, and respectively calculating an average value of 9 characteristic parameters in each type of data to serve as a final clustering center of each type;
and S4.2, analyzing and comparing the three clustering center data, and endowing the corresponding type to each data label through analysis and comparison.
5. The driving style identification model modeling and statistical method based on Internet of vehicles data and SVM of claim 1, characterized in that: the training of the SVM classification recognition model in the step S5 comprises the following steps:
s5.1 Pair of Data Training Dividing a data set, wherein 80% of data samples in the data set are used as a training set, and 20% of data samples are used as a verification set;
s5.2, training and learning the SVM machine learning model by using a training set, calling a Sklearn library by using a Python3 language, training the model by using a Sklearn.svm module, and identifying by using three SVM machine learning models, wherein a pair of multi-mode (one rest) is required due to the fact that three driving styles relate to multi-classification, namely, the decisionfussion = 'ovr', and finally, a decision is made on an identification result according to the result of each SVM machine learning model, and the MSE mean square error is used as a standard in the evaluation method;
s5.3, further adjusting the super parameters of the SVM by adopting methods such as grid search, three-fold cross validation and the like, wherein the parameters to be adjusted have a penalty coefficient C of a target function, a kernel function kernel and a coefficient gamma of the kernel function;
s5.4 after parameter adjustment, the optimal model parameters are as follows: { ' C ':1.793, ' kernel ': linear ' }, the model training accuracy is 97.1%, and the prediction accuracy is 98.2%.
6. The driving style identification model modeling and statistical method based on Internet of vehicles data and SVM of claim 1, characterized in that: in step S6, the applying step of the SVM driving style recognition model includes:
s6.1, acquiring Pre _ Data in the step S2, identifying by using the trained SVM driving style identification model, and returning a Data label;
and S6.2, mapping the returned data labels to corresponding driving style types, carrying out statistical analysis on the driving style types, obtaining proportion data of various types, and showing that the driver tends to a certain driving style type in a period of time through the proportion of the types.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211612078.XA CN115774942A (en) | 2022-12-13 | 2022-12-13 | Driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211612078.XA CN115774942A (en) | 2022-12-13 | 2022-12-13 | Driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115774942A true CN115774942A (en) | 2023-03-10 |
Family
ID=85392327
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211612078.XA Pending CN115774942A (en) | 2022-12-13 | 2022-12-13 | Driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115774942A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116502142A (en) * | 2023-07-03 | 2023-07-28 | 北京航空航天大学 | Driving style identification method based on input characteristic parameter selection |
-
2022
- 2022-12-13 CN CN202211612078.XA patent/CN115774942A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116502142A (en) * | 2023-07-03 | 2023-07-28 | 北京航空航天大学 | Driving style identification method based on input characteristic parameter selection |
| CN116502142B (en) * | 2023-07-03 | 2023-08-25 | 北京航空航天大学 | Driving style identification method based on input characteristic parameter selection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhu et al. | Typical-driving-style-oriented personalized adaptive cruise control design based on human driving data | |
| Liu et al. | Research on classification and recognition of driving styles based on feature engineering | |
| CN109840612A (en) | User's driving behavior analysis method and system | |
| CN110949398A (en) | Method for detecting abnormal driving behavior of first-vehicle drivers in vehicle formation driving | |
| CN111783826B (en) | Driving style classification method based on pre-classification and ensemble learning | |
| CN113188807B (en) | Automatic abs result judging algorithm | |
| CN110395245B (en) | Hybrid electric vehicle energy management system based on fixed route driving information | |
| CN113222385B (en) | Method for constructing and evaluating driving condition of electric automobile | |
| CN114493191B (en) | Driving behavior modeling analysis method based on network about vehicle data | |
| CN111563555A (en) | Driver driving behavior analysis method and system | |
| CN113920732B (en) | Road traffic accident risk early warning method for specific driving crowd | |
| CN110858312A (en) | Driver driving style classification method based on fuzzy C-means clustering algorithm | |
| CN113581188A (en) | Commercial vehicle driver driving style identification method based on Internet of vehicles data | |
| CN114817371A (en) | XGboost-based vehicle driving condition identification method | |
| CN105808857A (en) | Prediction method of automobile active safety system effectiveness on the basis of collision deformation depth | |
| CN115774942A (en) | Driving style identification model modeling and statistical method based on Internet of vehicles real vehicle data and SVM | |
| CN113297795A (en) | Method for constructing running condition of pure electric vehicle | |
| Liang et al. | Comprehensive identification of driving style based on vehicle's driving cycle recognition | |
| CN112985830A (en) | Abs result automatic judging algorithm | |
| CN115186594A (en) | Energy-saving speed optimization method under influence of man-vehicle-road coupling | |
| Halim et al. | Deep neural network-based identification of driving risk utilizing driver dependent vehicle driving features: A scheme for critical infrastructure protection | |
| CN110254435B (en) | Driving style recognition method | |
| Chu et al. | A review of driving style recognition methods from short-term and long-term perspectives | |
| CN113657432A (en) | Commercial vehicle driving behavior risk level identification method based on Internet of vehicles data | |
| Lee et al. | Individual stable driving pattern analysis for evaluating driver readiness at autonomous driving levels 2 and 3 |
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 |