Intelligent driving style identification method based on natural driving data

Abstract

The invention discloses a driving style intelligent identification method based on natural driving data, and the method comprises the steps: obtaining vehicle track data according to the natural driving data, and dividing the vehicle track data into a following data set and a lane changing data set; driving style characterization parameters in a car following state and a lane changing state are obtained, and a driving style principal component score matrix is obtained after dimension reduction processing; constructing a driving style clustering model and a driving style identification model based on a support vector machine according to the driving style characterization parameters; judging driving styles in a car following state and a lane changing state through a driving style clustering model, and judging the style of a driver by integrating the two driving states; the driving style identification model is used for identifying the driving style of the driver in the following state and the lane changing state. According to the invention, a complete and systematic method can be provided for actively identifying the driving styles of the driver in different scenes, and the method has the advantages of being comprehensive, systematic, accurate in identification, stable and efficient.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent traffic safety, and in particular to a driving style intelligent recognition method based on natural driving data. Background Art

[0002] The surge in the number of motor vehicles has provided people with a good transportation experience, but it is also accompanied by problems such as traffic congestion and environmental pollution. Therefore, in order to ensure efficient, convenient, green and safe travel, the development of intelligent transportation, unmanned driving and intelligent networked vehicle technology have become the current development trend. Driving style research is closely related to road safety, environmental protection and autonomous driving. Driving style will affect the driver's driving behavior, including speeding, running red lights, forced lane changes, etc. By studying driving style, we can identify high-risk drivers and take appropriate education and training in advance to prevent traffic accidents; different driving styles have different effects on driving fuel consumption. By studying driving style, we can put forward more environmentally friendly and energy-saving driving operation suggestions, reduce driving energy consumption and reduce driving costs. The study of driving style can also help the autonomous driving system to more fully understand the driver's intentions and make more reasonable decisions; design an autonomous driving system that is more in line with the driver's expectations and improve the driver's experience; and provide corresponding early warning and intervention mechanisms to effectively prevent driving accidents.

[0003] During the entire driving process, different driving behaviors such as free driving, following and lane changing correspond to different driving states. For different driving states, drivers may show different driving styles. In order to more effectively reveal the driving style differences of different types of drivers and obtain a more comprehensive driving style evaluation, it is necessary to classify the driving style and comprehensively evaluate the driving style of the driving data under different driving states.

[0004] The classification of driving styles is currently mainly divided into two categories: subjective driving style evaluation methods based on questionnaires and objective driving style evaluation methods based on driving behavior characteristics. For example, driving styles are classified through driving style questionnaires (DSQ) or driving style scales; driving styles are judged based on fuzzy comprehensive evaluation models or unsupervised learning algorithms. On this basis, the characteristics of different driving style categories are learned, and the use of machine learning algorithms can realize the identification of driving styles. However, most of the current research focuses on a single driving scenario. Drivers will show different driving styles in different driving scenarios. The driving style in a specific scenario cannot represent their overall driving style tendency. At present, there is no specific and systematic driving style identification method for multiple driving scenarios. Summary of the invention

[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a method for intelligently identifying driving styles based on natural driving data, using natural driving data sets and vehicle trajectory data in the following state and lane changing state to classify driving styles in different scenarios and identify driving styles, thereby providing strong technical support for actively judging the driver's driving style, identifying high-risk drivers, and preventing traffic accidents. In order to achieve the above-mentioned purpose and other advantages according to the present invention, a method for intelligently identifying driving styles based on natural driving data is provided, comprising:

[0006] Acquire vehicle trajectory data according to natural driving data, and divide the vehicle trajectory data into a car-following data set and a lane-changing data set;

[0007] Obtain the driving style characterization parameters in the following state and the lane changing state, and obtain the driving style principal component score matrix after dimensionality reduction processing;

[0008] According to the driving style characterization parameters, a driving style clustering model and a driving style identification model based on support vector machine are constructed;

[0009] The driving style clustering model is used to determine the driving style in the following state and the lane changing state, and the driver's style is determined by combining the two driving states;

[0010] The driving style recognition model is used to identify the driver's driving style in a car-following state and a lane-changing state.

[0011] Preferably, the vehicle trajectory data includes vehicle ID, vehicle speed, vehicle horizontal and vertical coordinates, vehicle running time, vehicle headway and vehicle headway.

[0012] Preferably, establishing the car-following data set and the lane-changing data set according to the natural driving data specifically comprises the following steps:

[0013] Step 1.1, set the driving state extraction rules, extract the natural driving vehicle trajectory data into a car-following data set and a lane-changing data set, and use the Savitzky-Golay filtering algorithm for smoothing;

[0014] Step 1.2: The rules for dividing the following vehicle group are as follows: the maximum distance between the following vehicles is 150m, the minimum distance between the following vehicles is 5m, the maximum headway between the following vehicles is 6s, the minimum duration of following vehicles is 15s, and the target vehicle must have a preceding vehicle;

[0015] Step 1.3: The rules for dividing the lane-changing vehicle group are as follows: the total lane-changing state duration is 12 seconds, and the frame when the lane is changed is used as the judgment frame. It is considered that the first 60 frames and the last 60 frames of the frame, a total of 120 frames, or 12 seconds, together constitute the lane-changing state. In addition, the lane-changing vehicle must have a preceding vehicle and must not have stopped in between.

[0016] Preferably, the driving style principal component score matrix is ​​established by specifically comprising the following steps:

[0017] Step 2.1, selecting feature parameters that can fully characterize the driving style according to the characteristics of the following state and the lane changing state to form a high-dimensional driving style feature parameter set;

[0018] Step 2.2, using correlation analysis on the following parameters and the lane changing parameters, deleting some characteristic parameters with low correlation, and preliminarily reducing the dimension of the driving style characteristic parameters;

[0019] Step 2.3: Use principal component analysis to reduce the dimension. To retain the information in the original data as much as possible, retain all principal components that make the cumulative contribution rate of the principal components exceed 80% as the dimensionality reduction result of the feature parameters, and obtain the driving style principal component score matrix.

[0020] The driving style classification model is established based on the feature data after dimensionality reduction, including the following steps:

[0021] Step 3.1, constructing a following state feature matrix A and a lane-changing state feature matrix B according to the principal components of the driving style feature parameter set after dimensionality reduction;

[0022] Step 3.2, cluster analysis is performed on the characteristic matrix A of the following state and the characteristic matrix B of the lane changing state respectively. According to the clustering results, the SSE and the silhouette coefficient are calculated, and the data of the following state data set and the lane changing state data set are divided into i categories. It is considered that there are i driving styles in the following state and the lane changing state, which are set as the driving style labels of the i-category following state and the lane changing state;

[0023] Step 3.3: According to the vehicle ID, calculate the distances between the characteristic principal component data of the same vehicle and the cluster centers of the following cluster and the lane-changing cluster, sum the distances of each type of driving style, and based on the minimum distance principle, the minimum value among the i types of driving styles is the corresponding style of the driver.

[0024] Preferably, the driving style identification model is established according to the driving style main component scores after dimensionality reduction and the driving style labels obtained by the driving style classification model, comprising the following steps:

[0025] Step 4.1, according to the following state feature matrix A obtained in step 3.1, the feature data is divided into a training set C and a test set D, the feature data of the training set C and the corresponding driving style labels obtained in step 3.2 are used as input, a support vector machine classification model is established, the historical feature data and the corresponding driving style labels are learned, and the model parameters are optimized through grid search and cross-validation;

[0026] Step 4.2, perform the same processing on the lane-changing state characteristic matrix B obtained in step 3.1;

[0027] Step 4.3: After the training model is established, the model is tested using the test set D and compared with the recognition accuracy of the random forest and KNN algorithms. If the test value reaches the expected accuracy, the driving style recognition model is established. At the same time, by comparing with the recognition accuracy before dimensionality reduction, it can be verified that dimensionality reduction helps improve the recognition accuracy of the model.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] (1) The method of the present invention provides a systematic method, including four steps of data processing, feature extraction, driving style classification, and driving style identification, forming a complete driving style identification scheme. The method is based on universally applicable natural driving vehicle trajectory data, has strong adaptability and transferability, and can be applied to different vehicles and driving environments. At the same time, the natural driving data is real vehicle driving data, which makes the identification of driving style more consistent with actual driving conditions and improves the practicality of the model.

[0030] (2) The method of the present invention uses mature data processing technology and machine learning algorithms, such as principal component analysis, clustering algorithm, and support vector machine algorithm. By comprehensively using the multi-dimensional characteristics of the vehicle's operating status, the algorithm can more comprehensively understand and predict driving behavior. By applying dimensionality reduction techniques such as principal component analysis, the algorithm reduces the computational complexity while retaining the most important information, thereby improving the efficiency of data processing. Using a clustering algorithm to classify driving styles is simple and efficient, and is easy to explain and implement, thereby improving the accuracy, stability, and efficiency of driving style identification.

[0031] (3) The method of the present invention not only analyzes a single driving scenario, but also comprehensively evaluates the driver's performance in different driving scenarios, providing a more comprehensive evaluation of driving style. In addition, although only two driving states, following and lane changing, are considered, the possibility of expanding to other driving states such as free driving and turning is reserved, providing a comprehensive driving style analysis. By integrating data from different scenarios, the method helps to identify the driver's personalized driving habits and style, providing a basis for personalized driving training and assistance. At the same time, it maintains better generalization ability and can adapt to a wider range of application environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A schematic diagram of a process of a method for intelligently identifying a driving style based on natural driving data according to the present invention;

[0033] Figure 2A flowchart of driving style classification and establishment of a driving style identification model is provided for an embodiment of a driving style intelligent identification method based on natural driving data according to the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] The present invention provides a driving style intelligent identification method based on natural driving data in order to improve the early warning capability of high-risk drivers and thus reduce the accident rate. The method includes four parts: data processing, feature extraction, driving style clustering and driving style identification. The data processing part is used to extract the natural driving data into the following and lane-changing vehicle groups and implement smoothing processing; the feature extraction part is used to extract the driving style characterization parameters and obtain the driving style principal component score matrix after dimensionality reduction processing; the driving style clustering part is used to construct a driving style clustering model to judge the driving style in the following state and the lane-changing state respectively, and comprehensively judge the driver's style in the two driving states; the driving style identification part is used to construct a driving style identification model based on a support vector machine to identify the driver's driving style in the following state and the lane-changing state.

[0036] The method of the present invention makes full use of natural driving data to realize the classification and identification of driving styles, uses the vehicle operation characteristics in the following state and the lane changing state to construct a driving style feature parameter set, and after correlation analysis and principal component analysis dimensionality reduction, constructs a clustering-based driving style clustering model and a support vector machine-based driving style identification model to realize the classification and identification of driving styles in the following state and the lane changing state, more effectively reveal the driving style differences of different categories of drivers, help identify high-risk drivers, so that appropriate education and training can be taken in advance to prevent the occurrence of traffic accidents, and provide a complete solution for actively identifying the driving style of drivers. The detection has the characteristics of high accuracy, high stability and high efficiency, and at the same time, the cost is low, and it has strong adaptability and portability, and can be applied to different vehicles and driving environments.

[0037] Example 1

[0038] Reference Figure 1 , a driving style intelligent recognition method based on natural driving data, comprising the following steps:

[0039] Step 1: Use a public natural driving data set to obtain vehicle trajectory data, divide the following data set and the lane changing data set according to the driving state extraction conditions, and perform outlier and noise processing on the extracted trajectory data; specifically, in this embodiment, the vehicle trajectory data includes: vehicle ID, vehicle speed, vehicle horizontal and vertical coordinates, vehicle running time, headway, and headway.

[0040] Step 2: Optimize characteristic parameters characterizing the driving styles in the following state and the lane changing state through statistical analysis to obtain a high-dimensional driving style characteristic parameter set, and perform dimensionality reduction twice through correlation analysis and principal component analysis to obtain a driving style principal component score matrix; specifically, in this embodiment, the driving style characteristic parameters of the following state include: speed, acceleration, following time and corresponding driving distance, time ratio above the mean vehicle speed, headway, headway, impact degree, and number of sudden accelerations and decelerations; the driving style characteristic parameters of the lane changing state include: speed, acceleration, time ratio above the mean vehicle speed, headway, headway, impact degree, lateral speed, lateral acceleration, lateral impact degree, lane changing time and corresponding driving distance, heading angle, and lane changeable gap.

[0041] By establishing an empty feature data set, the following trajectory data and lane-changing trajectory data are analyzed respectively, and the driving style characteristic parameters are calculated according to the vehicle trajectory data such as vehicle speed, horizontal and vertical coordinates, and time, and the acceleration, following time and corresponding driving distance (lane-changing time and corresponding driving distance), impact degree, number of sudden accelerations (sudden decelerations), lateral speed, lateral acceleration, lateral impact degree, heading angle and other parameters are obtained to form the driving style characteristic parameter set. Through two dimensionality reductions of correlation analysis and principal component analysis, a lower-dimensional principal component score matrix can be obtained for subsequent modeling.

[0042] Step 3: On the premise of obtaining the driving style characterization parameters, the k-means clustering method is used to perform cluster analysis on the principal component score matrix after dimensionality reduction, and the categories of driving styles in the following state and lane changing state are determined by SSE and silhouette coefficient. Labels are set for subsequent modeling, and the driver's style is determined based on the minimum distance criterion;

[0043] Step 4: According to the driving style clustering results, the driving style principal component score matrix and the corresponding driving style category labels after dimensionality reduction are input into the driving style recognition model based on the support vector machine. The model recognition accuracy is obtained through grid search and cross-validation training parameters. The effectiveness of the support vector machine model is verified by comparing it with other algorithms (random forest, KNN). At the same time, by comparing it with the recognition accuracy before dimensionality reduction, it can be verified that dimensionality reduction improves the recognition accuracy.

[0044] Specifically, in this embodiment, the car-following data set and the lane-changing data set in step 1 are established according to natural driving data, including the following steps:

[0045] Step 1.1, set the driving state extraction rules, extract the natural driving vehicle trajectory data into the following vehicle group and the lane-changing vehicle group, and use the Savitzky-Golay filtering algorithm for smoothing;

[0046] Step 1.2: The rules for dividing the following vehicle groups are as follows: the maximum spacing between following vehicles is 150m, the minimum spacing between following vehicles is 5m, the maximum headway time between following vehicles is 6s, the minimum duration of following vehicles is 15s, and the target vehicle must be preceded by a preceding vehicle.

[0047] Step 1.3: The rules for dividing the lane-changing vehicle group are as follows: the total lane-changing state duration is 12 seconds, and the frame when the lane is changed is used as the judgment frame. It is considered that the first 60 frames and the last 60 frames of the frame, a total of 120 frames, or 12 seconds, together constitute the lane-changing state. In addition, the lane-changing vehicle must have a preceding vehicle and must not have stopped in between.

[0048] Specifically, in this embodiment, the principal component score matrix of step 2 is established according to the high-dimensional driving style feature parameter set, including the following steps:

[0049] Step 2.1, selecting feature parameters that can fully characterize the driving style according to the characteristics of the following state and the lane changing state to form a high-dimensional driving style feature parameter set;

[0050] Step 2.2, using correlation analysis on the following parameters and the lane changing parameters, deleting some characteristic parameters with low correlation, and preliminarily reducing the dimension of the driving style characteristic parameters;

[0051] Step 2.3: Use principal component analysis to reduce the dimension. To retain the information in the original data as much as possible, retain all principal components that make the cumulative contribution rate of the principal components exceed 80% as the dimensionality reduction result of the feature parameters, and obtain the driving style principal component score matrix.

[0052] Specifically, in this embodiment, the driving style classification model in step 3 is established based on the feature data after dimensionality reduction, including the following steps:

[0053] Step 3.1, constructing a following state feature matrix A and a lane-changing state feature matrix B according to the principal components of the driving style feature parameter set after dimensionality reduction;

[0054] Step 3.2, cluster analysis is performed on the characteristic matrix A of the following state and the characteristic matrix B of the lane changing state respectively. According to the clustering results, the SSE and the silhouette coefficient are calculated, and the data of the following state data set and the lane changing state data set are divided into i categories. It is considered that there are i driving styles in the following state and the lane changing state, which are set as the driving style labels of the i-category following state and the lane changing state;

[0055] Step 3.3: According to the vehicle ID, calculate the distances between the characteristic principal component data of the same vehicle and the following cluster center and the lane-changing cluster center, sum the distances of each type of driving style, and based on the minimum distance principle, the minimum value among the i types of driving styles is the corresponding style of the driver.

[0056] Specifically, in this embodiment, the driving style identification model in step 4 is established according to the feature data after dimensionality reduction, including the following steps:

[0057] Step 4.1, according to the following state feature matrix A obtained in step 3.1, the feature data is divided into a training set C and a test set D, the feature data of the training set C and the corresponding driving style labels obtained in step 3.2 are used as input, a support vector machine classification model is established, the historical feature data and the corresponding driving style labels are learned, and the model parameters are optimized through grid search and cross-validation;

[0058] Step 4.2, perform the same processing on the lane-changing state characteristic matrix B obtained in step 3.1;

[0059] Step 4.3: After the training model is established, the model is tested using the test set D and compared with the recognition accuracy of the random forest and KNN algorithms. If the test value reaches the expected accuracy, the driving style recognition model is established. At the same time, by comparing the recognition accuracy before dimensionality reduction, it can be verified that dimensionality reduction helps improve the recognition accuracy of the model.

[0060] The present invention is based on natural driving trajectory data and machine learning algorithms to help judge the driver's driving style in different driving scenarios, and judge the driver's driving style according to the characteristic parameters that characterize the driving style. It can identify high-risk drivers and take appropriate education and training in advance to prevent traffic accidents. In addition, it can help the automatic driving system to more fully understand the driver's intentions, make more reasonable decisions, and provide corresponding early warning and intervention mechanisms to effectively prevent the occurrence of driving accidents and improve the safety of road vehicles.

[0061] The number of devices and processing scales described here are used to simplify the description of the present invention, and the application, modification and variation of the present invention will be obvious to those skilled in the art.

[0062] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and implementation modes. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A driving style intelligent recognition method based on natural driving data, characterized in that: include: Acquire vehicle trajectory data according to natural driving data, and divide the vehicle trajectory data into a car-following data set and a lane-changing data set; Obtain the driving style characterization parameters in the following state and the lane changing state, and obtain the driving style principal component score matrix after dimensionality reduction processing; According to the driving style characterization parameters, a driving style clustering model and a driving style identification model based on support vector machine are constructed; The driving style clustering model is used to determine the driving style in the following state and the lane changing state, and the driver's style is determined by combining the two driving states; The driving style recognition model is used to identify the driver's driving style in a car-following state and a lane-changing state.

2. The method for intelligently identifying driving style based on natural driving data according to claim 1, characterized in that: The vehicle trajectory data includes vehicle ID, vehicle speed, vehicle horizontal and vertical coordinates, vehicle running time, vehicle headway and vehicle headway.

3. The method for intelligently identifying driving style based on natural driving data according to claim 1, characterized in that: The establishment of the car-following dataset and the lane-changing dataset based on the natural driving data specifically includes the following steps: Step 1.1, set the driving state extraction rules, extract the natural driving vehicle trajectory data into a car-following data set and a lane-changing data set, and use the Savitzky-Golay filtering algorithm for smoothing; Step 1.2: The rules for dividing the following vehicle group are as follows: the maximum distance between the following vehicles is 150m, the minimum distance between the following vehicles is 5m, the maximum headway between the following vehicles is 6s, the minimum duration of following vehicles is 15s, and the target vehicle must have a preceding vehicle; Step 1.3: The rules for dividing the lane-changing vehicle group are as follows: the total lane-changing state duration is 12 seconds, and the frame when the lane is changed is used as the judgment frame. It is considered that the first 60 frames and the last 60 frames of the frame, a total of 120 frames, or 12 seconds, together constitute the lane-changing state. In addition, the lane-changing vehicle must have a preceding vehicle and must not have stopped in between.

4. The method for intelligently identifying driving style based on natural driving data according to claim 1, characterized in that: The establishment of the driving style principal component score matrix specifically includes the following steps: Step 2.1, selecting feature parameters that can fully characterize the driving style according to the characteristics of the following state and the lane changing state to form a high-dimensional driving style feature parameter set; Step 2.2, using correlation analysis on the following parameters and the lane changing parameters, deleting some characteristic parameters with low correlation, and preliminarily reducing the dimension of the driving style characteristic parameters; Step 2.3: Use principal component analysis to reduce the dimension. To retain the information in the original data as much as possible, retain all principal components that make the cumulative contribution rate of the principal components exceed 80% as the dimensionality reduction result of the feature parameters, and obtain the driving style principal component score matrix.

5. The method for intelligently identifying driving style based on natural driving data according to claim 4, characterized in that: The driving style characteristic parameters of the following state include speed, acceleration, following time and corresponding driving distance, proportion of time above average vehicle speed, headway, headway, impact degree, and number of sudden acceleration and deceleration.

6. The method for intelligently identifying driving style based on natural driving data according to claim 5, characterized in that: The driving style characteristic parameters of the lane changing state include speed, acceleration, proportion of time above the average vehicle speed, headway, headway, impact degree, lateral speed, lateral acceleration, lateral impact degree, lane changing time and corresponding driving distance, heading angle and lane changeable gap.

7. The method for intelligently identifying driving style based on natural driving data according to claim 1, characterized in that: The driving style classification model is established based on the feature data after dimensionality reduction, including the following steps: Step 3.1, constructing a following state feature matrix A and a lane-changing state feature matrix B according to the principal components of the driving style feature parameter set after dimensionality reduction; Step 3.2, cluster analysis is performed on the characteristic matrix A of the following state and the characteristic matrix B of the lane changing state respectively. According to the clustering results, the SSE and the silhouette coefficient are calculated, and the data of the following state data set and the lane changing state data set are divided into i categories. It is considered that there are i driving styles in the following state and the lane changing state, which are set as the driving style labels of the i-category following state and the lane changing state; Step 3.3: According to the vehicle ID, calculate the distances between the characteristic principal component data of the same vehicle and the following cluster center and the lane-changing cluster center, sum the distances of each type of driving style, and based on the minimum distance principle, the minimum value among the i types of driving styles is the corresponding style of the driver.

8. The method for intelligently identifying driving style based on natural driving data according to claim 1, characterized in that: The driving style identification model is established based on the driving style main component scores after dimensionality reduction and the driving style labels obtained by the driving style classification model, including the following steps: Step 4.1, according to the following state feature matrix A obtained in step 3.1, the feature data is divided into a training set C and a test set D, the feature data of the training set C and the corresponding driving style labels obtained in step 3.2 are used as input, a support vector machine classification model is established, the historical feature data and the corresponding driving style labels are learned, and the model parameters are optimized through grid search and cross-validation; Step 4.2, perform the same processing on the lane-changing state characteristic matrix B obtained in step 3.1; Step 4.3: After the training model is established, the model is tested using the test set D and compared with the recognition accuracy of the random forest and KNN algorithms. If the test value reaches the expected accuracy, the driving style recognition model is established. At the same time, by comparing with the recognition accuracy before dimensionality reduction, it can be verified that dimensionality reduction helps improve the recognition accuracy of the model.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images